xref: /linux/drivers/net/tun.c (revision c0e297dc61f8d4453e07afbea1fa8d0e67cd4a34)
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 #define TUN_VNET_BE     0x40000000
115 
116 #define TUN_FEATURES (IFF_NO_PI | IFF_ONE_QUEUE | IFF_VNET_HDR | \
117 		      IFF_MULTI_QUEUE)
118 #define GOODCOPY_LEN 128
119 
120 #define FLT_EXACT_COUNT 8
121 struct tap_filter {
122 	unsigned int    count;    /* Number of addrs. Zero means disabled */
123 	u32             mask[2];  /* Mask of the hashed addrs */
124 	unsigned char	addr[FLT_EXACT_COUNT][ETH_ALEN];
125 };
126 
127 /* MAX_TAP_QUEUES 256 is chosen to allow rx/tx queues to be equal
128  * to max number of VCPUs in guest. */
129 #define MAX_TAP_QUEUES 256
130 #define MAX_TAP_FLOWS  4096
131 
132 #define TUN_FLOW_EXPIRE (3 * HZ)
133 
134 /* A tun_file connects an open character device to a tuntap netdevice. It
135  * also contains all socket related structures (except sock_fprog and tap_filter)
136  * to serve as one transmit queue for tuntap device. The sock_fprog and
137  * tap_filter were kept in tun_struct since they were used for filtering for the
138  * netdevice not for a specific queue (at least I didn't see the requirement for
139  * this).
140  *
141  * RCU usage:
142  * The tun_file and tun_struct are loosely coupled, the pointer from one to the
143  * other can only be read while rcu_read_lock or rtnl_lock is held.
144  */
145 struct tun_file {
146 	struct sock sk;
147 	struct socket socket;
148 	struct socket_wq wq;
149 	struct tun_struct __rcu *tun;
150 	struct fasync_struct *fasync;
151 	/* only used for fasnyc */
152 	unsigned int flags;
153 	union {
154 		u16 queue_index;
155 		unsigned int ifindex;
156 	};
157 	struct list_head next;
158 	struct tun_struct *detached;
159 };
160 
161 struct tun_flow_entry {
162 	struct hlist_node hash_link;
163 	struct rcu_head rcu;
164 	struct tun_struct *tun;
165 
166 	u32 rxhash;
167 	u32 rps_rxhash;
168 	int queue_index;
169 	unsigned long updated;
170 };
171 
172 #define TUN_NUM_FLOW_ENTRIES 1024
173 
174 /* Since the socket were moved to tun_file, to preserve the behavior of persist
175  * device, socket filter, sndbuf and vnet header size were restore when the
176  * file were attached to a persist device.
177  */
178 struct tun_struct {
179 	struct tun_file __rcu	*tfiles[MAX_TAP_QUEUES];
180 	unsigned int            numqueues;
181 	unsigned int 		flags;
182 	kuid_t			owner;
183 	kgid_t			group;
184 
185 	struct net_device	*dev;
186 	netdev_features_t	set_features;
187 #define TUN_USER_FEATURES (NETIF_F_HW_CSUM|NETIF_F_TSO_ECN|NETIF_F_TSO| \
188 			  NETIF_F_TSO6|NETIF_F_UFO)
189 
190 	int			vnet_hdr_sz;
191 	int			sndbuf;
192 	struct tap_filter	txflt;
193 	struct sock_fprog	fprog;
194 	/* protected by rtnl lock */
195 	bool			filter_attached;
196 #ifdef TUN_DEBUG
197 	int debug;
198 #endif
199 	spinlock_t lock;
200 	struct hlist_head flows[TUN_NUM_FLOW_ENTRIES];
201 	struct timer_list flow_gc_timer;
202 	unsigned long ageing_time;
203 	unsigned int numdisabled;
204 	struct list_head disabled;
205 	void *security;
206 	u32 flow_count;
207 };
208 
209 #ifdef CONFIG_TUN_VNET_CROSS_LE
210 static inline bool tun_legacy_is_little_endian(struct tun_struct *tun)
211 {
212 	return tun->flags & TUN_VNET_BE ? false :
213 		virtio_legacy_is_little_endian();
214 }
215 
216 static long tun_get_vnet_be(struct tun_struct *tun, int __user *argp)
217 {
218 	int be = !!(tun->flags & TUN_VNET_BE);
219 
220 	if (put_user(be, argp))
221 		return -EFAULT;
222 
223 	return 0;
224 }
225 
226 static long tun_set_vnet_be(struct tun_struct *tun, int __user *argp)
227 {
228 	int be;
229 
230 	if (get_user(be, argp))
231 		return -EFAULT;
232 
233 	if (be)
234 		tun->flags |= TUN_VNET_BE;
235 	else
236 		tun->flags &= ~TUN_VNET_BE;
237 
238 	return 0;
239 }
240 #else
241 static inline bool tun_legacy_is_little_endian(struct tun_struct *tun)
242 {
243 	return virtio_legacy_is_little_endian();
244 }
245 
246 static long tun_get_vnet_be(struct tun_struct *tun, int __user *argp)
247 {
248 	return -EINVAL;
249 }
250 
251 static long tun_set_vnet_be(struct tun_struct *tun, int __user *argp)
252 {
253 	return -EINVAL;
254 }
255 #endif /* CONFIG_TUN_VNET_CROSS_LE */
256 
257 static inline bool tun_is_little_endian(struct tun_struct *tun)
258 {
259 	return tun->flags & TUN_VNET_LE ||
260 		tun_legacy_is_little_endian(tun);
261 }
262 
263 static inline u16 tun16_to_cpu(struct tun_struct *tun, __virtio16 val)
264 {
265 	return __virtio16_to_cpu(tun_is_little_endian(tun), val);
266 }
267 
268 static inline __virtio16 cpu_to_tun16(struct tun_struct *tun, u16 val)
269 {
270 	return __cpu_to_virtio16(tun_is_little_endian(tun), val);
271 }
272 
273 static inline u32 tun_hashfn(u32 rxhash)
274 {
275 	return rxhash & 0x3ff;
276 }
277 
278 static struct tun_flow_entry *tun_flow_find(struct hlist_head *head, u32 rxhash)
279 {
280 	struct tun_flow_entry *e;
281 
282 	hlist_for_each_entry_rcu(e, head, hash_link) {
283 		if (e->rxhash == rxhash)
284 			return e;
285 	}
286 	return NULL;
287 }
288 
289 static struct tun_flow_entry *tun_flow_create(struct tun_struct *tun,
290 					      struct hlist_head *head,
291 					      u32 rxhash, u16 queue_index)
292 {
293 	struct tun_flow_entry *e = kmalloc(sizeof(*e), GFP_ATOMIC);
294 
295 	if (e) {
296 		tun_debug(KERN_INFO, tun, "create flow: hash %u index %u\n",
297 			  rxhash, queue_index);
298 		e->updated = jiffies;
299 		e->rxhash = rxhash;
300 		e->rps_rxhash = 0;
301 		e->queue_index = queue_index;
302 		e->tun = tun;
303 		hlist_add_head_rcu(&e->hash_link, head);
304 		++tun->flow_count;
305 	}
306 	return e;
307 }
308 
309 static void tun_flow_delete(struct tun_struct *tun, struct tun_flow_entry *e)
310 {
311 	tun_debug(KERN_INFO, tun, "delete flow: hash %u index %u\n",
312 		  e->rxhash, e->queue_index);
313 	hlist_del_rcu(&e->hash_link);
314 	kfree_rcu(e, rcu);
315 	--tun->flow_count;
316 }
317 
318 static void tun_flow_flush(struct tun_struct *tun)
319 {
320 	int i;
321 
322 	spin_lock_bh(&tun->lock);
323 	for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
324 		struct tun_flow_entry *e;
325 		struct hlist_node *n;
326 
327 		hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link)
328 			tun_flow_delete(tun, e);
329 	}
330 	spin_unlock_bh(&tun->lock);
331 }
332 
333 static void tun_flow_delete_by_queue(struct tun_struct *tun, u16 queue_index)
334 {
335 	int i;
336 
337 	spin_lock_bh(&tun->lock);
338 	for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
339 		struct tun_flow_entry *e;
340 		struct hlist_node *n;
341 
342 		hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
343 			if (e->queue_index == queue_index)
344 				tun_flow_delete(tun, e);
345 		}
346 	}
347 	spin_unlock_bh(&tun->lock);
348 }
349 
350 static void tun_flow_cleanup(unsigned long data)
351 {
352 	struct tun_struct *tun = (struct tun_struct *)data;
353 	unsigned long delay = tun->ageing_time;
354 	unsigned long next_timer = jiffies + delay;
355 	unsigned long count = 0;
356 	int i;
357 
358 	tun_debug(KERN_INFO, tun, "tun_flow_cleanup\n");
359 
360 	spin_lock_bh(&tun->lock);
361 	for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
362 		struct tun_flow_entry *e;
363 		struct hlist_node *n;
364 
365 		hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
366 			unsigned long this_timer;
367 			count++;
368 			this_timer = e->updated + delay;
369 			if (time_before_eq(this_timer, jiffies))
370 				tun_flow_delete(tun, e);
371 			else if (time_before(this_timer, next_timer))
372 				next_timer = this_timer;
373 		}
374 	}
375 
376 	if (count)
377 		mod_timer(&tun->flow_gc_timer, round_jiffies_up(next_timer));
378 	spin_unlock_bh(&tun->lock);
379 }
380 
381 static void tun_flow_update(struct tun_struct *tun, u32 rxhash,
382 			    struct tun_file *tfile)
383 {
384 	struct hlist_head *head;
385 	struct tun_flow_entry *e;
386 	unsigned long delay = tun->ageing_time;
387 	u16 queue_index = tfile->queue_index;
388 
389 	if (!rxhash)
390 		return;
391 	else
392 		head = &tun->flows[tun_hashfn(rxhash)];
393 
394 	rcu_read_lock();
395 
396 	/* We may get a very small possibility of OOO during switching, not
397 	 * worth to optimize.*/
398 	if (tun->numqueues == 1 || tfile->detached)
399 		goto unlock;
400 
401 	e = tun_flow_find(head, rxhash);
402 	if (likely(e)) {
403 		/* TODO: keep queueing to old queue until it's empty? */
404 		e->queue_index = queue_index;
405 		e->updated = jiffies;
406 		sock_rps_record_flow_hash(e->rps_rxhash);
407 	} else {
408 		spin_lock_bh(&tun->lock);
409 		if (!tun_flow_find(head, rxhash) &&
410 		    tun->flow_count < MAX_TAP_FLOWS)
411 			tun_flow_create(tun, head, rxhash, queue_index);
412 
413 		if (!timer_pending(&tun->flow_gc_timer))
414 			mod_timer(&tun->flow_gc_timer,
415 				  round_jiffies_up(jiffies + delay));
416 		spin_unlock_bh(&tun->lock);
417 	}
418 
419 unlock:
420 	rcu_read_unlock();
421 }
422 
423 /**
424  * Save the hash received in the stack receive path and update the
425  * flow_hash table accordingly.
426  */
427 static inline void tun_flow_save_rps_rxhash(struct tun_flow_entry *e, u32 hash)
428 {
429 	if (unlikely(e->rps_rxhash != hash))
430 		e->rps_rxhash = hash;
431 }
432 
433 /* We try to identify a flow through its rxhash first. The reason that
434  * we do not check rxq no. is because some cards(e.g 82599), chooses
435  * the rxq based on the txq where the last packet of the flow comes. As
436  * the userspace application move between processors, we may get a
437  * different rxq no. here. If we could not get rxhash, then we would
438  * hope the rxq no. may help here.
439  */
440 static u16 tun_select_queue(struct net_device *dev, struct sk_buff *skb,
441 			    void *accel_priv, select_queue_fallback_t fallback)
442 {
443 	struct tun_struct *tun = netdev_priv(dev);
444 	struct tun_flow_entry *e;
445 	u32 txq = 0;
446 	u32 numqueues = 0;
447 
448 	rcu_read_lock();
449 	numqueues = ACCESS_ONCE(tun->numqueues);
450 
451 	txq = skb_get_hash(skb);
452 	if (txq) {
453 		e = tun_flow_find(&tun->flows[tun_hashfn(txq)], txq);
454 		if (e) {
455 			tun_flow_save_rps_rxhash(e, txq);
456 			txq = e->queue_index;
457 		} else
458 			/* use multiply and shift instead of expensive divide */
459 			txq = ((u64)txq * numqueues) >> 32;
460 	} else if (likely(skb_rx_queue_recorded(skb))) {
461 		txq = skb_get_rx_queue(skb);
462 		while (unlikely(txq >= numqueues))
463 			txq -= numqueues;
464 	}
465 
466 	rcu_read_unlock();
467 	return txq;
468 }
469 
470 static inline bool tun_not_capable(struct tun_struct *tun)
471 {
472 	const struct cred *cred = current_cred();
473 	struct net *net = dev_net(tun->dev);
474 
475 	return ((uid_valid(tun->owner) && !uid_eq(cred->euid, tun->owner)) ||
476 		  (gid_valid(tun->group) && !in_egroup_p(tun->group))) &&
477 		!ns_capable(net->user_ns, CAP_NET_ADMIN);
478 }
479 
480 static void tun_set_real_num_queues(struct tun_struct *tun)
481 {
482 	netif_set_real_num_tx_queues(tun->dev, tun->numqueues);
483 	netif_set_real_num_rx_queues(tun->dev, tun->numqueues);
484 }
485 
486 static void tun_disable_queue(struct tun_struct *tun, struct tun_file *tfile)
487 {
488 	tfile->detached = tun;
489 	list_add_tail(&tfile->next, &tun->disabled);
490 	++tun->numdisabled;
491 }
492 
493 static struct tun_struct *tun_enable_queue(struct tun_file *tfile)
494 {
495 	struct tun_struct *tun = tfile->detached;
496 
497 	tfile->detached = NULL;
498 	list_del_init(&tfile->next);
499 	--tun->numdisabled;
500 	return tun;
501 }
502 
503 static void tun_queue_purge(struct tun_file *tfile)
504 {
505 	skb_queue_purge(&tfile->sk.sk_receive_queue);
506 	skb_queue_purge(&tfile->sk.sk_error_queue);
507 }
508 
509 static void __tun_detach(struct tun_file *tfile, bool clean)
510 {
511 	struct tun_file *ntfile;
512 	struct tun_struct *tun;
513 
514 	tun = rtnl_dereference(tfile->tun);
515 
516 	if (tun && !tfile->detached) {
517 		u16 index = tfile->queue_index;
518 		BUG_ON(index >= tun->numqueues);
519 
520 		rcu_assign_pointer(tun->tfiles[index],
521 				   tun->tfiles[tun->numqueues - 1]);
522 		ntfile = rtnl_dereference(tun->tfiles[index]);
523 		ntfile->queue_index = index;
524 
525 		--tun->numqueues;
526 		if (clean) {
527 			RCU_INIT_POINTER(tfile->tun, NULL);
528 			sock_put(&tfile->sk);
529 		} else
530 			tun_disable_queue(tun, tfile);
531 
532 		synchronize_net();
533 		tun_flow_delete_by_queue(tun, tun->numqueues + 1);
534 		/* Drop read queue */
535 		tun_queue_purge(tfile);
536 		tun_set_real_num_queues(tun);
537 	} else if (tfile->detached && clean) {
538 		tun = tun_enable_queue(tfile);
539 		sock_put(&tfile->sk);
540 	}
541 
542 	if (clean) {
543 		if (tun && tun->numqueues == 0 && tun->numdisabled == 0) {
544 			netif_carrier_off(tun->dev);
545 
546 			if (!(tun->flags & IFF_PERSIST) &&
547 			    tun->dev->reg_state == NETREG_REGISTERED)
548 				unregister_netdevice(tun->dev);
549 		}
550 		sock_put(&tfile->sk);
551 	}
552 }
553 
554 static void tun_detach(struct tun_file *tfile, bool clean)
555 {
556 	rtnl_lock();
557 	__tun_detach(tfile, clean);
558 	rtnl_unlock();
559 }
560 
561 static void tun_detach_all(struct net_device *dev)
562 {
563 	struct tun_struct *tun = netdev_priv(dev);
564 	struct tun_file *tfile, *tmp;
565 	int i, n = tun->numqueues;
566 
567 	for (i = 0; i < n; i++) {
568 		tfile = rtnl_dereference(tun->tfiles[i]);
569 		BUG_ON(!tfile);
570 		tfile->socket.sk->sk_data_ready(tfile->socket.sk);
571 		RCU_INIT_POINTER(tfile->tun, NULL);
572 		--tun->numqueues;
573 	}
574 	list_for_each_entry(tfile, &tun->disabled, next) {
575 		tfile->socket.sk->sk_data_ready(tfile->socket.sk);
576 		RCU_INIT_POINTER(tfile->tun, NULL);
577 	}
578 	BUG_ON(tun->numqueues != 0);
579 
580 	synchronize_net();
581 	for (i = 0; i < n; i++) {
582 		tfile = rtnl_dereference(tun->tfiles[i]);
583 		/* Drop read queue */
584 		tun_queue_purge(tfile);
585 		sock_put(&tfile->sk);
586 	}
587 	list_for_each_entry_safe(tfile, tmp, &tun->disabled, next) {
588 		tun_enable_queue(tfile);
589 		tun_queue_purge(tfile);
590 		sock_put(&tfile->sk);
591 	}
592 	BUG_ON(tun->numdisabled != 0);
593 
594 	if (tun->flags & IFF_PERSIST)
595 		module_put(THIS_MODULE);
596 }
597 
598 static int tun_attach(struct tun_struct *tun, struct file *file, bool skip_filter)
599 {
600 	struct tun_file *tfile = file->private_data;
601 	int err;
602 
603 	err = security_tun_dev_attach(tfile->socket.sk, tun->security);
604 	if (err < 0)
605 		goto out;
606 
607 	err = -EINVAL;
608 	if (rtnl_dereference(tfile->tun) && !tfile->detached)
609 		goto out;
610 
611 	err = -EBUSY;
612 	if (!(tun->flags & IFF_MULTI_QUEUE) && tun->numqueues == 1)
613 		goto out;
614 
615 	err = -E2BIG;
616 	if (!tfile->detached &&
617 	    tun->numqueues + tun->numdisabled == MAX_TAP_QUEUES)
618 		goto out;
619 
620 	err = 0;
621 
622 	/* Re-attach the filter to persist device */
623 	if (!skip_filter && (tun->filter_attached == true)) {
624 		err = sk_attach_filter(&tun->fprog, tfile->socket.sk);
625 		if (!err)
626 			goto out;
627 	}
628 	tfile->queue_index = tun->numqueues;
629 	rcu_assign_pointer(tfile->tun, tun);
630 	rcu_assign_pointer(tun->tfiles[tun->numqueues], tfile);
631 	tun->numqueues++;
632 
633 	if (tfile->detached)
634 		tun_enable_queue(tfile);
635 	else
636 		sock_hold(&tfile->sk);
637 
638 	tun_set_real_num_queues(tun);
639 
640 	/* device is allowed to go away first, so no need to hold extra
641 	 * refcnt.
642 	 */
643 
644 out:
645 	return err;
646 }
647 
648 static struct tun_struct *__tun_get(struct tun_file *tfile)
649 {
650 	struct tun_struct *tun;
651 
652 	rcu_read_lock();
653 	tun = rcu_dereference(tfile->tun);
654 	if (tun)
655 		dev_hold(tun->dev);
656 	rcu_read_unlock();
657 
658 	return tun;
659 }
660 
661 static struct tun_struct *tun_get(struct file *file)
662 {
663 	return __tun_get(file->private_data);
664 }
665 
666 static void tun_put(struct tun_struct *tun)
667 {
668 	dev_put(tun->dev);
669 }
670 
671 /* TAP filtering */
672 static void addr_hash_set(u32 *mask, const u8 *addr)
673 {
674 	int n = ether_crc(ETH_ALEN, addr) >> 26;
675 	mask[n >> 5] |= (1 << (n & 31));
676 }
677 
678 static unsigned int addr_hash_test(const u32 *mask, const u8 *addr)
679 {
680 	int n = ether_crc(ETH_ALEN, addr) >> 26;
681 	return mask[n >> 5] & (1 << (n & 31));
682 }
683 
684 static int update_filter(struct tap_filter *filter, void __user *arg)
685 {
686 	struct { u8 u[ETH_ALEN]; } *addr;
687 	struct tun_filter uf;
688 	int err, alen, n, nexact;
689 
690 	if (copy_from_user(&uf, arg, sizeof(uf)))
691 		return -EFAULT;
692 
693 	if (!uf.count) {
694 		/* Disabled */
695 		filter->count = 0;
696 		return 0;
697 	}
698 
699 	alen = ETH_ALEN * uf.count;
700 	addr = kmalloc(alen, GFP_KERNEL);
701 	if (!addr)
702 		return -ENOMEM;
703 
704 	if (copy_from_user(addr, arg + sizeof(uf), alen)) {
705 		err = -EFAULT;
706 		goto done;
707 	}
708 
709 	/* The filter is updated without holding any locks. Which is
710 	 * perfectly safe. We disable it first and in the worst
711 	 * case we'll accept a few undesired packets. */
712 	filter->count = 0;
713 	wmb();
714 
715 	/* Use first set of addresses as an exact filter */
716 	for (n = 0; n < uf.count && n < FLT_EXACT_COUNT; n++)
717 		memcpy(filter->addr[n], addr[n].u, ETH_ALEN);
718 
719 	nexact = n;
720 
721 	/* Remaining multicast addresses are hashed,
722 	 * unicast will leave the filter disabled. */
723 	memset(filter->mask, 0, sizeof(filter->mask));
724 	for (; n < uf.count; n++) {
725 		if (!is_multicast_ether_addr(addr[n].u)) {
726 			err = 0; /* no filter */
727 			goto done;
728 		}
729 		addr_hash_set(filter->mask, addr[n].u);
730 	}
731 
732 	/* For ALLMULTI just set the mask to all ones.
733 	 * This overrides the mask populated above. */
734 	if ((uf.flags & TUN_FLT_ALLMULTI))
735 		memset(filter->mask, ~0, sizeof(filter->mask));
736 
737 	/* Now enable the filter */
738 	wmb();
739 	filter->count = nexact;
740 
741 	/* Return the number of exact filters */
742 	err = nexact;
743 
744 done:
745 	kfree(addr);
746 	return err;
747 }
748 
749 /* Returns: 0 - drop, !=0 - accept */
750 static int run_filter(struct tap_filter *filter, const struct sk_buff *skb)
751 {
752 	/* Cannot use eth_hdr(skb) here because skb_mac_hdr() is incorrect
753 	 * at this point. */
754 	struct ethhdr *eh = (struct ethhdr *) skb->data;
755 	int i;
756 
757 	/* Exact match */
758 	for (i = 0; i < filter->count; i++)
759 		if (ether_addr_equal(eh->h_dest, filter->addr[i]))
760 			return 1;
761 
762 	/* Inexact match (multicast only) */
763 	if (is_multicast_ether_addr(eh->h_dest))
764 		return addr_hash_test(filter->mask, eh->h_dest);
765 
766 	return 0;
767 }
768 
769 /*
770  * Checks whether the packet is accepted or not.
771  * Returns: 0 - drop, !=0 - accept
772  */
773 static int check_filter(struct tap_filter *filter, const struct sk_buff *skb)
774 {
775 	if (!filter->count)
776 		return 1;
777 
778 	return run_filter(filter, skb);
779 }
780 
781 /* Network device part of the driver */
782 
783 static const struct ethtool_ops tun_ethtool_ops;
784 
785 /* Net device detach from fd. */
786 static void tun_net_uninit(struct net_device *dev)
787 {
788 	tun_detach_all(dev);
789 }
790 
791 /* Net device open. */
792 static int tun_net_open(struct net_device *dev)
793 {
794 	netif_tx_start_all_queues(dev);
795 	return 0;
796 }
797 
798 /* Net device close. */
799 static int tun_net_close(struct net_device *dev)
800 {
801 	netif_tx_stop_all_queues(dev);
802 	return 0;
803 }
804 
805 /* Net device start xmit */
806 static netdev_tx_t tun_net_xmit(struct sk_buff *skb, struct net_device *dev)
807 {
808 	struct tun_struct *tun = netdev_priv(dev);
809 	int txq = skb->queue_mapping;
810 	struct tun_file *tfile;
811 	u32 numqueues = 0;
812 
813 	rcu_read_lock();
814 	tfile = rcu_dereference(tun->tfiles[txq]);
815 	numqueues = ACCESS_ONCE(tun->numqueues);
816 
817 	/* Drop packet if interface is not attached */
818 	if (txq >= numqueues)
819 		goto drop;
820 
821 	if (numqueues == 1) {
822 		/* Select queue was not called for the skbuff, so we extract the
823 		 * RPS hash and save it into the flow_table here.
824 		 */
825 		__u32 rxhash;
826 
827 		rxhash = skb_get_hash(skb);
828 		if (rxhash) {
829 			struct tun_flow_entry *e;
830 			e = tun_flow_find(&tun->flows[tun_hashfn(rxhash)],
831 					rxhash);
832 			if (e)
833 				tun_flow_save_rps_rxhash(e, rxhash);
834 		}
835 	}
836 
837 	tun_debug(KERN_INFO, tun, "tun_net_xmit %d\n", skb->len);
838 
839 	BUG_ON(!tfile);
840 
841 	/* Drop if the filter does not like it.
842 	 * This is a noop if the filter is disabled.
843 	 * Filter can be enabled only for the TAP devices. */
844 	if (!check_filter(&tun->txflt, skb))
845 		goto drop;
846 
847 	if (tfile->socket.sk->sk_filter &&
848 	    sk_filter(tfile->socket.sk, skb))
849 		goto drop;
850 
851 	/* Limit the number of packets queued by dividing txq length with the
852 	 * number of queues.
853 	 */
854 	if (skb_queue_len(&tfile->socket.sk->sk_receive_queue) * numqueues
855 			  >= dev->tx_queue_len)
856 		goto drop;
857 
858 	if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC)))
859 		goto drop;
860 
861 	if (skb->sk) {
862 		sock_tx_timestamp(skb->sk, &skb_shinfo(skb)->tx_flags);
863 		sw_tx_timestamp(skb);
864 	}
865 
866 	/* Orphan the skb - required as we might hang on to it
867 	 * for indefinite time.
868 	 */
869 	skb_orphan(skb);
870 
871 	nf_reset(skb);
872 
873 	/* Enqueue packet */
874 	skb_queue_tail(&tfile->socket.sk->sk_receive_queue, skb);
875 
876 	/* Notify and wake up reader process */
877 	if (tfile->flags & TUN_FASYNC)
878 		kill_fasync(&tfile->fasync, SIGIO, POLL_IN);
879 	tfile->socket.sk->sk_data_ready(tfile->socket.sk);
880 
881 	rcu_read_unlock();
882 	return NETDEV_TX_OK;
883 
884 drop:
885 	dev->stats.tx_dropped++;
886 	skb_tx_error(skb);
887 	kfree_skb(skb);
888 	rcu_read_unlock();
889 	return NET_XMIT_DROP;
890 }
891 
892 static void tun_net_mclist(struct net_device *dev)
893 {
894 	/*
895 	 * This callback is supposed to deal with mc filter in
896 	 * _rx_ path and has nothing to do with the _tx_ path.
897 	 * In rx path we always accept everything userspace gives us.
898 	 */
899 }
900 
901 #define MIN_MTU 68
902 #define MAX_MTU 65535
903 
904 static int
905 tun_net_change_mtu(struct net_device *dev, int new_mtu)
906 {
907 	if (new_mtu < MIN_MTU || new_mtu + dev->hard_header_len > MAX_MTU)
908 		return -EINVAL;
909 	dev->mtu = new_mtu;
910 	return 0;
911 }
912 
913 static netdev_features_t tun_net_fix_features(struct net_device *dev,
914 	netdev_features_t features)
915 {
916 	struct tun_struct *tun = netdev_priv(dev);
917 
918 	return (features & tun->set_features) | (features & ~TUN_USER_FEATURES);
919 }
920 #ifdef CONFIG_NET_POLL_CONTROLLER
921 static void tun_poll_controller(struct net_device *dev)
922 {
923 	/*
924 	 * Tun only receives frames when:
925 	 * 1) the char device endpoint gets data from user space
926 	 * 2) the tun socket gets a sendmsg call from user space
927 	 * Since both of those are synchronous operations, we are guaranteed
928 	 * never to have pending data when we poll for it
929 	 * so there is nothing to do here but return.
930 	 * We need this though so netpoll recognizes us as an interface that
931 	 * supports polling, which enables bridge devices in virt setups to
932 	 * still use netconsole
933 	 */
934 	return;
935 }
936 #endif
937 static const struct net_device_ops tun_netdev_ops = {
938 	.ndo_uninit		= tun_net_uninit,
939 	.ndo_open		= tun_net_open,
940 	.ndo_stop		= tun_net_close,
941 	.ndo_start_xmit		= tun_net_xmit,
942 	.ndo_change_mtu		= tun_net_change_mtu,
943 	.ndo_fix_features	= tun_net_fix_features,
944 	.ndo_select_queue	= tun_select_queue,
945 #ifdef CONFIG_NET_POLL_CONTROLLER
946 	.ndo_poll_controller	= tun_poll_controller,
947 #endif
948 };
949 
950 static const struct net_device_ops tap_netdev_ops = {
951 	.ndo_uninit		= tun_net_uninit,
952 	.ndo_open		= tun_net_open,
953 	.ndo_stop		= tun_net_close,
954 	.ndo_start_xmit		= tun_net_xmit,
955 	.ndo_change_mtu		= tun_net_change_mtu,
956 	.ndo_fix_features	= tun_net_fix_features,
957 	.ndo_set_rx_mode	= tun_net_mclist,
958 	.ndo_set_mac_address	= eth_mac_addr,
959 	.ndo_validate_addr	= eth_validate_addr,
960 	.ndo_select_queue	= tun_select_queue,
961 #ifdef CONFIG_NET_POLL_CONTROLLER
962 	.ndo_poll_controller	= tun_poll_controller,
963 #endif
964 };
965 
966 static void tun_flow_init(struct tun_struct *tun)
967 {
968 	int i;
969 
970 	for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++)
971 		INIT_HLIST_HEAD(&tun->flows[i]);
972 
973 	tun->ageing_time = TUN_FLOW_EXPIRE;
974 	setup_timer(&tun->flow_gc_timer, tun_flow_cleanup, (unsigned long)tun);
975 	mod_timer(&tun->flow_gc_timer,
976 		  round_jiffies_up(jiffies + tun->ageing_time));
977 }
978 
979 static void tun_flow_uninit(struct tun_struct *tun)
980 {
981 	del_timer_sync(&tun->flow_gc_timer);
982 	tun_flow_flush(tun);
983 }
984 
985 /* Initialize net device. */
986 static void tun_net_init(struct net_device *dev)
987 {
988 	struct tun_struct *tun = netdev_priv(dev);
989 
990 	switch (tun->flags & TUN_TYPE_MASK) {
991 	case IFF_TUN:
992 		dev->netdev_ops = &tun_netdev_ops;
993 
994 		/* Point-to-Point TUN Device */
995 		dev->hard_header_len = 0;
996 		dev->addr_len = 0;
997 		dev->mtu = 1500;
998 
999 		/* Zero header length */
1000 		dev->type = ARPHRD_NONE;
1001 		dev->flags = IFF_POINTOPOINT | IFF_NOARP | IFF_MULTICAST;
1002 		dev->tx_queue_len = TUN_READQ_SIZE;  /* We prefer our own queue length */
1003 		break;
1004 
1005 	case IFF_TAP:
1006 		dev->netdev_ops = &tap_netdev_ops;
1007 		/* Ethernet TAP Device */
1008 		ether_setup(dev);
1009 		dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1010 		dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
1011 
1012 		eth_hw_addr_random(dev);
1013 
1014 		dev->tx_queue_len = TUN_READQ_SIZE;  /* We prefer our own queue length */
1015 		break;
1016 	}
1017 }
1018 
1019 /* Character device part */
1020 
1021 /* Poll */
1022 static unsigned int tun_chr_poll(struct file *file, poll_table *wait)
1023 {
1024 	struct tun_file *tfile = file->private_data;
1025 	struct tun_struct *tun = __tun_get(tfile);
1026 	struct sock *sk;
1027 	unsigned int mask = 0;
1028 
1029 	if (!tun)
1030 		return POLLERR;
1031 
1032 	sk = tfile->socket.sk;
1033 
1034 	tun_debug(KERN_INFO, tun, "tun_chr_poll\n");
1035 
1036 	poll_wait(file, sk_sleep(sk), wait);
1037 
1038 	if (!skb_queue_empty(&sk->sk_receive_queue))
1039 		mask |= POLLIN | POLLRDNORM;
1040 
1041 	if (sock_writeable(sk) ||
1042 	    (!test_and_set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags) &&
1043 	     sock_writeable(sk)))
1044 		mask |= POLLOUT | POLLWRNORM;
1045 
1046 	if (tun->dev->reg_state != NETREG_REGISTERED)
1047 		mask = POLLERR;
1048 
1049 	tun_put(tun);
1050 	return mask;
1051 }
1052 
1053 /* prepad is the amount to reserve at front.  len is length after that.
1054  * linear is a hint as to how much to copy (usually headers). */
1055 static struct sk_buff *tun_alloc_skb(struct tun_file *tfile,
1056 				     size_t prepad, size_t len,
1057 				     size_t linear, int noblock)
1058 {
1059 	struct sock *sk = tfile->socket.sk;
1060 	struct sk_buff *skb;
1061 	int err;
1062 
1063 	/* Under a page?  Don't bother with paged skb. */
1064 	if (prepad + len < PAGE_SIZE || !linear)
1065 		linear = len;
1066 
1067 	skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
1068 				   &err, 0);
1069 	if (!skb)
1070 		return ERR_PTR(err);
1071 
1072 	skb_reserve(skb, prepad);
1073 	skb_put(skb, linear);
1074 	skb->data_len = len - linear;
1075 	skb->len += len - linear;
1076 
1077 	return skb;
1078 }
1079 
1080 /* Get packet from user space buffer */
1081 static ssize_t tun_get_user(struct tun_struct *tun, struct tun_file *tfile,
1082 			    void *msg_control, struct iov_iter *from,
1083 			    int noblock)
1084 {
1085 	struct tun_pi pi = { 0, cpu_to_be16(ETH_P_IP) };
1086 	struct sk_buff *skb;
1087 	size_t total_len = iov_iter_count(from);
1088 	size_t len = total_len, align = NET_SKB_PAD, linear;
1089 	struct virtio_net_hdr gso = { 0 };
1090 	int good_linear;
1091 	int copylen;
1092 	bool zerocopy = false;
1093 	int err;
1094 	u32 rxhash;
1095 	ssize_t n;
1096 
1097 	if (!(tun->flags & IFF_NO_PI)) {
1098 		if (len < sizeof(pi))
1099 			return -EINVAL;
1100 		len -= sizeof(pi);
1101 
1102 		n = copy_from_iter(&pi, sizeof(pi), from);
1103 		if (n != sizeof(pi))
1104 			return -EFAULT;
1105 	}
1106 
1107 	if (tun->flags & IFF_VNET_HDR) {
1108 		if (len < tun->vnet_hdr_sz)
1109 			return -EINVAL;
1110 		len -= tun->vnet_hdr_sz;
1111 
1112 		n = copy_from_iter(&gso, sizeof(gso), from);
1113 		if (n != sizeof(gso))
1114 			return -EFAULT;
1115 
1116 		if ((gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
1117 		    tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2 > tun16_to_cpu(tun, gso.hdr_len))
1118 			gso.hdr_len = cpu_to_tun16(tun, tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2);
1119 
1120 		if (tun16_to_cpu(tun, gso.hdr_len) > len)
1121 			return -EINVAL;
1122 		iov_iter_advance(from, tun->vnet_hdr_sz - sizeof(gso));
1123 	}
1124 
1125 	if ((tun->flags & TUN_TYPE_MASK) == IFF_TAP) {
1126 		align += NET_IP_ALIGN;
1127 		if (unlikely(len < ETH_HLEN ||
1128 			     (gso.hdr_len && tun16_to_cpu(tun, gso.hdr_len) < ETH_HLEN)))
1129 			return -EINVAL;
1130 	}
1131 
1132 	good_linear = SKB_MAX_HEAD(align);
1133 
1134 	if (msg_control) {
1135 		struct iov_iter i = *from;
1136 
1137 		/* There are 256 bytes to be copied in skb, so there is
1138 		 * enough room for skb expand head in case it is used.
1139 		 * The rest of the buffer is mapped from userspace.
1140 		 */
1141 		copylen = gso.hdr_len ? tun16_to_cpu(tun, gso.hdr_len) : GOODCOPY_LEN;
1142 		if (copylen > good_linear)
1143 			copylen = good_linear;
1144 		linear = copylen;
1145 		iov_iter_advance(&i, copylen);
1146 		if (iov_iter_npages(&i, INT_MAX) <= MAX_SKB_FRAGS)
1147 			zerocopy = true;
1148 	}
1149 
1150 	if (!zerocopy) {
1151 		copylen = len;
1152 		if (tun16_to_cpu(tun, gso.hdr_len) > good_linear)
1153 			linear = good_linear;
1154 		else
1155 			linear = tun16_to_cpu(tun, gso.hdr_len);
1156 	}
1157 
1158 	skb = tun_alloc_skb(tfile, align, copylen, linear, noblock);
1159 	if (IS_ERR(skb)) {
1160 		if (PTR_ERR(skb) != -EAGAIN)
1161 			tun->dev->stats.rx_dropped++;
1162 		return PTR_ERR(skb);
1163 	}
1164 
1165 	if (zerocopy)
1166 		err = zerocopy_sg_from_iter(skb, from);
1167 	else {
1168 		err = skb_copy_datagram_from_iter(skb, 0, from, len);
1169 		if (!err && msg_control) {
1170 			struct ubuf_info *uarg = msg_control;
1171 			uarg->callback(uarg, false);
1172 		}
1173 	}
1174 
1175 	if (err) {
1176 		tun->dev->stats.rx_dropped++;
1177 		kfree_skb(skb);
1178 		return -EFAULT;
1179 	}
1180 
1181 	if (gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
1182 		if (!skb_partial_csum_set(skb, tun16_to_cpu(tun, gso.csum_start),
1183 					  tun16_to_cpu(tun, gso.csum_offset))) {
1184 			tun->dev->stats.rx_frame_errors++;
1185 			kfree_skb(skb);
1186 			return -EINVAL;
1187 		}
1188 	}
1189 
1190 	switch (tun->flags & TUN_TYPE_MASK) {
1191 	case IFF_TUN:
1192 		if (tun->flags & IFF_NO_PI) {
1193 			switch (skb->data[0] & 0xf0) {
1194 			case 0x40:
1195 				pi.proto = htons(ETH_P_IP);
1196 				break;
1197 			case 0x60:
1198 				pi.proto = htons(ETH_P_IPV6);
1199 				break;
1200 			default:
1201 				tun->dev->stats.rx_dropped++;
1202 				kfree_skb(skb);
1203 				return -EINVAL;
1204 			}
1205 		}
1206 
1207 		skb_reset_mac_header(skb);
1208 		skb->protocol = pi.proto;
1209 		skb->dev = tun->dev;
1210 		break;
1211 	case IFF_TAP:
1212 		skb->protocol = eth_type_trans(skb, tun->dev);
1213 		break;
1214 	}
1215 
1216 	if (gso.gso_type != VIRTIO_NET_HDR_GSO_NONE) {
1217 		pr_debug("GSO!\n");
1218 		switch (gso.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
1219 		case VIRTIO_NET_HDR_GSO_TCPV4:
1220 			skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
1221 			break;
1222 		case VIRTIO_NET_HDR_GSO_TCPV6:
1223 			skb_shinfo(skb)->gso_type = SKB_GSO_TCPV6;
1224 			break;
1225 		case VIRTIO_NET_HDR_GSO_UDP:
1226 			skb_shinfo(skb)->gso_type = SKB_GSO_UDP;
1227 			break;
1228 		default:
1229 			tun->dev->stats.rx_frame_errors++;
1230 			kfree_skb(skb);
1231 			return -EINVAL;
1232 		}
1233 
1234 		if (gso.gso_type & VIRTIO_NET_HDR_GSO_ECN)
1235 			skb_shinfo(skb)->gso_type |= SKB_GSO_TCP_ECN;
1236 
1237 		skb_shinfo(skb)->gso_size = tun16_to_cpu(tun, gso.gso_size);
1238 		if (skb_shinfo(skb)->gso_size == 0) {
1239 			tun->dev->stats.rx_frame_errors++;
1240 			kfree_skb(skb);
1241 			return -EINVAL;
1242 		}
1243 
1244 		/* Header must be checked, and gso_segs computed. */
1245 		skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
1246 		skb_shinfo(skb)->gso_segs = 0;
1247 	}
1248 
1249 	/* copy skb_ubuf_info for callback when skb has no error */
1250 	if (zerocopy) {
1251 		skb_shinfo(skb)->destructor_arg = msg_control;
1252 		skb_shinfo(skb)->tx_flags |= SKBTX_DEV_ZEROCOPY;
1253 		skb_shinfo(skb)->tx_flags |= SKBTX_SHARED_FRAG;
1254 	}
1255 
1256 	skb_reset_network_header(skb);
1257 	skb_probe_transport_header(skb, 0);
1258 
1259 	rxhash = skb_get_hash(skb);
1260 	netif_rx_ni(skb);
1261 
1262 	tun->dev->stats.rx_packets++;
1263 	tun->dev->stats.rx_bytes += len;
1264 
1265 	tun_flow_update(tun, rxhash, tfile);
1266 	return total_len;
1267 }
1268 
1269 static ssize_t tun_chr_write_iter(struct kiocb *iocb, struct iov_iter *from)
1270 {
1271 	struct file *file = iocb->ki_filp;
1272 	struct tun_struct *tun = tun_get(file);
1273 	struct tun_file *tfile = file->private_data;
1274 	ssize_t result;
1275 
1276 	if (!tun)
1277 		return -EBADFD;
1278 
1279 	result = tun_get_user(tun, tfile, NULL, from, file->f_flags & O_NONBLOCK);
1280 
1281 	tun_put(tun);
1282 	return result;
1283 }
1284 
1285 /* Put packet to the user space buffer */
1286 static ssize_t tun_put_user(struct tun_struct *tun,
1287 			    struct tun_file *tfile,
1288 			    struct sk_buff *skb,
1289 			    struct iov_iter *iter)
1290 {
1291 	struct tun_pi pi = { 0, skb->protocol };
1292 	ssize_t total;
1293 	int vlan_offset = 0;
1294 	int vlan_hlen = 0;
1295 	int vnet_hdr_sz = 0;
1296 
1297 	if (skb_vlan_tag_present(skb))
1298 		vlan_hlen = VLAN_HLEN;
1299 
1300 	if (tun->flags & IFF_VNET_HDR)
1301 		vnet_hdr_sz = tun->vnet_hdr_sz;
1302 
1303 	total = skb->len + vlan_hlen + vnet_hdr_sz;
1304 
1305 	if (!(tun->flags & IFF_NO_PI)) {
1306 		if (iov_iter_count(iter) < sizeof(pi))
1307 			return -EINVAL;
1308 
1309 		total += sizeof(pi);
1310 		if (iov_iter_count(iter) < total) {
1311 			/* Packet will be striped */
1312 			pi.flags |= TUN_PKT_STRIP;
1313 		}
1314 
1315 		if (copy_to_iter(&pi, sizeof(pi), iter) != sizeof(pi))
1316 			return -EFAULT;
1317 	}
1318 
1319 	if (vnet_hdr_sz) {
1320 		struct virtio_net_hdr gso = { 0 }; /* no info leak */
1321 		if (iov_iter_count(iter) < vnet_hdr_sz)
1322 			return -EINVAL;
1323 
1324 		if (skb_is_gso(skb)) {
1325 			struct skb_shared_info *sinfo = skb_shinfo(skb);
1326 
1327 			/* This is a hint as to how much should be linear. */
1328 			gso.hdr_len = cpu_to_tun16(tun, skb_headlen(skb));
1329 			gso.gso_size = cpu_to_tun16(tun, sinfo->gso_size);
1330 			if (sinfo->gso_type & SKB_GSO_TCPV4)
1331 				gso.gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
1332 			else if (sinfo->gso_type & SKB_GSO_TCPV6)
1333 				gso.gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
1334 			else if (sinfo->gso_type & SKB_GSO_UDP)
1335 				gso.gso_type = VIRTIO_NET_HDR_GSO_UDP;
1336 			else {
1337 				pr_err("unexpected GSO type: "
1338 				       "0x%x, gso_size %d, hdr_len %d\n",
1339 				       sinfo->gso_type, tun16_to_cpu(tun, gso.gso_size),
1340 				       tun16_to_cpu(tun, gso.hdr_len));
1341 				print_hex_dump(KERN_ERR, "tun: ",
1342 					       DUMP_PREFIX_NONE,
1343 					       16, 1, skb->head,
1344 					       min((int)tun16_to_cpu(tun, gso.hdr_len), 64), true);
1345 				WARN_ON_ONCE(1);
1346 				return -EINVAL;
1347 			}
1348 			if (sinfo->gso_type & SKB_GSO_TCP_ECN)
1349 				gso.gso_type |= VIRTIO_NET_HDR_GSO_ECN;
1350 		} else
1351 			gso.gso_type = VIRTIO_NET_HDR_GSO_NONE;
1352 
1353 		if (skb->ip_summed == CHECKSUM_PARTIAL) {
1354 			gso.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
1355 			gso.csum_start = cpu_to_tun16(tun, skb_checksum_start_offset(skb) +
1356 						      vlan_hlen);
1357 			gso.csum_offset = cpu_to_tun16(tun, skb->csum_offset);
1358 		} else if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
1359 			gso.flags = VIRTIO_NET_HDR_F_DATA_VALID;
1360 		} /* else everything is zero */
1361 
1362 		if (copy_to_iter(&gso, sizeof(gso), iter) != sizeof(gso))
1363 			return -EFAULT;
1364 
1365 		iov_iter_advance(iter, vnet_hdr_sz - sizeof(gso));
1366 	}
1367 
1368 	if (vlan_hlen) {
1369 		int ret;
1370 		struct {
1371 			__be16 h_vlan_proto;
1372 			__be16 h_vlan_TCI;
1373 		} veth;
1374 
1375 		veth.h_vlan_proto = skb->vlan_proto;
1376 		veth.h_vlan_TCI = htons(skb_vlan_tag_get(skb));
1377 
1378 		vlan_offset = offsetof(struct vlan_ethhdr, h_vlan_proto);
1379 
1380 		ret = skb_copy_datagram_iter(skb, 0, iter, vlan_offset);
1381 		if (ret || !iov_iter_count(iter))
1382 			goto done;
1383 
1384 		ret = copy_to_iter(&veth, sizeof(veth), iter);
1385 		if (ret != sizeof(veth) || !iov_iter_count(iter))
1386 			goto done;
1387 	}
1388 
1389 	skb_copy_datagram_iter(skb, vlan_offset, iter, skb->len - vlan_offset);
1390 
1391 done:
1392 	tun->dev->stats.tx_packets++;
1393 	tun->dev->stats.tx_bytes += skb->len + vlan_hlen;
1394 
1395 	return total;
1396 }
1397 
1398 static ssize_t tun_do_read(struct tun_struct *tun, struct tun_file *tfile,
1399 			   struct iov_iter *to,
1400 			   int noblock)
1401 {
1402 	struct sk_buff *skb;
1403 	ssize_t ret;
1404 	int peeked, err, off = 0;
1405 
1406 	tun_debug(KERN_INFO, tun, "tun_do_read\n");
1407 
1408 	if (!iov_iter_count(to))
1409 		return 0;
1410 
1411 	if (tun->dev->reg_state != NETREG_REGISTERED)
1412 		return -EIO;
1413 
1414 	/* Read frames from queue */
1415 	skb = __skb_recv_datagram(tfile->socket.sk, noblock ? MSG_DONTWAIT : 0,
1416 				  &peeked, &off, &err);
1417 	if (!skb)
1418 		return err;
1419 
1420 	ret = tun_put_user(tun, tfile, skb, to);
1421 	if (unlikely(ret < 0))
1422 		kfree_skb(skb);
1423 	else
1424 		consume_skb(skb);
1425 
1426 	return ret;
1427 }
1428 
1429 static ssize_t tun_chr_read_iter(struct kiocb *iocb, struct iov_iter *to)
1430 {
1431 	struct file *file = iocb->ki_filp;
1432 	struct tun_file *tfile = file->private_data;
1433 	struct tun_struct *tun = __tun_get(tfile);
1434 	ssize_t len = iov_iter_count(to), ret;
1435 
1436 	if (!tun)
1437 		return -EBADFD;
1438 	ret = tun_do_read(tun, tfile, to, file->f_flags & O_NONBLOCK);
1439 	ret = min_t(ssize_t, ret, len);
1440 	if (ret > 0)
1441 		iocb->ki_pos = ret;
1442 	tun_put(tun);
1443 	return ret;
1444 }
1445 
1446 static void tun_free_netdev(struct net_device *dev)
1447 {
1448 	struct tun_struct *tun = netdev_priv(dev);
1449 
1450 	BUG_ON(!(list_empty(&tun->disabled)));
1451 	tun_flow_uninit(tun);
1452 	security_tun_dev_free_security(tun->security);
1453 	free_netdev(dev);
1454 }
1455 
1456 static void tun_setup(struct net_device *dev)
1457 {
1458 	struct tun_struct *tun = netdev_priv(dev);
1459 
1460 	tun->owner = INVALID_UID;
1461 	tun->group = INVALID_GID;
1462 
1463 	dev->ethtool_ops = &tun_ethtool_ops;
1464 	dev->destructor = tun_free_netdev;
1465 }
1466 
1467 /* Trivial set of netlink ops to allow deleting tun or tap
1468  * device with netlink.
1469  */
1470 static int tun_validate(struct nlattr *tb[], struct nlattr *data[])
1471 {
1472 	return -EINVAL;
1473 }
1474 
1475 static struct rtnl_link_ops tun_link_ops __read_mostly = {
1476 	.kind		= DRV_NAME,
1477 	.priv_size	= sizeof(struct tun_struct),
1478 	.setup		= tun_setup,
1479 	.validate	= tun_validate,
1480 };
1481 
1482 static void tun_sock_write_space(struct sock *sk)
1483 {
1484 	struct tun_file *tfile;
1485 	wait_queue_head_t *wqueue;
1486 
1487 	if (!sock_writeable(sk))
1488 		return;
1489 
1490 	if (!test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags))
1491 		return;
1492 
1493 	wqueue = sk_sleep(sk);
1494 	if (wqueue && waitqueue_active(wqueue))
1495 		wake_up_interruptible_sync_poll(wqueue, POLLOUT |
1496 						POLLWRNORM | POLLWRBAND);
1497 
1498 	tfile = container_of(sk, struct tun_file, sk);
1499 	kill_fasync(&tfile->fasync, SIGIO, POLL_OUT);
1500 }
1501 
1502 static int tun_sendmsg(struct socket *sock, struct msghdr *m, size_t total_len)
1503 {
1504 	int ret;
1505 	struct tun_file *tfile = container_of(sock, struct tun_file, socket);
1506 	struct tun_struct *tun = __tun_get(tfile);
1507 
1508 	if (!tun)
1509 		return -EBADFD;
1510 
1511 	ret = tun_get_user(tun, tfile, m->msg_control, &m->msg_iter,
1512 			   m->msg_flags & MSG_DONTWAIT);
1513 	tun_put(tun);
1514 	return ret;
1515 }
1516 
1517 static int tun_recvmsg(struct socket *sock, struct msghdr *m, size_t total_len,
1518 		       int flags)
1519 {
1520 	struct tun_file *tfile = container_of(sock, struct tun_file, socket);
1521 	struct tun_struct *tun = __tun_get(tfile);
1522 	int ret;
1523 
1524 	if (!tun)
1525 		return -EBADFD;
1526 
1527 	if (flags & ~(MSG_DONTWAIT|MSG_TRUNC|MSG_ERRQUEUE)) {
1528 		ret = -EINVAL;
1529 		goto out;
1530 	}
1531 	if (flags & MSG_ERRQUEUE) {
1532 		ret = sock_recv_errqueue(sock->sk, m, total_len,
1533 					 SOL_PACKET, TUN_TX_TIMESTAMP);
1534 		goto out;
1535 	}
1536 	ret = tun_do_read(tun, tfile, &m->msg_iter, flags & MSG_DONTWAIT);
1537 	if (ret > (ssize_t)total_len) {
1538 		m->msg_flags |= MSG_TRUNC;
1539 		ret = flags & MSG_TRUNC ? ret : total_len;
1540 	}
1541 out:
1542 	tun_put(tun);
1543 	return ret;
1544 }
1545 
1546 /* Ops structure to mimic raw sockets with tun */
1547 static const struct proto_ops tun_socket_ops = {
1548 	.sendmsg = tun_sendmsg,
1549 	.recvmsg = tun_recvmsg,
1550 };
1551 
1552 static struct proto tun_proto = {
1553 	.name		= "tun",
1554 	.owner		= THIS_MODULE,
1555 	.obj_size	= sizeof(struct tun_file),
1556 };
1557 
1558 static int tun_flags(struct tun_struct *tun)
1559 {
1560 	return tun->flags & (TUN_FEATURES | IFF_PERSIST | IFF_TUN | IFF_TAP);
1561 }
1562 
1563 static ssize_t tun_show_flags(struct device *dev, struct device_attribute *attr,
1564 			      char *buf)
1565 {
1566 	struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1567 	return sprintf(buf, "0x%x\n", tun_flags(tun));
1568 }
1569 
1570 static ssize_t tun_show_owner(struct device *dev, struct device_attribute *attr,
1571 			      char *buf)
1572 {
1573 	struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1574 	return uid_valid(tun->owner)?
1575 		sprintf(buf, "%u\n",
1576 			from_kuid_munged(current_user_ns(), tun->owner)):
1577 		sprintf(buf, "-1\n");
1578 }
1579 
1580 static ssize_t tun_show_group(struct device *dev, struct device_attribute *attr,
1581 			      char *buf)
1582 {
1583 	struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1584 	return gid_valid(tun->group) ?
1585 		sprintf(buf, "%u\n",
1586 			from_kgid_munged(current_user_ns(), tun->group)):
1587 		sprintf(buf, "-1\n");
1588 }
1589 
1590 static DEVICE_ATTR(tun_flags, 0444, tun_show_flags, NULL);
1591 static DEVICE_ATTR(owner, 0444, tun_show_owner, NULL);
1592 static DEVICE_ATTR(group, 0444, tun_show_group, NULL);
1593 
1594 static struct attribute *tun_dev_attrs[] = {
1595 	&dev_attr_tun_flags.attr,
1596 	&dev_attr_owner.attr,
1597 	&dev_attr_group.attr,
1598 	NULL
1599 };
1600 
1601 static const struct attribute_group tun_attr_group = {
1602 	.attrs = tun_dev_attrs
1603 };
1604 
1605 static int tun_set_iff(struct net *net, struct file *file, struct ifreq *ifr)
1606 {
1607 	struct tun_struct *tun;
1608 	struct tun_file *tfile = file->private_data;
1609 	struct net_device *dev;
1610 	int err;
1611 
1612 	if (tfile->detached)
1613 		return -EINVAL;
1614 
1615 	dev = __dev_get_by_name(net, ifr->ifr_name);
1616 	if (dev) {
1617 		if (ifr->ifr_flags & IFF_TUN_EXCL)
1618 			return -EBUSY;
1619 		if ((ifr->ifr_flags & IFF_TUN) && dev->netdev_ops == &tun_netdev_ops)
1620 			tun = netdev_priv(dev);
1621 		else if ((ifr->ifr_flags & IFF_TAP) && dev->netdev_ops == &tap_netdev_ops)
1622 			tun = netdev_priv(dev);
1623 		else
1624 			return -EINVAL;
1625 
1626 		if (!!(ifr->ifr_flags & IFF_MULTI_QUEUE) !=
1627 		    !!(tun->flags & IFF_MULTI_QUEUE))
1628 			return -EINVAL;
1629 
1630 		if (tun_not_capable(tun))
1631 			return -EPERM;
1632 		err = security_tun_dev_open(tun->security);
1633 		if (err < 0)
1634 			return err;
1635 
1636 		err = tun_attach(tun, file, ifr->ifr_flags & IFF_NOFILTER);
1637 		if (err < 0)
1638 			return err;
1639 
1640 		if (tun->flags & IFF_MULTI_QUEUE &&
1641 		    (tun->numqueues + tun->numdisabled > 1)) {
1642 			/* One or more queue has already been attached, no need
1643 			 * to initialize the device again.
1644 			 */
1645 			return 0;
1646 		}
1647 	}
1648 	else {
1649 		char *name;
1650 		unsigned long flags = 0;
1651 		int queues = ifr->ifr_flags & IFF_MULTI_QUEUE ?
1652 			     MAX_TAP_QUEUES : 1;
1653 
1654 		if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
1655 			return -EPERM;
1656 		err = security_tun_dev_create();
1657 		if (err < 0)
1658 			return err;
1659 
1660 		/* Set dev type */
1661 		if (ifr->ifr_flags & IFF_TUN) {
1662 			/* TUN device */
1663 			flags |= IFF_TUN;
1664 			name = "tun%d";
1665 		} else if (ifr->ifr_flags & IFF_TAP) {
1666 			/* TAP device */
1667 			flags |= IFF_TAP;
1668 			name = "tap%d";
1669 		} else
1670 			return -EINVAL;
1671 
1672 		if (*ifr->ifr_name)
1673 			name = ifr->ifr_name;
1674 
1675 		dev = alloc_netdev_mqs(sizeof(struct tun_struct), name,
1676 				       NET_NAME_UNKNOWN, tun_setup, queues,
1677 				       queues);
1678 
1679 		if (!dev)
1680 			return -ENOMEM;
1681 
1682 		dev_net_set(dev, net);
1683 		dev->rtnl_link_ops = &tun_link_ops;
1684 		dev->ifindex = tfile->ifindex;
1685 		dev->sysfs_groups[0] = &tun_attr_group;
1686 
1687 		tun = netdev_priv(dev);
1688 		tun->dev = dev;
1689 		tun->flags = flags;
1690 		tun->txflt.count = 0;
1691 		tun->vnet_hdr_sz = sizeof(struct virtio_net_hdr);
1692 
1693 		tun->filter_attached = false;
1694 		tun->sndbuf = tfile->socket.sk->sk_sndbuf;
1695 
1696 		spin_lock_init(&tun->lock);
1697 
1698 		err = security_tun_dev_alloc_security(&tun->security);
1699 		if (err < 0)
1700 			goto err_free_dev;
1701 
1702 		tun_net_init(dev);
1703 		tun_flow_init(tun);
1704 
1705 		dev->hw_features = NETIF_F_SG | NETIF_F_FRAGLIST |
1706 				   TUN_USER_FEATURES | NETIF_F_HW_VLAN_CTAG_TX |
1707 				   NETIF_F_HW_VLAN_STAG_TX;
1708 		dev->features = dev->hw_features;
1709 		dev->vlan_features = dev->features &
1710 				     ~(NETIF_F_HW_VLAN_CTAG_TX |
1711 				       NETIF_F_HW_VLAN_STAG_TX);
1712 
1713 		INIT_LIST_HEAD(&tun->disabled);
1714 		err = tun_attach(tun, file, false);
1715 		if (err < 0)
1716 			goto err_free_flow;
1717 
1718 		err = register_netdevice(tun->dev);
1719 		if (err < 0)
1720 			goto err_detach;
1721 	}
1722 
1723 	netif_carrier_on(tun->dev);
1724 
1725 	tun_debug(KERN_INFO, tun, "tun_set_iff\n");
1726 
1727 	tun->flags = (tun->flags & ~TUN_FEATURES) |
1728 		(ifr->ifr_flags & TUN_FEATURES);
1729 
1730 	/* Make sure persistent devices do not get stuck in
1731 	 * xoff state.
1732 	 */
1733 	if (netif_running(tun->dev))
1734 		netif_tx_wake_all_queues(tun->dev);
1735 
1736 	strcpy(ifr->ifr_name, tun->dev->name);
1737 	return 0;
1738 
1739 err_detach:
1740 	tun_detach_all(dev);
1741 err_free_flow:
1742 	tun_flow_uninit(tun);
1743 	security_tun_dev_free_security(tun->security);
1744 err_free_dev:
1745 	free_netdev(dev);
1746 	return err;
1747 }
1748 
1749 static void tun_get_iff(struct net *net, struct tun_struct *tun,
1750 		       struct ifreq *ifr)
1751 {
1752 	tun_debug(KERN_INFO, tun, "tun_get_iff\n");
1753 
1754 	strcpy(ifr->ifr_name, tun->dev->name);
1755 
1756 	ifr->ifr_flags = tun_flags(tun);
1757 
1758 }
1759 
1760 /* This is like a cut-down ethtool ops, except done via tun fd so no
1761  * privs required. */
1762 static int set_offload(struct tun_struct *tun, unsigned long arg)
1763 {
1764 	netdev_features_t features = 0;
1765 
1766 	if (arg & TUN_F_CSUM) {
1767 		features |= NETIF_F_HW_CSUM;
1768 		arg &= ~TUN_F_CSUM;
1769 
1770 		if (arg & (TUN_F_TSO4|TUN_F_TSO6)) {
1771 			if (arg & TUN_F_TSO_ECN) {
1772 				features |= NETIF_F_TSO_ECN;
1773 				arg &= ~TUN_F_TSO_ECN;
1774 			}
1775 			if (arg & TUN_F_TSO4)
1776 				features |= NETIF_F_TSO;
1777 			if (arg & TUN_F_TSO6)
1778 				features |= NETIF_F_TSO6;
1779 			arg &= ~(TUN_F_TSO4|TUN_F_TSO6);
1780 		}
1781 
1782 		if (arg & TUN_F_UFO) {
1783 			features |= NETIF_F_UFO;
1784 			arg &= ~TUN_F_UFO;
1785 		}
1786 	}
1787 
1788 	/* This gives the user a way to test for new features in future by
1789 	 * trying to set them. */
1790 	if (arg)
1791 		return -EINVAL;
1792 
1793 	tun->set_features = features;
1794 	netdev_update_features(tun->dev);
1795 
1796 	return 0;
1797 }
1798 
1799 static void tun_detach_filter(struct tun_struct *tun, int n)
1800 {
1801 	int i;
1802 	struct tun_file *tfile;
1803 
1804 	for (i = 0; i < n; i++) {
1805 		tfile = rtnl_dereference(tun->tfiles[i]);
1806 		sk_detach_filter(tfile->socket.sk);
1807 	}
1808 
1809 	tun->filter_attached = false;
1810 }
1811 
1812 static int tun_attach_filter(struct tun_struct *tun)
1813 {
1814 	int i, ret = 0;
1815 	struct tun_file *tfile;
1816 
1817 	for (i = 0; i < tun->numqueues; i++) {
1818 		tfile = rtnl_dereference(tun->tfiles[i]);
1819 		ret = sk_attach_filter(&tun->fprog, tfile->socket.sk);
1820 		if (ret) {
1821 			tun_detach_filter(tun, i);
1822 			return ret;
1823 		}
1824 	}
1825 
1826 	tun->filter_attached = true;
1827 	return ret;
1828 }
1829 
1830 static void tun_set_sndbuf(struct tun_struct *tun)
1831 {
1832 	struct tun_file *tfile;
1833 	int i;
1834 
1835 	for (i = 0; i < tun->numqueues; i++) {
1836 		tfile = rtnl_dereference(tun->tfiles[i]);
1837 		tfile->socket.sk->sk_sndbuf = tun->sndbuf;
1838 	}
1839 }
1840 
1841 static int tun_set_queue(struct file *file, struct ifreq *ifr)
1842 {
1843 	struct tun_file *tfile = file->private_data;
1844 	struct tun_struct *tun;
1845 	int ret = 0;
1846 
1847 	rtnl_lock();
1848 
1849 	if (ifr->ifr_flags & IFF_ATTACH_QUEUE) {
1850 		tun = tfile->detached;
1851 		if (!tun) {
1852 			ret = -EINVAL;
1853 			goto unlock;
1854 		}
1855 		ret = security_tun_dev_attach_queue(tun->security);
1856 		if (ret < 0)
1857 			goto unlock;
1858 		ret = tun_attach(tun, file, false);
1859 	} else if (ifr->ifr_flags & IFF_DETACH_QUEUE) {
1860 		tun = rtnl_dereference(tfile->tun);
1861 		if (!tun || !(tun->flags & IFF_MULTI_QUEUE) || tfile->detached)
1862 			ret = -EINVAL;
1863 		else
1864 			__tun_detach(tfile, false);
1865 	} else
1866 		ret = -EINVAL;
1867 
1868 unlock:
1869 	rtnl_unlock();
1870 	return ret;
1871 }
1872 
1873 static long __tun_chr_ioctl(struct file *file, unsigned int cmd,
1874 			    unsigned long arg, int ifreq_len)
1875 {
1876 	struct tun_file *tfile = file->private_data;
1877 	struct tun_struct *tun;
1878 	void __user* argp = (void __user*)arg;
1879 	struct ifreq ifr;
1880 	kuid_t owner;
1881 	kgid_t group;
1882 	int sndbuf;
1883 	int vnet_hdr_sz;
1884 	unsigned int ifindex;
1885 	int le;
1886 	int ret;
1887 
1888 	if (cmd == TUNSETIFF || cmd == TUNSETQUEUE || _IOC_TYPE(cmd) == 0x89) {
1889 		if (copy_from_user(&ifr, argp, ifreq_len))
1890 			return -EFAULT;
1891 	} else {
1892 		memset(&ifr, 0, sizeof(ifr));
1893 	}
1894 	if (cmd == TUNGETFEATURES) {
1895 		/* Currently this just means: "what IFF flags are valid?".
1896 		 * This is needed because we never checked for invalid flags on
1897 		 * TUNSETIFF.
1898 		 */
1899 		return put_user(IFF_TUN | IFF_TAP | TUN_FEATURES,
1900 				(unsigned int __user*)argp);
1901 	} else if (cmd == TUNSETQUEUE)
1902 		return tun_set_queue(file, &ifr);
1903 
1904 	ret = 0;
1905 	rtnl_lock();
1906 
1907 	tun = __tun_get(tfile);
1908 	if (cmd == TUNSETIFF && !tun) {
1909 		ifr.ifr_name[IFNAMSIZ-1] = '\0';
1910 
1911 		ret = tun_set_iff(sock_net(&tfile->sk), file, &ifr);
1912 
1913 		if (ret)
1914 			goto unlock;
1915 
1916 		if (copy_to_user(argp, &ifr, ifreq_len))
1917 			ret = -EFAULT;
1918 		goto unlock;
1919 	}
1920 	if (cmd == TUNSETIFINDEX) {
1921 		ret = -EPERM;
1922 		if (tun)
1923 			goto unlock;
1924 
1925 		ret = -EFAULT;
1926 		if (copy_from_user(&ifindex, argp, sizeof(ifindex)))
1927 			goto unlock;
1928 
1929 		ret = 0;
1930 		tfile->ifindex = ifindex;
1931 		goto unlock;
1932 	}
1933 
1934 	ret = -EBADFD;
1935 	if (!tun)
1936 		goto unlock;
1937 
1938 	tun_debug(KERN_INFO, tun, "tun_chr_ioctl cmd %u\n", cmd);
1939 
1940 	ret = 0;
1941 	switch (cmd) {
1942 	case TUNGETIFF:
1943 		tun_get_iff(current->nsproxy->net_ns, tun, &ifr);
1944 
1945 		if (tfile->detached)
1946 			ifr.ifr_flags |= IFF_DETACH_QUEUE;
1947 		if (!tfile->socket.sk->sk_filter)
1948 			ifr.ifr_flags |= IFF_NOFILTER;
1949 
1950 		if (copy_to_user(argp, &ifr, ifreq_len))
1951 			ret = -EFAULT;
1952 		break;
1953 
1954 	case TUNSETNOCSUM:
1955 		/* Disable/Enable checksum */
1956 
1957 		/* [unimplemented] */
1958 		tun_debug(KERN_INFO, tun, "ignored: set checksum %s\n",
1959 			  arg ? "disabled" : "enabled");
1960 		break;
1961 
1962 	case TUNSETPERSIST:
1963 		/* Disable/Enable persist mode. Keep an extra reference to the
1964 		 * module to prevent the module being unprobed.
1965 		 */
1966 		if (arg && !(tun->flags & IFF_PERSIST)) {
1967 			tun->flags |= IFF_PERSIST;
1968 			__module_get(THIS_MODULE);
1969 		}
1970 		if (!arg && (tun->flags & IFF_PERSIST)) {
1971 			tun->flags &= ~IFF_PERSIST;
1972 			module_put(THIS_MODULE);
1973 		}
1974 
1975 		tun_debug(KERN_INFO, tun, "persist %s\n",
1976 			  arg ? "enabled" : "disabled");
1977 		break;
1978 
1979 	case TUNSETOWNER:
1980 		/* Set owner of the device */
1981 		owner = make_kuid(current_user_ns(), arg);
1982 		if (!uid_valid(owner)) {
1983 			ret = -EINVAL;
1984 			break;
1985 		}
1986 		tun->owner = owner;
1987 		tun_debug(KERN_INFO, tun, "owner set to %u\n",
1988 			  from_kuid(&init_user_ns, tun->owner));
1989 		break;
1990 
1991 	case TUNSETGROUP:
1992 		/* Set group of the device */
1993 		group = make_kgid(current_user_ns(), arg);
1994 		if (!gid_valid(group)) {
1995 			ret = -EINVAL;
1996 			break;
1997 		}
1998 		tun->group = group;
1999 		tun_debug(KERN_INFO, tun, "group set to %u\n",
2000 			  from_kgid(&init_user_ns, tun->group));
2001 		break;
2002 
2003 	case TUNSETLINK:
2004 		/* Only allow setting the type when the interface is down */
2005 		if (tun->dev->flags & IFF_UP) {
2006 			tun_debug(KERN_INFO, tun,
2007 				  "Linktype set failed because interface is up\n");
2008 			ret = -EBUSY;
2009 		} else {
2010 			tun->dev->type = (int) arg;
2011 			tun_debug(KERN_INFO, tun, "linktype set to %d\n",
2012 				  tun->dev->type);
2013 			ret = 0;
2014 		}
2015 		break;
2016 
2017 #ifdef TUN_DEBUG
2018 	case TUNSETDEBUG:
2019 		tun->debug = arg;
2020 		break;
2021 #endif
2022 	case TUNSETOFFLOAD:
2023 		ret = set_offload(tun, arg);
2024 		break;
2025 
2026 	case TUNSETTXFILTER:
2027 		/* Can be set only for TAPs */
2028 		ret = -EINVAL;
2029 		if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
2030 			break;
2031 		ret = update_filter(&tun->txflt, (void __user *)arg);
2032 		break;
2033 
2034 	case SIOCGIFHWADDR:
2035 		/* Get hw address */
2036 		memcpy(ifr.ifr_hwaddr.sa_data, tun->dev->dev_addr, ETH_ALEN);
2037 		ifr.ifr_hwaddr.sa_family = tun->dev->type;
2038 		if (copy_to_user(argp, &ifr, ifreq_len))
2039 			ret = -EFAULT;
2040 		break;
2041 
2042 	case SIOCSIFHWADDR:
2043 		/* Set hw address */
2044 		tun_debug(KERN_DEBUG, tun, "set hw address: %pM\n",
2045 			  ifr.ifr_hwaddr.sa_data);
2046 
2047 		ret = dev_set_mac_address(tun->dev, &ifr.ifr_hwaddr);
2048 		break;
2049 
2050 	case TUNGETSNDBUF:
2051 		sndbuf = tfile->socket.sk->sk_sndbuf;
2052 		if (copy_to_user(argp, &sndbuf, sizeof(sndbuf)))
2053 			ret = -EFAULT;
2054 		break;
2055 
2056 	case TUNSETSNDBUF:
2057 		if (copy_from_user(&sndbuf, argp, sizeof(sndbuf))) {
2058 			ret = -EFAULT;
2059 			break;
2060 		}
2061 
2062 		tun->sndbuf = sndbuf;
2063 		tun_set_sndbuf(tun);
2064 		break;
2065 
2066 	case TUNGETVNETHDRSZ:
2067 		vnet_hdr_sz = tun->vnet_hdr_sz;
2068 		if (copy_to_user(argp, &vnet_hdr_sz, sizeof(vnet_hdr_sz)))
2069 			ret = -EFAULT;
2070 		break;
2071 
2072 	case TUNSETVNETHDRSZ:
2073 		if (copy_from_user(&vnet_hdr_sz, argp, sizeof(vnet_hdr_sz))) {
2074 			ret = -EFAULT;
2075 			break;
2076 		}
2077 		if (vnet_hdr_sz < (int)sizeof(struct virtio_net_hdr)) {
2078 			ret = -EINVAL;
2079 			break;
2080 		}
2081 
2082 		tun->vnet_hdr_sz = vnet_hdr_sz;
2083 		break;
2084 
2085 	case TUNGETVNETLE:
2086 		le = !!(tun->flags & TUN_VNET_LE);
2087 		if (put_user(le, (int __user *)argp))
2088 			ret = -EFAULT;
2089 		break;
2090 
2091 	case TUNSETVNETLE:
2092 		if (get_user(le, (int __user *)argp)) {
2093 			ret = -EFAULT;
2094 			break;
2095 		}
2096 		if (le)
2097 			tun->flags |= TUN_VNET_LE;
2098 		else
2099 			tun->flags &= ~TUN_VNET_LE;
2100 		break;
2101 
2102 	case TUNGETVNETBE:
2103 		ret = tun_get_vnet_be(tun, argp);
2104 		break;
2105 
2106 	case TUNSETVNETBE:
2107 		ret = tun_set_vnet_be(tun, argp);
2108 		break;
2109 
2110 	case TUNATTACHFILTER:
2111 		/* Can be set only for TAPs */
2112 		ret = -EINVAL;
2113 		if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
2114 			break;
2115 		ret = -EFAULT;
2116 		if (copy_from_user(&tun->fprog, argp, sizeof(tun->fprog)))
2117 			break;
2118 
2119 		ret = tun_attach_filter(tun);
2120 		break;
2121 
2122 	case TUNDETACHFILTER:
2123 		/* Can be set only for TAPs */
2124 		ret = -EINVAL;
2125 		if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
2126 			break;
2127 		ret = 0;
2128 		tun_detach_filter(tun, tun->numqueues);
2129 		break;
2130 
2131 	case TUNGETFILTER:
2132 		ret = -EINVAL;
2133 		if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
2134 			break;
2135 		ret = -EFAULT;
2136 		if (copy_to_user(argp, &tun->fprog, sizeof(tun->fprog)))
2137 			break;
2138 		ret = 0;
2139 		break;
2140 
2141 	default:
2142 		ret = -EINVAL;
2143 		break;
2144 	}
2145 
2146 unlock:
2147 	rtnl_unlock();
2148 	if (tun)
2149 		tun_put(tun);
2150 	return ret;
2151 }
2152 
2153 static long tun_chr_ioctl(struct file *file,
2154 			  unsigned int cmd, unsigned long arg)
2155 {
2156 	return __tun_chr_ioctl(file, cmd, arg, sizeof (struct ifreq));
2157 }
2158 
2159 #ifdef CONFIG_COMPAT
2160 static long tun_chr_compat_ioctl(struct file *file,
2161 			 unsigned int cmd, unsigned long arg)
2162 {
2163 	switch (cmd) {
2164 	case TUNSETIFF:
2165 	case TUNGETIFF:
2166 	case TUNSETTXFILTER:
2167 	case TUNGETSNDBUF:
2168 	case TUNSETSNDBUF:
2169 	case SIOCGIFHWADDR:
2170 	case SIOCSIFHWADDR:
2171 		arg = (unsigned long)compat_ptr(arg);
2172 		break;
2173 	default:
2174 		arg = (compat_ulong_t)arg;
2175 		break;
2176 	}
2177 
2178 	/*
2179 	 * compat_ifreq is shorter than ifreq, so we must not access beyond
2180 	 * the end of that structure. All fields that are used in this
2181 	 * driver are compatible though, we don't need to convert the
2182 	 * contents.
2183 	 */
2184 	return __tun_chr_ioctl(file, cmd, arg, sizeof(struct compat_ifreq));
2185 }
2186 #endif /* CONFIG_COMPAT */
2187 
2188 static int tun_chr_fasync(int fd, struct file *file, int on)
2189 {
2190 	struct tun_file *tfile = file->private_data;
2191 	int ret;
2192 
2193 	if ((ret = fasync_helper(fd, file, on, &tfile->fasync)) < 0)
2194 		goto out;
2195 
2196 	if (on) {
2197 		__f_setown(file, task_pid(current), PIDTYPE_PID, 0);
2198 		tfile->flags |= TUN_FASYNC;
2199 	} else
2200 		tfile->flags &= ~TUN_FASYNC;
2201 	ret = 0;
2202 out:
2203 	return ret;
2204 }
2205 
2206 static int tun_chr_open(struct inode *inode, struct file * file)
2207 {
2208 	struct net *net = current->nsproxy->net_ns;
2209 	struct tun_file *tfile;
2210 
2211 	DBG1(KERN_INFO, "tunX: tun_chr_open\n");
2212 
2213 	tfile = (struct tun_file *)sk_alloc(net, AF_UNSPEC, GFP_KERNEL,
2214 					    &tun_proto, 0);
2215 	if (!tfile)
2216 		return -ENOMEM;
2217 	RCU_INIT_POINTER(tfile->tun, NULL);
2218 	tfile->flags = 0;
2219 	tfile->ifindex = 0;
2220 
2221 	init_waitqueue_head(&tfile->wq.wait);
2222 	RCU_INIT_POINTER(tfile->socket.wq, &tfile->wq);
2223 
2224 	tfile->socket.file = file;
2225 	tfile->socket.ops = &tun_socket_ops;
2226 
2227 	sock_init_data(&tfile->socket, &tfile->sk);
2228 
2229 	tfile->sk.sk_write_space = tun_sock_write_space;
2230 	tfile->sk.sk_sndbuf = INT_MAX;
2231 
2232 	file->private_data = tfile;
2233 	INIT_LIST_HEAD(&tfile->next);
2234 
2235 	sock_set_flag(&tfile->sk, SOCK_ZEROCOPY);
2236 
2237 	return 0;
2238 }
2239 
2240 static int tun_chr_close(struct inode *inode, struct file *file)
2241 {
2242 	struct tun_file *tfile = file->private_data;
2243 
2244 	tun_detach(tfile, true);
2245 
2246 	return 0;
2247 }
2248 
2249 #ifdef CONFIG_PROC_FS
2250 static void tun_chr_show_fdinfo(struct seq_file *m, struct file *f)
2251 {
2252 	struct tun_struct *tun;
2253 	struct ifreq ifr;
2254 
2255 	memset(&ifr, 0, sizeof(ifr));
2256 
2257 	rtnl_lock();
2258 	tun = tun_get(f);
2259 	if (tun)
2260 		tun_get_iff(current->nsproxy->net_ns, tun, &ifr);
2261 	rtnl_unlock();
2262 
2263 	if (tun)
2264 		tun_put(tun);
2265 
2266 	seq_printf(m, "iff:\t%s\n", ifr.ifr_name);
2267 }
2268 #endif
2269 
2270 static const struct file_operations tun_fops = {
2271 	.owner	= THIS_MODULE,
2272 	.llseek = no_llseek,
2273 	.read_iter  = tun_chr_read_iter,
2274 	.write_iter = tun_chr_write_iter,
2275 	.poll	= tun_chr_poll,
2276 	.unlocked_ioctl	= tun_chr_ioctl,
2277 #ifdef CONFIG_COMPAT
2278 	.compat_ioctl = tun_chr_compat_ioctl,
2279 #endif
2280 	.open	= tun_chr_open,
2281 	.release = tun_chr_close,
2282 	.fasync = tun_chr_fasync,
2283 #ifdef CONFIG_PROC_FS
2284 	.show_fdinfo = tun_chr_show_fdinfo,
2285 #endif
2286 };
2287 
2288 static struct miscdevice tun_miscdev = {
2289 	.minor = TUN_MINOR,
2290 	.name = "tun",
2291 	.nodename = "net/tun",
2292 	.fops = &tun_fops,
2293 };
2294 
2295 /* ethtool interface */
2296 
2297 static int tun_get_settings(struct net_device *dev, struct ethtool_cmd *cmd)
2298 {
2299 	cmd->supported		= 0;
2300 	cmd->advertising	= 0;
2301 	ethtool_cmd_speed_set(cmd, SPEED_10);
2302 	cmd->duplex		= DUPLEX_FULL;
2303 	cmd->port		= PORT_TP;
2304 	cmd->phy_address	= 0;
2305 	cmd->transceiver	= XCVR_INTERNAL;
2306 	cmd->autoneg		= AUTONEG_DISABLE;
2307 	cmd->maxtxpkt		= 0;
2308 	cmd->maxrxpkt		= 0;
2309 	return 0;
2310 }
2311 
2312 static void tun_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
2313 {
2314 	struct tun_struct *tun = netdev_priv(dev);
2315 
2316 	strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
2317 	strlcpy(info->version, DRV_VERSION, sizeof(info->version));
2318 
2319 	switch (tun->flags & TUN_TYPE_MASK) {
2320 	case IFF_TUN:
2321 		strlcpy(info->bus_info, "tun", sizeof(info->bus_info));
2322 		break;
2323 	case IFF_TAP:
2324 		strlcpy(info->bus_info, "tap", sizeof(info->bus_info));
2325 		break;
2326 	}
2327 }
2328 
2329 static u32 tun_get_msglevel(struct net_device *dev)
2330 {
2331 #ifdef TUN_DEBUG
2332 	struct tun_struct *tun = netdev_priv(dev);
2333 	return tun->debug;
2334 #else
2335 	return -EOPNOTSUPP;
2336 #endif
2337 }
2338 
2339 static void tun_set_msglevel(struct net_device *dev, u32 value)
2340 {
2341 #ifdef TUN_DEBUG
2342 	struct tun_struct *tun = netdev_priv(dev);
2343 	tun->debug = value;
2344 #endif
2345 }
2346 
2347 static const struct ethtool_ops tun_ethtool_ops = {
2348 	.get_settings	= tun_get_settings,
2349 	.get_drvinfo	= tun_get_drvinfo,
2350 	.get_msglevel	= tun_get_msglevel,
2351 	.set_msglevel	= tun_set_msglevel,
2352 	.get_link	= ethtool_op_get_link,
2353 	.get_ts_info	= ethtool_op_get_ts_info,
2354 };
2355 
2356 
2357 static int __init tun_init(void)
2358 {
2359 	int ret = 0;
2360 
2361 	pr_info("%s, %s\n", DRV_DESCRIPTION, DRV_VERSION);
2362 	pr_info("%s\n", DRV_COPYRIGHT);
2363 
2364 	ret = rtnl_link_register(&tun_link_ops);
2365 	if (ret) {
2366 		pr_err("Can't register link_ops\n");
2367 		goto err_linkops;
2368 	}
2369 
2370 	ret = misc_register(&tun_miscdev);
2371 	if (ret) {
2372 		pr_err("Can't register misc device %d\n", TUN_MINOR);
2373 		goto err_misc;
2374 	}
2375 	return  0;
2376 err_misc:
2377 	rtnl_link_unregister(&tun_link_ops);
2378 err_linkops:
2379 	return ret;
2380 }
2381 
2382 static void tun_cleanup(void)
2383 {
2384 	misc_deregister(&tun_miscdev);
2385 	rtnl_link_unregister(&tun_link_ops);
2386 }
2387 
2388 /* Get an underlying socket object from tun file.  Returns error unless file is
2389  * attached to a device.  The returned object works like a packet socket, it
2390  * can be used for sock_sendmsg/sock_recvmsg.  The caller is responsible for
2391  * holding a reference to the file for as long as the socket is in use. */
2392 struct socket *tun_get_socket(struct file *file)
2393 {
2394 	struct tun_file *tfile;
2395 	if (file->f_op != &tun_fops)
2396 		return ERR_PTR(-EINVAL);
2397 	tfile = file->private_data;
2398 	if (!tfile)
2399 		return ERR_PTR(-EBADFD);
2400 	return &tfile->socket;
2401 }
2402 EXPORT_SYMBOL_GPL(tun_get_socket);
2403 
2404 module_init(tun_init);
2405 module_exit(tun_cleanup);
2406 MODULE_DESCRIPTION(DRV_DESCRIPTION);
2407 MODULE_AUTHOR(DRV_COPYRIGHT);
2408 MODULE_LICENSE("GPL");
2409 MODULE_ALIAS_MISCDEV(TUN_MINOR);
2410 MODULE_ALIAS("devname:net/tun");
2411