xref: /openbmc/linux/drivers/net/tun.c (revision 45471cd98decae5fced8b38e46c223f54a924814)
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/net_namespace.h>
69 #include <net/netns/generic.h>
70 #include <net/rtnetlink.h>
71 #include <net/sock.h>
72 #include <linux/seq_file.h>
73 #include <linux/uio.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 /* High bits in flags field are unused. */
113 #define TUN_VNET_LE     0x80000000
114 
115 #define TUN_FEATURES (IFF_NO_PI | IFF_ONE_QUEUE | IFF_VNET_HDR | \
116 		      IFF_MULTI_QUEUE)
117 #define GOODCOPY_LEN 128
118 
119 #define FLT_EXACT_COUNT 8
120 struct tap_filter {
121 	unsigned int    count;    /* Number of addrs. Zero means disabled */
122 	u32             mask[2];  /* Mask of the hashed addrs */
123 	unsigned char	addr[FLT_EXACT_COUNT][ETH_ALEN];
124 };
125 
126 /* MAX_TAP_QUEUES 256 is chosen to allow rx/tx queues to be equal
127  * to max number of VCPUs in guest. */
128 #define MAX_TAP_QUEUES 256
129 #define MAX_TAP_FLOWS  4096
130 
131 #define TUN_FLOW_EXPIRE (3 * HZ)
132 
133 /* A tun_file connects an open character device to a tuntap netdevice. It
134  * also contains all socket related structures (except sock_fprog and tap_filter)
135  * to serve as one transmit queue for tuntap device. The sock_fprog and
136  * tap_filter were kept in tun_struct since they were used for filtering for the
137  * netdevice not for a specific queue (at least I didn't see the requirement for
138  * this).
139  *
140  * RCU usage:
141  * The tun_file and tun_struct are loosely coupled, the pointer from one to the
142  * other can only be read while rcu_read_lock or rtnl_lock is held.
143  */
144 struct tun_file {
145 	struct sock sk;
146 	struct socket socket;
147 	struct socket_wq wq;
148 	struct tun_struct __rcu *tun;
149 	struct 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|NETIF_F_UFO)
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 & TUN_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 & TUN_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 	hlist_del_rcu(&e->hash_link);
259 	kfree_rcu(e, rcu);
260 	--tun->flow_count;
261 }
262 
263 static void tun_flow_flush(struct tun_struct *tun)
264 {
265 	int i;
266 
267 	spin_lock_bh(&tun->lock);
268 	for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
269 		struct tun_flow_entry *e;
270 		struct hlist_node *n;
271 
272 		hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link)
273 			tun_flow_delete(tun, e);
274 	}
275 	spin_unlock_bh(&tun->lock);
276 }
277 
278 static void tun_flow_delete_by_queue(struct tun_struct *tun, u16 queue_index)
279 {
280 	int i;
281 
282 	spin_lock_bh(&tun->lock);
283 	for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
284 		struct tun_flow_entry *e;
285 		struct hlist_node *n;
286 
287 		hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
288 			if (e->queue_index == queue_index)
289 				tun_flow_delete(tun, e);
290 		}
291 	}
292 	spin_unlock_bh(&tun->lock);
293 }
294 
295 static void tun_flow_cleanup(unsigned long data)
296 {
297 	struct tun_struct *tun = (struct tun_struct *)data;
298 	unsigned long delay = tun->ageing_time;
299 	unsigned long next_timer = jiffies + delay;
300 	unsigned long count = 0;
301 	int i;
302 
303 	tun_debug(KERN_INFO, tun, "tun_flow_cleanup\n");
304 
305 	spin_lock_bh(&tun->lock);
306 	for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
307 		struct tun_flow_entry *e;
308 		struct hlist_node *n;
309 
310 		hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
311 			unsigned long this_timer;
312 			count++;
313 			this_timer = e->updated + delay;
314 			if (time_before_eq(this_timer, jiffies))
315 				tun_flow_delete(tun, e);
316 			else if (time_before(this_timer, next_timer))
317 				next_timer = this_timer;
318 		}
319 	}
320 
321 	if (count)
322 		mod_timer(&tun->flow_gc_timer, round_jiffies_up(next_timer));
323 	spin_unlock_bh(&tun->lock);
324 }
325 
326 static void tun_flow_update(struct tun_struct *tun, u32 rxhash,
327 			    struct tun_file *tfile)
328 {
329 	struct hlist_head *head;
330 	struct tun_flow_entry *e;
331 	unsigned long delay = tun->ageing_time;
332 	u16 queue_index = tfile->queue_index;
333 
334 	if (!rxhash)
335 		return;
336 	else
337 		head = &tun->flows[tun_hashfn(rxhash)];
338 
339 	rcu_read_lock();
340 
341 	/* We may get a very small possibility of OOO during switching, not
342 	 * worth to optimize.*/
343 	if (tun->numqueues == 1 || tfile->detached)
344 		goto unlock;
345 
346 	e = tun_flow_find(head, rxhash);
347 	if (likely(e)) {
348 		/* TODO: keep queueing to old queue until it's empty? */
349 		e->queue_index = queue_index;
350 		e->updated = jiffies;
351 		sock_rps_record_flow_hash(e->rps_rxhash);
352 	} else {
353 		spin_lock_bh(&tun->lock);
354 		if (!tun_flow_find(head, rxhash) &&
355 		    tun->flow_count < MAX_TAP_FLOWS)
356 			tun_flow_create(tun, head, rxhash, queue_index);
357 
358 		if (!timer_pending(&tun->flow_gc_timer))
359 			mod_timer(&tun->flow_gc_timer,
360 				  round_jiffies_up(jiffies + delay));
361 		spin_unlock_bh(&tun->lock);
362 	}
363 
364 unlock:
365 	rcu_read_unlock();
366 }
367 
368 /**
369  * Save the hash received in the stack receive path and update the
370  * flow_hash table accordingly.
371  */
372 static inline void tun_flow_save_rps_rxhash(struct tun_flow_entry *e, u32 hash)
373 {
374 	if (unlikely(e->rps_rxhash != hash))
375 		e->rps_rxhash = hash;
376 }
377 
378 /* We try to identify a flow through its rxhash first. The reason that
379  * we do not check rxq no. is because some cards(e.g 82599), chooses
380  * the rxq based on the txq where the last packet of the flow comes. As
381  * the userspace application move between processors, we may get a
382  * different rxq no. here. If we could not get rxhash, then we would
383  * hope the rxq no. may help here.
384  */
385 static u16 tun_select_queue(struct net_device *dev, struct sk_buff *skb,
386 			    void *accel_priv, select_queue_fallback_t fallback)
387 {
388 	struct tun_struct *tun = netdev_priv(dev);
389 	struct tun_flow_entry *e;
390 	u32 txq = 0;
391 	u32 numqueues = 0;
392 
393 	rcu_read_lock();
394 	numqueues = ACCESS_ONCE(tun->numqueues);
395 
396 	txq = skb_get_hash(skb);
397 	if (txq) {
398 		e = tun_flow_find(&tun->flows[tun_hashfn(txq)], txq);
399 		if (e) {
400 			tun_flow_save_rps_rxhash(e, txq);
401 			txq = e->queue_index;
402 		} else
403 			/* use multiply and shift instead of expensive divide */
404 			txq = ((u64)txq * numqueues) >> 32;
405 	} else if (likely(skb_rx_queue_recorded(skb))) {
406 		txq = skb_get_rx_queue(skb);
407 		while (unlikely(txq >= numqueues))
408 			txq -= numqueues;
409 	}
410 
411 	rcu_read_unlock();
412 	return txq;
413 }
414 
415 static inline bool tun_not_capable(struct tun_struct *tun)
416 {
417 	const struct cred *cred = current_cred();
418 	struct net *net = dev_net(tun->dev);
419 
420 	return ((uid_valid(tun->owner) && !uid_eq(cred->euid, tun->owner)) ||
421 		  (gid_valid(tun->group) && !in_egroup_p(tun->group))) &&
422 		!ns_capable(net->user_ns, CAP_NET_ADMIN);
423 }
424 
425 static void tun_set_real_num_queues(struct tun_struct *tun)
426 {
427 	netif_set_real_num_tx_queues(tun->dev, tun->numqueues);
428 	netif_set_real_num_rx_queues(tun->dev, tun->numqueues);
429 }
430 
431 static void tun_disable_queue(struct tun_struct *tun, struct tun_file *tfile)
432 {
433 	tfile->detached = tun;
434 	list_add_tail(&tfile->next, &tun->disabled);
435 	++tun->numdisabled;
436 }
437 
438 static struct tun_struct *tun_enable_queue(struct tun_file *tfile)
439 {
440 	struct tun_struct *tun = tfile->detached;
441 
442 	tfile->detached = NULL;
443 	list_del_init(&tfile->next);
444 	--tun->numdisabled;
445 	return tun;
446 }
447 
448 static void tun_queue_purge(struct tun_file *tfile)
449 {
450 	skb_queue_purge(&tfile->sk.sk_receive_queue);
451 	skb_queue_purge(&tfile->sk.sk_error_queue);
452 }
453 
454 static void __tun_detach(struct tun_file *tfile, bool clean)
455 {
456 	struct tun_file *ntfile;
457 	struct tun_struct *tun;
458 
459 	tun = rtnl_dereference(tfile->tun);
460 
461 	if (tun && !tfile->detached) {
462 		u16 index = tfile->queue_index;
463 		BUG_ON(index >= tun->numqueues);
464 
465 		rcu_assign_pointer(tun->tfiles[index],
466 				   tun->tfiles[tun->numqueues - 1]);
467 		ntfile = rtnl_dereference(tun->tfiles[index]);
468 		ntfile->queue_index = index;
469 
470 		--tun->numqueues;
471 		if (clean) {
472 			RCU_INIT_POINTER(tfile->tun, NULL);
473 			sock_put(&tfile->sk);
474 		} else
475 			tun_disable_queue(tun, tfile);
476 
477 		synchronize_net();
478 		tun_flow_delete_by_queue(tun, tun->numqueues + 1);
479 		/* Drop read queue */
480 		tun_queue_purge(tfile);
481 		tun_set_real_num_queues(tun);
482 	} else if (tfile->detached && clean) {
483 		tun = tun_enable_queue(tfile);
484 		sock_put(&tfile->sk);
485 	}
486 
487 	if (clean) {
488 		if (tun && tun->numqueues == 0 && tun->numdisabled == 0) {
489 			netif_carrier_off(tun->dev);
490 
491 			if (!(tun->flags & IFF_PERSIST) &&
492 			    tun->dev->reg_state == NETREG_REGISTERED)
493 				unregister_netdevice(tun->dev);
494 		}
495 		sock_put(&tfile->sk);
496 	}
497 }
498 
499 static void tun_detach(struct tun_file *tfile, bool clean)
500 {
501 	rtnl_lock();
502 	__tun_detach(tfile, clean);
503 	rtnl_unlock();
504 }
505 
506 static void tun_detach_all(struct net_device *dev)
507 {
508 	struct tun_struct *tun = netdev_priv(dev);
509 	struct tun_file *tfile, *tmp;
510 	int i, n = tun->numqueues;
511 
512 	for (i = 0; i < n; i++) {
513 		tfile = rtnl_dereference(tun->tfiles[i]);
514 		BUG_ON(!tfile);
515 		tfile->socket.sk->sk_data_ready(tfile->socket.sk);
516 		RCU_INIT_POINTER(tfile->tun, NULL);
517 		--tun->numqueues;
518 	}
519 	list_for_each_entry(tfile, &tun->disabled, next) {
520 		tfile->socket.sk->sk_data_ready(tfile->socket.sk);
521 		RCU_INIT_POINTER(tfile->tun, NULL);
522 	}
523 	BUG_ON(tun->numqueues != 0);
524 
525 	synchronize_net();
526 	for (i = 0; i < n; i++) {
527 		tfile = rtnl_dereference(tun->tfiles[i]);
528 		/* Drop read queue */
529 		tun_queue_purge(tfile);
530 		sock_put(&tfile->sk);
531 	}
532 	list_for_each_entry_safe(tfile, tmp, &tun->disabled, next) {
533 		tun_enable_queue(tfile);
534 		tun_queue_purge(tfile);
535 		sock_put(&tfile->sk);
536 	}
537 	BUG_ON(tun->numdisabled != 0);
538 
539 	if (tun->flags & IFF_PERSIST)
540 		module_put(THIS_MODULE);
541 }
542 
543 static int tun_attach(struct tun_struct *tun, struct file *file, bool skip_filter)
544 {
545 	struct tun_file *tfile = file->private_data;
546 	int err;
547 
548 	err = security_tun_dev_attach(tfile->socket.sk, tun->security);
549 	if (err < 0)
550 		goto out;
551 
552 	err = -EINVAL;
553 	if (rtnl_dereference(tfile->tun) && !tfile->detached)
554 		goto out;
555 
556 	err = -EBUSY;
557 	if (!(tun->flags & IFF_MULTI_QUEUE) && tun->numqueues == 1)
558 		goto out;
559 
560 	err = -E2BIG;
561 	if (!tfile->detached &&
562 	    tun->numqueues + tun->numdisabled == MAX_TAP_QUEUES)
563 		goto out;
564 
565 	err = 0;
566 
567 	/* Re-attach the filter to persist device */
568 	if (!skip_filter && (tun->filter_attached == true)) {
569 		err = sk_attach_filter(&tun->fprog, tfile->socket.sk);
570 		if (!err)
571 			goto out;
572 	}
573 	tfile->queue_index = tun->numqueues;
574 	rcu_assign_pointer(tfile->tun, tun);
575 	rcu_assign_pointer(tun->tfiles[tun->numqueues], tfile);
576 	tun->numqueues++;
577 
578 	if (tfile->detached)
579 		tun_enable_queue(tfile);
580 	else
581 		sock_hold(&tfile->sk);
582 
583 	tun_set_real_num_queues(tun);
584 
585 	/* device is allowed to go away first, so no need to hold extra
586 	 * refcnt.
587 	 */
588 
589 out:
590 	return err;
591 }
592 
593 static struct tun_struct *__tun_get(struct tun_file *tfile)
594 {
595 	struct tun_struct *tun;
596 
597 	rcu_read_lock();
598 	tun = rcu_dereference(tfile->tun);
599 	if (tun)
600 		dev_hold(tun->dev);
601 	rcu_read_unlock();
602 
603 	return tun;
604 }
605 
606 static struct tun_struct *tun_get(struct file *file)
607 {
608 	return __tun_get(file->private_data);
609 }
610 
611 static void tun_put(struct tun_struct *tun)
612 {
613 	dev_put(tun->dev);
614 }
615 
616 /* TAP filtering */
617 static void addr_hash_set(u32 *mask, const u8 *addr)
618 {
619 	int n = ether_crc(ETH_ALEN, addr) >> 26;
620 	mask[n >> 5] |= (1 << (n & 31));
621 }
622 
623 static unsigned int addr_hash_test(const u32 *mask, const u8 *addr)
624 {
625 	int n = ether_crc(ETH_ALEN, addr) >> 26;
626 	return mask[n >> 5] & (1 << (n & 31));
627 }
628 
629 static int update_filter(struct tap_filter *filter, void __user *arg)
630 {
631 	struct { u8 u[ETH_ALEN]; } *addr;
632 	struct tun_filter uf;
633 	int err, alen, n, nexact;
634 
635 	if (copy_from_user(&uf, arg, sizeof(uf)))
636 		return -EFAULT;
637 
638 	if (!uf.count) {
639 		/* Disabled */
640 		filter->count = 0;
641 		return 0;
642 	}
643 
644 	alen = ETH_ALEN * uf.count;
645 	addr = kmalloc(alen, GFP_KERNEL);
646 	if (!addr)
647 		return -ENOMEM;
648 
649 	if (copy_from_user(addr, arg + sizeof(uf), alen)) {
650 		err = -EFAULT;
651 		goto done;
652 	}
653 
654 	/* The filter is updated without holding any locks. Which is
655 	 * perfectly safe. We disable it first and in the worst
656 	 * case we'll accept a few undesired packets. */
657 	filter->count = 0;
658 	wmb();
659 
660 	/* Use first set of addresses as an exact filter */
661 	for (n = 0; n < uf.count && n < FLT_EXACT_COUNT; n++)
662 		memcpy(filter->addr[n], addr[n].u, ETH_ALEN);
663 
664 	nexact = n;
665 
666 	/* Remaining multicast addresses are hashed,
667 	 * unicast will leave the filter disabled. */
668 	memset(filter->mask, 0, sizeof(filter->mask));
669 	for (; n < uf.count; n++) {
670 		if (!is_multicast_ether_addr(addr[n].u)) {
671 			err = 0; /* no filter */
672 			goto done;
673 		}
674 		addr_hash_set(filter->mask, addr[n].u);
675 	}
676 
677 	/* For ALLMULTI just set the mask to all ones.
678 	 * This overrides the mask populated above. */
679 	if ((uf.flags & TUN_FLT_ALLMULTI))
680 		memset(filter->mask, ~0, sizeof(filter->mask));
681 
682 	/* Now enable the filter */
683 	wmb();
684 	filter->count = nexact;
685 
686 	/* Return the number of exact filters */
687 	err = nexact;
688 
689 done:
690 	kfree(addr);
691 	return err;
692 }
693 
694 /* Returns: 0 - drop, !=0 - accept */
695 static int run_filter(struct tap_filter *filter, const struct sk_buff *skb)
696 {
697 	/* Cannot use eth_hdr(skb) here because skb_mac_hdr() is incorrect
698 	 * at this point. */
699 	struct ethhdr *eh = (struct ethhdr *) skb->data;
700 	int i;
701 
702 	/* Exact match */
703 	for (i = 0; i < filter->count; i++)
704 		if (ether_addr_equal(eh->h_dest, filter->addr[i]))
705 			return 1;
706 
707 	/* Inexact match (multicast only) */
708 	if (is_multicast_ether_addr(eh->h_dest))
709 		return addr_hash_test(filter->mask, eh->h_dest);
710 
711 	return 0;
712 }
713 
714 /*
715  * Checks whether the packet is accepted or not.
716  * Returns: 0 - drop, !=0 - accept
717  */
718 static int check_filter(struct tap_filter *filter, const struct sk_buff *skb)
719 {
720 	if (!filter->count)
721 		return 1;
722 
723 	return run_filter(filter, skb);
724 }
725 
726 /* Network device part of the driver */
727 
728 static const struct ethtool_ops tun_ethtool_ops;
729 
730 /* Net device detach from fd. */
731 static void tun_net_uninit(struct net_device *dev)
732 {
733 	tun_detach_all(dev);
734 }
735 
736 /* Net device open. */
737 static int tun_net_open(struct net_device *dev)
738 {
739 	netif_tx_start_all_queues(dev);
740 	return 0;
741 }
742 
743 /* Net device close. */
744 static int tun_net_close(struct net_device *dev)
745 {
746 	netif_tx_stop_all_queues(dev);
747 	return 0;
748 }
749 
750 /* Net device start xmit */
751 static netdev_tx_t tun_net_xmit(struct sk_buff *skb, struct net_device *dev)
752 {
753 	struct tun_struct *tun = netdev_priv(dev);
754 	int txq = skb->queue_mapping;
755 	struct tun_file *tfile;
756 	u32 numqueues = 0;
757 
758 	rcu_read_lock();
759 	tfile = rcu_dereference(tun->tfiles[txq]);
760 	numqueues = ACCESS_ONCE(tun->numqueues);
761 
762 	/* Drop packet if interface is not attached */
763 	if (txq >= numqueues)
764 		goto drop;
765 
766 	if (numqueues == 1) {
767 		/* Select queue was not called for the skbuff, so we extract the
768 		 * RPS hash and save it into the flow_table here.
769 		 */
770 		__u32 rxhash;
771 
772 		rxhash = skb_get_hash(skb);
773 		if (rxhash) {
774 			struct tun_flow_entry *e;
775 			e = tun_flow_find(&tun->flows[tun_hashfn(rxhash)],
776 					rxhash);
777 			if (e)
778 				tun_flow_save_rps_rxhash(e, rxhash);
779 		}
780 	}
781 
782 	tun_debug(KERN_INFO, tun, "tun_net_xmit %d\n", skb->len);
783 
784 	BUG_ON(!tfile);
785 
786 	/* Drop if the filter does not like it.
787 	 * This is a noop if the filter is disabled.
788 	 * Filter can be enabled only for the TAP devices. */
789 	if (!check_filter(&tun->txflt, skb))
790 		goto drop;
791 
792 	if (tfile->socket.sk->sk_filter &&
793 	    sk_filter(tfile->socket.sk, skb))
794 		goto drop;
795 
796 	/* Limit the number of packets queued by dividing txq length with the
797 	 * number of queues.
798 	 */
799 	if (skb_queue_len(&tfile->socket.sk->sk_receive_queue) * numqueues
800 			  >= dev->tx_queue_len)
801 		goto drop;
802 
803 	if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC)))
804 		goto drop;
805 
806 	if (skb->sk) {
807 		sock_tx_timestamp(skb->sk, &skb_shinfo(skb)->tx_flags);
808 		sw_tx_timestamp(skb);
809 	}
810 
811 	/* Orphan the skb - required as we might hang on to it
812 	 * for indefinite time.
813 	 */
814 	skb_orphan(skb);
815 
816 	nf_reset(skb);
817 
818 	/* Enqueue packet */
819 	skb_queue_tail(&tfile->socket.sk->sk_receive_queue, skb);
820 
821 	/* Notify and wake up reader process */
822 	if (tfile->flags & TUN_FASYNC)
823 		kill_fasync(&tfile->fasync, SIGIO, POLL_IN);
824 	tfile->socket.sk->sk_data_ready(tfile->socket.sk);
825 
826 	rcu_read_unlock();
827 	return NETDEV_TX_OK;
828 
829 drop:
830 	dev->stats.tx_dropped++;
831 	skb_tx_error(skb);
832 	kfree_skb(skb);
833 	rcu_read_unlock();
834 	return NET_XMIT_DROP;
835 }
836 
837 static void tun_net_mclist(struct net_device *dev)
838 {
839 	/*
840 	 * This callback is supposed to deal with mc filter in
841 	 * _rx_ path and has nothing to do with the _tx_ path.
842 	 * In rx path we always accept everything userspace gives us.
843 	 */
844 }
845 
846 #define MIN_MTU 68
847 #define MAX_MTU 65535
848 
849 static int
850 tun_net_change_mtu(struct net_device *dev, int new_mtu)
851 {
852 	if (new_mtu < MIN_MTU || new_mtu + dev->hard_header_len > MAX_MTU)
853 		return -EINVAL;
854 	dev->mtu = new_mtu;
855 	return 0;
856 }
857 
858 static netdev_features_t tun_net_fix_features(struct net_device *dev,
859 	netdev_features_t features)
860 {
861 	struct tun_struct *tun = netdev_priv(dev);
862 
863 	return (features & tun->set_features) | (features & ~TUN_USER_FEATURES);
864 }
865 #ifdef CONFIG_NET_POLL_CONTROLLER
866 static void tun_poll_controller(struct net_device *dev)
867 {
868 	/*
869 	 * Tun only receives frames when:
870 	 * 1) the char device endpoint gets data from user space
871 	 * 2) the tun socket gets a sendmsg call from user space
872 	 * Since both of those are synchronous operations, we are guaranteed
873 	 * never to have pending data when we poll for it
874 	 * so there is nothing to do here but return.
875 	 * We need this though so netpoll recognizes us as an interface that
876 	 * supports polling, which enables bridge devices in virt setups to
877 	 * still use netconsole
878 	 */
879 	return;
880 }
881 #endif
882 static const struct net_device_ops tun_netdev_ops = {
883 	.ndo_uninit		= tun_net_uninit,
884 	.ndo_open		= tun_net_open,
885 	.ndo_stop		= tun_net_close,
886 	.ndo_start_xmit		= tun_net_xmit,
887 	.ndo_change_mtu		= tun_net_change_mtu,
888 	.ndo_fix_features	= tun_net_fix_features,
889 	.ndo_select_queue	= tun_select_queue,
890 #ifdef CONFIG_NET_POLL_CONTROLLER
891 	.ndo_poll_controller	= tun_poll_controller,
892 #endif
893 };
894 
895 static const struct net_device_ops tap_netdev_ops = {
896 	.ndo_uninit		= tun_net_uninit,
897 	.ndo_open		= tun_net_open,
898 	.ndo_stop		= tun_net_close,
899 	.ndo_start_xmit		= tun_net_xmit,
900 	.ndo_change_mtu		= tun_net_change_mtu,
901 	.ndo_fix_features	= tun_net_fix_features,
902 	.ndo_set_rx_mode	= tun_net_mclist,
903 	.ndo_set_mac_address	= eth_mac_addr,
904 	.ndo_validate_addr	= eth_validate_addr,
905 	.ndo_select_queue	= tun_select_queue,
906 #ifdef CONFIG_NET_POLL_CONTROLLER
907 	.ndo_poll_controller	= tun_poll_controller,
908 #endif
909 };
910 
911 static void tun_flow_init(struct tun_struct *tun)
912 {
913 	int i;
914 
915 	for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++)
916 		INIT_HLIST_HEAD(&tun->flows[i]);
917 
918 	tun->ageing_time = TUN_FLOW_EXPIRE;
919 	setup_timer(&tun->flow_gc_timer, tun_flow_cleanup, (unsigned long)tun);
920 	mod_timer(&tun->flow_gc_timer,
921 		  round_jiffies_up(jiffies + tun->ageing_time));
922 }
923 
924 static void tun_flow_uninit(struct tun_struct *tun)
925 {
926 	del_timer_sync(&tun->flow_gc_timer);
927 	tun_flow_flush(tun);
928 }
929 
930 /* Initialize net device. */
931 static void tun_net_init(struct net_device *dev)
932 {
933 	struct tun_struct *tun = netdev_priv(dev);
934 
935 	switch (tun->flags & TUN_TYPE_MASK) {
936 	case IFF_TUN:
937 		dev->netdev_ops = &tun_netdev_ops;
938 
939 		/* Point-to-Point TUN Device */
940 		dev->hard_header_len = 0;
941 		dev->addr_len = 0;
942 		dev->mtu = 1500;
943 
944 		/* Zero header length */
945 		dev->type = ARPHRD_NONE;
946 		dev->flags = IFF_POINTOPOINT | IFF_NOARP | IFF_MULTICAST;
947 		dev->tx_queue_len = TUN_READQ_SIZE;  /* We prefer our own queue length */
948 		break;
949 
950 	case IFF_TAP:
951 		dev->netdev_ops = &tap_netdev_ops;
952 		/* Ethernet TAP Device */
953 		ether_setup(dev);
954 		dev->priv_flags &= ~IFF_TX_SKB_SHARING;
955 		dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
956 
957 		eth_hw_addr_random(dev);
958 
959 		dev->tx_queue_len = TUN_READQ_SIZE;  /* We prefer our own queue length */
960 		break;
961 	}
962 }
963 
964 /* Character device part */
965 
966 /* Poll */
967 static unsigned int tun_chr_poll(struct file *file, poll_table *wait)
968 {
969 	struct tun_file *tfile = file->private_data;
970 	struct tun_struct *tun = __tun_get(tfile);
971 	struct sock *sk;
972 	unsigned int mask = 0;
973 
974 	if (!tun)
975 		return POLLERR;
976 
977 	sk = tfile->socket.sk;
978 
979 	tun_debug(KERN_INFO, tun, "tun_chr_poll\n");
980 
981 	poll_wait(file, sk_sleep(sk), wait);
982 
983 	if (!skb_queue_empty(&sk->sk_receive_queue))
984 		mask |= POLLIN | POLLRDNORM;
985 
986 	if (sock_writeable(sk) ||
987 	    (!test_and_set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags) &&
988 	     sock_writeable(sk)))
989 		mask |= POLLOUT | POLLWRNORM;
990 
991 	if (tun->dev->reg_state != NETREG_REGISTERED)
992 		mask = POLLERR;
993 
994 	tun_put(tun);
995 	return mask;
996 }
997 
998 /* prepad is the amount to reserve at front.  len is length after that.
999  * linear is a hint as to how much to copy (usually headers). */
1000 static struct sk_buff *tun_alloc_skb(struct tun_file *tfile,
1001 				     size_t prepad, size_t len,
1002 				     size_t linear, int noblock)
1003 {
1004 	struct sock *sk = tfile->socket.sk;
1005 	struct sk_buff *skb;
1006 	int err;
1007 
1008 	/* Under a page?  Don't bother with paged skb. */
1009 	if (prepad + len < PAGE_SIZE || !linear)
1010 		linear = len;
1011 
1012 	skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
1013 				   &err, 0);
1014 	if (!skb)
1015 		return ERR_PTR(err);
1016 
1017 	skb_reserve(skb, prepad);
1018 	skb_put(skb, linear);
1019 	skb->data_len = len - linear;
1020 	skb->len += len - linear;
1021 
1022 	return skb;
1023 }
1024 
1025 /* Get packet from user space buffer */
1026 static ssize_t tun_get_user(struct tun_struct *tun, struct tun_file *tfile,
1027 			    void *msg_control, struct iov_iter *from,
1028 			    int noblock)
1029 {
1030 	struct tun_pi pi = { 0, cpu_to_be16(ETH_P_IP) };
1031 	struct sk_buff *skb;
1032 	size_t total_len = iov_iter_count(from);
1033 	size_t len = total_len, align = NET_SKB_PAD, linear;
1034 	struct virtio_net_hdr gso = { 0 };
1035 	int good_linear;
1036 	int copylen;
1037 	bool zerocopy = false;
1038 	int err;
1039 	u32 rxhash;
1040 	ssize_t n;
1041 
1042 	if (!(tun->flags & IFF_NO_PI)) {
1043 		if (len < sizeof(pi))
1044 			return -EINVAL;
1045 		len -= sizeof(pi);
1046 
1047 		n = copy_from_iter(&pi, sizeof(pi), from);
1048 		if (n != sizeof(pi))
1049 			return -EFAULT;
1050 	}
1051 
1052 	if (tun->flags & IFF_VNET_HDR) {
1053 		if (len < tun->vnet_hdr_sz)
1054 			return -EINVAL;
1055 		len -= tun->vnet_hdr_sz;
1056 
1057 		n = copy_from_iter(&gso, sizeof(gso), from);
1058 		if (n != sizeof(gso))
1059 			return -EFAULT;
1060 
1061 		if ((gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
1062 		    tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2 > tun16_to_cpu(tun, gso.hdr_len))
1063 			gso.hdr_len = cpu_to_tun16(tun, tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2);
1064 
1065 		if (tun16_to_cpu(tun, gso.hdr_len) > len)
1066 			return -EINVAL;
1067 		iov_iter_advance(from, tun->vnet_hdr_sz - sizeof(gso));
1068 	}
1069 
1070 	if ((tun->flags & TUN_TYPE_MASK) == IFF_TAP) {
1071 		align += NET_IP_ALIGN;
1072 		if (unlikely(len < ETH_HLEN ||
1073 			     (gso.hdr_len && tun16_to_cpu(tun, gso.hdr_len) < ETH_HLEN)))
1074 			return -EINVAL;
1075 	}
1076 
1077 	good_linear = SKB_MAX_HEAD(align);
1078 
1079 	if (msg_control) {
1080 		struct iov_iter i = *from;
1081 
1082 		/* There are 256 bytes to be copied in skb, so there is
1083 		 * enough room for skb expand head in case it is used.
1084 		 * The rest of the buffer is mapped from userspace.
1085 		 */
1086 		copylen = gso.hdr_len ? tun16_to_cpu(tun, gso.hdr_len) : GOODCOPY_LEN;
1087 		if (copylen > good_linear)
1088 			copylen = good_linear;
1089 		linear = copylen;
1090 		iov_iter_advance(&i, copylen);
1091 		if (iov_iter_npages(&i, INT_MAX) <= MAX_SKB_FRAGS)
1092 			zerocopy = true;
1093 	}
1094 
1095 	if (!zerocopy) {
1096 		copylen = len;
1097 		if (tun16_to_cpu(tun, gso.hdr_len) > good_linear)
1098 			linear = good_linear;
1099 		else
1100 			linear = tun16_to_cpu(tun, gso.hdr_len);
1101 	}
1102 
1103 	skb = tun_alloc_skb(tfile, align, copylen, linear, noblock);
1104 	if (IS_ERR(skb)) {
1105 		if (PTR_ERR(skb) != -EAGAIN)
1106 			tun->dev->stats.rx_dropped++;
1107 		return PTR_ERR(skb);
1108 	}
1109 
1110 	if (zerocopy)
1111 		err = zerocopy_sg_from_iter(skb, from);
1112 	else {
1113 		err = skb_copy_datagram_from_iter(skb, 0, from, len);
1114 		if (!err && msg_control) {
1115 			struct ubuf_info *uarg = msg_control;
1116 			uarg->callback(uarg, false);
1117 		}
1118 	}
1119 
1120 	if (err) {
1121 		tun->dev->stats.rx_dropped++;
1122 		kfree_skb(skb);
1123 		return -EFAULT;
1124 	}
1125 
1126 	if (gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
1127 		if (!skb_partial_csum_set(skb, tun16_to_cpu(tun, gso.csum_start),
1128 					  tun16_to_cpu(tun, gso.csum_offset))) {
1129 			tun->dev->stats.rx_frame_errors++;
1130 			kfree_skb(skb);
1131 			return -EINVAL;
1132 		}
1133 	}
1134 
1135 	switch (tun->flags & TUN_TYPE_MASK) {
1136 	case IFF_TUN:
1137 		if (tun->flags & IFF_NO_PI) {
1138 			switch (skb->data[0] & 0xf0) {
1139 			case 0x40:
1140 				pi.proto = htons(ETH_P_IP);
1141 				break;
1142 			case 0x60:
1143 				pi.proto = htons(ETH_P_IPV6);
1144 				break;
1145 			default:
1146 				tun->dev->stats.rx_dropped++;
1147 				kfree_skb(skb);
1148 				return -EINVAL;
1149 			}
1150 		}
1151 
1152 		skb_reset_mac_header(skb);
1153 		skb->protocol = pi.proto;
1154 		skb->dev = tun->dev;
1155 		break;
1156 	case IFF_TAP:
1157 		skb->protocol = eth_type_trans(skb, tun->dev);
1158 		break;
1159 	}
1160 
1161 	if (gso.gso_type != VIRTIO_NET_HDR_GSO_NONE) {
1162 		pr_debug("GSO!\n");
1163 		switch (gso.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
1164 		case VIRTIO_NET_HDR_GSO_TCPV4:
1165 			skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
1166 			break;
1167 		case VIRTIO_NET_HDR_GSO_TCPV6:
1168 			skb_shinfo(skb)->gso_type = SKB_GSO_TCPV6;
1169 			break;
1170 		case VIRTIO_NET_HDR_GSO_UDP:
1171 			skb_shinfo(skb)->gso_type = SKB_GSO_UDP;
1172 			break;
1173 		default:
1174 			tun->dev->stats.rx_frame_errors++;
1175 			kfree_skb(skb);
1176 			return -EINVAL;
1177 		}
1178 
1179 		if (gso.gso_type & VIRTIO_NET_HDR_GSO_ECN)
1180 			skb_shinfo(skb)->gso_type |= SKB_GSO_TCP_ECN;
1181 
1182 		skb_shinfo(skb)->gso_size = tun16_to_cpu(tun, gso.gso_size);
1183 		if (skb_shinfo(skb)->gso_size == 0) {
1184 			tun->dev->stats.rx_frame_errors++;
1185 			kfree_skb(skb);
1186 			return -EINVAL;
1187 		}
1188 
1189 		/* Header must be checked, and gso_segs computed. */
1190 		skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
1191 		skb_shinfo(skb)->gso_segs = 0;
1192 	}
1193 
1194 	/* copy skb_ubuf_info for callback when skb has no error */
1195 	if (zerocopy) {
1196 		skb_shinfo(skb)->destructor_arg = msg_control;
1197 		skb_shinfo(skb)->tx_flags |= SKBTX_DEV_ZEROCOPY;
1198 		skb_shinfo(skb)->tx_flags |= SKBTX_SHARED_FRAG;
1199 	}
1200 
1201 	skb_reset_network_header(skb);
1202 	skb_probe_transport_header(skb, 0);
1203 
1204 	rxhash = skb_get_hash(skb);
1205 	netif_rx_ni(skb);
1206 
1207 	tun->dev->stats.rx_packets++;
1208 	tun->dev->stats.rx_bytes += len;
1209 
1210 	tun_flow_update(tun, rxhash, tfile);
1211 	return total_len;
1212 }
1213 
1214 static ssize_t tun_chr_write_iter(struct kiocb *iocb, struct iov_iter *from)
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 	result = tun_get_user(tun, tfile, NULL, from, file->f_flags & O_NONBLOCK);
1225 
1226 	tun_put(tun);
1227 	return result;
1228 }
1229 
1230 /* Put packet to the user space buffer */
1231 static ssize_t tun_put_user(struct tun_struct *tun,
1232 			    struct tun_file *tfile,
1233 			    struct sk_buff *skb,
1234 			    struct iov_iter *iter)
1235 {
1236 	struct tun_pi pi = { 0, skb->protocol };
1237 	ssize_t total;
1238 	int vlan_offset = 0;
1239 	int vlan_hlen = 0;
1240 	int vnet_hdr_sz = 0;
1241 
1242 	if (skb_vlan_tag_present(skb))
1243 		vlan_hlen = VLAN_HLEN;
1244 
1245 	if (tun->flags & IFF_VNET_HDR)
1246 		vnet_hdr_sz = tun->vnet_hdr_sz;
1247 
1248 	total = skb->len + vlan_hlen + vnet_hdr_sz;
1249 
1250 	if (!(tun->flags & IFF_NO_PI)) {
1251 		if (iov_iter_count(iter) < sizeof(pi))
1252 			return -EINVAL;
1253 
1254 		total += sizeof(pi);
1255 		if (iov_iter_count(iter) < total) {
1256 			/* Packet will be striped */
1257 			pi.flags |= TUN_PKT_STRIP;
1258 		}
1259 
1260 		if (copy_to_iter(&pi, sizeof(pi), iter) != sizeof(pi))
1261 			return -EFAULT;
1262 	}
1263 
1264 	if (vnet_hdr_sz) {
1265 		struct virtio_net_hdr gso = { 0 }; /* no info leak */
1266 		if (iov_iter_count(iter) < vnet_hdr_sz)
1267 			return -EINVAL;
1268 
1269 		if (skb_is_gso(skb)) {
1270 			struct skb_shared_info *sinfo = skb_shinfo(skb);
1271 
1272 			/* This is a hint as to how much should be linear. */
1273 			gso.hdr_len = cpu_to_tun16(tun, skb_headlen(skb));
1274 			gso.gso_size = cpu_to_tun16(tun, sinfo->gso_size);
1275 			if (sinfo->gso_type & SKB_GSO_TCPV4)
1276 				gso.gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
1277 			else if (sinfo->gso_type & SKB_GSO_TCPV6)
1278 				gso.gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
1279 			else if (sinfo->gso_type & SKB_GSO_UDP)
1280 				gso.gso_type = VIRTIO_NET_HDR_GSO_UDP;
1281 			else {
1282 				pr_err("unexpected GSO type: "
1283 				       "0x%x, gso_size %d, hdr_len %d\n",
1284 				       sinfo->gso_type, tun16_to_cpu(tun, gso.gso_size),
1285 				       tun16_to_cpu(tun, gso.hdr_len));
1286 				print_hex_dump(KERN_ERR, "tun: ",
1287 					       DUMP_PREFIX_NONE,
1288 					       16, 1, skb->head,
1289 					       min((int)tun16_to_cpu(tun, gso.hdr_len), 64), true);
1290 				WARN_ON_ONCE(1);
1291 				return -EINVAL;
1292 			}
1293 			if (sinfo->gso_type & SKB_GSO_TCP_ECN)
1294 				gso.gso_type |= VIRTIO_NET_HDR_GSO_ECN;
1295 		} else
1296 			gso.gso_type = VIRTIO_NET_HDR_GSO_NONE;
1297 
1298 		if (skb->ip_summed == CHECKSUM_PARTIAL) {
1299 			gso.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
1300 			gso.csum_start = cpu_to_tun16(tun, skb_checksum_start_offset(skb) +
1301 						      vlan_hlen);
1302 			gso.csum_offset = cpu_to_tun16(tun, skb->csum_offset);
1303 		} else if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
1304 			gso.flags = VIRTIO_NET_HDR_F_DATA_VALID;
1305 		} /* else everything is zero */
1306 
1307 		if (copy_to_iter(&gso, sizeof(gso), iter) != sizeof(gso))
1308 			return -EFAULT;
1309 
1310 		iov_iter_advance(iter, vnet_hdr_sz - sizeof(gso));
1311 	}
1312 
1313 	if (vlan_hlen) {
1314 		int ret;
1315 		struct {
1316 			__be16 h_vlan_proto;
1317 			__be16 h_vlan_TCI;
1318 		} veth;
1319 
1320 		veth.h_vlan_proto = skb->vlan_proto;
1321 		veth.h_vlan_TCI = htons(skb_vlan_tag_get(skb));
1322 
1323 		vlan_offset = offsetof(struct vlan_ethhdr, h_vlan_proto);
1324 
1325 		ret = skb_copy_datagram_iter(skb, 0, iter, vlan_offset);
1326 		if (ret || !iov_iter_count(iter))
1327 			goto done;
1328 
1329 		ret = copy_to_iter(&veth, sizeof(veth), iter);
1330 		if (ret != sizeof(veth) || !iov_iter_count(iter))
1331 			goto done;
1332 	}
1333 
1334 	skb_copy_datagram_iter(skb, vlan_offset, iter, skb->len - vlan_offset);
1335 
1336 done:
1337 	tun->dev->stats.tx_packets++;
1338 	tun->dev->stats.tx_bytes += skb->len + vlan_hlen;
1339 
1340 	return total;
1341 }
1342 
1343 static ssize_t tun_do_read(struct tun_struct *tun, struct tun_file *tfile,
1344 			   struct iov_iter *to,
1345 			   int noblock)
1346 {
1347 	struct sk_buff *skb;
1348 	ssize_t ret;
1349 	int peeked, err, off = 0;
1350 
1351 	tun_debug(KERN_INFO, tun, "tun_do_read\n");
1352 
1353 	if (!iov_iter_count(to))
1354 		return 0;
1355 
1356 	if (tun->dev->reg_state != NETREG_REGISTERED)
1357 		return -EIO;
1358 
1359 	/* Read frames from queue */
1360 	skb = __skb_recv_datagram(tfile->socket.sk, noblock ? MSG_DONTWAIT : 0,
1361 				  &peeked, &off, &err);
1362 	if (!skb)
1363 		return err;
1364 
1365 	ret = tun_put_user(tun, tfile, skb, to);
1366 	if (unlikely(ret < 0))
1367 		kfree_skb(skb);
1368 	else
1369 		consume_skb(skb);
1370 
1371 	return ret;
1372 }
1373 
1374 static ssize_t tun_chr_read_iter(struct kiocb *iocb, struct iov_iter *to)
1375 {
1376 	struct file *file = iocb->ki_filp;
1377 	struct tun_file *tfile = file->private_data;
1378 	struct tun_struct *tun = __tun_get(tfile);
1379 	ssize_t len = iov_iter_count(to), ret;
1380 
1381 	if (!tun)
1382 		return -EBADFD;
1383 	ret = tun_do_read(tun, tfile, to, file->f_flags & O_NONBLOCK);
1384 	ret = min_t(ssize_t, ret, len);
1385 	if (ret > 0)
1386 		iocb->ki_pos = ret;
1387 	tun_put(tun);
1388 	return ret;
1389 }
1390 
1391 static void tun_free_netdev(struct net_device *dev)
1392 {
1393 	struct tun_struct *tun = netdev_priv(dev);
1394 
1395 	BUG_ON(!(list_empty(&tun->disabled)));
1396 	tun_flow_uninit(tun);
1397 	security_tun_dev_free_security(tun->security);
1398 	free_netdev(dev);
1399 }
1400 
1401 static void tun_setup(struct net_device *dev)
1402 {
1403 	struct tun_struct *tun = netdev_priv(dev);
1404 
1405 	tun->owner = INVALID_UID;
1406 	tun->group = INVALID_GID;
1407 
1408 	dev->ethtool_ops = &tun_ethtool_ops;
1409 	dev->destructor = tun_free_netdev;
1410 }
1411 
1412 /* Trivial set of netlink ops to allow deleting tun or tap
1413  * device with netlink.
1414  */
1415 static int tun_validate(struct nlattr *tb[], struct nlattr *data[])
1416 {
1417 	return -EINVAL;
1418 }
1419 
1420 static struct rtnl_link_ops tun_link_ops __read_mostly = {
1421 	.kind		= DRV_NAME,
1422 	.priv_size	= sizeof(struct tun_struct),
1423 	.setup		= tun_setup,
1424 	.validate	= tun_validate,
1425 };
1426 
1427 static void tun_sock_write_space(struct sock *sk)
1428 {
1429 	struct tun_file *tfile;
1430 	wait_queue_head_t *wqueue;
1431 
1432 	if (!sock_writeable(sk))
1433 		return;
1434 
1435 	if (!test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags))
1436 		return;
1437 
1438 	wqueue = sk_sleep(sk);
1439 	if (wqueue && waitqueue_active(wqueue))
1440 		wake_up_interruptible_sync_poll(wqueue, POLLOUT |
1441 						POLLWRNORM | POLLWRBAND);
1442 
1443 	tfile = container_of(sk, struct tun_file, sk);
1444 	kill_fasync(&tfile->fasync, SIGIO, POLL_OUT);
1445 }
1446 
1447 static int tun_sendmsg(struct socket *sock, struct msghdr *m, size_t total_len)
1448 {
1449 	int ret;
1450 	struct tun_file *tfile = container_of(sock, struct tun_file, socket);
1451 	struct tun_struct *tun = __tun_get(tfile);
1452 
1453 	if (!tun)
1454 		return -EBADFD;
1455 
1456 	ret = tun_get_user(tun, tfile, m->msg_control, &m->msg_iter,
1457 			   m->msg_flags & MSG_DONTWAIT);
1458 	tun_put(tun);
1459 	return ret;
1460 }
1461 
1462 static int tun_recvmsg(struct socket *sock, struct msghdr *m, size_t total_len,
1463 		       int flags)
1464 {
1465 	struct tun_file *tfile = container_of(sock, struct tun_file, socket);
1466 	struct tun_struct *tun = __tun_get(tfile);
1467 	int ret;
1468 
1469 	if (!tun)
1470 		return -EBADFD;
1471 
1472 	if (flags & ~(MSG_DONTWAIT|MSG_TRUNC|MSG_ERRQUEUE)) {
1473 		ret = -EINVAL;
1474 		goto out;
1475 	}
1476 	if (flags & MSG_ERRQUEUE) {
1477 		ret = sock_recv_errqueue(sock->sk, m, total_len,
1478 					 SOL_PACKET, TUN_TX_TIMESTAMP);
1479 		goto out;
1480 	}
1481 	ret = tun_do_read(tun, tfile, &m->msg_iter, flags & MSG_DONTWAIT);
1482 	if (ret > (ssize_t)total_len) {
1483 		m->msg_flags |= MSG_TRUNC;
1484 		ret = flags & MSG_TRUNC ? ret : total_len;
1485 	}
1486 out:
1487 	tun_put(tun);
1488 	return ret;
1489 }
1490 
1491 /* Ops structure to mimic raw sockets with tun */
1492 static const struct proto_ops tun_socket_ops = {
1493 	.sendmsg = tun_sendmsg,
1494 	.recvmsg = tun_recvmsg,
1495 };
1496 
1497 static struct proto tun_proto = {
1498 	.name		= "tun",
1499 	.owner		= THIS_MODULE,
1500 	.obj_size	= sizeof(struct tun_file),
1501 };
1502 
1503 static int tun_flags(struct tun_struct *tun)
1504 {
1505 	return tun->flags & (TUN_FEATURES | IFF_PERSIST | IFF_TUN | IFF_TAP);
1506 }
1507 
1508 static ssize_t tun_show_flags(struct device *dev, struct device_attribute *attr,
1509 			      char *buf)
1510 {
1511 	struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1512 	return sprintf(buf, "0x%x\n", tun_flags(tun));
1513 }
1514 
1515 static ssize_t tun_show_owner(struct device *dev, struct device_attribute *attr,
1516 			      char *buf)
1517 {
1518 	struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1519 	return uid_valid(tun->owner)?
1520 		sprintf(buf, "%u\n",
1521 			from_kuid_munged(current_user_ns(), tun->owner)):
1522 		sprintf(buf, "-1\n");
1523 }
1524 
1525 static ssize_t tun_show_group(struct device *dev, struct device_attribute *attr,
1526 			      char *buf)
1527 {
1528 	struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1529 	return gid_valid(tun->group) ?
1530 		sprintf(buf, "%u\n",
1531 			from_kgid_munged(current_user_ns(), tun->group)):
1532 		sprintf(buf, "-1\n");
1533 }
1534 
1535 static DEVICE_ATTR(tun_flags, 0444, tun_show_flags, NULL);
1536 static DEVICE_ATTR(owner, 0444, tun_show_owner, NULL);
1537 static DEVICE_ATTR(group, 0444, tun_show_group, NULL);
1538 
1539 static struct attribute *tun_dev_attrs[] = {
1540 	&dev_attr_tun_flags.attr,
1541 	&dev_attr_owner.attr,
1542 	&dev_attr_group.attr,
1543 	NULL
1544 };
1545 
1546 static const struct attribute_group tun_attr_group = {
1547 	.attrs = tun_dev_attrs
1548 };
1549 
1550 static int tun_set_iff(struct net *net, struct file *file, struct ifreq *ifr)
1551 {
1552 	struct tun_struct *tun;
1553 	struct tun_file *tfile = file->private_data;
1554 	struct net_device *dev;
1555 	int err;
1556 
1557 	if (tfile->detached)
1558 		return -EINVAL;
1559 
1560 	dev = __dev_get_by_name(net, ifr->ifr_name);
1561 	if (dev) {
1562 		if (ifr->ifr_flags & IFF_TUN_EXCL)
1563 			return -EBUSY;
1564 		if ((ifr->ifr_flags & IFF_TUN) && dev->netdev_ops == &tun_netdev_ops)
1565 			tun = netdev_priv(dev);
1566 		else if ((ifr->ifr_flags & IFF_TAP) && dev->netdev_ops == &tap_netdev_ops)
1567 			tun = netdev_priv(dev);
1568 		else
1569 			return -EINVAL;
1570 
1571 		if (!!(ifr->ifr_flags & IFF_MULTI_QUEUE) !=
1572 		    !!(tun->flags & IFF_MULTI_QUEUE))
1573 			return -EINVAL;
1574 
1575 		if (tun_not_capable(tun))
1576 			return -EPERM;
1577 		err = security_tun_dev_open(tun->security);
1578 		if (err < 0)
1579 			return err;
1580 
1581 		err = tun_attach(tun, file, ifr->ifr_flags & IFF_NOFILTER);
1582 		if (err < 0)
1583 			return err;
1584 
1585 		if (tun->flags & IFF_MULTI_QUEUE &&
1586 		    (tun->numqueues + tun->numdisabled > 1)) {
1587 			/* One or more queue has already been attached, no need
1588 			 * to initialize the device again.
1589 			 */
1590 			return 0;
1591 		}
1592 	}
1593 	else {
1594 		char *name;
1595 		unsigned long flags = 0;
1596 		int queues = ifr->ifr_flags & IFF_MULTI_QUEUE ?
1597 			     MAX_TAP_QUEUES : 1;
1598 
1599 		if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
1600 			return -EPERM;
1601 		err = security_tun_dev_create();
1602 		if (err < 0)
1603 			return err;
1604 
1605 		/* Set dev type */
1606 		if (ifr->ifr_flags & IFF_TUN) {
1607 			/* TUN device */
1608 			flags |= IFF_TUN;
1609 			name = "tun%d";
1610 		} else if (ifr->ifr_flags & IFF_TAP) {
1611 			/* TAP device */
1612 			flags |= IFF_TAP;
1613 			name = "tap%d";
1614 		} else
1615 			return -EINVAL;
1616 
1617 		if (*ifr->ifr_name)
1618 			name = ifr->ifr_name;
1619 
1620 		dev = alloc_netdev_mqs(sizeof(struct tun_struct), name,
1621 				       NET_NAME_UNKNOWN, tun_setup, queues,
1622 				       queues);
1623 
1624 		if (!dev)
1625 			return -ENOMEM;
1626 
1627 		dev_net_set(dev, net);
1628 		dev->rtnl_link_ops = &tun_link_ops;
1629 		dev->ifindex = tfile->ifindex;
1630 		dev->sysfs_groups[0] = &tun_attr_group;
1631 
1632 		tun = netdev_priv(dev);
1633 		tun->dev = dev;
1634 		tun->flags = flags;
1635 		tun->txflt.count = 0;
1636 		tun->vnet_hdr_sz = sizeof(struct virtio_net_hdr);
1637 
1638 		tun->filter_attached = false;
1639 		tun->sndbuf = tfile->socket.sk->sk_sndbuf;
1640 
1641 		spin_lock_init(&tun->lock);
1642 
1643 		err = security_tun_dev_alloc_security(&tun->security);
1644 		if (err < 0)
1645 			goto err_free_dev;
1646 
1647 		tun_net_init(dev);
1648 		tun_flow_init(tun);
1649 
1650 		dev->hw_features = NETIF_F_SG | NETIF_F_FRAGLIST |
1651 				   TUN_USER_FEATURES | NETIF_F_HW_VLAN_CTAG_TX |
1652 				   NETIF_F_HW_VLAN_STAG_TX;
1653 		dev->features = dev->hw_features;
1654 		dev->vlan_features = dev->features &
1655 				     ~(NETIF_F_HW_VLAN_CTAG_TX |
1656 				       NETIF_F_HW_VLAN_STAG_TX);
1657 
1658 		INIT_LIST_HEAD(&tun->disabled);
1659 		err = tun_attach(tun, file, false);
1660 		if (err < 0)
1661 			goto err_free_flow;
1662 
1663 		err = register_netdevice(tun->dev);
1664 		if (err < 0)
1665 			goto err_detach;
1666 	}
1667 
1668 	netif_carrier_on(tun->dev);
1669 
1670 	tun_debug(KERN_INFO, tun, "tun_set_iff\n");
1671 
1672 	tun->flags = (tun->flags & ~TUN_FEATURES) |
1673 		(ifr->ifr_flags & TUN_FEATURES);
1674 
1675 	/* Make sure persistent devices do not get stuck in
1676 	 * xoff state.
1677 	 */
1678 	if (netif_running(tun->dev))
1679 		netif_tx_wake_all_queues(tun->dev);
1680 
1681 	strcpy(ifr->ifr_name, tun->dev->name);
1682 	return 0;
1683 
1684 err_detach:
1685 	tun_detach_all(dev);
1686 err_free_flow:
1687 	tun_flow_uninit(tun);
1688 	security_tun_dev_free_security(tun->security);
1689 err_free_dev:
1690 	free_netdev(dev);
1691 	return err;
1692 }
1693 
1694 static void tun_get_iff(struct net *net, struct tun_struct *tun,
1695 		       struct ifreq *ifr)
1696 {
1697 	tun_debug(KERN_INFO, tun, "tun_get_iff\n");
1698 
1699 	strcpy(ifr->ifr_name, tun->dev->name);
1700 
1701 	ifr->ifr_flags = tun_flags(tun);
1702 
1703 }
1704 
1705 /* This is like a cut-down ethtool ops, except done via tun fd so no
1706  * privs required. */
1707 static int set_offload(struct tun_struct *tun, unsigned long arg)
1708 {
1709 	netdev_features_t features = 0;
1710 
1711 	if (arg & TUN_F_CSUM) {
1712 		features |= NETIF_F_HW_CSUM;
1713 		arg &= ~TUN_F_CSUM;
1714 
1715 		if (arg & (TUN_F_TSO4|TUN_F_TSO6)) {
1716 			if (arg & TUN_F_TSO_ECN) {
1717 				features |= NETIF_F_TSO_ECN;
1718 				arg &= ~TUN_F_TSO_ECN;
1719 			}
1720 			if (arg & TUN_F_TSO4)
1721 				features |= NETIF_F_TSO;
1722 			if (arg & TUN_F_TSO6)
1723 				features |= NETIF_F_TSO6;
1724 			arg &= ~(TUN_F_TSO4|TUN_F_TSO6);
1725 		}
1726 
1727 		if (arg & TUN_F_UFO) {
1728 			features |= NETIF_F_UFO;
1729 			arg &= ~TUN_F_UFO;
1730 		}
1731 	}
1732 
1733 	/* This gives the user a way to test for new features in future by
1734 	 * trying to set them. */
1735 	if (arg)
1736 		return -EINVAL;
1737 
1738 	tun->set_features = features;
1739 	netdev_update_features(tun->dev);
1740 
1741 	return 0;
1742 }
1743 
1744 static void tun_detach_filter(struct tun_struct *tun, int n)
1745 {
1746 	int i;
1747 	struct tun_file *tfile;
1748 
1749 	for (i = 0; i < n; i++) {
1750 		tfile = rtnl_dereference(tun->tfiles[i]);
1751 		sk_detach_filter(tfile->socket.sk);
1752 	}
1753 
1754 	tun->filter_attached = false;
1755 }
1756 
1757 static int tun_attach_filter(struct tun_struct *tun)
1758 {
1759 	int i, ret = 0;
1760 	struct tun_file *tfile;
1761 
1762 	for (i = 0; i < tun->numqueues; i++) {
1763 		tfile = rtnl_dereference(tun->tfiles[i]);
1764 		ret = sk_attach_filter(&tun->fprog, tfile->socket.sk);
1765 		if (ret) {
1766 			tun_detach_filter(tun, i);
1767 			return ret;
1768 		}
1769 	}
1770 
1771 	tun->filter_attached = true;
1772 	return ret;
1773 }
1774 
1775 static void tun_set_sndbuf(struct tun_struct *tun)
1776 {
1777 	struct tun_file *tfile;
1778 	int i;
1779 
1780 	for (i = 0; i < tun->numqueues; i++) {
1781 		tfile = rtnl_dereference(tun->tfiles[i]);
1782 		tfile->socket.sk->sk_sndbuf = tun->sndbuf;
1783 	}
1784 }
1785 
1786 static int tun_set_queue(struct file *file, struct ifreq *ifr)
1787 {
1788 	struct tun_file *tfile = file->private_data;
1789 	struct tun_struct *tun;
1790 	int ret = 0;
1791 
1792 	rtnl_lock();
1793 
1794 	if (ifr->ifr_flags & IFF_ATTACH_QUEUE) {
1795 		tun = tfile->detached;
1796 		if (!tun) {
1797 			ret = -EINVAL;
1798 			goto unlock;
1799 		}
1800 		ret = security_tun_dev_attach_queue(tun->security);
1801 		if (ret < 0)
1802 			goto unlock;
1803 		ret = tun_attach(tun, file, false);
1804 	} else if (ifr->ifr_flags & IFF_DETACH_QUEUE) {
1805 		tun = rtnl_dereference(tfile->tun);
1806 		if (!tun || !(tun->flags & IFF_MULTI_QUEUE) || tfile->detached)
1807 			ret = -EINVAL;
1808 		else
1809 			__tun_detach(tfile, false);
1810 	} else
1811 		ret = -EINVAL;
1812 
1813 unlock:
1814 	rtnl_unlock();
1815 	return ret;
1816 }
1817 
1818 static long __tun_chr_ioctl(struct file *file, unsigned int cmd,
1819 			    unsigned long arg, int ifreq_len)
1820 {
1821 	struct tun_file *tfile = file->private_data;
1822 	struct tun_struct *tun;
1823 	void __user* argp = (void __user*)arg;
1824 	struct ifreq ifr;
1825 	kuid_t owner;
1826 	kgid_t group;
1827 	int sndbuf;
1828 	int vnet_hdr_sz;
1829 	unsigned int ifindex;
1830 	int le;
1831 	int ret;
1832 
1833 	if (cmd == TUNSETIFF || cmd == TUNSETQUEUE || _IOC_TYPE(cmd) == 0x89) {
1834 		if (copy_from_user(&ifr, argp, ifreq_len))
1835 			return -EFAULT;
1836 	} else {
1837 		memset(&ifr, 0, sizeof(ifr));
1838 	}
1839 	if (cmd == TUNGETFEATURES) {
1840 		/* Currently this just means: "what IFF flags are valid?".
1841 		 * This is needed because we never checked for invalid flags on
1842 		 * TUNSETIFF.
1843 		 */
1844 		return put_user(IFF_TUN | IFF_TAP | TUN_FEATURES,
1845 				(unsigned int __user*)argp);
1846 	} else if (cmd == TUNSETQUEUE)
1847 		return tun_set_queue(file, &ifr);
1848 
1849 	ret = 0;
1850 	rtnl_lock();
1851 
1852 	tun = __tun_get(tfile);
1853 	if (cmd == TUNSETIFF && !tun) {
1854 		ifr.ifr_name[IFNAMSIZ-1] = '\0';
1855 
1856 		ret = tun_set_iff(sock_net(&tfile->sk), file, &ifr);
1857 
1858 		if (ret)
1859 			goto unlock;
1860 
1861 		if (copy_to_user(argp, &ifr, ifreq_len))
1862 			ret = -EFAULT;
1863 		goto unlock;
1864 	}
1865 	if (cmd == TUNSETIFINDEX) {
1866 		ret = -EPERM;
1867 		if (tun)
1868 			goto unlock;
1869 
1870 		ret = -EFAULT;
1871 		if (copy_from_user(&ifindex, argp, sizeof(ifindex)))
1872 			goto unlock;
1873 
1874 		ret = 0;
1875 		tfile->ifindex = ifindex;
1876 		goto unlock;
1877 	}
1878 
1879 	ret = -EBADFD;
1880 	if (!tun)
1881 		goto unlock;
1882 
1883 	tun_debug(KERN_INFO, tun, "tun_chr_ioctl cmd %u\n", cmd);
1884 
1885 	ret = 0;
1886 	switch (cmd) {
1887 	case TUNGETIFF:
1888 		tun_get_iff(current->nsproxy->net_ns, tun, &ifr);
1889 
1890 		if (tfile->detached)
1891 			ifr.ifr_flags |= IFF_DETACH_QUEUE;
1892 		if (!tfile->socket.sk->sk_filter)
1893 			ifr.ifr_flags |= IFF_NOFILTER;
1894 
1895 		if (copy_to_user(argp, &ifr, ifreq_len))
1896 			ret = -EFAULT;
1897 		break;
1898 
1899 	case TUNSETNOCSUM:
1900 		/* Disable/Enable checksum */
1901 
1902 		/* [unimplemented] */
1903 		tun_debug(KERN_INFO, tun, "ignored: set checksum %s\n",
1904 			  arg ? "disabled" : "enabled");
1905 		break;
1906 
1907 	case TUNSETPERSIST:
1908 		/* Disable/Enable persist mode. Keep an extra reference to the
1909 		 * module to prevent the module being unprobed.
1910 		 */
1911 		if (arg && !(tun->flags & IFF_PERSIST)) {
1912 			tun->flags |= IFF_PERSIST;
1913 			__module_get(THIS_MODULE);
1914 		}
1915 		if (!arg && (tun->flags & IFF_PERSIST)) {
1916 			tun->flags &= ~IFF_PERSIST;
1917 			module_put(THIS_MODULE);
1918 		}
1919 
1920 		tun_debug(KERN_INFO, tun, "persist %s\n",
1921 			  arg ? "enabled" : "disabled");
1922 		break;
1923 
1924 	case TUNSETOWNER:
1925 		/* Set owner of the device */
1926 		owner = make_kuid(current_user_ns(), arg);
1927 		if (!uid_valid(owner)) {
1928 			ret = -EINVAL;
1929 			break;
1930 		}
1931 		tun->owner = owner;
1932 		tun_debug(KERN_INFO, tun, "owner set to %u\n",
1933 			  from_kuid(&init_user_ns, tun->owner));
1934 		break;
1935 
1936 	case TUNSETGROUP:
1937 		/* Set group of the device */
1938 		group = make_kgid(current_user_ns(), arg);
1939 		if (!gid_valid(group)) {
1940 			ret = -EINVAL;
1941 			break;
1942 		}
1943 		tun->group = group;
1944 		tun_debug(KERN_INFO, tun, "group set to %u\n",
1945 			  from_kgid(&init_user_ns, tun->group));
1946 		break;
1947 
1948 	case TUNSETLINK:
1949 		/* Only allow setting the type when the interface is down */
1950 		if (tun->dev->flags & IFF_UP) {
1951 			tun_debug(KERN_INFO, tun,
1952 				  "Linktype set failed because interface is up\n");
1953 			ret = -EBUSY;
1954 		} else {
1955 			tun->dev->type = (int) arg;
1956 			tun_debug(KERN_INFO, tun, "linktype set to %d\n",
1957 				  tun->dev->type);
1958 			ret = 0;
1959 		}
1960 		break;
1961 
1962 #ifdef TUN_DEBUG
1963 	case TUNSETDEBUG:
1964 		tun->debug = arg;
1965 		break;
1966 #endif
1967 	case TUNSETOFFLOAD:
1968 		ret = set_offload(tun, arg);
1969 		break;
1970 
1971 	case TUNSETTXFILTER:
1972 		/* Can be set only for TAPs */
1973 		ret = -EINVAL;
1974 		if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
1975 			break;
1976 		ret = update_filter(&tun->txflt, (void __user *)arg);
1977 		break;
1978 
1979 	case SIOCGIFHWADDR:
1980 		/* Get hw address */
1981 		memcpy(ifr.ifr_hwaddr.sa_data, tun->dev->dev_addr, ETH_ALEN);
1982 		ifr.ifr_hwaddr.sa_family = tun->dev->type;
1983 		if (copy_to_user(argp, &ifr, ifreq_len))
1984 			ret = -EFAULT;
1985 		break;
1986 
1987 	case SIOCSIFHWADDR:
1988 		/* Set hw address */
1989 		tun_debug(KERN_DEBUG, tun, "set hw address: %pM\n",
1990 			  ifr.ifr_hwaddr.sa_data);
1991 
1992 		ret = dev_set_mac_address(tun->dev, &ifr.ifr_hwaddr);
1993 		break;
1994 
1995 	case TUNGETSNDBUF:
1996 		sndbuf = tfile->socket.sk->sk_sndbuf;
1997 		if (copy_to_user(argp, &sndbuf, sizeof(sndbuf)))
1998 			ret = -EFAULT;
1999 		break;
2000 
2001 	case TUNSETSNDBUF:
2002 		if (copy_from_user(&sndbuf, argp, sizeof(sndbuf))) {
2003 			ret = -EFAULT;
2004 			break;
2005 		}
2006 
2007 		tun->sndbuf = sndbuf;
2008 		tun_set_sndbuf(tun);
2009 		break;
2010 
2011 	case TUNGETVNETHDRSZ:
2012 		vnet_hdr_sz = tun->vnet_hdr_sz;
2013 		if (copy_to_user(argp, &vnet_hdr_sz, sizeof(vnet_hdr_sz)))
2014 			ret = -EFAULT;
2015 		break;
2016 
2017 	case TUNSETVNETHDRSZ:
2018 		if (copy_from_user(&vnet_hdr_sz, argp, sizeof(vnet_hdr_sz))) {
2019 			ret = -EFAULT;
2020 			break;
2021 		}
2022 		if (vnet_hdr_sz < (int)sizeof(struct virtio_net_hdr)) {
2023 			ret = -EINVAL;
2024 			break;
2025 		}
2026 
2027 		tun->vnet_hdr_sz = vnet_hdr_sz;
2028 		break;
2029 
2030 	case TUNGETVNETLE:
2031 		le = !!(tun->flags & TUN_VNET_LE);
2032 		if (put_user(le, (int __user *)argp))
2033 			ret = -EFAULT;
2034 		break;
2035 
2036 	case TUNSETVNETLE:
2037 		if (get_user(le, (int __user *)argp)) {
2038 			ret = -EFAULT;
2039 			break;
2040 		}
2041 		if (le)
2042 			tun->flags |= TUN_VNET_LE;
2043 		else
2044 			tun->flags &= ~TUN_VNET_LE;
2045 		break;
2046 
2047 	case TUNATTACHFILTER:
2048 		/* Can be set only for TAPs */
2049 		ret = -EINVAL;
2050 		if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
2051 			break;
2052 		ret = -EFAULT;
2053 		if (copy_from_user(&tun->fprog, argp, sizeof(tun->fprog)))
2054 			break;
2055 
2056 		ret = tun_attach_filter(tun);
2057 		break;
2058 
2059 	case TUNDETACHFILTER:
2060 		/* Can be set only for TAPs */
2061 		ret = -EINVAL;
2062 		if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
2063 			break;
2064 		ret = 0;
2065 		tun_detach_filter(tun, tun->numqueues);
2066 		break;
2067 
2068 	case TUNGETFILTER:
2069 		ret = -EINVAL;
2070 		if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
2071 			break;
2072 		ret = -EFAULT;
2073 		if (copy_to_user(argp, &tun->fprog, sizeof(tun->fprog)))
2074 			break;
2075 		ret = 0;
2076 		break;
2077 
2078 	default:
2079 		ret = -EINVAL;
2080 		break;
2081 	}
2082 
2083 unlock:
2084 	rtnl_unlock();
2085 	if (tun)
2086 		tun_put(tun);
2087 	return ret;
2088 }
2089 
2090 static long tun_chr_ioctl(struct file *file,
2091 			  unsigned int cmd, unsigned long arg)
2092 {
2093 	return __tun_chr_ioctl(file, cmd, arg, sizeof (struct ifreq));
2094 }
2095 
2096 #ifdef CONFIG_COMPAT
2097 static long tun_chr_compat_ioctl(struct file *file,
2098 			 unsigned int cmd, unsigned long arg)
2099 {
2100 	switch (cmd) {
2101 	case TUNSETIFF:
2102 	case TUNGETIFF:
2103 	case TUNSETTXFILTER:
2104 	case TUNGETSNDBUF:
2105 	case TUNSETSNDBUF:
2106 	case SIOCGIFHWADDR:
2107 	case SIOCSIFHWADDR:
2108 		arg = (unsigned long)compat_ptr(arg);
2109 		break;
2110 	default:
2111 		arg = (compat_ulong_t)arg;
2112 		break;
2113 	}
2114 
2115 	/*
2116 	 * compat_ifreq is shorter than ifreq, so we must not access beyond
2117 	 * the end of that structure. All fields that are used in this
2118 	 * driver are compatible though, we don't need to convert the
2119 	 * contents.
2120 	 */
2121 	return __tun_chr_ioctl(file, cmd, arg, sizeof(struct compat_ifreq));
2122 }
2123 #endif /* CONFIG_COMPAT */
2124 
2125 static int tun_chr_fasync(int fd, struct file *file, int on)
2126 {
2127 	struct tun_file *tfile = file->private_data;
2128 	int ret;
2129 
2130 	if ((ret = fasync_helper(fd, file, on, &tfile->fasync)) < 0)
2131 		goto out;
2132 
2133 	if (on) {
2134 		__f_setown(file, task_pid(current), PIDTYPE_PID, 0);
2135 		tfile->flags |= TUN_FASYNC;
2136 	} else
2137 		tfile->flags &= ~TUN_FASYNC;
2138 	ret = 0;
2139 out:
2140 	return ret;
2141 }
2142 
2143 static int tun_chr_open(struct inode *inode, struct file * file)
2144 {
2145 	struct net *net = current->nsproxy->net_ns;
2146 	struct tun_file *tfile;
2147 
2148 	DBG1(KERN_INFO, "tunX: tun_chr_open\n");
2149 
2150 	tfile = (struct tun_file *)sk_alloc(net, AF_UNSPEC, GFP_KERNEL,
2151 					    &tun_proto, 0);
2152 	if (!tfile)
2153 		return -ENOMEM;
2154 	RCU_INIT_POINTER(tfile->tun, NULL);
2155 	tfile->flags = 0;
2156 	tfile->ifindex = 0;
2157 
2158 	init_waitqueue_head(&tfile->wq.wait);
2159 	RCU_INIT_POINTER(tfile->socket.wq, &tfile->wq);
2160 
2161 	tfile->socket.file = file;
2162 	tfile->socket.ops = &tun_socket_ops;
2163 
2164 	sock_init_data(&tfile->socket, &tfile->sk);
2165 
2166 	tfile->sk.sk_write_space = tun_sock_write_space;
2167 	tfile->sk.sk_sndbuf = INT_MAX;
2168 
2169 	file->private_data = tfile;
2170 	INIT_LIST_HEAD(&tfile->next);
2171 
2172 	sock_set_flag(&tfile->sk, SOCK_ZEROCOPY);
2173 
2174 	return 0;
2175 }
2176 
2177 static int tun_chr_close(struct inode *inode, struct file *file)
2178 {
2179 	struct tun_file *tfile = file->private_data;
2180 
2181 	tun_detach(tfile, true);
2182 
2183 	return 0;
2184 }
2185 
2186 #ifdef CONFIG_PROC_FS
2187 static void tun_chr_show_fdinfo(struct seq_file *m, struct file *f)
2188 {
2189 	struct tun_struct *tun;
2190 	struct ifreq ifr;
2191 
2192 	memset(&ifr, 0, sizeof(ifr));
2193 
2194 	rtnl_lock();
2195 	tun = tun_get(f);
2196 	if (tun)
2197 		tun_get_iff(current->nsproxy->net_ns, tun, &ifr);
2198 	rtnl_unlock();
2199 
2200 	if (tun)
2201 		tun_put(tun);
2202 
2203 	seq_printf(m, "iff:\t%s\n", ifr.ifr_name);
2204 }
2205 #endif
2206 
2207 static const struct file_operations tun_fops = {
2208 	.owner	= THIS_MODULE,
2209 	.llseek = no_llseek,
2210 	.read_iter  = tun_chr_read_iter,
2211 	.write_iter = tun_chr_write_iter,
2212 	.poll	= tun_chr_poll,
2213 	.unlocked_ioctl	= tun_chr_ioctl,
2214 #ifdef CONFIG_COMPAT
2215 	.compat_ioctl = tun_chr_compat_ioctl,
2216 #endif
2217 	.open	= tun_chr_open,
2218 	.release = tun_chr_close,
2219 	.fasync = tun_chr_fasync,
2220 #ifdef CONFIG_PROC_FS
2221 	.show_fdinfo = tun_chr_show_fdinfo,
2222 #endif
2223 };
2224 
2225 static struct miscdevice tun_miscdev = {
2226 	.minor = TUN_MINOR,
2227 	.name = "tun",
2228 	.nodename = "net/tun",
2229 	.fops = &tun_fops,
2230 };
2231 
2232 /* ethtool interface */
2233 
2234 static int tun_get_settings(struct net_device *dev, struct ethtool_cmd *cmd)
2235 {
2236 	cmd->supported		= 0;
2237 	cmd->advertising	= 0;
2238 	ethtool_cmd_speed_set(cmd, SPEED_10);
2239 	cmd->duplex		= DUPLEX_FULL;
2240 	cmd->port		= PORT_TP;
2241 	cmd->phy_address	= 0;
2242 	cmd->transceiver	= XCVR_INTERNAL;
2243 	cmd->autoneg		= AUTONEG_DISABLE;
2244 	cmd->maxtxpkt		= 0;
2245 	cmd->maxrxpkt		= 0;
2246 	return 0;
2247 }
2248 
2249 static void tun_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
2250 {
2251 	struct tun_struct *tun = netdev_priv(dev);
2252 
2253 	strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
2254 	strlcpy(info->version, DRV_VERSION, sizeof(info->version));
2255 
2256 	switch (tun->flags & TUN_TYPE_MASK) {
2257 	case IFF_TUN:
2258 		strlcpy(info->bus_info, "tun", sizeof(info->bus_info));
2259 		break;
2260 	case IFF_TAP:
2261 		strlcpy(info->bus_info, "tap", sizeof(info->bus_info));
2262 		break;
2263 	}
2264 }
2265 
2266 static u32 tun_get_msglevel(struct net_device *dev)
2267 {
2268 #ifdef TUN_DEBUG
2269 	struct tun_struct *tun = netdev_priv(dev);
2270 	return tun->debug;
2271 #else
2272 	return -EOPNOTSUPP;
2273 #endif
2274 }
2275 
2276 static void tun_set_msglevel(struct net_device *dev, u32 value)
2277 {
2278 #ifdef TUN_DEBUG
2279 	struct tun_struct *tun = netdev_priv(dev);
2280 	tun->debug = value;
2281 #endif
2282 }
2283 
2284 static const struct ethtool_ops tun_ethtool_ops = {
2285 	.get_settings	= tun_get_settings,
2286 	.get_drvinfo	= tun_get_drvinfo,
2287 	.get_msglevel	= tun_get_msglevel,
2288 	.set_msglevel	= tun_set_msglevel,
2289 	.get_link	= ethtool_op_get_link,
2290 	.get_ts_info	= ethtool_op_get_ts_info,
2291 };
2292 
2293 
2294 static int __init tun_init(void)
2295 {
2296 	int ret = 0;
2297 
2298 	pr_info("%s, %s\n", DRV_DESCRIPTION, DRV_VERSION);
2299 	pr_info("%s\n", DRV_COPYRIGHT);
2300 
2301 	ret = rtnl_link_register(&tun_link_ops);
2302 	if (ret) {
2303 		pr_err("Can't register link_ops\n");
2304 		goto err_linkops;
2305 	}
2306 
2307 	ret = misc_register(&tun_miscdev);
2308 	if (ret) {
2309 		pr_err("Can't register misc device %d\n", TUN_MINOR);
2310 		goto err_misc;
2311 	}
2312 	return  0;
2313 err_misc:
2314 	rtnl_link_unregister(&tun_link_ops);
2315 err_linkops:
2316 	return ret;
2317 }
2318 
2319 static void tun_cleanup(void)
2320 {
2321 	misc_deregister(&tun_miscdev);
2322 	rtnl_link_unregister(&tun_link_ops);
2323 }
2324 
2325 /* Get an underlying socket object from tun file.  Returns error unless file is
2326  * attached to a device.  The returned object works like a packet socket, it
2327  * can be used for sock_sendmsg/sock_recvmsg.  The caller is responsible for
2328  * holding a reference to the file for as long as the socket is in use. */
2329 struct socket *tun_get_socket(struct file *file)
2330 {
2331 	struct tun_file *tfile;
2332 	if (file->f_op != &tun_fops)
2333 		return ERR_PTR(-EINVAL);
2334 	tfile = file->private_data;
2335 	if (!tfile)
2336 		return ERR_PTR(-EBADFD);
2337 	return &tfile->socket;
2338 }
2339 EXPORT_SYMBOL_GPL(tun_get_socket);
2340 
2341 module_init(tun_init);
2342 module_exit(tun_cleanup);
2343 MODULE_DESCRIPTION(DRV_DESCRIPTION);
2344 MODULE_AUTHOR(DRV_COPYRIGHT);
2345 MODULE_LICENSE("GPL");
2346 MODULE_ALIAS_MISCDEV(TUN_MINOR);
2347 MODULE_ALIAS("devname:net/tun");
2348