xref: /linux/drivers/net/tun.c (revision 4c62e9764ab403d42f9b8871b1241fe7812f19d4)
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/crc32.h>
64 #include <linux/nsproxy.h>
65 #include <linux/virtio_net.h>
66 #include <linux/rcupdate.h>
67 #include <net/net_namespace.h>
68 #include <net/netns/generic.h>
69 #include <net/rtnetlink.h>
70 #include <net/sock.h>
71 
72 #include <asm/uaccess.h>
73 
74 /* Uncomment to enable debugging */
75 /* #define TUN_DEBUG 1 */
76 
77 #ifdef TUN_DEBUG
78 static int debug;
79 
80 #define tun_debug(level, tun, fmt, args...)			\
81 do {								\
82 	if (tun->debug)						\
83 		netdev_printk(level, tun->dev, fmt, ##args);	\
84 } while (0)
85 #define DBG1(level, fmt, args...)				\
86 do {								\
87 	if (debug == 2)						\
88 		printk(level fmt, ##args);			\
89 } while (0)
90 #else
91 #define tun_debug(level, tun, fmt, args...)			\
92 do {								\
93 	if (0)							\
94 		netdev_printk(level, tun->dev, fmt, ##args);	\
95 } while (0)
96 #define DBG1(level, fmt, args...)				\
97 do {								\
98 	if (0)							\
99 		printk(level fmt, ##args);			\
100 } while (0)
101 #endif
102 
103 #define GOODCOPY_LEN 128
104 
105 #define FLT_EXACT_COUNT 8
106 struct tap_filter {
107 	unsigned int    count;    /* Number of addrs. Zero means disabled */
108 	u32             mask[2];  /* Mask of the hashed addrs */
109 	unsigned char	addr[FLT_EXACT_COUNT][ETH_ALEN];
110 };
111 
112 /* 1024 is probably a high enough limit: modern hypervisors seem to support on
113  * the order of 100-200 CPUs so this leaves us some breathing space if we want
114  * to match a queue per guest CPU.
115  */
116 #define MAX_TAP_QUEUES 1024
117 
118 #define TUN_FLOW_EXPIRE (3 * HZ)
119 
120 /* A tun_file connects an open character device to a tuntap netdevice. It
121  * also contains all socket related strctures (except sock_fprog and tap_filter)
122  * to serve as one transmit queue for tuntap device. The sock_fprog and
123  * tap_filter were kept in tun_struct since they were used for filtering for the
124  * netdevice not for a specific queue (at least I didn't see the requirement for
125  * this).
126  *
127  * RCU usage:
128  * The tun_file and tun_struct are loosely coupled, the pointer from one to the
129  * other can only be read while rcu_read_lock or rtnl_lock is held.
130  */
131 struct tun_file {
132 	struct sock sk;
133 	struct socket socket;
134 	struct socket_wq wq;
135 	struct tun_struct __rcu *tun;
136 	struct net *net;
137 	struct fasync_struct *fasync;
138 	/* only used for fasnyc */
139 	unsigned int flags;
140 	u16 queue_index;
141 	struct list_head next;
142 	struct tun_struct *detached;
143 };
144 
145 struct tun_flow_entry {
146 	struct hlist_node hash_link;
147 	struct rcu_head rcu;
148 	struct tun_struct *tun;
149 
150 	u32 rxhash;
151 	int queue_index;
152 	unsigned long updated;
153 };
154 
155 #define TUN_NUM_FLOW_ENTRIES 1024
156 
157 /* Since the socket were moved to tun_file, to preserve the behavior of persist
158  * device, socket filter, sndbuf and vnet header size were restore when the
159  * file were attached to a persist device.
160  */
161 struct tun_struct {
162 	struct tun_file __rcu	*tfiles[MAX_TAP_QUEUES];
163 	unsigned int            numqueues;
164 	unsigned int 		flags;
165 	kuid_t			owner;
166 	kgid_t			group;
167 
168 	struct net_device	*dev;
169 	netdev_features_t	set_features;
170 #define TUN_USER_FEATURES (NETIF_F_HW_CSUM|NETIF_F_TSO_ECN|NETIF_F_TSO| \
171 			  NETIF_F_TSO6|NETIF_F_UFO)
172 
173 	int			vnet_hdr_sz;
174 	int			sndbuf;
175 	struct tap_filter	txflt;
176 	struct sock_fprog	fprog;
177 	/* protected by rtnl lock */
178 	bool			filter_attached;
179 #ifdef TUN_DEBUG
180 	int debug;
181 #endif
182 	spinlock_t lock;
183 	struct hlist_head flows[TUN_NUM_FLOW_ENTRIES];
184 	struct timer_list flow_gc_timer;
185 	unsigned long ageing_time;
186 	unsigned int numdisabled;
187 	struct list_head disabled;
188 };
189 
190 static inline u32 tun_hashfn(u32 rxhash)
191 {
192 	return rxhash & 0x3ff;
193 }
194 
195 static struct tun_flow_entry *tun_flow_find(struct hlist_head *head, u32 rxhash)
196 {
197 	struct tun_flow_entry *e;
198 	struct hlist_node *n;
199 
200 	hlist_for_each_entry_rcu(e, n, head, hash_link) {
201 		if (e->rxhash == rxhash)
202 			return e;
203 	}
204 	return NULL;
205 }
206 
207 static struct tun_flow_entry *tun_flow_create(struct tun_struct *tun,
208 					      struct hlist_head *head,
209 					      u32 rxhash, u16 queue_index)
210 {
211 	struct tun_flow_entry *e = kmalloc(sizeof(*e), GFP_ATOMIC);
212 
213 	if (e) {
214 		tun_debug(KERN_INFO, tun, "create flow: hash %u index %u\n",
215 			  rxhash, queue_index);
216 		e->updated = jiffies;
217 		e->rxhash = rxhash;
218 		e->queue_index = queue_index;
219 		e->tun = tun;
220 		hlist_add_head_rcu(&e->hash_link, head);
221 	}
222 	return e;
223 }
224 
225 static void tun_flow_delete(struct tun_struct *tun, struct tun_flow_entry *e)
226 {
227 	tun_debug(KERN_INFO, tun, "delete flow: hash %u index %u\n",
228 		  e->rxhash, e->queue_index);
229 	hlist_del_rcu(&e->hash_link);
230 	kfree_rcu(e, rcu);
231 }
232 
233 static void tun_flow_flush(struct tun_struct *tun)
234 {
235 	int i;
236 
237 	spin_lock_bh(&tun->lock);
238 	for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
239 		struct tun_flow_entry *e;
240 		struct hlist_node *h, *n;
241 
242 		hlist_for_each_entry_safe(e, h, n, &tun->flows[i], hash_link)
243 			tun_flow_delete(tun, e);
244 	}
245 	spin_unlock_bh(&tun->lock);
246 }
247 
248 static void tun_flow_delete_by_queue(struct tun_struct *tun, u16 queue_index)
249 {
250 	int i;
251 
252 	spin_lock_bh(&tun->lock);
253 	for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
254 		struct tun_flow_entry *e;
255 		struct hlist_node *h, *n;
256 
257 		hlist_for_each_entry_safe(e, h, n, &tun->flows[i], hash_link) {
258 			if (e->queue_index == queue_index)
259 				tun_flow_delete(tun, e);
260 		}
261 	}
262 	spin_unlock_bh(&tun->lock);
263 }
264 
265 static void tun_flow_cleanup(unsigned long data)
266 {
267 	struct tun_struct *tun = (struct tun_struct *)data;
268 	unsigned long delay = tun->ageing_time;
269 	unsigned long next_timer = jiffies + delay;
270 	unsigned long count = 0;
271 	int i;
272 
273 	tun_debug(KERN_INFO, tun, "tun_flow_cleanup\n");
274 
275 	spin_lock_bh(&tun->lock);
276 	for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
277 		struct tun_flow_entry *e;
278 		struct hlist_node *h, *n;
279 
280 		hlist_for_each_entry_safe(e, h, n, &tun->flows[i], hash_link) {
281 			unsigned long this_timer;
282 			count++;
283 			this_timer = e->updated + delay;
284 			if (time_before_eq(this_timer, jiffies))
285 				tun_flow_delete(tun, e);
286 			else if (time_before(this_timer, next_timer))
287 				next_timer = this_timer;
288 		}
289 	}
290 
291 	if (count)
292 		mod_timer(&tun->flow_gc_timer, round_jiffies_up(next_timer));
293 	spin_unlock_bh(&tun->lock);
294 }
295 
296 static void tun_flow_update(struct tun_struct *tun, u32 rxhash,
297 			    u16 queue_index)
298 {
299 	struct hlist_head *head;
300 	struct tun_flow_entry *e;
301 	unsigned long delay = tun->ageing_time;
302 
303 	if (!rxhash)
304 		return;
305 	else
306 		head = &tun->flows[tun_hashfn(rxhash)];
307 
308 	rcu_read_lock();
309 
310 	if (tun->numqueues == 1)
311 		goto unlock;
312 
313 	e = tun_flow_find(head, rxhash);
314 	if (likely(e)) {
315 		/* TODO: keep queueing to old queue until it's empty? */
316 		e->queue_index = queue_index;
317 		e->updated = jiffies;
318 	} else {
319 		spin_lock_bh(&tun->lock);
320 		if (!tun_flow_find(head, rxhash))
321 			tun_flow_create(tun, head, rxhash, queue_index);
322 
323 		if (!timer_pending(&tun->flow_gc_timer))
324 			mod_timer(&tun->flow_gc_timer,
325 				  round_jiffies_up(jiffies + delay));
326 		spin_unlock_bh(&tun->lock);
327 	}
328 
329 unlock:
330 	rcu_read_unlock();
331 }
332 
333 /* We try to identify a flow through its rxhash first. The reason that
334  * we do not check rxq no. is becuase some cards(e.g 82599), chooses
335  * the rxq based on the txq where the last packet of the flow comes. As
336  * the userspace application move between processors, we may get a
337  * different rxq no. here. If we could not get rxhash, then we would
338  * hope the rxq no. may help here.
339  */
340 static u16 tun_select_queue(struct net_device *dev, struct sk_buff *skb)
341 {
342 	struct tun_struct *tun = netdev_priv(dev);
343 	struct tun_flow_entry *e;
344 	u32 txq = 0;
345 	u32 numqueues = 0;
346 
347 	rcu_read_lock();
348 	numqueues = tun->numqueues;
349 
350 	txq = skb_get_rxhash(skb);
351 	if (txq) {
352 		e = tun_flow_find(&tun->flows[tun_hashfn(txq)], txq);
353 		if (e)
354 			txq = e->queue_index;
355 		else
356 			/* use multiply and shift instead of expensive divide */
357 			txq = ((u64)txq * numqueues) >> 32;
358 	} else if (likely(skb_rx_queue_recorded(skb))) {
359 		txq = skb_get_rx_queue(skb);
360 		while (unlikely(txq >= numqueues))
361 			txq -= numqueues;
362 	}
363 
364 	rcu_read_unlock();
365 	return txq;
366 }
367 
368 static inline bool tun_not_capable(struct tun_struct *tun)
369 {
370 	const struct cred *cred = current_cred();
371 	struct net *net = dev_net(tun->dev);
372 
373 	return ((uid_valid(tun->owner) && !uid_eq(cred->euid, tun->owner)) ||
374 		  (gid_valid(tun->group) && !in_egroup_p(tun->group))) &&
375 		!ns_capable(net->user_ns, CAP_NET_ADMIN);
376 }
377 
378 static void tun_set_real_num_queues(struct tun_struct *tun)
379 {
380 	netif_set_real_num_tx_queues(tun->dev, tun->numqueues);
381 	netif_set_real_num_rx_queues(tun->dev, tun->numqueues);
382 }
383 
384 static void tun_disable_queue(struct tun_struct *tun, struct tun_file *tfile)
385 {
386 	tfile->detached = tun;
387 	list_add_tail(&tfile->next, &tun->disabled);
388 	++tun->numdisabled;
389 }
390 
391 static struct tun_struct *tun_enable_queue(struct tun_file *tfile)
392 {
393 	struct tun_struct *tun = tfile->detached;
394 
395 	tfile->detached = NULL;
396 	list_del_init(&tfile->next);
397 	--tun->numdisabled;
398 	return tun;
399 }
400 
401 static void __tun_detach(struct tun_file *tfile, bool clean)
402 {
403 	struct tun_file *ntfile;
404 	struct tun_struct *tun;
405 	struct net_device *dev;
406 
407 	tun = rcu_dereference_protected(tfile->tun,
408 					lockdep_rtnl_is_held());
409 	if (tun) {
410 		u16 index = tfile->queue_index;
411 		BUG_ON(index >= tun->numqueues);
412 		dev = tun->dev;
413 
414 		rcu_assign_pointer(tun->tfiles[index],
415 				   tun->tfiles[tun->numqueues - 1]);
416 		rcu_assign_pointer(tfile->tun, NULL);
417 		ntfile = rcu_dereference_protected(tun->tfiles[index],
418 						   lockdep_rtnl_is_held());
419 		ntfile->queue_index = index;
420 
421 		--tun->numqueues;
422 		if (clean)
423 			sock_put(&tfile->sk);
424 		else
425 			tun_disable_queue(tun, tfile);
426 
427 		synchronize_net();
428 		tun_flow_delete_by_queue(tun, tun->numqueues + 1);
429 		/* Drop read queue */
430 		skb_queue_purge(&tfile->sk.sk_receive_queue);
431 		tun_set_real_num_queues(tun);
432 	} else if (tfile->detached && clean)
433 		tun = tun_enable_queue(tfile);
434 
435 	if (clean) {
436 		if (tun && tun->numqueues == 0 && tun->numdisabled == 0 &&
437 		    !(tun->flags & TUN_PERSIST))
438 			if (tun->dev->reg_state == NETREG_REGISTERED)
439 				unregister_netdevice(tun->dev);
440 
441 		BUG_ON(!test_bit(SOCK_EXTERNALLY_ALLOCATED,
442 				 &tfile->socket.flags));
443 		sk_release_kernel(&tfile->sk);
444 	}
445 }
446 
447 static void tun_detach(struct tun_file *tfile, bool clean)
448 {
449 	rtnl_lock();
450 	__tun_detach(tfile, clean);
451 	rtnl_unlock();
452 }
453 
454 static void tun_detach_all(struct net_device *dev)
455 {
456 	struct tun_struct *tun = netdev_priv(dev);
457 	struct tun_file *tfile, *tmp;
458 	int i, n = tun->numqueues;
459 
460 	for (i = 0; i < n; i++) {
461 		tfile = rcu_dereference_protected(tun->tfiles[i],
462 						  lockdep_rtnl_is_held());
463 		BUG_ON(!tfile);
464 		wake_up_all(&tfile->wq.wait);
465 		rcu_assign_pointer(tfile->tun, NULL);
466 		--tun->numqueues;
467 	}
468 	BUG_ON(tun->numqueues != 0);
469 
470 	synchronize_net();
471 	for (i = 0; i < n; i++) {
472 		tfile = rcu_dereference_protected(tun->tfiles[i],
473 						  lockdep_rtnl_is_held());
474 		/* Drop read queue */
475 		skb_queue_purge(&tfile->sk.sk_receive_queue);
476 		sock_put(&tfile->sk);
477 	}
478 	list_for_each_entry_safe(tfile, tmp, &tun->disabled, next) {
479 		tun_enable_queue(tfile);
480 		skb_queue_purge(&tfile->sk.sk_receive_queue);
481 		sock_put(&tfile->sk);
482 	}
483 	BUG_ON(tun->numdisabled != 0);
484 }
485 
486 static int tun_attach(struct tun_struct *tun, struct file *file)
487 {
488 	struct tun_file *tfile = file->private_data;
489 	int err;
490 
491 	err = -EINVAL;
492 	if (rcu_dereference_protected(tfile->tun, lockdep_rtnl_is_held()))
493 		goto out;
494 
495 	err = -EBUSY;
496 	if (!(tun->flags & TUN_TAP_MQ) && tun->numqueues == 1)
497 		goto out;
498 
499 	err = -E2BIG;
500 	if (!tfile->detached &&
501 	    tun->numqueues + tun->numdisabled == MAX_TAP_QUEUES)
502 		goto out;
503 
504 	err = 0;
505 
506 	/* Re-attach the filter to presist device */
507 	if (tun->filter_attached == true) {
508 		err = sk_attach_filter(&tun->fprog, tfile->socket.sk);
509 		if (!err)
510 			goto out;
511 	}
512 	tfile->queue_index = tun->numqueues;
513 	rcu_assign_pointer(tfile->tun, tun);
514 	rcu_assign_pointer(tun->tfiles[tun->numqueues], tfile);
515 	tun->numqueues++;
516 
517 	if (tfile->detached)
518 		tun_enable_queue(tfile);
519 	else
520 		sock_hold(&tfile->sk);
521 
522 	tun_set_real_num_queues(tun);
523 
524 	/* device is allowed to go away first, so no need to hold extra
525 	 * refcnt.
526 	 */
527 
528 out:
529 	return err;
530 }
531 
532 static struct tun_struct *__tun_get(struct tun_file *tfile)
533 {
534 	struct tun_struct *tun;
535 
536 	rcu_read_lock();
537 	tun = rcu_dereference(tfile->tun);
538 	if (tun)
539 		dev_hold(tun->dev);
540 	rcu_read_unlock();
541 
542 	return tun;
543 }
544 
545 static struct tun_struct *tun_get(struct file *file)
546 {
547 	return __tun_get(file->private_data);
548 }
549 
550 static void tun_put(struct tun_struct *tun)
551 {
552 	dev_put(tun->dev);
553 }
554 
555 /* TAP filtering */
556 static void addr_hash_set(u32 *mask, const u8 *addr)
557 {
558 	int n = ether_crc(ETH_ALEN, addr) >> 26;
559 	mask[n >> 5] |= (1 << (n & 31));
560 }
561 
562 static unsigned int addr_hash_test(const u32 *mask, const u8 *addr)
563 {
564 	int n = ether_crc(ETH_ALEN, addr) >> 26;
565 	return mask[n >> 5] & (1 << (n & 31));
566 }
567 
568 static int update_filter(struct tap_filter *filter, void __user *arg)
569 {
570 	struct { u8 u[ETH_ALEN]; } *addr;
571 	struct tun_filter uf;
572 	int err, alen, n, nexact;
573 
574 	if (copy_from_user(&uf, arg, sizeof(uf)))
575 		return -EFAULT;
576 
577 	if (!uf.count) {
578 		/* Disabled */
579 		filter->count = 0;
580 		return 0;
581 	}
582 
583 	alen = ETH_ALEN * uf.count;
584 	addr = kmalloc(alen, GFP_KERNEL);
585 	if (!addr)
586 		return -ENOMEM;
587 
588 	if (copy_from_user(addr, arg + sizeof(uf), alen)) {
589 		err = -EFAULT;
590 		goto done;
591 	}
592 
593 	/* The filter is updated without holding any locks. Which is
594 	 * perfectly safe. We disable it first and in the worst
595 	 * case we'll accept a few undesired packets. */
596 	filter->count = 0;
597 	wmb();
598 
599 	/* Use first set of addresses as an exact filter */
600 	for (n = 0; n < uf.count && n < FLT_EXACT_COUNT; n++)
601 		memcpy(filter->addr[n], addr[n].u, ETH_ALEN);
602 
603 	nexact = n;
604 
605 	/* Remaining multicast addresses are hashed,
606 	 * unicast will leave the filter disabled. */
607 	memset(filter->mask, 0, sizeof(filter->mask));
608 	for (; n < uf.count; n++) {
609 		if (!is_multicast_ether_addr(addr[n].u)) {
610 			err = 0; /* no filter */
611 			goto done;
612 		}
613 		addr_hash_set(filter->mask, addr[n].u);
614 	}
615 
616 	/* For ALLMULTI just set the mask to all ones.
617 	 * This overrides the mask populated above. */
618 	if ((uf.flags & TUN_FLT_ALLMULTI))
619 		memset(filter->mask, ~0, sizeof(filter->mask));
620 
621 	/* Now enable the filter */
622 	wmb();
623 	filter->count = nexact;
624 
625 	/* Return the number of exact filters */
626 	err = nexact;
627 
628 done:
629 	kfree(addr);
630 	return err;
631 }
632 
633 /* Returns: 0 - drop, !=0 - accept */
634 static int run_filter(struct tap_filter *filter, const struct sk_buff *skb)
635 {
636 	/* Cannot use eth_hdr(skb) here because skb_mac_hdr() is incorrect
637 	 * at this point. */
638 	struct ethhdr *eh = (struct ethhdr *) skb->data;
639 	int i;
640 
641 	/* Exact match */
642 	for (i = 0; i < filter->count; i++)
643 		if (ether_addr_equal(eh->h_dest, filter->addr[i]))
644 			return 1;
645 
646 	/* Inexact match (multicast only) */
647 	if (is_multicast_ether_addr(eh->h_dest))
648 		return addr_hash_test(filter->mask, eh->h_dest);
649 
650 	return 0;
651 }
652 
653 /*
654  * Checks whether the packet is accepted or not.
655  * Returns: 0 - drop, !=0 - accept
656  */
657 static int check_filter(struct tap_filter *filter, const struct sk_buff *skb)
658 {
659 	if (!filter->count)
660 		return 1;
661 
662 	return run_filter(filter, skb);
663 }
664 
665 /* Network device part of the driver */
666 
667 static const struct ethtool_ops tun_ethtool_ops;
668 
669 /* Net device detach from fd. */
670 static void tun_net_uninit(struct net_device *dev)
671 {
672 	tun_detach_all(dev);
673 }
674 
675 /* Net device open. */
676 static int tun_net_open(struct net_device *dev)
677 {
678 	netif_tx_start_all_queues(dev);
679 	return 0;
680 }
681 
682 /* Net device close. */
683 static int tun_net_close(struct net_device *dev)
684 {
685 	netif_tx_stop_all_queues(dev);
686 	return 0;
687 }
688 
689 /* Net device start xmit */
690 static netdev_tx_t tun_net_xmit(struct sk_buff *skb, struct net_device *dev)
691 {
692 	struct tun_struct *tun = netdev_priv(dev);
693 	int txq = skb->queue_mapping;
694 	struct tun_file *tfile;
695 
696 	rcu_read_lock();
697 	tfile = rcu_dereference(tun->tfiles[txq]);
698 
699 	/* Drop packet if interface is not attached */
700 	if (txq >= tun->numqueues)
701 		goto drop;
702 
703 	tun_debug(KERN_INFO, tun, "tun_net_xmit %d\n", skb->len);
704 
705 	BUG_ON(!tfile);
706 
707 	/* Drop if the filter does not like it.
708 	 * This is a noop if the filter is disabled.
709 	 * Filter can be enabled only for the TAP devices. */
710 	if (!check_filter(&tun->txflt, skb))
711 		goto drop;
712 
713 	if (tfile->socket.sk->sk_filter &&
714 	    sk_filter(tfile->socket.sk, skb))
715 		goto drop;
716 
717 	/* Limit the number of packets queued by dividing txq length with the
718 	 * number of queues.
719 	 */
720 	if (skb_queue_len(&tfile->socket.sk->sk_receive_queue)
721 			  >= dev->tx_queue_len / tun->numqueues)
722 		goto drop;
723 
724 	/* Orphan the skb - required as we might hang on to it
725 	 * for indefinite time. */
726 	if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC)))
727 		goto drop;
728 	skb_orphan(skb);
729 
730 	/* Enqueue packet */
731 	skb_queue_tail(&tfile->socket.sk->sk_receive_queue, skb);
732 
733 	/* Notify and wake up reader process */
734 	if (tfile->flags & TUN_FASYNC)
735 		kill_fasync(&tfile->fasync, SIGIO, POLL_IN);
736 	wake_up_interruptible_poll(&tfile->wq.wait, POLLIN |
737 				   POLLRDNORM | POLLRDBAND);
738 
739 	rcu_read_unlock();
740 	return NETDEV_TX_OK;
741 
742 drop:
743 	dev->stats.tx_dropped++;
744 	skb_tx_error(skb);
745 	kfree_skb(skb);
746 	rcu_read_unlock();
747 	return NETDEV_TX_OK;
748 }
749 
750 static void tun_net_mclist(struct net_device *dev)
751 {
752 	/*
753 	 * This callback is supposed to deal with mc filter in
754 	 * _rx_ path and has nothing to do with the _tx_ path.
755 	 * In rx path we always accept everything userspace gives us.
756 	 */
757 }
758 
759 #define MIN_MTU 68
760 #define MAX_MTU 65535
761 
762 static int
763 tun_net_change_mtu(struct net_device *dev, int new_mtu)
764 {
765 	if (new_mtu < MIN_MTU || new_mtu + dev->hard_header_len > MAX_MTU)
766 		return -EINVAL;
767 	dev->mtu = new_mtu;
768 	return 0;
769 }
770 
771 static netdev_features_t tun_net_fix_features(struct net_device *dev,
772 	netdev_features_t features)
773 {
774 	struct tun_struct *tun = netdev_priv(dev);
775 
776 	return (features & tun->set_features) | (features & ~TUN_USER_FEATURES);
777 }
778 #ifdef CONFIG_NET_POLL_CONTROLLER
779 static void tun_poll_controller(struct net_device *dev)
780 {
781 	/*
782 	 * Tun only receives frames when:
783 	 * 1) the char device endpoint gets data from user space
784 	 * 2) the tun socket gets a sendmsg call from user space
785 	 * Since both of those are syncronous operations, we are guaranteed
786 	 * never to have pending data when we poll for it
787 	 * so theres nothing to do here but return.
788 	 * We need this though so netpoll recognizes us as an interface that
789 	 * supports polling, which enables bridge devices in virt setups to
790 	 * still use netconsole
791 	 */
792 	return;
793 }
794 #endif
795 static const struct net_device_ops tun_netdev_ops = {
796 	.ndo_uninit		= tun_net_uninit,
797 	.ndo_open		= tun_net_open,
798 	.ndo_stop		= tun_net_close,
799 	.ndo_start_xmit		= tun_net_xmit,
800 	.ndo_change_mtu		= tun_net_change_mtu,
801 	.ndo_fix_features	= tun_net_fix_features,
802 	.ndo_select_queue	= tun_select_queue,
803 #ifdef CONFIG_NET_POLL_CONTROLLER
804 	.ndo_poll_controller	= tun_poll_controller,
805 #endif
806 };
807 
808 static const struct net_device_ops tap_netdev_ops = {
809 	.ndo_uninit		= tun_net_uninit,
810 	.ndo_open		= tun_net_open,
811 	.ndo_stop		= tun_net_close,
812 	.ndo_start_xmit		= tun_net_xmit,
813 	.ndo_change_mtu		= tun_net_change_mtu,
814 	.ndo_fix_features	= tun_net_fix_features,
815 	.ndo_set_rx_mode	= tun_net_mclist,
816 	.ndo_set_mac_address	= eth_mac_addr,
817 	.ndo_validate_addr	= eth_validate_addr,
818 	.ndo_select_queue	= tun_select_queue,
819 #ifdef CONFIG_NET_POLL_CONTROLLER
820 	.ndo_poll_controller	= tun_poll_controller,
821 #endif
822 };
823 
824 static int tun_flow_init(struct tun_struct *tun)
825 {
826 	int i;
827 
828 	for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++)
829 		INIT_HLIST_HEAD(&tun->flows[i]);
830 
831 	tun->ageing_time = TUN_FLOW_EXPIRE;
832 	setup_timer(&tun->flow_gc_timer, tun_flow_cleanup, (unsigned long)tun);
833 	mod_timer(&tun->flow_gc_timer,
834 		  round_jiffies_up(jiffies + tun->ageing_time));
835 
836 	return 0;
837 }
838 
839 static void tun_flow_uninit(struct tun_struct *tun)
840 {
841 	del_timer_sync(&tun->flow_gc_timer);
842 	tun_flow_flush(tun);
843 }
844 
845 /* Initialize net device. */
846 static void tun_net_init(struct net_device *dev)
847 {
848 	struct tun_struct *tun = netdev_priv(dev);
849 
850 	switch (tun->flags & TUN_TYPE_MASK) {
851 	case TUN_TUN_DEV:
852 		dev->netdev_ops = &tun_netdev_ops;
853 
854 		/* Point-to-Point TUN Device */
855 		dev->hard_header_len = 0;
856 		dev->addr_len = 0;
857 		dev->mtu = 1500;
858 
859 		/* Zero header length */
860 		dev->type = ARPHRD_NONE;
861 		dev->flags = IFF_POINTOPOINT | IFF_NOARP | IFF_MULTICAST;
862 		dev->tx_queue_len = TUN_READQ_SIZE;  /* We prefer our own queue length */
863 		break;
864 
865 	case TUN_TAP_DEV:
866 		dev->netdev_ops = &tap_netdev_ops;
867 		/* Ethernet TAP Device */
868 		ether_setup(dev);
869 		dev->priv_flags &= ~IFF_TX_SKB_SHARING;
870 		dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
871 
872 		eth_hw_addr_random(dev);
873 
874 		dev->tx_queue_len = TUN_READQ_SIZE;  /* We prefer our own queue length */
875 		break;
876 	}
877 }
878 
879 /* Character device part */
880 
881 /* Poll */
882 static unsigned int tun_chr_poll(struct file *file, poll_table *wait)
883 {
884 	struct tun_file *tfile = file->private_data;
885 	struct tun_struct *tun = __tun_get(tfile);
886 	struct sock *sk;
887 	unsigned int mask = 0;
888 
889 	if (!tun)
890 		return POLLERR;
891 
892 	sk = tfile->socket.sk;
893 
894 	tun_debug(KERN_INFO, tun, "tun_chr_poll\n");
895 
896 	poll_wait(file, &tfile->wq.wait, wait);
897 
898 	if (!skb_queue_empty(&sk->sk_receive_queue))
899 		mask |= POLLIN | POLLRDNORM;
900 
901 	if (sock_writeable(sk) ||
902 	    (!test_and_set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags) &&
903 	     sock_writeable(sk)))
904 		mask |= POLLOUT | POLLWRNORM;
905 
906 	if (tun->dev->reg_state != NETREG_REGISTERED)
907 		mask = POLLERR;
908 
909 	tun_put(tun);
910 	return mask;
911 }
912 
913 /* prepad is the amount to reserve at front.  len is length after that.
914  * linear is a hint as to how much to copy (usually headers). */
915 static struct sk_buff *tun_alloc_skb(struct tun_file *tfile,
916 				     size_t prepad, size_t len,
917 				     size_t linear, int noblock)
918 {
919 	struct sock *sk = tfile->socket.sk;
920 	struct sk_buff *skb;
921 	int err;
922 
923 	/* Under a page?  Don't bother with paged skb. */
924 	if (prepad + len < PAGE_SIZE || !linear)
925 		linear = len;
926 
927 	skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
928 				   &err);
929 	if (!skb)
930 		return ERR_PTR(err);
931 
932 	skb_reserve(skb, prepad);
933 	skb_put(skb, linear);
934 	skb->data_len = len - linear;
935 	skb->len += len - linear;
936 
937 	return skb;
938 }
939 
940 /* set skb frags from iovec, this can move to core network code for reuse */
941 static int zerocopy_sg_from_iovec(struct sk_buff *skb, const struct iovec *from,
942 				  int offset, size_t count)
943 {
944 	int len = iov_length(from, count) - offset;
945 	int copy = skb_headlen(skb);
946 	int size, offset1 = 0;
947 	int i = 0;
948 
949 	/* Skip over from offset */
950 	while (count && (offset >= from->iov_len)) {
951 		offset -= from->iov_len;
952 		++from;
953 		--count;
954 	}
955 
956 	/* copy up to skb headlen */
957 	while (count && (copy > 0)) {
958 		size = min_t(unsigned int, copy, from->iov_len - offset);
959 		if (copy_from_user(skb->data + offset1, from->iov_base + offset,
960 				   size))
961 			return -EFAULT;
962 		if (copy > size) {
963 			++from;
964 			--count;
965 			offset = 0;
966 		} else
967 			offset += size;
968 		copy -= size;
969 		offset1 += size;
970 	}
971 
972 	if (len == offset1)
973 		return 0;
974 
975 	while (count--) {
976 		struct page *page[MAX_SKB_FRAGS];
977 		int num_pages;
978 		unsigned long base;
979 		unsigned long truesize;
980 
981 		len = from->iov_len - offset;
982 		if (!len) {
983 			offset = 0;
984 			++from;
985 			continue;
986 		}
987 		base = (unsigned long)from->iov_base + offset;
988 		size = ((base & ~PAGE_MASK) + len + ~PAGE_MASK) >> PAGE_SHIFT;
989 		if (i + size > MAX_SKB_FRAGS)
990 			return -EMSGSIZE;
991 		num_pages = get_user_pages_fast(base, size, 0, &page[i]);
992 		if (num_pages != size) {
993 			for (i = 0; i < num_pages; i++)
994 				put_page(page[i]);
995 			return -EFAULT;
996 		}
997 		truesize = size * PAGE_SIZE;
998 		skb->data_len += len;
999 		skb->len += len;
1000 		skb->truesize += truesize;
1001 		atomic_add(truesize, &skb->sk->sk_wmem_alloc);
1002 		while (len) {
1003 			int off = base & ~PAGE_MASK;
1004 			int size = min_t(int, len, PAGE_SIZE - off);
1005 			__skb_fill_page_desc(skb, i, page[i], off, size);
1006 			skb_shinfo(skb)->nr_frags++;
1007 			/* increase sk_wmem_alloc */
1008 			base += size;
1009 			len -= size;
1010 			i++;
1011 		}
1012 		offset = 0;
1013 		++from;
1014 	}
1015 	return 0;
1016 }
1017 
1018 /* Get packet from user space buffer */
1019 static ssize_t tun_get_user(struct tun_struct *tun, struct tun_file *tfile,
1020 			    void *msg_control, const struct iovec *iv,
1021 			    size_t total_len, size_t count, int noblock)
1022 {
1023 	struct tun_pi pi = { 0, cpu_to_be16(ETH_P_IP) };
1024 	struct sk_buff *skb;
1025 	size_t len = total_len, align = NET_SKB_PAD;
1026 	struct virtio_net_hdr gso = { 0 };
1027 	int offset = 0;
1028 	int copylen;
1029 	bool zerocopy = false;
1030 	int err;
1031 	u32 rxhash;
1032 
1033 	if (!(tun->flags & TUN_NO_PI)) {
1034 		if ((len -= sizeof(pi)) > total_len)
1035 			return -EINVAL;
1036 
1037 		if (memcpy_fromiovecend((void *)&pi, iv, 0, sizeof(pi)))
1038 			return -EFAULT;
1039 		offset += sizeof(pi);
1040 	}
1041 
1042 	if (tun->flags & TUN_VNET_HDR) {
1043 		if ((len -= tun->vnet_hdr_sz) > total_len)
1044 			return -EINVAL;
1045 
1046 		if (memcpy_fromiovecend((void *)&gso, iv, offset, sizeof(gso)))
1047 			return -EFAULT;
1048 
1049 		if ((gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
1050 		    gso.csum_start + gso.csum_offset + 2 > gso.hdr_len)
1051 			gso.hdr_len = gso.csum_start + gso.csum_offset + 2;
1052 
1053 		if (gso.hdr_len > len)
1054 			return -EINVAL;
1055 		offset += tun->vnet_hdr_sz;
1056 	}
1057 
1058 	if ((tun->flags & TUN_TYPE_MASK) == TUN_TAP_DEV) {
1059 		align += NET_IP_ALIGN;
1060 		if (unlikely(len < ETH_HLEN ||
1061 			     (gso.hdr_len && gso.hdr_len < ETH_HLEN)))
1062 			return -EINVAL;
1063 	}
1064 
1065 	if (msg_control)
1066 		zerocopy = true;
1067 
1068 	if (zerocopy) {
1069 		/* Userspace may produce vectors with count greater than
1070 		 * MAX_SKB_FRAGS, so we need to linearize parts of the skb
1071 		 * to let the rest of data to be fit in the frags.
1072 		 */
1073 		if (count > MAX_SKB_FRAGS) {
1074 			copylen = iov_length(iv, count - MAX_SKB_FRAGS);
1075 			if (copylen < offset)
1076 				copylen = 0;
1077 			else
1078 				copylen -= offset;
1079 		} else
1080 				copylen = 0;
1081 		/* There are 256 bytes to be copied in skb, so there is enough
1082 		 * room for skb expand head in case it is used.
1083 		 * The rest of the buffer is mapped from userspace.
1084 		 */
1085 		if (copylen < gso.hdr_len)
1086 			copylen = gso.hdr_len;
1087 		if (!copylen)
1088 			copylen = GOODCOPY_LEN;
1089 	} else
1090 		copylen = len;
1091 
1092 	skb = tun_alloc_skb(tfile, align, copylen, gso.hdr_len, noblock);
1093 	if (IS_ERR(skb)) {
1094 		if (PTR_ERR(skb) != -EAGAIN)
1095 			tun->dev->stats.rx_dropped++;
1096 		return PTR_ERR(skb);
1097 	}
1098 
1099 	if (zerocopy)
1100 		err = zerocopy_sg_from_iovec(skb, iv, offset, count);
1101 	else
1102 		err = skb_copy_datagram_from_iovec(skb, 0, iv, offset, len);
1103 
1104 	if (err) {
1105 		tun->dev->stats.rx_dropped++;
1106 		kfree_skb(skb);
1107 		return -EFAULT;
1108 	}
1109 
1110 	if (gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
1111 		if (!skb_partial_csum_set(skb, gso.csum_start,
1112 					  gso.csum_offset)) {
1113 			tun->dev->stats.rx_frame_errors++;
1114 			kfree_skb(skb);
1115 			return -EINVAL;
1116 		}
1117 	}
1118 
1119 	switch (tun->flags & TUN_TYPE_MASK) {
1120 	case TUN_TUN_DEV:
1121 		if (tun->flags & TUN_NO_PI) {
1122 			switch (skb->data[0] & 0xf0) {
1123 			case 0x40:
1124 				pi.proto = htons(ETH_P_IP);
1125 				break;
1126 			case 0x60:
1127 				pi.proto = htons(ETH_P_IPV6);
1128 				break;
1129 			default:
1130 				tun->dev->stats.rx_dropped++;
1131 				kfree_skb(skb);
1132 				return -EINVAL;
1133 			}
1134 		}
1135 
1136 		skb_reset_mac_header(skb);
1137 		skb->protocol = pi.proto;
1138 		skb->dev = tun->dev;
1139 		break;
1140 	case TUN_TAP_DEV:
1141 		skb->protocol = eth_type_trans(skb, tun->dev);
1142 		break;
1143 	}
1144 
1145 	if (gso.gso_type != VIRTIO_NET_HDR_GSO_NONE) {
1146 		pr_debug("GSO!\n");
1147 		switch (gso.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
1148 		case VIRTIO_NET_HDR_GSO_TCPV4:
1149 			skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
1150 			break;
1151 		case VIRTIO_NET_HDR_GSO_TCPV6:
1152 			skb_shinfo(skb)->gso_type = SKB_GSO_TCPV6;
1153 			break;
1154 		case VIRTIO_NET_HDR_GSO_UDP:
1155 			skb_shinfo(skb)->gso_type = SKB_GSO_UDP;
1156 			break;
1157 		default:
1158 			tun->dev->stats.rx_frame_errors++;
1159 			kfree_skb(skb);
1160 			return -EINVAL;
1161 		}
1162 
1163 		if (gso.gso_type & VIRTIO_NET_HDR_GSO_ECN)
1164 			skb_shinfo(skb)->gso_type |= SKB_GSO_TCP_ECN;
1165 
1166 		skb_shinfo(skb)->gso_size = gso.gso_size;
1167 		if (skb_shinfo(skb)->gso_size == 0) {
1168 			tun->dev->stats.rx_frame_errors++;
1169 			kfree_skb(skb);
1170 			return -EINVAL;
1171 		}
1172 
1173 		/* Header must be checked, and gso_segs computed. */
1174 		skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
1175 		skb_shinfo(skb)->gso_segs = 0;
1176 	}
1177 
1178 	/* copy skb_ubuf_info for callback when skb has no error */
1179 	if (zerocopy) {
1180 		skb_shinfo(skb)->destructor_arg = msg_control;
1181 		skb_shinfo(skb)->tx_flags |= SKBTX_DEV_ZEROCOPY;
1182 	}
1183 
1184 	skb_reset_network_header(skb);
1185 	rxhash = skb_get_rxhash(skb);
1186 	netif_rx_ni(skb);
1187 
1188 	tun->dev->stats.rx_packets++;
1189 	tun->dev->stats.rx_bytes += len;
1190 
1191 	tun_flow_update(tun, rxhash, tfile->queue_index);
1192 	return total_len;
1193 }
1194 
1195 static ssize_t tun_chr_aio_write(struct kiocb *iocb, const struct iovec *iv,
1196 			      unsigned long count, loff_t pos)
1197 {
1198 	struct file *file = iocb->ki_filp;
1199 	struct tun_struct *tun = tun_get(file);
1200 	struct tun_file *tfile = file->private_data;
1201 	ssize_t result;
1202 
1203 	if (!tun)
1204 		return -EBADFD;
1205 
1206 	tun_debug(KERN_INFO, tun, "tun_chr_write %ld\n", count);
1207 
1208 	result = tun_get_user(tun, tfile, NULL, iv, iov_length(iv, count),
1209 			      count, file->f_flags & O_NONBLOCK);
1210 
1211 	tun_put(tun);
1212 	return result;
1213 }
1214 
1215 /* Put packet to the user space buffer */
1216 static ssize_t tun_put_user(struct tun_struct *tun,
1217 			    struct tun_file *tfile,
1218 			    struct sk_buff *skb,
1219 			    const struct iovec *iv, int len)
1220 {
1221 	struct tun_pi pi = { 0, skb->protocol };
1222 	ssize_t total = 0;
1223 
1224 	if (!(tun->flags & TUN_NO_PI)) {
1225 		if ((len -= sizeof(pi)) < 0)
1226 			return -EINVAL;
1227 
1228 		if (len < skb->len) {
1229 			/* Packet will be striped */
1230 			pi.flags |= TUN_PKT_STRIP;
1231 		}
1232 
1233 		if (memcpy_toiovecend(iv, (void *) &pi, 0, sizeof(pi)))
1234 			return -EFAULT;
1235 		total += sizeof(pi);
1236 	}
1237 
1238 	if (tun->flags & TUN_VNET_HDR) {
1239 		struct virtio_net_hdr gso = { 0 }; /* no info leak */
1240 		if ((len -= tun->vnet_hdr_sz) < 0)
1241 			return -EINVAL;
1242 
1243 		if (skb_is_gso(skb)) {
1244 			struct skb_shared_info *sinfo = skb_shinfo(skb);
1245 
1246 			/* This is a hint as to how much should be linear. */
1247 			gso.hdr_len = skb_headlen(skb);
1248 			gso.gso_size = sinfo->gso_size;
1249 			if (sinfo->gso_type & SKB_GSO_TCPV4)
1250 				gso.gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
1251 			else if (sinfo->gso_type & SKB_GSO_TCPV6)
1252 				gso.gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
1253 			else if (sinfo->gso_type & SKB_GSO_UDP)
1254 				gso.gso_type = VIRTIO_NET_HDR_GSO_UDP;
1255 			else {
1256 				pr_err("unexpected GSO type: "
1257 				       "0x%x, gso_size %d, hdr_len %d\n",
1258 				       sinfo->gso_type, gso.gso_size,
1259 				       gso.hdr_len);
1260 				print_hex_dump(KERN_ERR, "tun: ",
1261 					       DUMP_PREFIX_NONE,
1262 					       16, 1, skb->head,
1263 					       min((int)gso.hdr_len, 64), true);
1264 				WARN_ON_ONCE(1);
1265 				return -EINVAL;
1266 			}
1267 			if (sinfo->gso_type & SKB_GSO_TCP_ECN)
1268 				gso.gso_type |= VIRTIO_NET_HDR_GSO_ECN;
1269 		} else
1270 			gso.gso_type = VIRTIO_NET_HDR_GSO_NONE;
1271 
1272 		if (skb->ip_summed == CHECKSUM_PARTIAL) {
1273 			gso.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
1274 			gso.csum_start = skb_checksum_start_offset(skb);
1275 			gso.csum_offset = skb->csum_offset;
1276 		} else if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
1277 			gso.flags = VIRTIO_NET_HDR_F_DATA_VALID;
1278 		} /* else everything is zero */
1279 
1280 		if (unlikely(memcpy_toiovecend(iv, (void *)&gso, total,
1281 					       sizeof(gso))))
1282 			return -EFAULT;
1283 		total += tun->vnet_hdr_sz;
1284 	}
1285 
1286 	len = min_t(int, skb->len, len);
1287 
1288 	skb_copy_datagram_const_iovec(skb, 0, iv, total, len);
1289 	total += skb->len;
1290 
1291 	tun->dev->stats.tx_packets++;
1292 	tun->dev->stats.tx_bytes += len;
1293 
1294 	return total;
1295 }
1296 
1297 static ssize_t tun_do_read(struct tun_struct *tun, struct tun_file *tfile,
1298 			   struct kiocb *iocb, const struct iovec *iv,
1299 			   ssize_t len, int noblock)
1300 {
1301 	DECLARE_WAITQUEUE(wait, current);
1302 	struct sk_buff *skb;
1303 	ssize_t ret = 0;
1304 
1305 	tun_debug(KERN_INFO, tun, "tun_do_read\n");
1306 
1307 	if (unlikely(!noblock))
1308 		add_wait_queue(&tfile->wq.wait, &wait);
1309 	while (len) {
1310 		current->state = TASK_INTERRUPTIBLE;
1311 
1312 		/* Read frames from the queue */
1313 		if (!(skb = skb_dequeue(&tfile->socket.sk->sk_receive_queue))) {
1314 			if (noblock) {
1315 				ret = -EAGAIN;
1316 				break;
1317 			}
1318 			if (signal_pending(current)) {
1319 				ret = -ERESTARTSYS;
1320 				break;
1321 			}
1322 			if (tun->dev->reg_state != NETREG_REGISTERED) {
1323 				ret = -EIO;
1324 				break;
1325 			}
1326 
1327 			/* Nothing to read, let's sleep */
1328 			schedule();
1329 			continue;
1330 		}
1331 
1332 		ret = tun_put_user(tun, tfile, skb, iv, len);
1333 		kfree_skb(skb);
1334 		break;
1335 	}
1336 
1337 	current->state = TASK_RUNNING;
1338 	if (unlikely(!noblock))
1339 		remove_wait_queue(&tfile->wq.wait, &wait);
1340 
1341 	return ret;
1342 }
1343 
1344 static ssize_t tun_chr_aio_read(struct kiocb *iocb, const struct iovec *iv,
1345 			    unsigned long count, loff_t pos)
1346 {
1347 	struct file *file = iocb->ki_filp;
1348 	struct tun_file *tfile = file->private_data;
1349 	struct tun_struct *tun = __tun_get(tfile);
1350 	ssize_t len, ret;
1351 
1352 	if (!tun)
1353 		return -EBADFD;
1354 	len = iov_length(iv, count);
1355 	if (len < 0) {
1356 		ret = -EINVAL;
1357 		goto out;
1358 	}
1359 
1360 	ret = tun_do_read(tun, tfile, iocb, iv, len,
1361 			  file->f_flags & O_NONBLOCK);
1362 	ret = min_t(ssize_t, ret, len);
1363 out:
1364 	tun_put(tun);
1365 	return ret;
1366 }
1367 
1368 static void tun_free_netdev(struct net_device *dev)
1369 {
1370 	struct tun_struct *tun = netdev_priv(dev);
1371 
1372 	BUG_ON(!(list_empty(&tun->disabled)));
1373 	tun_flow_uninit(tun);
1374 	free_netdev(dev);
1375 }
1376 
1377 static void tun_setup(struct net_device *dev)
1378 {
1379 	struct tun_struct *tun = netdev_priv(dev);
1380 
1381 	tun->owner = INVALID_UID;
1382 	tun->group = INVALID_GID;
1383 
1384 	dev->ethtool_ops = &tun_ethtool_ops;
1385 	dev->destructor = tun_free_netdev;
1386 }
1387 
1388 /* Trivial set of netlink ops to allow deleting tun or tap
1389  * device with netlink.
1390  */
1391 static int tun_validate(struct nlattr *tb[], struct nlattr *data[])
1392 {
1393 	return -EINVAL;
1394 }
1395 
1396 static struct rtnl_link_ops tun_link_ops __read_mostly = {
1397 	.kind		= DRV_NAME,
1398 	.priv_size	= sizeof(struct tun_struct),
1399 	.setup		= tun_setup,
1400 	.validate	= tun_validate,
1401 };
1402 
1403 static void tun_sock_write_space(struct sock *sk)
1404 {
1405 	struct tun_file *tfile;
1406 	wait_queue_head_t *wqueue;
1407 
1408 	if (!sock_writeable(sk))
1409 		return;
1410 
1411 	if (!test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags))
1412 		return;
1413 
1414 	wqueue = sk_sleep(sk);
1415 	if (wqueue && waitqueue_active(wqueue))
1416 		wake_up_interruptible_sync_poll(wqueue, POLLOUT |
1417 						POLLWRNORM | POLLWRBAND);
1418 
1419 	tfile = container_of(sk, struct tun_file, sk);
1420 	kill_fasync(&tfile->fasync, SIGIO, POLL_OUT);
1421 }
1422 
1423 static int tun_sendmsg(struct kiocb *iocb, struct socket *sock,
1424 		       struct msghdr *m, size_t total_len)
1425 {
1426 	int ret;
1427 	struct tun_file *tfile = container_of(sock, struct tun_file, socket);
1428 	struct tun_struct *tun = __tun_get(tfile);
1429 
1430 	if (!tun)
1431 		return -EBADFD;
1432 	ret = tun_get_user(tun, tfile, m->msg_control, m->msg_iov, total_len,
1433 			   m->msg_iovlen, m->msg_flags & MSG_DONTWAIT);
1434 	tun_put(tun);
1435 	return ret;
1436 }
1437 
1438 
1439 static int tun_recvmsg(struct kiocb *iocb, struct socket *sock,
1440 		       struct msghdr *m, size_t total_len,
1441 		       int flags)
1442 {
1443 	struct tun_file *tfile = container_of(sock, struct tun_file, socket);
1444 	struct tun_struct *tun = __tun_get(tfile);
1445 	int ret;
1446 
1447 	if (!tun)
1448 		return -EBADFD;
1449 
1450 	if (flags & ~(MSG_DONTWAIT|MSG_TRUNC))
1451 		return -EINVAL;
1452 	ret = tun_do_read(tun, tfile, iocb, m->msg_iov, total_len,
1453 			  flags & MSG_DONTWAIT);
1454 	if (ret > total_len) {
1455 		m->msg_flags |= MSG_TRUNC;
1456 		ret = flags & MSG_TRUNC ? ret : total_len;
1457 	}
1458 	tun_put(tun);
1459 	return ret;
1460 }
1461 
1462 static int tun_release(struct socket *sock)
1463 {
1464 	if (sock->sk)
1465 		sock_put(sock->sk);
1466 	return 0;
1467 }
1468 
1469 /* Ops structure to mimic raw sockets with tun */
1470 static const struct proto_ops tun_socket_ops = {
1471 	.sendmsg = tun_sendmsg,
1472 	.recvmsg = tun_recvmsg,
1473 	.release = tun_release,
1474 };
1475 
1476 static struct proto tun_proto = {
1477 	.name		= "tun",
1478 	.owner		= THIS_MODULE,
1479 	.obj_size	= sizeof(struct tun_file),
1480 };
1481 
1482 static int tun_flags(struct tun_struct *tun)
1483 {
1484 	int flags = 0;
1485 
1486 	if (tun->flags & TUN_TUN_DEV)
1487 		flags |= IFF_TUN;
1488 	else
1489 		flags |= IFF_TAP;
1490 
1491 	if (tun->flags & TUN_NO_PI)
1492 		flags |= IFF_NO_PI;
1493 
1494 	/* This flag has no real effect.  We track the value for backwards
1495 	 * compatibility.
1496 	 */
1497 	if (tun->flags & TUN_ONE_QUEUE)
1498 		flags |= IFF_ONE_QUEUE;
1499 
1500 	if (tun->flags & TUN_VNET_HDR)
1501 		flags |= IFF_VNET_HDR;
1502 
1503 	if (tun->flags & TUN_TAP_MQ)
1504 		flags |= IFF_MULTI_QUEUE;
1505 
1506 	return flags;
1507 }
1508 
1509 static ssize_t tun_show_flags(struct device *dev, struct device_attribute *attr,
1510 			      char *buf)
1511 {
1512 	struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1513 	return sprintf(buf, "0x%x\n", tun_flags(tun));
1514 }
1515 
1516 static ssize_t tun_show_owner(struct device *dev, struct device_attribute *attr,
1517 			      char *buf)
1518 {
1519 	struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1520 	return uid_valid(tun->owner)?
1521 		sprintf(buf, "%u\n",
1522 			from_kuid_munged(current_user_ns(), tun->owner)):
1523 		sprintf(buf, "-1\n");
1524 }
1525 
1526 static ssize_t tun_show_group(struct device *dev, struct device_attribute *attr,
1527 			      char *buf)
1528 {
1529 	struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1530 	return gid_valid(tun->group) ?
1531 		sprintf(buf, "%u\n",
1532 			from_kgid_munged(current_user_ns(), tun->group)):
1533 		sprintf(buf, "-1\n");
1534 }
1535 
1536 static DEVICE_ATTR(tun_flags, 0444, tun_show_flags, NULL);
1537 static DEVICE_ATTR(owner, 0444, tun_show_owner, NULL);
1538 static DEVICE_ATTR(group, 0444, tun_show_group, NULL);
1539 
1540 static int tun_set_iff(struct net *net, struct file *file, struct ifreq *ifr)
1541 {
1542 	struct tun_struct *tun;
1543 	struct tun_file *tfile = file->private_data;
1544 	struct net_device *dev;
1545 	int err;
1546 
1547 	dev = __dev_get_by_name(net, ifr->ifr_name);
1548 	if (dev) {
1549 		if (ifr->ifr_flags & IFF_TUN_EXCL)
1550 			return -EBUSY;
1551 		if ((ifr->ifr_flags & IFF_TUN) && dev->netdev_ops == &tun_netdev_ops)
1552 			tun = netdev_priv(dev);
1553 		else if ((ifr->ifr_flags & IFF_TAP) && dev->netdev_ops == &tap_netdev_ops)
1554 			tun = netdev_priv(dev);
1555 		else
1556 			return -EINVAL;
1557 
1558 		if (tun_not_capable(tun))
1559 			return -EPERM;
1560 		err = security_tun_dev_attach(tfile->socket.sk);
1561 		if (err < 0)
1562 			return err;
1563 
1564 		err = tun_attach(tun, file);
1565 		if (err < 0)
1566 			return err;
1567 
1568 		if (tun->flags & TUN_TAP_MQ &&
1569 		    (tun->numqueues + tun->numdisabled > 1))
1570 			return err;
1571 	}
1572 	else {
1573 		char *name;
1574 		unsigned long flags = 0;
1575 
1576 		if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
1577 			return -EPERM;
1578 		err = security_tun_dev_create();
1579 		if (err < 0)
1580 			return err;
1581 
1582 		/* Set dev type */
1583 		if (ifr->ifr_flags & IFF_TUN) {
1584 			/* TUN device */
1585 			flags |= TUN_TUN_DEV;
1586 			name = "tun%d";
1587 		} else if (ifr->ifr_flags & IFF_TAP) {
1588 			/* TAP device */
1589 			flags |= TUN_TAP_DEV;
1590 			name = "tap%d";
1591 		} else
1592 			return -EINVAL;
1593 
1594 		if (*ifr->ifr_name)
1595 			name = ifr->ifr_name;
1596 
1597 		dev = alloc_netdev_mqs(sizeof(struct tun_struct), name,
1598 				       tun_setup,
1599 				       MAX_TAP_QUEUES, MAX_TAP_QUEUES);
1600 		if (!dev)
1601 			return -ENOMEM;
1602 
1603 		dev_net_set(dev, net);
1604 		dev->rtnl_link_ops = &tun_link_ops;
1605 
1606 		tun = netdev_priv(dev);
1607 		tun->dev = dev;
1608 		tun->flags = flags;
1609 		tun->txflt.count = 0;
1610 		tun->vnet_hdr_sz = sizeof(struct virtio_net_hdr);
1611 
1612 		tun->filter_attached = false;
1613 		tun->sndbuf = tfile->socket.sk->sk_sndbuf;
1614 
1615 		spin_lock_init(&tun->lock);
1616 
1617 		security_tun_dev_post_create(&tfile->sk);
1618 
1619 		tun_net_init(dev);
1620 
1621 		err = tun_flow_init(tun);
1622 		if (err < 0)
1623 			goto err_free_dev;
1624 
1625 		dev->hw_features = NETIF_F_SG | NETIF_F_FRAGLIST |
1626 			TUN_USER_FEATURES;
1627 		dev->features = dev->hw_features;
1628 
1629 		INIT_LIST_HEAD(&tun->disabled);
1630 		err = tun_attach(tun, file);
1631 		if (err < 0)
1632 			goto err_free_dev;
1633 
1634 		err = register_netdevice(tun->dev);
1635 		if (err < 0)
1636 			goto err_free_dev;
1637 
1638 		if (device_create_file(&tun->dev->dev, &dev_attr_tun_flags) ||
1639 		    device_create_file(&tun->dev->dev, &dev_attr_owner) ||
1640 		    device_create_file(&tun->dev->dev, &dev_attr_group))
1641 			pr_err("Failed to create tun sysfs files\n");
1642 
1643 		netif_carrier_on(tun->dev);
1644 	}
1645 
1646 	tun_debug(KERN_INFO, tun, "tun_set_iff\n");
1647 
1648 	if (ifr->ifr_flags & IFF_NO_PI)
1649 		tun->flags |= TUN_NO_PI;
1650 	else
1651 		tun->flags &= ~TUN_NO_PI;
1652 
1653 	/* This flag has no real effect.  We track the value for backwards
1654 	 * compatibility.
1655 	 */
1656 	if (ifr->ifr_flags & IFF_ONE_QUEUE)
1657 		tun->flags |= TUN_ONE_QUEUE;
1658 	else
1659 		tun->flags &= ~TUN_ONE_QUEUE;
1660 
1661 	if (ifr->ifr_flags & IFF_VNET_HDR)
1662 		tun->flags |= TUN_VNET_HDR;
1663 	else
1664 		tun->flags &= ~TUN_VNET_HDR;
1665 
1666 	if (ifr->ifr_flags & IFF_MULTI_QUEUE)
1667 		tun->flags |= TUN_TAP_MQ;
1668 	else
1669 		tun->flags &= ~TUN_TAP_MQ;
1670 
1671 	/* Make sure persistent devices do not get stuck in
1672 	 * xoff state.
1673 	 */
1674 	if (netif_running(tun->dev))
1675 		netif_tx_wake_all_queues(tun->dev);
1676 
1677 	strcpy(ifr->ifr_name, tun->dev->name);
1678 	return 0;
1679 
1680  err_free_dev:
1681 	free_netdev(dev);
1682 	return err;
1683 }
1684 
1685 static void tun_get_iff(struct net *net, struct tun_struct *tun,
1686 		       struct ifreq *ifr)
1687 {
1688 	tun_debug(KERN_INFO, tun, "tun_get_iff\n");
1689 
1690 	strcpy(ifr->ifr_name, tun->dev->name);
1691 
1692 	ifr->ifr_flags = tun_flags(tun);
1693 
1694 }
1695 
1696 /* This is like a cut-down ethtool ops, except done via tun fd so no
1697  * privs required. */
1698 static int set_offload(struct tun_struct *tun, unsigned long arg)
1699 {
1700 	netdev_features_t features = 0;
1701 
1702 	if (arg & TUN_F_CSUM) {
1703 		features |= NETIF_F_HW_CSUM;
1704 		arg &= ~TUN_F_CSUM;
1705 
1706 		if (arg & (TUN_F_TSO4|TUN_F_TSO6)) {
1707 			if (arg & TUN_F_TSO_ECN) {
1708 				features |= NETIF_F_TSO_ECN;
1709 				arg &= ~TUN_F_TSO_ECN;
1710 			}
1711 			if (arg & TUN_F_TSO4)
1712 				features |= NETIF_F_TSO;
1713 			if (arg & TUN_F_TSO6)
1714 				features |= NETIF_F_TSO6;
1715 			arg &= ~(TUN_F_TSO4|TUN_F_TSO6);
1716 		}
1717 
1718 		if (arg & TUN_F_UFO) {
1719 			features |= NETIF_F_UFO;
1720 			arg &= ~TUN_F_UFO;
1721 		}
1722 	}
1723 
1724 	/* This gives the user a way to test for new features in future by
1725 	 * trying to set them. */
1726 	if (arg)
1727 		return -EINVAL;
1728 
1729 	tun->set_features = features;
1730 	netdev_update_features(tun->dev);
1731 
1732 	return 0;
1733 }
1734 
1735 static void tun_detach_filter(struct tun_struct *tun, int n)
1736 {
1737 	int i;
1738 	struct tun_file *tfile;
1739 
1740 	for (i = 0; i < n; i++) {
1741 		tfile = rcu_dereference_protected(tun->tfiles[i],
1742 						  lockdep_rtnl_is_held());
1743 		sk_detach_filter(tfile->socket.sk);
1744 	}
1745 
1746 	tun->filter_attached = false;
1747 }
1748 
1749 static int tun_attach_filter(struct tun_struct *tun)
1750 {
1751 	int i, ret = 0;
1752 	struct tun_file *tfile;
1753 
1754 	for (i = 0; i < tun->numqueues; i++) {
1755 		tfile = rcu_dereference_protected(tun->tfiles[i],
1756 						  lockdep_rtnl_is_held());
1757 		ret = sk_attach_filter(&tun->fprog, tfile->socket.sk);
1758 		if (ret) {
1759 			tun_detach_filter(tun, i);
1760 			return ret;
1761 		}
1762 	}
1763 
1764 	tun->filter_attached = true;
1765 	return ret;
1766 }
1767 
1768 static void tun_set_sndbuf(struct tun_struct *tun)
1769 {
1770 	struct tun_file *tfile;
1771 	int i;
1772 
1773 	for (i = 0; i < tun->numqueues; i++) {
1774 		tfile = rcu_dereference_protected(tun->tfiles[i],
1775 						lockdep_rtnl_is_held());
1776 		tfile->socket.sk->sk_sndbuf = tun->sndbuf;
1777 	}
1778 }
1779 
1780 static int tun_set_queue(struct file *file, struct ifreq *ifr)
1781 {
1782 	struct tun_file *tfile = file->private_data;
1783 	struct tun_struct *tun;
1784 	int ret = 0;
1785 
1786 	rtnl_lock();
1787 
1788 	if (ifr->ifr_flags & IFF_ATTACH_QUEUE) {
1789 		tun = tfile->detached;
1790 		if (!tun)
1791 			ret = -EINVAL;
1792 		else if (tun_not_capable(tun))
1793 			ret = -EPERM;
1794 		else
1795 			ret = tun_attach(tun, file);
1796 	} else if (ifr->ifr_flags & IFF_DETACH_QUEUE) {
1797 		tun = rcu_dereference_protected(tfile->tun,
1798 						lockdep_rtnl_is_held());
1799 		if (!tun || !(tun->flags & TUN_TAP_MQ))
1800 			ret = -EINVAL;
1801 		else
1802 			__tun_detach(tfile, false);
1803 	} else
1804 		ret = -EINVAL;
1805 
1806 	rtnl_unlock();
1807 	return ret;
1808 }
1809 
1810 static long __tun_chr_ioctl(struct file *file, unsigned int cmd,
1811 			    unsigned long arg, int ifreq_len)
1812 {
1813 	struct tun_file *tfile = file->private_data;
1814 	struct tun_struct *tun;
1815 	void __user* argp = (void __user*)arg;
1816 	struct ifreq ifr;
1817 	kuid_t owner;
1818 	kgid_t group;
1819 	int sndbuf;
1820 	int vnet_hdr_sz;
1821 	int ret;
1822 
1823 	if (cmd == TUNSETIFF || cmd == TUNSETQUEUE || _IOC_TYPE(cmd) == 0x89) {
1824 		if (copy_from_user(&ifr, argp, ifreq_len))
1825 			return -EFAULT;
1826 	} else {
1827 		memset(&ifr, 0, sizeof(ifr));
1828 	}
1829 	if (cmd == TUNGETFEATURES) {
1830 		/* Currently this just means: "what IFF flags are valid?".
1831 		 * This is needed because we never checked for invalid flags on
1832 		 * TUNSETIFF. */
1833 		return put_user(IFF_TUN | IFF_TAP | IFF_NO_PI | IFF_ONE_QUEUE |
1834 				IFF_VNET_HDR | IFF_MULTI_QUEUE,
1835 				(unsigned int __user*)argp);
1836 	} else if (cmd == TUNSETQUEUE)
1837 		return tun_set_queue(file, &ifr);
1838 
1839 	ret = 0;
1840 	rtnl_lock();
1841 
1842 	tun = __tun_get(tfile);
1843 	if (cmd == TUNSETIFF && !tun) {
1844 		ifr.ifr_name[IFNAMSIZ-1] = '\0';
1845 
1846 		ret = tun_set_iff(tfile->net, file, &ifr);
1847 
1848 		if (ret)
1849 			goto unlock;
1850 
1851 		if (copy_to_user(argp, &ifr, ifreq_len))
1852 			ret = -EFAULT;
1853 		goto unlock;
1854 	}
1855 
1856 	ret = -EBADFD;
1857 	if (!tun)
1858 		goto unlock;
1859 
1860 	tun_debug(KERN_INFO, tun, "tun_chr_ioctl cmd %u\n", cmd);
1861 
1862 	ret = 0;
1863 	switch (cmd) {
1864 	case TUNGETIFF:
1865 		tun_get_iff(current->nsproxy->net_ns, tun, &ifr);
1866 
1867 		if (copy_to_user(argp, &ifr, ifreq_len))
1868 			ret = -EFAULT;
1869 		break;
1870 
1871 	case TUNSETNOCSUM:
1872 		/* Disable/Enable checksum */
1873 
1874 		/* [unimplemented] */
1875 		tun_debug(KERN_INFO, tun, "ignored: set checksum %s\n",
1876 			  arg ? "disabled" : "enabled");
1877 		break;
1878 
1879 	case TUNSETPERSIST:
1880 		/* Disable/Enable persist mode. Keep an extra reference to the
1881 		 * module to prevent the module being unprobed.
1882 		 */
1883 		if (arg) {
1884 			tun->flags |= TUN_PERSIST;
1885 			__module_get(THIS_MODULE);
1886 		} else {
1887 			tun->flags &= ~TUN_PERSIST;
1888 			module_put(THIS_MODULE);
1889 		}
1890 
1891 		tun_debug(KERN_INFO, tun, "persist %s\n",
1892 			  arg ? "enabled" : "disabled");
1893 		break;
1894 
1895 	case TUNSETOWNER:
1896 		/* Set owner of the device */
1897 		owner = make_kuid(current_user_ns(), arg);
1898 		if (!uid_valid(owner)) {
1899 			ret = -EINVAL;
1900 			break;
1901 		}
1902 		tun->owner = owner;
1903 		tun_debug(KERN_INFO, tun, "owner set to %u\n",
1904 			  from_kuid(&init_user_ns, tun->owner));
1905 		break;
1906 
1907 	case TUNSETGROUP:
1908 		/* Set group of the device */
1909 		group = make_kgid(current_user_ns(), arg);
1910 		if (!gid_valid(group)) {
1911 			ret = -EINVAL;
1912 			break;
1913 		}
1914 		tun->group = group;
1915 		tun_debug(KERN_INFO, tun, "group set to %u\n",
1916 			  from_kgid(&init_user_ns, tun->group));
1917 		break;
1918 
1919 	case TUNSETLINK:
1920 		/* Only allow setting the type when the interface is down */
1921 		if (tun->dev->flags & IFF_UP) {
1922 			tun_debug(KERN_INFO, tun,
1923 				  "Linktype set failed because interface is up\n");
1924 			ret = -EBUSY;
1925 		} else {
1926 			tun->dev->type = (int) arg;
1927 			tun_debug(KERN_INFO, tun, "linktype set to %d\n",
1928 				  tun->dev->type);
1929 			ret = 0;
1930 		}
1931 		break;
1932 
1933 #ifdef TUN_DEBUG
1934 	case TUNSETDEBUG:
1935 		tun->debug = arg;
1936 		break;
1937 #endif
1938 	case TUNSETOFFLOAD:
1939 		ret = set_offload(tun, arg);
1940 		break;
1941 
1942 	case TUNSETTXFILTER:
1943 		/* Can be set only for TAPs */
1944 		ret = -EINVAL;
1945 		if ((tun->flags & TUN_TYPE_MASK) != TUN_TAP_DEV)
1946 			break;
1947 		ret = update_filter(&tun->txflt, (void __user *)arg);
1948 		break;
1949 
1950 	case SIOCGIFHWADDR:
1951 		/* Get hw address */
1952 		memcpy(ifr.ifr_hwaddr.sa_data, tun->dev->dev_addr, ETH_ALEN);
1953 		ifr.ifr_hwaddr.sa_family = tun->dev->type;
1954 		if (copy_to_user(argp, &ifr, ifreq_len))
1955 			ret = -EFAULT;
1956 		break;
1957 
1958 	case SIOCSIFHWADDR:
1959 		/* Set hw address */
1960 		tun_debug(KERN_DEBUG, tun, "set hw address: %pM\n",
1961 			  ifr.ifr_hwaddr.sa_data);
1962 
1963 		ret = dev_set_mac_address(tun->dev, &ifr.ifr_hwaddr);
1964 		break;
1965 
1966 	case TUNGETSNDBUF:
1967 		sndbuf = tfile->socket.sk->sk_sndbuf;
1968 		if (copy_to_user(argp, &sndbuf, sizeof(sndbuf)))
1969 			ret = -EFAULT;
1970 		break;
1971 
1972 	case TUNSETSNDBUF:
1973 		if (copy_from_user(&sndbuf, argp, sizeof(sndbuf))) {
1974 			ret = -EFAULT;
1975 			break;
1976 		}
1977 
1978 		tun->sndbuf = sndbuf;
1979 		tun_set_sndbuf(tun);
1980 		break;
1981 
1982 	case TUNGETVNETHDRSZ:
1983 		vnet_hdr_sz = tun->vnet_hdr_sz;
1984 		if (copy_to_user(argp, &vnet_hdr_sz, sizeof(vnet_hdr_sz)))
1985 			ret = -EFAULT;
1986 		break;
1987 
1988 	case TUNSETVNETHDRSZ:
1989 		if (copy_from_user(&vnet_hdr_sz, argp, sizeof(vnet_hdr_sz))) {
1990 			ret = -EFAULT;
1991 			break;
1992 		}
1993 		if (vnet_hdr_sz < (int)sizeof(struct virtio_net_hdr)) {
1994 			ret = -EINVAL;
1995 			break;
1996 		}
1997 
1998 		tun->vnet_hdr_sz = vnet_hdr_sz;
1999 		break;
2000 
2001 	case TUNATTACHFILTER:
2002 		/* Can be set only for TAPs */
2003 		ret = -EINVAL;
2004 		if ((tun->flags & TUN_TYPE_MASK) != TUN_TAP_DEV)
2005 			break;
2006 		ret = -EFAULT;
2007 		if (copy_from_user(&tun->fprog, argp, sizeof(tun->fprog)))
2008 			break;
2009 
2010 		ret = tun_attach_filter(tun);
2011 		break;
2012 
2013 	case TUNDETACHFILTER:
2014 		/* Can be set only for TAPs */
2015 		ret = -EINVAL;
2016 		if ((tun->flags & TUN_TYPE_MASK) != TUN_TAP_DEV)
2017 			break;
2018 		ret = 0;
2019 		tun_detach_filter(tun, tun->numqueues);
2020 		break;
2021 
2022 	default:
2023 		ret = -EINVAL;
2024 		break;
2025 	}
2026 
2027 unlock:
2028 	rtnl_unlock();
2029 	if (tun)
2030 		tun_put(tun);
2031 	return ret;
2032 }
2033 
2034 static long tun_chr_ioctl(struct file *file,
2035 			  unsigned int cmd, unsigned long arg)
2036 {
2037 	return __tun_chr_ioctl(file, cmd, arg, sizeof (struct ifreq));
2038 }
2039 
2040 #ifdef CONFIG_COMPAT
2041 static long tun_chr_compat_ioctl(struct file *file,
2042 			 unsigned int cmd, unsigned long arg)
2043 {
2044 	switch (cmd) {
2045 	case TUNSETIFF:
2046 	case TUNGETIFF:
2047 	case TUNSETTXFILTER:
2048 	case TUNGETSNDBUF:
2049 	case TUNSETSNDBUF:
2050 	case SIOCGIFHWADDR:
2051 	case SIOCSIFHWADDR:
2052 		arg = (unsigned long)compat_ptr(arg);
2053 		break;
2054 	default:
2055 		arg = (compat_ulong_t)arg;
2056 		break;
2057 	}
2058 
2059 	/*
2060 	 * compat_ifreq is shorter than ifreq, so we must not access beyond
2061 	 * the end of that structure. All fields that are used in this
2062 	 * driver are compatible though, we don't need to convert the
2063 	 * contents.
2064 	 */
2065 	return __tun_chr_ioctl(file, cmd, arg, sizeof(struct compat_ifreq));
2066 }
2067 #endif /* CONFIG_COMPAT */
2068 
2069 static int tun_chr_fasync(int fd, struct file *file, int on)
2070 {
2071 	struct tun_file *tfile = file->private_data;
2072 	int ret;
2073 
2074 	if ((ret = fasync_helper(fd, file, on, &tfile->fasync)) < 0)
2075 		goto out;
2076 
2077 	if (on) {
2078 		ret = __f_setown(file, task_pid(current), PIDTYPE_PID, 0);
2079 		if (ret)
2080 			goto out;
2081 		tfile->flags |= TUN_FASYNC;
2082 	} else
2083 		tfile->flags &= ~TUN_FASYNC;
2084 	ret = 0;
2085 out:
2086 	return ret;
2087 }
2088 
2089 static int tun_chr_open(struct inode *inode, struct file * file)
2090 {
2091 	struct tun_file *tfile;
2092 
2093 	DBG1(KERN_INFO, "tunX: tun_chr_open\n");
2094 
2095 	tfile = (struct tun_file *)sk_alloc(&init_net, AF_UNSPEC, GFP_KERNEL,
2096 					    &tun_proto);
2097 	if (!tfile)
2098 		return -ENOMEM;
2099 	rcu_assign_pointer(tfile->tun, NULL);
2100 	tfile->net = get_net(current->nsproxy->net_ns);
2101 	tfile->flags = 0;
2102 
2103 	rcu_assign_pointer(tfile->socket.wq, &tfile->wq);
2104 	init_waitqueue_head(&tfile->wq.wait);
2105 
2106 	tfile->socket.file = file;
2107 	tfile->socket.ops = &tun_socket_ops;
2108 
2109 	sock_init_data(&tfile->socket, &tfile->sk);
2110 	sk_change_net(&tfile->sk, tfile->net);
2111 
2112 	tfile->sk.sk_write_space = tun_sock_write_space;
2113 	tfile->sk.sk_sndbuf = INT_MAX;
2114 
2115 	file->private_data = tfile;
2116 	set_bit(SOCK_EXTERNALLY_ALLOCATED, &tfile->socket.flags);
2117 	INIT_LIST_HEAD(&tfile->next);
2118 
2119 	return 0;
2120 }
2121 
2122 static int tun_chr_close(struct inode *inode, struct file *file)
2123 {
2124 	struct tun_file *tfile = file->private_data;
2125 	struct net *net = tfile->net;
2126 
2127 	tun_detach(tfile, true);
2128 	put_net(net);
2129 
2130 	return 0;
2131 }
2132 
2133 static const struct file_operations tun_fops = {
2134 	.owner	= THIS_MODULE,
2135 	.llseek = no_llseek,
2136 	.read  = do_sync_read,
2137 	.aio_read  = tun_chr_aio_read,
2138 	.write = do_sync_write,
2139 	.aio_write = tun_chr_aio_write,
2140 	.poll	= tun_chr_poll,
2141 	.unlocked_ioctl	= tun_chr_ioctl,
2142 #ifdef CONFIG_COMPAT
2143 	.compat_ioctl = tun_chr_compat_ioctl,
2144 #endif
2145 	.open	= tun_chr_open,
2146 	.release = tun_chr_close,
2147 	.fasync = tun_chr_fasync
2148 };
2149 
2150 static struct miscdevice tun_miscdev = {
2151 	.minor = TUN_MINOR,
2152 	.name = "tun",
2153 	.nodename = "net/tun",
2154 	.fops = &tun_fops,
2155 };
2156 
2157 /* ethtool interface */
2158 
2159 static int tun_get_settings(struct net_device *dev, struct ethtool_cmd *cmd)
2160 {
2161 	cmd->supported		= 0;
2162 	cmd->advertising	= 0;
2163 	ethtool_cmd_speed_set(cmd, SPEED_10);
2164 	cmd->duplex		= DUPLEX_FULL;
2165 	cmd->port		= PORT_TP;
2166 	cmd->phy_address	= 0;
2167 	cmd->transceiver	= XCVR_INTERNAL;
2168 	cmd->autoneg		= AUTONEG_DISABLE;
2169 	cmd->maxtxpkt		= 0;
2170 	cmd->maxrxpkt		= 0;
2171 	return 0;
2172 }
2173 
2174 static void tun_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
2175 {
2176 	struct tun_struct *tun = netdev_priv(dev);
2177 
2178 	strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
2179 	strlcpy(info->version, DRV_VERSION, sizeof(info->version));
2180 
2181 	switch (tun->flags & TUN_TYPE_MASK) {
2182 	case TUN_TUN_DEV:
2183 		strlcpy(info->bus_info, "tun", sizeof(info->bus_info));
2184 		break;
2185 	case TUN_TAP_DEV:
2186 		strlcpy(info->bus_info, "tap", sizeof(info->bus_info));
2187 		break;
2188 	}
2189 }
2190 
2191 static u32 tun_get_msglevel(struct net_device *dev)
2192 {
2193 #ifdef TUN_DEBUG
2194 	struct tun_struct *tun = netdev_priv(dev);
2195 	return tun->debug;
2196 #else
2197 	return -EOPNOTSUPP;
2198 #endif
2199 }
2200 
2201 static void tun_set_msglevel(struct net_device *dev, u32 value)
2202 {
2203 #ifdef TUN_DEBUG
2204 	struct tun_struct *tun = netdev_priv(dev);
2205 	tun->debug = value;
2206 #endif
2207 }
2208 
2209 static const struct ethtool_ops tun_ethtool_ops = {
2210 	.get_settings	= tun_get_settings,
2211 	.get_drvinfo	= tun_get_drvinfo,
2212 	.get_msglevel	= tun_get_msglevel,
2213 	.set_msglevel	= tun_set_msglevel,
2214 	.get_link	= ethtool_op_get_link,
2215 };
2216 
2217 
2218 static int __init tun_init(void)
2219 {
2220 	int ret = 0;
2221 
2222 	pr_info("%s, %s\n", DRV_DESCRIPTION, DRV_VERSION);
2223 	pr_info("%s\n", DRV_COPYRIGHT);
2224 
2225 	ret = rtnl_link_register(&tun_link_ops);
2226 	if (ret) {
2227 		pr_err("Can't register link_ops\n");
2228 		goto err_linkops;
2229 	}
2230 
2231 	ret = misc_register(&tun_miscdev);
2232 	if (ret) {
2233 		pr_err("Can't register misc device %d\n", TUN_MINOR);
2234 		goto err_misc;
2235 	}
2236 	return  0;
2237 err_misc:
2238 	rtnl_link_unregister(&tun_link_ops);
2239 err_linkops:
2240 	return ret;
2241 }
2242 
2243 static void tun_cleanup(void)
2244 {
2245 	misc_deregister(&tun_miscdev);
2246 	rtnl_link_unregister(&tun_link_ops);
2247 }
2248 
2249 /* Get an underlying socket object from tun file.  Returns error unless file is
2250  * attached to a device.  The returned object works like a packet socket, it
2251  * can be used for sock_sendmsg/sock_recvmsg.  The caller is responsible for
2252  * holding a reference to the file for as long as the socket is in use. */
2253 struct socket *tun_get_socket(struct file *file)
2254 {
2255 	struct tun_file *tfile;
2256 	if (file->f_op != &tun_fops)
2257 		return ERR_PTR(-EINVAL);
2258 	tfile = file->private_data;
2259 	if (!tfile)
2260 		return ERR_PTR(-EBADFD);
2261 	return &tfile->socket;
2262 }
2263 EXPORT_SYMBOL_GPL(tun_get_socket);
2264 
2265 module_init(tun_init);
2266 module_exit(tun_cleanup);
2267 MODULE_DESCRIPTION(DRV_DESCRIPTION);
2268 MODULE_AUTHOR(DRV_COPYRIGHT);
2269 MODULE_LICENSE("GPL");
2270 MODULE_ALIAS_MISCDEV(TUN_MINOR);
2271 MODULE_ALIAS("devname:net/tun");
2272