xref: /linux/drivers/net/tun.c (revision b889fcf63cb62e7fdb7816565e28f44dbe4a76a5)
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 kmem_cache *flow_cache;
184 	struct hlist_head flows[TUN_NUM_FLOW_ENTRIES];
185 	struct timer_list flow_gc_timer;
186 	unsigned long ageing_time;
187 	unsigned int numdisabled;
188 	struct list_head disabled;
189 };
190 
191 static inline u32 tun_hashfn(u32 rxhash)
192 {
193 	return rxhash & 0x3ff;
194 }
195 
196 static struct tun_flow_entry *tun_flow_find(struct hlist_head *head, u32 rxhash)
197 {
198 	struct tun_flow_entry *e;
199 	struct hlist_node *n;
200 
201 	hlist_for_each_entry_rcu(e, n, 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 = kmem_cache_alloc(tun->flow_cache,
213 						    GFP_ATOMIC);
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 	}
223 	return e;
224 }
225 
226 static void tun_flow_free(struct rcu_head *head)
227 {
228 	struct tun_flow_entry *e
229 		= container_of(head, struct tun_flow_entry, rcu);
230 	kmem_cache_free(e->tun->flow_cache, e);
231 }
232 
233 static void tun_flow_delete(struct tun_struct *tun, struct tun_flow_entry *e)
234 {
235 	tun_debug(KERN_INFO, tun, "delete flow: hash %u index %u\n",
236 		  e->rxhash, e->queue_index);
237 	hlist_del_rcu(&e->hash_link);
238 	call_rcu(&e->rcu, tun_flow_free);
239 }
240 
241 static void tun_flow_flush(struct tun_struct *tun)
242 {
243 	int i;
244 
245 	spin_lock_bh(&tun->lock);
246 	for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
247 		struct tun_flow_entry *e;
248 		struct hlist_node *h, *n;
249 
250 		hlist_for_each_entry_safe(e, h, n, &tun->flows[i], hash_link)
251 			tun_flow_delete(tun, e);
252 	}
253 	spin_unlock_bh(&tun->lock);
254 }
255 
256 static void tun_flow_delete_by_queue(struct tun_struct *tun, u16 queue_index)
257 {
258 	int i;
259 
260 	spin_lock_bh(&tun->lock);
261 	for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
262 		struct tun_flow_entry *e;
263 		struct hlist_node *h, *n;
264 
265 		hlist_for_each_entry_safe(e, h, n, &tun->flows[i], hash_link) {
266 			if (e->queue_index == queue_index)
267 				tun_flow_delete(tun, e);
268 		}
269 	}
270 	spin_unlock_bh(&tun->lock);
271 }
272 
273 static void tun_flow_cleanup(unsigned long data)
274 {
275 	struct tun_struct *tun = (struct tun_struct *)data;
276 	unsigned long delay = tun->ageing_time;
277 	unsigned long next_timer = jiffies + delay;
278 	unsigned long count = 0;
279 	int i;
280 
281 	tun_debug(KERN_INFO, tun, "tun_flow_cleanup\n");
282 
283 	spin_lock_bh(&tun->lock);
284 	for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
285 		struct tun_flow_entry *e;
286 		struct hlist_node *h, *n;
287 
288 		hlist_for_each_entry_safe(e, h, n, &tun->flows[i], hash_link) {
289 			unsigned long this_timer;
290 			count++;
291 			this_timer = e->updated + delay;
292 			if (time_before_eq(this_timer, jiffies))
293 				tun_flow_delete(tun, e);
294 			else if (time_before(this_timer, next_timer))
295 				next_timer = this_timer;
296 		}
297 	}
298 
299 	if (count)
300 		mod_timer(&tun->flow_gc_timer, round_jiffies_up(next_timer));
301 	spin_unlock_bh(&tun->lock);
302 }
303 
304 static void tun_flow_update(struct tun_struct *tun, u32 rxhash,
305 			    u16 queue_index)
306 {
307 	struct hlist_head *head;
308 	struct tun_flow_entry *e;
309 	unsigned long delay = tun->ageing_time;
310 
311 	if (!rxhash)
312 		return;
313 	else
314 		head = &tun->flows[tun_hashfn(rxhash)];
315 
316 	rcu_read_lock();
317 
318 	if (tun->numqueues == 1)
319 		goto unlock;
320 
321 	e = tun_flow_find(head, rxhash);
322 	if (likely(e)) {
323 		/* TODO: keep queueing to old queue until it's empty? */
324 		e->queue_index = queue_index;
325 		e->updated = jiffies;
326 	} else {
327 		spin_lock_bh(&tun->lock);
328 		if (!tun_flow_find(head, rxhash))
329 			tun_flow_create(tun, head, rxhash, queue_index);
330 
331 		if (!timer_pending(&tun->flow_gc_timer))
332 			mod_timer(&tun->flow_gc_timer,
333 				  round_jiffies_up(jiffies + delay));
334 		spin_unlock_bh(&tun->lock);
335 	}
336 
337 unlock:
338 	rcu_read_unlock();
339 }
340 
341 /* We try to identify a flow through its rxhash first. The reason that
342  * we do not check rxq no. is becuase some cards(e.g 82599), chooses
343  * the rxq based on the txq where the last packet of the flow comes. As
344  * the userspace application move between processors, we may get a
345  * different rxq no. here. If we could not get rxhash, then we would
346  * hope the rxq no. may help here.
347  */
348 static u16 tun_select_queue(struct net_device *dev, struct sk_buff *skb)
349 {
350 	struct tun_struct *tun = netdev_priv(dev);
351 	struct tun_flow_entry *e;
352 	u32 txq = 0;
353 	u32 numqueues = 0;
354 
355 	rcu_read_lock();
356 	numqueues = tun->numqueues;
357 
358 	txq = skb_get_rxhash(skb);
359 	if (txq) {
360 		e = tun_flow_find(&tun->flows[tun_hashfn(txq)], txq);
361 		if (e)
362 			txq = e->queue_index;
363 		else
364 			/* use multiply and shift instead of expensive divide */
365 			txq = ((u64)txq * numqueues) >> 32;
366 	} else if (likely(skb_rx_queue_recorded(skb))) {
367 		txq = skb_get_rx_queue(skb);
368 		while (unlikely(txq >= numqueues))
369 			txq -= numqueues;
370 	}
371 
372 	rcu_read_unlock();
373 	return txq;
374 }
375 
376 static inline bool tun_not_capable(struct tun_struct *tun)
377 {
378 	const struct cred *cred = current_cred();
379 	struct net *net = dev_net(tun->dev);
380 
381 	return ((uid_valid(tun->owner) && !uid_eq(cred->euid, tun->owner)) ||
382 		  (gid_valid(tun->group) && !in_egroup_p(tun->group))) &&
383 		!ns_capable(net->user_ns, CAP_NET_ADMIN);
384 }
385 
386 static void tun_set_real_num_queues(struct tun_struct *tun)
387 {
388 	netif_set_real_num_tx_queues(tun->dev, tun->numqueues);
389 	netif_set_real_num_rx_queues(tun->dev, tun->numqueues);
390 }
391 
392 static void tun_disable_queue(struct tun_struct *tun, struct tun_file *tfile)
393 {
394 	tfile->detached = tun;
395 	list_add_tail(&tfile->next, &tun->disabled);
396 	++tun->numdisabled;
397 }
398 
399 static struct tun_struct *tun_enable_queue(struct tun_file *tfile)
400 {
401 	struct tun_struct *tun = tfile->detached;
402 
403 	tfile->detached = NULL;
404 	list_del_init(&tfile->next);
405 	--tun->numdisabled;
406 	return tun;
407 }
408 
409 static void __tun_detach(struct tun_file *tfile, bool clean)
410 {
411 	struct tun_file *ntfile;
412 	struct tun_struct *tun;
413 	struct net_device *dev;
414 
415 	tun = rcu_dereference_protected(tfile->tun,
416 					lockdep_rtnl_is_held());
417 	if (tun) {
418 		u16 index = tfile->queue_index;
419 		BUG_ON(index >= tun->numqueues);
420 		dev = tun->dev;
421 
422 		rcu_assign_pointer(tun->tfiles[index],
423 				   tun->tfiles[tun->numqueues - 1]);
424 		rcu_assign_pointer(tfile->tun, NULL);
425 		ntfile = rcu_dereference_protected(tun->tfiles[index],
426 						   lockdep_rtnl_is_held());
427 		ntfile->queue_index = index;
428 
429 		--tun->numqueues;
430 		if (clean)
431 			sock_put(&tfile->sk);
432 		else
433 			tun_disable_queue(tun, tfile);
434 
435 		synchronize_net();
436 		tun_flow_delete_by_queue(tun, tun->numqueues + 1);
437 		/* Drop read queue */
438 		skb_queue_purge(&tfile->sk.sk_receive_queue);
439 		tun_set_real_num_queues(tun);
440 	} else if (tfile->detached && clean)
441 		tun = tun_enable_queue(tfile);
442 
443 	if (clean) {
444 		if (tun && tun->numqueues == 0 && tun->numdisabled == 0 &&
445 		    !(tun->flags & TUN_PERSIST))
446 			if (tun->dev->reg_state == NETREG_REGISTERED)
447 				unregister_netdevice(tun->dev);
448 
449 		BUG_ON(!test_bit(SOCK_EXTERNALLY_ALLOCATED,
450 				 &tfile->socket.flags));
451 		sk_release_kernel(&tfile->sk);
452 	}
453 }
454 
455 static void tun_detach(struct tun_file *tfile, bool clean)
456 {
457 	rtnl_lock();
458 	__tun_detach(tfile, clean);
459 	rtnl_unlock();
460 }
461 
462 static void tun_detach_all(struct net_device *dev)
463 {
464 	struct tun_struct *tun = netdev_priv(dev);
465 	struct tun_file *tfile, *tmp;
466 	int i, n = tun->numqueues;
467 
468 	for (i = 0; i < n; i++) {
469 		tfile = rcu_dereference_protected(tun->tfiles[i],
470 						  lockdep_rtnl_is_held());
471 		BUG_ON(!tfile);
472 		wake_up_all(&tfile->wq.wait);
473 		rcu_assign_pointer(tfile->tun, NULL);
474 		--tun->numqueues;
475 	}
476 	BUG_ON(tun->numqueues != 0);
477 
478 	synchronize_net();
479 	for (i = 0; i < n; i++) {
480 		tfile = rcu_dereference_protected(tun->tfiles[i],
481 						  lockdep_rtnl_is_held());
482 		/* Drop read queue */
483 		skb_queue_purge(&tfile->sk.sk_receive_queue);
484 		sock_put(&tfile->sk);
485 	}
486 	list_for_each_entry_safe(tfile, tmp, &tun->disabled, next) {
487 		tun_enable_queue(tfile);
488 		skb_queue_purge(&tfile->sk.sk_receive_queue);
489 		sock_put(&tfile->sk);
490 	}
491 	BUG_ON(tun->numdisabled != 0);
492 }
493 
494 static int tun_attach(struct tun_struct *tun, struct file *file)
495 {
496 	struct tun_file *tfile = file->private_data;
497 	int err;
498 
499 	err = -EINVAL;
500 	if (rcu_dereference_protected(tfile->tun, lockdep_rtnl_is_held()))
501 		goto out;
502 
503 	err = -EBUSY;
504 	if (!(tun->flags & TUN_TAP_MQ) && tun->numqueues == 1)
505 		goto out;
506 
507 	err = -E2BIG;
508 	if (!tfile->detached &&
509 	    tun->numqueues + tun->numdisabled == MAX_TAP_QUEUES)
510 		goto out;
511 
512 	err = 0;
513 
514 	/* Re-attach the filter to presist device */
515 	if (tun->filter_attached == true) {
516 		err = sk_attach_filter(&tun->fprog, tfile->socket.sk);
517 		if (!err)
518 			goto out;
519 	}
520 	tfile->queue_index = tun->numqueues;
521 	rcu_assign_pointer(tfile->tun, tun);
522 	rcu_assign_pointer(tun->tfiles[tun->numqueues], tfile);
523 	tun->numqueues++;
524 
525 	if (tfile->detached)
526 		tun_enable_queue(tfile);
527 	else
528 		sock_hold(&tfile->sk);
529 
530 	tun_set_real_num_queues(tun);
531 
532 	/* device is allowed to go away first, so no need to hold extra
533 	 * refcnt.
534 	 */
535 
536 out:
537 	return err;
538 }
539 
540 static struct tun_struct *__tun_get(struct tun_file *tfile)
541 {
542 	struct tun_struct *tun;
543 
544 	rcu_read_lock();
545 	tun = rcu_dereference(tfile->tun);
546 	if (tun)
547 		dev_hold(tun->dev);
548 	rcu_read_unlock();
549 
550 	return tun;
551 }
552 
553 static struct tun_struct *tun_get(struct file *file)
554 {
555 	return __tun_get(file->private_data);
556 }
557 
558 static void tun_put(struct tun_struct *tun)
559 {
560 	dev_put(tun->dev);
561 }
562 
563 /* TAP filtering */
564 static void addr_hash_set(u32 *mask, const u8 *addr)
565 {
566 	int n = ether_crc(ETH_ALEN, addr) >> 26;
567 	mask[n >> 5] |= (1 << (n & 31));
568 }
569 
570 static unsigned int addr_hash_test(const u32 *mask, const u8 *addr)
571 {
572 	int n = ether_crc(ETH_ALEN, addr) >> 26;
573 	return mask[n >> 5] & (1 << (n & 31));
574 }
575 
576 static int update_filter(struct tap_filter *filter, void __user *arg)
577 {
578 	struct { u8 u[ETH_ALEN]; } *addr;
579 	struct tun_filter uf;
580 	int err, alen, n, nexact;
581 
582 	if (copy_from_user(&uf, arg, sizeof(uf)))
583 		return -EFAULT;
584 
585 	if (!uf.count) {
586 		/* Disabled */
587 		filter->count = 0;
588 		return 0;
589 	}
590 
591 	alen = ETH_ALEN * uf.count;
592 	addr = kmalloc(alen, GFP_KERNEL);
593 	if (!addr)
594 		return -ENOMEM;
595 
596 	if (copy_from_user(addr, arg + sizeof(uf), alen)) {
597 		err = -EFAULT;
598 		goto done;
599 	}
600 
601 	/* The filter is updated without holding any locks. Which is
602 	 * perfectly safe. We disable it first and in the worst
603 	 * case we'll accept a few undesired packets. */
604 	filter->count = 0;
605 	wmb();
606 
607 	/* Use first set of addresses as an exact filter */
608 	for (n = 0; n < uf.count && n < FLT_EXACT_COUNT; n++)
609 		memcpy(filter->addr[n], addr[n].u, ETH_ALEN);
610 
611 	nexact = n;
612 
613 	/* Remaining multicast addresses are hashed,
614 	 * unicast will leave the filter disabled. */
615 	memset(filter->mask, 0, sizeof(filter->mask));
616 	for (; n < uf.count; n++) {
617 		if (!is_multicast_ether_addr(addr[n].u)) {
618 			err = 0; /* no filter */
619 			goto done;
620 		}
621 		addr_hash_set(filter->mask, addr[n].u);
622 	}
623 
624 	/* For ALLMULTI just set the mask to all ones.
625 	 * This overrides the mask populated above. */
626 	if ((uf.flags & TUN_FLT_ALLMULTI))
627 		memset(filter->mask, ~0, sizeof(filter->mask));
628 
629 	/* Now enable the filter */
630 	wmb();
631 	filter->count = nexact;
632 
633 	/* Return the number of exact filters */
634 	err = nexact;
635 
636 done:
637 	kfree(addr);
638 	return err;
639 }
640 
641 /* Returns: 0 - drop, !=0 - accept */
642 static int run_filter(struct tap_filter *filter, const struct sk_buff *skb)
643 {
644 	/* Cannot use eth_hdr(skb) here because skb_mac_hdr() is incorrect
645 	 * at this point. */
646 	struct ethhdr *eh = (struct ethhdr *) skb->data;
647 	int i;
648 
649 	/* Exact match */
650 	for (i = 0; i < filter->count; i++)
651 		if (ether_addr_equal(eh->h_dest, filter->addr[i]))
652 			return 1;
653 
654 	/* Inexact match (multicast only) */
655 	if (is_multicast_ether_addr(eh->h_dest))
656 		return addr_hash_test(filter->mask, eh->h_dest);
657 
658 	return 0;
659 }
660 
661 /*
662  * Checks whether the packet is accepted or not.
663  * Returns: 0 - drop, !=0 - accept
664  */
665 static int check_filter(struct tap_filter *filter, const struct sk_buff *skb)
666 {
667 	if (!filter->count)
668 		return 1;
669 
670 	return run_filter(filter, skb);
671 }
672 
673 /* Network device part of the driver */
674 
675 static const struct ethtool_ops tun_ethtool_ops;
676 
677 /* Net device detach from fd. */
678 static void tun_net_uninit(struct net_device *dev)
679 {
680 	tun_detach_all(dev);
681 }
682 
683 /* Net device open. */
684 static int tun_net_open(struct net_device *dev)
685 {
686 	netif_tx_start_all_queues(dev);
687 	return 0;
688 }
689 
690 /* Net device close. */
691 static int tun_net_close(struct net_device *dev)
692 {
693 	netif_tx_stop_all_queues(dev);
694 	return 0;
695 }
696 
697 /* Net device start xmit */
698 static netdev_tx_t tun_net_xmit(struct sk_buff *skb, struct net_device *dev)
699 {
700 	struct tun_struct *tun = netdev_priv(dev);
701 	int txq = skb->queue_mapping;
702 	struct tun_file *tfile;
703 
704 	rcu_read_lock();
705 	tfile = rcu_dereference(tun->tfiles[txq]);
706 
707 	/* Drop packet if interface is not attached */
708 	if (txq >= tun->numqueues)
709 		goto drop;
710 
711 	tun_debug(KERN_INFO, tun, "tun_net_xmit %d\n", skb->len);
712 
713 	BUG_ON(!tfile);
714 
715 	/* Drop if the filter does not like it.
716 	 * This is a noop if the filter is disabled.
717 	 * Filter can be enabled only for the TAP devices. */
718 	if (!check_filter(&tun->txflt, skb))
719 		goto drop;
720 
721 	if (tfile->socket.sk->sk_filter &&
722 	    sk_filter(tfile->socket.sk, skb))
723 		goto drop;
724 
725 	/* Limit the number of packets queued by dividing txq length with the
726 	 * number of queues.
727 	 */
728 	if (skb_queue_len(&tfile->socket.sk->sk_receive_queue)
729 			  >= dev->tx_queue_len / tun->numqueues)
730 		goto drop;
731 
732 	/* Orphan the skb - required as we might hang on to it
733 	 * for indefinite time. */
734 	if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC)))
735 		goto drop;
736 	skb_orphan(skb);
737 
738 	/* Enqueue packet */
739 	skb_queue_tail(&tfile->socket.sk->sk_receive_queue, skb);
740 
741 	/* Notify and wake up reader process */
742 	if (tfile->flags & TUN_FASYNC)
743 		kill_fasync(&tfile->fasync, SIGIO, POLL_IN);
744 	wake_up_interruptible_poll(&tfile->wq.wait, POLLIN |
745 				   POLLRDNORM | POLLRDBAND);
746 
747 	rcu_read_unlock();
748 	return NETDEV_TX_OK;
749 
750 drop:
751 	dev->stats.tx_dropped++;
752 	skb_tx_error(skb);
753 	kfree_skb(skb);
754 	rcu_read_unlock();
755 	return NETDEV_TX_OK;
756 }
757 
758 static void tun_net_mclist(struct net_device *dev)
759 {
760 	/*
761 	 * This callback is supposed to deal with mc filter in
762 	 * _rx_ path and has nothing to do with the _tx_ path.
763 	 * In rx path we always accept everything userspace gives us.
764 	 */
765 }
766 
767 #define MIN_MTU 68
768 #define MAX_MTU 65535
769 
770 static int
771 tun_net_change_mtu(struct net_device *dev, int new_mtu)
772 {
773 	if (new_mtu < MIN_MTU || new_mtu + dev->hard_header_len > MAX_MTU)
774 		return -EINVAL;
775 	dev->mtu = new_mtu;
776 	return 0;
777 }
778 
779 static netdev_features_t tun_net_fix_features(struct net_device *dev,
780 	netdev_features_t features)
781 {
782 	struct tun_struct *tun = netdev_priv(dev);
783 
784 	return (features & tun->set_features) | (features & ~TUN_USER_FEATURES);
785 }
786 #ifdef CONFIG_NET_POLL_CONTROLLER
787 static void tun_poll_controller(struct net_device *dev)
788 {
789 	/*
790 	 * Tun only receives frames when:
791 	 * 1) the char device endpoint gets data from user space
792 	 * 2) the tun socket gets a sendmsg call from user space
793 	 * Since both of those are syncronous operations, we are guaranteed
794 	 * never to have pending data when we poll for it
795 	 * so theres nothing to do here but return.
796 	 * We need this though so netpoll recognizes us as an interface that
797 	 * supports polling, which enables bridge devices in virt setups to
798 	 * still use netconsole
799 	 */
800 	return;
801 }
802 #endif
803 static const struct net_device_ops tun_netdev_ops = {
804 	.ndo_uninit		= tun_net_uninit,
805 	.ndo_open		= tun_net_open,
806 	.ndo_stop		= tun_net_close,
807 	.ndo_start_xmit		= tun_net_xmit,
808 	.ndo_change_mtu		= tun_net_change_mtu,
809 	.ndo_fix_features	= tun_net_fix_features,
810 	.ndo_select_queue	= tun_select_queue,
811 #ifdef CONFIG_NET_POLL_CONTROLLER
812 	.ndo_poll_controller	= tun_poll_controller,
813 #endif
814 };
815 
816 static const struct net_device_ops tap_netdev_ops = {
817 	.ndo_uninit		= tun_net_uninit,
818 	.ndo_open		= tun_net_open,
819 	.ndo_stop		= tun_net_close,
820 	.ndo_start_xmit		= tun_net_xmit,
821 	.ndo_change_mtu		= tun_net_change_mtu,
822 	.ndo_fix_features	= tun_net_fix_features,
823 	.ndo_set_rx_mode	= tun_net_mclist,
824 	.ndo_set_mac_address	= eth_mac_addr,
825 	.ndo_validate_addr	= eth_validate_addr,
826 	.ndo_select_queue	= tun_select_queue,
827 #ifdef CONFIG_NET_POLL_CONTROLLER
828 	.ndo_poll_controller	= tun_poll_controller,
829 #endif
830 };
831 
832 static int tun_flow_init(struct tun_struct *tun)
833 {
834 	int i;
835 
836 	tun->flow_cache = kmem_cache_create("tun_flow_cache",
837 					    sizeof(struct tun_flow_entry), 0, 0,
838 					    NULL);
839 	if (!tun->flow_cache)
840 		return -ENOMEM;
841 
842 	for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++)
843 		INIT_HLIST_HEAD(&tun->flows[i]);
844 
845 	tun->ageing_time = TUN_FLOW_EXPIRE;
846 	setup_timer(&tun->flow_gc_timer, tun_flow_cleanup, (unsigned long)tun);
847 	mod_timer(&tun->flow_gc_timer,
848 		  round_jiffies_up(jiffies + tun->ageing_time));
849 
850 	return 0;
851 }
852 
853 static void tun_flow_uninit(struct tun_struct *tun)
854 {
855 	del_timer_sync(&tun->flow_gc_timer);
856 	tun_flow_flush(tun);
857 
858 	/* Wait for completion of call_rcu()'s */
859 	rcu_barrier();
860 	kmem_cache_destroy(tun->flow_cache);
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 			for (i = 0; i < num_pages; i++)
1012 				put_page(page[i]);
1013 			return -EFAULT;
1014 		}
1015 		truesize = size * PAGE_SIZE;
1016 		skb->data_len += len;
1017 		skb->len += len;
1018 		skb->truesize += truesize;
1019 		atomic_add(truesize, &skb->sk->sk_wmem_alloc);
1020 		while (len) {
1021 			int off = base & ~PAGE_MASK;
1022 			int size = min_t(int, len, PAGE_SIZE - off);
1023 			__skb_fill_page_desc(skb, i, page[i], off, size);
1024 			skb_shinfo(skb)->nr_frags++;
1025 			/* increase sk_wmem_alloc */
1026 			base += size;
1027 			len -= size;
1028 			i++;
1029 		}
1030 		offset = 0;
1031 		++from;
1032 	}
1033 	return 0;
1034 }
1035 
1036 /* Get packet from user space buffer */
1037 static ssize_t tun_get_user(struct tun_struct *tun, struct tun_file *tfile,
1038 			    void *msg_control, const struct iovec *iv,
1039 			    size_t total_len, size_t count, int noblock)
1040 {
1041 	struct tun_pi pi = { 0, cpu_to_be16(ETH_P_IP) };
1042 	struct sk_buff *skb;
1043 	size_t len = total_len, align = NET_SKB_PAD;
1044 	struct virtio_net_hdr gso = { 0 };
1045 	int offset = 0;
1046 	int copylen;
1047 	bool zerocopy = false;
1048 	int err;
1049 	u32 rxhash;
1050 
1051 	if (!(tun->flags & TUN_NO_PI)) {
1052 		if ((len -= sizeof(pi)) > total_len)
1053 			return -EINVAL;
1054 
1055 		if (memcpy_fromiovecend((void *)&pi, iv, 0, sizeof(pi)))
1056 			return -EFAULT;
1057 		offset += sizeof(pi);
1058 	}
1059 
1060 	if (tun->flags & TUN_VNET_HDR) {
1061 		if ((len -= tun->vnet_hdr_sz) > total_len)
1062 			return -EINVAL;
1063 
1064 		if (memcpy_fromiovecend((void *)&gso, iv, offset, sizeof(gso)))
1065 			return -EFAULT;
1066 
1067 		if ((gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
1068 		    gso.csum_start + gso.csum_offset + 2 > gso.hdr_len)
1069 			gso.hdr_len = gso.csum_start + gso.csum_offset + 2;
1070 
1071 		if (gso.hdr_len > len)
1072 			return -EINVAL;
1073 		offset += tun->vnet_hdr_sz;
1074 	}
1075 
1076 	if ((tun->flags & TUN_TYPE_MASK) == TUN_TAP_DEV) {
1077 		align += NET_IP_ALIGN;
1078 		if (unlikely(len < ETH_HLEN ||
1079 			     (gso.hdr_len && gso.hdr_len < ETH_HLEN)))
1080 			return -EINVAL;
1081 	}
1082 
1083 	if (msg_control)
1084 		zerocopy = true;
1085 
1086 	if (zerocopy) {
1087 		/* Userspace may produce vectors with count greater than
1088 		 * MAX_SKB_FRAGS, so we need to linearize parts of the skb
1089 		 * to let the rest of data to be fit in the frags.
1090 		 */
1091 		if (count > MAX_SKB_FRAGS) {
1092 			copylen = iov_length(iv, count - MAX_SKB_FRAGS);
1093 			if (copylen < offset)
1094 				copylen = 0;
1095 			else
1096 				copylen -= offset;
1097 		} else
1098 				copylen = 0;
1099 		/* There are 256 bytes to be copied in skb, so there is enough
1100 		 * room for skb expand head in case it is used.
1101 		 * The rest of the buffer is mapped from userspace.
1102 		 */
1103 		if (copylen < gso.hdr_len)
1104 			copylen = gso.hdr_len;
1105 		if (!copylen)
1106 			copylen = GOODCOPY_LEN;
1107 	} else
1108 		copylen = len;
1109 
1110 	skb = tun_alloc_skb(tfile, align, copylen, gso.hdr_len, noblock);
1111 	if (IS_ERR(skb)) {
1112 		if (PTR_ERR(skb) != -EAGAIN)
1113 			tun->dev->stats.rx_dropped++;
1114 		return PTR_ERR(skb);
1115 	}
1116 
1117 	if (zerocopy)
1118 		err = zerocopy_sg_from_iovec(skb, iv, offset, count);
1119 	else
1120 		err = skb_copy_datagram_from_iovec(skb, 0, iv, offset, len);
1121 
1122 	if (err) {
1123 		tun->dev->stats.rx_dropped++;
1124 		kfree_skb(skb);
1125 		return -EFAULT;
1126 	}
1127 
1128 	if (gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
1129 		if (!skb_partial_csum_set(skb, gso.csum_start,
1130 					  gso.csum_offset)) {
1131 			tun->dev->stats.rx_frame_errors++;
1132 			kfree_skb(skb);
1133 			return -EINVAL;
1134 		}
1135 	}
1136 
1137 	switch (tun->flags & TUN_TYPE_MASK) {
1138 	case TUN_TUN_DEV:
1139 		if (tun->flags & TUN_NO_PI) {
1140 			switch (skb->data[0] & 0xf0) {
1141 			case 0x40:
1142 				pi.proto = htons(ETH_P_IP);
1143 				break;
1144 			case 0x60:
1145 				pi.proto = htons(ETH_P_IPV6);
1146 				break;
1147 			default:
1148 				tun->dev->stats.rx_dropped++;
1149 				kfree_skb(skb);
1150 				return -EINVAL;
1151 			}
1152 		}
1153 
1154 		skb_reset_mac_header(skb);
1155 		skb->protocol = pi.proto;
1156 		skb->dev = tun->dev;
1157 		break;
1158 	case TUN_TAP_DEV:
1159 		skb->protocol = eth_type_trans(skb, tun->dev);
1160 		break;
1161 	}
1162 
1163 	if (gso.gso_type != VIRTIO_NET_HDR_GSO_NONE) {
1164 		pr_debug("GSO!\n");
1165 		switch (gso.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
1166 		case VIRTIO_NET_HDR_GSO_TCPV4:
1167 			skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
1168 			break;
1169 		case VIRTIO_NET_HDR_GSO_TCPV6:
1170 			skb_shinfo(skb)->gso_type = SKB_GSO_TCPV6;
1171 			break;
1172 		case VIRTIO_NET_HDR_GSO_UDP:
1173 			skb_shinfo(skb)->gso_type = SKB_GSO_UDP;
1174 			break;
1175 		default:
1176 			tun->dev->stats.rx_frame_errors++;
1177 			kfree_skb(skb);
1178 			return -EINVAL;
1179 		}
1180 
1181 		if (gso.gso_type & VIRTIO_NET_HDR_GSO_ECN)
1182 			skb_shinfo(skb)->gso_type |= SKB_GSO_TCP_ECN;
1183 
1184 		skb_shinfo(skb)->gso_size = gso.gso_size;
1185 		if (skb_shinfo(skb)->gso_size == 0) {
1186 			tun->dev->stats.rx_frame_errors++;
1187 			kfree_skb(skb);
1188 			return -EINVAL;
1189 		}
1190 
1191 		/* Header must be checked, and gso_segs computed. */
1192 		skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
1193 		skb_shinfo(skb)->gso_segs = 0;
1194 	}
1195 
1196 	/* copy skb_ubuf_info for callback when skb has no error */
1197 	if (zerocopy) {
1198 		skb_shinfo(skb)->destructor_arg = msg_control;
1199 		skb_shinfo(skb)->tx_flags |= SKBTX_DEV_ZEROCOPY;
1200 	}
1201 
1202 	skb_reset_network_header(skb);
1203 	rxhash = skb_get_rxhash(skb);
1204 	netif_rx_ni(skb);
1205 
1206 	tun->dev->stats.rx_packets++;
1207 	tun->dev->stats.rx_bytes += len;
1208 
1209 	tun_flow_update(tun, rxhash, tfile->queue_index);
1210 	return total_len;
1211 }
1212 
1213 static ssize_t tun_chr_aio_write(struct kiocb *iocb, const struct iovec *iv,
1214 			      unsigned long count, loff_t pos)
1215 {
1216 	struct file *file = iocb->ki_filp;
1217 	struct tun_struct *tun = tun_get(file);
1218 	struct tun_file *tfile = file->private_data;
1219 	ssize_t result;
1220 
1221 	if (!tun)
1222 		return -EBADFD;
1223 
1224 	tun_debug(KERN_INFO, tun, "tun_chr_write %ld\n", count);
1225 
1226 	result = tun_get_user(tun, tfile, NULL, iv, iov_length(iv, count),
1227 			      count, file->f_flags & O_NONBLOCK);
1228 
1229 	tun_put(tun);
1230 	return result;
1231 }
1232 
1233 /* Put packet to the user space buffer */
1234 static ssize_t tun_put_user(struct tun_struct *tun,
1235 			    struct tun_file *tfile,
1236 			    struct sk_buff *skb,
1237 			    const struct iovec *iv, int len)
1238 {
1239 	struct tun_pi pi = { 0, skb->protocol };
1240 	ssize_t total = 0;
1241 
1242 	if (!(tun->flags & TUN_NO_PI)) {
1243 		if ((len -= sizeof(pi)) < 0)
1244 			return -EINVAL;
1245 
1246 		if (len < skb->len) {
1247 			/* Packet will be striped */
1248 			pi.flags |= TUN_PKT_STRIP;
1249 		}
1250 
1251 		if (memcpy_toiovecend(iv, (void *) &pi, 0, sizeof(pi)))
1252 			return -EFAULT;
1253 		total += sizeof(pi);
1254 	}
1255 
1256 	if (tun->flags & TUN_VNET_HDR) {
1257 		struct virtio_net_hdr gso = { 0 }; /* no info leak */
1258 		if ((len -= tun->vnet_hdr_sz) < 0)
1259 			return -EINVAL;
1260 
1261 		if (skb_is_gso(skb)) {
1262 			struct skb_shared_info *sinfo = skb_shinfo(skb);
1263 
1264 			/* This is a hint as to how much should be linear. */
1265 			gso.hdr_len = skb_headlen(skb);
1266 			gso.gso_size = sinfo->gso_size;
1267 			if (sinfo->gso_type & SKB_GSO_TCPV4)
1268 				gso.gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
1269 			else if (sinfo->gso_type & SKB_GSO_TCPV6)
1270 				gso.gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
1271 			else if (sinfo->gso_type & SKB_GSO_UDP)
1272 				gso.gso_type = VIRTIO_NET_HDR_GSO_UDP;
1273 			else {
1274 				pr_err("unexpected GSO type: "
1275 				       "0x%x, gso_size %d, hdr_len %d\n",
1276 				       sinfo->gso_type, gso.gso_size,
1277 				       gso.hdr_len);
1278 				print_hex_dump(KERN_ERR, "tun: ",
1279 					       DUMP_PREFIX_NONE,
1280 					       16, 1, skb->head,
1281 					       min((int)gso.hdr_len, 64), true);
1282 				WARN_ON_ONCE(1);
1283 				return -EINVAL;
1284 			}
1285 			if (sinfo->gso_type & SKB_GSO_TCP_ECN)
1286 				gso.gso_type |= VIRTIO_NET_HDR_GSO_ECN;
1287 		} else
1288 			gso.gso_type = VIRTIO_NET_HDR_GSO_NONE;
1289 
1290 		if (skb->ip_summed == CHECKSUM_PARTIAL) {
1291 			gso.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
1292 			gso.csum_start = skb_checksum_start_offset(skb);
1293 			gso.csum_offset = skb->csum_offset;
1294 		} else if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
1295 			gso.flags = VIRTIO_NET_HDR_F_DATA_VALID;
1296 		} /* else everything is zero */
1297 
1298 		if (unlikely(memcpy_toiovecend(iv, (void *)&gso, total,
1299 					       sizeof(gso))))
1300 			return -EFAULT;
1301 		total += tun->vnet_hdr_sz;
1302 	}
1303 
1304 	len = min_t(int, skb->len, len);
1305 
1306 	skb_copy_datagram_const_iovec(skb, 0, iv, total, len);
1307 	total += skb->len;
1308 
1309 	tun->dev->stats.tx_packets++;
1310 	tun->dev->stats.tx_bytes += len;
1311 
1312 	return total;
1313 }
1314 
1315 static ssize_t tun_do_read(struct tun_struct *tun, struct tun_file *tfile,
1316 			   struct kiocb *iocb, const struct iovec *iv,
1317 			   ssize_t len, int noblock)
1318 {
1319 	DECLARE_WAITQUEUE(wait, current);
1320 	struct sk_buff *skb;
1321 	ssize_t ret = 0;
1322 
1323 	tun_debug(KERN_INFO, tun, "tun_do_read\n");
1324 
1325 	if (unlikely(!noblock))
1326 		add_wait_queue(&tfile->wq.wait, &wait);
1327 	while (len) {
1328 		current->state = TASK_INTERRUPTIBLE;
1329 
1330 		/* Read frames from the queue */
1331 		if (!(skb = skb_dequeue(&tfile->socket.sk->sk_receive_queue))) {
1332 			if (noblock) {
1333 				ret = -EAGAIN;
1334 				break;
1335 			}
1336 			if (signal_pending(current)) {
1337 				ret = -ERESTARTSYS;
1338 				break;
1339 			}
1340 			if (tun->dev->reg_state != NETREG_REGISTERED) {
1341 				ret = -EIO;
1342 				break;
1343 			}
1344 
1345 			/* Nothing to read, let's sleep */
1346 			schedule();
1347 			continue;
1348 		}
1349 
1350 		ret = tun_put_user(tun, tfile, skb, iv, len);
1351 		kfree_skb(skb);
1352 		break;
1353 	}
1354 
1355 	current->state = TASK_RUNNING;
1356 	if (unlikely(!noblock))
1357 		remove_wait_queue(&tfile->wq.wait, &wait);
1358 
1359 	return ret;
1360 }
1361 
1362 static ssize_t tun_chr_aio_read(struct kiocb *iocb, const struct iovec *iv,
1363 			    unsigned long count, loff_t pos)
1364 {
1365 	struct file *file = iocb->ki_filp;
1366 	struct tun_file *tfile = file->private_data;
1367 	struct tun_struct *tun = __tun_get(tfile);
1368 	ssize_t len, ret;
1369 
1370 	if (!tun)
1371 		return -EBADFD;
1372 	len = iov_length(iv, count);
1373 	if (len < 0) {
1374 		ret = -EINVAL;
1375 		goto out;
1376 	}
1377 
1378 	ret = tun_do_read(tun, tfile, iocb, iv, len,
1379 			  file->f_flags & O_NONBLOCK);
1380 	ret = min_t(ssize_t, ret, len);
1381 out:
1382 	tun_put(tun);
1383 	return ret;
1384 }
1385 
1386 static void tun_free_netdev(struct net_device *dev)
1387 {
1388 	struct tun_struct *tun = netdev_priv(dev);
1389 
1390 	BUG_ON(!(list_empty(&tun->disabled)));
1391 	tun_flow_uninit(tun);
1392 	free_netdev(dev);
1393 }
1394 
1395 static void tun_setup(struct net_device *dev)
1396 {
1397 	struct tun_struct *tun = netdev_priv(dev);
1398 
1399 	tun->owner = INVALID_UID;
1400 	tun->group = INVALID_GID;
1401 
1402 	dev->ethtool_ops = &tun_ethtool_ops;
1403 	dev->destructor = tun_free_netdev;
1404 }
1405 
1406 /* Trivial set of netlink ops to allow deleting tun or tap
1407  * device with netlink.
1408  */
1409 static int tun_validate(struct nlattr *tb[], struct nlattr *data[])
1410 {
1411 	return -EINVAL;
1412 }
1413 
1414 static struct rtnl_link_ops tun_link_ops __read_mostly = {
1415 	.kind		= DRV_NAME,
1416 	.priv_size	= sizeof(struct tun_struct),
1417 	.setup		= tun_setup,
1418 	.validate	= tun_validate,
1419 };
1420 
1421 static void tun_sock_write_space(struct sock *sk)
1422 {
1423 	struct tun_file *tfile;
1424 	wait_queue_head_t *wqueue;
1425 
1426 	if (!sock_writeable(sk))
1427 		return;
1428 
1429 	if (!test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags))
1430 		return;
1431 
1432 	wqueue = sk_sleep(sk);
1433 	if (wqueue && waitqueue_active(wqueue))
1434 		wake_up_interruptible_sync_poll(wqueue, POLLOUT |
1435 						POLLWRNORM | POLLWRBAND);
1436 
1437 	tfile = container_of(sk, struct tun_file, sk);
1438 	kill_fasync(&tfile->fasync, SIGIO, POLL_OUT);
1439 }
1440 
1441 static int tun_sendmsg(struct kiocb *iocb, struct socket *sock,
1442 		       struct msghdr *m, size_t total_len)
1443 {
1444 	int ret;
1445 	struct tun_file *tfile = container_of(sock, struct tun_file, socket);
1446 	struct tun_struct *tun = __tun_get(tfile);
1447 
1448 	if (!tun)
1449 		return -EBADFD;
1450 	ret = tun_get_user(tun, tfile, m->msg_control, m->msg_iov, total_len,
1451 			   m->msg_iovlen, m->msg_flags & MSG_DONTWAIT);
1452 	tun_put(tun);
1453 	return ret;
1454 }
1455 
1456 
1457 static int tun_recvmsg(struct kiocb *iocb, struct socket *sock,
1458 		       struct msghdr *m, size_t total_len,
1459 		       int flags)
1460 {
1461 	struct tun_file *tfile = container_of(sock, struct tun_file, socket);
1462 	struct tun_struct *tun = __tun_get(tfile);
1463 	int ret;
1464 
1465 	if (!tun)
1466 		return -EBADFD;
1467 
1468 	if (flags & ~(MSG_DONTWAIT|MSG_TRUNC))
1469 		return -EINVAL;
1470 	ret = tun_do_read(tun, tfile, iocb, m->msg_iov, total_len,
1471 			  flags & MSG_DONTWAIT);
1472 	if (ret > total_len) {
1473 		m->msg_flags |= MSG_TRUNC;
1474 		ret = flags & MSG_TRUNC ? ret : total_len;
1475 	}
1476 	tun_put(tun);
1477 	return ret;
1478 }
1479 
1480 static int tun_release(struct socket *sock)
1481 {
1482 	if (sock->sk)
1483 		sock_put(sock->sk);
1484 	return 0;
1485 }
1486 
1487 /* Ops structure to mimic raw sockets with tun */
1488 static const struct proto_ops tun_socket_ops = {
1489 	.sendmsg = tun_sendmsg,
1490 	.recvmsg = tun_recvmsg,
1491 	.release = tun_release,
1492 };
1493 
1494 static struct proto tun_proto = {
1495 	.name		= "tun",
1496 	.owner		= THIS_MODULE,
1497 	.obj_size	= sizeof(struct tun_file),
1498 };
1499 
1500 static int tun_flags(struct tun_struct *tun)
1501 {
1502 	int flags = 0;
1503 
1504 	if (tun->flags & TUN_TUN_DEV)
1505 		flags |= IFF_TUN;
1506 	else
1507 		flags |= IFF_TAP;
1508 
1509 	if (tun->flags & TUN_NO_PI)
1510 		flags |= IFF_NO_PI;
1511 
1512 	/* This flag has no real effect.  We track the value for backwards
1513 	 * compatibility.
1514 	 */
1515 	if (tun->flags & TUN_ONE_QUEUE)
1516 		flags |= IFF_ONE_QUEUE;
1517 
1518 	if (tun->flags & TUN_VNET_HDR)
1519 		flags |= IFF_VNET_HDR;
1520 
1521 	if (tun->flags & TUN_TAP_MQ)
1522 		flags |= IFF_MULTI_QUEUE;
1523 
1524 	return flags;
1525 }
1526 
1527 static ssize_t tun_show_flags(struct device *dev, struct device_attribute *attr,
1528 			      char *buf)
1529 {
1530 	struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1531 	return sprintf(buf, "0x%x\n", tun_flags(tun));
1532 }
1533 
1534 static ssize_t tun_show_owner(struct device *dev, struct device_attribute *attr,
1535 			      char *buf)
1536 {
1537 	struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1538 	return uid_valid(tun->owner)?
1539 		sprintf(buf, "%u\n",
1540 			from_kuid_munged(current_user_ns(), tun->owner)):
1541 		sprintf(buf, "-1\n");
1542 }
1543 
1544 static ssize_t tun_show_group(struct device *dev, struct device_attribute *attr,
1545 			      char *buf)
1546 {
1547 	struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1548 	return gid_valid(tun->group) ?
1549 		sprintf(buf, "%u\n",
1550 			from_kgid_munged(current_user_ns(), tun->group)):
1551 		sprintf(buf, "-1\n");
1552 }
1553 
1554 static DEVICE_ATTR(tun_flags, 0444, tun_show_flags, NULL);
1555 static DEVICE_ATTR(owner, 0444, tun_show_owner, NULL);
1556 static DEVICE_ATTR(group, 0444, tun_show_group, NULL);
1557 
1558 static int tun_set_iff(struct net *net, struct file *file, struct ifreq *ifr)
1559 {
1560 	struct tun_struct *tun;
1561 	struct tun_file *tfile = file->private_data;
1562 	struct net_device *dev;
1563 	int err;
1564 
1565 	dev = __dev_get_by_name(net, ifr->ifr_name);
1566 	if (dev) {
1567 		if (ifr->ifr_flags & IFF_TUN_EXCL)
1568 			return -EBUSY;
1569 		if ((ifr->ifr_flags & IFF_TUN) && dev->netdev_ops == &tun_netdev_ops)
1570 			tun = netdev_priv(dev);
1571 		else if ((ifr->ifr_flags & IFF_TAP) && dev->netdev_ops == &tap_netdev_ops)
1572 			tun = netdev_priv(dev);
1573 		else
1574 			return -EINVAL;
1575 
1576 		if (tun_not_capable(tun))
1577 			return -EPERM;
1578 		err = security_tun_dev_attach(tfile->socket.sk);
1579 		if (err < 0)
1580 			return err;
1581 
1582 		err = tun_attach(tun, file);
1583 		if (err < 0)
1584 			return err;
1585 
1586 		if (tun->flags & TUN_TAP_MQ &&
1587 		    (tun->numqueues + tun->numdisabled > 1))
1588 			return err;
1589 	}
1590 	else {
1591 		char *name;
1592 		unsigned long flags = 0;
1593 
1594 		if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
1595 			return -EPERM;
1596 		err = security_tun_dev_create();
1597 		if (err < 0)
1598 			return err;
1599 
1600 		/* Set dev type */
1601 		if (ifr->ifr_flags & IFF_TUN) {
1602 			/* TUN device */
1603 			flags |= TUN_TUN_DEV;
1604 			name = "tun%d";
1605 		} else if (ifr->ifr_flags & IFF_TAP) {
1606 			/* TAP device */
1607 			flags |= TUN_TAP_DEV;
1608 			name = "tap%d";
1609 		} else
1610 			return -EINVAL;
1611 
1612 		if (*ifr->ifr_name)
1613 			name = ifr->ifr_name;
1614 
1615 		dev = alloc_netdev_mqs(sizeof(struct tun_struct), name,
1616 				       tun_setup,
1617 				       MAX_TAP_QUEUES, MAX_TAP_QUEUES);
1618 		if (!dev)
1619 			return -ENOMEM;
1620 
1621 		dev_net_set(dev, net);
1622 		dev->rtnl_link_ops = &tun_link_ops;
1623 
1624 		tun = netdev_priv(dev);
1625 		tun->dev = dev;
1626 		tun->flags = flags;
1627 		tun->txflt.count = 0;
1628 		tun->vnet_hdr_sz = sizeof(struct virtio_net_hdr);
1629 
1630 		tun->filter_attached = false;
1631 		tun->sndbuf = tfile->socket.sk->sk_sndbuf;
1632 
1633 		spin_lock_init(&tun->lock);
1634 
1635 		security_tun_dev_post_create(&tfile->sk);
1636 
1637 		tun_net_init(dev);
1638 
1639 		err = tun_flow_init(tun);
1640 		if (err < 0)
1641 			goto err_free_dev;
1642 
1643 		dev->hw_features = NETIF_F_SG | NETIF_F_FRAGLIST |
1644 			TUN_USER_FEATURES;
1645 		dev->features = dev->hw_features;
1646 
1647 		INIT_LIST_HEAD(&tun->disabled);
1648 		err = tun_attach(tun, file);
1649 		if (err < 0)
1650 			goto err_free_dev;
1651 
1652 		err = register_netdevice(tun->dev);
1653 		if (err < 0)
1654 			goto err_free_dev;
1655 
1656 		if (device_create_file(&tun->dev->dev, &dev_attr_tun_flags) ||
1657 		    device_create_file(&tun->dev->dev, &dev_attr_owner) ||
1658 		    device_create_file(&tun->dev->dev, &dev_attr_group))
1659 			pr_err("Failed to create tun sysfs files\n");
1660 
1661 		netif_carrier_on(tun->dev);
1662 	}
1663 
1664 	tun_debug(KERN_INFO, tun, "tun_set_iff\n");
1665 
1666 	if (ifr->ifr_flags & IFF_NO_PI)
1667 		tun->flags |= TUN_NO_PI;
1668 	else
1669 		tun->flags &= ~TUN_NO_PI;
1670 
1671 	/* This flag has no real effect.  We track the value for backwards
1672 	 * compatibility.
1673 	 */
1674 	if (ifr->ifr_flags & IFF_ONE_QUEUE)
1675 		tun->flags |= TUN_ONE_QUEUE;
1676 	else
1677 		tun->flags &= ~TUN_ONE_QUEUE;
1678 
1679 	if (ifr->ifr_flags & IFF_VNET_HDR)
1680 		tun->flags |= TUN_VNET_HDR;
1681 	else
1682 		tun->flags &= ~TUN_VNET_HDR;
1683 
1684 	if (ifr->ifr_flags & IFF_MULTI_QUEUE)
1685 		tun->flags |= TUN_TAP_MQ;
1686 	else
1687 		tun->flags &= ~TUN_TAP_MQ;
1688 
1689 	/* Make sure persistent devices do not get stuck in
1690 	 * xoff state.
1691 	 */
1692 	if (netif_running(tun->dev))
1693 		netif_tx_wake_all_queues(tun->dev);
1694 
1695 	strcpy(ifr->ifr_name, tun->dev->name);
1696 	return 0;
1697 
1698  err_free_dev:
1699 	free_netdev(dev);
1700 	return err;
1701 }
1702 
1703 static void tun_get_iff(struct net *net, struct tun_struct *tun,
1704 		       struct ifreq *ifr)
1705 {
1706 	tun_debug(KERN_INFO, tun, "tun_get_iff\n");
1707 
1708 	strcpy(ifr->ifr_name, tun->dev->name);
1709 
1710 	ifr->ifr_flags = tun_flags(tun);
1711 
1712 }
1713 
1714 /* This is like a cut-down ethtool ops, except done via tun fd so no
1715  * privs required. */
1716 static int set_offload(struct tun_struct *tun, unsigned long arg)
1717 {
1718 	netdev_features_t features = 0;
1719 
1720 	if (arg & TUN_F_CSUM) {
1721 		features |= NETIF_F_HW_CSUM;
1722 		arg &= ~TUN_F_CSUM;
1723 
1724 		if (arg & (TUN_F_TSO4|TUN_F_TSO6)) {
1725 			if (arg & TUN_F_TSO_ECN) {
1726 				features |= NETIF_F_TSO_ECN;
1727 				arg &= ~TUN_F_TSO_ECN;
1728 			}
1729 			if (arg & TUN_F_TSO4)
1730 				features |= NETIF_F_TSO;
1731 			if (arg & TUN_F_TSO6)
1732 				features |= NETIF_F_TSO6;
1733 			arg &= ~(TUN_F_TSO4|TUN_F_TSO6);
1734 		}
1735 
1736 		if (arg & TUN_F_UFO) {
1737 			features |= NETIF_F_UFO;
1738 			arg &= ~TUN_F_UFO;
1739 		}
1740 	}
1741 
1742 	/* This gives the user a way to test for new features in future by
1743 	 * trying to set them. */
1744 	if (arg)
1745 		return -EINVAL;
1746 
1747 	tun->set_features = features;
1748 	netdev_update_features(tun->dev);
1749 
1750 	return 0;
1751 }
1752 
1753 static void tun_detach_filter(struct tun_struct *tun, int n)
1754 {
1755 	int i;
1756 	struct tun_file *tfile;
1757 
1758 	for (i = 0; i < n; i++) {
1759 		tfile = rcu_dereference_protected(tun->tfiles[i],
1760 						  lockdep_rtnl_is_held());
1761 		sk_detach_filter(tfile->socket.sk);
1762 	}
1763 
1764 	tun->filter_attached = false;
1765 }
1766 
1767 static int tun_attach_filter(struct tun_struct *tun)
1768 {
1769 	int i, ret = 0;
1770 	struct tun_file *tfile;
1771 
1772 	for (i = 0; i < tun->numqueues; i++) {
1773 		tfile = rcu_dereference_protected(tun->tfiles[i],
1774 						  lockdep_rtnl_is_held());
1775 		ret = sk_attach_filter(&tun->fprog, tfile->socket.sk);
1776 		if (ret) {
1777 			tun_detach_filter(tun, i);
1778 			return ret;
1779 		}
1780 	}
1781 
1782 	tun->filter_attached = true;
1783 	return ret;
1784 }
1785 
1786 static void tun_set_sndbuf(struct tun_struct *tun)
1787 {
1788 	struct tun_file *tfile;
1789 	int i;
1790 
1791 	for (i = 0; i < tun->numqueues; i++) {
1792 		tfile = rcu_dereference_protected(tun->tfiles[i],
1793 						lockdep_rtnl_is_held());
1794 		tfile->socket.sk->sk_sndbuf = tun->sndbuf;
1795 	}
1796 }
1797 
1798 static int tun_set_queue(struct file *file, struct ifreq *ifr)
1799 {
1800 	struct tun_file *tfile = file->private_data;
1801 	struct tun_struct *tun;
1802 	int ret = 0;
1803 
1804 	rtnl_lock();
1805 
1806 	if (ifr->ifr_flags & IFF_ATTACH_QUEUE) {
1807 		tun = tfile->detached;
1808 		if (!tun)
1809 			ret = -EINVAL;
1810 		else if (tun_not_capable(tun))
1811 			ret = -EPERM;
1812 		else
1813 			ret = tun_attach(tun, file);
1814 	} else if (ifr->ifr_flags & IFF_DETACH_QUEUE) {
1815 		tun = rcu_dereference_protected(tfile->tun,
1816 						lockdep_rtnl_is_held());
1817 		if (!tun || !(tun->flags & TUN_TAP_MQ))
1818 			ret = -EINVAL;
1819 		else
1820 			__tun_detach(tfile, false);
1821 	} else
1822 		ret = -EINVAL;
1823 
1824 	rtnl_unlock();
1825 	return ret;
1826 }
1827 
1828 static long __tun_chr_ioctl(struct file *file, unsigned int cmd,
1829 			    unsigned long arg, int ifreq_len)
1830 {
1831 	struct tun_file *tfile = file->private_data;
1832 	struct tun_struct *tun;
1833 	void __user* argp = (void __user*)arg;
1834 	struct ifreq ifr;
1835 	kuid_t owner;
1836 	kgid_t group;
1837 	int sndbuf;
1838 	int vnet_hdr_sz;
1839 	int ret;
1840 
1841 	if (cmd == TUNSETIFF || cmd == TUNSETQUEUE || _IOC_TYPE(cmd) == 0x89) {
1842 		if (copy_from_user(&ifr, argp, ifreq_len))
1843 			return -EFAULT;
1844 	} else {
1845 		memset(&ifr, 0, sizeof(ifr));
1846 	}
1847 	if (cmd == TUNGETFEATURES) {
1848 		/* Currently this just means: "what IFF flags are valid?".
1849 		 * This is needed because we never checked for invalid flags on
1850 		 * TUNSETIFF. */
1851 		return put_user(IFF_TUN | IFF_TAP | IFF_NO_PI | IFF_ONE_QUEUE |
1852 				IFF_VNET_HDR | IFF_MULTI_QUEUE,
1853 				(unsigned int __user*)argp);
1854 	} else if (cmd == TUNSETQUEUE)
1855 		return tun_set_queue(file, &ifr);
1856 
1857 	ret = 0;
1858 	rtnl_lock();
1859 
1860 	tun = __tun_get(tfile);
1861 	if (cmd == TUNSETIFF && !tun) {
1862 		ifr.ifr_name[IFNAMSIZ-1] = '\0';
1863 
1864 		ret = tun_set_iff(tfile->net, file, &ifr);
1865 
1866 		if (ret)
1867 			goto unlock;
1868 
1869 		if (copy_to_user(argp, &ifr, ifreq_len))
1870 			ret = -EFAULT;
1871 		goto unlock;
1872 	}
1873 
1874 	ret = -EBADFD;
1875 	if (!tun)
1876 		goto unlock;
1877 
1878 	tun_debug(KERN_INFO, tun, "tun_chr_ioctl cmd %u\n", cmd);
1879 
1880 	ret = 0;
1881 	switch (cmd) {
1882 	case TUNGETIFF:
1883 		tun_get_iff(current->nsproxy->net_ns, tun, &ifr);
1884 
1885 		if (copy_to_user(argp, &ifr, ifreq_len))
1886 			ret = -EFAULT;
1887 		break;
1888 
1889 	case TUNSETNOCSUM:
1890 		/* Disable/Enable checksum */
1891 
1892 		/* [unimplemented] */
1893 		tun_debug(KERN_INFO, tun, "ignored: set checksum %s\n",
1894 			  arg ? "disabled" : "enabled");
1895 		break;
1896 
1897 	case TUNSETPERSIST:
1898 		/* Disable/Enable persist mode. Keep an extra reference to the
1899 		 * module to prevent the module being unprobed.
1900 		 */
1901 		if (arg) {
1902 			tun->flags |= TUN_PERSIST;
1903 			__module_get(THIS_MODULE);
1904 		} else {
1905 			tun->flags &= ~TUN_PERSIST;
1906 			module_put(THIS_MODULE);
1907 		}
1908 
1909 		tun_debug(KERN_INFO, tun, "persist %s\n",
1910 			  arg ? "enabled" : "disabled");
1911 		break;
1912 
1913 	case TUNSETOWNER:
1914 		/* Set owner of the device */
1915 		owner = make_kuid(current_user_ns(), arg);
1916 		if (!uid_valid(owner)) {
1917 			ret = -EINVAL;
1918 			break;
1919 		}
1920 		tun->owner = owner;
1921 		tun_debug(KERN_INFO, tun, "owner set to %u\n",
1922 			  from_kuid(&init_user_ns, tun->owner));
1923 		break;
1924 
1925 	case TUNSETGROUP:
1926 		/* Set group of the device */
1927 		group = make_kgid(current_user_ns(), arg);
1928 		if (!gid_valid(group)) {
1929 			ret = -EINVAL;
1930 			break;
1931 		}
1932 		tun->group = group;
1933 		tun_debug(KERN_INFO, tun, "group set to %u\n",
1934 			  from_kgid(&init_user_ns, tun->group));
1935 		break;
1936 
1937 	case TUNSETLINK:
1938 		/* Only allow setting the type when the interface is down */
1939 		if (tun->dev->flags & IFF_UP) {
1940 			tun_debug(KERN_INFO, tun,
1941 				  "Linktype set failed because interface is up\n");
1942 			ret = -EBUSY;
1943 		} else {
1944 			tun->dev->type = (int) arg;
1945 			tun_debug(KERN_INFO, tun, "linktype set to %d\n",
1946 				  tun->dev->type);
1947 			ret = 0;
1948 		}
1949 		break;
1950 
1951 #ifdef TUN_DEBUG
1952 	case TUNSETDEBUG:
1953 		tun->debug = arg;
1954 		break;
1955 #endif
1956 	case TUNSETOFFLOAD:
1957 		ret = set_offload(tun, arg);
1958 		break;
1959 
1960 	case TUNSETTXFILTER:
1961 		/* Can be set only for TAPs */
1962 		ret = -EINVAL;
1963 		if ((tun->flags & TUN_TYPE_MASK) != TUN_TAP_DEV)
1964 			break;
1965 		ret = update_filter(&tun->txflt, (void __user *)arg);
1966 		break;
1967 
1968 	case SIOCGIFHWADDR:
1969 		/* Get hw address */
1970 		memcpy(ifr.ifr_hwaddr.sa_data, tun->dev->dev_addr, ETH_ALEN);
1971 		ifr.ifr_hwaddr.sa_family = tun->dev->type;
1972 		if (copy_to_user(argp, &ifr, ifreq_len))
1973 			ret = -EFAULT;
1974 		break;
1975 
1976 	case SIOCSIFHWADDR:
1977 		/* Set hw address */
1978 		tun_debug(KERN_DEBUG, tun, "set hw address: %pM\n",
1979 			  ifr.ifr_hwaddr.sa_data);
1980 
1981 		ret = dev_set_mac_address(tun->dev, &ifr.ifr_hwaddr);
1982 		break;
1983 
1984 	case TUNGETSNDBUF:
1985 		sndbuf = tfile->socket.sk->sk_sndbuf;
1986 		if (copy_to_user(argp, &sndbuf, sizeof(sndbuf)))
1987 			ret = -EFAULT;
1988 		break;
1989 
1990 	case TUNSETSNDBUF:
1991 		if (copy_from_user(&sndbuf, argp, sizeof(sndbuf))) {
1992 			ret = -EFAULT;
1993 			break;
1994 		}
1995 
1996 		tun->sndbuf = sndbuf;
1997 		tun_set_sndbuf(tun);
1998 		break;
1999 
2000 	case TUNGETVNETHDRSZ:
2001 		vnet_hdr_sz = tun->vnet_hdr_sz;
2002 		if (copy_to_user(argp, &vnet_hdr_sz, sizeof(vnet_hdr_sz)))
2003 			ret = -EFAULT;
2004 		break;
2005 
2006 	case TUNSETVNETHDRSZ:
2007 		if (copy_from_user(&vnet_hdr_sz, argp, sizeof(vnet_hdr_sz))) {
2008 			ret = -EFAULT;
2009 			break;
2010 		}
2011 		if (vnet_hdr_sz < (int)sizeof(struct virtio_net_hdr)) {
2012 			ret = -EINVAL;
2013 			break;
2014 		}
2015 
2016 		tun->vnet_hdr_sz = vnet_hdr_sz;
2017 		break;
2018 
2019 	case TUNATTACHFILTER:
2020 		/* Can be set only for TAPs */
2021 		ret = -EINVAL;
2022 		if ((tun->flags & TUN_TYPE_MASK) != TUN_TAP_DEV)
2023 			break;
2024 		ret = -EFAULT;
2025 		if (copy_from_user(&tun->fprog, argp, sizeof(tun->fprog)))
2026 			break;
2027 
2028 		ret = tun_attach_filter(tun);
2029 		break;
2030 
2031 	case TUNDETACHFILTER:
2032 		/* Can be set only for TAPs */
2033 		ret = -EINVAL;
2034 		if ((tun->flags & TUN_TYPE_MASK) != TUN_TAP_DEV)
2035 			break;
2036 		ret = 0;
2037 		tun_detach_filter(tun, tun->numqueues);
2038 		break;
2039 
2040 	default:
2041 		ret = -EINVAL;
2042 		break;
2043 	}
2044 
2045 unlock:
2046 	rtnl_unlock();
2047 	if (tun)
2048 		tun_put(tun);
2049 	return ret;
2050 }
2051 
2052 static long tun_chr_ioctl(struct file *file,
2053 			  unsigned int cmd, unsigned long arg)
2054 {
2055 	return __tun_chr_ioctl(file, cmd, arg, sizeof (struct ifreq));
2056 }
2057 
2058 #ifdef CONFIG_COMPAT
2059 static long tun_chr_compat_ioctl(struct file *file,
2060 			 unsigned int cmd, unsigned long arg)
2061 {
2062 	switch (cmd) {
2063 	case TUNSETIFF:
2064 	case TUNGETIFF:
2065 	case TUNSETTXFILTER:
2066 	case TUNGETSNDBUF:
2067 	case TUNSETSNDBUF:
2068 	case SIOCGIFHWADDR:
2069 	case SIOCSIFHWADDR:
2070 		arg = (unsigned long)compat_ptr(arg);
2071 		break;
2072 	default:
2073 		arg = (compat_ulong_t)arg;
2074 		break;
2075 	}
2076 
2077 	/*
2078 	 * compat_ifreq is shorter than ifreq, so we must not access beyond
2079 	 * the end of that structure. All fields that are used in this
2080 	 * driver are compatible though, we don't need to convert the
2081 	 * contents.
2082 	 */
2083 	return __tun_chr_ioctl(file, cmd, arg, sizeof(struct compat_ifreq));
2084 }
2085 #endif /* CONFIG_COMPAT */
2086 
2087 static int tun_chr_fasync(int fd, struct file *file, int on)
2088 {
2089 	struct tun_file *tfile = file->private_data;
2090 	int ret;
2091 
2092 	if ((ret = fasync_helper(fd, file, on, &tfile->fasync)) < 0)
2093 		goto out;
2094 
2095 	if (on) {
2096 		ret = __f_setown(file, task_pid(current), PIDTYPE_PID, 0);
2097 		if (ret)
2098 			goto out;
2099 		tfile->flags |= TUN_FASYNC;
2100 	} else
2101 		tfile->flags &= ~TUN_FASYNC;
2102 	ret = 0;
2103 out:
2104 	return ret;
2105 }
2106 
2107 static int tun_chr_open(struct inode *inode, struct file * file)
2108 {
2109 	struct tun_file *tfile;
2110 
2111 	DBG1(KERN_INFO, "tunX: tun_chr_open\n");
2112 
2113 	tfile = (struct tun_file *)sk_alloc(&init_net, AF_UNSPEC, GFP_KERNEL,
2114 					    &tun_proto);
2115 	if (!tfile)
2116 		return -ENOMEM;
2117 	rcu_assign_pointer(tfile->tun, NULL);
2118 	tfile->net = get_net(current->nsproxy->net_ns);
2119 	tfile->flags = 0;
2120 
2121 	rcu_assign_pointer(tfile->socket.wq, &tfile->wq);
2122 	init_waitqueue_head(&tfile->wq.wait);
2123 
2124 	tfile->socket.file = file;
2125 	tfile->socket.ops = &tun_socket_ops;
2126 
2127 	sock_init_data(&tfile->socket, &tfile->sk);
2128 	sk_change_net(&tfile->sk, tfile->net);
2129 
2130 	tfile->sk.sk_write_space = tun_sock_write_space;
2131 	tfile->sk.sk_sndbuf = INT_MAX;
2132 
2133 	file->private_data = tfile;
2134 	set_bit(SOCK_EXTERNALLY_ALLOCATED, &tfile->socket.flags);
2135 	INIT_LIST_HEAD(&tfile->next);
2136 
2137 	return 0;
2138 }
2139 
2140 static int tun_chr_close(struct inode *inode, struct file *file)
2141 {
2142 	struct tun_file *tfile = file->private_data;
2143 	struct net *net = tfile->net;
2144 
2145 	tun_detach(tfile, true);
2146 	put_net(net);
2147 
2148 	return 0;
2149 }
2150 
2151 static const struct file_operations tun_fops = {
2152 	.owner	= THIS_MODULE,
2153 	.llseek = no_llseek,
2154 	.read  = do_sync_read,
2155 	.aio_read  = tun_chr_aio_read,
2156 	.write = do_sync_write,
2157 	.aio_write = tun_chr_aio_write,
2158 	.poll	= tun_chr_poll,
2159 	.unlocked_ioctl	= tun_chr_ioctl,
2160 #ifdef CONFIG_COMPAT
2161 	.compat_ioctl = tun_chr_compat_ioctl,
2162 #endif
2163 	.open	= tun_chr_open,
2164 	.release = tun_chr_close,
2165 	.fasync = tun_chr_fasync
2166 };
2167 
2168 static struct miscdevice tun_miscdev = {
2169 	.minor = TUN_MINOR,
2170 	.name = "tun",
2171 	.nodename = "net/tun",
2172 	.fops = &tun_fops,
2173 };
2174 
2175 /* ethtool interface */
2176 
2177 static int tun_get_settings(struct net_device *dev, struct ethtool_cmd *cmd)
2178 {
2179 	cmd->supported		= 0;
2180 	cmd->advertising	= 0;
2181 	ethtool_cmd_speed_set(cmd, SPEED_10);
2182 	cmd->duplex		= DUPLEX_FULL;
2183 	cmd->port		= PORT_TP;
2184 	cmd->phy_address	= 0;
2185 	cmd->transceiver	= XCVR_INTERNAL;
2186 	cmd->autoneg		= AUTONEG_DISABLE;
2187 	cmd->maxtxpkt		= 0;
2188 	cmd->maxrxpkt		= 0;
2189 	return 0;
2190 }
2191 
2192 static void tun_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
2193 {
2194 	struct tun_struct *tun = netdev_priv(dev);
2195 
2196 	strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
2197 	strlcpy(info->version, DRV_VERSION, sizeof(info->version));
2198 
2199 	switch (tun->flags & TUN_TYPE_MASK) {
2200 	case TUN_TUN_DEV:
2201 		strlcpy(info->bus_info, "tun", sizeof(info->bus_info));
2202 		break;
2203 	case TUN_TAP_DEV:
2204 		strlcpy(info->bus_info, "tap", sizeof(info->bus_info));
2205 		break;
2206 	}
2207 }
2208 
2209 static u32 tun_get_msglevel(struct net_device *dev)
2210 {
2211 #ifdef TUN_DEBUG
2212 	struct tun_struct *tun = netdev_priv(dev);
2213 	return tun->debug;
2214 #else
2215 	return -EOPNOTSUPP;
2216 #endif
2217 }
2218 
2219 static void tun_set_msglevel(struct net_device *dev, u32 value)
2220 {
2221 #ifdef TUN_DEBUG
2222 	struct tun_struct *tun = netdev_priv(dev);
2223 	tun->debug = value;
2224 #endif
2225 }
2226 
2227 static const struct ethtool_ops tun_ethtool_ops = {
2228 	.get_settings	= tun_get_settings,
2229 	.get_drvinfo	= tun_get_drvinfo,
2230 	.get_msglevel	= tun_get_msglevel,
2231 	.set_msglevel	= tun_set_msglevel,
2232 	.get_link	= ethtool_op_get_link,
2233 };
2234 
2235 
2236 static int __init tun_init(void)
2237 {
2238 	int ret = 0;
2239 
2240 	pr_info("%s, %s\n", DRV_DESCRIPTION, DRV_VERSION);
2241 	pr_info("%s\n", DRV_COPYRIGHT);
2242 
2243 	ret = rtnl_link_register(&tun_link_ops);
2244 	if (ret) {
2245 		pr_err("Can't register link_ops\n");
2246 		goto err_linkops;
2247 	}
2248 
2249 	ret = misc_register(&tun_miscdev);
2250 	if (ret) {
2251 		pr_err("Can't register misc device %d\n", TUN_MINOR);
2252 		goto err_misc;
2253 	}
2254 	return  0;
2255 err_misc:
2256 	rtnl_link_unregister(&tun_link_ops);
2257 err_linkops:
2258 	return ret;
2259 }
2260 
2261 static void tun_cleanup(void)
2262 {
2263 	misc_deregister(&tun_miscdev);
2264 	rtnl_link_unregister(&tun_link_ops);
2265 }
2266 
2267 /* Get an underlying socket object from tun file.  Returns error unless file is
2268  * attached to a device.  The returned object works like a packet socket, it
2269  * can be used for sock_sendmsg/sock_recvmsg.  The caller is responsible for
2270  * holding a reference to the file for as long as the socket is in use. */
2271 struct socket *tun_get_socket(struct file *file)
2272 {
2273 	struct tun_file *tfile;
2274 	if (file->f_op != &tun_fops)
2275 		return ERR_PTR(-EINVAL);
2276 	tfile = file->private_data;
2277 	if (!tfile)
2278 		return ERR_PTR(-EBADFD);
2279 	return &tfile->socket;
2280 }
2281 EXPORT_SYMBOL_GPL(tun_get_socket);
2282 
2283 module_init(tun_init);
2284 module_exit(tun_cleanup);
2285 MODULE_DESCRIPTION(DRV_DESCRIPTION);
2286 MODULE_AUTHOR(DRV_COPYRIGHT);
2287 MODULE_LICENSE("GPL");
2288 MODULE_ALIAS_MISCDEV(TUN_MINOR);
2289 MODULE_ALIAS("devname:net/tun");
2290