xref: /linux/drivers/net/tun.c (revision 31a1b26f16e822577def5402ffc79cfe4aed2db9)
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 /* DEFAULT_MAX_NUM_RSS_QUEUES were choosed to let the rx/tx queues allocated for
113  * the netdevice to be fit in one page. So we can make sure the success of
114  * memory allocation. TODO: increase the limit. */
115 #define MAX_TAP_QUEUES DEFAULT_MAX_NUM_RSS_QUEUES
116 #define MAX_TAP_FLOWS  4096
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 	void *security;
189 	u32 flow_count;
190 };
191 
192 static inline u32 tun_hashfn(u32 rxhash)
193 {
194 	return rxhash & 0x3ff;
195 }
196 
197 static struct tun_flow_entry *tun_flow_find(struct hlist_head *head, u32 rxhash)
198 {
199 	struct tun_flow_entry *e;
200 
201 	hlist_for_each_entry_rcu(e, head, hash_link) {
202 		if (e->rxhash == rxhash)
203 			return e;
204 	}
205 	return NULL;
206 }
207 
208 static struct tun_flow_entry *tun_flow_create(struct tun_struct *tun,
209 					      struct hlist_head *head,
210 					      u32 rxhash, u16 queue_index)
211 {
212 	struct tun_flow_entry *e = kmalloc(sizeof(*e), GFP_ATOMIC);
213 
214 	if (e) {
215 		tun_debug(KERN_INFO, tun, "create flow: hash %u index %u\n",
216 			  rxhash, queue_index);
217 		e->updated = jiffies;
218 		e->rxhash = rxhash;
219 		e->queue_index = queue_index;
220 		e->tun = tun;
221 		hlist_add_head_rcu(&e->hash_link, head);
222 		++tun->flow_count;
223 	}
224 	return e;
225 }
226 
227 static void tun_flow_delete(struct tun_struct *tun, struct tun_flow_entry *e)
228 {
229 	tun_debug(KERN_INFO, tun, "delete flow: hash %u index %u\n",
230 		  e->rxhash, e->queue_index);
231 	hlist_del_rcu(&e->hash_link);
232 	kfree_rcu(e, rcu);
233 	--tun->flow_count;
234 }
235 
236 static void tun_flow_flush(struct tun_struct *tun)
237 {
238 	int i;
239 
240 	spin_lock_bh(&tun->lock);
241 	for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
242 		struct tun_flow_entry *e;
243 		struct hlist_node *n;
244 
245 		hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link)
246 			tun_flow_delete(tun, e);
247 	}
248 	spin_unlock_bh(&tun->lock);
249 }
250 
251 static void tun_flow_delete_by_queue(struct tun_struct *tun, u16 queue_index)
252 {
253 	int i;
254 
255 	spin_lock_bh(&tun->lock);
256 	for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
257 		struct tun_flow_entry *e;
258 		struct hlist_node *n;
259 
260 		hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
261 			if (e->queue_index == queue_index)
262 				tun_flow_delete(tun, e);
263 		}
264 	}
265 	spin_unlock_bh(&tun->lock);
266 }
267 
268 static void tun_flow_cleanup(unsigned long data)
269 {
270 	struct tun_struct *tun = (struct tun_struct *)data;
271 	unsigned long delay = tun->ageing_time;
272 	unsigned long next_timer = jiffies + delay;
273 	unsigned long count = 0;
274 	int i;
275 
276 	tun_debug(KERN_INFO, tun, "tun_flow_cleanup\n");
277 
278 	spin_lock_bh(&tun->lock);
279 	for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
280 		struct tun_flow_entry *e;
281 		struct hlist_node *n;
282 
283 		hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
284 			unsigned long this_timer;
285 			count++;
286 			this_timer = e->updated + delay;
287 			if (time_before_eq(this_timer, jiffies))
288 				tun_flow_delete(tun, e);
289 			else if (time_before(this_timer, next_timer))
290 				next_timer = this_timer;
291 		}
292 	}
293 
294 	if (count)
295 		mod_timer(&tun->flow_gc_timer, round_jiffies_up(next_timer));
296 	spin_unlock_bh(&tun->lock);
297 }
298 
299 static void tun_flow_update(struct tun_struct *tun, u32 rxhash,
300 			    struct tun_file *tfile)
301 {
302 	struct hlist_head *head;
303 	struct tun_flow_entry *e;
304 	unsigned long delay = tun->ageing_time;
305 	u16 queue_index = tfile->queue_index;
306 
307 	if (!rxhash)
308 		return;
309 	else
310 		head = &tun->flows[tun_hashfn(rxhash)];
311 
312 	rcu_read_lock();
313 
314 	/* We may get a very small possibility of OOO during switching, not
315 	 * worth to optimize.*/
316 	if (tun->numqueues == 1 || tfile->detached)
317 		goto unlock;
318 
319 	e = tun_flow_find(head, rxhash);
320 	if (likely(e)) {
321 		/* TODO: keep queueing to old queue until it's empty? */
322 		e->queue_index = queue_index;
323 		e->updated = jiffies;
324 	} else {
325 		spin_lock_bh(&tun->lock);
326 		if (!tun_flow_find(head, rxhash) &&
327 		    tun->flow_count < MAX_TAP_FLOWS)
328 			tun_flow_create(tun, head, rxhash, queue_index);
329 
330 		if (!timer_pending(&tun->flow_gc_timer))
331 			mod_timer(&tun->flow_gc_timer,
332 				  round_jiffies_up(jiffies + delay));
333 		spin_unlock_bh(&tun->lock);
334 	}
335 
336 unlock:
337 	rcu_read_unlock();
338 }
339 
340 /* We try to identify a flow through its rxhash first. The reason that
341  * we do not check rxq no. is becuase some cards(e.g 82599), chooses
342  * the rxq based on the txq where the last packet of the flow comes. As
343  * the userspace application move between processors, we may get a
344  * different rxq no. here. If we could not get rxhash, then we would
345  * hope the rxq no. may help here.
346  */
347 static u16 tun_select_queue(struct net_device *dev, struct sk_buff *skb)
348 {
349 	struct tun_struct *tun = netdev_priv(dev);
350 	struct tun_flow_entry *e;
351 	u32 txq = 0;
352 	u32 numqueues = 0;
353 
354 	rcu_read_lock();
355 	numqueues = ACCESS_ONCE(tun->numqueues);
356 
357 	txq = skb_get_rxhash(skb);
358 	if (txq) {
359 		e = tun_flow_find(&tun->flows[tun_hashfn(txq)], txq);
360 		if (e)
361 			txq = e->queue_index;
362 		else
363 			/* use multiply and shift instead of expensive divide */
364 			txq = ((u64)txq * numqueues) >> 32;
365 	} else if (likely(skb_rx_queue_recorded(skb))) {
366 		txq = skb_get_rx_queue(skb);
367 		while (unlikely(txq >= numqueues))
368 			txq -= numqueues;
369 	}
370 
371 	rcu_read_unlock();
372 	return txq;
373 }
374 
375 static inline bool tun_not_capable(struct tun_struct *tun)
376 {
377 	const struct cred *cred = current_cred();
378 	struct net *net = dev_net(tun->dev);
379 
380 	return ((uid_valid(tun->owner) && !uid_eq(cred->euid, tun->owner)) ||
381 		  (gid_valid(tun->group) && !in_egroup_p(tun->group))) &&
382 		!ns_capable(net->user_ns, CAP_NET_ADMIN);
383 }
384 
385 static void tun_set_real_num_queues(struct tun_struct *tun)
386 {
387 	netif_set_real_num_tx_queues(tun->dev, tun->numqueues);
388 	netif_set_real_num_rx_queues(tun->dev, tun->numqueues);
389 }
390 
391 static void tun_disable_queue(struct tun_struct *tun, struct tun_file *tfile)
392 {
393 	tfile->detached = tun;
394 	list_add_tail(&tfile->next, &tun->disabled);
395 	++tun->numdisabled;
396 }
397 
398 static struct tun_struct *tun_enable_queue(struct tun_file *tfile)
399 {
400 	struct tun_struct *tun = tfile->detached;
401 
402 	tfile->detached = NULL;
403 	list_del_init(&tfile->next);
404 	--tun->numdisabled;
405 	return tun;
406 }
407 
408 static void __tun_detach(struct tun_file *tfile, bool clean)
409 {
410 	struct tun_file *ntfile;
411 	struct tun_struct *tun;
412 
413 	tun = rtnl_dereference(tfile->tun);
414 
415 	if (tun && !tfile->detached) {
416 		u16 index = tfile->queue_index;
417 		BUG_ON(index >= tun->numqueues);
418 
419 		rcu_assign_pointer(tun->tfiles[index],
420 				   tun->tfiles[tun->numqueues - 1]);
421 		ntfile = rtnl_dereference(tun->tfiles[index]);
422 		ntfile->queue_index = index;
423 
424 		--tun->numqueues;
425 		if (clean) {
426 			rcu_assign_pointer(tfile->tun, NULL);
427 			sock_put(&tfile->sk);
428 		} else
429 			tun_disable_queue(tun, tfile);
430 
431 		synchronize_net();
432 		tun_flow_delete_by_queue(tun, tun->numqueues + 1);
433 		/* Drop read queue */
434 		skb_queue_purge(&tfile->sk.sk_receive_queue);
435 		tun_set_real_num_queues(tun);
436 	} else if (tfile->detached && clean) {
437 		tun = tun_enable_queue(tfile);
438 		sock_put(&tfile->sk);
439 	}
440 
441 	if (clean) {
442 		if (tun && tun->numqueues == 0 && tun->numdisabled == 0) {
443 			netif_carrier_off(tun->dev);
444 
445 			if (!(tun->flags & TUN_PERSIST) &&
446 			    tun->dev->reg_state == NETREG_REGISTERED)
447 				unregister_netdevice(tun->dev);
448 		}
449 
450 		BUG_ON(!test_bit(SOCK_EXTERNALLY_ALLOCATED,
451 				 &tfile->socket.flags));
452 		sk_release_kernel(&tfile->sk);
453 	}
454 }
455 
456 static void tun_detach(struct tun_file *tfile, bool clean)
457 {
458 	rtnl_lock();
459 	__tun_detach(tfile, clean);
460 	rtnl_unlock();
461 }
462 
463 static void tun_detach_all(struct net_device *dev)
464 {
465 	struct tun_struct *tun = netdev_priv(dev);
466 	struct tun_file *tfile, *tmp;
467 	int i, n = tun->numqueues;
468 
469 	for (i = 0; i < n; i++) {
470 		tfile = rtnl_dereference(tun->tfiles[i]);
471 		BUG_ON(!tfile);
472 		wake_up_all(&tfile->wq.wait);
473 		rcu_assign_pointer(tfile->tun, NULL);
474 		--tun->numqueues;
475 	}
476 	list_for_each_entry(tfile, &tun->disabled, next) {
477 		wake_up_all(&tfile->wq.wait);
478 		rcu_assign_pointer(tfile->tun, NULL);
479 	}
480 	BUG_ON(tun->numqueues != 0);
481 
482 	synchronize_net();
483 	for (i = 0; i < n; i++) {
484 		tfile = rtnl_dereference(tun->tfiles[i]);
485 		/* Drop read queue */
486 		skb_queue_purge(&tfile->sk.sk_receive_queue);
487 		sock_put(&tfile->sk);
488 	}
489 	list_for_each_entry_safe(tfile, tmp, &tun->disabled, next) {
490 		tun_enable_queue(tfile);
491 		skb_queue_purge(&tfile->sk.sk_receive_queue);
492 		sock_put(&tfile->sk);
493 	}
494 	BUG_ON(tun->numdisabled != 0);
495 
496 	if (tun->flags & TUN_PERSIST)
497 		module_put(THIS_MODULE);
498 }
499 
500 static int tun_attach(struct tun_struct *tun, struct file *file)
501 {
502 	struct tun_file *tfile = file->private_data;
503 	int err;
504 
505 	err = security_tun_dev_attach(tfile->socket.sk, tun->security);
506 	if (err < 0)
507 		goto out;
508 
509 	err = -EINVAL;
510 	if (rtnl_dereference(tfile->tun) && !tfile->detached)
511 		goto out;
512 
513 	err = -EBUSY;
514 	if (!(tun->flags & TUN_TAP_MQ) && tun->numqueues == 1)
515 		goto out;
516 
517 	err = -E2BIG;
518 	if (!tfile->detached &&
519 	    tun->numqueues + tun->numdisabled == MAX_TAP_QUEUES)
520 		goto out;
521 
522 	err = 0;
523 
524 	/* Re-attach the filter to presist device */
525 	if (tun->filter_attached == true) {
526 		err = sk_attach_filter(&tun->fprog, tfile->socket.sk);
527 		if (!err)
528 			goto out;
529 	}
530 	tfile->queue_index = tun->numqueues;
531 	rcu_assign_pointer(tfile->tun, tun);
532 	rcu_assign_pointer(tun->tfiles[tun->numqueues], tfile);
533 	tun->numqueues++;
534 
535 	if (tfile->detached)
536 		tun_enable_queue(tfile);
537 	else
538 		sock_hold(&tfile->sk);
539 
540 	tun_set_real_num_queues(tun);
541 
542 	/* device is allowed to go away first, so no need to hold extra
543 	 * refcnt.
544 	 */
545 
546 out:
547 	return err;
548 }
549 
550 static struct tun_struct *__tun_get(struct tun_file *tfile)
551 {
552 	struct tun_struct *tun;
553 
554 	rcu_read_lock();
555 	tun = rcu_dereference(tfile->tun);
556 	if (tun)
557 		dev_hold(tun->dev);
558 	rcu_read_unlock();
559 
560 	return tun;
561 }
562 
563 static struct tun_struct *tun_get(struct file *file)
564 {
565 	return __tun_get(file->private_data);
566 }
567 
568 static void tun_put(struct tun_struct *tun)
569 {
570 	dev_put(tun->dev);
571 }
572 
573 /* TAP filtering */
574 static void addr_hash_set(u32 *mask, const u8 *addr)
575 {
576 	int n = ether_crc(ETH_ALEN, addr) >> 26;
577 	mask[n >> 5] |= (1 << (n & 31));
578 }
579 
580 static unsigned int addr_hash_test(const u32 *mask, const u8 *addr)
581 {
582 	int n = ether_crc(ETH_ALEN, addr) >> 26;
583 	return mask[n >> 5] & (1 << (n & 31));
584 }
585 
586 static int update_filter(struct tap_filter *filter, void __user *arg)
587 {
588 	struct { u8 u[ETH_ALEN]; } *addr;
589 	struct tun_filter uf;
590 	int err, alen, n, nexact;
591 
592 	if (copy_from_user(&uf, arg, sizeof(uf)))
593 		return -EFAULT;
594 
595 	if (!uf.count) {
596 		/* Disabled */
597 		filter->count = 0;
598 		return 0;
599 	}
600 
601 	alen = ETH_ALEN * uf.count;
602 	addr = kmalloc(alen, GFP_KERNEL);
603 	if (!addr)
604 		return -ENOMEM;
605 
606 	if (copy_from_user(addr, arg + sizeof(uf), alen)) {
607 		err = -EFAULT;
608 		goto done;
609 	}
610 
611 	/* The filter is updated without holding any locks. Which is
612 	 * perfectly safe. We disable it first and in the worst
613 	 * case we'll accept a few undesired packets. */
614 	filter->count = 0;
615 	wmb();
616 
617 	/* Use first set of addresses as an exact filter */
618 	for (n = 0; n < uf.count && n < FLT_EXACT_COUNT; n++)
619 		memcpy(filter->addr[n], addr[n].u, ETH_ALEN);
620 
621 	nexact = n;
622 
623 	/* Remaining multicast addresses are hashed,
624 	 * unicast will leave the filter disabled. */
625 	memset(filter->mask, 0, sizeof(filter->mask));
626 	for (; n < uf.count; n++) {
627 		if (!is_multicast_ether_addr(addr[n].u)) {
628 			err = 0; /* no filter */
629 			goto done;
630 		}
631 		addr_hash_set(filter->mask, addr[n].u);
632 	}
633 
634 	/* For ALLMULTI just set the mask to all ones.
635 	 * This overrides the mask populated above. */
636 	if ((uf.flags & TUN_FLT_ALLMULTI))
637 		memset(filter->mask, ~0, sizeof(filter->mask));
638 
639 	/* Now enable the filter */
640 	wmb();
641 	filter->count = nexact;
642 
643 	/* Return the number of exact filters */
644 	err = nexact;
645 
646 done:
647 	kfree(addr);
648 	return err;
649 }
650 
651 /* Returns: 0 - drop, !=0 - accept */
652 static int run_filter(struct tap_filter *filter, const struct sk_buff *skb)
653 {
654 	/* Cannot use eth_hdr(skb) here because skb_mac_hdr() is incorrect
655 	 * at this point. */
656 	struct ethhdr *eh = (struct ethhdr *) skb->data;
657 	int i;
658 
659 	/* Exact match */
660 	for (i = 0; i < filter->count; i++)
661 		if (ether_addr_equal(eh->h_dest, filter->addr[i]))
662 			return 1;
663 
664 	/* Inexact match (multicast only) */
665 	if (is_multicast_ether_addr(eh->h_dest))
666 		return addr_hash_test(filter->mask, eh->h_dest);
667 
668 	return 0;
669 }
670 
671 /*
672  * Checks whether the packet is accepted or not.
673  * Returns: 0 - drop, !=0 - accept
674  */
675 static int check_filter(struct tap_filter *filter, const struct sk_buff *skb)
676 {
677 	if (!filter->count)
678 		return 1;
679 
680 	return run_filter(filter, skb);
681 }
682 
683 /* Network device part of the driver */
684 
685 static const struct ethtool_ops tun_ethtool_ops;
686 
687 /* Net device detach from fd. */
688 static void tun_net_uninit(struct net_device *dev)
689 {
690 	tun_detach_all(dev);
691 }
692 
693 /* Net device open. */
694 static int tun_net_open(struct net_device *dev)
695 {
696 	netif_tx_start_all_queues(dev);
697 	return 0;
698 }
699 
700 /* Net device close. */
701 static int tun_net_close(struct net_device *dev)
702 {
703 	netif_tx_stop_all_queues(dev);
704 	return 0;
705 }
706 
707 /* Net device start xmit */
708 static netdev_tx_t tun_net_xmit(struct sk_buff *skb, struct net_device *dev)
709 {
710 	struct tun_struct *tun = netdev_priv(dev);
711 	int txq = skb->queue_mapping;
712 	struct tun_file *tfile;
713 
714 	rcu_read_lock();
715 	tfile = rcu_dereference(tun->tfiles[txq]);
716 
717 	/* Drop packet if interface is not attached */
718 	if (txq >= tun->numqueues)
719 		goto drop;
720 
721 	tun_debug(KERN_INFO, tun, "tun_net_xmit %d\n", skb->len);
722 
723 	BUG_ON(!tfile);
724 
725 	/* Drop if the filter does not like it.
726 	 * This is a noop if the filter is disabled.
727 	 * Filter can be enabled only for the TAP devices. */
728 	if (!check_filter(&tun->txflt, skb))
729 		goto drop;
730 
731 	if (tfile->socket.sk->sk_filter &&
732 	    sk_filter(tfile->socket.sk, skb))
733 		goto drop;
734 
735 	/* Limit the number of packets queued by dividing txq length with the
736 	 * number of queues.
737 	 */
738 	if (skb_queue_len(&tfile->socket.sk->sk_receive_queue)
739 			  >= dev->tx_queue_len / tun->numqueues)
740 		goto drop;
741 
742 	/* Orphan the skb - required as we might hang on to it
743 	 * for indefinite time. */
744 	if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC)))
745 		goto drop;
746 	skb_orphan(skb);
747 
748 	nf_reset(skb);
749 
750 	/* Enqueue packet */
751 	skb_queue_tail(&tfile->socket.sk->sk_receive_queue, skb);
752 
753 	/* Notify and wake up reader process */
754 	if (tfile->flags & TUN_FASYNC)
755 		kill_fasync(&tfile->fasync, SIGIO, POLL_IN);
756 	wake_up_interruptible_poll(&tfile->wq.wait, POLLIN |
757 				   POLLRDNORM | POLLRDBAND);
758 
759 	rcu_read_unlock();
760 	return NETDEV_TX_OK;
761 
762 drop:
763 	dev->stats.tx_dropped++;
764 	skb_tx_error(skb);
765 	kfree_skb(skb);
766 	rcu_read_unlock();
767 	return NETDEV_TX_OK;
768 }
769 
770 static void tun_net_mclist(struct net_device *dev)
771 {
772 	/*
773 	 * This callback is supposed to deal with mc filter in
774 	 * _rx_ path and has nothing to do with the _tx_ path.
775 	 * In rx path we always accept everything userspace gives us.
776 	 */
777 }
778 
779 #define MIN_MTU 68
780 #define MAX_MTU 65535
781 
782 static int
783 tun_net_change_mtu(struct net_device *dev, int new_mtu)
784 {
785 	if (new_mtu < MIN_MTU || new_mtu + dev->hard_header_len > MAX_MTU)
786 		return -EINVAL;
787 	dev->mtu = new_mtu;
788 	return 0;
789 }
790 
791 static netdev_features_t tun_net_fix_features(struct net_device *dev,
792 	netdev_features_t features)
793 {
794 	struct tun_struct *tun = netdev_priv(dev);
795 
796 	return (features & tun->set_features) | (features & ~TUN_USER_FEATURES);
797 }
798 #ifdef CONFIG_NET_POLL_CONTROLLER
799 static void tun_poll_controller(struct net_device *dev)
800 {
801 	/*
802 	 * Tun only receives frames when:
803 	 * 1) the char device endpoint gets data from user space
804 	 * 2) the tun socket gets a sendmsg call from user space
805 	 * Since both of those are syncronous operations, we are guaranteed
806 	 * never to have pending data when we poll for it
807 	 * so theres nothing to do here but return.
808 	 * We need this though so netpoll recognizes us as an interface that
809 	 * supports polling, which enables bridge devices in virt setups to
810 	 * still use netconsole
811 	 */
812 	return;
813 }
814 #endif
815 static const struct net_device_ops tun_netdev_ops = {
816 	.ndo_uninit		= tun_net_uninit,
817 	.ndo_open		= tun_net_open,
818 	.ndo_stop		= tun_net_close,
819 	.ndo_start_xmit		= tun_net_xmit,
820 	.ndo_change_mtu		= tun_net_change_mtu,
821 	.ndo_fix_features	= tun_net_fix_features,
822 	.ndo_select_queue	= tun_select_queue,
823 #ifdef CONFIG_NET_POLL_CONTROLLER
824 	.ndo_poll_controller	= tun_poll_controller,
825 #endif
826 };
827 
828 static const struct net_device_ops tap_netdev_ops = {
829 	.ndo_uninit		= tun_net_uninit,
830 	.ndo_open		= tun_net_open,
831 	.ndo_stop		= tun_net_close,
832 	.ndo_start_xmit		= tun_net_xmit,
833 	.ndo_change_mtu		= tun_net_change_mtu,
834 	.ndo_fix_features	= tun_net_fix_features,
835 	.ndo_set_rx_mode	= tun_net_mclist,
836 	.ndo_set_mac_address	= eth_mac_addr,
837 	.ndo_validate_addr	= eth_validate_addr,
838 	.ndo_select_queue	= tun_select_queue,
839 #ifdef CONFIG_NET_POLL_CONTROLLER
840 	.ndo_poll_controller	= tun_poll_controller,
841 #endif
842 };
843 
844 static void tun_flow_init(struct tun_struct *tun)
845 {
846 	int i;
847 
848 	for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++)
849 		INIT_HLIST_HEAD(&tun->flows[i]);
850 
851 	tun->ageing_time = TUN_FLOW_EXPIRE;
852 	setup_timer(&tun->flow_gc_timer, tun_flow_cleanup, (unsigned long)tun);
853 	mod_timer(&tun->flow_gc_timer,
854 		  round_jiffies_up(jiffies + tun->ageing_time));
855 }
856 
857 static void tun_flow_uninit(struct tun_struct *tun)
858 {
859 	del_timer_sync(&tun->flow_gc_timer);
860 	tun_flow_flush(tun);
861 }
862 
863 /* Initialize net device. */
864 static void tun_net_init(struct net_device *dev)
865 {
866 	struct tun_struct *tun = netdev_priv(dev);
867 
868 	switch (tun->flags & TUN_TYPE_MASK) {
869 	case TUN_TUN_DEV:
870 		dev->netdev_ops = &tun_netdev_ops;
871 
872 		/* Point-to-Point TUN Device */
873 		dev->hard_header_len = 0;
874 		dev->addr_len = 0;
875 		dev->mtu = 1500;
876 
877 		/* Zero header length */
878 		dev->type = ARPHRD_NONE;
879 		dev->flags = IFF_POINTOPOINT | IFF_NOARP | IFF_MULTICAST;
880 		dev->tx_queue_len = TUN_READQ_SIZE;  /* We prefer our own queue length */
881 		break;
882 
883 	case TUN_TAP_DEV:
884 		dev->netdev_ops = &tap_netdev_ops;
885 		/* Ethernet TAP Device */
886 		ether_setup(dev);
887 		dev->priv_flags &= ~IFF_TX_SKB_SHARING;
888 		dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
889 
890 		eth_hw_addr_random(dev);
891 
892 		dev->tx_queue_len = TUN_READQ_SIZE;  /* We prefer our own queue length */
893 		break;
894 	}
895 }
896 
897 /* Character device part */
898 
899 /* Poll */
900 static unsigned int tun_chr_poll(struct file *file, poll_table *wait)
901 {
902 	struct tun_file *tfile = file->private_data;
903 	struct tun_struct *tun = __tun_get(tfile);
904 	struct sock *sk;
905 	unsigned int mask = 0;
906 
907 	if (!tun)
908 		return POLLERR;
909 
910 	sk = tfile->socket.sk;
911 
912 	tun_debug(KERN_INFO, tun, "tun_chr_poll\n");
913 
914 	poll_wait(file, &tfile->wq.wait, wait);
915 
916 	if (!skb_queue_empty(&sk->sk_receive_queue))
917 		mask |= POLLIN | POLLRDNORM;
918 
919 	if (sock_writeable(sk) ||
920 	    (!test_and_set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags) &&
921 	     sock_writeable(sk)))
922 		mask |= POLLOUT | POLLWRNORM;
923 
924 	if (tun->dev->reg_state != NETREG_REGISTERED)
925 		mask = POLLERR;
926 
927 	tun_put(tun);
928 	return mask;
929 }
930 
931 /* prepad is the amount to reserve at front.  len is length after that.
932  * linear is a hint as to how much to copy (usually headers). */
933 static struct sk_buff *tun_alloc_skb(struct tun_file *tfile,
934 				     size_t prepad, size_t len,
935 				     size_t linear, int noblock)
936 {
937 	struct sock *sk = tfile->socket.sk;
938 	struct sk_buff *skb;
939 	int err;
940 
941 	/* Under a page?  Don't bother with paged skb. */
942 	if (prepad + len < PAGE_SIZE || !linear)
943 		linear = len;
944 
945 	skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
946 				   &err);
947 	if (!skb)
948 		return ERR_PTR(err);
949 
950 	skb_reserve(skb, prepad);
951 	skb_put(skb, linear);
952 	skb->data_len = len - linear;
953 	skb->len += len - linear;
954 
955 	return skb;
956 }
957 
958 /* set skb frags from iovec, this can move to core network code for reuse */
959 static int zerocopy_sg_from_iovec(struct sk_buff *skb, const struct iovec *from,
960 				  int offset, size_t count)
961 {
962 	int len = iov_length(from, count) - offset;
963 	int copy = skb_headlen(skb);
964 	int size, offset1 = 0;
965 	int i = 0;
966 
967 	/* Skip over from offset */
968 	while (count && (offset >= from->iov_len)) {
969 		offset -= from->iov_len;
970 		++from;
971 		--count;
972 	}
973 
974 	/* copy up to skb headlen */
975 	while (count && (copy > 0)) {
976 		size = min_t(unsigned int, copy, from->iov_len - offset);
977 		if (copy_from_user(skb->data + offset1, from->iov_base + offset,
978 				   size))
979 			return -EFAULT;
980 		if (copy > size) {
981 			++from;
982 			--count;
983 			offset = 0;
984 		} else
985 			offset += size;
986 		copy -= size;
987 		offset1 += size;
988 	}
989 
990 	if (len == offset1)
991 		return 0;
992 
993 	while (count--) {
994 		struct page *page[MAX_SKB_FRAGS];
995 		int num_pages;
996 		unsigned long base;
997 		unsigned long truesize;
998 
999 		len = from->iov_len - offset;
1000 		if (!len) {
1001 			offset = 0;
1002 			++from;
1003 			continue;
1004 		}
1005 		base = (unsigned long)from->iov_base + offset;
1006 		size = ((base & ~PAGE_MASK) + len + ~PAGE_MASK) >> PAGE_SHIFT;
1007 		if (i + size > MAX_SKB_FRAGS)
1008 			return -EMSGSIZE;
1009 		num_pages = get_user_pages_fast(base, size, 0, &page[i]);
1010 		if (num_pages != size) {
1011 			int j;
1012 
1013 			for (j = 0; j < num_pages; j++)
1014 				put_page(page[i + j]);
1015 			return -EFAULT;
1016 		}
1017 		truesize = size * PAGE_SIZE;
1018 		skb->data_len += len;
1019 		skb->len += len;
1020 		skb->truesize += truesize;
1021 		atomic_add(truesize, &skb->sk->sk_wmem_alloc);
1022 		while (len) {
1023 			int off = base & ~PAGE_MASK;
1024 			int size = min_t(int, len, PAGE_SIZE - off);
1025 			__skb_fill_page_desc(skb, i, page[i], off, size);
1026 			skb_shinfo(skb)->nr_frags++;
1027 			/* increase sk_wmem_alloc */
1028 			base += size;
1029 			len -= size;
1030 			i++;
1031 		}
1032 		offset = 0;
1033 		++from;
1034 	}
1035 	return 0;
1036 }
1037 
1038 static unsigned long iov_pages(const struct iovec *iv, int offset,
1039 			       unsigned long nr_segs)
1040 {
1041 	unsigned long seg, base;
1042 	int pages = 0, len, size;
1043 
1044 	while (nr_segs && (offset >= iv->iov_len)) {
1045 		offset -= iv->iov_len;
1046 		++iv;
1047 		--nr_segs;
1048 	}
1049 
1050 	for (seg = 0; seg < nr_segs; seg++) {
1051 		base = (unsigned long)iv[seg].iov_base + offset;
1052 		len = iv[seg].iov_len - offset;
1053 		size = ((base & ~PAGE_MASK) + len + ~PAGE_MASK) >> PAGE_SHIFT;
1054 		pages += size;
1055 		offset = 0;
1056 	}
1057 
1058 	return pages;
1059 }
1060 
1061 /* Get packet from user space buffer */
1062 static ssize_t tun_get_user(struct tun_struct *tun, struct tun_file *tfile,
1063 			    void *msg_control, const struct iovec *iv,
1064 			    size_t total_len, size_t count, int noblock)
1065 {
1066 	struct tun_pi pi = { 0, cpu_to_be16(ETH_P_IP) };
1067 	struct sk_buff *skb;
1068 	size_t len = total_len, align = NET_SKB_PAD, linear;
1069 	struct virtio_net_hdr gso = { 0 };
1070 	int offset = 0;
1071 	int copylen;
1072 	bool zerocopy = false;
1073 	int err;
1074 	u32 rxhash;
1075 
1076 	if (!(tun->flags & TUN_NO_PI)) {
1077 		if ((len -= sizeof(pi)) > total_len)
1078 			return -EINVAL;
1079 
1080 		if (memcpy_fromiovecend((void *)&pi, iv, 0, sizeof(pi)))
1081 			return -EFAULT;
1082 		offset += sizeof(pi);
1083 	}
1084 
1085 	if (tun->flags & TUN_VNET_HDR) {
1086 		if ((len -= tun->vnet_hdr_sz) > total_len)
1087 			return -EINVAL;
1088 
1089 		if (memcpy_fromiovecend((void *)&gso, iv, offset, sizeof(gso)))
1090 			return -EFAULT;
1091 
1092 		if ((gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
1093 		    gso.csum_start + gso.csum_offset + 2 > gso.hdr_len)
1094 			gso.hdr_len = gso.csum_start + gso.csum_offset + 2;
1095 
1096 		if (gso.hdr_len > len)
1097 			return -EINVAL;
1098 		offset += tun->vnet_hdr_sz;
1099 	}
1100 
1101 	if ((tun->flags & TUN_TYPE_MASK) == TUN_TAP_DEV) {
1102 		align += NET_IP_ALIGN;
1103 		if (unlikely(len < ETH_HLEN ||
1104 			     (gso.hdr_len && gso.hdr_len < ETH_HLEN)))
1105 			return -EINVAL;
1106 	}
1107 
1108 	if (msg_control) {
1109 		/* There are 256 bytes to be copied in skb, so there is
1110 		 * enough room for skb expand head in case it is used.
1111 		 * The rest of the buffer is mapped from userspace.
1112 		 */
1113 		copylen = gso.hdr_len ? gso.hdr_len : GOODCOPY_LEN;
1114 		linear = copylen;
1115 		if (iov_pages(iv, offset + copylen, count) <= MAX_SKB_FRAGS)
1116 			zerocopy = true;
1117 	}
1118 
1119 	if (!zerocopy) {
1120 		copylen = len;
1121 		linear = gso.hdr_len;
1122 	}
1123 
1124 	skb = tun_alloc_skb(tfile, align, copylen, linear, noblock);
1125 	if (IS_ERR(skb)) {
1126 		if (PTR_ERR(skb) != -EAGAIN)
1127 			tun->dev->stats.rx_dropped++;
1128 		return PTR_ERR(skb);
1129 	}
1130 
1131 	if (zerocopy)
1132 		err = zerocopy_sg_from_iovec(skb, iv, offset, count);
1133 	else {
1134 		err = skb_copy_datagram_from_iovec(skb, 0, iv, offset, len);
1135 		if (!err && msg_control) {
1136 			struct ubuf_info *uarg = msg_control;
1137 			uarg->callback(uarg, false);
1138 		}
1139 	}
1140 
1141 	if (err) {
1142 		tun->dev->stats.rx_dropped++;
1143 		kfree_skb(skb);
1144 		return -EFAULT;
1145 	}
1146 
1147 	if (gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
1148 		if (!skb_partial_csum_set(skb, gso.csum_start,
1149 					  gso.csum_offset)) {
1150 			tun->dev->stats.rx_frame_errors++;
1151 			kfree_skb(skb);
1152 			return -EINVAL;
1153 		}
1154 	}
1155 
1156 	switch (tun->flags & TUN_TYPE_MASK) {
1157 	case TUN_TUN_DEV:
1158 		if (tun->flags & TUN_NO_PI) {
1159 			switch (skb->data[0] & 0xf0) {
1160 			case 0x40:
1161 				pi.proto = htons(ETH_P_IP);
1162 				break;
1163 			case 0x60:
1164 				pi.proto = htons(ETH_P_IPV6);
1165 				break;
1166 			default:
1167 				tun->dev->stats.rx_dropped++;
1168 				kfree_skb(skb);
1169 				return -EINVAL;
1170 			}
1171 		}
1172 
1173 		skb_reset_mac_header(skb);
1174 		skb->protocol = pi.proto;
1175 		skb->dev = tun->dev;
1176 		break;
1177 	case TUN_TAP_DEV:
1178 		skb->protocol = eth_type_trans(skb, tun->dev);
1179 		break;
1180 	}
1181 
1182 	if (gso.gso_type != VIRTIO_NET_HDR_GSO_NONE) {
1183 		pr_debug("GSO!\n");
1184 		switch (gso.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
1185 		case VIRTIO_NET_HDR_GSO_TCPV4:
1186 			skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
1187 			break;
1188 		case VIRTIO_NET_HDR_GSO_TCPV6:
1189 			skb_shinfo(skb)->gso_type = SKB_GSO_TCPV6;
1190 			break;
1191 		case VIRTIO_NET_HDR_GSO_UDP:
1192 			skb_shinfo(skb)->gso_type = SKB_GSO_UDP;
1193 			break;
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 = 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_reset_network_header(skb);
1223 	skb_probe_transport_header(skb, 0);
1224 
1225 	rxhash = skb_get_rxhash(skb);
1226 	netif_rx_ni(skb);
1227 
1228 	tun->dev->stats.rx_packets++;
1229 	tun->dev->stats.rx_bytes += len;
1230 
1231 	tun_flow_update(tun, rxhash, tfile);
1232 	return total_len;
1233 }
1234 
1235 static ssize_t tun_chr_aio_write(struct kiocb *iocb, const struct iovec *iv,
1236 			      unsigned long count, loff_t pos)
1237 {
1238 	struct file *file = iocb->ki_filp;
1239 	struct tun_struct *tun = tun_get(file);
1240 	struct tun_file *tfile = file->private_data;
1241 	ssize_t result;
1242 
1243 	if (!tun)
1244 		return -EBADFD;
1245 
1246 	tun_debug(KERN_INFO, tun, "tun_chr_write %ld\n", count);
1247 
1248 	result = tun_get_user(tun, tfile, NULL, iv, iov_length(iv, count),
1249 			      count, file->f_flags & O_NONBLOCK);
1250 
1251 	tun_put(tun);
1252 	return result;
1253 }
1254 
1255 /* Put packet to the user space buffer */
1256 static ssize_t tun_put_user(struct tun_struct *tun,
1257 			    struct tun_file *tfile,
1258 			    struct sk_buff *skb,
1259 			    const struct iovec *iv, int len)
1260 {
1261 	struct tun_pi pi = { 0, skb->protocol };
1262 	ssize_t total = 0;
1263 
1264 	if (!(tun->flags & TUN_NO_PI)) {
1265 		if ((len -= sizeof(pi)) < 0)
1266 			return -EINVAL;
1267 
1268 		if (len < skb->len) {
1269 			/* Packet will be striped */
1270 			pi.flags |= TUN_PKT_STRIP;
1271 		}
1272 
1273 		if (memcpy_toiovecend(iv, (void *) &pi, 0, sizeof(pi)))
1274 			return -EFAULT;
1275 		total += sizeof(pi);
1276 	}
1277 
1278 	if (tun->flags & TUN_VNET_HDR) {
1279 		struct virtio_net_hdr gso = { 0 }; /* no info leak */
1280 		if ((len -= tun->vnet_hdr_sz) < 0)
1281 			return -EINVAL;
1282 
1283 		if (skb_is_gso(skb)) {
1284 			struct skb_shared_info *sinfo = skb_shinfo(skb);
1285 
1286 			/* This is a hint as to how much should be linear. */
1287 			gso.hdr_len = skb_headlen(skb);
1288 			gso.gso_size = sinfo->gso_size;
1289 			if (sinfo->gso_type & SKB_GSO_TCPV4)
1290 				gso.gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
1291 			else if (sinfo->gso_type & SKB_GSO_TCPV6)
1292 				gso.gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
1293 			else if (sinfo->gso_type & SKB_GSO_UDP)
1294 				gso.gso_type = VIRTIO_NET_HDR_GSO_UDP;
1295 			else {
1296 				pr_err("unexpected GSO type: "
1297 				       "0x%x, gso_size %d, hdr_len %d\n",
1298 				       sinfo->gso_type, gso.gso_size,
1299 				       gso.hdr_len);
1300 				print_hex_dump(KERN_ERR, "tun: ",
1301 					       DUMP_PREFIX_NONE,
1302 					       16, 1, skb->head,
1303 					       min((int)gso.hdr_len, 64), true);
1304 				WARN_ON_ONCE(1);
1305 				return -EINVAL;
1306 			}
1307 			if (sinfo->gso_type & SKB_GSO_TCP_ECN)
1308 				gso.gso_type |= VIRTIO_NET_HDR_GSO_ECN;
1309 		} else
1310 			gso.gso_type = VIRTIO_NET_HDR_GSO_NONE;
1311 
1312 		if (skb->ip_summed == CHECKSUM_PARTIAL) {
1313 			gso.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
1314 			gso.csum_start = skb_checksum_start_offset(skb);
1315 			gso.csum_offset = skb->csum_offset;
1316 		} else if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
1317 			gso.flags = VIRTIO_NET_HDR_F_DATA_VALID;
1318 		} /* else everything is zero */
1319 
1320 		if (unlikely(memcpy_toiovecend(iv, (void *)&gso, total,
1321 					       sizeof(gso))))
1322 			return -EFAULT;
1323 		total += tun->vnet_hdr_sz;
1324 	}
1325 
1326 	len = min_t(int, skb->len, len);
1327 
1328 	skb_copy_datagram_const_iovec(skb, 0, iv, total, len);
1329 	total += skb->len;
1330 
1331 	tun->dev->stats.tx_packets++;
1332 	tun->dev->stats.tx_bytes += len;
1333 
1334 	return total;
1335 }
1336 
1337 static ssize_t tun_do_read(struct tun_struct *tun, struct tun_file *tfile,
1338 			   struct kiocb *iocb, const struct iovec *iv,
1339 			   ssize_t len, int noblock)
1340 {
1341 	DECLARE_WAITQUEUE(wait, current);
1342 	struct sk_buff *skb;
1343 	ssize_t ret = 0;
1344 
1345 	tun_debug(KERN_INFO, tun, "tun_do_read\n");
1346 
1347 	if (unlikely(!noblock))
1348 		add_wait_queue(&tfile->wq.wait, &wait);
1349 	while (len) {
1350 		current->state = TASK_INTERRUPTIBLE;
1351 
1352 		/* Read frames from the queue */
1353 		if (!(skb = skb_dequeue(&tfile->socket.sk->sk_receive_queue))) {
1354 			if (noblock) {
1355 				ret = -EAGAIN;
1356 				break;
1357 			}
1358 			if (signal_pending(current)) {
1359 				ret = -ERESTARTSYS;
1360 				break;
1361 			}
1362 			if (tun->dev->reg_state != NETREG_REGISTERED) {
1363 				ret = -EIO;
1364 				break;
1365 			}
1366 
1367 			/* Nothing to read, let's sleep */
1368 			schedule();
1369 			continue;
1370 		}
1371 
1372 		ret = tun_put_user(tun, tfile, skb, iv, len);
1373 		kfree_skb(skb);
1374 		break;
1375 	}
1376 
1377 	current->state = TASK_RUNNING;
1378 	if (unlikely(!noblock))
1379 		remove_wait_queue(&tfile->wq.wait, &wait);
1380 
1381 	return ret;
1382 }
1383 
1384 static ssize_t tun_chr_aio_read(struct kiocb *iocb, const struct iovec *iv,
1385 			    unsigned long count, loff_t pos)
1386 {
1387 	struct file *file = iocb->ki_filp;
1388 	struct tun_file *tfile = file->private_data;
1389 	struct tun_struct *tun = __tun_get(tfile);
1390 	ssize_t len, ret;
1391 
1392 	if (!tun)
1393 		return -EBADFD;
1394 	len = iov_length(iv, count);
1395 	if (len < 0) {
1396 		ret = -EINVAL;
1397 		goto out;
1398 	}
1399 
1400 	ret = tun_do_read(tun, tfile, iocb, iv, len,
1401 			  file->f_flags & O_NONBLOCK);
1402 	ret = min_t(ssize_t, ret, len);
1403 out:
1404 	tun_put(tun);
1405 	return ret;
1406 }
1407 
1408 static void tun_free_netdev(struct net_device *dev)
1409 {
1410 	struct tun_struct *tun = netdev_priv(dev);
1411 
1412 	BUG_ON(!(list_empty(&tun->disabled)));
1413 	tun_flow_uninit(tun);
1414 	security_tun_dev_free_security(tun->security);
1415 	free_netdev(dev);
1416 }
1417 
1418 static void tun_setup(struct net_device *dev)
1419 {
1420 	struct tun_struct *tun = netdev_priv(dev);
1421 
1422 	tun->owner = INVALID_UID;
1423 	tun->group = INVALID_GID;
1424 
1425 	dev->ethtool_ops = &tun_ethtool_ops;
1426 	dev->destructor = tun_free_netdev;
1427 }
1428 
1429 /* Trivial set of netlink ops to allow deleting tun or tap
1430  * device with netlink.
1431  */
1432 static int tun_validate(struct nlattr *tb[], struct nlattr *data[])
1433 {
1434 	return -EINVAL;
1435 }
1436 
1437 static struct rtnl_link_ops tun_link_ops __read_mostly = {
1438 	.kind		= DRV_NAME,
1439 	.priv_size	= sizeof(struct tun_struct),
1440 	.setup		= tun_setup,
1441 	.validate	= tun_validate,
1442 };
1443 
1444 static void tun_sock_write_space(struct sock *sk)
1445 {
1446 	struct tun_file *tfile;
1447 	wait_queue_head_t *wqueue;
1448 
1449 	if (!sock_writeable(sk))
1450 		return;
1451 
1452 	if (!test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags))
1453 		return;
1454 
1455 	wqueue = sk_sleep(sk);
1456 	if (wqueue && waitqueue_active(wqueue))
1457 		wake_up_interruptible_sync_poll(wqueue, POLLOUT |
1458 						POLLWRNORM | POLLWRBAND);
1459 
1460 	tfile = container_of(sk, struct tun_file, sk);
1461 	kill_fasync(&tfile->fasync, SIGIO, POLL_OUT);
1462 }
1463 
1464 static int tun_sendmsg(struct kiocb *iocb, struct socket *sock,
1465 		       struct msghdr *m, size_t total_len)
1466 {
1467 	int ret;
1468 	struct tun_file *tfile = container_of(sock, struct tun_file, socket);
1469 	struct tun_struct *tun = __tun_get(tfile);
1470 
1471 	if (!tun)
1472 		return -EBADFD;
1473 	ret = tun_get_user(tun, tfile, m->msg_control, m->msg_iov, total_len,
1474 			   m->msg_iovlen, m->msg_flags & MSG_DONTWAIT);
1475 	tun_put(tun);
1476 	return ret;
1477 }
1478 
1479 
1480 static int tun_recvmsg(struct kiocb *iocb, struct socket *sock,
1481 		       struct msghdr *m, size_t total_len,
1482 		       int flags)
1483 {
1484 	struct tun_file *tfile = container_of(sock, struct tun_file, socket);
1485 	struct tun_struct *tun = __tun_get(tfile);
1486 	int ret;
1487 
1488 	if (!tun)
1489 		return -EBADFD;
1490 
1491 	if (flags & ~(MSG_DONTWAIT|MSG_TRUNC)) {
1492 		ret = -EINVAL;
1493 		goto out;
1494 	}
1495 	ret = tun_do_read(tun, tfile, iocb, m->msg_iov, total_len,
1496 			  flags & MSG_DONTWAIT);
1497 	if (ret > total_len) {
1498 		m->msg_flags |= MSG_TRUNC;
1499 		ret = flags & MSG_TRUNC ? ret : total_len;
1500 	}
1501 out:
1502 	tun_put(tun);
1503 	return ret;
1504 }
1505 
1506 static int tun_release(struct socket *sock)
1507 {
1508 	if (sock->sk)
1509 		sock_put(sock->sk);
1510 	return 0;
1511 }
1512 
1513 /* Ops structure to mimic raw sockets with tun */
1514 static const struct proto_ops tun_socket_ops = {
1515 	.sendmsg = tun_sendmsg,
1516 	.recvmsg = tun_recvmsg,
1517 	.release = tun_release,
1518 };
1519 
1520 static struct proto tun_proto = {
1521 	.name		= "tun",
1522 	.owner		= THIS_MODULE,
1523 	.obj_size	= sizeof(struct tun_file),
1524 };
1525 
1526 static int tun_flags(struct tun_struct *tun)
1527 {
1528 	int flags = 0;
1529 
1530 	if (tun->flags & TUN_TUN_DEV)
1531 		flags |= IFF_TUN;
1532 	else
1533 		flags |= IFF_TAP;
1534 
1535 	if (tun->flags & TUN_NO_PI)
1536 		flags |= IFF_NO_PI;
1537 
1538 	/* This flag has no real effect.  We track the value for backwards
1539 	 * compatibility.
1540 	 */
1541 	if (tun->flags & TUN_ONE_QUEUE)
1542 		flags |= IFF_ONE_QUEUE;
1543 
1544 	if (tun->flags & TUN_VNET_HDR)
1545 		flags |= IFF_VNET_HDR;
1546 
1547 	if (tun->flags & TUN_TAP_MQ)
1548 		flags |= IFF_MULTI_QUEUE;
1549 
1550 	if (tun->flags & TUN_PERSIST)
1551 		flags |= IFF_PERSIST;
1552 
1553 	return flags;
1554 }
1555 
1556 static ssize_t tun_show_flags(struct device *dev, struct device_attribute *attr,
1557 			      char *buf)
1558 {
1559 	struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1560 	return sprintf(buf, "0x%x\n", tun_flags(tun));
1561 }
1562 
1563 static ssize_t tun_show_owner(struct device *dev, struct device_attribute *attr,
1564 			      char *buf)
1565 {
1566 	struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1567 	return uid_valid(tun->owner)?
1568 		sprintf(buf, "%u\n",
1569 			from_kuid_munged(current_user_ns(), tun->owner)):
1570 		sprintf(buf, "-1\n");
1571 }
1572 
1573 static ssize_t tun_show_group(struct device *dev, struct device_attribute *attr,
1574 			      char *buf)
1575 {
1576 	struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1577 	return gid_valid(tun->group) ?
1578 		sprintf(buf, "%u\n",
1579 			from_kgid_munged(current_user_ns(), tun->group)):
1580 		sprintf(buf, "-1\n");
1581 }
1582 
1583 static DEVICE_ATTR(tun_flags, 0444, tun_show_flags, NULL);
1584 static DEVICE_ATTR(owner, 0444, tun_show_owner, NULL);
1585 static DEVICE_ATTR(group, 0444, tun_show_group, NULL);
1586 
1587 static int tun_set_iff(struct net *net, struct file *file, struct ifreq *ifr)
1588 {
1589 	struct tun_struct *tun;
1590 	struct tun_file *tfile = file->private_data;
1591 	struct net_device *dev;
1592 	int err;
1593 
1594 	if (tfile->detached)
1595 		return -EINVAL;
1596 
1597 	dev = __dev_get_by_name(net, ifr->ifr_name);
1598 	if (dev) {
1599 		if (ifr->ifr_flags & IFF_TUN_EXCL)
1600 			return -EBUSY;
1601 		if ((ifr->ifr_flags & IFF_TUN) && dev->netdev_ops == &tun_netdev_ops)
1602 			tun = netdev_priv(dev);
1603 		else if ((ifr->ifr_flags & IFF_TAP) && dev->netdev_ops == &tap_netdev_ops)
1604 			tun = netdev_priv(dev);
1605 		else
1606 			return -EINVAL;
1607 
1608 		if (!!(ifr->ifr_flags & IFF_MULTI_QUEUE) !=
1609 		    !!(tun->flags & TUN_TAP_MQ))
1610 			return -EINVAL;
1611 
1612 		if (tun_not_capable(tun))
1613 			return -EPERM;
1614 		err = security_tun_dev_open(tun->security);
1615 		if (err < 0)
1616 			return err;
1617 
1618 		err = tun_attach(tun, file);
1619 		if (err < 0)
1620 			return err;
1621 
1622 		if (tun->flags & TUN_TAP_MQ &&
1623 		    (tun->numqueues + tun->numdisabled > 1)) {
1624 			/* One or more queue has already been attached, no need
1625 			 * to initialize the device again.
1626 			 */
1627 			return 0;
1628 		}
1629 	}
1630 	else {
1631 		char *name;
1632 		unsigned long flags = 0;
1633 		int queues = ifr->ifr_flags & IFF_MULTI_QUEUE ?
1634 			     MAX_TAP_QUEUES : 1;
1635 
1636 		if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
1637 			return -EPERM;
1638 		err = security_tun_dev_create();
1639 		if (err < 0)
1640 			return err;
1641 
1642 		/* Set dev type */
1643 		if (ifr->ifr_flags & IFF_TUN) {
1644 			/* TUN device */
1645 			flags |= TUN_TUN_DEV;
1646 			name = "tun%d";
1647 		} else if (ifr->ifr_flags & IFF_TAP) {
1648 			/* TAP device */
1649 			flags |= TUN_TAP_DEV;
1650 			name = "tap%d";
1651 		} else
1652 			return -EINVAL;
1653 
1654 		if (*ifr->ifr_name)
1655 			name = ifr->ifr_name;
1656 
1657 		dev = alloc_netdev_mqs(sizeof(struct tun_struct), name,
1658 				       tun_setup, queues, queues);
1659 
1660 		if (!dev)
1661 			return -ENOMEM;
1662 
1663 		dev_net_set(dev, net);
1664 		dev->rtnl_link_ops = &tun_link_ops;
1665 
1666 		tun = netdev_priv(dev);
1667 		tun->dev = dev;
1668 		tun->flags = flags;
1669 		tun->txflt.count = 0;
1670 		tun->vnet_hdr_sz = sizeof(struct virtio_net_hdr);
1671 
1672 		tun->filter_attached = false;
1673 		tun->sndbuf = tfile->socket.sk->sk_sndbuf;
1674 
1675 		spin_lock_init(&tun->lock);
1676 
1677 		err = security_tun_dev_alloc_security(&tun->security);
1678 		if (err < 0)
1679 			goto err_free_dev;
1680 
1681 		tun_net_init(dev);
1682 		tun_flow_init(tun);
1683 
1684 		dev->hw_features = NETIF_F_SG | NETIF_F_FRAGLIST |
1685 			TUN_USER_FEATURES;
1686 		dev->features = dev->hw_features;
1687 		dev->vlan_features = dev->features;
1688 
1689 		INIT_LIST_HEAD(&tun->disabled);
1690 		err = tun_attach(tun, file);
1691 		if (err < 0)
1692 			goto err_free_dev;
1693 
1694 		err = register_netdevice(tun->dev);
1695 		if (err < 0)
1696 			goto err_free_dev;
1697 
1698 		if (device_create_file(&tun->dev->dev, &dev_attr_tun_flags) ||
1699 		    device_create_file(&tun->dev->dev, &dev_attr_owner) ||
1700 		    device_create_file(&tun->dev->dev, &dev_attr_group))
1701 			pr_err("Failed to create tun sysfs files\n");
1702 	}
1703 
1704 	netif_carrier_on(tun->dev);
1705 
1706 	tun_debug(KERN_INFO, tun, "tun_set_iff\n");
1707 
1708 	if (ifr->ifr_flags & IFF_NO_PI)
1709 		tun->flags |= TUN_NO_PI;
1710 	else
1711 		tun->flags &= ~TUN_NO_PI;
1712 
1713 	/* This flag has no real effect.  We track the value for backwards
1714 	 * compatibility.
1715 	 */
1716 	if (ifr->ifr_flags & IFF_ONE_QUEUE)
1717 		tun->flags |= TUN_ONE_QUEUE;
1718 	else
1719 		tun->flags &= ~TUN_ONE_QUEUE;
1720 
1721 	if (ifr->ifr_flags & IFF_VNET_HDR)
1722 		tun->flags |= TUN_VNET_HDR;
1723 	else
1724 		tun->flags &= ~TUN_VNET_HDR;
1725 
1726 	if (ifr->ifr_flags & IFF_MULTI_QUEUE)
1727 		tun->flags |= TUN_TAP_MQ;
1728 	else
1729 		tun->flags &= ~TUN_TAP_MQ;
1730 
1731 	/* Make sure persistent devices do not get stuck in
1732 	 * xoff state.
1733 	 */
1734 	if (netif_running(tun->dev))
1735 		netif_tx_wake_all_queues(tun->dev);
1736 
1737 	strcpy(ifr->ifr_name, tun->dev->name);
1738 	return 0;
1739 
1740  err_free_dev:
1741 	free_netdev(dev);
1742 	return err;
1743 }
1744 
1745 static void tun_get_iff(struct net *net, struct tun_struct *tun,
1746 		       struct ifreq *ifr)
1747 {
1748 	tun_debug(KERN_INFO, tun, "tun_get_iff\n");
1749 
1750 	strcpy(ifr->ifr_name, tun->dev->name);
1751 
1752 	ifr->ifr_flags = tun_flags(tun);
1753 
1754 }
1755 
1756 /* This is like a cut-down ethtool ops, except done via tun fd so no
1757  * privs required. */
1758 static int set_offload(struct tun_struct *tun, unsigned long arg)
1759 {
1760 	netdev_features_t features = 0;
1761 
1762 	if (arg & TUN_F_CSUM) {
1763 		features |= NETIF_F_HW_CSUM;
1764 		arg &= ~TUN_F_CSUM;
1765 
1766 		if (arg & (TUN_F_TSO4|TUN_F_TSO6)) {
1767 			if (arg & TUN_F_TSO_ECN) {
1768 				features |= NETIF_F_TSO_ECN;
1769 				arg &= ~TUN_F_TSO_ECN;
1770 			}
1771 			if (arg & TUN_F_TSO4)
1772 				features |= NETIF_F_TSO;
1773 			if (arg & TUN_F_TSO6)
1774 				features |= NETIF_F_TSO6;
1775 			arg &= ~(TUN_F_TSO4|TUN_F_TSO6);
1776 		}
1777 
1778 		if (arg & TUN_F_UFO) {
1779 			features |= NETIF_F_UFO;
1780 			arg &= ~TUN_F_UFO;
1781 		}
1782 	}
1783 
1784 	/* This gives the user a way to test for new features in future by
1785 	 * trying to set them. */
1786 	if (arg)
1787 		return -EINVAL;
1788 
1789 	tun->set_features = features;
1790 	netdev_update_features(tun->dev);
1791 
1792 	return 0;
1793 }
1794 
1795 static void tun_detach_filter(struct tun_struct *tun, int n)
1796 {
1797 	int i;
1798 	struct tun_file *tfile;
1799 
1800 	for (i = 0; i < n; i++) {
1801 		tfile = rtnl_dereference(tun->tfiles[i]);
1802 		sk_detach_filter(tfile->socket.sk);
1803 	}
1804 
1805 	tun->filter_attached = false;
1806 }
1807 
1808 static int tun_attach_filter(struct tun_struct *tun)
1809 {
1810 	int i, ret = 0;
1811 	struct tun_file *tfile;
1812 
1813 	for (i = 0; i < tun->numqueues; i++) {
1814 		tfile = rtnl_dereference(tun->tfiles[i]);
1815 		ret = sk_attach_filter(&tun->fprog, tfile->socket.sk);
1816 		if (ret) {
1817 			tun_detach_filter(tun, i);
1818 			return ret;
1819 		}
1820 	}
1821 
1822 	tun->filter_attached = true;
1823 	return ret;
1824 }
1825 
1826 static void tun_set_sndbuf(struct tun_struct *tun)
1827 {
1828 	struct tun_file *tfile;
1829 	int i;
1830 
1831 	for (i = 0; i < tun->numqueues; i++) {
1832 		tfile = rtnl_dereference(tun->tfiles[i]);
1833 		tfile->socket.sk->sk_sndbuf = tun->sndbuf;
1834 	}
1835 }
1836 
1837 static int tun_set_queue(struct file *file, struct ifreq *ifr)
1838 {
1839 	struct tun_file *tfile = file->private_data;
1840 	struct tun_struct *tun;
1841 	int ret = 0;
1842 
1843 	rtnl_lock();
1844 
1845 	if (ifr->ifr_flags & IFF_ATTACH_QUEUE) {
1846 		tun = tfile->detached;
1847 		if (!tun) {
1848 			ret = -EINVAL;
1849 			goto unlock;
1850 		}
1851 		ret = security_tun_dev_attach_queue(tun->security);
1852 		if (ret < 0)
1853 			goto unlock;
1854 		ret = tun_attach(tun, file);
1855 	} else if (ifr->ifr_flags & IFF_DETACH_QUEUE) {
1856 		tun = rtnl_dereference(tfile->tun);
1857 		if (!tun || !(tun->flags & TUN_TAP_MQ) || tfile->detached)
1858 			ret = -EINVAL;
1859 		else
1860 			__tun_detach(tfile, false);
1861 	} else
1862 		ret = -EINVAL;
1863 
1864 unlock:
1865 	rtnl_unlock();
1866 	return ret;
1867 }
1868 
1869 static long __tun_chr_ioctl(struct file *file, unsigned int cmd,
1870 			    unsigned long arg, int ifreq_len)
1871 {
1872 	struct tun_file *tfile = file->private_data;
1873 	struct tun_struct *tun;
1874 	void __user* argp = (void __user*)arg;
1875 	struct ifreq ifr;
1876 	kuid_t owner;
1877 	kgid_t group;
1878 	int sndbuf;
1879 	int vnet_hdr_sz;
1880 	int ret;
1881 
1882 	if (cmd == TUNSETIFF || cmd == TUNSETQUEUE || _IOC_TYPE(cmd) == 0x89) {
1883 		if (copy_from_user(&ifr, argp, ifreq_len))
1884 			return -EFAULT;
1885 	} else {
1886 		memset(&ifr, 0, sizeof(ifr));
1887 	}
1888 	if (cmd == TUNGETFEATURES) {
1889 		/* Currently this just means: "what IFF flags are valid?".
1890 		 * This is needed because we never checked for invalid flags on
1891 		 * TUNSETIFF. */
1892 		return put_user(IFF_TUN | IFF_TAP | IFF_NO_PI | IFF_ONE_QUEUE |
1893 				IFF_VNET_HDR | IFF_MULTI_QUEUE,
1894 				(unsigned int __user*)argp);
1895 	} else if (cmd == TUNSETQUEUE)
1896 		return tun_set_queue(file, &ifr);
1897 
1898 	ret = 0;
1899 	rtnl_lock();
1900 
1901 	tun = __tun_get(tfile);
1902 	if (cmd == TUNSETIFF && !tun) {
1903 		ifr.ifr_name[IFNAMSIZ-1] = '\0';
1904 
1905 		ret = tun_set_iff(tfile->net, file, &ifr);
1906 
1907 		if (ret)
1908 			goto unlock;
1909 
1910 		if (copy_to_user(argp, &ifr, ifreq_len))
1911 			ret = -EFAULT;
1912 		goto unlock;
1913 	}
1914 
1915 	ret = -EBADFD;
1916 	if (!tun)
1917 		goto unlock;
1918 
1919 	tun_debug(KERN_INFO, tun, "tun_chr_ioctl cmd %u\n", cmd);
1920 
1921 	ret = 0;
1922 	switch (cmd) {
1923 	case TUNGETIFF:
1924 		tun_get_iff(current->nsproxy->net_ns, tun, &ifr);
1925 
1926 		if (copy_to_user(argp, &ifr, ifreq_len))
1927 			ret = -EFAULT;
1928 		break;
1929 
1930 	case TUNSETNOCSUM:
1931 		/* Disable/Enable checksum */
1932 
1933 		/* [unimplemented] */
1934 		tun_debug(KERN_INFO, tun, "ignored: set checksum %s\n",
1935 			  arg ? "disabled" : "enabled");
1936 		break;
1937 
1938 	case TUNSETPERSIST:
1939 		/* Disable/Enable persist mode. Keep an extra reference to the
1940 		 * module to prevent the module being unprobed.
1941 		 */
1942 		if (arg && !(tun->flags & TUN_PERSIST)) {
1943 			tun->flags |= TUN_PERSIST;
1944 			__module_get(THIS_MODULE);
1945 		}
1946 		if (!arg && (tun->flags & TUN_PERSIST)) {
1947 			tun->flags &= ~TUN_PERSIST;
1948 			module_put(THIS_MODULE);
1949 		}
1950 
1951 		tun_debug(KERN_INFO, tun, "persist %s\n",
1952 			  arg ? "enabled" : "disabled");
1953 		break;
1954 
1955 	case TUNSETOWNER:
1956 		/* Set owner of the device */
1957 		owner = make_kuid(current_user_ns(), arg);
1958 		if (!uid_valid(owner)) {
1959 			ret = -EINVAL;
1960 			break;
1961 		}
1962 		tun->owner = owner;
1963 		tun_debug(KERN_INFO, tun, "owner set to %u\n",
1964 			  from_kuid(&init_user_ns, tun->owner));
1965 		break;
1966 
1967 	case TUNSETGROUP:
1968 		/* Set group of the device */
1969 		group = make_kgid(current_user_ns(), arg);
1970 		if (!gid_valid(group)) {
1971 			ret = -EINVAL;
1972 			break;
1973 		}
1974 		tun->group = group;
1975 		tun_debug(KERN_INFO, tun, "group set to %u\n",
1976 			  from_kgid(&init_user_ns, tun->group));
1977 		break;
1978 
1979 	case TUNSETLINK:
1980 		/* Only allow setting the type when the interface is down */
1981 		if (tun->dev->flags & IFF_UP) {
1982 			tun_debug(KERN_INFO, tun,
1983 				  "Linktype set failed because interface is up\n");
1984 			ret = -EBUSY;
1985 		} else {
1986 			tun->dev->type = (int) arg;
1987 			tun_debug(KERN_INFO, tun, "linktype set to %d\n",
1988 				  tun->dev->type);
1989 			ret = 0;
1990 		}
1991 		break;
1992 
1993 #ifdef TUN_DEBUG
1994 	case TUNSETDEBUG:
1995 		tun->debug = arg;
1996 		break;
1997 #endif
1998 	case TUNSETOFFLOAD:
1999 		ret = set_offload(tun, arg);
2000 		break;
2001 
2002 	case TUNSETTXFILTER:
2003 		/* Can be set only for TAPs */
2004 		ret = -EINVAL;
2005 		if ((tun->flags & TUN_TYPE_MASK) != TUN_TAP_DEV)
2006 			break;
2007 		ret = update_filter(&tun->txflt, (void __user *)arg);
2008 		break;
2009 
2010 	case SIOCGIFHWADDR:
2011 		/* Get hw address */
2012 		memcpy(ifr.ifr_hwaddr.sa_data, tun->dev->dev_addr, ETH_ALEN);
2013 		ifr.ifr_hwaddr.sa_family = tun->dev->type;
2014 		if (copy_to_user(argp, &ifr, ifreq_len))
2015 			ret = -EFAULT;
2016 		break;
2017 
2018 	case SIOCSIFHWADDR:
2019 		/* Set hw address */
2020 		tun_debug(KERN_DEBUG, tun, "set hw address: %pM\n",
2021 			  ifr.ifr_hwaddr.sa_data);
2022 
2023 		ret = dev_set_mac_address(tun->dev, &ifr.ifr_hwaddr);
2024 		break;
2025 
2026 	case TUNGETSNDBUF:
2027 		sndbuf = tfile->socket.sk->sk_sndbuf;
2028 		if (copy_to_user(argp, &sndbuf, sizeof(sndbuf)))
2029 			ret = -EFAULT;
2030 		break;
2031 
2032 	case TUNSETSNDBUF:
2033 		if (copy_from_user(&sndbuf, argp, sizeof(sndbuf))) {
2034 			ret = -EFAULT;
2035 			break;
2036 		}
2037 
2038 		tun->sndbuf = sndbuf;
2039 		tun_set_sndbuf(tun);
2040 		break;
2041 
2042 	case TUNGETVNETHDRSZ:
2043 		vnet_hdr_sz = tun->vnet_hdr_sz;
2044 		if (copy_to_user(argp, &vnet_hdr_sz, sizeof(vnet_hdr_sz)))
2045 			ret = -EFAULT;
2046 		break;
2047 
2048 	case TUNSETVNETHDRSZ:
2049 		if (copy_from_user(&vnet_hdr_sz, argp, sizeof(vnet_hdr_sz))) {
2050 			ret = -EFAULT;
2051 			break;
2052 		}
2053 		if (vnet_hdr_sz < (int)sizeof(struct virtio_net_hdr)) {
2054 			ret = -EINVAL;
2055 			break;
2056 		}
2057 
2058 		tun->vnet_hdr_sz = vnet_hdr_sz;
2059 		break;
2060 
2061 	case TUNATTACHFILTER:
2062 		/* Can be set only for TAPs */
2063 		ret = -EINVAL;
2064 		if ((tun->flags & TUN_TYPE_MASK) != TUN_TAP_DEV)
2065 			break;
2066 		ret = -EFAULT;
2067 		if (copy_from_user(&tun->fprog, argp, sizeof(tun->fprog)))
2068 			break;
2069 
2070 		ret = tun_attach_filter(tun);
2071 		break;
2072 
2073 	case TUNDETACHFILTER:
2074 		/* Can be set only for TAPs */
2075 		ret = -EINVAL;
2076 		if ((tun->flags & TUN_TYPE_MASK) != TUN_TAP_DEV)
2077 			break;
2078 		ret = 0;
2079 		tun_detach_filter(tun, tun->numqueues);
2080 		break;
2081 
2082 	default:
2083 		ret = -EINVAL;
2084 		break;
2085 	}
2086 
2087 unlock:
2088 	rtnl_unlock();
2089 	if (tun)
2090 		tun_put(tun);
2091 	return ret;
2092 }
2093 
2094 static long tun_chr_ioctl(struct file *file,
2095 			  unsigned int cmd, unsigned long arg)
2096 {
2097 	return __tun_chr_ioctl(file, cmd, arg, sizeof (struct ifreq));
2098 }
2099 
2100 #ifdef CONFIG_COMPAT
2101 static long tun_chr_compat_ioctl(struct file *file,
2102 			 unsigned int cmd, unsigned long arg)
2103 {
2104 	switch (cmd) {
2105 	case TUNSETIFF:
2106 	case TUNGETIFF:
2107 	case TUNSETTXFILTER:
2108 	case TUNGETSNDBUF:
2109 	case TUNSETSNDBUF:
2110 	case SIOCGIFHWADDR:
2111 	case SIOCSIFHWADDR:
2112 		arg = (unsigned long)compat_ptr(arg);
2113 		break;
2114 	default:
2115 		arg = (compat_ulong_t)arg;
2116 		break;
2117 	}
2118 
2119 	/*
2120 	 * compat_ifreq is shorter than ifreq, so we must not access beyond
2121 	 * the end of that structure. All fields that are used in this
2122 	 * driver are compatible though, we don't need to convert the
2123 	 * contents.
2124 	 */
2125 	return __tun_chr_ioctl(file, cmd, arg, sizeof(struct compat_ifreq));
2126 }
2127 #endif /* CONFIG_COMPAT */
2128 
2129 static int tun_chr_fasync(int fd, struct file *file, int on)
2130 {
2131 	struct tun_file *tfile = file->private_data;
2132 	int ret;
2133 
2134 	if ((ret = fasync_helper(fd, file, on, &tfile->fasync)) < 0)
2135 		goto out;
2136 
2137 	if (on) {
2138 		ret = __f_setown(file, task_pid(current), PIDTYPE_PID, 0);
2139 		if (ret)
2140 			goto out;
2141 		tfile->flags |= TUN_FASYNC;
2142 	} else
2143 		tfile->flags &= ~TUN_FASYNC;
2144 	ret = 0;
2145 out:
2146 	return ret;
2147 }
2148 
2149 static int tun_chr_open(struct inode *inode, struct file * file)
2150 {
2151 	struct tun_file *tfile;
2152 
2153 	DBG1(KERN_INFO, "tunX: tun_chr_open\n");
2154 
2155 	tfile = (struct tun_file *)sk_alloc(&init_net, AF_UNSPEC, GFP_KERNEL,
2156 					    &tun_proto);
2157 	if (!tfile)
2158 		return -ENOMEM;
2159 	rcu_assign_pointer(tfile->tun, NULL);
2160 	tfile->net = get_net(current->nsproxy->net_ns);
2161 	tfile->flags = 0;
2162 
2163 	rcu_assign_pointer(tfile->socket.wq, &tfile->wq);
2164 	init_waitqueue_head(&tfile->wq.wait);
2165 
2166 	tfile->socket.file = file;
2167 	tfile->socket.ops = &tun_socket_ops;
2168 
2169 	sock_init_data(&tfile->socket, &tfile->sk);
2170 	sk_change_net(&tfile->sk, tfile->net);
2171 
2172 	tfile->sk.sk_write_space = tun_sock_write_space;
2173 	tfile->sk.sk_sndbuf = INT_MAX;
2174 
2175 	file->private_data = tfile;
2176 	set_bit(SOCK_EXTERNALLY_ALLOCATED, &tfile->socket.flags);
2177 	INIT_LIST_HEAD(&tfile->next);
2178 
2179 	sock_set_flag(&tfile->sk, SOCK_ZEROCOPY);
2180 
2181 	return 0;
2182 }
2183 
2184 static int tun_chr_close(struct inode *inode, struct file *file)
2185 {
2186 	struct tun_file *tfile = file->private_data;
2187 	struct net *net = tfile->net;
2188 
2189 	tun_detach(tfile, true);
2190 	put_net(net);
2191 
2192 	return 0;
2193 }
2194 
2195 static const struct file_operations tun_fops = {
2196 	.owner	= THIS_MODULE,
2197 	.llseek = no_llseek,
2198 	.read  = do_sync_read,
2199 	.aio_read  = tun_chr_aio_read,
2200 	.write = do_sync_write,
2201 	.aio_write = tun_chr_aio_write,
2202 	.poll	= tun_chr_poll,
2203 	.unlocked_ioctl	= tun_chr_ioctl,
2204 #ifdef CONFIG_COMPAT
2205 	.compat_ioctl = tun_chr_compat_ioctl,
2206 #endif
2207 	.open	= tun_chr_open,
2208 	.release = tun_chr_close,
2209 	.fasync = tun_chr_fasync
2210 };
2211 
2212 static struct miscdevice tun_miscdev = {
2213 	.minor = TUN_MINOR,
2214 	.name = "tun",
2215 	.nodename = "net/tun",
2216 	.fops = &tun_fops,
2217 };
2218 
2219 /* ethtool interface */
2220 
2221 static int tun_get_settings(struct net_device *dev, struct ethtool_cmd *cmd)
2222 {
2223 	cmd->supported		= 0;
2224 	cmd->advertising	= 0;
2225 	ethtool_cmd_speed_set(cmd, SPEED_10);
2226 	cmd->duplex		= DUPLEX_FULL;
2227 	cmd->port		= PORT_TP;
2228 	cmd->phy_address	= 0;
2229 	cmd->transceiver	= XCVR_INTERNAL;
2230 	cmd->autoneg		= AUTONEG_DISABLE;
2231 	cmd->maxtxpkt		= 0;
2232 	cmd->maxrxpkt		= 0;
2233 	return 0;
2234 }
2235 
2236 static void tun_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
2237 {
2238 	struct tun_struct *tun = netdev_priv(dev);
2239 
2240 	strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
2241 	strlcpy(info->version, DRV_VERSION, sizeof(info->version));
2242 
2243 	switch (tun->flags & TUN_TYPE_MASK) {
2244 	case TUN_TUN_DEV:
2245 		strlcpy(info->bus_info, "tun", sizeof(info->bus_info));
2246 		break;
2247 	case TUN_TAP_DEV:
2248 		strlcpy(info->bus_info, "tap", sizeof(info->bus_info));
2249 		break;
2250 	}
2251 }
2252 
2253 static u32 tun_get_msglevel(struct net_device *dev)
2254 {
2255 #ifdef TUN_DEBUG
2256 	struct tun_struct *tun = netdev_priv(dev);
2257 	return tun->debug;
2258 #else
2259 	return -EOPNOTSUPP;
2260 #endif
2261 }
2262 
2263 static void tun_set_msglevel(struct net_device *dev, u32 value)
2264 {
2265 #ifdef TUN_DEBUG
2266 	struct tun_struct *tun = netdev_priv(dev);
2267 	tun->debug = value;
2268 #endif
2269 }
2270 
2271 static const struct ethtool_ops tun_ethtool_ops = {
2272 	.get_settings	= tun_get_settings,
2273 	.get_drvinfo	= tun_get_drvinfo,
2274 	.get_msglevel	= tun_get_msglevel,
2275 	.set_msglevel	= tun_set_msglevel,
2276 	.get_link	= ethtool_op_get_link,
2277 };
2278 
2279 
2280 static int __init tun_init(void)
2281 {
2282 	int ret = 0;
2283 
2284 	pr_info("%s, %s\n", DRV_DESCRIPTION, DRV_VERSION);
2285 	pr_info("%s\n", DRV_COPYRIGHT);
2286 
2287 	ret = rtnl_link_register(&tun_link_ops);
2288 	if (ret) {
2289 		pr_err("Can't register link_ops\n");
2290 		goto err_linkops;
2291 	}
2292 
2293 	ret = misc_register(&tun_miscdev);
2294 	if (ret) {
2295 		pr_err("Can't register misc device %d\n", TUN_MINOR);
2296 		goto err_misc;
2297 	}
2298 	return  0;
2299 err_misc:
2300 	rtnl_link_unregister(&tun_link_ops);
2301 err_linkops:
2302 	return ret;
2303 }
2304 
2305 static void tun_cleanup(void)
2306 {
2307 	misc_deregister(&tun_miscdev);
2308 	rtnl_link_unregister(&tun_link_ops);
2309 }
2310 
2311 /* Get an underlying socket object from tun file.  Returns error unless file is
2312  * attached to a device.  The returned object works like a packet socket, it
2313  * can be used for sock_sendmsg/sock_recvmsg.  The caller is responsible for
2314  * holding a reference to the file for as long as the socket is in use. */
2315 struct socket *tun_get_socket(struct file *file)
2316 {
2317 	struct tun_file *tfile;
2318 	if (file->f_op != &tun_fops)
2319 		return ERR_PTR(-EINVAL);
2320 	tfile = file->private_data;
2321 	if (!tfile)
2322 		return ERR_PTR(-EBADFD);
2323 	return &tfile->socket;
2324 }
2325 EXPORT_SYMBOL_GPL(tun_get_socket);
2326 
2327 module_init(tun_init);
2328 module_exit(tun_cleanup);
2329 MODULE_DESCRIPTION(DRV_DESCRIPTION);
2330 MODULE_AUTHOR(DRV_COPYRIGHT);
2331 MODULE_LICENSE("GPL");
2332 MODULE_ALIAS_MISCDEV(TUN_MINOR);
2333 MODULE_ALIAS("devname:net/tun");
2334