xref: /openbmc/linux/drivers/net/tun.c (revision 4f3db074)
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 net *net;
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 static inline u16 tun16_to_cpu(struct tun_struct *tun, __virtio16 val)
210 {
211 	return __virtio16_to_cpu(tun->flags & TUN_VNET_LE, val);
212 }
213 
214 static inline __virtio16 cpu_to_tun16(struct tun_struct *tun, u16 val)
215 {
216 	return __cpu_to_virtio16(tun->flags & TUN_VNET_LE, val);
217 }
218 
219 static inline u32 tun_hashfn(u32 rxhash)
220 {
221 	return rxhash & 0x3ff;
222 }
223 
224 static struct tun_flow_entry *tun_flow_find(struct hlist_head *head, u32 rxhash)
225 {
226 	struct tun_flow_entry *e;
227 
228 	hlist_for_each_entry_rcu(e, head, hash_link) {
229 		if (e->rxhash == rxhash)
230 			return e;
231 	}
232 	return NULL;
233 }
234 
235 static struct tun_flow_entry *tun_flow_create(struct tun_struct *tun,
236 					      struct hlist_head *head,
237 					      u32 rxhash, u16 queue_index)
238 {
239 	struct tun_flow_entry *e = kmalloc(sizeof(*e), GFP_ATOMIC);
240 
241 	if (e) {
242 		tun_debug(KERN_INFO, tun, "create flow: hash %u index %u\n",
243 			  rxhash, queue_index);
244 		e->updated = jiffies;
245 		e->rxhash = rxhash;
246 		e->rps_rxhash = 0;
247 		e->queue_index = queue_index;
248 		e->tun = tun;
249 		hlist_add_head_rcu(&e->hash_link, head);
250 		++tun->flow_count;
251 	}
252 	return e;
253 }
254 
255 static void tun_flow_delete(struct tun_struct *tun, struct tun_flow_entry *e)
256 {
257 	tun_debug(KERN_INFO, tun, "delete flow: hash %u index %u\n",
258 		  e->rxhash, e->queue_index);
259 	hlist_del_rcu(&e->hash_link);
260 	kfree_rcu(e, rcu);
261 	--tun->flow_count;
262 }
263 
264 static void tun_flow_flush(struct tun_struct *tun)
265 {
266 	int i;
267 
268 	spin_lock_bh(&tun->lock);
269 	for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
270 		struct tun_flow_entry *e;
271 		struct hlist_node *n;
272 
273 		hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link)
274 			tun_flow_delete(tun, e);
275 	}
276 	spin_unlock_bh(&tun->lock);
277 }
278 
279 static void tun_flow_delete_by_queue(struct tun_struct *tun, u16 queue_index)
280 {
281 	int i;
282 
283 	spin_lock_bh(&tun->lock);
284 	for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
285 		struct tun_flow_entry *e;
286 		struct hlist_node *n;
287 
288 		hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
289 			if (e->queue_index == queue_index)
290 				tun_flow_delete(tun, e);
291 		}
292 	}
293 	spin_unlock_bh(&tun->lock);
294 }
295 
296 static void tun_flow_cleanup(unsigned long data)
297 {
298 	struct tun_struct *tun = (struct tun_struct *)data;
299 	unsigned long delay = tun->ageing_time;
300 	unsigned long next_timer = jiffies + delay;
301 	unsigned long count = 0;
302 	int i;
303 
304 	tun_debug(KERN_INFO, tun, "tun_flow_cleanup\n");
305 
306 	spin_lock_bh(&tun->lock);
307 	for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
308 		struct tun_flow_entry *e;
309 		struct hlist_node *n;
310 
311 		hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
312 			unsigned long this_timer;
313 			count++;
314 			this_timer = e->updated + delay;
315 			if (time_before_eq(this_timer, jiffies))
316 				tun_flow_delete(tun, e);
317 			else if (time_before(this_timer, next_timer))
318 				next_timer = this_timer;
319 		}
320 	}
321 
322 	if (count)
323 		mod_timer(&tun->flow_gc_timer, round_jiffies_up(next_timer));
324 	spin_unlock_bh(&tun->lock);
325 }
326 
327 static void tun_flow_update(struct tun_struct *tun, u32 rxhash,
328 			    struct tun_file *tfile)
329 {
330 	struct hlist_head *head;
331 	struct tun_flow_entry *e;
332 	unsigned long delay = tun->ageing_time;
333 	u16 queue_index = tfile->queue_index;
334 
335 	if (!rxhash)
336 		return;
337 	else
338 		head = &tun->flows[tun_hashfn(rxhash)];
339 
340 	rcu_read_lock();
341 
342 	/* We may get a very small possibility of OOO during switching, not
343 	 * worth to optimize.*/
344 	if (tun->numqueues == 1 || tfile->detached)
345 		goto unlock;
346 
347 	e = tun_flow_find(head, rxhash);
348 	if (likely(e)) {
349 		/* TODO: keep queueing to old queue until it's empty? */
350 		e->queue_index = queue_index;
351 		e->updated = jiffies;
352 		sock_rps_record_flow_hash(e->rps_rxhash);
353 	} else {
354 		spin_lock_bh(&tun->lock);
355 		if (!tun_flow_find(head, rxhash) &&
356 		    tun->flow_count < MAX_TAP_FLOWS)
357 			tun_flow_create(tun, head, rxhash, queue_index);
358 
359 		if (!timer_pending(&tun->flow_gc_timer))
360 			mod_timer(&tun->flow_gc_timer,
361 				  round_jiffies_up(jiffies + delay));
362 		spin_unlock_bh(&tun->lock);
363 	}
364 
365 unlock:
366 	rcu_read_unlock();
367 }
368 
369 /**
370  * Save the hash received in the stack receive path and update the
371  * flow_hash table accordingly.
372  */
373 static inline void tun_flow_save_rps_rxhash(struct tun_flow_entry *e, u32 hash)
374 {
375 	if (unlikely(e->rps_rxhash != hash))
376 		e->rps_rxhash = hash;
377 }
378 
379 /* We try to identify a flow through its rxhash first. The reason that
380  * we do not check rxq no. is because some cards(e.g 82599), chooses
381  * the rxq based on the txq where the last packet of the flow comes. As
382  * the userspace application move between processors, we may get a
383  * different rxq no. here. If we could not get rxhash, then we would
384  * hope the rxq no. may help here.
385  */
386 static u16 tun_select_queue(struct net_device *dev, struct sk_buff *skb,
387 			    void *accel_priv, select_queue_fallback_t fallback)
388 {
389 	struct tun_struct *tun = netdev_priv(dev);
390 	struct tun_flow_entry *e;
391 	u32 txq = 0;
392 	u32 numqueues = 0;
393 
394 	rcu_read_lock();
395 	numqueues = ACCESS_ONCE(tun->numqueues);
396 
397 	txq = skb_get_hash(skb);
398 	if (txq) {
399 		e = tun_flow_find(&tun->flows[tun_hashfn(txq)], txq);
400 		if (e) {
401 			tun_flow_save_rps_rxhash(e, txq);
402 			txq = e->queue_index;
403 		} else
404 			/* use multiply and shift instead of expensive divide */
405 			txq = ((u64)txq * numqueues) >> 32;
406 	} else if (likely(skb_rx_queue_recorded(skb))) {
407 		txq = skb_get_rx_queue(skb);
408 		while (unlikely(txq >= numqueues))
409 			txq -= numqueues;
410 	}
411 
412 	rcu_read_unlock();
413 	return txq;
414 }
415 
416 static inline bool tun_not_capable(struct tun_struct *tun)
417 {
418 	const struct cred *cred = current_cred();
419 	struct net *net = dev_net(tun->dev);
420 
421 	return ((uid_valid(tun->owner) && !uid_eq(cred->euid, tun->owner)) ||
422 		  (gid_valid(tun->group) && !in_egroup_p(tun->group))) &&
423 		!ns_capable(net->user_ns, CAP_NET_ADMIN);
424 }
425 
426 static void tun_set_real_num_queues(struct tun_struct *tun)
427 {
428 	netif_set_real_num_tx_queues(tun->dev, tun->numqueues);
429 	netif_set_real_num_rx_queues(tun->dev, tun->numqueues);
430 }
431 
432 static void tun_disable_queue(struct tun_struct *tun, struct tun_file *tfile)
433 {
434 	tfile->detached = tun;
435 	list_add_tail(&tfile->next, &tun->disabled);
436 	++tun->numdisabled;
437 }
438 
439 static struct tun_struct *tun_enable_queue(struct tun_file *tfile)
440 {
441 	struct tun_struct *tun = tfile->detached;
442 
443 	tfile->detached = NULL;
444 	list_del_init(&tfile->next);
445 	--tun->numdisabled;
446 	return tun;
447 }
448 
449 static void tun_queue_purge(struct tun_file *tfile)
450 {
451 	skb_queue_purge(&tfile->sk.sk_receive_queue);
452 	skb_queue_purge(&tfile->sk.sk_error_queue);
453 }
454 
455 static void __tun_detach(struct tun_file *tfile, bool clean)
456 {
457 	struct tun_file *ntfile;
458 	struct tun_struct *tun;
459 
460 	tun = rtnl_dereference(tfile->tun);
461 
462 	if (tun && !tfile->detached) {
463 		u16 index = tfile->queue_index;
464 		BUG_ON(index >= tun->numqueues);
465 
466 		rcu_assign_pointer(tun->tfiles[index],
467 				   tun->tfiles[tun->numqueues - 1]);
468 		ntfile = rtnl_dereference(tun->tfiles[index]);
469 		ntfile->queue_index = index;
470 
471 		--tun->numqueues;
472 		if (clean) {
473 			RCU_INIT_POINTER(tfile->tun, NULL);
474 			sock_put(&tfile->sk);
475 		} else
476 			tun_disable_queue(tun, tfile);
477 
478 		synchronize_net();
479 		tun_flow_delete_by_queue(tun, tun->numqueues + 1);
480 		/* Drop read queue */
481 		tun_queue_purge(tfile);
482 		tun_set_real_num_queues(tun);
483 	} else if (tfile->detached && clean) {
484 		tun = tun_enable_queue(tfile);
485 		sock_put(&tfile->sk);
486 	}
487 
488 	if (clean) {
489 		if (tun && tun->numqueues == 0 && tun->numdisabled == 0) {
490 			netif_carrier_off(tun->dev);
491 
492 			if (!(tun->flags & IFF_PERSIST) &&
493 			    tun->dev->reg_state == NETREG_REGISTERED)
494 				unregister_netdevice(tun->dev);
495 		}
496 
497 		BUG_ON(!test_bit(SOCK_EXTERNALLY_ALLOCATED,
498 				 &tfile->socket.flags));
499 		sk_release_kernel(&tfile->sk);
500 	}
501 }
502 
503 static void tun_detach(struct tun_file *tfile, bool clean)
504 {
505 	rtnl_lock();
506 	__tun_detach(tfile, clean);
507 	rtnl_unlock();
508 }
509 
510 static void tun_detach_all(struct net_device *dev)
511 {
512 	struct tun_struct *tun = netdev_priv(dev);
513 	struct tun_file *tfile, *tmp;
514 	int i, n = tun->numqueues;
515 
516 	for (i = 0; i < n; i++) {
517 		tfile = rtnl_dereference(tun->tfiles[i]);
518 		BUG_ON(!tfile);
519 		tfile->socket.sk->sk_data_ready(tfile->socket.sk);
520 		RCU_INIT_POINTER(tfile->tun, NULL);
521 		--tun->numqueues;
522 	}
523 	list_for_each_entry(tfile, &tun->disabled, next) {
524 		tfile->socket.sk->sk_data_ready(tfile->socket.sk);
525 		RCU_INIT_POINTER(tfile->tun, NULL);
526 	}
527 	BUG_ON(tun->numqueues != 0);
528 
529 	synchronize_net();
530 	for (i = 0; i < n; i++) {
531 		tfile = rtnl_dereference(tun->tfiles[i]);
532 		/* Drop read queue */
533 		tun_queue_purge(tfile);
534 		sock_put(&tfile->sk);
535 	}
536 	list_for_each_entry_safe(tfile, tmp, &tun->disabled, next) {
537 		tun_enable_queue(tfile);
538 		tun_queue_purge(tfile);
539 		sock_put(&tfile->sk);
540 	}
541 	BUG_ON(tun->numdisabled != 0);
542 
543 	if (tun->flags & IFF_PERSIST)
544 		module_put(THIS_MODULE);
545 }
546 
547 static int tun_attach(struct tun_struct *tun, struct file *file, bool skip_filter)
548 {
549 	struct tun_file *tfile = file->private_data;
550 	int err;
551 
552 	err = security_tun_dev_attach(tfile->socket.sk, tun->security);
553 	if (err < 0)
554 		goto out;
555 
556 	err = -EINVAL;
557 	if (rtnl_dereference(tfile->tun) && !tfile->detached)
558 		goto out;
559 
560 	err = -EBUSY;
561 	if (!(tun->flags & IFF_MULTI_QUEUE) && tun->numqueues == 1)
562 		goto out;
563 
564 	err = -E2BIG;
565 	if (!tfile->detached &&
566 	    tun->numqueues + tun->numdisabled == MAX_TAP_QUEUES)
567 		goto out;
568 
569 	err = 0;
570 
571 	/* Re-attach the filter to persist device */
572 	if (!skip_filter && (tun->filter_attached == true)) {
573 		err = sk_attach_filter(&tun->fprog, tfile->socket.sk);
574 		if (!err)
575 			goto out;
576 	}
577 	tfile->queue_index = tun->numqueues;
578 	rcu_assign_pointer(tfile->tun, tun);
579 	rcu_assign_pointer(tun->tfiles[tun->numqueues], tfile);
580 	tun->numqueues++;
581 
582 	if (tfile->detached)
583 		tun_enable_queue(tfile);
584 	else
585 		sock_hold(&tfile->sk);
586 
587 	tun_set_real_num_queues(tun);
588 
589 	/* device is allowed to go away first, so no need to hold extra
590 	 * refcnt.
591 	 */
592 
593 out:
594 	return err;
595 }
596 
597 static struct tun_struct *__tun_get(struct tun_file *tfile)
598 {
599 	struct tun_struct *tun;
600 
601 	rcu_read_lock();
602 	tun = rcu_dereference(tfile->tun);
603 	if (tun)
604 		dev_hold(tun->dev);
605 	rcu_read_unlock();
606 
607 	return tun;
608 }
609 
610 static struct tun_struct *tun_get(struct file *file)
611 {
612 	return __tun_get(file->private_data);
613 }
614 
615 static void tun_put(struct tun_struct *tun)
616 {
617 	dev_put(tun->dev);
618 }
619 
620 /* TAP filtering */
621 static void addr_hash_set(u32 *mask, const u8 *addr)
622 {
623 	int n = ether_crc(ETH_ALEN, addr) >> 26;
624 	mask[n >> 5] |= (1 << (n & 31));
625 }
626 
627 static unsigned int addr_hash_test(const u32 *mask, const u8 *addr)
628 {
629 	int n = ether_crc(ETH_ALEN, addr) >> 26;
630 	return mask[n >> 5] & (1 << (n & 31));
631 }
632 
633 static int update_filter(struct tap_filter *filter, void __user *arg)
634 {
635 	struct { u8 u[ETH_ALEN]; } *addr;
636 	struct tun_filter uf;
637 	int err, alen, n, nexact;
638 
639 	if (copy_from_user(&uf, arg, sizeof(uf)))
640 		return -EFAULT;
641 
642 	if (!uf.count) {
643 		/* Disabled */
644 		filter->count = 0;
645 		return 0;
646 	}
647 
648 	alen = ETH_ALEN * uf.count;
649 	addr = kmalloc(alen, GFP_KERNEL);
650 	if (!addr)
651 		return -ENOMEM;
652 
653 	if (copy_from_user(addr, arg + sizeof(uf), alen)) {
654 		err = -EFAULT;
655 		goto done;
656 	}
657 
658 	/* The filter is updated without holding any locks. Which is
659 	 * perfectly safe. We disable it first and in the worst
660 	 * case we'll accept a few undesired packets. */
661 	filter->count = 0;
662 	wmb();
663 
664 	/* Use first set of addresses as an exact filter */
665 	for (n = 0; n < uf.count && n < FLT_EXACT_COUNT; n++)
666 		memcpy(filter->addr[n], addr[n].u, ETH_ALEN);
667 
668 	nexact = n;
669 
670 	/* Remaining multicast addresses are hashed,
671 	 * unicast will leave the filter disabled. */
672 	memset(filter->mask, 0, sizeof(filter->mask));
673 	for (; n < uf.count; n++) {
674 		if (!is_multicast_ether_addr(addr[n].u)) {
675 			err = 0; /* no filter */
676 			goto done;
677 		}
678 		addr_hash_set(filter->mask, addr[n].u);
679 	}
680 
681 	/* For ALLMULTI just set the mask to all ones.
682 	 * This overrides the mask populated above. */
683 	if ((uf.flags & TUN_FLT_ALLMULTI))
684 		memset(filter->mask, ~0, sizeof(filter->mask));
685 
686 	/* Now enable the filter */
687 	wmb();
688 	filter->count = nexact;
689 
690 	/* Return the number of exact filters */
691 	err = nexact;
692 
693 done:
694 	kfree(addr);
695 	return err;
696 }
697 
698 /* Returns: 0 - drop, !=0 - accept */
699 static int run_filter(struct tap_filter *filter, const struct sk_buff *skb)
700 {
701 	/* Cannot use eth_hdr(skb) here because skb_mac_hdr() is incorrect
702 	 * at this point. */
703 	struct ethhdr *eh = (struct ethhdr *) skb->data;
704 	int i;
705 
706 	/* Exact match */
707 	for (i = 0; i < filter->count; i++)
708 		if (ether_addr_equal(eh->h_dest, filter->addr[i]))
709 			return 1;
710 
711 	/* Inexact match (multicast only) */
712 	if (is_multicast_ether_addr(eh->h_dest))
713 		return addr_hash_test(filter->mask, eh->h_dest);
714 
715 	return 0;
716 }
717 
718 /*
719  * Checks whether the packet is accepted or not.
720  * Returns: 0 - drop, !=0 - accept
721  */
722 static int check_filter(struct tap_filter *filter, const struct sk_buff *skb)
723 {
724 	if (!filter->count)
725 		return 1;
726 
727 	return run_filter(filter, skb);
728 }
729 
730 /* Network device part of the driver */
731 
732 static const struct ethtool_ops tun_ethtool_ops;
733 
734 /* Net device detach from fd. */
735 static void tun_net_uninit(struct net_device *dev)
736 {
737 	tun_detach_all(dev);
738 }
739 
740 /* Net device open. */
741 static int tun_net_open(struct net_device *dev)
742 {
743 	netif_tx_start_all_queues(dev);
744 	return 0;
745 }
746 
747 /* Net device close. */
748 static int tun_net_close(struct net_device *dev)
749 {
750 	netif_tx_stop_all_queues(dev);
751 	return 0;
752 }
753 
754 /* Net device start xmit */
755 static netdev_tx_t tun_net_xmit(struct sk_buff *skb, struct net_device *dev)
756 {
757 	struct tun_struct *tun = netdev_priv(dev);
758 	int txq = skb->queue_mapping;
759 	struct tun_file *tfile;
760 	u32 numqueues = 0;
761 
762 	rcu_read_lock();
763 	tfile = rcu_dereference(tun->tfiles[txq]);
764 	numqueues = ACCESS_ONCE(tun->numqueues);
765 
766 	/* Drop packet if interface is not attached */
767 	if (txq >= numqueues)
768 		goto drop;
769 
770 	if (numqueues == 1) {
771 		/* Select queue was not called for the skbuff, so we extract the
772 		 * RPS hash and save it into the flow_table here.
773 		 */
774 		__u32 rxhash;
775 
776 		rxhash = skb_get_hash(skb);
777 		if (rxhash) {
778 			struct tun_flow_entry *e;
779 			e = tun_flow_find(&tun->flows[tun_hashfn(rxhash)],
780 					rxhash);
781 			if (e)
782 				tun_flow_save_rps_rxhash(e, rxhash);
783 		}
784 	}
785 
786 	tun_debug(KERN_INFO, tun, "tun_net_xmit %d\n", skb->len);
787 
788 	BUG_ON(!tfile);
789 
790 	/* Drop if the filter does not like it.
791 	 * This is a noop if the filter is disabled.
792 	 * Filter can be enabled only for the TAP devices. */
793 	if (!check_filter(&tun->txflt, skb))
794 		goto drop;
795 
796 	if (tfile->socket.sk->sk_filter &&
797 	    sk_filter(tfile->socket.sk, skb))
798 		goto drop;
799 
800 	/* Limit the number of packets queued by dividing txq length with the
801 	 * number of queues.
802 	 */
803 	if (skb_queue_len(&tfile->socket.sk->sk_receive_queue) * numqueues
804 			  >= dev->tx_queue_len)
805 		goto drop;
806 
807 	if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC)))
808 		goto drop;
809 
810 	if (skb->sk) {
811 		sock_tx_timestamp(skb->sk, &skb_shinfo(skb)->tx_flags);
812 		sw_tx_timestamp(skb);
813 	}
814 
815 	/* Orphan the skb - required as we might hang on to it
816 	 * for indefinite time.
817 	 */
818 	skb_orphan(skb);
819 
820 	nf_reset(skb);
821 
822 	/* Enqueue packet */
823 	skb_queue_tail(&tfile->socket.sk->sk_receive_queue, skb);
824 
825 	/* Notify and wake up reader process */
826 	if (tfile->flags & TUN_FASYNC)
827 		kill_fasync(&tfile->fasync, SIGIO, POLL_IN);
828 	tfile->socket.sk->sk_data_ready(tfile->socket.sk);
829 
830 	rcu_read_unlock();
831 	return NETDEV_TX_OK;
832 
833 drop:
834 	dev->stats.tx_dropped++;
835 	skb_tx_error(skb);
836 	kfree_skb(skb);
837 	rcu_read_unlock();
838 	return NET_XMIT_DROP;
839 }
840 
841 static void tun_net_mclist(struct net_device *dev)
842 {
843 	/*
844 	 * This callback is supposed to deal with mc filter in
845 	 * _rx_ path and has nothing to do with the _tx_ path.
846 	 * In rx path we always accept everything userspace gives us.
847 	 */
848 }
849 
850 #define MIN_MTU 68
851 #define MAX_MTU 65535
852 
853 static int
854 tun_net_change_mtu(struct net_device *dev, int new_mtu)
855 {
856 	if (new_mtu < MIN_MTU || new_mtu + dev->hard_header_len > MAX_MTU)
857 		return -EINVAL;
858 	dev->mtu = new_mtu;
859 	return 0;
860 }
861 
862 static netdev_features_t tun_net_fix_features(struct net_device *dev,
863 	netdev_features_t features)
864 {
865 	struct tun_struct *tun = netdev_priv(dev);
866 
867 	return (features & tun->set_features) | (features & ~TUN_USER_FEATURES);
868 }
869 #ifdef CONFIG_NET_POLL_CONTROLLER
870 static void tun_poll_controller(struct net_device *dev)
871 {
872 	/*
873 	 * Tun only receives frames when:
874 	 * 1) the char device endpoint gets data from user space
875 	 * 2) the tun socket gets a sendmsg call from user space
876 	 * Since both of those are synchronous operations, we are guaranteed
877 	 * never to have pending data when we poll for it
878 	 * so there is nothing to do here but return.
879 	 * We need this though so netpoll recognizes us as an interface that
880 	 * supports polling, which enables bridge devices in virt setups to
881 	 * still use netconsole
882 	 */
883 	return;
884 }
885 #endif
886 static const struct net_device_ops tun_netdev_ops = {
887 	.ndo_uninit		= tun_net_uninit,
888 	.ndo_open		= tun_net_open,
889 	.ndo_stop		= tun_net_close,
890 	.ndo_start_xmit		= tun_net_xmit,
891 	.ndo_change_mtu		= tun_net_change_mtu,
892 	.ndo_fix_features	= tun_net_fix_features,
893 	.ndo_select_queue	= tun_select_queue,
894 #ifdef CONFIG_NET_POLL_CONTROLLER
895 	.ndo_poll_controller	= tun_poll_controller,
896 #endif
897 };
898 
899 static const struct net_device_ops tap_netdev_ops = {
900 	.ndo_uninit		= tun_net_uninit,
901 	.ndo_open		= tun_net_open,
902 	.ndo_stop		= tun_net_close,
903 	.ndo_start_xmit		= tun_net_xmit,
904 	.ndo_change_mtu		= tun_net_change_mtu,
905 	.ndo_fix_features	= tun_net_fix_features,
906 	.ndo_set_rx_mode	= tun_net_mclist,
907 	.ndo_set_mac_address	= eth_mac_addr,
908 	.ndo_validate_addr	= eth_validate_addr,
909 	.ndo_select_queue	= tun_select_queue,
910 #ifdef CONFIG_NET_POLL_CONTROLLER
911 	.ndo_poll_controller	= tun_poll_controller,
912 #endif
913 };
914 
915 static void tun_flow_init(struct tun_struct *tun)
916 {
917 	int i;
918 
919 	for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++)
920 		INIT_HLIST_HEAD(&tun->flows[i]);
921 
922 	tun->ageing_time = TUN_FLOW_EXPIRE;
923 	setup_timer(&tun->flow_gc_timer, tun_flow_cleanup, (unsigned long)tun);
924 	mod_timer(&tun->flow_gc_timer,
925 		  round_jiffies_up(jiffies + tun->ageing_time));
926 }
927 
928 static void tun_flow_uninit(struct tun_struct *tun)
929 {
930 	del_timer_sync(&tun->flow_gc_timer);
931 	tun_flow_flush(tun);
932 }
933 
934 /* Initialize net device. */
935 static void tun_net_init(struct net_device *dev)
936 {
937 	struct tun_struct *tun = netdev_priv(dev);
938 
939 	switch (tun->flags & TUN_TYPE_MASK) {
940 	case IFF_TUN:
941 		dev->netdev_ops = &tun_netdev_ops;
942 
943 		/* Point-to-Point TUN Device */
944 		dev->hard_header_len = 0;
945 		dev->addr_len = 0;
946 		dev->mtu = 1500;
947 
948 		/* Zero header length */
949 		dev->type = ARPHRD_NONE;
950 		dev->flags = IFF_POINTOPOINT | IFF_NOARP | IFF_MULTICAST;
951 		dev->tx_queue_len = TUN_READQ_SIZE;  /* We prefer our own queue length */
952 		break;
953 
954 	case IFF_TAP:
955 		dev->netdev_ops = &tap_netdev_ops;
956 		/* Ethernet TAP Device */
957 		ether_setup(dev);
958 		dev->priv_flags &= ~IFF_TX_SKB_SHARING;
959 		dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
960 
961 		eth_hw_addr_random(dev);
962 
963 		dev->tx_queue_len = TUN_READQ_SIZE;  /* We prefer our own queue length */
964 		break;
965 	}
966 }
967 
968 /* Character device part */
969 
970 /* Poll */
971 static unsigned int tun_chr_poll(struct file *file, poll_table *wait)
972 {
973 	struct tun_file *tfile = file->private_data;
974 	struct tun_struct *tun = __tun_get(tfile);
975 	struct sock *sk;
976 	unsigned int mask = 0;
977 
978 	if (!tun)
979 		return POLLERR;
980 
981 	sk = tfile->socket.sk;
982 
983 	tun_debug(KERN_INFO, tun, "tun_chr_poll\n");
984 
985 	poll_wait(file, sk_sleep(sk), wait);
986 
987 	if (!skb_queue_empty(&sk->sk_receive_queue))
988 		mask |= POLLIN | POLLRDNORM;
989 
990 	if (sock_writeable(sk) ||
991 	    (!test_and_set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags) &&
992 	     sock_writeable(sk)))
993 		mask |= POLLOUT | POLLWRNORM;
994 
995 	if (tun->dev->reg_state != NETREG_REGISTERED)
996 		mask = POLLERR;
997 
998 	tun_put(tun);
999 	return mask;
1000 }
1001 
1002 /* prepad is the amount to reserve at front.  len is length after that.
1003  * linear is a hint as to how much to copy (usually headers). */
1004 static struct sk_buff *tun_alloc_skb(struct tun_file *tfile,
1005 				     size_t prepad, size_t len,
1006 				     size_t linear, int noblock)
1007 {
1008 	struct sock *sk = tfile->socket.sk;
1009 	struct sk_buff *skb;
1010 	int err;
1011 
1012 	/* Under a page?  Don't bother with paged skb. */
1013 	if (prepad + len < PAGE_SIZE || !linear)
1014 		linear = len;
1015 
1016 	skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
1017 				   &err, 0);
1018 	if (!skb)
1019 		return ERR_PTR(err);
1020 
1021 	skb_reserve(skb, prepad);
1022 	skb_put(skb, linear);
1023 	skb->data_len = len - linear;
1024 	skb->len += len - linear;
1025 
1026 	return skb;
1027 }
1028 
1029 /* Get packet from user space buffer */
1030 static ssize_t tun_get_user(struct tun_struct *tun, struct tun_file *tfile,
1031 			    void *msg_control, struct iov_iter *from,
1032 			    int noblock)
1033 {
1034 	struct tun_pi pi = { 0, cpu_to_be16(ETH_P_IP) };
1035 	struct sk_buff *skb;
1036 	size_t total_len = iov_iter_count(from);
1037 	size_t len = total_len, align = NET_SKB_PAD, linear;
1038 	struct virtio_net_hdr gso = { 0 };
1039 	int good_linear;
1040 	int copylen;
1041 	bool zerocopy = false;
1042 	int err;
1043 	u32 rxhash;
1044 	ssize_t n;
1045 
1046 	if (!(tun->flags & IFF_NO_PI)) {
1047 		if (len < sizeof(pi))
1048 			return -EINVAL;
1049 		len -= sizeof(pi);
1050 
1051 		n = copy_from_iter(&pi, sizeof(pi), from);
1052 		if (n != sizeof(pi))
1053 			return -EFAULT;
1054 	}
1055 
1056 	if (tun->flags & IFF_VNET_HDR) {
1057 		if (len < tun->vnet_hdr_sz)
1058 			return -EINVAL;
1059 		len -= tun->vnet_hdr_sz;
1060 
1061 		n = copy_from_iter(&gso, sizeof(gso), from);
1062 		if (n != sizeof(gso))
1063 			return -EFAULT;
1064 
1065 		if ((gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
1066 		    tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2 > tun16_to_cpu(tun, gso.hdr_len))
1067 			gso.hdr_len = cpu_to_tun16(tun, tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2);
1068 
1069 		if (tun16_to_cpu(tun, gso.hdr_len) > len)
1070 			return -EINVAL;
1071 		iov_iter_advance(from, tun->vnet_hdr_sz - sizeof(gso));
1072 	}
1073 
1074 	if ((tun->flags & TUN_TYPE_MASK) == IFF_TAP) {
1075 		align += NET_IP_ALIGN;
1076 		if (unlikely(len < ETH_HLEN ||
1077 			     (gso.hdr_len && tun16_to_cpu(tun, gso.hdr_len) < ETH_HLEN)))
1078 			return -EINVAL;
1079 	}
1080 
1081 	good_linear = SKB_MAX_HEAD(align);
1082 
1083 	if (msg_control) {
1084 		struct iov_iter i = *from;
1085 
1086 		/* There are 256 bytes to be copied in skb, so there is
1087 		 * enough room for skb expand head in case it is used.
1088 		 * The rest of the buffer is mapped from userspace.
1089 		 */
1090 		copylen = gso.hdr_len ? tun16_to_cpu(tun, gso.hdr_len) : GOODCOPY_LEN;
1091 		if (copylen > good_linear)
1092 			copylen = good_linear;
1093 		linear = copylen;
1094 		iov_iter_advance(&i, copylen);
1095 		if (iov_iter_npages(&i, INT_MAX) <= MAX_SKB_FRAGS)
1096 			zerocopy = true;
1097 	}
1098 
1099 	if (!zerocopy) {
1100 		copylen = len;
1101 		if (tun16_to_cpu(tun, gso.hdr_len) > good_linear)
1102 			linear = good_linear;
1103 		else
1104 			linear = tun16_to_cpu(tun, gso.hdr_len);
1105 	}
1106 
1107 	skb = tun_alloc_skb(tfile, align, copylen, linear, noblock);
1108 	if (IS_ERR(skb)) {
1109 		if (PTR_ERR(skb) != -EAGAIN)
1110 			tun->dev->stats.rx_dropped++;
1111 		return PTR_ERR(skb);
1112 	}
1113 
1114 	if (zerocopy)
1115 		err = zerocopy_sg_from_iter(skb, from);
1116 	else {
1117 		err = skb_copy_datagram_from_iter(skb, 0, from, len);
1118 		if (!err && msg_control) {
1119 			struct ubuf_info *uarg = msg_control;
1120 			uarg->callback(uarg, false);
1121 		}
1122 	}
1123 
1124 	if (err) {
1125 		tun->dev->stats.rx_dropped++;
1126 		kfree_skb(skb);
1127 		return -EFAULT;
1128 	}
1129 
1130 	if (gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
1131 		if (!skb_partial_csum_set(skb, tun16_to_cpu(tun, gso.csum_start),
1132 					  tun16_to_cpu(tun, gso.csum_offset))) {
1133 			tun->dev->stats.rx_frame_errors++;
1134 			kfree_skb(skb);
1135 			return -EINVAL;
1136 		}
1137 	}
1138 
1139 	switch (tun->flags & TUN_TYPE_MASK) {
1140 	case IFF_TUN:
1141 		if (tun->flags & IFF_NO_PI) {
1142 			switch (skb->data[0] & 0xf0) {
1143 			case 0x40:
1144 				pi.proto = htons(ETH_P_IP);
1145 				break;
1146 			case 0x60:
1147 				pi.proto = htons(ETH_P_IPV6);
1148 				break;
1149 			default:
1150 				tun->dev->stats.rx_dropped++;
1151 				kfree_skb(skb);
1152 				return -EINVAL;
1153 			}
1154 		}
1155 
1156 		skb_reset_mac_header(skb);
1157 		skb->protocol = pi.proto;
1158 		skb->dev = tun->dev;
1159 		break;
1160 	case IFF_TAP:
1161 		skb->protocol = eth_type_trans(skb, tun->dev);
1162 		break;
1163 	}
1164 
1165 	if (gso.gso_type != VIRTIO_NET_HDR_GSO_NONE) {
1166 		pr_debug("GSO!\n");
1167 		switch (gso.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
1168 		case VIRTIO_NET_HDR_GSO_TCPV4:
1169 			skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
1170 			break;
1171 		case VIRTIO_NET_HDR_GSO_TCPV6:
1172 			skb_shinfo(skb)->gso_type = SKB_GSO_TCPV6;
1173 			break;
1174 		case VIRTIO_NET_HDR_GSO_UDP:
1175 			skb_shinfo(skb)->gso_type = SKB_GSO_UDP;
1176 			break;
1177 		default:
1178 			tun->dev->stats.rx_frame_errors++;
1179 			kfree_skb(skb);
1180 			return -EINVAL;
1181 		}
1182 
1183 		if (gso.gso_type & VIRTIO_NET_HDR_GSO_ECN)
1184 			skb_shinfo(skb)->gso_type |= SKB_GSO_TCP_ECN;
1185 
1186 		skb_shinfo(skb)->gso_size = tun16_to_cpu(tun, gso.gso_size);
1187 		if (skb_shinfo(skb)->gso_size == 0) {
1188 			tun->dev->stats.rx_frame_errors++;
1189 			kfree_skb(skb);
1190 			return -EINVAL;
1191 		}
1192 
1193 		/* Header must be checked, and gso_segs computed. */
1194 		skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
1195 		skb_shinfo(skb)->gso_segs = 0;
1196 	}
1197 
1198 	/* copy skb_ubuf_info for callback when skb has no error */
1199 	if (zerocopy) {
1200 		skb_shinfo(skb)->destructor_arg = msg_control;
1201 		skb_shinfo(skb)->tx_flags |= SKBTX_DEV_ZEROCOPY;
1202 		skb_shinfo(skb)->tx_flags |= SKBTX_SHARED_FRAG;
1203 	}
1204 
1205 	skb_reset_network_header(skb);
1206 	skb_probe_transport_header(skb, 0);
1207 
1208 	rxhash = skb_get_hash(skb);
1209 	netif_rx_ni(skb);
1210 
1211 	tun->dev->stats.rx_packets++;
1212 	tun->dev->stats.rx_bytes += len;
1213 
1214 	tun_flow_update(tun, rxhash, tfile);
1215 	return total_len;
1216 }
1217 
1218 static ssize_t tun_chr_write_iter(struct kiocb *iocb, struct iov_iter *from)
1219 {
1220 	struct file *file = iocb->ki_filp;
1221 	struct tun_struct *tun = tun_get(file);
1222 	struct tun_file *tfile = file->private_data;
1223 	ssize_t result;
1224 
1225 	if (!tun)
1226 		return -EBADFD;
1227 
1228 	result = tun_get_user(tun, tfile, NULL, from, file->f_flags & O_NONBLOCK);
1229 
1230 	tun_put(tun);
1231 	return result;
1232 }
1233 
1234 /* Put packet to the user space buffer */
1235 static ssize_t tun_put_user(struct tun_struct *tun,
1236 			    struct tun_file *tfile,
1237 			    struct sk_buff *skb,
1238 			    struct iov_iter *iter)
1239 {
1240 	struct tun_pi pi = { 0, skb->protocol };
1241 	ssize_t total;
1242 	int vlan_offset = 0;
1243 	int vlan_hlen = 0;
1244 	int vnet_hdr_sz = 0;
1245 
1246 	if (skb_vlan_tag_present(skb))
1247 		vlan_hlen = VLAN_HLEN;
1248 
1249 	if (tun->flags & IFF_VNET_HDR)
1250 		vnet_hdr_sz = tun->vnet_hdr_sz;
1251 
1252 	total = skb->len + vlan_hlen + vnet_hdr_sz;
1253 
1254 	if (!(tun->flags & IFF_NO_PI)) {
1255 		if (iov_iter_count(iter) < sizeof(pi))
1256 			return -EINVAL;
1257 
1258 		total += sizeof(pi);
1259 		if (iov_iter_count(iter) < total) {
1260 			/* Packet will be striped */
1261 			pi.flags |= TUN_PKT_STRIP;
1262 		}
1263 
1264 		if (copy_to_iter(&pi, sizeof(pi), iter) != sizeof(pi))
1265 			return -EFAULT;
1266 	}
1267 
1268 	if (vnet_hdr_sz) {
1269 		struct virtio_net_hdr gso = { 0 }; /* no info leak */
1270 		if (iov_iter_count(iter) < vnet_hdr_sz)
1271 			return -EINVAL;
1272 
1273 		if (skb_is_gso(skb)) {
1274 			struct skb_shared_info *sinfo = skb_shinfo(skb);
1275 
1276 			/* This is a hint as to how much should be linear. */
1277 			gso.hdr_len = cpu_to_tun16(tun, skb_headlen(skb));
1278 			gso.gso_size = cpu_to_tun16(tun, sinfo->gso_size);
1279 			if (sinfo->gso_type & SKB_GSO_TCPV4)
1280 				gso.gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
1281 			else if (sinfo->gso_type & SKB_GSO_TCPV6)
1282 				gso.gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
1283 			else if (sinfo->gso_type & SKB_GSO_UDP)
1284 				gso.gso_type = VIRTIO_NET_HDR_GSO_UDP;
1285 			else {
1286 				pr_err("unexpected GSO type: "
1287 				       "0x%x, gso_size %d, hdr_len %d\n",
1288 				       sinfo->gso_type, tun16_to_cpu(tun, gso.gso_size),
1289 				       tun16_to_cpu(tun, gso.hdr_len));
1290 				print_hex_dump(KERN_ERR, "tun: ",
1291 					       DUMP_PREFIX_NONE,
1292 					       16, 1, skb->head,
1293 					       min((int)tun16_to_cpu(tun, gso.hdr_len), 64), true);
1294 				WARN_ON_ONCE(1);
1295 				return -EINVAL;
1296 			}
1297 			if (sinfo->gso_type & SKB_GSO_TCP_ECN)
1298 				gso.gso_type |= VIRTIO_NET_HDR_GSO_ECN;
1299 		} else
1300 			gso.gso_type = VIRTIO_NET_HDR_GSO_NONE;
1301 
1302 		if (skb->ip_summed == CHECKSUM_PARTIAL) {
1303 			gso.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
1304 			gso.csum_start = cpu_to_tun16(tun, skb_checksum_start_offset(skb) +
1305 						      vlan_hlen);
1306 			gso.csum_offset = cpu_to_tun16(tun, skb->csum_offset);
1307 		} else if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
1308 			gso.flags = VIRTIO_NET_HDR_F_DATA_VALID;
1309 		} /* else everything is zero */
1310 
1311 		if (copy_to_iter(&gso, sizeof(gso), iter) != sizeof(gso))
1312 			return -EFAULT;
1313 
1314 		iov_iter_advance(iter, vnet_hdr_sz - sizeof(gso));
1315 	}
1316 
1317 	if (vlan_hlen) {
1318 		int ret;
1319 		struct {
1320 			__be16 h_vlan_proto;
1321 			__be16 h_vlan_TCI;
1322 		} veth;
1323 
1324 		veth.h_vlan_proto = skb->vlan_proto;
1325 		veth.h_vlan_TCI = htons(skb_vlan_tag_get(skb));
1326 
1327 		vlan_offset = offsetof(struct vlan_ethhdr, h_vlan_proto);
1328 
1329 		ret = skb_copy_datagram_iter(skb, 0, iter, vlan_offset);
1330 		if (ret || !iov_iter_count(iter))
1331 			goto done;
1332 
1333 		ret = copy_to_iter(&veth, sizeof(veth), iter);
1334 		if (ret != sizeof(veth) || !iov_iter_count(iter))
1335 			goto done;
1336 	}
1337 
1338 	skb_copy_datagram_iter(skb, vlan_offset, iter, skb->len - vlan_offset);
1339 
1340 done:
1341 	tun->dev->stats.tx_packets++;
1342 	tun->dev->stats.tx_bytes += skb->len + vlan_hlen;
1343 
1344 	return total;
1345 }
1346 
1347 static ssize_t tun_do_read(struct tun_struct *tun, struct tun_file *tfile,
1348 			   struct iov_iter *to,
1349 			   int noblock)
1350 {
1351 	struct sk_buff *skb;
1352 	ssize_t ret;
1353 	int peeked, err, off = 0;
1354 
1355 	tun_debug(KERN_INFO, tun, "tun_do_read\n");
1356 
1357 	if (!iov_iter_count(to))
1358 		return 0;
1359 
1360 	if (tun->dev->reg_state != NETREG_REGISTERED)
1361 		return -EIO;
1362 
1363 	/* Read frames from queue */
1364 	skb = __skb_recv_datagram(tfile->socket.sk, noblock ? MSG_DONTWAIT : 0,
1365 				  &peeked, &off, &err);
1366 	if (!skb)
1367 		return err;
1368 
1369 	ret = tun_put_user(tun, tfile, skb, to);
1370 	if (unlikely(ret < 0))
1371 		kfree_skb(skb);
1372 	else
1373 		consume_skb(skb);
1374 
1375 	return ret;
1376 }
1377 
1378 static ssize_t tun_chr_read_iter(struct kiocb *iocb, struct iov_iter *to)
1379 {
1380 	struct file *file = iocb->ki_filp;
1381 	struct tun_file *tfile = file->private_data;
1382 	struct tun_struct *tun = __tun_get(tfile);
1383 	ssize_t len = iov_iter_count(to), ret;
1384 
1385 	if (!tun)
1386 		return -EBADFD;
1387 	ret = tun_do_read(tun, tfile, to, file->f_flags & O_NONBLOCK);
1388 	ret = min_t(ssize_t, ret, len);
1389 	if (ret > 0)
1390 		iocb->ki_pos = ret;
1391 	tun_put(tun);
1392 	return ret;
1393 }
1394 
1395 static void tun_free_netdev(struct net_device *dev)
1396 {
1397 	struct tun_struct *tun = netdev_priv(dev);
1398 
1399 	BUG_ON(!(list_empty(&tun->disabled)));
1400 	tun_flow_uninit(tun);
1401 	security_tun_dev_free_security(tun->security);
1402 	free_netdev(dev);
1403 }
1404 
1405 static void tun_setup(struct net_device *dev)
1406 {
1407 	struct tun_struct *tun = netdev_priv(dev);
1408 
1409 	tun->owner = INVALID_UID;
1410 	tun->group = INVALID_GID;
1411 
1412 	dev->ethtool_ops = &tun_ethtool_ops;
1413 	dev->destructor = tun_free_netdev;
1414 }
1415 
1416 /* Trivial set of netlink ops to allow deleting tun or tap
1417  * device with netlink.
1418  */
1419 static int tun_validate(struct nlattr *tb[], struct nlattr *data[])
1420 {
1421 	return -EINVAL;
1422 }
1423 
1424 static struct rtnl_link_ops tun_link_ops __read_mostly = {
1425 	.kind		= DRV_NAME,
1426 	.priv_size	= sizeof(struct tun_struct),
1427 	.setup		= tun_setup,
1428 	.validate	= tun_validate,
1429 };
1430 
1431 static void tun_sock_write_space(struct sock *sk)
1432 {
1433 	struct tun_file *tfile;
1434 	wait_queue_head_t *wqueue;
1435 
1436 	if (!sock_writeable(sk))
1437 		return;
1438 
1439 	if (!test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags))
1440 		return;
1441 
1442 	wqueue = sk_sleep(sk);
1443 	if (wqueue && waitqueue_active(wqueue))
1444 		wake_up_interruptible_sync_poll(wqueue, POLLOUT |
1445 						POLLWRNORM | POLLWRBAND);
1446 
1447 	tfile = container_of(sk, struct tun_file, sk);
1448 	kill_fasync(&tfile->fasync, SIGIO, POLL_OUT);
1449 }
1450 
1451 static int tun_sendmsg(struct socket *sock, struct msghdr *m, size_t total_len)
1452 {
1453 	int ret;
1454 	struct tun_file *tfile = container_of(sock, struct tun_file, socket);
1455 	struct tun_struct *tun = __tun_get(tfile);
1456 
1457 	if (!tun)
1458 		return -EBADFD;
1459 
1460 	ret = tun_get_user(tun, tfile, m->msg_control, &m->msg_iter,
1461 			   m->msg_flags & MSG_DONTWAIT);
1462 	tun_put(tun);
1463 	return ret;
1464 }
1465 
1466 static int tun_recvmsg(struct socket *sock, struct msghdr *m, size_t total_len,
1467 		       int flags)
1468 {
1469 	struct tun_file *tfile = container_of(sock, struct tun_file, socket);
1470 	struct tun_struct *tun = __tun_get(tfile);
1471 	int ret;
1472 
1473 	if (!tun)
1474 		return -EBADFD;
1475 
1476 	if (flags & ~(MSG_DONTWAIT|MSG_TRUNC|MSG_ERRQUEUE)) {
1477 		ret = -EINVAL;
1478 		goto out;
1479 	}
1480 	if (flags & MSG_ERRQUEUE) {
1481 		ret = sock_recv_errqueue(sock->sk, m, total_len,
1482 					 SOL_PACKET, TUN_TX_TIMESTAMP);
1483 		goto out;
1484 	}
1485 	ret = tun_do_read(tun, tfile, &m->msg_iter, flags & MSG_DONTWAIT);
1486 	if (ret > (ssize_t)total_len) {
1487 		m->msg_flags |= MSG_TRUNC;
1488 		ret = flags & MSG_TRUNC ? ret : total_len;
1489 	}
1490 out:
1491 	tun_put(tun);
1492 	return ret;
1493 }
1494 
1495 static int tun_release(struct socket *sock)
1496 {
1497 	if (sock->sk)
1498 		sock_put(sock->sk);
1499 	return 0;
1500 }
1501 
1502 /* Ops structure to mimic raw sockets with tun */
1503 static const struct proto_ops tun_socket_ops = {
1504 	.sendmsg = tun_sendmsg,
1505 	.recvmsg = tun_recvmsg,
1506 	.release = tun_release,
1507 };
1508 
1509 static struct proto tun_proto = {
1510 	.name		= "tun",
1511 	.owner		= THIS_MODULE,
1512 	.obj_size	= sizeof(struct tun_file),
1513 };
1514 
1515 static int tun_flags(struct tun_struct *tun)
1516 {
1517 	return tun->flags & (TUN_FEATURES | IFF_PERSIST | IFF_TUN | IFF_TAP);
1518 }
1519 
1520 static ssize_t tun_show_flags(struct device *dev, struct device_attribute *attr,
1521 			      char *buf)
1522 {
1523 	struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1524 	return sprintf(buf, "0x%x\n", tun_flags(tun));
1525 }
1526 
1527 static ssize_t tun_show_owner(struct device *dev, struct device_attribute *attr,
1528 			      char *buf)
1529 {
1530 	struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1531 	return uid_valid(tun->owner)?
1532 		sprintf(buf, "%u\n",
1533 			from_kuid_munged(current_user_ns(), tun->owner)):
1534 		sprintf(buf, "-1\n");
1535 }
1536 
1537 static ssize_t tun_show_group(struct device *dev, struct device_attribute *attr,
1538 			      char *buf)
1539 {
1540 	struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1541 	return gid_valid(tun->group) ?
1542 		sprintf(buf, "%u\n",
1543 			from_kgid_munged(current_user_ns(), tun->group)):
1544 		sprintf(buf, "-1\n");
1545 }
1546 
1547 static DEVICE_ATTR(tun_flags, 0444, tun_show_flags, NULL);
1548 static DEVICE_ATTR(owner, 0444, tun_show_owner, NULL);
1549 static DEVICE_ATTR(group, 0444, tun_show_group, NULL);
1550 
1551 static struct attribute *tun_dev_attrs[] = {
1552 	&dev_attr_tun_flags.attr,
1553 	&dev_attr_owner.attr,
1554 	&dev_attr_group.attr,
1555 	NULL
1556 };
1557 
1558 static const struct attribute_group tun_attr_group = {
1559 	.attrs = tun_dev_attrs
1560 };
1561 
1562 static int tun_set_iff(struct net *net, struct file *file, struct ifreq *ifr)
1563 {
1564 	struct tun_struct *tun;
1565 	struct tun_file *tfile = file->private_data;
1566 	struct net_device *dev;
1567 	int err;
1568 
1569 	if (tfile->detached)
1570 		return -EINVAL;
1571 
1572 	dev = __dev_get_by_name(net, ifr->ifr_name);
1573 	if (dev) {
1574 		if (ifr->ifr_flags & IFF_TUN_EXCL)
1575 			return -EBUSY;
1576 		if ((ifr->ifr_flags & IFF_TUN) && dev->netdev_ops == &tun_netdev_ops)
1577 			tun = netdev_priv(dev);
1578 		else if ((ifr->ifr_flags & IFF_TAP) && dev->netdev_ops == &tap_netdev_ops)
1579 			tun = netdev_priv(dev);
1580 		else
1581 			return -EINVAL;
1582 
1583 		if (!!(ifr->ifr_flags & IFF_MULTI_QUEUE) !=
1584 		    !!(tun->flags & IFF_MULTI_QUEUE))
1585 			return -EINVAL;
1586 
1587 		if (tun_not_capable(tun))
1588 			return -EPERM;
1589 		err = security_tun_dev_open(tun->security);
1590 		if (err < 0)
1591 			return err;
1592 
1593 		err = tun_attach(tun, file, ifr->ifr_flags & IFF_NOFILTER);
1594 		if (err < 0)
1595 			return err;
1596 
1597 		if (tun->flags & IFF_MULTI_QUEUE &&
1598 		    (tun->numqueues + tun->numdisabled > 1)) {
1599 			/* One or more queue has already been attached, no need
1600 			 * to initialize the device again.
1601 			 */
1602 			return 0;
1603 		}
1604 	}
1605 	else {
1606 		char *name;
1607 		unsigned long flags = 0;
1608 		int queues = ifr->ifr_flags & IFF_MULTI_QUEUE ?
1609 			     MAX_TAP_QUEUES : 1;
1610 
1611 		if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
1612 			return -EPERM;
1613 		err = security_tun_dev_create();
1614 		if (err < 0)
1615 			return err;
1616 
1617 		/* Set dev type */
1618 		if (ifr->ifr_flags & IFF_TUN) {
1619 			/* TUN device */
1620 			flags |= IFF_TUN;
1621 			name = "tun%d";
1622 		} else if (ifr->ifr_flags & IFF_TAP) {
1623 			/* TAP device */
1624 			flags |= IFF_TAP;
1625 			name = "tap%d";
1626 		} else
1627 			return -EINVAL;
1628 
1629 		if (*ifr->ifr_name)
1630 			name = ifr->ifr_name;
1631 
1632 		dev = alloc_netdev_mqs(sizeof(struct tun_struct), name,
1633 				       NET_NAME_UNKNOWN, tun_setup, queues,
1634 				       queues);
1635 
1636 		if (!dev)
1637 			return -ENOMEM;
1638 
1639 		dev_net_set(dev, net);
1640 		dev->rtnl_link_ops = &tun_link_ops;
1641 		dev->ifindex = tfile->ifindex;
1642 		dev->sysfs_groups[0] = &tun_attr_group;
1643 
1644 		tun = netdev_priv(dev);
1645 		tun->dev = dev;
1646 		tun->flags = flags;
1647 		tun->txflt.count = 0;
1648 		tun->vnet_hdr_sz = sizeof(struct virtio_net_hdr);
1649 
1650 		tun->filter_attached = false;
1651 		tun->sndbuf = tfile->socket.sk->sk_sndbuf;
1652 
1653 		spin_lock_init(&tun->lock);
1654 
1655 		err = security_tun_dev_alloc_security(&tun->security);
1656 		if (err < 0)
1657 			goto err_free_dev;
1658 
1659 		tun_net_init(dev);
1660 		tun_flow_init(tun);
1661 
1662 		dev->hw_features = NETIF_F_SG | NETIF_F_FRAGLIST |
1663 				   TUN_USER_FEATURES | NETIF_F_HW_VLAN_CTAG_TX |
1664 				   NETIF_F_HW_VLAN_STAG_TX;
1665 		dev->features = dev->hw_features;
1666 		dev->vlan_features = dev->features &
1667 				     ~(NETIF_F_HW_VLAN_CTAG_TX |
1668 				       NETIF_F_HW_VLAN_STAG_TX);
1669 
1670 		INIT_LIST_HEAD(&tun->disabled);
1671 		err = tun_attach(tun, file, false);
1672 		if (err < 0)
1673 			goto err_free_flow;
1674 
1675 		err = register_netdevice(tun->dev);
1676 		if (err < 0)
1677 			goto err_detach;
1678 	}
1679 
1680 	netif_carrier_on(tun->dev);
1681 
1682 	tun_debug(KERN_INFO, tun, "tun_set_iff\n");
1683 
1684 	tun->flags = (tun->flags & ~TUN_FEATURES) |
1685 		(ifr->ifr_flags & TUN_FEATURES);
1686 
1687 	/* Make sure persistent devices do not get stuck in
1688 	 * xoff state.
1689 	 */
1690 	if (netif_running(tun->dev))
1691 		netif_tx_wake_all_queues(tun->dev);
1692 
1693 	strcpy(ifr->ifr_name, tun->dev->name);
1694 	return 0;
1695 
1696 err_detach:
1697 	tun_detach_all(dev);
1698 err_free_flow:
1699 	tun_flow_uninit(tun);
1700 	security_tun_dev_free_security(tun->security);
1701 err_free_dev:
1702 	free_netdev(dev);
1703 	return err;
1704 }
1705 
1706 static void tun_get_iff(struct net *net, struct tun_struct *tun,
1707 		       struct ifreq *ifr)
1708 {
1709 	tun_debug(KERN_INFO, tun, "tun_get_iff\n");
1710 
1711 	strcpy(ifr->ifr_name, tun->dev->name);
1712 
1713 	ifr->ifr_flags = tun_flags(tun);
1714 
1715 }
1716 
1717 /* This is like a cut-down ethtool ops, except done via tun fd so no
1718  * privs required. */
1719 static int set_offload(struct tun_struct *tun, unsigned long arg)
1720 {
1721 	netdev_features_t features = 0;
1722 
1723 	if (arg & TUN_F_CSUM) {
1724 		features |= NETIF_F_HW_CSUM;
1725 		arg &= ~TUN_F_CSUM;
1726 
1727 		if (arg & (TUN_F_TSO4|TUN_F_TSO6)) {
1728 			if (arg & TUN_F_TSO_ECN) {
1729 				features |= NETIF_F_TSO_ECN;
1730 				arg &= ~TUN_F_TSO_ECN;
1731 			}
1732 			if (arg & TUN_F_TSO4)
1733 				features |= NETIF_F_TSO;
1734 			if (arg & TUN_F_TSO6)
1735 				features |= NETIF_F_TSO6;
1736 			arg &= ~(TUN_F_TSO4|TUN_F_TSO6);
1737 		}
1738 
1739 		if (arg & TUN_F_UFO) {
1740 			features |= NETIF_F_UFO;
1741 			arg &= ~TUN_F_UFO;
1742 		}
1743 	}
1744 
1745 	/* This gives the user a way to test for new features in future by
1746 	 * trying to set them. */
1747 	if (arg)
1748 		return -EINVAL;
1749 
1750 	tun->set_features = features;
1751 	netdev_update_features(tun->dev);
1752 
1753 	return 0;
1754 }
1755 
1756 static void tun_detach_filter(struct tun_struct *tun, int n)
1757 {
1758 	int i;
1759 	struct tun_file *tfile;
1760 
1761 	for (i = 0; i < n; i++) {
1762 		tfile = rtnl_dereference(tun->tfiles[i]);
1763 		sk_detach_filter(tfile->socket.sk);
1764 	}
1765 
1766 	tun->filter_attached = false;
1767 }
1768 
1769 static int tun_attach_filter(struct tun_struct *tun)
1770 {
1771 	int i, ret = 0;
1772 	struct tun_file *tfile;
1773 
1774 	for (i = 0; i < tun->numqueues; i++) {
1775 		tfile = rtnl_dereference(tun->tfiles[i]);
1776 		ret = sk_attach_filter(&tun->fprog, tfile->socket.sk);
1777 		if (ret) {
1778 			tun_detach_filter(tun, i);
1779 			return ret;
1780 		}
1781 	}
1782 
1783 	tun->filter_attached = true;
1784 	return ret;
1785 }
1786 
1787 static void tun_set_sndbuf(struct tun_struct *tun)
1788 {
1789 	struct tun_file *tfile;
1790 	int i;
1791 
1792 	for (i = 0; i < tun->numqueues; i++) {
1793 		tfile = rtnl_dereference(tun->tfiles[i]);
1794 		tfile->socket.sk->sk_sndbuf = tun->sndbuf;
1795 	}
1796 }
1797 
1798 static int tun_set_queue(struct file *file, struct ifreq *ifr)
1799 {
1800 	struct tun_file *tfile = file->private_data;
1801 	struct tun_struct *tun;
1802 	int ret = 0;
1803 
1804 	rtnl_lock();
1805 
1806 	if (ifr->ifr_flags & IFF_ATTACH_QUEUE) {
1807 		tun = tfile->detached;
1808 		if (!tun) {
1809 			ret = -EINVAL;
1810 			goto unlock;
1811 		}
1812 		ret = security_tun_dev_attach_queue(tun->security);
1813 		if (ret < 0)
1814 			goto unlock;
1815 		ret = tun_attach(tun, file, false);
1816 	} else if (ifr->ifr_flags & IFF_DETACH_QUEUE) {
1817 		tun = rtnl_dereference(tfile->tun);
1818 		if (!tun || !(tun->flags & IFF_MULTI_QUEUE) || tfile->detached)
1819 			ret = -EINVAL;
1820 		else
1821 			__tun_detach(tfile, false);
1822 	} else
1823 		ret = -EINVAL;
1824 
1825 unlock:
1826 	rtnl_unlock();
1827 	return ret;
1828 }
1829 
1830 static long __tun_chr_ioctl(struct file *file, unsigned int cmd,
1831 			    unsigned long arg, int ifreq_len)
1832 {
1833 	struct tun_file *tfile = file->private_data;
1834 	struct tun_struct *tun;
1835 	void __user* argp = (void __user*)arg;
1836 	struct ifreq ifr;
1837 	kuid_t owner;
1838 	kgid_t group;
1839 	int sndbuf;
1840 	int vnet_hdr_sz;
1841 	unsigned int ifindex;
1842 	int le;
1843 	int ret;
1844 
1845 	if (cmd == TUNSETIFF || cmd == TUNSETQUEUE || _IOC_TYPE(cmd) == 0x89) {
1846 		if (copy_from_user(&ifr, argp, ifreq_len))
1847 			return -EFAULT;
1848 	} else {
1849 		memset(&ifr, 0, sizeof(ifr));
1850 	}
1851 	if (cmd == TUNGETFEATURES) {
1852 		/* Currently this just means: "what IFF flags are valid?".
1853 		 * This is needed because we never checked for invalid flags on
1854 		 * TUNSETIFF.
1855 		 */
1856 		return put_user(IFF_TUN | IFF_TAP | TUN_FEATURES,
1857 				(unsigned int __user*)argp);
1858 	} else if (cmd == TUNSETQUEUE)
1859 		return tun_set_queue(file, &ifr);
1860 
1861 	ret = 0;
1862 	rtnl_lock();
1863 
1864 	tun = __tun_get(tfile);
1865 	if (cmd == TUNSETIFF && !tun) {
1866 		ifr.ifr_name[IFNAMSIZ-1] = '\0';
1867 
1868 		ret = tun_set_iff(tfile->net, file, &ifr);
1869 
1870 		if (ret)
1871 			goto unlock;
1872 
1873 		if (copy_to_user(argp, &ifr, ifreq_len))
1874 			ret = -EFAULT;
1875 		goto unlock;
1876 	}
1877 	if (cmd == TUNSETIFINDEX) {
1878 		ret = -EPERM;
1879 		if (tun)
1880 			goto unlock;
1881 
1882 		ret = -EFAULT;
1883 		if (copy_from_user(&ifindex, argp, sizeof(ifindex)))
1884 			goto unlock;
1885 
1886 		ret = 0;
1887 		tfile->ifindex = ifindex;
1888 		goto unlock;
1889 	}
1890 
1891 	ret = -EBADFD;
1892 	if (!tun)
1893 		goto unlock;
1894 
1895 	tun_debug(KERN_INFO, tun, "tun_chr_ioctl cmd %u\n", cmd);
1896 
1897 	ret = 0;
1898 	switch (cmd) {
1899 	case TUNGETIFF:
1900 		tun_get_iff(current->nsproxy->net_ns, tun, &ifr);
1901 
1902 		if (tfile->detached)
1903 			ifr.ifr_flags |= IFF_DETACH_QUEUE;
1904 		if (!tfile->socket.sk->sk_filter)
1905 			ifr.ifr_flags |= IFF_NOFILTER;
1906 
1907 		if (copy_to_user(argp, &ifr, ifreq_len))
1908 			ret = -EFAULT;
1909 		break;
1910 
1911 	case TUNSETNOCSUM:
1912 		/* Disable/Enable checksum */
1913 
1914 		/* [unimplemented] */
1915 		tun_debug(KERN_INFO, tun, "ignored: set checksum %s\n",
1916 			  arg ? "disabled" : "enabled");
1917 		break;
1918 
1919 	case TUNSETPERSIST:
1920 		/* Disable/Enable persist mode. Keep an extra reference to the
1921 		 * module to prevent the module being unprobed.
1922 		 */
1923 		if (arg && !(tun->flags & IFF_PERSIST)) {
1924 			tun->flags |= IFF_PERSIST;
1925 			__module_get(THIS_MODULE);
1926 		}
1927 		if (!arg && (tun->flags & IFF_PERSIST)) {
1928 			tun->flags &= ~IFF_PERSIST;
1929 			module_put(THIS_MODULE);
1930 		}
1931 
1932 		tun_debug(KERN_INFO, tun, "persist %s\n",
1933 			  arg ? "enabled" : "disabled");
1934 		break;
1935 
1936 	case TUNSETOWNER:
1937 		/* Set owner of the device */
1938 		owner = make_kuid(current_user_ns(), arg);
1939 		if (!uid_valid(owner)) {
1940 			ret = -EINVAL;
1941 			break;
1942 		}
1943 		tun->owner = owner;
1944 		tun_debug(KERN_INFO, tun, "owner set to %u\n",
1945 			  from_kuid(&init_user_ns, tun->owner));
1946 		break;
1947 
1948 	case TUNSETGROUP:
1949 		/* Set group of the device */
1950 		group = make_kgid(current_user_ns(), arg);
1951 		if (!gid_valid(group)) {
1952 			ret = -EINVAL;
1953 			break;
1954 		}
1955 		tun->group = group;
1956 		tun_debug(KERN_INFO, tun, "group set to %u\n",
1957 			  from_kgid(&init_user_ns, tun->group));
1958 		break;
1959 
1960 	case TUNSETLINK:
1961 		/* Only allow setting the type when the interface is down */
1962 		if (tun->dev->flags & IFF_UP) {
1963 			tun_debug(KERN_INFO, tun,
1964 				  "Linktype set failed because interface is up\n");
1965 			ret = -EBUSY;
1966 		} else {
1967 			tun->dev->type = (int) arg;
1968 			tun_debug(KERN_INFO, tun, "linktype set to %d\n",
1969 				  tun->dev->type);
1970 			ret = 0;
1971 		}
1972 		break;
1973 
1974 #ifdef TUN_DEBUG
1975 	case TUNSETDEBUG:
1976 		tun->debug = arg;
1977 		break;
1978 #endif
1979 	case TUNSETOFFLOAD:
1980 		ret = set_offload(tun, arg);
1981 		break;
1982 
1983 	case TUNSETTXFILTER:
1984 		/* Can be set only for TAPs */
1985 		ret = -EINVAL;
1986 		if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
1987 			break;
1988 		ret = update_filter(&tun->txflt, (void __user *)arg);
1989 		break;
1990 
1991 	case SIOCGIFHWADDR:
1992 		/* Get hw address */
1993 		memcpy(ifr.ifr_hwaddr.sa_data, tun->dev->dev_addr, ETH_ALEN);
1994 		ifr.ifr_hwaddr.sa_family = tun->dev->type;
1995 		if (copy_to_user(argp, &ifr, ifreq_len))
1996 			ret = -EFAULT;
1997 		break;
1998 
1999 	case SIOCSIFHWADDR:
2000 		/* Set hw address */
2001 		tun_debug(KERN_DEBUG, tun, "set hw address: %pM\n",
2002 			  ifr.ifr_hwaddr.sa_data);
2003 
2004 		ret = dev_set_mac_address(tun->dev, &ifr.ifr_hwaddr);
2005 		break;
2006 
2007 	case TUNGETSNDBUF:
2008 		sndbuf = tfile->socket.sk->sk_sndbuf;
2009 		if (copy_to_user(argp, &sndbuf, sizeof(sndbuf)))
2010 			ret = -EFAULT;
2011 		break;
2012 
2013 	case TUNSETSNDBUF:
2014 		if (copy_from_user(&sndbuf, argp, sizeof(sndbuf))) {
2015 			ret = -EFAULT;
2016 			break;
2017 		}
2018 
2019 		tun->sndbuf = sndbuf;
2020 		tun_set_sndbuf(tun);
2021 		break;
2022 
2023 	case TUNGETVNETHDRSZ:
2024 		vnet_hdr_sz = tun->vnet_hdr_sz;
2025 		if (copy_to_user(argp, &vnet_hdr_sz, sizeof(vnet_hdr_sz)))
2026 			ret = -EFAULT;
2027 		break;
2028 
2029 	case TUNSETVNETHDRSZ:
2030 		if (copy_from_user(&vnet_hdr_sz, argp, sizeof(vnet_hdr_sz))) {
2031 			ret = -EFAULT;
2032 			break;
2033 		}
2034 		if (vnet_hdr_sz < (int)sizeof(struct virtio_net_hdr)) {
2035 			ret = -EINVAL;
2036 			break;
2037 		}
2038 
2039 		tun->vnet_hdr_sz = vnet_hdr_sz;
2040 		break;
2041 
2042 	case TUNGETVNETLE:
2043 		le = !!(tun->flags & TUN_VNET_LE);
2044 		if (put_user(le, (int __user *)argp))
2045 			ret = -EFAULT;
2046 		break;
2047 
2048 	case TUNSETVNETLE:
2049 		if (get_user(le, (int __user *)argp)) {
2050 			ret = -EFAULT;
2051 			break;
2052 		}
2053 		if (le)
2054 			tun->flags |= TUN_VNET_LE;
2055 		else
2056 			tun->flags &= ~TUN_VNET_LE;
2057 		break;
2058 
2059 	case TUNATTACHFILTER:
2060 		/* Can be set only for TAPs */
2061 		ret = -EINVAL;
2062 		if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
2063 			break;
2064 		ret = -EFAULT;
2065 		if (copy_from_user(&tun->fprog, argp, sizeof(tun->fprog)))
2066 			break;
2067 
2068 		ret = tun_attach_filter(tun);
2069 		break;
2070 
2071 	case TUNDETACHFILTER:
2072 		/* Can be set only for TAPs */
2073 		ret = -EINVAL;
2074 		if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
2075 			break;
2076 		ret = 0;
2077 		tun_detach_filter(tun, tun->numqueues);
2078 		break;
2079 
2080 	case TUNGETFILTER:
2081 		ret = -EINVAL;
2082 		if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
2083 			break;
2084 		ret = -EFAULT;
2085 		if (copy_to_user(argp, &tun->fprog, sizeof(tun->fprog)))
2086 			break;
2087 		ret = 0;
2088 		break;
2089 
2090 	default:
2091 		ret = -EINVAL;
2092 		break;
2093 	}
2094 
2095 unlock:
2096 	rtnl_unlock();
2097 	if (tun)
2098 		tun_put(tun);
2099 	return ret;
2100 }
2101 
2102 static long tun_chr_ioctl(struct file *file,
2103 			  unsigned int cmd, unsigned long arg)
2104 {
2105 	return __tun_chr_ioctl(file, cmd, arg, sizeof (struct ifreq));
2106 }
2107 
2108 #ifdef CONFIG_COMPAT
2109 static long tun_chr_compat_ioctl(struct file *file,
2110 			 unsigned int cmd, unsigned long arg)
2111 {
2112 	switch (cmd) {
2113 	case TUNSETIFF:
2114 	case TUNGETIFF:
2115 	case TUNSETTXFILTER:
2116 	case TUNGETSNDBUF:
2117 	case TUNSETSNDBUF:
2118 	case SIOCGIFHWADDR:
2119 	case SIOCSIFHWADDR:
2120 		arg = (unsigned long)compat_ptr(arg);
2121 		break;
2122 	default:
2123 		arg = (compat_ulong_t)arg;
2124 		break;
2125 	}
2126 
2127 	/*
2128 	 * compat_ifreq is shorter than ifreq, so we must not access beyond
2129 	 * the end of that structure. All fields that are used in this
2130 	 * driver are compatible though, we don't need to convert the
2131 	 * contents.
2132 	 */
2133 	return __tun_chr_ioctl(file, cmd, arg, sizeof(struct compat_ifreq));
2134 }
2135 #endif /* CONFIG_COMPAT */
2136 
2137 static int tun_chr_fasync(int fd, struct file *file, int on)
2138 {
2139 	struct tun_file *tfile = file->private_data;
2140 	int ret;
2141 
2142 	if ((ret = fasync_helper(fd, file, on, &tfile->fasync)) < 0)
2143 		goto out;
2144 
2145 	if (on) {
2146 		__f_setown(file, task_pid(current), PIDTYPE_PID, 0);
2147 		tfile->flags |= TUN_FASYNC;
2148 	} else
2149 		tfile->flags &= ~TUN_FASYNC;
2150 	ret = 0;
2151 out:
2152 	return ret;
2153 }
2154 
2155 static int tun_chr_open(struct inode *inode, struct file * file)
2156 {
2157 	struct tun_file *tfile;
2158 
2159 	DBG1(KERN_INFO, "tunX: tun_chr_open\n");
2160 
2161 	tfile = (struct tun_file *)sk_alloc(&init_net, AF_UNSPEC, GFP_KERNEL,
2162 					    &tun_proto);
2163 	if (!tfile)
2164 		return -ENOMEM;
2165 	RCU_INIT_POINTER(tfile->tun, NULL);
2166 	tfile->net = get_net(current->nsproxy->net_ns);
2167 	tfile->flags = 0;
2168 	tfile->ifindex = 0;
2169 
2170 	init_waitqueue_head(&tfile->wq.wait);
2171 	RCU_INIT_POINTER(tfile->socket.wq, &tfile->wq);
2172 
2173 	tfile->socket.file = file;
2174 	tfile->socket.ops = &tun_socket_ops;
2175 
2176 	sock_init_data(&tfile->socket, &tfile->sk);
2177 	sk_change_net(&tfile->sk, tfile->net);
2178 
2179 	tfile->sk.sk_write_space = tun_sock_write_space;
2180 	tfile->sk.sk_sndbuf = INT_MAX;
2181 
2182 	file->private_data = tfile;
2183 	set_bit(SOCK_EXTERNALLY_ALLOCATED, &tfile->socket.flags);
2184 	INIT_LIST_HEAD(&tfile->next);
2185 
2186 	sock_set_flag(&tfile->sk, SOCK_ZEROCOPY);
2187 
2188 	return 0;
2189 }
2190 
2191 static int tun_chr_close(struct inode *inode, struct file *file)
2192 {
2193 	struct tun_file *tfile = file->private_data;
2194 	struct net *net = tfile->net;
2195 
2196 	tun_detach(tfile, true);
2197 	put_net(net);
2198 
2199 	return 0;
2200 }
2201 
2202 #ifdef CONFIG_PROC_FS
2203 static void tun_chr_show_fdinfo(struct seq_file *m, struct file *f)
2204 {
2205 	struct tun_struct *tun;
2206 	struct ifreq ifr;
2207 
2208 	memset(&ifr, 0, sizeof(ifr));
2209 
2210 	rtnl_lock();
2211 	tun = tun_get(f);
2212 	if (tun)
2213 		tun_get_iff(current->nsproxy->net_ns, tun, &ifr);
2214 	rtnl_unlock();
2215 
2216 	if (tun)
2217 		tun_put(tun);
2218 
2219 	seq_printf(m, "iff:\t%s\n", ifr.ifr_name);
2220 }
2221 #endif
2222 
2223 static const struct file_operations tun_fops = {
2224 	.owner	= THIS_MODULE,
2225 	.llseek = no_llseek,
2226 	.read_iter  = tun_chr_read_iter,
2227 	.write_iter = tun_chr_write_iter,
2228 	.poll	= tun_chr_poll,
2229 	.unlocked_ioctl	= tun_chr_ioctl,
2230 #ifdef CONFIG_COMPAT
2231 	.compat_ioctl = tun_chr_compat_ioctl,
2232 #endif
2233 	.open	= tun_chr_open,
2234 	.release = tun_chr_close,
2235 	.fasync = tun_chr_fasync,
2236 #ifdef CONFIG_PROC_FS
2237 	.show_fdinfo = tun_chr_show_fdinfo,
2238 #endif
2239 };
2240 
2241 static struct miscdevice tun_miscdev = {
2242 	.minor = TUN_MINOR,
2243 	.name = "tun",
2244 	.nodename = "net/tun",
2245 	.fops = &tun_fops,
2246 };
2247 
2248 /* ethtool interface */
2249 
2250 static int tun_get_settings(struct net_device *dev, struct ethtool_cmd *cmd)
2251 {
2252 	cmd->supported		= 0;
2253 	cmd->advertising	= 0;
2254 	ethtool_cmd_speed_set(cmd, SPEED_10);
2255 	cmd->duplex		= DUPLEX_FULL;
2256 	cmd->port		= PORT_TP;
2257 	cmd->phy_address	= 0;
2258 	cmd->transceiver	= XCVR_INTERNAL;
2259 	cmd->autoneg		= AUTONEG_DISABLE;
2260 	cmd->maxtxpkt		= 0;
2261 	cmd->maxrxpkt		= 0;
2262 	return 0;
2263 }
2264 
2265 static void tun_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
2266 {
2267 	struct tun_struct *tun = netdev_priv(dev);
2268 
2269 	strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
2270 	strlcpy(info->version, DRV_VERSION, sizeof(info->version));
2271 
2272 	switch (tun->flags & TUN_TYPE_MASK) {
2273 	case IFF_TUN:
2274 		strlcpy(info->bus_info, "tun", sizeof(info->bus_info));
2275 		break;
2276 	case IFF_TAP:
2277 		strlcpy(info->bus_info, "tap", sizeof(info->bus_info));
2278 		break;
2279 	}
2280 }
2281 
2282 static u32 tun_get_msglevel(struct net_device *dev)
2283 {
2284 #ifdef TUN_DEBUG
2285 	struct tun_struct *tun = netdev_priv(dev);
2286 	return tun->debug;
2287 #else
2288 	return -EOPNOTSUPP;
2289 #endif
2290 }
2291 
2292 static void tun_set_msglevel(struct net_device *dev, u32 value)
2293 {
2294 #ifdef TUN_DEBUG
2295 	struct tun_struct *tun = netdev_priv(dev);
2296 	tun->debug = value;
2297 #endif
2298 }
2299 
2300 static const struct ethtool_ops tun_ethtool_ops = {
2301 	.get_settings	= tun_get_settings,
2302 	.get_drvinfo	= tun_get_drvinfo,
2303 	.get_msglevel	= tun_get_msglevel,
2304 	.set_msglevel	= tun_set_msglevel,
2305 	.get_link	= ethtool_op_get_link,
2306 	.get_ts_info	= ethtool_op_get_ts_info,
2307 };
2308 
2309 
2310 static int __init tun_init(void)
2311 {
2312 	int ret = 0;
2313 
2314 	pr_info("%s, %s\n", DRV_DESCRIPTION, DRV_VERSION);
2315 	pr_info("%s\n", DRV_COPYRIGHT);
2316 
2317 	ret = rtnl_link_register(&tun_link_ops);
2318 	if (ret) {
2319 		pr_err("Can't register link_ops\n");
2320 		goto err_linkops;
2321 	}
2322 
2323 	ret = misc_register(&tun_miscdev);
2324 	if (ret) {
2325 		pr_err("Can't register misc device %d\n", TUN_MINOR);
2326 		goto err_misc;
2327 	}
2328 	return  0;
2329 err_misc:
2330 	rtnl_link_unregister(&tun_link_ops);
2331 err_linkops:
2332 	return ret;
2333 }
2334 
2335 static void tun_cleanup(void)
2336 {
2337 	misc_deregister(&tun_miscdev);
2338 	rtnl_link_unregister(&tun_link_ops);
2339 }
2340 
2341 /* Get an underlying socket object from tun file.  Returns error unless file is
2342  * attached to a device.  The returned object works like a packet socket, it
2343  * can be used for sock_sendmsg/sock_recvmsg.  The caller is responsible for
2344  * holding a reference to the file for as long as the socket is in use. */
2345 struct socket *tun_get_socket(struct file *file)
2346 {
2347 	struct tun_file *tfile;
2348 	if (file->f_op != &tun_fops)
2349 		return ERR_PTR(-EINVAL);
2350 	tfile = file->private_data;
2351 	if (!tfile)
2352 		return ERR_PTR(-EBADFD);
2353 	return &tfile->socket;
2354 }
2355 EXPORT_SYMBOL_GPL(tun_get_socket);
2356 
2357 module_init(tun_init);
2358 module_exit(tun_cleanup);
2359 MODULE_DESCRIPTION(DRV_DESCRIPTION);
2360 MODULE_AUTHOR(DRV_COPYRIGHT);
2361 MODULE_LICENSE("GPL");
2362 MODULE_ALIAS_MISCDEV(TUN_MINOR);
2363 MODULE_ALIAS("devname:net/tun");
2364