xref: /linux/drivers/net/tun.c (revision 957e3facd147510f2cf8780e38606f1d707f0e33)
1 /*
2  *  TUN - Universal TUN/TAP device driver.
3  *  Copyright (C) 1999-2002 Maxim Krasnyansky <maxk@qualcomm.com>
4  *
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.
9  *
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.
14  *
15  *  $Id: tun.c,v 1.15 2002/03/01 02:44:24 maxk Exp $
16  */
17 
18 /*
19  *  Changes:
20  *
21  *  Mike Kershaw <dragorn@kismetwireless.net> 2005/08/14
22  *    Add TUNSETLINK ioctl to set the link encapsulation
23  *
24  *  Mark Smith <markzzzsmith@yahoo.com.au>
25  *    Use eth_random_addr() for tap MAC address.
26  *
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.
30  *    Added ethtool API.
31  *    Minor cleanups
32  *
33  *  Daniel Podlejski <underley@underley.eu.org>
34  *    Modifications for 2.3.99-pre5 kernel.
35  */
36 
37 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
38 
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>"
43 
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>
59 #include <linux/if.h>
60 #include <linux/if_arp.h>
61 #include <linux/if_ether.h>
62 #include <linux/if_tun.h>
63 #include <linux/if_vlan.h>
64 #include <linux/crc32.h>
65 #include <linux/nsproxy.h>
66 #include <linux/virtio_net.h>
67 #include <linux/rcupdate.h>
68 #include <net/ipv6.h>
69 #include <net/net_namespace.h>
70 #include <net/netns/generic.h>
71 #include <net/rtnetlink.h>
72 #include <net/sock.h>
73 #include <linux/seq_file.h>
74 #include <linux/uio.h>
75 
76 #include <asm/uaccess.h>
77 
78 /* Uncomment to enable debugging */
79 /* #define TUN_DEBUG 1 */
80 
81 #ifdef TUN_DEBUG
82 static int debug;
83 
84 #define tun_debug(level, tun, fmt, args...)			\
85 do {								\
86 	if (tun->debug)						\
87 		netdev_printk(level, tun->dev, fmt, ##args);	\
88 } while (0)
89 #define DBG1(level, fmt, args...)				\
90 do {								\
91 	if (debug == 2)						\
92 		printk(level fmt, ##args);			\
93 } while (0)
94 #else
95 #define tun_debug(level, tun, fmt, args...)			\
96 do {								\
97 	if (0)							\
98 		netdev_printk(level, tun->dev, fmt, ##args);	\
99 } while (0)
100 #define DBG1(level, fmt, args...)				\
101 do {								\
102 	if (0)							\
103 		printk(level fmt, ##args);			\
104 } while (0)
105 #endif
106 
107 /* TUN device flags */
108 
109 /* IFF_ATTACH_QUEUE is never stored in device flags,
110  * overload it to mean fasync when stored there.
111  */
112 #define TUN_FASYNC	IFF_ATTACH_QUEUE
113 
114 #define TUN_FEATURES (IFF_NO_PI | IFF_ONE_QUEUE | IFF_VNET_HDR | \
115 		      IFF_VNET_LE | IFF_MULTI_QUEUE)
116 #define GOODCOPY_LEN 128
117 
118 #define FLT_EXACT_COUNT 8
119 struct tap_filter {
120 	unsigned int    count;    /* Number of addrs. Zero means disabled */
121 	u32             mask[2];  /* Mask of the hashed addrs */
122 	unsigned char	addr[FLT_EXACT_COUNT][ETH_ALEN];
123 };
124 
125 /* DEFAULT_MAX_NUM_RSS_QUEUES were chosen to let the rx/tx queues allocated for
126  * the netdevice to be fit in one page. So we can make sure the success of
127  * memory allocation. TODO: increase the limit. */
128 #define MAX_TAP_QUEUES DEFAULT_MAX_NUM_RSS_QUEUES
129 #define MAX_TAP_FLOWS  4096
130 
131 #define TUN_FLOW_EXPIRE (3 * HZ)
132 
133 /* A tun_file connects an open character device to a tuntap netdevice. It
134  * also contains all socket related structures (except sock_fprog and tap_filter)
135  * to serve as one transmit queue for tuntap device. The sock_fprog and
136  * tap_filter were kept in tun_struct since they were used for filtering for the
137  * netdevice not for a specific queue (at least I didn't see the requirement for
138  * this).
139  *
140  * RCU usage:
141  * The tun_file and tun_struct are loosely coupled, the pointer from one to the
142  * other can only be read while rcu_read_lock or rtnl_lock is held.
143  */
144 struct tun_file {
145 	struct sock sk;
146 	struct socket socket;
147 	struct socket_wq wq;
148 	struct tun_struct __rcu *tun;
149 	struct net *net;
150 	struct fasync_struct *fasync;
151 	/* only used for fasnyc */
152 	unsigned int flags;
153 	union {
154 		u16 queue_index;
155 		unsigned int ifindex;
156 	};
157 	struct list_head next;
158 	struct tun_struct *detached;
159 };
160 
161 struct tun_flow_entry {
162 	struct hlist_node hash_link;
163 	struct rcu_head rcu;
164 	struct tun_struct *tun;
165 
166 	u32 rxhash;
167 	u32 rps_rxhash;
168 	int queue_index;
169 	unsigned long updated;
170 };
171 
172 #define TUN_NUM_FLOW_ENTRIES 1024
173 
174 /* Since the socket were moved to tun_file, to preserve the behavior of persist
175  * device, socket filter, sndbuf and vnet header size were restore when the
176  * file were attached to a persist device.
177  */
178 struct tun_struct {
179 	struct tun_file __rcu	*tfiles[MAX_TAP_QUEUES];
180 	unsigned int            numqueues;
181 	unsigned int 		flags;
182 	kuid_t			owner;
183 	kgid_t			group;
184 
185 	struct net_device	*dev;
186 	netdev_features_t	set_features;
187 #define TUN_USER_FEATURES (NETIF_F_HW_CSUM|NETIF_F_TSO_ECN|NETIF_F_TSO| \
188 			  NETIF_F_TSO6)
189 
190 	int			vnet_hdr_sz;
191 	int			sndbuf;
192 	struct tap_filter	txflt;
193 	struct sock_fprog	fprog;
194 	/* protected by rtnl lock */
195 	bool			filter_attached;
196 #ifdef TUN_DEBUG
197 	int debug;
198 #endif
199 	spinlock_t lock;
200 	struct hlist_head flows[TUN_NUM_FLOW_ENTRIES];
201 	struct timer_list flow_gc_timer;
202 	unsigned long ageing_time;
203 	unsigned int numdisabled;
204 	struct list_head disabled;
205 	void *security;
206 	u32 flow_count;
207 };
208 
209 static inline u16 tun16_to_cpu(struct tun_struct *tun, __virtio16 val)
210 {
211 	return __virtio16_to_cpu(tun->flags & IFF_VNET_LE, val);
212 }
213 
214 static inline __virtio16 cpu_to_tun16(struct tun_struct *tun, u16 val)
215 {
216 	return __cpu_to_virtio16(tun->flags & IFF_VNET_LE, val);
217 }
218 
219 static inline u32 tun_hashfn(u32 rxhash)
220 {
221 	return rxhash & 0x3ff;
222 }
223 
224 static struct tun_flow_entry *tun_flow_find(struct hlist_head *head, u32 rxhash)
225 {
226 	struct tun_flow_entry *e;
227 
228 	hlist_for_each_entry_rcu(e, head, hash_link) {
229 		if (e->rxhash == rxhash)
230 			return e;
231 	}
232 	return NULL;
233 }
234 
235 static struct tun_flow_entry *tun_flow_create(struct tun_struct *tun,
236 					      struct hlist_head *head,
237 					      u32 rxhash, u16 queue_index)
238 {
239 	struct tun_flow_entry *e = kmalloc(sizeof(*e), GFP_ATOMIC);
240 
241 	if (e) {
242 		tun_debug(KERN_INFO, tun, "create flow: hash %u index %u\n",
243 			  rxhash, queue_index);
244 		e->updated = jiffies;
245 		e->rxhash = rxhash;
246 		e->rps_rxhash = 0;
247 		e->queue_index = queue_index;
248 		e->tun = tun;
249 		hlist_add_head_rcu(&e->hash_link, head);
250 		++tun->flow_count;
251 	}
252 	return e;
253 }
254 
255 static void tun_flow_delete(struct tun_struct *tun, struct tun_flow_entry *e)
256 {
257 	tun_debug(KERN_INFO, tun, "delete flow: hash %u index %u\n",
258 		  e->rxhash, e->queue_index);
259 	sock_rps_reset_flow_hash(e->rps_rxhash);
260 	hlist_del_rcu(&e->hash_link);
261 	kfree_rcu(e, rcu);
262 	--tun->flow_count;
263 }
264 
265 static void tun_flow_flush(struct tun_struct *tun)
266 {
267 	int i;
268 
269 	spin_lock_bh(&tun->lock);
270 	for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
271 		struct tun_flow_entry *e;
272 		struct hlist_node *n;
273 
274 		hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link)
275 			tun_flow_delete(tun, e);
276 	}
277 	spin_unlock_bh(&tun->lock);
278 }
279 
280 static void tun_flow_delete_by_queue(struct tun_struct *tun, u16 queue_index)
281 {
282 	int i;
283 
284 	spin_lock_bh(&tun->lock);
285 	for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
286 		struct tun_flow_entry *e;
287 		struct hlist_node *n;
288 
289 		hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
290 			if (e->queue_index == queue_index)
291 				tun_flow_delete(tun, e);
292 		}
293 	}
294 	spin_unlock_bh(&tun->lock);
295 }
296 
297 static void tun_flow_cleanup(unsigned long data)
298 {
299 	struct tun_struct *tun = (struct tun_struct *)data;
300 	unsigned long delay = tun->ageing_time;
301 	unsigned long next_timer = jiffies + delay;
302 	unsigned long count = 0;
303 	int i;
304 
305 	tun_debug(KERN_INFO, tun, "tun_flow_cleanup\n");
306 
307 	spin_lock_bh(&tun->lock);
308 	for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
309 		struct tun_flow_entry *e;
310 		struct hlist_node *n;
311 
312 		hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
313 			unsigned long this_timer;
314 			count++;
315 			this_timer = e->updated + delay;
316 			if (time_before_eq(this_timer, jiffies))
317 				tun_flow_delete(tun, e);
318 			else if (time_before(this_timer, next_timer))
319 				next_timer = this_timer;
320 		}
321 	}
322 
323 	if (count)
324 		mod_timer(&tun->flow_gc_timer, round_jiffies_up(next_timer));
325 	spin_unlock_bh(&tun->lock);
326 }
327 
328 static void tun_flow_update(struct tun_struct *tun, u32 rxhash,
329 			    struct tun_file *tfile)
330 {
331 	struct hlist_head *head;
332 	struct tun_flow_entry *e;
333 	unsigned long delay = tun->ageing_time;
334 	u16 queue_index = tfile->queue_index;
335 
336 	if (!rxhash)
337 		return;
338 	else
339 		head = &tun->flows[tun_hashfn(rxhash)];
340 
341 	rcu_read_lock();
342 
343 	/* We may get a very small possibility of OOO during switching, not
344 	 * worth to optimize.*/
345 	if (tun->numqueues == 1 || tfile->detached)
346 		goto unlock;
347 
348 	e = tun_flow_find(head, rxhash);
349 	if (likely(e)) {
350 		/* TODO: keep queueing to old queue until it's empty? */
351 		e->queue_index = queue_index;
352 		e->updated = jiffies;
353 		sock_rps_record_flow_hash(e->rps_rxhash);
354 	} else {
355 		spin_lock_bh(&tun->lock);
356 		if (!tun_flow_find(head, rxhash) &&
357 		    tun->flow_count < MAX_TAP_FLOWS)
358 			tun_flow_create(tun, head, rxhash, queue_index);
359 
360 		if (!timer_pending(&tun->flow_gc_timer))
361 			mod_timer(&tun->flow_gc_timer,
362 				  round_jiffies_up(jiffies + delay));
363 		spin_unlock_bh(&tun->lock);
364 	}
365 
366 unlock:
367 	rcu_read_unlock();
368 }
369 
370 /**
371  * Save the hash received in the stack receive path and update the
372  * flow_hash table accordingly.
373  */
374 static inline void tun_flow_save_rps_rxhash(struct tun_flow_entry *e, u32 hash)
375 {
376 	if (unlikely(e->rps_rxhash != hash)) {
377 		sock_rps_reset_flow_hash(e->rps_rxhash);
378 		e->rps_rxhash = hash;
379 	}
380 }
381 
382 /* We try to identify a flow through its rxhash first. The reason that
383  * we do not check rxq no. is because some cards(e.g 82599), chooses
384  * the rxq based on the txq where the last packet of the flow comes. As
385  * the userspace application move between processors, we may get a
386  * different rxq no. here. If we could not get rxhash, then we would
387  * hope the rxq no. may help here.
388  */
389 static u16 tun_select_queue(struct net_device *dev, struct sk_buff *skb,
390 			    void *accel_priv, select_queue_fallback_t fallback)
391 {
392 	struct tun_struct *tun = netdev_priv(dev);
393 	struct tun_flow_entry *e;
394 	u32 txq = 0;
395 	u32 numqueues = 0;
396 
397 	rcu_read_lock();
398 	numqueues = ACCESS_ONCE(tun->numqueues);
399 
400 	txq = skb_get_hash(skb);
401 	if (txq) {
402 		e = tun_flow_find(&tun->flows[tun_hashfn(txq)], txq);
403 		if (e) {
404 			tun_flow_save_rps_rxhash(e, txq);
405 			txq = e->queue_index;
406 		} else
407 			/* use multiply and shift instead of expensive divide */
408 			txq = ((u64)txq * numqueues) >> 32;
409 	} else if (likely(skb_rx_queue_recorded(skb))) {
410 		txq = skb_get_rx_queue(skb);
411 		while (unlikely(txq >= numqueues))
412 			txq -= numqueues;
413 	}
414 
415 	rcu_read_unlock();
416 	return txq;
417 }
418 
419 static inline bool tun_not_capable(struct tun_struct *tun)
420 {
421 	const struct cred *cred = current_cred();
422 	struct net *net = dev_net(tun->dev);
423 
424 	return ((uid_valid(tun->owner) && !uid_eq(cred->euid, tun->owner)) ||
425 		  (gid_valid(tun->group) && !in_egroup_p(tun->group))) &&
426 		!ns_capable(net->user_ns, CAP_NET_ADMIN);
427 }
428 
429 static void tun_set_real_num_queues(struct tun_struct *tun)
430 {
431 	netif_set_real_num_tx_queues(tun->dev, tun->numqueues);
432 	netif_set_real_num_rx_queues(tun->dev, tun->numqueues);
433 }
434 
435 static void tun_disable_queue(struct tun_struct *tun, struct tun_file *tfile)
436 {
437 	tfile->detached = tun;
438 	list_add_tail(&tfile->next, &tun->disabled);
439 	++tun->numdisabled;
440 }
441 
442 static struct tun_struct *tun_enable_queue(struct tun_file *tfile)
443 {
444 	struct tun_struct *tun = tfile->detached;
445 
446 	tfile->detached = NULL;
447 	list_del_init(&tfile->next);
448 	--tun->numdisabled;
449 	return tun;
450 }
451 
452 static void tun_queue_purge(struct tun_file *tfile)
453 {
454 	skb_queue_purge(&tfile->sk.sk_receive_queue);
455 	skb_queue_purge(&tfile->sk.sk_error_queue);
456 }
457 
458 static void __tun_detach(struct tun_file *tfile, bool clean)
459 {
460 	struct tun_file *ntfile;
461 	struct tun_struct *tun;
462 
463 	tun = rtnl_dereference(tfile->tun);
464 
465 	if (tun && !tfile->detached) {
466 		u16 index = tfile->queue_index;
467 		BUG_ON(index >= tun->numqueues);
468 
469 		rcu_assign_pointer(tun->tfiles[index],
470 				   tun->tfiles[tun->numqueues - 1]);
471 		ntfile = rtnl_dereference(tun->tfiles[index]);
472 		ntfile->queue_index = index;
473 
474 		--tun->numqueues;
475 		if (clean) {
476 			RCU_INIT_POINTER(tfile->tun, NULL);
477 			sock_put(&tfile->sk);
478 		} else
479 			tun_disable_queue(tun, tfile);
480 
481 		synchronize_net();
482 		tun_flow_delete_by_queue(tun, tun->numqueues + 1);
483 		/* Drop read queue */
484 		tun_queue_purge(tfile);
485 		tun_set_real_num_queues(tun);
486 	} else if (tfile->detached && clean) {
487 		tun = tun_enable_queue(tfile);
488 		sock_put(&tfile->sk);
489 	}
490 
491 	if (clean) {
492 		if (tun && tun->numqueues == 0 && tun->numdisabled == 0) {
493 			netif_carrier_off(tun->dev);
494 
495 			if (!(tun->flags & IFF_PERSIST) &&
496 			    tun->dev->reg_state == NETREG_REGISTERED)
497 				unregister_netdevice(tun->dev);
498 		}
499 
500 		BUG_ON(!test_bit(SOCK_EXTERNALLY_ALLOCATED,
501 				 &tfile->socket.flags));
502 		sk_release_kernel(&tfile->sk);
503 	}
504 }
505 
506 static void tun_detach(struct tun_file *tfile, bool clean)
507 {
508 	rtnl_lock();
509 	__tun_detach(tfile, clean);
510 	rtnl_unlock();
511 }
512 
513 static void tun_detach_all(struct net_device *dev)
514 {
515 	struct tun_struct *tun = netdev_priv(dev);
516 	struct tun_file *tfile, *tmp;
517 	int i, n = tun->numqueues;
518 
519 	for (i = 0; i < n; i++) {
520 		tfile = rtnl_dereference(tun->tfiles[i]);
521 		BUG_ON(!tfile);
522 		tfile->socket.sk->sk_data_ready(tfile->socket.sk);
523 		RCU_INIT_POINTER(tfile->tun, NULL);
524 		--tun->numqueues;
525 	}
526 	list_for_each_entry(tfile, &tun->disabled, next) {
527 		tfile->socket.sk->sk_data_ready(tfile->socket.sk);
528 		RCU_INIT_POINTER(tfile->tun, NULL);
529 	}
530 	BUG_ON(tun->numqueues != 0);
531 
532 	synchronize_net();
533 	for (i = 0; i < n; i++) {
534 		tfile = rtnl_dereference(tun->tfiles[i]);
535 		/* Drop read queue */
536 		tun_queue_purge(tfile);
537 		sock_put(&tfile->sk);
538 	}
539 	list_for_each_entry_safe(tfile, tmp, &tun->disabled, next) {
540 		tun_enable_queue(tfile);
541 		tun_queue_purge(tfile);
542 		sock_put(&tfile->sk);
543 	}
544 	BUG_ON(tun->numdisabled != 0);
545 
546 	if (tun->flags & IFF_PERSIST)
547 		module_put(THIS_MODULE);
548 }
549 
550 static int tun_attach(struct tun_struct *tun, struct file *file, bool skip_filter)
551 {
552 	struct tun_file *tfile = file->private_data;
553 	int err;
554 
555 	err = security_tun_dev_attach(tfile->socket.sk, tun->security);
556 	if (err < 0)
557 		goto out;
558 
559 	err = -EINVAL;
560 	if (rtnl_dereference(tfile->tun) && !tfile->detached)
561 		goto out;
562 
563 	err = -EBUSY;
564 	if (!(tun->flags & IFF_MULTI_QUEUE) && tun->numqueues == 1)
565 		goto out;
566 
567 	err = -E2BIG;
568 	if (!tfile->detached &&
569 	    tun->numqueues + tun->numdisabled == MAX_TAP_QUEUES)
570 		goto out;
571 
572 	err = 0;
573 
574 	/* Re-attach the filter to persist device */
575 	if (!skip_filter && (tun->filter_attached == true)) {
576 		err = sk_attach_filter(&tun->fprog, tfile->socket.sk);
577 		if (!err)
578 			goto out;
579 	}
580 	tfile->queue_index = tun->numqueues;
581 	rcu_assign_pointer(tfile->tun, tun);
582 	rcu_assign_pointer(tun->tfiles[tun->numqueues], tfile);
583 	tun->numqueues++;
584 
585 	if (tfile->detached)
586 		tun_enable_queue(tfile);
587 	else
588 		sock_hold(&tfile->sk);
589 
590 	tun_set_real_num_queues(tun);
591 
592 	/* device is allowed to go away first, so no need to hold extra
593 	 * refcnt.
594 	 */
595 
596 out:
597 	return err;
598 }
599 
600 static struct tun_struct *__tun_get(struct tun_file *tfile)
601 {
602 	struct tun_struct *tun;
603 
604 	rcu_read_lock();
605 	tun = rcu_dereference(tfile->tun);
606 	if (tun)
607 		dev_hold(tun->dev);
608 	rcu_read_unlock();
609 
610 	return tun;
611 }
612 
613 static struct tun_struct *tun_get(struct file *file)
614 {
615 	return __tun_get(file->private_data);
616 }
617 
618 static void tun_put(struct tun_struct *tun)
619 {
620 	dev_put(tun->dev);
621 }
622 
623 /* TAP filtering */
624 static void addr_hash_set(u32 *mask, const u8 *addr)
625 {
626 	int n = ether_crc(ETH_ALEN, addr) >> 26;
627 	mask[n >> 5] |= (1 << (n & 31));
628 }
629 
630 static unsigned int addr_hash_test(const u32 *mask, const u8 *addr)
631 {
632 	int n = ether_crc(ETH_ALEN, addr) >> 26;
633 	return mask[n >> 5] & (1 << (n & 31));
634 }
635 
636 static int update_filter(struct tap_filter *filter, void __user *arg)
637 {
638 	struct { u8 u[ETH_ALEN]; } *addr;
639 	struct tun_filter uf;
640 	int err, alen, n, nexact;
641 
642 	if (copy_from_user(&uf, arg, sizeof(uf)))
643 		return -EFAULT;
644 
645 	if (!uf.count) {
646 		/* Disabled */
647 		filter->count = 0;
648 		return 0;
649 	}
650 
651 	alen = ETH_ALEN * uf.count;
652 	addr = kmalloc(alen, GFP_KERNEL);
653 	if (!addr)
654 		return -ENOMEM;
655 
656 	if (copy_from_user(addr, arg + sizeof(uf), alen)) {
657 		err = -EFAULT;
658 		goto done;
659 	}
660 
661 	/* The filter is updated without holding any locks. Which is
662 	 * perfectly safe. We disable it first and in the worst
663 	 * case we'll accept a few undesired packets. */
664 	filter->count = 0;
665 	wmb();
666 
667 	/* Use first set of addresses as an exact filter */
668 	for (n = 0; n < uf.count && n < FLT_EXACT_COUNT; n++)
669 		memcpy(filter->addr[n], addr[n].u, ETH_ALEN);
670 
671 	nexact = n;
672 
673 	/* Remaining multicast addresses are hashed,
674 	 * unicast will leave the filter disabled. */
675 	memset(filter->mask, 0, sizeof(filter->mask));
676 	for (; n < uf.count; n++) {
677 		if (!is_multicast_ether_addr(addr[n].u)) {
678 			err = 0; /* no filter */
679 			goto done;
680 		}
681 		addr_hash_set(filter->mask, addr[n].u);
682 	}
683 
684 	/* For ALLMULTI just set the mask to all ones.
685 	 * This overrides the mask populated above. */
686 	if ((uf.flags & TUN_FLT_ALLMULTI))
687 		memset(filter->mask, ~0, sizeof(filter->mask));
688 
689 	/* Now enable the filter */
690 	wmb();
691 	filter->count = nexact;
692 
693 	/* Return the number of exact filters */
694 	err = nexact;
695 
696 done:
697 	kfree(addr);
698 	return err;
699 }
700 
701 /* Returns: 0 - drop, !=0 - accept */
702 static int run_filter(struct tap_filter *filter, const struct sk_buff *skb)
703 {
704 	/* Cannot use eth_hdr(skb) here because skb_mac_hdr() is incorrect
705 	 * at this point. */
706 	struct ethhdr *eh = (struct ethhdr *) skb->data;
707 	int i;
708 
709 	/* Exact match */
710 	for (i = 0; i < filter->count; i++)
711 		if (ether_addr_equal(eh->h_dest, filter->addr[i]))
712 			return 1;
713 
714 	/* Inexact match (multicast only) */
715 	if (is_multicast_ether_addr(eh->h_dest))
716 		return addr_hash_test(filter->mask, eh->h_dest);
717 
718 	return 0;
719 }
720 
721 /*
722  * Checks whether the packet is accepted or not.
723  * Returns: 0 - drop, !=0 - accept
724  */
725 static int check_filter(struct tap_filter *filter, const struct sk_buff *skb)
726 {
727 	if (!filter->count)
728 		return 1;
729 
730 	return run_filter(filter, skb);
731 }
732 
733 /* Network device part of the driver */
734 
735 static const struct ethtool_ops tun_ethtool_ops;
736 
737 /* Net device detach from fd. */
738 static void tun_net_uninit(struct net_device *dev)
739 {
740 	tun_detach_all(dev);
741 }
742 
743 /* Net device open. */
744 static int tun_net_open(struct net_device *dev)
745 {
746 	netif_tx_start_all_queues(dev);
747 	return 0;
748 }
749 
750 /* Net device close. */
751 static int tun_net_close(struct net_device *dev)
752 {
753 	netif_tx_stop_all_queues(dev);
754 	return 0;
755 }
756 
757 /* Net device start xmit */
758 static netdev_tx_t tun_net_xmit(struct sk_buff *skb, struct net_device *dev)
759 {
760 	struct tun_struct *tun = netdev_priv(dev);
761 	int txq = skb->queue_mapping;
762 	struct tun_file *tfile;
763 	u32 numqueues = 0;
764 
765 	rcu_read_lock();
766 	tfile = rcu_dereference(tun->tfiles[txq]);
767 	numqueues = ACCESS_ONCE(tun->numqueues);
768 
769 	/* Drop packet if interface is not attached */
770 	if (txq >= numqueues)
771 		goto drop;
772 
773 	if (numqueues == 1) {
774 		/* Select queue was not called for the skbuff, so we extract the
775 		 * RPS hash and save it into the flow_table here.
776 		 */
777 		__u32 rxhash;
778 
779 		rxhash = skb_get_hash(skb);
780 		if (rxhash) {
781 			struct tun_flow_entry *e;
782 			e = tun_flow_find(&tun->flows[tun_hashfn(rxhash)],
783 					rxhash);
784 			if (e)
785 				tun_flow_save_rps_rxhash(e, rxhash);
786 		}
787 	}
788 
789 	tun_debug(KERN_INFO, tun, "tun_net_xmit %d\n", skb->len);
790 
791 	BUG_ON(!tfile);
792 
793 	/* Drop if the filter does not like it.
794 	 * This is a noop if the filter is disabled.
795 	 * Filter can be enabled only for the TAP devices. */
796 	if (!check_filter(&tun->txflt, skb))
797 		goto drop;
798 
799 	if (tfile->socket.sk->sk_filter &&
800 	    sk_filter(tfile->socket.sk, skb))
801 		goto drop;
802 
803 	/* Limit the number of packets queued by dividing txq length with the
804 	 * number of queues.
805 	 */
806 	if (skb_queue_len(&tfile->socket.sk->sk_receive_queue) * numqueues
807 			  >= dev->tx_queue_len)
808 		goto drop;
809 
810 	if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC)))
811 		goto drop;
812 
813 	if (skb->sk) {
814 		sock_tx_timestamp(skb->sk, &skb_shinfo(skb)->tx_flags);
815 		sw_tx_timestamp(skb);
816 	}
817 
818 	/* Orphan the skb - required as we might hang on to it
819 	 * for indefinite time.
820 	 */
821 	skb_orphan(skb);
822 
823 	nf_reset(skb);
824 
825 	/* Enqueue packet */
826 	skb_queue_tail(&tfile->socket.sk->sk_receive_queue, skb);
827 
828 	/* Notify and wake up reader process */
829 	if (tfile->flags & TUN_FASYNC)
830 		kill_fasync(&tfile->fasync, SIGIO, POLL_IN);
831 	tfile->socket.sk->sk_data_ready(tfile->socket.sk);
832 
833 	rcu_read_unlock();
834 	return NETDEV_TX_OK;
835 
836 drop:
837 	dev->stats.tx_dropped++;
838 	skb_tx_error(skb);
839 	kfree_skb(skb);
840 	rcu_read_unlock();
841 	return NET_XMIT_DROP;
842 }
843 
844 static void tun_net_mclist(struct net_device *dev)
845 {
846 	/*
847 	 * This callback is supposed to deal with mc filter in
848 	 * _rx_ path and has nothing to do with the _tx_ path.
849 	 * In rx path we always accept everything userspace gives us.
850 	 */
851 }
852 
853 #define MIN_MTU 68
854 #define MAX_MTU 65535
855 
856 static int
857 tun_net_change_mtu(struct net_device *dev, int new_mtu)
858 {
859 	if (new_mtu < MIN_MTU || new_mtu + dev->hard_header_len > MAX_MTU)
860 		return -EINVAL;
861 	dev->mtu = new_mtu;
862 	return 0;
863 }
864 
865 static netdev_features_t tun_net_fix_features(struct net_device *dev,
866 	netdev_features_t features)
867 {
868 	struct tun_struct *tun = netdev_priv(dev);
869 
870 	return (features & tun->set_features) | (features & ~TUN_USER_FEATURES);
871 }
872 #ifdef CONFIG_NET_POLL_CONTROLLER
873 static void tun_poll_controller(struct net_device *dev)
874 {
875 	/*
876 	 * Tun only receives frames when:
877 	 * 1) the char device endpoint gets data from user space
878 	 * 2) the tun socket gets a sendmsg call from user space
879 	 * Since both of those are synchronous operations, we are guaranteed
880 	 * never to have pending data when we poll for it
881 	 * so there is nothing to do here but return.
882 	 * We need this though so netpoll recognizes us as an interface that
883 	 * supports polling, which enables bridge devices in virt setups to
884 	 * still use netconsole
885 	 */
886 	return;
887 }
888 #endif
889 static const struct net_device_ops tun_netdev_ops = {
890 	.ndo_uninit		= tun_net_uninit,
891 	.ndo_open		= tun_net_open,
892 	.ndo_stop		= tun_net_close,
893 	.ndo_start_xmit		= tun_net_xmit,
894 	.ndo_change_mtu		= tun_net_change_mtu,
895 	.ndo_fix_features	= tun_net_fix_features,
896 	.ndo_select_queue	= tun_select_queue,
897 #ifdef CONFIG_NET_POLL_CONTROLLER
898 	.ndo_poll_controller	= tun_poll_controller,
899 #endif
900 };
901 
902 static const struct net_device_ops tap_netdev_ops = {
903 	.ndo_uninit		= tun_net_uninit,
904 	.ndo_open		= tun_net_open,
905 	.ndo_stop		= tun_net_close,
906 	.ndo_start_xmit		= tun_net_xmit,
907 	.ndo_change_mtu		= tun_net_change_mtu,
908 	.ndo_fix_features	= tun_net_fix_features,
909 	.ndo_set_rx_mode	= tun_net_mclist,
910 	.ndo_set_mac_address	= eth_mac_addr,
911 	.ndo_validate_addr	= eth_validate_addr,
912 	.ndo_select_queue	= tun_select_queue,
913 #ifdef CONFIG_NET_POLL_CONTROLLER
914 	.ndo_poll_controller	= tun_poll_controller,
915 #endif
916 };
917 
918 static void tun_flow_init(struct tun_struct *tun)
919 {
920 	int i;
921 
922 	for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++)
923 		INIT_HLIST_HEAD(&tun->flows[i]);
924 
925 	tun->ageing_time = TUN_FLOW_EXPIRE;
926 	setup_timer(&tun->flow_gc_timer, tun_flow_cleanup, (unsigned long)tun);
927 	mod_timer(&tun->flow_gc_timer,
928 		  round_jiffies_up(jiffies + tun->ageing_time));
929 }
930 
931 static void tun_flow_uninit(struct tun_struct *tun)
932 {
933 	del_timer_sync(&tun->flow_gc_timer);
934 	tun_flow_flush(tun);
935 }
936 
937 /* Initialize net device. */
938 static void tun_net_init(struct net_device *dev)
939 {
940 	struct tun_struct *tun = netdev_priv(dev);
941 
942 	switch (tun->flags & TUN_TYPE_MASK) {
943 	case IFF_TUN:
944 		dev->netdev_ops = &tun_netdev_ops;
945 
946 		/* Point-to-Point TUN Device */
947 		dev->hard_header_len = 0;
948 		dev->addr_len = 0;
949 		dev->mtu = 1500;
950 
951 		/* Zero header length */
952 		dev->type = ARPHRD_NONE;
953 		dev->flags = IFF_POINTOPOINT | IFF_NOARP | IFF_MULTICAST;
954 		dev->tx_queue_len = TUN_READQ_SIZE;  /* We prefer our own queue length */
955 		break;
956 
957 	case IFF_TAP:
958 		dev->netdev_ops = &tap_netdev_ops;
959 		/* Ethernet TAP Device */
960 		ether_setup(dev);
961 		dev->priv_flags &= ~IFF_TX_SKB_SHARING;
962 		dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
963 
964 		eth_hw_addr_random(dev);
965 
966 		dev->tx_queue_len = TUN_READQ_SIZE;  /* We prefer our own queue length */
967 		break;
968 	}
969 }
970 
971 /* Character device part */
972 
973 /* Poll */
974 static unsigned int tun_chr_poll(struct file *file, poll_table *wait)
975 {
976 	struct tun_file *tfile = file->private_data;
977 	struct tun_struct *tun = __tun_get(tfile);
978 	struct sock *sk;
979 	unsigned int mask = 0;
980 
981 	if (!tun)
982 		return POLLERR;
983 
984 	sk = tfile->socket.sk;
985 
986 	tun_debug(KERN_INFO, tun, "tun_chr_poll\n");
987 
988 	poll_wait(file, sk_sleep(sk), wait);
989 
990 	if (!skb_queue_empty(&sk->sk_receive_queue))
991 		mask |= POLLIN | POLLRDNORM;
992 
993 	if (sock_writeable(sk) ||
994 	    (!test_and_set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags) &&
995 	     sock_writeable(sk)))
996 		mask |= POLLOUT | POLLWRNORM;
997 
998 	if (tun->dev->reg_state != NETREG_REGISTERED)
999 		mask = POLLERR;
1000 
1001 	tun_put(tun);
1002 	return mask;
1003 }
1004 
1005 /* prepad is the amount to reserve at front.  len is length after that.
1006  * linear is a hint as to how much to copy (usually headers). */
1007 static struct sk_buff *tun_alloc_skb(struct tun_file *tfile,
1008 				     size_t prepad, size_t len,
1009 				     size_t linear, int noblock)
1010 {
1011 	struct sock *sk = tfile->socket.sk;
1012 	struct sk_buff *skb;
1013 	int err;
1014 
1015 	/* Under a page?  Don't bother with paged skb. */
1016 	if (prepad + len < PAGE_SIZE || !linear)
1017 		linear = len;
1018 
1019 	skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
1020 				   &err, 0);
1021 	if (!skb)
1022 		return ERR_PTR(err);
1023 
1024 	skb_reserve(skb, prepad);
1025 	skb_put(skb, linear);
1026 	skb->data_len = len - linear;
1027 	skb->len += len - linear;
1028 
1029 	return skb;
1030 }
1031 
1032 /* Get packet from user space buffer */
1033 static ssize_t tun_get_user(struct tun_struct *tun, struct tun_file *tfile,
1034 			    void *msg_control, struct iov_iter *from,
1035 			    int noblock)
1036 {
1037 	struct tun_pi pi = { 0, cpu_to_be16(ETH_P_IP) };
1038 	struct sk_buff *skb;
1039 	size_t total_len = iov_iter_count(from);
1040 	size_t len = total_len, align = NET_SKB_PAD, linear;
1041 	struct virtio_net_hdr gso = { 0 };
1042 	int good_linear;
1043 	int copylen;
1044 	bool zerocopy = false;
1045 	int err;
1046 	u32 rxhash;
1047 	ssize_t n;
1048 
1049 	if (!(tun->flags & IFF_NO_PI)) {
1050 		if (len < sizeof(pi))
1051 			return -EINVAL;
1052 		len -= sizeof(pi);
1053 
1054 		n = copy_from_iter(&pi, sizeof(pi), from);
1055 		if (n != sizeof(pi))
1056 			return -EFAULT;
1057 	}
1058 
1059 	if (tun->flags & IFF_VNET_HDR) {
1060 		if (len < tun->vnet_hdr_sz)
1061 			return -EINVAL;
1062 		len -= tun->vnet_hdr_sz;
1063 
1064 		n = copy_from_iter(&gso, sizeof(gso), from);
1065 		if (n != sizeof(gso))
1066 			return -EFAULT;
1067 
1068 		if ((gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
1069 		    tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2 > tun16_to_cpu(tun, gso.hdr_len))
1070 			gso.hdr_len = cpu_to_tun16(tun, tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2);
1071 
1072 		if (tun16_to_cpu(tun, gso.hdr_len) > len)
1073 			return -EINVAL;
1074 		iov_iter_advance(from, tun->vnet_hdr_sz - sizeof(gso));
1075 	}
1076 
1077 	if ((tun->flags & TUN_TYPE_MASK) == IFF_TAP) {
1078 		align += NET_IP_ALIGN;
1079 		if (unlikely(len < ETH_HLEN ||
1080 			     (gso.hdr_len && tun16_to_cpu(tun, gso.hdr_len) < ETH_HLEN)))
1081 			return -EINVAL;
1082 	}
1083 
1084 	good_linear = SKB_MAX_HEAD(align);
1085 
1086 	if (msg_control) {
1087 		struct iov_iter i = *from;
1088 
1089 		/* There are 256 bytes to be copied in skb, so there is
1090 		 * enough room for skb expand head in case it is used.
1091 		 * The rest of the buffer is mapped from userspace.
1092 		 */
1093 		copylen = gso.hdr_len ? tun16_to_cpu(tun, gso.hdr_len) : GOODCOPY_LEN;
1094 		if (copylen > good_linear)
1095 			copylen = good_linear;
1096 		linear = copylen;
1097 		iov_iter_advance(&i, copylen);
1098 		if (iov_iter_npages(&i, INT_MAX) <= MAX_SKB_FRAGS)
1099 			zerocopy = true;
1100 	}
1101 
1102 	if (!zerocopy) {
1103 		copylen = len;
1104 		if (tun16_to_cpu(tun, gso.hdr_len) > good_linear)
1105 			linear = good_linear;
1106 		else
1107 			linear = tun16_to_cpu(tun, gso.hdr_len);
1108 	}
1109 
1110 	skb = tun_alloc_skb(tfile, align, copylen, linear, noblock);
1111 	if (IS_ERR(skb)) {
1112 		if (PTR_ERR(skb) != -EAGAIN)
1113 			tun->dev->stats.rx_dropped++;
1114 		return PTR_ERR(skb);
1115 	}
1116 
1117 	if (zerocopy)
1118 		err = zerocopy_sg_from_iter(skb, from);
1119 	else {
1120 		err = skb_copy_datagram_from_iter(skb, 0, from, len);
1121 		if (!err && msg_control) {
1122 			struct ubuf_info *uarg = msg_control;
1123 			uarg->callback(uarg, false);
1124 		}
1125 	}
1126 
1127 	if (err) {
1128 		tun->dev->stats.rx_dropped++;
1129 		kfree_skb(skb);
1130 		return -EFAULT;
1131 	}
1132 
1133 	if (gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
1134 		if (!skb_partial_csum_set(skb, tun16_to_cpu(tun, gso.csum_start),
1135 					  tun16_to_cpu(tun, gso.csum_offset))) {
1136 			tun->dev->stats.rx_frame_errors++;
1137 			kfree_skb(skb);
1138 			return -EINVAL;
1139 		}
1140 	}
1141 
1142 	switch (tun->flags & TUN_TYPE_MASK) {
1143 	case IFF_TUN:
1144 		if (tun->flags & IFF_NO_PI) {
1145 			switch (skb->data[0] & 0xf0) {
1146 			case 0x40:
1147 				pi.proto = htons(ETH_P_IP);
1148 				break;
1149 			case 0x60:
1150 				pi.proto = htons(ETH_P_IPV6);
1151 				break;
1152 			default:
1153 				tun->dev->stats.rx_dropped++;
1154 				kfree_skb(skb);
1155 				return -EINVAL;
1156 			}
1157 		}
1158 
1159 		skb_reset_mac_header(skb);
1160 		skb->protocol = pi.proto;
1161 		skb->dev = tun->dev;
1162 		break;
1163 	case IFF_TAP:
1164 		skb->protocol = eth_type_trans(skb, tun->dev);
1165 		break;
1166 	}
1167 
1168 	skb_reset_network_header(skb);
1169 
1170 	if (gso.gso_type != VIRTIO_NET_HDR_GSO_NONE) {
1171 		pr_debug("GSO!\n");
1172 		switch (gso.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
1173 		case VIRTIO_NET_HDR_GSO_TCPV4:
1174 			skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
1175 			break;
1176 		case VIRTIO_NET_HDR_GSO_TCPV6:
1177 			skb_shinfo(skb)->gso_type = SKB_GSO_TCPV6;
1178 			break;
1179 		case VIRTIO_NET_HDR_GSO_UDP:
1180 		{
1181 			static bool warned;
1182 
1183 			if (!warned) {
1184 				warned = true;
1185 				netdev_warn(tun->dev,
1186 					    "%s: using disabled UFO feature; please fix this program\n",
1187 					    current->comm);
1188 			}
1189 			skb_shinfo(skb)->gso_type = SKB_GSO_UDP;
1190 			if (skb->protocol == htons(ETH_P_IPV6))
1191 				ipv6_proxy_select_ident(skb);
1192 			break;
1193 		}
1194 		default:
1195 			tun->dev->stats.rx_frame_errors++;
1196 			kfree_skb(skb);
1197 			return -EINVAL;
1198 		}
1199 
1200 		if (gso.gso_type & VIRTIO_NET_HDR_GSO_ECN)
1201 			skb_shinfo(skb)->gso_type |= SKB_GSO_TCP_ECN;
1202 
1203 		skb_shinfo(skb)->gso_size = tun16_to_cpu(tun, gso.gso_size);
1204 		if (skb_shinfo(skb)->gso_size == 0) {
1205 			tun->dev->stats.rx_frame_errors++;
1206 			kfree_skb(skb);
1207 			return -EINVAL;
1208 		}
1209 
1210 		/* Header must be checked, and gso_segs computed. */
1211 		skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
1212 		skb_shinfo(skb)->gso_segs = 0;
1213 	}
1214 
1215 	/* copy skb_ubuf_info for callback when skb has no error */
1216 	if (zerocopy) {
1217 		skb_shinfo(skb)->destructor_arg = msg_control;
1218 		skb_shinfo(skb)->tx_flags |= SKBTX_DEV_ZEROCOPY;
1219 		skb_shinfo(skb)->tx_flags |= SKBTX_SHARED_FRAG;
1220 	}
1221 
1222 	skb_probe_transport_header(skb, 0);
1223 
1224 	rxhash = skb_get_hash(skb);
1225 	netif_rx_ni(skb);
1226 
1227 	tun->dev->stats.rx_packets++;
1228 	tun->dev->stats.rx_bytes += len;
1229 
1230 	tun_flow_update(tun, rxhash, tfile);
1231 	return total_len;
1232 }
1233 
1234 static ssize_t tun_chr_write_iter(struct kiocb *iocb, struct iov_iter *from)
1235 {
1236 	struct file *file = iocb->ki_filp;
1237 	struct tun_struct *tun = tun_get(file);
1238 	struct tun_file *tfile = file->private_data;
1239 	ssize_t result;
1240 
1241 	if (!tun)
1242 		return -EBADFD;
1243 
1244 	result = tun_get_user(tun, tfile, NULL, from, file->f_flags & O_NONBLOCK);
1245 
1246 	tun_put(tun);
1247 	return result;
1248 }
1249 
1250 /* Put packet to the user space buffer */
1251 static ssize_t tun_put_user(struct tun_struct *tun,
1252 			    struct tun_file *tfile,
1253 			    struct sk_buff *skb,
1254 			    struct iov_iter *iter)
1255 {
1256 	struct tun_pi pi = { 0, skb->protocol };
1257 	ssize_t total;
1258 	int vlan_offset = 0;
1259 	int vlan_hlen = 0;
1260 	int vnet_hdr_sz = 0;
1261 
1262 	if (vlan_tx_tag_present(skb))
1263 		vlan_hlen = VLAN_HLEN;
1264 
1265 	if (tun->flags & IFF_VNET_HDR)
1266 		vnet_hdr_sz = tun->vnet_hdr_sz;
1267 
1268 	total = skb->len + vlan_hlen + vnet_hdr_sz;
1269 
1270 	if (!(tun->flags & IFF_NO_PI)) {
1271 		if (iov_iter_count(iter) < sizeof(pi))
1272 			return -EINVAL;
1273 
1274 		total += sizeof(pi);
1275 		if (iov_iter_count(iter) < total) {
1276 			/* Packet will be striped */
1277 			pi.flags |= TUN_PKT_STRIP;
1278 		}
1279 
1280 		if (copy_to_iter(&pi, sizeof(pi), iter) != sizeof(pi))
1281 			return -EFAULT;
1282 	}
1283 
1284 	if (vnet_hdr_sz) {
1285 		struct virtio_net_hdr gso = { 0 }; /* no info leak */
1286 		if (iov_iter_count(iter) < vnet_hdr_sz)
1287 			return -EINVAL;
1288 
1289 		if (skb_is_gso(skb)) {
1290 			struct skb_shared_info *sinfo = skb_shinfo(skb);
1291 
1292 			/* This is a hint as to how much should be linear. */
1293 			gso.hdr_len = cpu_to_tun16(tun, skb_headlen(skb));
1294 			gso.gso_size = cpu_to_tun16(tun, sinfo->gso_size);
1295 			if (sinfo->gso_type & SKB_GSO_TCPV4)
1296 				gso.gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
1297 			else if (sinfo->gso_type & SKB_GSO_TCPV6)
1298 				gso.gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
1299 			else {
1300 				pr_err("unexpected GSO type: "
1301 				       "0x%x, gso_size %d, hdr_len %d\n",
1302 				       sinfo->gso_type, tun16_to_cpu(tun, gso.gso_size),
1303 				       tun16_to_cpu(tun, gso.hdr_len));
1304 				print_hex_dump(KERN_ERR, "tun: ",
1305 					       DUMP_PREFIX_NONE,
1306 					       16, 1, skb->head,
1307 					       min((int)tun16_to_cpu(tun, gso.hdr_len), 64), true);
1308 				WARN_ON_ONCE(1);
1309 				return -EINVAL;
1310 			}
1311 			if (sinfo->gso_type & SKB_GSO_TCP_ECN)
1312 				gso.gso_type |= VIRTIO_NET_HDR_GSO_ECN;
1313 		} else
1314 			gso.gso_type = VIRTIO_NET_HDR_GSO_NONE;
1315 
1316 		if (skb->ip_summed == CHECKSUM_PARTIAL) {
1317 			gso.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
1318 			gso.csum_start = cpu_to_tun16(tun, skb_checksum_start_offset(skb) +
1319 						      vlan_hlen);
1320 			gso.csum_offset = cpu_to_tun16(tun, skb->csum_offset);
1321 		} else if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
1322 			gso.flags = VIRTIO_NET_HDR_F_DATA_VALID;
1323 		} /* else everything is zero */
1324 
1325 		if (copy_to_iter(&gso, sizeof(gso), iter) != sizeof(gso))
1326 			return -EFAULT;
1327 
1328 		iov_iter_advance(iter, vnet_hdr_sz - sizeof(gso));
1329 	}
1330 
1331 	if (vlan_hlen) {
1332 		int ret;
1333 		struct {
1334 			__be16 h_vlan_proto;
1335 			__be16 h_vlan_TCI;
1336 		} veth;
1337 
1338 		veth.h_vlan_proto = skb->vlan_proto;
1339 		veth.h_vlan_TCI = htons(vlan_tx_tag_get(skb));
1340 
1341 		vlan_offset = offsetof(struct vlan_ethhdr, h_vlan_proto);
1342 
1343 		ret = skb_copy_datagram_iter(skb, 0, iter, vlan_offset);
1344 		if (ret || !iov_iter_count(iter))
1345 			goto done;
1346 
1347 		ret = copy_to_iter(&veth, sizeof(veth), iter);
1348 		if (ret != sizeof(veth) || !iov_iter_count(iter))
1349 			goto done;
1350 	}
1351 
1352 	skb_copy_datagram_iter(skb, vlan_offset, iter, skb->len - vlan_offset);
1353 
1354 done:
1355 	tun->dev->stats.tx_packets++;
1356 	tun->dev->stats.tx_bytes += skb->len + vlan_hlen;
1357 
1358 	return total;
1359 }
1360 
1361 static ssize_t tun_do_read(struct tun_struct *tun, struct tun_file *tfile,
1362 			   struct iov_iter *to,
1363 			   int noblock)
1364 {
1365 	struct sk_buff *skb;
1366 	ssize_t ret;
1367 	int peeked, err, off = 0;
1368 
1369 	tun_debug(KERN_INFO, tun, "tun_do_read\n");
1370 
1371 	if (!iov_iter_count(to))
1372 		return 0;
1373 
1374 	if (tun->dev->reg_state != NETREG_REGISTERED)
1375 		return -EIO;
1376 
1377 	/* Read frames from queue */
1378 	skb = __skb_recv_datagram(tfile->socket.sk, noblock ? MSG_DONTWAIT : 0,
1379 				  &peeked, &off, &err);
1380 	if (!skb)
1381 		return 0;
1382 
1383 	ret = tun_put_user(tun, tfile, skb, to);
1384 	if (unlikely(ret < 0))
1385 		kfree_skb(skb);
1386 	else
1387 		consume_skb(skb);
1388 
1389 	return ret;
1390 }
1391 
1392 static ssize_t tun_chr_read_iter(struct kiocb *iocb, struct iov_iter *to)
1393 {
1394 	struct file *file = iocb->ki_filp;
1395 	struct tun_file *tfile = file->private_data;
1396 	struct tun_struct *tun = __tun_get(tfile);
1397 	ssize_t len = iov_iter_count(to), ret;
1398 
1399 	if (!tun)
1400 		return -EBADFD;
1401 	ret = tun_do_read(tun, tfile, to, file->f_flags & O_NONBLOCK);
1402 	ret = min_t(ssize_t, ret, len);
1403 	if (ret > 0)
1404 		iocb->ki_pos = ret;
1405 	tun_put(tun);
1406 	return ret;
1407 }
1408 
1409 static void tun_free_netdev(struct net_device *dev)
1410 {
1411 	struct tun_struct *tun = netdev_priv(dev);
1412 
1413 	BUG_ON(!(list_empty(&tun->disabled)));
1414 	tun_flow_uninit(tun);
1415 	security_tun_dev_free_security(tun->security);
1416 	free_netdev(dev);
1417 }
1418 
1419 static void tun_setup(struct net_device *dev)
1420 {
1421 	struct tun_struct *tun = netdev_priv(dev);
1422 
1423 	tun->owner = INVALID_UID;
1424 	tun->group = INVALID_GID;
1425 
1426 	dev->ethtool_ops = &tun_ethtool_ops;
1427 	dev->destructor = tun_free_netdev;
1428 }
1429 
1430 /* Trivial set of netlink ops to allow deleting tun or tap
1431  * device with netlink.
1432  */
1433 static int tun_validate(struct nlattr *tb[], struct nlattr *data[])
1434 {
1435 	return -EINVAL;
1436 }
1437 
1438 static struct rtnl_link_ops tun_link_ops __read_mostly = {
1439 	.kind		= DRV_NAME,
1440 	.priv_size	= sizeof(struct tun_struct),
1441 	.setup		= tun_setup,
1442 	.validate	= tun_validate,
1443 };
1444 
1445 static void tun_sock_write_space(struct sock *sk)
1446 {
1447 	struct tun_file *tfile;
1448 	wait_queue_head_t *wqueue;
1449 
1450 	if (!sock_writeable(sk))
1451 		return;
1452 
1453 	if (!test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags))
1454 		return;
1455 
1456 	wqueue = sk_sleep(sk);
1457 	if (wqueue && waitqueue_active(wqueue))
1458 		wake_up_interruptible_sync_poll(wqueue, POLLOUT |
1459 						POLLWRNORM | POLLWRBAND);
1460 
1461 	tfile = container_of(sk, struct tun_file, sk);
1462 	kill_fasync(&tfile->fasync, SIGIO, POLL_OUT);
1463 }
1464 
1465 static int tun_sendmsg(struct kiocb *iocb, struct socket *sock,
1466 		       struct msghdr *m, size_t total_len)
1467 {
1468 	int ret;
1469 	struct tun_file *tfile = container_of(sock, struct tun_file, socket);
1470 	struct tun_struct *tun = __tun_get(tfile);
1471 
1472 	if (!tun)
1473 		return -EBADFD;
1474 
1475 	ret = tun_get_user(tun, tfile, m->msg_control, &m->msg_iter,
1476 			   m->msg_flags & MSG_DONTWAIT);
1477 	tun_put(tun);
1478 	return ret;
1479 }
1480 
1481 static int tun_recvmsg(struct kiocb *iocb, struct socket *sock,
1482 		       struct msghdr *m, size_t total_len,
1483 		       int flags)
1484 {
1485 	struct tun_file *tfile = container_of(sock, struct tun_file, socket);
1486 	struct tun_struct *tun = __tun_get(tfile);
1487 	int ret;
1488 
1489 	if (!tun)
1490 		return -EBADFD;
1491 
1492 	if (flags & ~(MSG_DONTWAIT|MSG_TRUNC|MSG_ERRQUEUE)) {
1493 		ret = -EINVAL;
1494 		goto out;
1495 	}
1496 	if (flags & MSG_ERRQUEUE) {
1497 		ret = sock_recv_errqueue(sock->sk, m, total_len,
1498 					 SOL_PACKET, TUN_TX_TIMESTAMP);
1499 		goto out;
1500 	}
1501 	ret = tun_do_read(tun, tfile, &m->msg_iter, flags & MSG_DONTWAIT);
1502 	if (ret > total_len) {
1503 		m->msg_flags |= MSG_TRUNC;
1504 		ret = flags & MSG_TRUNC ? ret : total_len;
1505 	}
1506 out:
1507 	tun_put(tun);
1508 	return ret;
1509 }
1510 
1511 static int tun_release(struct socket *sock)
1512 {
1513 	if (sock->sk)
1514 		sock_put(sock->sk);
1515 	return 0;
1516 }
1517 
1518 /* Ops structure to mimic raw sockets with tun */
1519 static const struct proto_ops tun_socket_ops = {
1520 	.sendmsg = tun_sendmsg,
1521 	.recvmsg = tun_recvmsg,
1522 	.release = tun_release,
1523 };
1524 
1525 static struct proto tun_proto = {
1526 	.name		= "tun",
1527 	.owner		= THIS_MODULE,
1528 	.obj_size	= sizeof(struct tun_file),
1529 };
1530 
1531 static int tun_flags(struct tun_struct *tun)
1532 {
1533 	return tun->flags & (TUN_FEATURES | IFF_PERSIST | IFF_TUN | IFF_TAP);
1534 }
1535 
1536 static ssize_t tun_show_flags(struct device *dev, struct device_attribute *attr,
1537 			      char *buf)
1538 {
1539 	struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1540 	return sprintf(buf, "0x%x\n", tun_flags(tun));
1541 }
1542 
1543 static ssize_t tun_show_owner(struct device *dev, struct device_attribute *attr,
1544 			      char *buf)
1545 {
1546 	struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1547 	return uid_valid(tun->owner)?
1548 		sprintf(buf, "%u\n",
1549 			from_kuid_munged(current_user_ns(), tun->owner)):
1550 		sprintf(buf, "-1\n");
1551 }
1552 
1553 static ssize_t tun_show_group(struct device *dev, struct device_attribute *attr,
1554 			      char *buf)
1555 {
1556 	struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1557 	return gid_valid(tun->group) ?
1558 		sprintf(buf, "%u\n",
1559 			from_kgid_munged(current_user_ns(), tun->group)):
1560 		sprintf(buf, "-1\n");
1561 }
1562 
1563 static DEVICE_ATTR(tun_flags, 0444, tun_show_flags, NULL);
1564 static DEVICE_ATTR(owner, 0444, tun_show_owner, NULL);
1565 static DEVICE_ATTR(group, 0444, tun_show_group, NULL);
1566 
1567 static int tun_set_iff(struct net *net, struct file *file, struct ifreq *ifr)
1568 {
1569 	struct tun_struct *tun;
1570 	struct tun_file *tfile = file->private_data;
1571 	struct net_device *dev;
1572 	int err;
1573 
1574 	if (tfile->detached)
1575 		return -EINVAL;
1576 
1577 	dev = __dev_get_by_name(net, ifr->ifr_name);
1578 	if (dev) {
1579 		if (ifr->ifr_flags & IFF_TUN_EXCL)
1580 			return -EBUSY;
1581 		if ((ifr->ifr_flags & IFF_TUN) && dev->netdev_ops == &tun_netdev_ops)
1582 			tun = netdev_priv(dev);
1583 		else if ((ifr->ifr_flags & IFF_TAP) && dev->netdev_ops == &tap_netdev_ops)
1584 			tun = netdev_priv(dev);
1585 		else
1586 			return -EINVAL;
1587 
1588 		if (!!(ifr->ifr_flags & IFF_MULTI_QUEUE) !=
1589 		    !!(tun->flags & IFF_MULTI_QUEUE))
1590 			return -EINVAL;
1591 
1592 		if (tun_not_capable(tun))
1593 			return -EPERM;
1594 		err = security_tun_dev_open(tun->security);
1595 		if (err < 0)
1596 			return err;
1597 
1598 		err = tun_attach(tun, file, ifr->ifr_flags & IFF_NOFILTER);
1599 		if (err < 0)
1600 			return err;
1601 
1602 		if (tun->flags & IFF_MULTI_QUEUE &&
1603 		    (tun->numqueues + tun->numdisabled > 1)) {
1604 			/* One or more queue has already been attached, no need
1605 			 * to initialize the device again.
1606 			 */
1607 			return 0;
1608 		}
1609 	}
1610 	else {
1611 		char *name;
1612 		unsigned long flags = 0;
1613 		int queues = ifr->ifr_flags & IFF_MULTI_QUEUE ?
1614 			     MAX_TAP_QUEUES : 1;
1615 
1616 		if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
1617 			return -EPERM;
1618 		err = security_tun_dev_create();
1619 		if (err < 0)
1620 			return err;
1621 
1622 		/* Set dev type */
1623 		if (ifr->ifr_flags & IFF_TUN) {
1624 			/* TUN device */
1625 			flags |= IFF_TUN;
1626 			name = "tun%d";
1627 		} else if (ifr->ifr_flags & IFF_TAP) {
1628 			/* TAP device */
1629 			flags |= IFF_TAP;
1630 			name = "tap%d";
1631 		} else
1632 			return -EINVAL;
1633 
1634 		if (*ifr->ifr_name)
1635 			name = ifr->ifr_name;
1636 
1637 		dev = alloc_netdev_mqs(sizeof(struct tun_struct), name,
1638 				       NET_NAME_UNKNOWN, tun_setup, queues,
1639 				       queues);
1640 
1641 		if (!dev)
1642 			return -ENOMEM;
1643 
1644 		dev_net_set(dev, net);
1645 		dev->rtnl_link_ops = &tun_link_ops;
1646 		dev->ifindex = tfile->ifindex;
1647 
1648 		tun = netdev_priv(dev);
1649 		tun->dev = dev;
1650 		tun->flags = flags;
1651 		tun->txflt.count = 0;
1652 		tun->vnet_hdr_sz = sizeof(struct virtio_net_hdr);
1653 
1654 		tun->filter_attached = false;
1655 		tun->sndbuf = tfile->socket.sk->sk_sndbuf;
1656 
1657 		spin_lock_init(&tun->lock);
1658 
1659 		err = security_tun_dev_alloc_security(&tun->security);
1660 		if (err < 0)
1661 			goto err_free_dev;
1662 
1663 		tun_net_init(dev);
1664 		tun_flow_init(tun);
1665 
1666 		dev->hw_features = NETIF_F_SG | NETIF_F_FRAGLIST |
1667 				   TUN_USER_FEATURES | NETIF_F_HW_VLAN_CTAG_TX |
1668 				   NETIF_F_HW_VLAN_STAG_TX;
1669 		dev->features = dev->hw_features;
1670 		dev->vlan_features = dev->features &
1671 				     ~(NETIF_F_HW_VLAN_CTAG_TX |
1672 				       NETIF_F_HW_VLAN_STAG_TX);
1673 
1674 		INIT_LIST_HEAD(&tun->disabled);
1675 		err = tun_attach(tun, file, false);
1676 		if (err < 0)
1677 			goto err_free_flow;
1678 
1679 		err = register_netdevice(tun->dev);
1680 		if (err < 0)
1681 			goto err_detach;
1682 
1683 		if (device_create_file(&tun->dev->dev, &dev_attr_tun_flags) ||
1684 		    device_create_file(&tun->dev->dev, &dev_attr_owner) ||
1685 		    device_create_file(&tun->dev->dev, &dev_attr_group))
1686 			pr_err("Failed to create tun sysfs files\n");
1687 	}
1688 
1689 	netif_carrier_on(tun->dev);
1690 
1691 	tun_debug(KERN_INFO, tun, "tun_set_iff\n");
1692 
1693 	tun->flags = (tun->flags & ~TUN_FEATURES) |
1694 		(ifr->ifr_flags & TUN_FEATURES);
1695 
1696 	/* Make sure persistent devices do not get stuck in
1697 	 * xoff state.
1698 	 */
1699 	if (netif_running(tun->dev))
1700 		netif_tx_wake_all_queues(tun->dev);
1701 
1702 	strcpy(ifr->ifr_name, tun->dev->name);
1703 	return 0;
1704 
1705 err_detach:
1706 	tun_detach_all(dev);
1707 err_free_flow:
1708 	tun_flow_uninit(tun);
1709 	security_tun_dev_free_security(tun->security);
1710 err_free_dev:
1711 	free_netdev(dev);
1712 	return err;
1713 }
1714 
1715 static void tun_get_iff(struct net *net, struct tun_struct *tun,
1716 		       struct ifreq *ifr)
1717 {
1718 	tun_debug(KERN_INFO, tun, "tun_get_iff\n");
1719 
1720 	strcpy(ifr->ifr_name, tun->dev->name);
1721 
1722 	ifr->ifr_flags = tun_flags(tun);
1723 
1724 }
1725 
1726 /* This is like a cut-down ethtool ops, except done via tun fd so no
1727  * privs required. */
1728 static int set_offload(struct tun_struct *tun, unsigned long arg)
1729 {
1730 	netdev_features_t features = 0;
1731 
1732 	if (arg & TUN_F_CSUM) {
1733 		features |= NETIF_F_HW_CSUM;
1734 		arg &= ~TUN_F_CSUM;
1735 
1736 		if (arg & (TUN_F_TSO4|TUN_F_TSO6)) {
1737 			if (arg & TUN_F_TSO_ECN) {
1738 				features |= NETIF_F_TSO_ECN;
1739 				arg &= ~TUN_F_TSO_ECN;
1740 			}
1741 			if (arg & TUN_F_TSO4)
1742 				features |= NETIF_F_TSO;
1743 			if (arg & TUN_F_TSO6)
1744 				features |= NETIF_F_TSO6;
1745 			arg &= ~(TUN_F_TSO4|TUN_F_TSO6);
1746 		}
1747 	}
1748 
1749 	/* This gives the user a way to test for new features in future by
1750 	 * trying to set them. */
1751 	if (arg)
1752 		return -EINVAL;
1753 
1754 	tun->set_features = features;
1755 	netdev_update_features(tun->dev);
1756 
1757 	return 0;
1758 }
1759 
1760 static void tun_detach_filter(struct tun_struct *tun, int n)
1761 {
1762 	int i;
1763 	struct tun_file *tfile;
1764 
1765 	for (i = 0; i < n; i++) {
1766 		tfile = rtnl_dereference(tun->tfiles[i]);
1767 		sk_detach_filter(tfile->socket.sk);
1768 	}
1769 
1770 	tun->filter_attached = false;
1771 }
1772 
1773 static int tun_attach_filter(struct tun_struct *tun)
1774 {
1775 	int i, ret = 0;
1776 	struct tun_file *tfile;
1777 
1778 	for (i = 0; i < tun->numqueues; i++) {
1779 		tfile = rtnl_dereference(tun->tfiles[i]);
1780 		ret = sk_attach_filter(&tun->fprog, tfile->socket.sk);
1781 		if (ret) {
1782 			tun_detach_filter(tun, i);
1783 			return ret;
1784 		}
1785 	}
1786 
1787 	tun->filter_attached = true;
1788 	return ret;
1789 }
1790 
1791 static void tun_set_sndbuf(struct tun_struct *tun)
1792 {
1793 	struct tun_file *tfile;
1794 	int i;
1795 
1796 	for (i = 0; i < tun->numqueues; i++) {
1797 		tfile = rtnl_dereference(tun->tfiles[i]);
1798 		tfile->socket.sk->sk_sndbuf = tun->sndbuf;
1799 	}
1800 }
1801 
1802 static int tun_set_queue(struct file *file, struct ifreq *ifr)
1803 {
1804 	struct tun_file *tfile = file->private_data;
1805 	struct tun_struct *tun;
1806 	int ret = 0;
1807 
1808 	rtnl_lock();
1809 
1810 	if (ifr->ifr_flags & IFF_ATTACH_QUEUE) {
1811 		tun = tfile->detached;
1812 		if (!tun) {
1813 			ret = -EINVAL;
1814 			goto unlock;
1815 		}
1816 		ret = security_tun_dev_attach_queue(tun->security);
1817 		if (ret < 0)
1818 			goto unlock;
1819 		ret = tun_attach(tun, file, false);
1820 	} else if (ifr->ifr_flags & IFF_DETACH_QUEUE) {
1821 		tun = rtnl_dereference(tfile->tun);
1822 		if (!tun || !(tun->flags & IFF_MULTI_QUEUE) || tfile->detached)
1823 			ret = -EINVAL;
1824 		else
1825 			__tun_detach(tfile, false);
1826 	} else
1827 		ret = -EINVAL;
1828 
1829 unlock:
1830 	rtnl_unlock();
1831 	return ret;
1832 }
1833 
1834 static long __tun_chr_ioctl(struct file *file, unsigned int cmd,
1835 			    unsigned long arg, int ifreq_len)
1836 {
1837 	struct tun_file *tfile = file->private_data;
1838 	struct tun_struct *tun;
1839 	void __user* argp = (void __user*)arg;
1840 	struct ifreq ifr;
1841 	kuid_t owner;
1842 	kgid_t group;
1843 	int sndbuf;
1844 	int vnet_hdr_sz;
1845 	unsigned int ifindex;
1846 	int ret;
1847 
1848 	if (cmd == TUNSETIFF || cmd == TUNSETQUEUE || _IOC_TYPE(cmd) == 0x89) {
1849 		if (copy_from_user(&ifr, argp, ifreq_len))
1850 			return -EFAULT;
1851 	} else {
1852 		memset(&ifr, 0, sizeof(ifr));
1853 	}
1854 	if (cmd == TUNGETFEATURES) {
1855 		/* Currently this just means: "what IFF flags are valid?".
1856 		 * This is needed because we never checked for invalid flags on
1857 		 * TUNSETIFF.
1858 		 */
1859 		return put_user(IFF_TUN | IFF_TAP | TUN_FEATURES,
1860 				(unsigned int __user*)argp);
1861 	} else if (cmd == TUNSETQUEUE)
1862 		return tun_set_queue(file, &ifr);
1863 
1864 	ret = 0;
1865 	rtnl_lock();
1866 
1867 	tun = __tun_get(tfile);
1868 	if (cmd == TUNSETIFF && !tun) {
1869 		ifr.ifr_name[IFNAMSIZ-1] = '\0';
1870 
1871 		ret = tun_set_iff(tfile->net, file, &ifr);
1872 
1873 		if (ret)
1874 			goto unlock;
1875 
1876 		if (copy_to_user(argp, &ifr, ifreq_len))
1877 			ret = -EFAULT;
1878 		goto unlock;
1879 	}
1880 	if (cmd == TUNSETIFINDEX) {
1881 		ret = -EPERM;
1882 		if (tun)
1883 			goto unlock;
1884 
1885 		ret = -EFAULT;
1886 		if (copy_from_user(&ifindex, argp, sizeof(ifindex)))
1887 			goto unlock;
1888 
1889 		ret = 0;
1890 		tfile->ifindex = ifindex;
1891 		goto unlock;
1892 	}
1893 
1894 	ret = -EBADFD;
1895 	if (!tun)
1896 		goto unlock;
1897 
1898 	tun_debug(KERN_INFO, tun, "tun_chr_ioctl cmd %u\n", cmd);
1899 
1900 	ret = 0;
1901 	switch (cmd) {
1902 	case TUNGETIFF:
1903 		tun_get_iff(current->nsproxy->net_ns, tun, &ifr);
1904 
1905 		if (tfile->detached)
1906 			ifr.ifr_flags |= IFF_DETACH_QUEUE;
1907 		if (!tfile->socket.sk->sk_filter)
1908 			ifr.ifr_flags |= IFF_NOFILTER;
1909 
1910 		if (copy_to_user(argp, &ifr, ifreq_len))
1911 			ret = -EFAULT;
1912 		break;
1913 
1914 	case TUNSETNOCSUM:
1915 		/* Disable/Enable checksum */
1916 
1917 		/* [unimplemented] */
1918 		tun_debug(KERN_INFO, tun, "ignored: set checksum %s\n",
1919 			  arg ? "disabled" : "enabled");
1920 		break;
1921 
1922 	case TUNSETPERSIST:
1923 		/* Disable/Enable persist mode. Keep an extra reference to the
1924 		 * module to prevent the module being unprobed.
1925 		 */
1926 		if (arg && !(tun->flags & IFF_PERSIST)) {
1927 			tun->flags |= IFF_PERSIST;
1928 			__module_get(THIS_MODULE);
1929 		}
1930 		if (!arg && (tun->flags & IFF_PERSIST)) {
1931 			tun->flags &= ~IFF_PERSIST;
1932 			module_put(THIS_MODULE);
1933 		}
1934 
1935 		tun_debug(KERN_INFO, tun, "persist %s\n",
1936 			  arg ? "enabled" : "disabled");
1937 		break;
1938 
1939 	case TUNSETOWNER:
1940 		/* Set owner of the device */
1941 		owner = make_kuid(current_user_ns(), arg);
1942 		if (!uid_valid(owner)) {
1943 			ret = -EINVAL;
1944 			break;
1945 		}
1946 		tun->owner = owner;
1947 		tun_debug(KERN_INFO, tun, "owner set to %u\n",
1948 			  from_kuid(&init_user_ns, tun->owner));
1949 		break;
1950 
1951 	case TUNSETGROUP:
1952 		/* Set group of the device */
1953 		group = make_kgid(current_user_ns(), arg);
1954 		if (!gid_valid(group)) {
1955 			ret = -EINVAL;
1956 			break;
1957 		}
1958 		tun->group = group;
1959 		tun_debug(KERN_INFO, tun, "group set to %u\n",
1960 			  from_kgid(&init_user_ns, tun->group));
1961 		break;
1962 
1963 	case TUNSETLINK:
1964 		/* Only allow setting the type when the interface is down */
1965 		if (tun->dev->flags & IFF_UP) {
1966 			tun_debug(KERN_INFO, tun,
1967 				  "Linktype set failed because interface is up\n");
1968 			ret = -EBUSY;
1969 		} else {
1970 			tun->dev->type = (int) arg;
1971 			tun_debug(KERN_INFO, tun, "linktype set to %d\n",
1972 				  tun->dev->type);
1973 			ret = 0;
1974 		}
1975 		break;
1976 
1977 #ifdef TUN_DEBUG
1978 	case TUNSETDEBUG:
1979 		tun->debug = arg;
1980 		break;
1981 #endif
1982 	case TUNSETOFFLOAD:
1983 		ret = set_offload(tun, arg);
1984 		break;
1985 
1986 	case TUNSETTXFILTER:
1987 		/* Can be set only for TAPs */
1988 		ret = -EINVAL;
1989 		if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
1990 			break;
1991 		ret = update_filter(&tun->txflt, (void __user *)arg);
1992 		break;
1993 
1994 	case SIOCGIFHWADDR:
1995 		/* Get hw address */
1996 		memcpy(ifr.ifr_hwaddr.sa_data, tun->dev->dev_addr, ETH_ALEN);
1997 		ifr.ifr_hwaddr.sa_family = tun->dev->type;
1998 		if (copy_to_user(argp, &ifr, ifreq_len))
1999 			ret = -EFAULT;
2000 		break;
2001 
2002 	case SIOCSIFHWADDR:
2003 		/* Set hw address */
2004 		tun_debug(KERN_DEBUG, tun, "set hw address: %pM\n",
2005 			  ifr.ifr_hwaddr.sa_data);
2006 
2007 		ret = dev_set_mac_address(tun->dev, &ifr.ifr_hwaddr);
2008 		break;
2009 
2010 	case TUNGETSNDBUF:
2011 		sndbuf = tfile->socket.sk->sk_sndbuf;
2012 		if (copy_to_user(argp, &sndbuf, sizeof(sndbuf)))
2013 			ret = -EFAULT;
2014 		break;
2015 
2016 	case TUNSETSNDBUF:
2017 		if (copy_from_user(&sndbuf, argp, sizeof(sndbuf))) {
2018 			ret = -EFAULT;
2019 			break;
2020 		}
2021 
2022 		tun->sndbuf = sndbuf;
2023 		tun_set_sndbuf(tun);
2024 		break;
2025 
2026 	case TUNGETVNETHDRSZ:
2027 		vnet_hdr_sz = tun->vnet_hdr_sz;
2028 		if (copy_to_user(argp, &vnet_hdr_sz, sizeof(vnet_hdr_sz)))
2029 			ret = -EFAULT;
2030 		break;
2031 
2032 	case TUNSETVNETHDRSZ:
2033 		if (copy_from_user(&vnet_hdr_sz, argp, sizeof(vnet_hdr_sz))) {
2034 			ret = -EFAULT;
2035 			break;
2036 		}
2037 		if (vnet_hdr_sz < (int)sizeof(struct virtio_net_hdr)) {
2038 			ret = -EINVAL;
2039 			break;
2040 		}
2041 
2042 		tun->vnet_hdr_sz = vnet_hdr_sz;
2043 		break;
2044 
2045 	case TUNATTACHFILTER:
2046 		/* Can be set only for TAPs */
2047 		ret = -EINVAL;
2048 		if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
2049 			break;
2050 		ret = -EFAULT;
2051 		if (copy_from_user(&tun->fprog, argp, sizeof(tun->fprog)))
2052 			break;
2053 
2054 		ret = tun_attach_filter(tun);
2055 		break;
2056 
2057 	case TUNDETACHFILTER:
2058 		/* Can be set only for TAPs */
2059 		ret = -EINVAL;
2060 		if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
2061 			break;
2062 		ret = 0;
2063 		tun_detach_filter(tun, tun->numqueues);
2064 		break;
2065 
2066 	case TUNGETFILTER:
2067 		ret = -EINVAL;
2068 		if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
2069 			break;
2070 		ret = -EFAULT;
2071 		if (copy_to_user(argp, &tun->fprog, sizeof(tun->fprog)))
2072 			break;
2073 		ret = 0;
2074 		break;
2075 
2076 	default:
2077 		ret = -EINVAL;
2078 		break;
2079 	}
2080 
2081 unlock:
2082 	rtnl_unlock();
2083 	if (tun)
2084 		tun_put(tun);
2085 	return ret;
2086 }
2087 
2088 static long tun_chr_ioctl(struct file *file,
2089 			  unsigned int cmd, unsigned long arg)
2090 {
2091 	return __tun_chr_ioctl(file, cmd, arg, sizeof (struct ifreq));
2092 }
2093 
2094 #ifdef CONFIG_COMPAT
2095 static long tun_chr_compat_ioctl(struct file *file,
2096 			 unsigned int cmd, unsigned long arg)
2097 {
2098 	switch (cmd) {
2099 	case TUNSETIFF:
2100 	case TUNGETIFF:
2101 	case TUNSETTXFILTER:
2102 	case TUNGETSNDBUF:
2103 	case TUNSETSNDBUF:
2104 	case SIOCGIFHWADDR:
2105 	case SIOCSIFHWADDR:
2106 		arg = (unsigned long)compat_ptr(arg);
2107 		break;
2108 	default:
2109 		arg = (compat_ulong_t)arg;
2110 		break;
2111 	}
2112 
2113 	/*
2114 	 * compat_ifreq is shorter than ifreq, so we must not access beyond
2115 	 * the end of that structure. All fields that are used in this
2116 	 * driver are compatible though, we don't need to convert the
2117 	 * contents.
2118 	 */
2119 	return __tun_chr_ioctl(file, cmd, arg, sizeof(struct compat_ifreq));
2120 }
2121 #endif /* CONFIG_COMPAT */
2122 
2123 static int tun_chr_fasync(int fd, struct file *file, int on)
2124 {
2125 	struct tun_file *tfile = file->private_data;
2126 	int ret;
2127 
2128 	if ((ret = fasync_helper(fd, file, on, &tfile->fasync)) < 0)
2129 		goto out;
2130 
2131 	if (on) {
2132 		__f_setown(file, task_pid(current), PIDTYPE_PID, 0);
2133 		tfile->flags |= TUN_FASYNC;
2134 	} else
2135 		tfile->flags &= ~TUN_FASYNC;
2136 	ret = 0;
2137 out:
2138 	return ret;
2139 }
2140 
2141 static int tun_chr_open(struct inode *inode, struct file * file)
2142 {
2143 	struct tun_file *tfile;
2144 
2145 	DBG1(KERN_INFO, "tunX: tun_chr_open\n");
2146 
2147 	tfile = (struct tun_file *)sk_alloc(&init_net, AF_UNSPEC, GFP_KERNEL,
2148 					    &tun_proto);
2149 	if (!tfile)
2150 		return -ENOMEM;
2151 	RCU_INIT_POINTER(tfile->tun, NULL);
2152 	tfile->net = get_net(current->nsproxy->net_ns);
2153 	tfile->flags = 0;
2154 	tfile->ifindex = 0;
2155 
2156 	init_waitqueue_head(&tfile->wq.wait);
2157 	RCU_INIT_POINTER(tfile->socket.wq, &tfile->wq);
2158 
2159 	tfile->socket.file = file;
2160 	tfile->socket.ops = &tun_socket_ops;
2161 
2162 	sock_init_data(&tfile->socket, &tfile->sk);
2163 	sk_change_net(&tfile->sk, tfile->net);
2164 
2165 	tfile->sk.sk_write_space = tun_sock_write_space;
2166 	tfile->sk.sk_sndbuf = INT_MAX;
2167 
2168 	file->private_data = tfile;
2169 	set_bit(SOCK_EXTERNALLY_ALLOCATED, &tfile->socket.flags);
2170 	INIT_LIST_HEAD(&tfile->next);
2171 
2172 	sock_set_flag(&tfile->sk, SOCK_ZEROCOPY);
2173 
2174 	return 0;
2175 }
2176 
2177 static int tun_chr_close(struct inode *inode, struct file *file)
2178 {
2179 	struct tun_file *tfile = file->private_data;
2180 	struct net *net = tfile->net;
2181 
2182 	tun_detach(tfile, true);
2183 	put_net(net);
2184 
2185 	return 0;
2186 }
2187 
2188 #ifdef CONFIG_PROC_FS
2189 static void tun_chr_show_fdinfo(struct seq_file *m, struct file *f)
2190 {
2191 	struct tun_struct *tun;
2192 	struct ifreq ifr;
2193 
2194 	memset(&ifr, 0, sizeof(ifr));
2195 
2196 	rtnl_lock();
2197 	tun = tun_get(f);
2198 	if (tun)
2199 		tun_get_iff(current->nsproxy->net_ns, tun, &ifr);
2200 	rtnl_unlock();
2201 
2202 	if (tun)
2203 		tun_put(tun);
2204 
2205 	seq_printf(m, "iff:\t%s\n", ifr.ifr_name);
2206 }
2207 #endif
2208 
2209 static const struct file_operations tun_fops = {
2210 	.owner	= THIS_MODULE,
2211 	.llseek = no_llseek,
2212 	.read  = new_sync_read,
2213 	.write = new_sync_write,
2214 	.read_iter  = tun_chr_read_iter,
2215 	.write_iter = tun_chr_write_iter,
2216 	.poll	= tun_chr_poll,
2217 	.unlocked_ioctl	= tun_chr_ioctl,
2218 #ifdef CONFIG_COMPAT
2219 	.compat_ioctl = tun_chr_compat_ioctl,
2220 #endif
2221 	.open	= tun_chr_open,
2222 	.release = tun_chr_close,
2223 	.fasync = tun_chr_fasync,
2224 #ifdef CONFIG_PROC_FS
2225 	.show_fdinfo = tun_chr_show_fdinfo,
2226 #endif
2227 };
2228 
2229 static struct miscdevice tun_miscdev = {
2230 	.minor = TUN_MINOR,
2231 	.name = "tun",
2232 	.nodename = "net/tun",
2233 	.fops = &tun_fops,
2234 };
2235 
2236 /* ethtool interface */
2237 
2238 static int tun_get_settings(struct net_device *dev, struct ethtool_cmd *cmd)
2239 {
2240 	cmd->supported		= 0;
2241 	cmd->advertising	= 0;
2242 	ethtool_cmd_speed_set(cmd, SPEED_10);
2243 	cmd->duplex		= DUPLEX_FULL;
2244 	cmd->port		= PORT_TP;
2245 	cmd->phy_address	= 0;
2246 	cmd->transceiver	= XCVR_INTERNAL;
2247 	cmd->autoneg		= AUTONEG_DISABLE;
2248 	cmd->maxtxpkt		= 0;
2249 	cmd->maxrxpkt		= 0;
2250 	return 0;
2251 }
2252 
2253 static void tun_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
2254 {
2255 	struct tun_struct *tun = netdev_priv(dev);
2256 
2257 	strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
2258 	strlcpy(info->version, DRV_VERSION, sizeof(info->version));
2259 
2260 	switch (tun->flags & TUN_TYPE_MASK) {
2261 	case IFF_TUN:
2262 		strlcpy(info->bus_info, "tun", sizeof(info->bus_info));
2263 		break;
2264 	case IFF_TAP:
2265 		strlcpy(info->bus_info, "tap", sizeof(info->bus_info));
2266 		break;
2267 	}
2268 }
2269 
2270 static u32 tun_get_msglevel(struct net_device *dev)
2271 {
2272 #ifdef TUN_DEBUG
2273 	struct tun_struct *tun = netdev_priv(dev);
2274 	return tun->debug;
2275 #else
2276 	return -EOPNOTSUPP;
2277 #endif
2278 }
2279 
2280 static void tun_set_msglevel(struct net_device *dev, u32 value)
2281 {
2282 #ifdef TUN_DEBUG
2283 	struct tun_struct *tun = netdev_priv(dev);
2284 	tun->debug = value;
2285 #endif
2286 }
2287 
2288 static const struct ethtool_ops tun_ethtool_ops = {
2289 	.get_settings	= tun_get_settings,
2290 	.get_drvinfo	= tun_get_drvinfo,
2291 	.get_msglevel	= tun_get_msglevel,
2292 	.set_msglevel	= tun_set_msglevel,
2293 	.get_link	= ethtool_op_get_link,
2294 	.get_ts_info	= ethtool_op_get_ts_info,
2295 };
2296 
2297 
2298 static int __init tun_init(void)
2299 {
2300 	int ret = 0;
2301 
2302 	pr_info("%s, %s\n", DRV_DESCRIPTION, DRV_VERSION);
2303 	pr_info("%s\n", DRV_COPYRIGHT);
2304 
2305 	ret = rtnl_link_register(&tun_link_ops);
2306 	if (ret) {
2307 		pr_err("Can't register link_ops\n");
2308 		goto err_linkops;
2309 	}
2310 
2311 	ret = misc_register(&tun_miscdev);
2312 	if (ret) {
2313 		pr_err("Can't register misc device %d\n", TUN_MINOR);
2314 		goto err_misc;
2315 	}
2316 	return  0;
2317 err_misc:
2318 	rtnl_link_unregister(&tun_link_ops);
2319 err_linkops:
2320 	return ret;
2321 }
2322 
2323 static void tun_cleanup(void)
2324 {
2325 	misc_deregister(&tun_miscdev);
2326 	rtnl_link_unregister(&tun_link_ops);
2327 }
2328 
2329 /* Get an underlying socket object from tun file.  Returns error unless file is
2330  * attached to a device.  The returned object works like a packet socket, it
2331  * can be used for sock_sendmsg/sock_recvmsg.  The caller is responsible for
2332  * holding a reference to the file for as long as the socket is in use. */
2333 struct socket *tun_get_socket(struct file *file)
2334 {
2335 	struct tun_file *tfile;
2336 	if (file->f_op != &tun_fops)
2337 		return ERR_PTR(-EINVAL);
2338 	tfile = file->private_data;
2339 	if (!tfile)
2340 		return ERR_PTR(-EBADFD);
2341 	return &tfile->socket;
2342 }
2343 EXPORT_SYMBOL_GPL(tun_get_socket);
2344 
2345 module_init(tun_init);
2346 module_exit(tun_cleanup);
2347 MODULE_DESCRIPTION(DRV_DESCRIPTION);
2348 MODULE_AUTHOR(DRV_COPYRIGHT);
2349 MODULE_LICENSE("GPL");
2350 MODULE_ALIAS_MISCDEV(TUN_MINOR);
2351 MODULE_ALIAS("devname:net/tun");
2352