xref: /linux/drivers/net/tun.c (revision 2d7ce0e8a704d1aedc4462817a4eade1c3b00fbe)
1 /*
2  *  TUN - Universal TUN/TAP device driver.
3  *  Copyright (C) 1999-2002 Maxim Krasnyansky <maxk@qualcomm.com>
4  *
5  *  This program is free software; you can redistribute it and/or modify
6  *  it under the terms of the GNU General Public License as published by
7  *  the Free Software Foundation; either version 2 of the License, or
8  *  (at your option) any later version.
9  *
10  *  This program is distributed in the hope that it will be useful,
11  *  but WITHOUT ANY WARRANTY; without even the implied warranty of
12  *  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
13  *  GNU General Public License for more details.
14  *
15  *  $Id: tun.c,v 1.15 2002/03/01 02:44:24 maxk Exp $
16  */
17 
18 /*
19  *  Changes:
20  *
21  *  Mike Kershaw <dragorn@kismetwireless.net> 2005/08/14
22  *    Add TUNSETLINK ioctl to set the link encapsulation
23  *
24  *  Mark Smith <markzzzsmith@yahoo.com.au>
25  *    Use eth_random_addr() for tap MAC address.
26  *
27  *  Harald Roelle <harald.roelle@ifi.lmu.de>  2004/04/20
28  *    Fixes in packet dropping, queue length setting and queue wakeup.
29  *    Increased default tx queue length.
30  *    Added ethtool API.
31  *    Minor cleanups
32  *
33  *  Daniel Podlejski <underley@underley.eu.org>
34  *    Modifications for 2.3.99-pre5 kernel.
35  */
36 
37 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
38 
39 #define DRV_NAME	"tun"
40 #define DRV_VERSION	"1.6"
41 #define DRV_DESCRIPTION	"Universal TUN/TAP device driver"
42 #define DRV_COPYRIGHT	"(C) 1999-2004 Max Krasnyansky <maxk@qualcomm.com>"
43 
44 #include <linux/module.h>
45 #include <linux/errno.h>
46 #include <linux/kernel.h>
47 #include <linux/major.h>
48 #include <linux/slab.h>
49 #include <linux/poll.h>
50 #include <linux/fcntl.h>
51 #include <linux/init.h>
52 #include <linux/skbuff.h>
53 #include <linux/netdevice.h>
54 #include <linux/etherdevice.h>
55 #include <linux/miscdevice.h>
56 #include <linux/ethtool.h>
57 #include <linux/rtnetlink.h>
58 #include <linux/compat.h>
59 #include <linux/if.h>
60 #include <linux/if_arp.h>
61 #include <linux/if_ether.h>
62 #include <linux/if_tun.h>
63 #include <linux/if_vlan.h>
64 #include <linux/crc32.h>
65 #include <linux/nsproxy.h>
66 #include <linux/virtio_net.h>
67 #include <linux/rcupdate.h>
68 #include <net/ipv6.h>
69 #include <net/net_namespace.h>
70 #include <net/netns/generic.h>
71 #include <net/rtnetlink.h>
72 #include <net/sock.h>
73 #include <linux/seq_file.h>
74 
75 #include <asm/uaccess.h>
76 
77 /* Uncomment to enable debugging */
78 /* #define TUN_DEBUG 1 */
79 
80 #ifdef TUN_DEBUG
81 static int debug;
82 
83 #define tun_debug(level, tun, fmt, args...)			\
84 do {								\
85 	if (tun->debug)						\
86 		netdev_printk(level, tun->dev, fmt, ##args);	\
87 } while (0)
88 #define DBG1(level, fmt, args...)				\
89 do {								\
90 	if (debug == 2)						\
91 		printk(level fmt, ##args);			\
92 } while (0)
93 #else
94 #define tun_debug(level, tun, fmt, args...)			\
95 do {								\
96 	if (0)							\
97 		netdev_printk(level, tun->dev, fmt, ##args);	\
98 } while (0)
99 #define DBG1(level, fmt, args...)				\
100 do {								\
101 	if (0)							\
102 		printk(level fmt, ##args);			\
103 } while (0)
104 #endif
105 
106 /* TUN device flags */
107 
108 /* IFF_ATTACH_QUEUE is never stored in device flags,
109  * overload it to mean fasync when stored there.
110  */
111 #define TUN_FASYNC	IFF_ATTACH_QUEUE
112 
113 #define TUN_FEATURES (IFF_NO_PI | IFF_ONE_QUEUE | IFF_VNET_HDR | \
114 		      IFF_VNET_LE | IFF_MULTI_QUEUE)
115 #define GOODCOPY_LEN 128
116 
117 #define FLT_EXACT_COUNT 8
118 struct tap_filter {
119 	unsigned int    count;    /* Number of addrs. Zero means disabled */
120 	u32             mask[2];  /* Mask of the hashed addrs */
121 	unsigned char	addr[FLT_EXACT_COUNT][ETH_ALEN];
122 };
123 
124 /* DEFAULT_MAX_NUM_RSS_QUEUES were chosen to let the rx/tx queues allocated for
125  * the netdevice to be fit in one page. So we can make sure the success of
126  * memory allocation. TODO: increase the limit. */
127 #define MAX_TAP_QUEUES DEFAULT_MAX_NUM_RSS_QUEUES
128 #define MAX_TAP_FLOWS  4096
129 
130 #define TUN_FLOW_EXPIRE (3 * HZ)
131 
132 /* A tun_file connects an open character device to a tuntap netdevice. It
133  * also contains all socket related structures (except sock_fprog and tap_filter)
134  * to serve as one transmit queue for tuntap device. The sock_fprog and
135  * tap_filter were kept in tun_struct since they were used for filtering for the
136  * netdevice not for a specific queue (at least I didn't see the requirement for
137  * this).
138  *
139  * RCU usage:
140  * The tun_file and tun_struct are loosely coupled, the pointer from one to the
141  * other can only be read while rcu_read_lock or rtnl_lock is held.
142  */
143 struct tun_file {
144 	struct sock sk;
145 	struct socket socket;
146 	struct socket_wq wq;
147 	struct tun_struct __rcu *tun;
148 	struct net *net;
149 	struct fasync_struct *fasync;
150 	/* only used for fasnyc */
151 	unsigned int flags;
152 	union {
153 		u16 queue_index;
154 		unsigned int ifindex;
155 	};
156 	struct list_head next;
157 	struct tun_struct *detached;
158 };
159 
160 struct tun_flow_entry {
161 	struct hlist_node hash_link;
162 	struct rcu_head rcu;
163 	struct tun_struct *tun;
164 
165 	u32 rxhash;
166 	u32 rps_rxhash;
167 	int queue_index;
168 	unsigned long updated;
169 };
170 
171 #define TUN_NUM_FLOW_ENTRIES 1024
172 
173 /* Since the socket were moved to tun_file, to preserve the behavior of persist
174  * device, socket filter, sndbuf and vnet header size were restore when the
175  * file were attached to a persist device.
176  */
177 struct tun_struct {
178 	struct tun_file __rcu	*tfiles[MAX_TAP_QUEUES];
179 	unsigned int            numqueues;
180 	unsigned int 		flags;
181 	kuid_t			owner;
182 	kgid_t			group;
183 
184 	struct net_device	*dev;
185 	netdev_features_t	set_features;
186 #define TUN_USER_FEATURES (NETIF_F_HW_CSUM|NETIF_F_TSO_ECN|NETIF_F_TSO| \
187 			  NETIF_F_TSO6)
188 
189 	int			vnet_hdr_sz;
190 	int			sndbuf;
191 	struct tap_filter	txflt;
192 	struct sock_fprog	fprog;
193 	/* protected by rtnl lock */
194 	bool			filter_attached;
195 #ifdef TUN_DEBUG
196 	int debug;
197 #endif
198 	spinlock_t lock;
199 	struct hlist_head flows[TUN_NUM_FLOW_ENTRIES];
200 	struct timer_list flow_gc_timer;
201 	unsigned long ageing_time;
202 	unsigned int numdisabled;
203 	struct list_head disabled;
204 	void *security;
205 	u32 flow_count;
206 };
207 
208 static inline u16 tun16_to_cpu(struct tun_struct *tun, __virtio16 val)
209 {
210 	return __virtio16_to_cpu(tun->flags & IFF_VNET_LE, val);
211 }
212 
213 static inline __virtio16 cpu_to_tun16(struct tun_struct *tun, u16 val)
214 {
215 	return __cpu_to_virtio16(tun->flags & IFF_VNET_LE, val);
216 }
217 
218 static inline u32 tun_hashfn(u32 rxhash)
219 {
220 	return rxhash & 0x3ff;
221 }
222 
223 static struct tun_flow_entry *tun_flow_find(struct hlist_head *head, u32 rxhash)
224 {
225 	struct tun_flow_entry *e;
226 
227 	hlist_for_each_entry_rcu(e, head, hash_link) {
228 		if (e->rxhash == rxhash)
229 			return e;
230 	}
231 	return NULL;
232 }
233 
234 static struct tun_flow_entry *tun_flow_create(struct tun_struct *tun,
235 					      struct hlist_head *head,
236 					      u32 rxhash, u16 queue_index)
237 {
238 	struct tun_flow_entry *e = kmalloc(sizeof(*e), GFP_ATOMIC);
239 
240 	if (e) {
241 		tun_debug(KERN_INFO, tun, "create flow: hash %u index %u\n",
242 			  rxhash, queue_index);
243 		e->updated = jiffies;
244 		e->rxhash = rxhash;
245 		e->rps_rxhash = 0;
246 		e->queue_index = queue_index;
247 		e->tun = tun;
248 		hlist_add_head_rcu(&e->hash_link, head);
249 		++tun->flow_count;
250 	}
251 	return e;
252 }
253 
254 static void tun_flow_delete(struct tun_struct *tun, struct tun_flow_entry *e)
255 {
256 	tun_debug(KERN_INFO, tun, "delete flow: hash %u index %u\n",
257 		  e->rxhash, e->queue_index);
258 	sock_rps_reset_flow_hash(e->rps_rxhash);
259 	hlist_del_rcu(&e->hash_link);
260 	kfree_rcu(e, rcu);
261 	--tun->flow_count;
262 }
263 
264 static void tun_flow_flush(struct tun_struct *tun)
265 {
266 	int i;
267 
268 	spin_lock_bh(&tun->lock);
269 	for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
270 		struct tun_flow_entry *e;
271 		struct hlist_node *n;
272 
273 		hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link)
274 			tun_flow_delete(tun, e);
275 	}
276 	spin_unlock_bh(&tun->lock);
277 }
278 
279 static void tun_flow_delete_by_queue(struct tun_struct *tun, u16 queue_index)
280 {
281 	int i;
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 *n;
287 
288 		hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
289 			if (e->queue_index == queue_index)
290 				tun_flow_delete(tun, e);
291 		}
292 	}
293 	spin_unlock_bh(&tun->lock);
294 }
295 
296 static void tun_flow_cleanup(unsigned long data)
297 {
298 	struct tun_struct *tun = (struct tun_struct *)data;
299 	unsigned long delay = tun->ageing_time;
300 	unsigned long next_timer = jiffies + delay;
301 	unsigned long count = 0;
302 	int i;
303 
304 	tun_debug(KERN_INFO, tun, "tun_flow_cleanup\n");
305 
306 	spin_lock_bh(&tun->lock);
307 	for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
308 		struct tun_flow_entry *e;
309 		struct hlist_node *n;
310 
311 		hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
312 			unsigned long this_timer;
313 			count++;
314 			this_timer = e->updated + delay;
315 			if (time_before_eq(this_timer, jiffies))
316 				tun_flow_delete(tun, e);
317 			else if (time_before(this_timer, next_timer))
318 				next_timer = this_timer;
319 		}
320 	}
321 
322 	if (count)
323 		mod_timer(&tun->flow_gc_timer, round_jiffies_up(next_timer));
324 	spin_unlock_bh(&tun->lock);
325 }
326 
327 static void tun_flow_update(struct tun_struct *tun, u32 rxhash,
328 			    struct tun_file *tfile)
329 {
330 	struct hlist_head *head;
331 	struct tun_flow_entry *e;
332 	unsigned long delay = tun->ageing_time;
333 	u16 queue_index = tfile->queue_index;
334 
335 	if (!rxhash)
336 		return;
337 	else
338 		head = &tun->flows[tun_hashfn(rxhash)];
339 
340 	rcu_read_lock();
341 
342 	/* We may get a very small possibility of OOO during switching, not
343 	 * worth to optimize.*/
344 	if (tun->numqueues == 1 || tfile->detached)
345 		goto unlock;
346 
347 	e = tun_flow_find(head, rxhash);
348 	if (likely(e)) {
349 		/* TODO: keep queueing to old queue until it's empty? */
350 		e->queue_index = queue_index;
351 		e->updated = jiffies;
352 		sock_rps_record_flow_hash(e->rps_rxhash);
353 	} else {
354 		spin_lock_bh(&tun->lock);
355 		if (!tun_flow_find(head, rxhash) &&
356 		    tun->flow_count < MAX_TAP_FLOWS)
357 			tun_flow_create(tun, head, rxhash, queue_index);
358 
359 		if (!timer_pending(&tun->flow_gc_timer))
360 			mod_timer(&tun->flow_gc_timer,
361 				  round_jiffies_up(jiffies + delay));
362 		spin_unlock_bh(&tun->lock);
363 	}
364 
365 unlock:
366 	rcu_read_unlock();
367 }
368 
369 /**
370  * Save the hash received in the stack receive path and update the
371  * flow_hash table accordingly.
372  */
373 static inline void tun_flow_save_rps_rxhash(struct tun_flow_entry *e, u32 hash)
374 {
375 	if (unlikely(e->rps_rxhash != hash)) {
376 		sock_rps_reset_flow_hash(e->rps_rxhash);
377 		e->rps_rxhash = hash;
378 	}
379 }
380 
381 /* We try to identify a flow through its rxhash first. The reason that
382  * we do not check rxq no. is because some cards(e.g 82599), chooses
383  * the rxq based on the txq where the last packet of the flow comes. As
384  * the userspace application move between processors, we may get a
385  * different rxq no. here. If we could not get rxhash, then we would
386  * hope the rxq no. may help here.
387  */
388 static u16 tun_select_queue(struct net_device *dev, struct sk_buff *skb,
389 			    void *accel_priv, select_queue_fallback_t fallback)
390 {
391 	struct tun_struct *tun = netdev_priv(dev);
392 	struct tun_flow_entry *e;
393 	u32 txq = 0;
394 	u32 numqueues = 0;
395 
396 	rcu_read_lock();
397 	numqueues = ACCESS_ONCE(tun->numqueues);
398 
399 	txq = skb_get_hash(skb);
400 	if (txq) {
401 		e = tun_flow_find(&tun->flows[tun_hashfn(txq)], txq);
402 		if (e) {
403 			tun_flow_save_rps_rxhash(e, txq);
404 			txq = e->queue_index;
405 		} else
406 			/* use multiply and shift instead of expensive divide */
407 			txq = ((u64)txq * numqueues) >> 32;
408 	} else if (likely(skb_rx_queue_recorded(skb))) {
409 		txq = skb_get_rx_queue(skb);
410 		while (unlikely(txq >= numqueues))
411 			txq -= numqueues;
412 	}
413 
414 	rcu_read_unlock();
415 	return txq;
416 }
417 
418 static inline bool tun_not_capable(struct tun_struct *tun)
419 {
420 	const struct cred *cred = current_cred();
421 	struct net *net = dev_net(tun->dev);
422 
423 	return ((uid_valid(tun->owner) && !uid_eq(cred->euid, tun->owner)) ||
424 		  (gid_valid(tun->group) && !in_egroup_p(tun->group))) &&
425 		!ns_capable(net->user_ns, CAP_NET_ADMIN);
426 }
427 
428 static void tun_set_real_num_queues(struct tun_struct *tun)
429 {
430 	netif_set_real_num_tx_queues(tun->dev, tun->numqueues);
431 	netif_set_real_num_rx_queues(tun->dev, tun->numqueues);
432 }
433 
434 static void tun_disable_queue(struct tun_struct *tun, struct tun_file *tfile)
435 {
436 	tfile->detached = tun;
437 	list_add_tail(&tfile->next, &tun->disabled);
438 	++tun->numdisabled;
439 }
440 
441 static struct tun_struct *tun_enable_queue(struct tun_file *tfile)
442 {
443 	struct tun_struct *tun = tfile->detached;
444 
445 	tfile->detached = NULL;
446 	list_del_init(&tfile->next);
447 	--tun->numdisabled;
448 	return tun;
449 }
450 
451 static void tun_queue_purge(struct tun_file *tfile)
452 {
453 	skb_queue_purge(&tfile->sk.sk_receive_queue);
454 	skb_queue_purge(&tfile->sk.sk_error_queue);
455 }
456 
457 static void __tun_detach(struct tun_file *tfile, bool clean)
458 {
459 	struct tun_file *ntfile;
460 	struct tun_struct *tun;
461 
462 	tun = rtnl_dereference(tfile->tun);
463 
464 	if (tun && !tfile->detached) {
465 		u16 index = tfile->queue_index;
466 		BUG_ON(index >= tun->numqueues);
467 
468 		rcu_assign_pointer(tun->tfiles[index],
469 				   tun->tfiles[tun->numqueues - 1]);
470 		ntfile = rtnl_dereference(tun->tfiles[index]);
471 		ntfile->queue_index = index;
472 
473 		--tun->numqueues;
474 		if (clean) {
475 			RCU_INIT_POINTER(tfile->tun, NULL);
476 			sock_put(&tfile->sk);
477 		} else
478 			tun_disable_queue(tun, tfile);
479 
480 		synchronize_net();
481 		tun_flow_delete_by_queue(tun, tun->numqueues + 1);
482 		/* Drop read queue */
483 		tun_queue_purge(tfile);
484 		tun_set_real_num_queues(tun);
485 	} else if (tfile->detached && clean) {
486 		tun = tun_enable_queue(tfile);
487 		sock_put(&tfile->sk);
488 	}
489 
490 	if (clean) {
491 		if (tun && tun->numqueues == 0 && tun->numdisabled == 0) {
492 			netif_carrier_off(tun->dev);
493 
494 			if (!(tun->flags & IFF_PERSIST) &&
495 			    tun->dev->reg_state == NETREG_REGISTERED)
496 				unregister_netdevice(tun->dev);
497 		}
498 
499 		BUG_ON(!test_bit(SOCK_EXTERNALLY_ALLOCATED,
500 				 &tfile->socket.flags));
501 		sk_release_kernel(&tfile->sk);
502 	}
503 }
504 
505 static void tun_detach(struct tun_file *tfile, bool clean)
506 {
507 	rtnl_lock();
508 	__tun_detach(tfile, clean);
509 	rtnl_unlock();
510 }
511 
512 static void tun_detach_all(struct net_device *dev)
513 {
514 	struct tun_struct *tun = netdev_priv(dev);
515 	struct tun_file *tfile, *tmp;
516 	int i, n = tun->numqueues;
517 
518 	for (i = 0; i < n; i++) {
519 		tfile = rtnl_dereference(tun->tfiles[i]);
520 		BUG_ON(!tfile);
521 		tfile->socket.sk->sk_data_ready(tfile->socket.sk);
522 		RCU_INIT_POINTER(tfile->tun, NULL);
523 		--tun->numqueues;
524 	}
525 	list_for_each_entry(tfile, &tun->disabled, next) {
526 		tfile->socket.sk->sk_data_ready(tfile->socket.sk);
527 		RCU_INIT_POINTER(tfile->tun, NULL);
528 	}
529 	BUG_ON(tun->numqueues != 0);
530 
531 	synchronize_net();
532 	for (i = 0; i < n; i++) {
533 		tfile = rtnl_dereference(tun->tfiles[i]);
534 		/* Drop read queue */
535 		tun_queue_purge(tfile);
536 		sock_put(&tfile->sk);
537 	}
538 	list_for_each_entry_safe(tfile, tmp, &tun->disabled, next) {
539 		tun_enable_queue(tfile);
540 		tun_queue_purge(tfile);
541 		sock_put(&tfile->sk);
542 	}
543 	BUG_ON(tun->numdisabled != 0);
544 
545 	if (tun->flags & IFF_PERSIST)
546 		module_put(THIS_MODULE);
547 }
548 
549 static int tun_attach(struct tun_struct *tun, struct file *file, bool skip_filter)
550 {
551 	struct tun_file *tfile = file->private_data;
552 	int err;
553 
554 	err = security_tun_dev_attach(tfile->socket.sk, tun->security);
555 	if (err < 0)
556 		goto out;
557 
558 	err = -EINVAL;
559 	if (rtnl_dereference(tfile->tun) && !tfile->detached)
560 		goto out;
561 
562 	err = -EBUSY;
563 	if (!(tun->flags & IFF_MULTI_QUEUE) && tun->numqueues == 1)
564 		goto out;
565 
566 	err = -E2BIG;
567 	if (!tfile->detached &&
568 	    tun->numqueues + tun->numdisabled == MAX_TAP_QUEUES)
569 		goto out;
570 
571 	err = 0;
572 
573 	/* Re-attach the filter to persist device */
574 	if (!skip_filter && (tun->filter_attached == true)) {
575 		err = sk_attach_filter(&tun->fprog, tfile->socket.sk);
576 		if (!err)
577 			goto out;
578 	}
579 	tfile->queue_index = tun->numqueues;
580 	rcu_assign_pointer(tfile->tun, tun);
581 	rcu_assign_pointer(tun->tfiles[tun->numqueues], tfile);
582 	tun->numqueues++;
583 
584 	if (tfile->detached)
585 		tun_enable_queue(tfile);
586 	else
587 		sock_hold(&tfile->sk);
588 
589 	tun_set_real_num_queues(tun);
590 
591 	/* device is allowed to go away first, so no need to hold extra
592 	 * refcnt.
593 	 */
594 
595 out:
596 	return err;
597 }
598 
599 static struct tun_struct *__tun_get(struct tun_file *tfile)
600 {
601 	struct tun_struct *tun;
602 
603 	rcu_read_lock();
604 	tun = rcu_dereference(tfile->tun);
605 	if (tun)
606 		dev_hold(tun->dev);
607 	rcu_read_unlock();
608 
609 	return tun;
610 }
611 
612 static struct tun_struct *tun_get(struct file *file)
613 {
614 	return __tun_get(file->private_data);
615 }
616 
617 static void tun_put(struct tun_struct *tun)
618 {
619 	dev_put(tun->dev);
620 }
621 
622 /* TAP filtering */
623 static void addr_hash_set(u32 *mask, const u8 *addr)
624 {
625 	int n = ether_crc(ETH_ALEN, addr) >> 26;
626 	mask[n >> 5] |= (1 << (n & 31));
627 }
628 
629 static unsigned int addr_hash_test(const u32 *mask, const u8 *addr)
630 {
631 	int n = ether_crc(ETH_ALEN, addr) >> 26;
632 	return mask[n >> 5] & (1 << (n & 31));
633 }
634 
635 static int update_filter(struct tap_filter *filter, void __user *arg)
636 {
637 	struct { u8 u[ETH_ALEN]; } *addr;
638 	struct tun_filter uf;
639 	int err, alen, n, nexact;
640 
641 	if (copy_from_user(&uf, arg, sizeof(uf)))
642 		return -EFAULT;
643 
644 	if (!uf.count) {
645 		/* Disabled */
646 		filter->count = 0;
647 		return 0;
648 	}
649 
650 	alen = ETH_ALEN * uf.count;
651 	addr = kmalloc(alen, GFP_KERNEL);
652 	if (!addr)
653 		return -ENOMEM;
654 
655 	if (copy_from_user(addr, arg + sizeof(uf), alen)) {
656 		err = -EFAULT;
657 		goto done;
658 	}
659 
660 	/* The filter is updated without holding any locks. Which is
661 	 * perfectly safe. We disable it first and in the worst
662 	 * case we'll accept a few undesired packets. */
663 	filter->count = 0;
664 	wmb();
665 
666 	/* Use first set of addresses as an exact filter */
667 	for (n = 0; n < uf.count && n < FLT_EXACT_COUNT; n++)
668 		memcpy(filter->addr[n], addr[n].u, ETH_ALEN);
669 
670 	nexact = n;
671 
672 	/* Remaining multicast addresses are hashed,
673 	 * unicast will leave the filter disabled. */
674 	memset(filter->mask, 0, sizeof(filter->mask));
675 	for (; n < uf.count; n++) {
676 		if (!is_multicast_ether_addr(addr[n].u)) {
677 			err = 0; /* no filter */
678 			goto done;
679 		}
680 		addr_hash_set(filter->mask, addr[n].u);
681 	}
682 
683 	/* For ALLMULTI just set the mask to all ones.
684 	 * This overrides the mask populated above. */
685 	if ((uf.flags & TUN_FLT_ALLMULTI))
686 		memset(filter->mask, ~0, sizeof(filter->mask));
687 
688 	/* Now enable the filter */
689 	wmb();
690 	filter->count = nexact;
691 
692 	/* Return the number of exact filters */
693 	err = nexact;
694 
695 done:
696 	kfree(addr);
697 	return err;
698 }
699 
700 /* Returns: 0 - drop, !=0 - accept */
701 static int run_filter(struct tap_filter *filter, const struct sk_buff *skb)
702 {
703 	/* Cannot use eth_hdr(skb) here because skb_mac_hdr() is incorrect
704 	 * at this point. */
705 	struct ethhdr *eh = (struct ethhdr *) skb->data;
706 	int i;
707 
708 	/* Exact match */
709 	for (i = 0; i < filter->count; i++)
710 		if (ether_addr_equal(eh->h_dest, filter->addr[i]))
711 			return 1;
712 
713 	/* Inexact match (multicast only) */
714 	if (is_multicast_ether_addr(eh->h_dest))
715 		return addr_hash_test(filter->mask, eh->h_dest);
716 
717 	return 0;
718 }
719 
720 /*
721  * Checks whether the packet is accepted or not.
722  * Returns: 0 - drop, !=0 - accept
723  */
724 static int check_filter(struct tap_filter *filter, const struct sk_buff *skb)
725 {
726 	if (!filter->count)
727 		return 1;
728 
729 	return run_filter(filter, skb);
730 }
731 
732 /* Network device part of the driver */
733 
734 static const struct ethtool_ops tun_ethtool_ops;
735 
736 /* Net device detach from fd. */
737 static void tun_net_uninit(struct net_device *dev)
738 {
739 	tun_detach_all(dev);
740 }
741 
742 /* Net device open. */
743 static int tun_net_open(struct net_device *dev)
744 {
745 	netif_tx_start_all_queues(dev);
746 	return 0;
747 }
748 
749 /* Net device close. */
750 static int tun_net_close(struct net_device *dev)
751 {
752 	netif_tx_stop_all_queues(dev);
753 	return 0;
754 }
755 
756 /* Net device start xmit */
757 static netdev_tx_t tun_net_xmit(struct sk_buff *skb, struct net_device *dev)
758 {
759 	struct tun_struct *tun = netdev_priv(dev);
760 	int txq = skb->queue_mapping;
761 	struct tun_file *tfile;
762 	u32 numqueues = 0;
763 
764 	rcu_read_lock();
765 	tfile = rcu_dereference(tun->tfiles[txq]);
766 	numqueues = ACCESS_ONCE(tun->numqueues);
767 
768 	/* Drop packet if interface is not attached */
769 	if (txq >= numqueues)
770 		goto drop;
771 
772 	if (numqueues == 1) {
773 		/* Select queue was not called for the skbuff, so we extract the
774 		 * RPS hash and save it into the flow_table here.
775 		 */
776 		__u32 rxhash;
777 
778 		rxhash = skb_get_hash(skb);
779 		if (rxhash) {
780 			struct tun_flow_entry *e;
781 			e = tun_flow_find(&tun->flows[tun_hashfn(rxhash)],
782 					rxhash);
783 			if (e)
784 				tun_flow_save_rps_rxhash(e, rxhash);
785 		}
786 	}
787 
788 	tun_debug(KERN_INFO, tun, "tun_net_xmit %d\n", skb->len);
789 
790 	BUG_ON(!tfile);
791 
792 	/* Drop if the filter does not like it.
793 	 * This is a noop if the filter is disabled.
794 	 * Filter can be enabled only for the TAP devices. */
795 	if (!check_filter(&tun->txflt, skb))
796 		goto drop;
797 
798 	if (tfile->socket.sk->sk_filter &&
799 	    sk_filter(tfile->socket.sk, skb))
800 		goto drop;
801 
802 	/* Limit the number of packets queued by dividing txq length with the
803 	 * number of queues.
804 	 */
805 	if (skb_queue_len(&tfile->socket.sk->sk_receive_queue) * numqueues
806 			  >= dev->tx_queue_len)
807 		goto drop;
808 
809 	if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC)))
810 		goto drop;
811 
812 	if (skb->sk) {
813 		sock_tx_timestamp(skb->sk, &skb_shinfo(skb)->tx_flags);
814 		sw_tx_timestamp(skb);
815 	}
816 
817 	/* Orphan the skb - required as we might hang on to it
818 	 * for indefinite time.
819 	 */
820 	skb_orphan(skb);
821 
822 	nf_reset(skb);
823 
824 	/* Enqueue packet */
825 	skb_queue_tail(&tfile->socket.sk->sk_receive_queue, skb);
826 
827 	/* Notify and wake up reader process */
828 	if (tfile->flags & TUN_FASYNC)
829 		kill_fasync(&tfile->fasync, SIGIO, POLL_IN);
830 	tfile->socket.sk->sk_data_ready(tfile->socket.sk);
831 
832 	rcu_read_unlock();
833 	return NETDEV_TX_OK;
834 
835 drop:
836 	dev->stats.tx_dropped++;
837 	skb_tx_error(skb);
838 	kfree_skb(skb);
839 	rcu_read_unlock();
840 	return NETDEV_TX_OK;
841 }
842 
843 static void tun_net_mclist(struct net_device *dev)
844 {
845 	/*
846 	 * This callback is supposed to deal with mc filter in
847 	 * _rx_ path and has nothing to do with the _tx_ path.
848 	 * In rx path we always accept everything userspace gives us.
849 	 */
850 }
851 
852 #define MIN_MTU 68
853 #define MAX_MTU 65535
854 
855 static int
856 tun_net_change_mtu(struct net_device *dev, int new_mtu)
857 {
858 	if (new_mtu < MIN_MTU || new_mtu + dev->hard_header_len > MAX_MTU)
859 		return -EINVAL;
860 	dev->mtu = new_mtu;
861 	return 0;
862 }
863 
864 static netdev_features_t tun_net_fix_features(struct net_device *dev,
865 	netdev_features_t features)
866 {
867 	struct tun_struct *tun = netdev_priv(dev);
868 
869 	return (features & tun->set_features) | (features & ~TUN_USER_FEATURES);
870 }
871 #ifdef CONFIG_NET_POLL_CONTROLLER
872 static void tun_poll_controller(struct net_device *dev)
873 {
874 	/*
875 	 * Tun only receives frames when:
876 	 * 1) the char device endpoint gets data from user space
877 	 * 2) the tun socket gets a sendmsg call from user space
878 	 * Since both of those are synchronous operations, we are guaranteed
879 	 * never to have pending data when we poll for it
880 	 * so there is nothing to do here but return.
881 	 * We need this though so netpoll recognizes us as an interface that
882 	 * supports polling, which enables bridge devices in virt setups to
883 	 * still use netconsole
884 	 */
885 	return;
886 }
887 #endif
888 static const struct net_device_ops tun_netdev_ops = {
889 	.ndo_uninit		= tun_net_uninit,
890 	.ndo_open		= tun_net_open,
891 	.ndo_stop		= tun_net_close,
892 	.ndo_start_xmit		= tun_net_xmit,
893 	.ndo_change_mtu		= tun_net_change_mtu,
894 	.ndo_fix_features	= tun_net_fix_features,
895 	.ndo_select_queue	= tun_select_queue,
896 #ifdef CONFIG_NET_POLL_CONTROLLER
897 	.ndo_poll_controller	= tun_poll_controller,
898 #endif
899 };
900 
901 static const struct net_device_ops tap_netdev_ops = {
902 	.ndo_uninit		= tun_net_uninit,
903 	.ndo_open		= tun_net_open,
904 	.ndo_stop		= tun_net_close,
905 	.ndo_start_xmit		= tun_net_xmit,
906 	.ndo_change_mtu		= tun_net_change_mtu,
907 	.ndo_fix_features	= tun_net_fix_features,
908 	.ndo_set_rx_mode	= tun_net_mclist,
909 	.ndo_set_mac_address	= eth_mac_addr,
910 	.ndo_validate_addr	= eth_validate_addr,
911 	.ndo_select_queue	= tun_select_queue,
912 #ifdef CONFIG_NET_POLL_CONTROLLER
913 	.ndo_poll_controller	= tun_poll_controller,
914 #endif
915 };
916 
917 static void tun_flow_init(struct tun_struct *tun)
918 {
919 	int i;
920 
921 	for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++)
922 		INIT_HLIST_HEAD(&tun->flows[i]);
923 
924 	tun->ageing_time = TUN_FLOW_EXPIRE;
925 	setup_timer(&tun->flow_gc_timer, tun_flow_cleanup, (unsigned long)tun);
926 	mod_timer(&tun->flow_gc_timer,
927 		  round_jiffies_up(jiffies + tun->ageing_time));
928 }
929 
930 static void tun_flow_uninit(struct tun_struct *tun)
931 {
932 	del_timer_sync(&tun->flow_gc_timer);
933 	tun_flow_flush(tun);
934 }
935 
936 /* Initialize net device. */
937 static void tun_net_init(struct net_device *dev)
938 {
939 	struct tun_struct *tun = netdev_priv(dev);
940 
941 	switch (tun->flags & TUN_TYPE_MASK) {
942 	case IFF_TUN:
943 		dev->netdev_ops = &tun_netdev_ops;
944 
945 		/* Point-to-Point TUN Device */
946 		dev->hard_header_len = 0;
947 		dev->addr_len = 0;
948 		dev->mtu = 1500;
949 
950 		/* Zero header length */
951 		dev->type = ARPHRD_NONE;
952 		dev->flags = IFF_POINTOPOINT | IFF_NOARP | IFF_MULTICAST;
953 		dev->tx_queue_len = TUN_READQ_SIZE;  /* We prefer our own queue length */
954 		break;
955 
956 	case IFF_TAP:
957 		dev->netdev_ops = &tap_netdev_ops;
958 		/* Ethernet TAP Device */
959 		ether_setup(dev);
960 		dev->priv_flags &= ~IFF_TX_SKB_SHARING;
961 		dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
962 
963 		eth_hw_addr_random(dev);
964 
965 		dev->tx_queue_len = TUN_READQ_SIZE;  /* We prefer our own queue length */
966 		break;
967 	}
968 }
969 
970 /* Character device part */
971 
972 /* Poll */
973 static unsigned int tun_chr_poll(struct file *file, poll_table *wait)
974 {
975 	struct tun_file *tfile = file->private_data;
976 	struct tun_struct *tun = __tun_get(tfile);
977 	struct sock *sk;
978 	unsigned int mask = 0;
979 
980 	if (!tun)
981 		return POLLERR;
982 
983 	sk = tfile->socket.sk;
984 
985 	tun_debug(KERN_INFO, tun, "tun_chr_poll\n");
986 
987 	poll_wait(file, sk_sleep(sk), wait);
988 
989 	if (!skb_queue_empty(&sk->sk_receive_queue))
990 		mask |= POLLIN | POLLRDNORM;
991 
992 	if (sock_writeable(sk) ||
993 	    (!test_and_set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags) &&
994 	     sock_writeable(sk)))
995 		mask |= POLLOUT | POLLWRNORM;
996 
997 	if (tun->dev->reg_state != NETREG_REGISTERED)
998 		mask = POLLERR;
999 
1000 	tun_put(tun);
1001 	return mask;
1002 }
1003 
1004 /* prepad is the amount to reserve at front.  len is length after that.
1005  * linear is a hint as to how much to copy (usually headers). */
1006 static struct sk_buff *tun_alloc_skb(struct tun_file *tfile,
1007 				     size_t prepad, size_t len,
1008 				     size_t linear, int noblock)
1009 {
1010 	struct sock *sk = tfile->socket.sk;
1011 	struct sk_buff *skb;
1012 	int err;
1013 
1014 	/* Under a page?  Don't bother with paged skb. */
1015 	if (prepad + len < PAGE_SIZE || !linear)
1016 		linear = len;
1017 
1018 	skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
1019 				   &err, 0);
1020 	if (!skb)
1021 		return ERR_PTR(err);
1022 
1023 	skb_reserve(skb, prepad);
1024 	skb_put(skb, linear);
1025 	skb->data_len = len - linear;
1026 	skb->len += len - linear;
1027 
1028 	return skb;
1029 }
1030 
1031 /* Get packet from user space buffer */
1032 static ssize_t tun_get_user(struct tun_struct *tun, struct tun_file *tfile,
1033 			    void *msg_control, const struct iovec *iv,
1034 			    size_t total_len, size_t count, int noblock)
1035 {
1036 	struct tun_pi pi = { 0, cpu_to_be16(ETH_P_IP) };
1037 	struct sk_buff *skb;
1038 	size_t len = total_len, align = NET_SKB_PAD, linear;
1039 	struct virtio_net_hdr gso = { 0 };
1040 	int good_linear;
1041 	int offset = 0;
1042 	int copylen;
1043 	bool zerocopy = false;
1044 	int err;
1045 	u32 rxhash;
1046 
1047 	if (!(tun->flags & IFF_NO_PI)) {
1048 		if (len < sizeof(pi))
1049 			return -EINVAL;
1050 		len -= sizeof(pi);
1051 
1052 		if (memcpy_fromiovecend((void *)&pi, iv, 0, sizeof(pi)))
1053 			return -EFAULT;
1054 		offset += sizeof(pi);
1055 	}
1056 
1057 	if (tun->flags & IFF_VNET_HDR) {
1058 		if (len < tun->vnet_hdr_sz)
1059 			return -EINVAL;
1060 		len -= tun->vnet_hdr_sz;
1061 
1062 		if (memcpy_fromiovecend((void *)&gso, iv, offset, sizeof(gso)))
1063 			return -EFAULT;
1064 
1065 		if ((gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
1066 		    tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2 > tun16_to_cpu(tun, gso.hdr_len))
1067 			gso.hdr_len = cpu_to_tun16(tun, tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2);
1068 
1069 		if (tun16_to_cpu(tun, gso.hdr_len) > len)
1070 			return -EINVAL;
1071 		offset += tun->vnet_hdr_sz;
1072 	}
1073 
1074 	if ((tun->flags & TUN_TYPE_MASK) == IFF_TAP) {
1075 		align += NET_IP_ALIGN;
1076 		if (unlikely(len < ETH_HLEN ||
1077 			     (gso.hdr_len && tun16_to_cpu(tun, gso.hdr_len) < ETH_HLEN)))
1078 			return -EINVAL;
1079 	}
1080 
1081 	good_linear = SKB_MAX_HEAD(align);
1082 
1083 	if (msg_control) {
1084 		/* There are 256 bytes to be copied in skb, so there is
1085 		 * enough room for skb expand head in case it is used.
1086 		 * The rest of the buffer is mapped from userspace.
1087 		 */
1088 		copylen = gso.hdr_len ? tun16_to_cpu(tun, gso.hdr_len) : GOODCOPY_LEN;
1089 		if (copylen > good_linear)
1090 			copylen = good_linear;
1091 		linear = copylen;
1092 		if (iov_pages(iv, offset + copylen, count) <= MAX_SKB_FRAGS)
1093 			zerocopy = true;
1094 	}
1095 
1096 	if (!zerocopy) {
1097 		copylen = len;
1098 		if (tun16_to_cpu(tun, gso.hdr_len) > good_linear)
1099 			linear = good_linear;
1100 		else
1101 			linear = tun16_to_cpu(tun, gso.hdr_len);
1102 	}
1103 
1104 	skb = tun_alloc_skb(tfile, align, copylen, linear, noblock);
1105 	if (IS_ERR(skb)) {
1106 		if (PTR_ERR(skb) != -EAGAIN)
1107 			tun->dev->stats.rx_dropped++;
1108 		return PTR_ERR(skb);
1109 	}
1110 
1111 	if (zerocopy)
1112 		err = zerocopy_sg_from_iovec(skb, iv, offset, count);
1113 	else {
1114 		err = skb_copy_datagram_from_iovec(skb, 0, iv, offset, len);
1115 		if (!err && msg_control) {
1116 			struct ubuf_info *uarg = msg_control;
1117 			uarg->callback(uarg, false);
1118 		}
1119 	}
1120 
1121 	if (err) {
1122 		tun->dev->stats.rx_dropped++;
1123 		kfree_skb(skb);
1124 		return -EFAULT;
1125 	}
1126 
1127 	if (gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
1128 		if (!skb_partial_csum_set(skb, tun16_to_cpu(tun, gso.csum_start),
1129 					  tun16_to_cpu(tun, gso.csum_offset))) {
1130 			tun->dev->stats.rx_frame_errors++;
1131 			kfree_skb(skb);
1132 			return -EINVAL;
1133 		}
1134 	}
1135 
1136 	switch (tun->flags & TUN_TYPE_MASK) {
1137 	case IFF_TUN:
1138 		if (tun->flags & IFF_NO_PI) {
1139 			switch (skb->data[0] & 0xf0) {
1140 			case 0x40:
1141 				pi.proto = htons(ETH_P_IP);
1142 				break;
1143 			case 0x60:
1144 				pi.proto = htons(ETH_P_IPV6);
1145 				break;
1146 			default:
1147 				tun->dev->stats.rx_dropped++;
1148 				kfree_skb(skb);
1149 				return -EINVAL;
1150 			}
1151 		}
1152 
1153 		skb_reset_mac_header(skb);
1154 		skb->protocol = pi.proto;
1155 		skb->dev = tun->dev;
1156 		break;
1157 	case IFF_TAP:
1158 		skb->protocol = eth_type_trans(skb, tun->dev);
1159 		break;
1160 	}
1161 
1162 	skb_reset_network_header(skb);
1163 
1164 	if (gso.gso_type != VIRTIO_NET_HDR_GSO_NONE) {
1165 		pr_debug("GSO!\n");
1166 		switch (gso.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
1167 		case VIRTIO_NET_HDR_GSO_TCPV4:
1168 			skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
1169 			break;
1170 		case VIRTIO_NET_HDR_GSO_TCPV6:
1171 			skb_shinfo(skb)->gso_type = SKB_GSO_TCPV6;
1172 			break;
1173 		case VIRTIO_NET_HDR_GSO_UDP:
1174 		{
1175 			static bool warned;
1176 
1177 			if (!warned) {
1178 				warned = true;
1179 				netdev_warn(tun->dev,
1180 					    "%s: using disabled UFO feature; please fix this program\n",
1181 					    current->comm);
1182 			}
1183 			skb_shinfo(skb)->gso_type = SKB_GSO_UDP;
1184 			if (skb->protocol == htons(ETH_P_IPV6))
1185 				ipv6_proxy_select_ident(skb);
1186 			break;
1187 		}
1188 		default:
1189 			tun->dev->stats.rx_frame_errors++;
1190 			kfree_skb(skb);
1191 			return -EINVAL;
1192 		}
1193 
1194 		if (gso.gso_type & VIRTIO_NET_HDR_GSO_ECN)
1195 			skb_shinfo(skb)->gso_type |= SKB_GSO_TCP_ECN;
1196 
1197 		skb_shinfo(skb)->gso_size = tun16_to_cpu(tun, gso.gso_size);
1198 		if (skb_shinfo(skb)->gso_size == 0) {
1199 			tun->dev->stats.rx_frame_errors++;
1200 			kfree_skb(skb);
1201 			return -EINVAL;
1202 		}
1203 
1204 		/* Header must be checked, and gso_segs computed. */
1205 		skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
1206 		skb_shinfo(skb)->gso_segs = 0;
1207 	}
1208 
1209 	/* copy skb_ubuf_info for callback when skb has no error */
1210 	if (zerocopy) {
1211 		skb_shinfo(skb)->destructor_arg = msg_control;
1212 		skb_shinfo(skb)->tx_flags |= SKBTX_DEV_ZEROCOPY;
1213 		skb_shinfo(skb)->tx_flags |= SKBTX_SHARED_FRAG;
1214 	}
1215 
1216 	skb_probe_transport_header(skb, 0);
1217 
1218 	rxhash = skb_get_hash(skb);
1219 	netif_rx_ni(skb);
1220 
1221 	tun->dev->stats.rx_packets++;
1222 	tun->dev->stats.rx_bytes += len;
1223 
1224 	tun_flow_update(tun, rxhash, tfile);
1225 	return total_len;
1226 }
1227 
1228 static ssize_t tun_chr_aio_write(struct kiocb *iocb, const struct iovec *iv,
1229 			      unsigned long count, loff_t pos)
1230 {
1231 	struct file *file = iocb->ki_filp;
1232 	struct tun_struct *tun = tun_get(file);
1233 	struct tun_file *tfile = file->private_data;
1234 	ssize_t result;
1235 
1236 	if (!tun)
1237 		return -EBADFD;
1238 
1239 	tun_debug(KERN_INFO, tun, "tun_chr_write %ld\n", count);
1240 
1241 	result = tun_get_user(tun, tfile, NULL, iv, iov_length(iv, count),
1242 			      count, file->f_flags & O_NONBLOCK);
1243 
1244 	tun_put(tun);
1245 	return result;
1246 }
1247 
1248 /* Put packet to the user space buffer */
1249 static ssize_t tun_put_user(struct tun_struct *tun,
1250 			    struct tun_file *tfile,
1251 			    struct sk_buff *skb,
1252 			    const struct iovec *iv, int len)
1253 {
1254 	struct tun_pi pi = { 0, skb->protocol };
1255 	ssize_t total = 0;
1256 	int vlan_offset = 0, copied;
1257 	int vlan_hlen = 0;
1258 	int vnet_hdr_sz = 0;
1259 
1260 	if (vlan_tx_tag_present(skb))
1261 		vlan_hlen = VLAN_HLEN;
1262 
1263 	if (tun->flags & IFF_VNET_HDR)
1264 		vnet_hdr_sz = tun->vnet_hdr_sz;
1265 
1266 	if (!(tun->flags & IFF_NO_PI)) {
1267 		if ((len -= sizeof(pi)) < 0)
1268 			return -EINVAL;
1269 
1270 		if (len < skb->len + vlan_hlen + vnet_hdr_sz) {
1271 			/* Packet will be striped */
1272 			pi.flags |= TUN_PKT_STRIP;
1273 		}
1274 
1275 		if (memcpy_toiovecend(iv, (void *) &pi, 0, sizeof(pi)))
1276 			return -EFAULT;
1277 		total += sizeof(pi);
1278 	}
1279 
1280 	if (vnet_hdr_sz) {
1281 		struct virtio_net_hdr gso = { 0 }; /* no info leak */
1282 		if ((len -= vnet_hdr_sz) < 0)
1283 			return -EINVAL;
1284 
1285 		if (skb_is_gso(skb)) {
1286 			struct skb_shared_info *sinfo = skb_shinfo(skb);
1287 
1288 			/* This is a hint as to how much should be linear. */
1289 			gso.hdr_len = cpu_to_tun16(tun, skb_headlen(skb));
1290 			gso.gso_size = cpu_to_tun16(tun, sinfo->gso_size);
1291 			if (sinfo->gso_type & SKB_GSO_TCPV4)
1292 				gso.gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
1293 			else if (sinfo->gso_type & SKB_GSO_TCPV6)
1294 				gso.gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
1295 			else {
1296 				pr_err("unexpected GSO type: "
1297 				       "0x%x, gso_size %d, hdr_len %d\n",
1298 				       sinfo->gso_type, tun16_to_cpu(tun, gso.gso_size),
1299 				       tun16_to_cpu(tun, gso.hdr_len));
1300 				print_hex_dump(KERN_ERR, "tun: ",
1301 					       DUMP_PREFIX_NONE,
1302 					       16, 1, skb->head,
1303 					       min((int)tun16_to_cpu(tun, 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 = cpu_to_tun16(tun, skb_checksum_start_offset(skb) +
1315 						      vlan_hlen);
1316 			gso.csum_offset = cpu_to_tun16(tun, skb->csum_offset);
1317 		} else if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
1318 			gso.flags = VIRTIO_NET_HDR_F_DATA_VALID;
1319 		} /* else everything is zero */
1320 
1321 		if (unlikely(memcpy_toiovecend(iv, (void *)&gso, total,
1322 					       sizeof(gso))))
1323 			return -EFAULT;
1324 		total += vnet_hdr_sz;
1325 	}
1326 
1327 	copied = total;
1328 	len = min_t(int, skb->len + vlan_hlen, len);
1329 	total += skb->len + vlan_hlen;
1330 	if (vlan_hlen) {
1331 		int copy, ret;
1332 		struct {
1333 			__be16 h_vlan_proto;
1334 			__be16 h_vlan_TCI;
1335 		} veth;
1336 
1337 		veth.h_vlan_proto = skb->vlan_proto;
1338 		veth.h_vlan_TCI = htons(vlan_tx_tag_get(skb));
1339 
1340 		vlan_offset = offsetof(struct vlan_ethhdr, h_vlan_proto);
1341 
1342 		copy = min_t(int, vlan_offset, len);
1343 		ret = skb_copy_datagram_const_iovec(skb, 0, iv, copied, copy);
1344 		len -= copy;
1345 		copied += copy;
1346 		if (ret || !len)
1347 			goto done;
1348 
1349 		copy = min_t(int, sizeof(veth), len);
1350 		ret = memcpy_toiovecend(iv, (void *)&veth, copied, copy);
1351 		len -= copy;
1352 		copied += copy;
1353 		if (ret || !len)
1354 			goto done;
1355 	}
1356 
1357 	skb_copy_datagram_const_iovec(skb, vlan_offset, iv, copied, len);
1358 
1359 done:
1360 	tun->dev->stats.tx_packets++;
1361 	tun->dev->stats.tx_bytes += len;
1362 
1363 	return total;
1364 }
1365 
1366 static ssize_t tun_do_read(struct tun_struct *tun, struct tun_file *tfile,
1367 			   const struct iovec *iv, ssize_t len, int noblock)
1368 {
1369 	struct sk_buff *skb;
1370 	ssize_t ret = 0;
1371 	int peeked, err, off = 0;
1372 
1373 	tun_debug(KERN_INFO, tun, "tun_do_read\n");
1374 
1375 	if (!len)
1376 		return ret;
1377 
1378 	if (tun->dev->reg_state != NETREG_REGISTERED)
1379 		return -EIO;
1380 
1381 	/* Read frames from queue */
1382 	skb = __skb_recv_datagram(tfile->socket.sk, noblock ? MSG_DONTWAIT : 0,
1383 				  &peeked, &off, &err);
1384 	if (skb) {
1385 		ret = tun_put_user(tun, tfile, skb, iv, len);
1386 		kfree_skb(skb);
1387 	} else
1388 		ret = err;
1389 
1390 	return ret;
1391 }
1392 
1393 static ssize_t tun_chr_aio_read(struct kiocb *iocb, const struct iovec *iv,
1394 			    unsigned long count, loff_t pos)
1395 {
1396 	struct file *file = iocb->ki_filp;
1397 	struct tun_file *tfile = file->private_data;
1398 	struct tun_struct *tun = __tun_get(tfile);
1399 	ssize_t len, ret;
1400 
1401 	if (!tun)
1402 		return -EBADFD;
1403 	len = iov_length(iv, count);
1404 	if (len < 0) {
1405 		ret = -EINVAL;
1406 		goto out;
1407 	}
1408 
1409 	ret = tun_do_read(tun, tfile, iv, len,
1410 			  file->f_flags & O_NONBLOCK);
1411 	ret = min_t(ssize_t, ret, len);
1412 	if (ret > 0)
1413 		iocb->ki_pos = ret;
1414 out:
1415 	tun_put(tun);
1416 	return ret;
1417 }
1418 
1419 static void tun_free_netdev(struct net_device *dev)
1420 {
1421 	struct tun_struct *tun = netdev_priv(dev);
1422 
1423 	BUG_ON(!(list_empty(&tun->disabled)));
1424 	tun_flow_uninit(tun);
1425 	security_tun_dev_free_security(tun->security);
1426 	free_netdev(dev);
1427 }
1428 
1429 static void tun_setup(struct net_device *dev)
1430 {
1431 	struct tun_struct *tun = netdev_priv(dev);
1432 
1433 	tun->owner = INVALID_UID;
1434 	tun->group = INVALID_GID;
1435 
1436 	dev->ethtool_ops = &tun_ethtool_ops;
1437 	dev->destructor = tun_free_netdev;
1438 }
1439 
1440 /* Trivial set of netlink ops to allow deleting tun or tap
1441  * device with netlink.
1442  */
1443 static int tun_validate(struct nlattr *tb[], struct nlattr *data[])
1444 {
1445 	return -EINVAL;
1446 }
1447 
1448 static struct rtnl_link_ops tun_link_ops __read_mostly = {
1449 	.kind		= DRV_NAME,
1450 	.priv_size	= sizeof(struct tun_struct),
1451 	.setup		= tun_setup,
1452 	.validate	= tun_validate,
1453 };
1454 
1455 static void tun_sock_write_space(struct sock *sk)
1456 {
1457 	struct tun_file *tfile;
1458 	wait_queue_head_t *wqueue;
1459 
1460 	if (!sock_writeable(sk))
1461 		return;
1462 
1463 	if (!test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags))
1464 		return;
1465 
1466 	wqueue = sk_sleep(sk);
1467 	if (wqueue && waitqueue_active(wqueue))
1468 		wake_up_interruptible_sync_poll(wqueue, POLLOUT |
1469 						POLLWRNORM | POLLWRBAND);
1470 
1471 	tfile = container_of(sk, struct tun_file, sk);
1472 	kill_fasync(&tfile->fasync, SIGIO, POLL_OUT);
1473 }
1474 
1475 static int tun_sendmsg(struct kiocb *iocb, struct socket *sock,
1476 		       struct msghdr *m, size_t total_len)
1477 {
1478 	int ret;
1479 	struct tun_file *tfile = container_of(sock, struct tun_file, socket);
1480 	struct tun_struct *tun = __tun_get(tfile);
1481 
1482 	if (!tun)
1483 		return -EBADFD;
1484 	ret = tun_get_user(tun, tfile, m->msg_control, m->msg_iov, total_len,
1485 			   m->msg_iovlen, m->msg_flags & MSG_DONTWAIT);
1486 	tun_put(tun);
1487 	return ret;
1488 }
1489 
1490 static int tun_recvmsg(struct kiocb *iocb, struct socket *sock,
1491 		       struct msghdr *m, size_t total_len,
1492 		       int flags)
1493 {
1494 	struct tun_file *tfile = container_of(sock, struct tun_file, socket);
1495 	struct tun_struct *tun = __tun_get(tfile);
1496 	int ret;
1497 
1498 	if (!tun)
1499 		return -EBADFD;
1500 
1501 	if (flags & ~(MSG_DONTWAIT|MSG_TRUNC|MSG_ERRQUEUE)) {
1502 		ret = -EINVAL;
1503 		goto out;
1504 	}
1505 	if (flags & MSG_ERRQUEUE) {
1506 		ret = sock_recv_errqueue(sock->sk, m, total_len,
1507 					 SOL_PACKET, TUN_TX_TIMESTAMP);
1508 		goto out;
1509 	}
1510 	ret = tun_do_read(tun, tfile, m->msg_iov, total_len,
1511 			  flags & MSG_DONTWAIT);
1512 	if (ret > total_len) {
1513 		m->msg_flags |= MSG_TRUNC;
1514 		ret = flags & MSG_TRUNC ? ret : total_len;
1515 	}
1516 out:
1517 	tun_put(tun);
1518 	return ret;
1519 }
1520 
1521 static int tun_release(struct socket *sock)
1522 {
1523 	if (sock->sk)
1524 		sock_put(sock->sk);
1525 	return 0;
1526 }
1527 
1528 /* Ops structure to mimic raw sockets with tun */
1529 static const struct proto_ops tun_socket_ops = {
1530 	.sendmsg = tun_sendmsg,
1531 	.recvmsg = tun_recvmsg,
1532 	.release = tun_release,
1533 };
1534 
1535 static struct proto tun_proto = {
1536 	.name		= "tun",
1537 	.owner		= THIS_MODULE,
1538 	.obj_size	= sizeof(struct tun_file),
1539 };
1540 
1541 static int tun_flags(struct tun_struct *tun)
1542 {
1543 	return tun->flags & (TUN_FEATURES | IFF_PERSIST | IFF_TUN | IFF_TAP);
1544 }
1545 
1546 static ssize_t tun_show_flags(struct device *dev, struct device_attribute *attr,
1547 			      char *buf)
1548 {
1549 	struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1550 	return sprintf(buf, "0x%x\n", tun_flags(tun));
1551 }
1552 
1553 static ssize_t tun_show_owner(struct device *dev, struct device_attribute *attr,
1554 			      char *buf)
1555 {
1556 	struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1557 	return uid_valid(tun->owner)?
1558 		sprintf(buf, "%u\n",
1559 			from_kuid_munged(current_user_ns(), tun->owner)):
1560 		sprintf(buf, "-1\n");
1561 }
1562 
1563 static ssize_t tun_show_group(struct device *dev, struct device_attribute *attr,
1564 			      char *buf)
1565 {
1566 	struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1567 	return gid_valid(tun->group) ?
1568 		sprintf(buf, "%u\n",
1569 			from_kgid_munged(current_user_ns(), tun->group)):
1570 		sprintf(buf, "-1\n");
1571 }
1572 
1573 static DEVICE_ATTR(tun_flags, 0444, tun_show_flags, NULL);
1574 static DEVICE_ATTR(owner, 0444, tun_show_owner, NULL);
1575 static DEVICE_ATTR(group, 0444, tun_show_group, NULL);
1576 
1577 static int tun_set_iff(struct net *net, struct file *file, struct ifreq *ifr)
1578 {
1579 	struct tun_struct *tun;
1580 	struct tun_file *tfile = file->private_data;
1581 	struct net_device *dev;
1582 	int err;
1583 
1584 	if (tfile->detached)
1585 		return -EINVAL;
1586 
1587 	dev = __dev_get_by_name(net, ifr->ifr_name);
1588 	if (dev) {
1589 		if (ifr->ifr_flags & IFF_TUN_EXCL)
1590 			return -EBUSY;
1591 		if ((ifr->ifr_flags & IFF_TUN) && dev->netdev_ops == &tun_netdev_ops)
1592 			tun = netdev_priv(dev);
1593 		else if ((ifr->ifr_flags & IFF_TAP) && dev->netdev_ops == &tap_netdev_ops)
1594 			tun = netdev_priv(dev);
1595 		else
1596 			return -EINVAL;
1597 
1598 		if (!!(ifr->ifr_flags & IFF_MULTI_QUEUE) !=
1599 		    !!(tun->flags & IFF_MULTI_QUEUE))
1600 			return -EINVAL;
1601 
1602 		if (tun_not_capable(tun))
1603 			return -EPERM;
1604 		err = security_tun_dev_open(tun->security);
1605 		if (err < 0)
1606 			return err;
1607 
1608 		err = tun_attach(tun, file, ifr->ifr_flags & IFF_NOFILTER);
1609 		if (err < 0)
1610 			return err;
1611 
1612 		if (tun->flags & IFF_MULTI_QUEUE &&
1613 		    (tun->numqueues + tun->numdisabled > 1)) {
1614 			/* One or more queue has already been attached, no need
1615 			 * to initialize the device again.
1616 			 */
1617 			return 0;
1618 		}
1619 	}
1620 	else {
1621 		char *name;
1622 		unsigned long flags = 0;
1623 		int queues = ifr->ifr_flags & IFF_MULTI_QUEUE ?
1624 			     MAX_TAP_QUEUES : 1;
1625 
1626 		if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
1627 			return -EPERM;
1628 		err = security_tun_dev_create();
1629 		if (err < 0)
1630 			return err;
1631 
1632 		/* Set dev type */
1633 		if (ifr->ifr_flags & IFF_TUN) {
1634 			/* TUN device */
1635 			flags |= IFF_TUN;
1636 			name = "tun%d";
1637 		} else if (ifr->ifr_flags & IFF_TAP) {
1638 			/* TAP device */
1639 			flags |= IFF_TAP;
1640 			name = "tap%d";
1641 		} else
1642 			return -EINVAL;
1643 
1644 		if (*ifr->ifr_name)
1645 			name = ifr->ifr_name;
1646 
1647 		dev = alloc_netdev_mqs(sizeof(struct tun_struct), name,
1648 				       NET_NAME_UNKNOWN, tun_setup, queues,
1649 				       queues);
1650 
1651 		if (!dev)
1652 			return -ENOMEM;
1653 
1654 		dev_net_set(dev, net);
1655 		dev->rtnl_link_ops = &tun_link_ops;
1656 		dev->ifindex = tfile->ifindex;
1657 
1658 		tun = netdev_priv(dev);
1659 		tun->dev = dev;
1660 		tun->flags = flags;
1661 		tun->txflt.count = 0;
1662 		tun->vnet_hdr_sz = sizeof(struct virtio_net_hdr);
1663 
1664 		tun->filter_attached = false;
1665 		tun->sndbuf = tfile->socket.sk->sk_sndbuf;
1666 
1667 		spin_lock_init(&tun->lock);
1668 
1669 		err = security_tun_dev_alloc_security(&tun->security);
1670 		if (err < 0)
1671 			goto err_free_dev;
1672 
1673 		tun_net_init(dev);
1674 		tun_flow_init(tun);
1675 
1676 		dev->hw_features = NETIF_F_SG | NETIF_F_FRAGLIST |
1677 				   TUN_USER_FEATURES | NETIF_F_HW_VLAN_CTAG_TX |
1678 				   NETIF_F_HW_VLAN_STAG_TX;
1679 		dev->features = dev->hw_features;
1680 		dev->vlan_features = dev->features &
1681 				     ~(NETIF_F_HW_VLAN_CTAG_TX |
1682 				       NETIF_F_HW_VLAN_STAG_TX);
1683 
1684 		INIT_LIST_HEAD(&tun->disabled);
1685 		err = tun_attach(tun, file, false);
1686 		if (err < 0)
1687 			goto err_free_flow;
1688 
1689 		err = register_netdevice(tun->dev);
1690 		if (err < 0)
1691 			goto err_detach;
1692 
1693 		if (device_create_file(&tun->dev->dev, &dev_attr_tun_flags) ||
1694 		    device_create_file(&tun->dev->dev, &dev_attr_owner) ||
1695 		    device_create_file(&tun->dev->dev, &dev_attr_group))
1696 			pr_err("Failed to create tun sysfs files\n");
1697 	}
1698 
1699 	netif_carrier_on(tun->dev);
1700 
1701 	tun_debug(KERN_INFO, tun, "tun_set_iff\n");
1702 
1703 	tun->flags = (tun->flags & ~TUN_FEATURES) |
1704 		(ifr->ifr_flags & TUN_FEATURES);
1705 
1706 	/* Make sure persistent devices do not get stuck in
1707 	 * xoff state.
1708 	 */
1709 	if (netif_running(tun->dev))
1710 		netif_tx_wake_all_queues(tun->dev);
1711 
1712 	strcpy(ifr->ifr_name, tun->dev->name);
1713 	return 0;
1714 
1715 err_detach:
1716 	tun_detach_all(dev);
1717 err_free_flow:
1718 	tun_flow_uninit(tun);
1719 	security_tun_dev_free_security(tun->security);
1720 err_free_dev:
1721 	free_netdev(dev);
1722 	return err;
1723 }
1724 
1725 static void tun_get_iff(struct net *net, struct tun_struct *tun,
1726 		       struct ifreq *ifr)
1727 {
1728 	tun_debug(KERN_INFO, tun, "tun_get_iff\n");
1729 
1730 	strcpy(ifr->ifr_name, tun->dev->name);
1731 
1732 	ifr->ifr_flags = tun_flags(tun);
1733 
1734 }
1735 
1736 /* This is like a cut-down ethtool ops, except done via tun fd so no
1737  * privs required. */
1738 static int set_offload(struct tun_struct *tun, unsigned long arg)
1739 {
1740 	netdev_features_t features = 0;
1741 
1742 	if (arg & TUN_F_CSUM) {
1743 		features |= NETIF_F_HW_CSUM;
1744 		arg &= ~TUN_F_CSUM;
1745 
1746 		if (arg & (TUN_F_TSO4|TUN_F_TSO6)) {
1747 			if (arg & TUN_F_TSO_ECN) {
1748 				features |= NETIF_F_TSO_ECN;
1749 				arg &= ~TUN_F_TSO_ECN;
1750 			}
1751 			if (arg & TUN_F_TSO4)
1752 				features |= NETIF_F_TSO;
1753 			if (arg & TUN_F_TSO6)
1754 				features |= NETIF_F_TSO6;
1755 			arg &= ~(TUN_F_TSO4|TUN_F_TSO6);
1756 		}
1757 	}
1758 
1759 	/* This gives the user a way to test for new features in future by
1760 	 * trying to set them. */
1761 	if (arg)
1762 		return -EINVAL;
1763 
1764 	tun->set_features = features;
1765 	netdev_update_features(tun->dev);
1766 
1767 	return 0;
1768 }
1769 
1770 static void tun_detach_filter(struct tun_struct *tun, int n)
1771 {
1772 	int i;
1773 	struct tun_file *tfile;
1774 
1775 	for (i = 0; i < n; i++) {
1776 		tfile = rtnl_dereference(tun->tfiles[i]);
1777 		sk_detach_filter(tfile->socket.sk);
1778 	}
1779 
1780 	tun->filter_attached = false;
1781 }
1782 
1783 static int tun_attach_filter(struct tun_struct *tun)
1784 {
1785 	int i, ret = 0;
1786 	struct tun_file *tfile;
1787 
1788 	for (i = 0; i < tun->numqueues; i++) {
1789 		tfile = rtnl_dereference(tun->tfiles[i]);
1790 		ret = sk_attach_filter(&tun->fprog, tfile->socket.sk);
1791 		if (ret) {
1792 			tun_detach_filter(tun, i);
1793 			return ret;
1794 		}
1795 	}
1796 
1797 	tun->filter_attached = true;
1798 	return ret;
1799 }
1800 
1801 static void tun_set_sndbuf(struct tun_struct *tun)
1802 {
1803 	struct tun_file *tfile;
1804 	int i;
1805 
1806 	for (i = 0; i < tun->numqueues; i++) {
1807 		tfile = rtnl_dereference(tun->tfiles[i]);
1808 		tfile->socket.sk->sk_sndbuf = tun->sndbuf;
1809 	}
1810 }
1811 
1812 static int tun_set_queue(struct file *file, struct ifreq *ifr)
1813 {
1814 	struct tun_file *tfile = file->private_data;
1815 	struct tun_struct *tun;
1816 	int ret = 0;
1817 
1818 	rtnl_lock();
1819 
1820 	if (ifr->ifr_flags & IFF_ATTACH_QUEUE) {
1821 		tun = tfile->detached;
1822 		if (!tun) {
1823 			ret = -EINVAL;
1824 			goto unlock;
1825 		}
1826 		ret = security_tun_dev_attach_queue(tun->security);
1827 		if (ret < 0)
1828 			goto unlock;
1829 		ret = tun_attach(tun, file, false);
1830 	} else if (ifr->ifr_flags & IFF_DETACH_QUEUE) {
1831 		tun = rtnl_dereference(tfile->tun);
1832 		if (!tun || !(tun->flags & IFF_MULTI_QUEUE) || tfile->detached)
1833 			ret = -EINVAL;
1834 		else
1835 			__tun_detach(tfile, false);
1836 	} else
1837 		ret = -EINVAL;
1838 
1839 unlock:
1840 	rtnl_unlock();
1841 	return ret;
1842 }
1843 
1844 static long __tun_chr_ioctl(struct file *file, unsigned int cmd,
1845 			    unsigned long arg, int ifreq_len)
1846 {
1847 	struct tun_file *tfile = file->private_data;
1848 	struct tun_struct *tun;
1849 	void __user* argp = (void __user*)arg;
1850 	struct ifreq ifr;
1851 	kuid_t owner;
1852 	kgid_t group;
1853 	int sndbuf;
1854 	int vnet_hdr_sz;
1855 	unsigned int ifindex;
1856 	int ret;
1857 
1858 	if (cmd == TUNSETIFF || cmd == TUNSETQUEUE || _IOC_TYPE(cmd) == 0x89) {
1859 		if (copy_from_user(&ifr, argp, ifreq_len))
1860 			return -EFAULT;
1861 	} else {
1862 		memset(&ifr, 0, sizeof(ifr));
1863 	}
1864 	if (cmd == TUNGETFEATURES) {
1865 		/* Currently this just means: "what IFF flags are valid?".
1866 		 * This is needed because we never checked for invalid flags on
1867 		 * TUNSETIFF.
1868 		 */
1869 		return put_user(IFF_TUN | IFF_TAP | TUN_FEATURES,
1870 				(unsigned int __user*)argp);
1871 	} else if (cmd == TUNSETQUEUE)
1872 		return tun_set_queue(file, &ifr);
1873 
1874 	ret = 0;
1875 	rtnl_lock();
1876 
1877 	tun = __tun_get(tfile);
1878 	if (cmd == TUNSETIFF && !tun) {
1879 		ifr.ifr_name[IFNAMSIZ-1] = '\0';
1880 
1881 		ret = tun_set_iff(tfile->net, file, &ifr);
1882 
1883 		if (ret)
1884 			goto unlock;
1885 
1886 		if (copy_to_user(argp, &ifr, ifreq_len))
1887 			ret = -EFAULT;
1888 		goto unlock;
1889 	}
1890 	if (cmd == TUNSETIFINDEX) {
1891 		ret = -EPERM;
1892 		if (tun)
1893 			goto unlock;
1894 
1895 		ret = -EFAULT;
1896 		if (copy_from_user(&ifindex, argp, sizeof(ifindex)))
1897 			goto unlock;
1898 
1899 		ret = 0;
1900 		tfile->ifindex = ifindex;
1901 		goto unlock;
1902 	}
1903 
1904 	ret = -EBADFD;
1905 	if (!tun)
1906 		goto unlock;
1907 
1908 	tun_debug(KERN_INFO, tun, "tun_chr_ioctl cmd %u\n", cmd);
1909 
1910 	ret = 0;
1911 	switch (cmd) {
1912 	case TUNGETIFF:
1913 		tun_get_iff(current->nsproxy->net_ns, tun, &ifr);
1914 
1915 		if (tfile->detached)
1916 			ifr.ifr_flags |= IFF_DETACH_QUEUE;
1917 		if (!tfile->socket.sk->sk_filter)
1918 			ifr.ifr_flags |= IFF_NOFILTER;
1919 
1920 		if (copy_to_user(argp, &ifr, ifreq_len))
1921 			ret = -EFAULT;
1922 		break;
1923 
1924 	case TUNSETNOCSUM:
1925 		/* Disable/Enable checksum */
1926 
1927 		/* [unimplemented] */
1928 		tun_debug(KERN_INFO, tun, "ignored: set checksum %s\n",
1929 			  arg ? "disabled" : "enabled");
1930 		break;
1931 
1932 	case TUNSETPERSIST:
1933 		/* Disable/Enable persist mode. Keep an extra reference to the
1934 		 * module to prevent the module being unprobed.
1935 		 */
1936 		if (arg && !(tun->flags & IFF_PERSIST)) {
1937 			tun->flags |= IFF_PERSIST;
1938 			__module_get(THIS_MODULE);
1939 		}
1940 		if (!arg && (tun->flags & IFF_PERSIST)) {
1941 			tun->flags &= ~IFF_PERSIST;
1942 			module_put(THIS_MODULE);
1943 		}
1944 
1945 		tun_debug(KERN_INFO, tun, "persist %s\n",
1946 			  arg ? "enabled" : "disabled");
1947 		break;
1948 
1949 	case TUNSETOWNER:
1950 		/* Set owner of the device */
1951 		owner = make_kuid(current_user_ns(), arg);
1952 		if (!uid_valid(owner)) {
1953 			ret = -EINVAL;
1954 			break;
1955 		}
1956 		tun->owner = owner;
1957 		tun_debug(KERN_INFO, tun, "owner set to %u\n",
1958 			  from_kuid(&init_user_ns, tun->owner));
1959 		break;
1960 
1961 	case TUNSETGROUP:
1962 		/* Set group of the device */
1963 		group = make_kgid(current_user_ns(), arg);
1964 		if (!gid_valid(group)) {
1965 			ret = -EINVAL;
1966 			break;
1967 		}
1968 		tun->group = group;
1969 		tun_debug(KERN_INFO, tun, "group set to %u\n",
1970 			  from_kgid(&init_user_ns, tun->group));
1971 		break;
1972 
1973 	case TUNSETLINK:
1974 		/* Only allow setting the type when the interface is down */
1975 		if (tun->dev->flags & IFF_UP) {
1976 			tun_debug(KERN_INFO, tun,
1977 				  "Linktype set failed because interface is up\n");
1978 			ret = -EBUSY;
1979 		} else {
1980 			tun->dev->type = (int) arg;
1981 			tun_debug(KERN_INFO, tun, "linktype set to %d\n",
1982 				  tun->dev->type);
1983 			ret = 0;
1984 		}
1985 		break;
1986 
1987 #ifdef TUN_DEBUG
1988 	case TUNSETDEBUG:
1989 		tun->debug = arg;
1990 		break;
1991 #endif
1992 	case TUNSETOFFLOAD:
1993 		ret = set_offload(tun, arg);
1994 		break;
1995 
1996 	case TUNSETTXFILTER:
1997 		/* Can be set only for TAPs */
1998 		ret = -EINVAL;
1999 		if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
2000 			break;
2001 		ret = update_filter(&tun->txflt, (void __user *)arg);
2002 		break;
2003 
2004 	case SIOCGIFHWADDR:
2005 		/* Get hw address */
2006 		memcpy(ifr.ifr_hwaddr.sa_data, tun->dev->dev_addr, ETH_ALEN);
2007 		ifr.ifr_hwaddr.sa_family = tun->dev->type;
2008 		if (copy_to_user(argp, &ifr, ifreq_len))
2009 			ret = -EFAULT;
2010 		break;
2011 
2012 	case SIOCSIFHWADDR:
2013 		/* Set hw address */
2014 		tun_debug(KERN_DEBUG, tun, "set hw address: %pM\n",
2015 			  ifr.ifr_hwaddr.sa_data);
2016 
2017 		ret = dev_set_mac_address(tun->dev, &ifr.ifr_hwaddr);
2018 		break;
2019 
2020 	case TUNGETSNDBUF:
2021 		sndbuf = tfile->socket.sk->sk_sndbuf;
2022 		if (copy_to_user(argp, &sndbuf, sizeof(sndbuf)))
2023 			ret = -EFAULT;
2024 		break;
2025 
2026 	case TUNSETSNDBUF:
2027 		if (copy_from_user(&sndbuf, argp, sizeof(sndbuf))) {
2028 			ret = -EFAULT;
2029 			break;
2030 		}
2031 
2032 		tun->sndbuf = sndbuf;
2033 		tun_set_sndbuf(tun);
2034 		break;
2035 
2036 	case TUNGETVNETHDRSZ:
2037 		vnet_hdr_sz = tun->vnet_hdr_sz;
2038 		if (copy_to_user(argp, &vnet_hdr_sz, sizeof(vnet_hdr_sz)))
2039 			ret = -EFAULT;
2040 		break;
2041 
2042 	case TUNSETVNETHDRSZ:
2043 		if (copy_from_user(&vnet_hdr_sz, argp, sizeof(vnet_hdr_sz))) {
2044 			ret = -EFAULT;
2045 			break;
2046 		}
2047 		if (vnet_hdr_sz < (int)sizeof(struct virtio_net_hdr)) {
2048 			ret = -EINVAL;
2049 			break;
2050 		}
2051 
2052 		tun->vnet_hdr_sz = vnet_hdr_sz;
2053 		break;
2054 
2055 	case TUNATTACHFILTER:
2056 		/* Can be set only for TAPs */
2057 		ret = -EINVAL;
2058 		if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
2059 			break;
2060 		ret = -EFAULT;
2061 		if (copy_from_user(&tun->fprog, argp, sizeof(tun->fprog)))
2062 			break;
2063 
2064 		ret = tun_attach_filter(tun);
2065 		break;
2066 
2067 	case TUNDETACHFILTER:
2068 		/* Can be set only for TAPs */
2069 		ret = -EINVAL;
2070 		if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
2071 			break;
2072 		ret = 0;
2073 		tun_detach_filter(tun, tun->numqueues);
2074 		break;
2075 
2076 	case TUNGETFILTER:
2077 		ret = -EINVAL;
2078 		if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
2079 			break;
2080 		ret = -EFAULT;
2081 		if (copy_to_user(argp, &tun->fprog, sizeof(tun->fprog)))
2082 			break;
2083 		ret = 0;
2084 		break;
2085 
2086 	default:
2087 		ret = -EINVAL;
2088 		break;
2089 	}
2090 
2091 unlock:
2092 	rtnl_unlock();
2093 	if (tun)
2094 		tun_put(tun);
2095 	return ret;
2096 }
2097 
2098 static long tun_chr_ioctl(struct file *file,
2099 			  unsigned int cmd, unsigned long arg)
2100 {
2101 	return __tun_chr_ioctl(file, cmd, arg, sizeof (struct ifreq));
2102 }
2103 
2104 #ifdef CONFIG_COMPAT
2105 static long tun_chr_compat_ioctl(struct file *file,
2106 			 unsigned int cmd, unsigned long arg)
2107 {
2108 	switch (cmd) {
2109 	case TUNSETIFF:
2110 	case TUNGETIFF:
2111 	case TUNSETTXFILTER:
2112 	case TUNGETSNDBUF:
2113 	case TUNSETSNDBUF:
2114 	case SIOCGIFHWADDR:
2115 	case SIOCSIFHWADDR:
2116 		arg = (unsigned long)compat_ptr(arg);
2117 		break;
2118 	default:
2119 		arg = (compat_ulong_t)arg;
2120 		break;
2121 	}
2122 
2123 	/*
2124 	 * compat_ifreq is shorter than ifreq, so we must not access beyond
2125 	 * the end of that structure. All fields that are used in this
2126 	 * driver are compatible though, we don't need to convert the
2127 	 * contents.
2128 	 */
2129 	return __tun_chr_ioctl(file, cmd, arg, sizeof(struct compat_ifreq));
2130 }
2131 #endif /* CONFIG_COMPAT */
2132 
2133 static int tun_chr_fasync(int fd, struct file *file, int on)
2134 {
2135 	struct tun_file *tfile = file->private_data;
2136 	int ret;
2137 
2138 	if ((ret = fasync_helper(fd, file, on, &tfile->fasync)) < 0)
2139 		goto out;
2140 
2141 	if (on) {
2142 		__f_setown(file, task_pid(current), PIDTYPE_PID, 0);
2143 		tfile->flags |= TUN_FASYNC;
2144 	} else
2145 		tfile->flags &= ~TUN_FASYNC;
2146 	ret = 0;
2147 out:
2148 	return ret;
2149 }
2150 
2151 static int tun_chr_open(struct inode *inode, struct file * file)
2152 {
2153 	struct tun_file *tfile;
2154 
2155 	DBG1(KERN_INFO, "tunX: tun_chr_open\n");
2156 
2157 	tfile = (struct tun_file *)sk_alloc(&init_net, AF_UNSPEC, GFP_KERNEL,
2158 					    &tun_proto);
2159 	if (!tfile)
2160 		return -ENOMEM;
2161 	RCU_INIT_POINTER(tfile->tun, NULL);
2162 	tfile->net = get_net(current->nsproxy->net_ns);
2163 	tfile->flags = 0;
2164 	tfile->ifindex = 0;
2165 
2166 	init_waitqueue_head(&tfile->wq.wait);
2167 	RCU_INIT_POINTER(tfile->socket.wq, &tfile->wq);
2168 
2169 	tfile->socket.file = file;
2170 	tfile->socket.ops = &tun_socket_ops;
2171 
2172 	sock_init_data(&tfile->socket, &tfile->sk);
2173 	sk_change_net(&tfile->sk, tfile->net);
2174 
2175 	tfile->sk.sk_write_space = tun_sock_write_space;
2176 	tfile->sk.sk_sndbuf = INT_MAX;
2177 
2178 	file->private_data = tfile;
2179 	set_bit(SOCK_EXTERNALLY_ALLOCATED, &tfile->socket.flags);
2180 	INIT_LIST_HEAD(&tfile->next);
2181 
2182 	sock_set_flag(&tfile->sk, SOCK_ZEROCOPY);
2183 
2184 	return 0;
2185 }
2186 
2187 static int tun_chr_close(struct inode *inode, struct file *file)
2188 {
2189 	struct tun_file *tfile = file->private_data;
2190 	struct net *net = tfile->net;
2191 
2192 	tun_detach(tfile, true);
2193 	put_net(net);
2194 
2195 	return 0;
2196 }
2197 
2198 #ifdef CONFIG_PROC_FS
2199 static int tun_chr_show_fdinfo(struct seq_file *m, struct file *f)
2200 {
2201 	struct tun_struct *tun;
2202 	struct ifreq ifr;
2203 
2204 	memset(&ifr, 0, sizeof(ifr));
2205 
2206 	rtnl_lock();
2207 	tun = tun_get(f);
2208 	if (tun)
2209 		tun_get_iff(current->nsproxy->net_ns, tun, &ifr);
2210 	rtnl_unlock();
2211 
2212 	if (tun)
2213 		tun_put(tun);
2214 
2215 	return seq_printf(m, "iff:\t%s\n", ifr.ifr_name);
2216 }
2217 #endif
2218 
2219 static const struct file_operations tun_fops = {
2220 	.owner	= THIS_MODULE,
2221 	.llseek = no_llseek,
2222 	.read  = do_sync_read,
2223 	.aio_read  = tun_chr_aio_read,
2224 	.write = do_sync_write,
2225 	.aio_write = tun_chr_aio_write,
2226 	.poll	= tun_chr_poll,
2227 	.unlocked_ioctl	= tun_chr_ioctl,
2228 #ifdef CONFIG_COMPAT
2229 	.compat_ioctl = tun_chr_compat_ioctl,
2230 #endif
2231 	.open	= tun_chr_open,
2232 	.release = tun_chr_close,
2233 	.fasync = tun_chr_fasync,
2234 #ifdef CONFIG_PROC_FS
2235 	.show_fdinfo = tun_chr_show_fdinfo,
2236 #endif
2237 };
2238 
2239 static struct miscdevice tun_miscdev = {
2240 	.minor = TUN_MINOR,
2241 	.name = "tun",
2242 	.nodename = "net/tun",
2243 	.fops = &tun_fops,
2244 };
2245 
2246 /* ethtool interface */
2247 
2248 static int tun_get_settings(struct net_device *dev, struct ethtool_cmd *cmd)
2249 {
2250 	cmd->supported		= 0;
2251 	cmd->advertising	= 0;
2252 	ethtool_cmd_speed_set(cmd, SPEED_10);
2253 	cmd->duplex		= DUPLEX_FULL;
2254 	cmd->port		= PORT_TP;
2255 	cmd->phy_address	= 0;
2256 	cmd->transceiver	= XCVR_INTERNAL;
2257 	cmd->autoneg		= AUTONEG_DISABLE;
2258 	cmd->maxtxpkt		= 0;
2259 	cmd->maxrxpkt		= 0;
2260 	return 0;
2261 }
2262 
2263 static void tun_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
2264 {
2265 	struct tun_struct *tun = netdev_priv(dev);
2266 
2267 	strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
2268 	strlcpy(info->version, DRV_VERSION, sizeof(info->version));
2269 
2270 	switch (tun->flags & TUN_TYPE_MASK) {
2271 	case IFF_TUN:
2272 		strlcpy(info->bus_info, "tun", sizeof(info->bus_info));
2273 		break;
2274 	case IFF_TAP:
2275 		strlcpy(info->bus_info, "tap", sizeof(info->bus_info));
2276 		break;
2277 	}
2278 }
2279 
2280 static u32 tun_get_msglevel(struct net_device *dev)
2281 {
2282 #ifdef TUN_DEBUG
2283 	struct tun_struct *tun = netdev_priv(dev);
2284 	return tun->debug;
2285 #else
2286 	return -EOPNOTSUPP;
2287 #endif
2288 }
2289 
2290 static void tun_set_msglevel(struct net_device *dev, u32 value)
2291 {
2292 #ifdef TUN_DEBUG
2293 	struct tun_struct *tun = netdev_priv(dev);
2294 	tun->debug = value;
2295 #endif
2296 }
2297 
2298 static const struct ethtool_ops tun_ethtool_ops = {
2299 	.get_settings	= tun_get_settings,
2300 	.get_drvinfo	= tun_get_drvinfo,
2301 	.get_msglevel	= tun_get_msglevel,
2302 	.set_msglevel	= tun_set_msglevel,
2303 	.get_link	= ethtool_op_get_link,
2304 	.get_ts_info	= ethtool_op_get_ts_info,
2305 };
2306 
2307 
2308 static int __init tun_init(void)
2309 {
2310 	int ret = 0;
2311 
2312 	pr_info("%s, %s\n", DRV_DESCRIPTION, DRV_VERSION);
2313 	pr_info("%s\n", DRV_COPYRIGHT);
2314 
2315 	ret = rtnl_link_register(&tun_link_ops);
2316 	if (ret) {
2317 		pr_err("Can't register link_ops\n");
2318 		goto err_linkops;
2319 	}
2320 
2321 	ret = misc_register(&tun_miscdev);
2322 	if (ret) {
2323 		pr_err("Can't register misc device %d\n", TUN_MINOR);
2324 		goto err_misc;
2325 	}
2326 	return  0;
2327 err_misc:
2328 	rtnl_link_unregister(&tun_link_ops);
2329 err_linkops:
2330 	return ret;
2331 }
2332 
2333 static void tun_cleanup(void)
2334 {
2335 	misc_deregister(&tun_miscdev);
2336 	rtnl_link_unregister(&tun_link_ops);
2337 }
2338 
2339 /* Get an underlying socket object from tun file.  Returns error unless file is
2340  * attached to a device.  The returned object works like a packet socket, it
2341  * can be used for sock_sendmsg/sock_recvmsg.  The caller is responsible for
2342  * holding a reference to the file for as long as the socket is in use. */
2343 struct socket *tun_get_socket(struct file *file)
2344 {
2345 	struct tun_file *tfile;
2346 	if (file->f_op != &tun_fops)
2347 		return ERR_PTR(-EINVAL);
2348 	tfile = file->private_data;
2349 	if (!tfile)
2350 		return ERR_PTR(-EBADFD);
2351 	return &tfile->socket;
2352 }
2353 EXPORT_SYMBOL_GPL(tun_get_socket);
2354 
2355 module_init(tun_init);
2356 module_exit(tun_cleanup);
2357 MODULE_DESCRIPTION(DRV_DESCRIPTION);
2358 MODULE_AUTHOR(DRV_COPYRIGHT);
2359 MODULE_LICENSE("GPL");
2360 MODULE_ALIAS_MISCDEV(TUN_MINOR);
2361 MODULE_ALIAS("devname:net/tun");
2362