xref: /openbmc/linux/drivers/net/tun.c (revision a8f9bfdf982e2b1fb9f094e4de9ab08c57f3d2fd)
1 /*
2  *  TUN - Universal TUN/TAP device driver.
3  *  Copyright (C) 1999-2002 Maxim Krasnyansky <maxk@qualcomm.com>
4  *
5  *  This program is free software; you can redistribute it and/or modify
6  *  it under the terms of the GNU General Public License as published by
7  *  the Free Software Foundation; either version 2 of the License, or
8  *  (at your option) any later version.
9  *
10  *  This program is distributed in the hope that it will be useful,
11  *  but WITHOUT ANY WARRANTY; without even the implied warranty of
12  *  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
13  *  GNU General Public License for more details.
14  *
15  *  $Id: tun.c,v 1.15 2002/03/01 02:44:24 maxk Exp $
16  */
17 
18 /*
19  *  Changes:
20  *
21  *  Mike Kershaw <dragorn@kismetwireless.net> 2005/08/14
22  *    Add TUNSETLINK ioctl to set the link encapsulation
23  *
24  *  Mark Smith <markzzzsmith@yahoo.com.au>
25  *    Use eth_random_addr() for tap MAC address.
26  *
27  *  Harald Roelle <harald.roelle@ifi.lmu.de>  2004/04/20
28  *    Fixes in packet dropping, queue length setting and queue wakeup.
29  *    Increased default tx queue length.
30  *    Added ethtool API.
31  *    Minor cleanups
32  *
33  *  Daniel Podlejski <underley@underley.eu.org>
34  *    Modifications for 2.3.99-pre5 kernel.
35  */
36 
37 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
38 
39 #define DRV_NAME	"tun"
40 #define DRV_VERSION	"1.6"
41 #define DRV_DESCRIPTION	"Universal TUN/TAP device driver"
42 #define DRV_COPYRIGHT	"(C) 1999-2004 Max Krasnyansky <maxk@qualcomm.com>"
43 
44 #include <linux/module.h>
45 #include <linux/errno.h>
46 #include <linux/kernel.h>
47 #include <linux/major.h>
48 #include <linux/slab.h>
49 #include <linux/poll.h>
50 #include <linux/fcntl.h>
51 #include <linux/init.h>
52 #include <linux/skbuff.h>
53 #include <linux/netdevice.h>
54 #include <linux/etherdevice.h>
55 #include <linux/miscdevice.h>
56 #include <linux/ethtool.h>
57 #include <linux/rtnetlink.h>
58 #include <linux/compat.h>
59 #include <linux/if.h>
60 #include <linux/if_arp.h>
61 #include <linux/if_ether.h>
62 #include <linux/if_tun.h>
63 #include <linux/if_vlan.h>
64 #include <linux/crc32.h>
65 #include <linux/nsproxy.h>
66 #include <linux/virtio_net.h>
67 #include <linux/rcupdate.h>
68 #include <net/ipv6.h>
69 #include <net/net_namespace.h>
70 #include <net/netns/generic.h>
71 #include <net/rtnetlink.h>
72 #include <net/sock.h>
73 #include <linux/seq_file.h>
74 
75 #include <asm/uaccess.h>
76 
77 /* Uncomment to enable debugging */
78 /* #define TUN_DEBUG 1 */
79 
80 #ifdef TUN_DEBUG
81 static int debug;
82 
83 #define tun_debug(level, tun, fmt, args...)			\
84 do {								\
85 	if (tun->debug)						\
86 		netdev_printk(level, tun->dev, fmt, ##args);	\
87 } while (0)
88 #define DBG1(level, fmt, args...)				\
89 do {								\
90 	if (debug == 2)						\
91 		printk(level fmt, ##args);			\
92 } while (0)
93 #else
94 #define tun_debug(level, tun, fmt, args...)			\
95 do {								\
96 	if (0)							\
97 		netdev_printk(level, tun->dev, fmt, ##args);	\
98 } while (0)
99 #define DBG1(level, fmt, args...)				\
100 do {								\
101 	if (0)							\
102 		printk(level fmt, ##args);			\
103 } while (0)
104 #endif
105 
106 #define GOODCOPY_LEN 128
107 
108 #define FLT_EXACT_COUNT 8
109 struct tap_filter {
110 	unsigned int    count;    /* Number of addrs. Zero means disabled */
111 	u32             mask[2];  /* Mask of the hashed addrs */
112 	unsigned char	addr[FLT_EXACT_COUNT][ETH_ALEN];
113 };
114 
115 /* DEFAULT_MAX_NUM_RSS_QUEUES were chosen to let the rx/tx queues allocated for
116  * the netdevice to be fit in one page. So we can make sure the success of
117  * memory allocation. TODO: increase the limit. */
118 #define MAX_TAP_QUEUES DEFAULT_MAX_NUM_RSS_QUEUES
119 #define MAX_TAP_FLOWS  4096
120 
121 #define TUN_FLOW_EXPIRE (3 * HZ)
122 
123 /* A tun_file connects an open character device to a tuntap netdevice. It
124  * also contains all socket related structures (except sock_fprog and tap_filter)
125  * to serve as one transmit queue for tuntap device. The sock_fprog and
126  * tap_filter were kept in tun_struct since they were used for filtering for the
127  * netdevice not for a specific queue (at least I didn't see the requirement for
128  * this).
129  *
130  * RCU usage:
131  * The tun_file and tun_struct are loosely coupled, the pointer from one to the
132  * other can only be read while rcu_read_lock or rtnl_lock is held.
133  */
134 struct tun_file {
135 	struct sock sk;
136 	struct socket socket;
137 	struct socket_wq wq;
138 	struct tun_struct __rcu *tun;
139 	struct net *net;
140 	struct fasync_struct *fasync;
141 	/* only used for fasnyc */
142 	unsigned int flags;
143 	union {
144 		u16 queue_index;
145 		unsigned int ifindex;
146 	};
147 	struct list_head next;
148 	struct tun_struct *detached;
149 };
150 
151 struct tun_flow_entry {
152 	struct hlist_node hash_link;
153 	struct rcu_head rcu;
154 	struct tun_struct *tun;
155 
156 	u32 rxhash;
157 	u32 rps_rxhash;
158 	int queue_index;
159 	unsigned long updated;
160 };
161 
162 #define TUN_NUM_FLOW_ENTRIES 1024
163 
164 /* Since the socket were moved to tun_file, to preserve the behavior of persist
165  * device, socket filter, sndbuf and vnet header size were restore when the
166  * file were attached to a persist device.
167  */
168 struct tun_struct {
169 	struct tun_file __rcu	*tfiles[MAX_TAP_QUEUES];
170 	unsigned int            numqueues;
171 	unsigned int 		flags;
172 	kuid_t			owner;
173 	kgid_t			group;
174 
175 	struct net_device	*dev;
176 	netdev_features_t	set_features;
177 #define TUN_USER_FEATURES (NETIF_F_HW_CSUM|NETIF_F_TSO_ECN|NETIF_F_TSO| \
178 			  NETIF_F_TSO6)
179 
180 	int			vnet_hdr_sz;
181 	int			sndbuf;
182 	struct tap_filter	txflt;
183 	struct sock_fprog	fprog;
184 	/* protected by rtnl lock */
185 	bool			filter_attached;
186 #ifdef TUN_DEBUG
187 	int debug;
188 #endif
189 	spinlock_t lock;
190 	struct hlist_head flows[TUN_NUM_FLOW_ENTRIES];
191 	struct timer_list flow_gc_timer;
192 	unsigned long ageing_time;
193 	unsigned int numdisabled;
194 	struct list_head disabled;
195 	void *security;
196 	u32 flow_count;
197 };
198 
199 static inline u32 tun_hashfn(u32 rxhash)
200 {
201 	return rxhash & 0x3ff;
202 }
203 
204 static struct tun_flow_entry *tun_flow_find(struct hlist_head *head, u32 rxhash)
205 {
206 	struct tun_flow_entry *e;
207 
208 	hlist_for_each_entry_rcu(e, head, hash_link) {
209 		if (e->rxhash == rxhash)
210 			return e;
211 	}
212 	return NULL;
213 }
214 
215 static struct tun_flow_entry *tun_flow_create(struct tun_struct *tun,
216 					      struct hlist_head *head,
217 					      u32 rxhash, u16 queue_index)
218 {
219 	struct tun_flow_entry *e = kmalloc(sizeof(*e), GFP_ATOMIC);
220 
221 	if (e) {
222 		tun_debug(KERN_INFO, tun, "create flow: hash %u index %u\n",
223 			  rxhash, queue_index);
224 		e->updated = jiffies;
225 		e->rxhash = rxhash;
226 		e->rps_rxhash = 0;
227 		e->queue_index = queue_index;
228 		e->tun = tun;
229 		hlist_add_head_rcu(&e->hash_link, head);
230 		++tun->flow_count;
231 	}
232 	return e;
233 }
234 
235 static void tun_flow_delete(struct tun_struct *tun, struct tun_flow_entry *e)
236 {
237 	tun_debug(KERN_INFO, tun, "delete flow: hash %u index %u\n",
238 		  e->rxhash, e->queue_index);
239 	sock_rps_reset_flow_hash(e->rps_rxhash);
240 	hlist_del_rcu(&e->hash_link);
241 	kfree_rcu(e, rcu);
242 	--tun->flow_count;
243 }
244 
245 static void tun_flow_flush(struct tun_struct *tun)
246 {
247 	int i;
248 
249 	spin_lock_bh(&tun->lock);
250 	for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
251 		struct tun_flow_entry *e;
252 		struct hlist_node *n;
253 
254 		hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link)
255 			tun_flow_delete(tun, e);
256 	}
257 	spin_unlock_bh(&tun->lock);
258 }
259 
260 static void tun_flow_delete_by_queue(struct tun_struct *tun, u16 queue_index)
261 {
262 	int i;
263 
264 	spin_lock_bh(&tun->lock);
265 	for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
266 		struct tun_flow_entry *e;
267 		struct hlist_node *n;
268 
269 		hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
270 			if (e->queue_index == queue_index)
271 				tun_flow_delete(tun, e);
272 		}
273 	}
274 	spin_unlock_bh(&tun->lock);
275 }
276 
277 static void tun_flow_cleanup(unsigned long data)
278 {
279 	struct tun_struct *tun = (struct tun_struct *)data;
280 	unsigned long delay = tun->ageing_time;
281 	unsigned long next_timer = jiffies + delay;
282 	unsigned long count = 0;
283 	int i;
284 
285 	tun_debug(KERN_INFO, tun, "tun_flow_cleanup\n");
286 
287 	spin_lock_bh(&tun->lock);
288 	for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
289 		struct tun_flow_entry *e;
290 		struct hlist_node *n;
291 
292 		hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
293 			unsigned long this_timer;
294 			count++;
295 			this_timer = e->updated + delay;
296 			if (time_before_eq(this_timer, jiffies))
297 				tun_flow_delete(tun, e);
298 			else if (time_before(this_timer, next_timer))
299 				next_timer = this_timer;
300 		}
301 	}
302 
303 	if (count)
304 		mod_timer(&tun->flow_gc_timer, round_jiffies_up(next_timer));
305 	spin_unlock_bh(&tun->lock);
306 }
307 
308 static void tun_flow_update(struct tun_struct *tun, u32 rxhash,
309 			    struct tun_file *tfile)
310 {
311 	struct hlist_head *head;
312 	struct tun_flow_entry *e;
313 	unsigned long delay = tun->ageing_time;
314 	u16 queue_index = tfile->queue_index;
315 
316 	if (!rxhash)
317 		return;
318 	else
319 		head = &tun->flows[tun_hashfn(rxhash)];
320 
321 	rcu_read_lock();
322 
323 	/* We may get a very small possibility of OOO during switching, not
324 	 * worth to optimize.*/
325 	if (tun->numqueues == 1 || tfile->detached)
326 		goto unlock;
327 
328 	e = tun_flow_find(head, rxhash);
329 	if (likely(e)) {
330 		/* TODO: keep queueing to old queue until it's empty? */
331 		e->queue_index = queue_index;
332 		e->updated = jiffies;
333 		sock_rps_record_flow_hash(e->rps_rxhash);
334 	} else {
335 		spin_lock_bh(&tun->lock);
336 		if (!tun_flow_find(head, rxhash) &&
337 		    tun->flow_count < MAX_TAP_FLOWS)
338 			tun_flow_create(tun, head, rxhash, queue_index);
339 
340 		if (!timer_pending(&tun->flow_gc_timer))
341 			mod_timer(&tun->flow_gc_timer,
342 				  round_jiffies_up(jiffies + delay));
343 		spin_unlock_bh(&tun->lock);
344 	}
345 
346 unlock:
347 	rcu_read_unlock();
348 }
349 
350 /**
351  * Save the hash received in the stack receive path and update the
352  * flow_hash table accordingly.
353  */
354 static inline void tun_flow_save_rps_rxhash(struct tun_flow_entry *e, u32 hash)
355 {
356 	if (unlikely(e->rps_rxhash != hash)) {
357 		sock_rps_reset_flow_hash(e->rps_rxhash);
358 		e->rps_rxhash = hash;
359 	}
360 }
361 
362 /* We try to identify a flow through its rxhash first. The reason that
363  * we do not check rxq no. is because some cards(e.g 82599), chooses
364  * the rxq based on the txq where the last packet of the flow comes. As
365  * the userspace application move between processors, we may get a
366  * different rxq no. here. If we could not get rxhash, then we would
367  * hope the rxq no. may help here.
368  */
369 static u16 tun_select_queue(struct net_device *dev, struct sk_buff *skb,
370 			    void *accel_priv, select_queue_fallback_t fallback)
371 {
372 	struct tun_struct *tun = netdev_priv(dev);
373 	struct tun_flow_entry *e;
374 	u32 txq = 0;
375 	u32 numqueues = 0;
376 
377 	rcu_read_lock();
378 	numqueues = ACCESS_ONCE(tun->numqueues);
379 
380 	txq = skb_get_hash(skb);
381 	if (txq) {
382 		e = tun_flow_find(&tun->flows[tun_hashfn(txq)], txq);
383 		if (e) {
384 			tun_flow_save_rps_rxhash(e, txq);
385 			txq = e->queue_index;
386 		} else
387 			/* use multiply and shift instead of expensive divide */
388 			txq = ((u64)txq * numqueues) >> 32;
389 	} else if (likely(skb_rx_queue_recorded(skb))) {
390 		txq = skb_get_rx_queue(skb);
391 		while (unlikely(txq >= numqueues))
392 			txq -= numqueues;
393 	}
394 
395 	rcu_read_unlock();
396 	return txq;
397 }
398 
399 static inline bool tun_not_capable(struct tun_struct *tun)
400 {
401 	const struct cred *cred = current_cred();
402 	struct net *net = dev_net(tun->dev);
403 
404 	return ((uid_valid(tun->owner) && !uid_eq(cred->euid, tun->owner)) ||
405 		  (gid_valid(tun->group) && !in_egroup_p(tun->group))) &&
406 		!ns_capable(net->user_ns, CAP_NET_ADMIN);
407 }
408 
409 static void tun_set_real_num_queues(struct tun_struct *tun)
410 {
411 	netif_set_real_num_tx_queues(tun->dev, tun->numqueues);
412 	netif_set_real_num_rx_queues(tun->dev, tun->numqueues);
413 }
414 
415 static void tun_disable_queue(struct tun_struct *tun, struct tun_file *tfile)
416 {
417 	tfile->detached = tun;
418 	list_add_tail(&tfile->next, &tun->disabled);
419 	++tun->numdisabled;
420 }
421 
422 static struct tun_struct *tun_enable_queue(struct tun_file *tfile)
423 {
424 	struct tun_struct *tun = tfile->detached;
425 
426 	tfile->detached = NULL;
427 	list_del_init(&tfile->next);
428 	--tun->numdisabled;
429 	return tun;
430 }
431 
432 static void tun_queue_purge(struct tun_file *tfile)
433 {
434 	skb_queue_purge(&tfile->sk.sk_receive_queue);
435 	skb_queue_purge(&tfile->sk.sk_error_queue);
436 }
437 
438 static void __tun_detach(struct tun_file *tfile, bool clean)
439 {
440 	struct tun_file *ntfile;
441 	struct tun_struct *tun;
442 
443 	tun = rtnl_dereference(tfile->tun);
444 
445 	if (tun && !tfile->detached) {
446 		u16 index = tfile->queue_index;
447 		BUG_ON(index >= tun->numqueues);
448 
449 		rcu_assign_pointer(tun->tfiles[index],
450 				   tun->tfiles[tun->numqueues - 1]);
451 		ntfile = rtnl_dereference(tun->tfiles[index]);
452 		ntfile->queue_index = index;
453 
454 		--tun->numqueues;
455 		if (clean) {
456 			RCU_INIT_POINTER(tfile->tun, NULL);
457 			sock_put(&tfile->sk);
458 		} else
459 			tun_disable_queue(tun, tfile);
460 
461 		synchronize_net();
462 		tun_flow_delete_by_queue(tun, tun->numqueues + 1);
463 		/* Drop read queue */
464 		tun_queue_purge(tfile);
465 		tun_set_real_num_queues(tun);
466 	} else if (tfile->detached && clean) {
467 		tun = tun_enable_queue(tfile);
468 		sock_put(&tfile->sk);
469 	}
470 
471 	if (clean) {
472 		if (tun && tun->numqueues == 0 && tun->numdisabled == 0) {
473 			netif_carrier_off(tun->dev);
474 
475 			if (!(tun->flags & TUN_PERSIST) &&
476 			    tun->dev->reg_state == NETREG_REGISTERED)
477 				unregister_netdevice(tun->dev);
478 		}
479 
480 		BUG_ON(!test_bit(SOCK_EXTERNALLY_ALLOCATED,
481 				 &tfile->socket.flags));
482 		sk_release_kernel(&tfile->sk);
483 	}
484 }
485 
486 static void tun_detach(struct tun_file *tfile, bool clean)
487 {
488 	rtnl_lock();
489 	__tun_detach(tfile, clean);
490 	rtnl_unlock();
491 }
492 
493 static void tun_detach_all(struct net_device *dev)
494 {
495 	struct tun_struct *tun = netdev_priv(dev);
496 	struct tun_file *tfile, *tmp;
497 	int i, n = tun->numqueues;
498 
499 	for (i = 0; i < n; i++) {
500 		tfile = rtnl_dereference(tun->tfiles[i]);
501 		BUG_ON(!tfile);
502 		tfile->socket.sk->sk_data_ready(tfile->socket.sk);
503 		RCU_INIT_POINTER(tfile->tun, NULL);
504 		--tun->numqueues;
505 	}
506 	list_for_each_entry(tfile, &tun->disabled, next) {
507 		tfile->socket.sk->sk_data_ready(tfile->socket.sk);
508 		RCU_INIT_POINTER(tfile->tun, NULL);
509 	}
510 	BUG_ON(tun->numqueues != 0);
511 
512 	synchronize_net();
513 	for (i = 0; i < n; i++) {
514 		tfile = rtnl_dereference(tun->tfiles[i]);
515 		/* Drop read queue */
516 		tun_queue_purge(tfile);
517 		sock_put(&tfile->sk);
518 	}
519 	list_for_each_entry_safe(tfile, tmp, &tun->disabled, next) {
520 		tun_enable_queue(tfile);
521 		tun_queue_purge(tfile);
522 		sock_put(&tfile->sk);
523 	}
524 	BUG_ON(tun->numdisabled != 0);
525 
526 	if (tun->flags & TUN_PERSIST)
527 		module_put(THIS_MODULE);
528 }
529 
530 static int tun_attach(struct tun_struct *tun, struct file *file, bool skip_filter)
531 {
532 	struct tun_file *tfile = file->private_data;
533 	int err;
534 
535 	err = security_tun_dev_attach(tfile->socket.sk, tun->security);
536 	if (err < 0)
537 		goto out;
538 
539 	err = -EINVAL;
540 	if (rtnl_dereference(tfile->tun) && !tfile->detached)
541 		goto out;
542 
543 	err = -EBUSY;
544 	if (!(tun->flags & TUN_TAP_MQ) && tun->numqueues == 1)
545 		goto out;
546 
547 	err = -E2BIG;
548 	if (!tfile->detached &&
549 	    tun->numqueues + tun->numdisabled == MAX_TAP_QUEUES)
550 		goto out;
551 
552 	err = 0;
553 
554 	/* Re-attach the filter to persist device */
555 	if (!skip_filter && (tun->filter_attached == true)) {
556 		err = sk_attach_filter(&tun->fprog, tfile->socket.sk);
557 		if (!err)
558 			goto out;
559 	}
560 	tfile->queue_index = tun->numqueues;
561 	rcu_assign_pointer(tfile->tun, tun);
562 	rcu_assign_pointer(tun->tfiles[tun->numqueues], tfile);
563 	tun->numqueues++;
564 
565 	if (tfile->detached)
566 		tun_enable_queue(tfile);
567 	else
568 		sock_hold(&tfile->sk);
569 
570 	tun_set_real_num_queues(tun);
571 
572 	/* device is allowed to go away first, so no need to hold extra
573 	 * refcnt.
574 	 */
575 
576 out:
577 	return err;
578 }
579 
580 static struct tun_struct *__tun_get(struct tun_file *tfile)
581 {
582 	struct tun_struct *tun;
583 
584 	rcu_read_lock();
585 	tun = rcu_dereference(tfile->tun);
586 	if (tun)
587 		dev_hold(tun->dev);
588 	rcu_read_unlock();
589 
590 	return tun;
591 }
592 
593 static struct tun_struct *tun_get(struct file *file)
594 {
595 	return __tun_get(file->private_data);
596 }
597 
598 static void tun_put(struct tun_struct *tun)
599 {
600 	dev_put(tun->dev);
601 }
602 
603 /* TAP filtering */
604 static void addr_hash_set(u32 *mask, const u8 *addr)
605 {
606 	int n = ether_crc(ETH_ALEN, addr) >> 26;
607 	mask[n >> 5] |= (1 << (n & 31));
608 }
609 
610 static unsigned int addr_hash_test(const u32 *mask, const u8 *addr)
611 {
612 	int n = ether_crc(ETH_ALEN, addr) >> 26;
613 	return mask[n >> 5] & (1 << (n & 31));
614 }
615 
616 static int update_filter(struct tap_filter *filter, void __user *arg)
617 {
618 	struct { u8 u[ETH_ALEN]; } *addr;
619 	struct tun_filter uf;
620 	int err, alen, n, nexact;
621 
622 	if (copy_from_user(&uf, arg, sizeof(uf)))
623 		return -EFAULT;
624 
625 	if (!uf.count) {
626 		/* Disabled */
627 		filter->count = 0;
628 		return 0;
629 	}
630 
631 	alen = ETH_ALEN * uf.count;
632 	addr = kmalloc(alen, GFP_KERNEL);
633 	if (!addr)
634 		return -ENOMEM;
635 
636 	if (copy_from_user(addr, arg + sizeof(uf), alen)) {
637 		err = -EFAULT;
638 		goto done;
639 	}
640 
641 	/* The filter is updated without holding any locks. Which is
642 	 * perfectly safe. We disable it first and in the worst
643 	 * case we'll accept a few undesired packets. */
644 	filter->count = 0;
645 	wmb();
646 
647 	/* Use first set of addresses as an exact filter */
648 	for (n = 0; n < uf.count && n < FLT_EXACT_COUNT; n++)
649 		memcpy(filter->addr[n], addr[n].u, ETH_ALEN);
650 
651 	nexact = n;
652 
653 	/* Remaining multicast addresses are hashed,
654 	 * unicast will leave the filter disabled. */
655 	memset(filter->mask, 0, sizeof(filter->mask));
656 	for (; n < uf.count; n++) {
657 		if (!is_multicast_ether_addr(addr[n].u)) {
658 			err = 0; /* no filter */
659 			goto done;
660 		}
661 		addr_hash_set(filter->mask, addr[n].u);
662 	}
663 
664 	/* For ALLMULTI just set the mask to all ones.
665 	 * This overrides the mask populated above. */
666 	if ((uf.flags & TUN_FLT_ALLMULTI))
667 		memset(filter->mask, ~0, sizeof(filter->mask));
668 
669 	/* Now enable the filter */
670 	wmb();
671 	filter->count = nexact;
672 
673 	/* Return the number of exact filters */
674 	err = nexact;
675 
676 done:
677 	kfree(addr);
678 	return err;
679 }
680 
681 /* Returns: 0 - drop, !=0 - accept */
682 static int run_filter(struct tap_filter *filter, const struct sk_buff *skb)
683 {
684 	/* Cannot use eth_hdr(skb) here because skb_mac_hdr() is incorrect
685 	 * at this point. */
686 	struct ethhdr *eh = (struct ethhdr *) skb->data;
687 	int i;
688 
689 	/* Exact match */
690 	for (i = 0; i < filter->count; i++)
691 		if (ether_addr_equal(eh->h_dest, filter->addr[i]))
692 			return 1;
693 
694 	/* Inexact match (multicast only) */
695 	if (is_multicast_ether_addr(eh->h_dest))
696 		return addr_hash_test(filter->mask, eh->h_dest);
697 
698 	return 0;
699 }
700 
701 /*
702  * Checks whether the packet is accepted or not.
703  * Returns: 0 - drop, !=0 - accept
704  */
705 static int check_filter(struct tap_filter *filter, const struct sk_buff *skb)
706 {
707 	if (!filter->count)
708 		return 1;
709 
710 	return run_filter(filter, skb);
711 }
712 
713 /* Network device part of the driver */
714 
715 static const struct ethtool_ops tun_ethtool_ops;
716 
717 /* Net device detach from fd. */
718 static void tun_net_uninit(struct net_device *dev)
719 {
720 	tun_detach_all(dev);
721 }
722 
723 /* Net device open. */
724 static int tun_net_open(struct net_device *dev)
725 {
726 	netif_tx_start_all_queues(dev);
727 	return 0;
728 }
729 
730 /* Net device close. */
731 static int tun_net_close(struct net_device *dev)
732 {
733 	netif_tx_stop_all_queues(dev);
734 	return 0;
735 }
736 
737 /* Net device start xmit */
738 static netdev_tx_t tun_net_xmit(struct sk_buff *skb, struct net_device *dev)
739 {
740 	struct tun_struct *tun = netdev_priv(dev);
741 	int txq = skb->queue_mapping;
742 	struct tun_file *tfile;
743 	u32 numqueues = 0;
744 
745 	rcu_read_lock();
746 	tfile = rcu_dereference(tun->tfiles[txq]);
747 	numqueues = ACCESS_ONCE(tun->numqueues);
748 
749 	/* Drop packet if interface is not attached */
750 	if (txq >= numqueues)
751 		goto drop;
752 
753 	if (numqueues == 1) {
754 		/* Select queue was not called for the skbuff, so we extract the
755 		 * RPS hash and save it into the flow_table here.
756 		 */
757 		__u32 rxhash;
758 
759 		rxhash = skb_get_hash(skb);
760 		if (rxhash) {
761 			struct tun_flow_entry *e;
762 			e = tun_flow_find(&tun->flows[tun_hashfn(rxhash)],
763 					rxhash);
764 			if (e)
765 				tun_flow_save_rps_rxhash(e, rxhash);
766 		}
767 	}
768 
769 	tun_debug(KERN_INFO, tun, "tun_net_xmit %d\n", skb->len);
770 
771 	BUG_ON(!tfile);
772 
773 	/* Drop if the filter does not like it.
774 	 * This is a noop if the filter is disabled.
775 	 * Filter can be enabled only for the TAP devices. */
776 	if (!check_filter(&tun->txflt, skb))
777 		goto drop;
778 
779 	if (tfile->socket.sk->sk_filter &&
780 	    sk_filter(tfile->socket.sk, skb))
781 		goto drop;
782 
783 	/* Limit the number of packets queued by dividing txq length with the
784 	 * number of queues.
785 	 */
786 	if (skb_queue_len(&tfile->socket.sk->sk_receive_queue) * numqueues
787 			  >= dev->tx_queue_len)
788 		goto drop;
789 
790 	if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC)))
791 		goto drop;
792 
793 	if (skb->sk) {
794 		sock_tx_timestamp(skb->sk, &skb_shinfo(skb)->tx_flags);
795 		sw_tx_timestamp(skb);
796 	}
797 
798 	/* Orphan the skb - required as we might hang on to it
799 	 * for indefinite time.
800 	 */
801 	skb_orphan(skb);
802 
803 	nf_reset(skb);
804 
805 	/* Enqueue packet */
806 	skb_queue_tail(&tfile->socket.sk->sk_receive_queue, skb);
807 
808 	/* Notify and wake up reader process */
809 	if (tfile->flags & TUN_FASYNC)
810 		kill_fasync(&tfile->fasync, SIGIO, POLL_IN);
811 	tfile->socket.sk->sk_data_ready(tfile->socket.sk);
812 
813 	rcu_read_unlock();
814 	return NETDEV_TX_OK;
815 
816 drop:
817 	dev->stats.tx_dropped++;
818 	skb_tx_error(skb);
819 	kfree_skb(skb);
820 	rcu_read_unlock();
821 	return NETDEV_TX_OK;
822 }
823 
824 static void tun_net_mclist(struct net_device *dev)
825 {
826 	/*
827 	 * This callback is supposed to deal with mc filter in
828 	 * _rx_ path and has nothing to do with the _tx_ path.
829 	 * In rx path we always accept everything userspace gives us.
830 	 */
831 }
832 
833 #define MIN_MTU 68
834 #define MAX_MTU 65535
835 
836 static int
837 tun_net_change_mtu(struct net_device *dev, int new_mtu)
838 {
839 	if (new_mtu < MIN_MTU || new_mtu + dev->hard_header_len > MAX_MTU)
840 		return -EINVAL;
841 	dev->mtu = new_mtu;
842 	return 0;
843 }
844 
845 static netdev_features_t tun_net_fix_features(struct net_device *dev,
846 	netdev_features_t features)
847 {
848 	struct tun_struct *tun = netdev_priv(dev);
849 
850 	return (features & tun->set_features) | (features & ~TUN_USER_FEATURES);
851 }
852 #ifdef CONFIG_NET_POLL_CONTROLLER
853 static void tun_poll_controller(struct net_device *dev)
854 {
855 	/*
856 	 * Tun only receives frames when:
857 	 * 1) the char device endpoint gets data from user space
858 	 * 2) the tun socket gets a sendmsg call from user space
859 	 * Since both of those are synchronous operations, we are guaranteed
860 	 * never to have pending data when we poll for it
861 	 * so there is nothing to do here but return.
862 	 * We need this though so netpoll recognizes us as an interface that
863 	 * supports polling, which enables bridge devices in virt setups to
864 	 * still use netconsole
865 	 */
866 	return;
867 }
868 #endif
869 static const struct net_device_ops tun_netdev_ops = {
870 	.ndo_uninit		= tun_net_uninit,
871 	.ndo_open		= tun_net_open,
872 	.ndo_stop		= tun_net_close,
873 	.ndo_start_xmit		= tun_net_xmit,
874 	.ndo_change_mtu		= tun_net_change_mtu,
875 	.ndo_fix_features	= tun_net_fix_features,
876 	.ndo_select_queue	= tun_select_queue,
877 #ifdef CONFIG_NET_POLL_CONTROLLER
878 	.ndo_poll_controller	= tun_poll_controller,
879 #endif
880 };
881 
882 static const struct net_device_ops tap_netdev_ops = {
883 	.ndo_uninit		= tun_net_uninit,
884 	.ndo_open		= tun_net_open,
885 	.ndo_stop		= tun_net_close,
886 	.ndo_start_xmit		= tun_net_xmit,
887 	.ndo_change_mtu		= tun_net_change_mtu,
888 	.ndo_fix_features	= tun_net_fix_features,
889 	.ndo_set_rx_mode	= tun_net_mclist,
890 	.ndo_set_mac_address	= eth_mac_addr,
891 	.ndo_validate_addr	= eth_validate_addr,
892 	.ndo_select_queue	= tun_select_queue,
893 #ifdef CONFIG_NET_POLL_CONTROLLER
894 	.ndo_poll_controller	= tun_poll_controller,
895 #endif
896 };
897 
898 static void tun_flow_init(struct tun_struct *tun)
899 {
900 	int i;
901 
902 	for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++)
903 		INIT_HLIST_HEAD(&tun->flows[i]);
904 
905 	tun->ageing_time = TUN_FLOW_EXPIRE;
906 	setup_timer(&tun->flow_gc_timer, tun_flow_cleanup, (unsigned long)tun);
907 	mod_timer(&tun->flow_gc_timer,
908 		  round_jiffies_up(jiffies + tun->ageing_time));
909 }
910 
911 static void tun_flow_uninit(struct tun_struct *tun)
912 {
913 	del_timer_sync(&tun->flow_gc_timer);
914 	tun_flow_flush(tun);
915 }
916 
917 /* Initialize net device. */
918 static void tun_net_init(struct net_device *dev)
919 {
920 	struct tun_struct *tun = netdev_priv(dev);
921 
922 	switch (tun->flags & TUN_TYPE_MASK) {
923 	case TUN_TUN_DEV:
924 		dev->netdev_ops = &tun_netdev_ops;
925 
926 		/* Point-to-Point TUN Device */
927 		dev->hard_header_len = 0;
928 		dev->addr_len = 0;
929 		dev->mtu = 1500;
930 
931 		/* Zero header length */
932 		dev->type = ARPHRD_NONE;
933 		dev->flags = IFF_POINTOPOINT | IFF_NOARP | IFF_MULTICAST;
934 		dev->tx_queue_len = TUN_READQ_SIZE;  /* We prefer our own queue length */
935 		break;
936 
937 	case TUN_TAP_DEV:
938 		dev->netdev_ops = &tap_netdev_ops;
939 		/* Ethernet TAP Device */
940 		ether_setup(dev);
941 		dev->priv_flags &= ~IFF_TX_SKB_SHARING;
942 		dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
943 
944 		eth_hw_addr_random(dev);
945 
946 		dev->tx_queue_len = TUN_READQ_SIZE;  /* We prefer our own queue length */
947 		break;
948 	}
949 }
950 
951 /* Character device part */
952 
953 /* Poll */
954 static unsigned int tun_chr_poll(struct file *file, poll_table *wait)
955 {
956 	struct tun_file *tfile = file->private_data;
957 	struct tun_struct *tun = __tun_get(tfile);
958 	struct sock *sk;
959 	unsigned int mask = 0;
960 
961 	if (!tun)
962 		return POLLERR;
963 
964 	sk = tfile->socket.sk;
965 
966 	tun_debug(KERN_INFO, tun, "tun_chr_poll\n");
967 
968 	poll_wait(file, sk_sleep(sk), wait);
969 
970 	if (!skb_queue_empty(&sk->sk_receive_queue))
971 		mask |= POLLIN | POLLRDNORM;
972 
973 	if (sock_writeable(sk) ||
974 	    (!test_and_set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags) &&
975 	     sock_writeable(sk)))
976 		mask |= POLLOUT | POLLWRNORM;
977 
978 	if (tun->dev->reg_state != NETREG_REGISTERED)
979 		mask = POLLERR;
980 
981 	tun_put(tun);
982 	return mask;
983 }
984 
985 /* prepad is the amount to reserve at front.  len is length after that.
986  * linear is a hint as to how much to copy (usually headers). */
987 static struct sk_buff *tun_alloc_skb(struct tun_file *tfile,
988 				     size_t prepad, size_t len,
989 				     size_t linear, int noblock)
990 {
991 	struct sock *sk = tfile->socket.sk;
992 	struct sk_buff *skb;
993 	int err;
994 
995 	/* Under a page?  Don't bother with paged skb. */
996 	if (prepad + len < PAGE_SIZE || !linear)
997 		linear = len;
998 
999 	skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
1000 				   &err, 0);
1001 	if (!skb)
1002 		return ERR_PTR(err);
1003 
1004 	skb_reserve(skb, prepad);
1005 	skb_put(skb, linear);
1006 	skb->data_len = len - linear;
1007 	skb->len += len - linear;
1008 
1009 	return skb;
1010 }
1011 
1012 /* Get packet from user space buffer */
1013 static ssize_t tun_get_user(struct tun_struct *tun, struct tun_file *tfile,
1014 			    void *msg_control, const struct iovec *iv,
1015 			    size_t total_len, size_t count, int noblock)
1016 {
1017 	struct tun_pi pi = { 0, cpu_to_be16(ETH_P_IP) };
1018 	struct sk_buff *skb;
1019 	size_t len = total_len, align = NET_SKB_PAD, linear;
1020 	struct virtio_net_hdr gso = { 0 };
1021 	int good_linear;
1022 	int offset = 0;
1023 	int copylen;
1024 	bool zerocopy = false;
1025 	int err;
1026 	u32 rxhash;
1027 
1028 	if (!(tun->flags & TUN_NO_PI)) {
1029 		if (len < sizeof(pi))
1030 			return -EINVAL;
1031 		len -= sizeof(pi);
1032 
1033 		if (memcpy_fromiovecend((void *)&pi, iv, 0, sizeof(pi)))
1034 			return -EFAULT;
1035 		offset += sizeof(pi);
1036 	}
1037 
1038 	if (tun->flags & TUN_VNET_HDR) {
1039 		if (len < tun->vnet_hdr_sz)
1040 			return -EINVAL;
1041 		len -= tun->vnet_hdr_sz;
1042 
1043 		if (memcpy_fromiovecend((void *)&gso, iv, offset, sizeof(gso)))
1044 			return -EFAULT;
1045 
1046 		if ((gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
1047 		    gso.csum_start + gso.csum_offset + 2 > gso.hdr_len)
1048 			gso.hdr_len = gso.csum_start + gso.csum_offset + 2;
1049 
1050 		if (gso.hdr_len > len)
1051 			return -EINVAL;
1052 		offset += tun->vnet_hdr_sz;
1053 	}
1054 
1055 	if ((tun->flags & TUN_TYPE_MASK) == TUN_TAP_DEV) {
1056 		align += NET_IP_ALIGN;
1057 		if (unlikely(len < ETH_HLEN ||
1058 			     (gso.hdr_len && gso.hdr_len < ETH_HLEN)))
1059 			return -EINVAL;
1060 	}
1061 
1062 	good_linear = SKB_MAX_HEAD(align);
1063 
1064 	if (msg_control) {
1065 		/* There are 256 bytes to be copied in skb, so there is
1066 		 * enough room for skb expand head in case it is used.
1067 		 * The rest of the buffer is mapped from userspace.
1068 		 */
1069 		copylen = gso.hdr_len ? gso.hdr_len : GOODCOPY_LEN;
1070 		if (copylen > good_linear)
1071 			copylen = good_linear;
1072 		linear = copylen;
1073 		if (iov_pages(iv, offset + copylen, count) <= MAX_SKB_FRAGS)
1074 			zerocopy = true;
1075 	}
1076 
1077 	if (!zerocopy) {
1078 		copylen = len;
1079 		if (gso.hdr_len > good_linear)
1080 			linear = good_linear;
1081 		else
1082 			linear = gso.hdr_len;
1083 	}
1084 
1085 	skb = tun_alloc_skb(tfile, align, copylen, linear, noblock);
1086 	if (IS_ERR(skb)) {
1087 		if (PTR_ERR(skb) != -EAGAIN)
1088 			tun->dev->stats.rx_dropped++;
1089 		return PTR_ERR(skb);
1090 	}
1091 
1092 	if (zerocopy)
1093 		err = zerocopy_sg_from_iovec(skb, iv, offset, count);
1094 	else {
1095 		err = skb_copy_datagram_from_iovec(skb, 0, iv, offset, len);
1096 		if (!err && msg_control) {
1097 			struct ubuf_info *uarg = msg_control;
1098 			uarg->callback(uarg, false);
1099 		}
1100 	}
1101 
1102 	if (err) {
1103 		tun->dev->stats.rx_dropped++;
1104 		kfree_skb(skb);
1105 		return -EFAULT;
1106 	}
1107 
1108 	if (gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
1109 		if (!skb_partial_csum_set(skb, gso.csum_start,
1110 					  gso.csum_offset)) {
1111 			tun->dev->stats.rx_frame_errors++;
1112 			kfree_skb(skb);
1113 			return -EINVAL;
1114 		}
1115 	}
1116 
1117 	switch (tun->flags & TUN_TYPE_MASK) {
1118 	case TUN_TUN_DEV:
1119 		if (tun->flags & TUN_NO_PI) {
1120 			switch (skb->data[0] & 0xf0) {
1121 			case 0x40:
1122 				pi.proto = htons(ETH_P_IP);
1123 				break;
1124 			case 0x60:
1125 				pi.proto = htons(ETH_P_IPV6);
1126 				break;
1127 			default:
1128 				tun->dev->stats.rx_dropped++;
1129 				kfree_skb(skb);
1130 				return -EINVAL;
1131 			}
1132 		}
1133 
1134 		skb_reset_mac_header(skb);
1135 		skb->protocol = pi.proto;
1136 		skb->dev = tun->dev;
1137 		break;
1138 	case TUN_TAP_DEV:
1139 		skb->protocol = eth_type_trans(skb, tun->dev);
1140 		break;
1141 	}
1142 
1143 	skb_reset_network_header(skb);
1144 
1145 	if (gso.gso_type != VIRTIO_NET_HDR_GSO_NONE) {
1146 		pr_debug("GSO!\n");
1147 		switch (gso.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
1148 		case VIRTIO_NET_HDR_GSO_TCPV4:
1149 			skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
1150 			break;
1151 		case VIRTIO_NET_HDR_GSO_TCPV6:
1152 			skb_shinfo(skb)->gso_type = SKB_GSO_TCPV6;
1153 			break;
1154 		case VIRTIO_NET_HDR_GSO_UDP:
1155 		{
1156 			static bool warned;
1157 
1158 			if (!warned) {
1159 				warned = true;
1160 				netdev_warn(tun->dev,
1161 					    "%s: using disabled UFO feature; please fix this program\n",
1162 					    current->comm);
1163 			}
1164 			skb_shinfo(skb)->gso_type = SKB_GSO_UDP;
1165 			if (skb->protocol == htons(ETH_P_IPV6))
1166 				ipv6_proxy_select_ident(skb);
1167 			break;
1168 		}
1169 		default:
1170 			tun->dev->stats.rx_frame_errors++;
1171 			kfree_skb(skb);
1172 			return -EINVAL;
1173 		}
1174 
1175 		if (gso.gso_type & VIRTIO_NET_HDR_GSO_ECN)
1176 			skb_shinfo(skb)->gso_type |= SKB_GSO_TCP_ECN;
1177 
1178 		skb_shinfo(skb)->gso_size = gso.gso_size;
1179 		if (skb_shinfo(skb)->gso_size == 0) {
1180 			tun->dev->stats.rx_frame_errors++;
1181 			kfree_skb(skb);
1182 			return -EINVAL;
1183 		}
1184 
1185 		/* Header must be checked, and gso_segs computed. */
1186 		skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
1187 		skb_shinfo(skb)->gso_segs = 0;
1188 	}
1189 
1190 	/* copy skb_ubuf_info for callback when skb has no error */
1191 	if (zerocopy) {
1192 		skb_shinfo(skb)->destructor_arg = msg_control;
1193 		skb_shinfo(skb)->tx_flags |= SKBTX_DEV_ZEROCOPY;
1194 		skb_shinfo(skb)->tx_flags |= SKBTX_SHARED_FRAG;
1195 	}
1196 
1197 	skb_probe_transport_header(skb, 0);
1198 
1199 	rxhash = skb_get_hash(skb);
1200 	netif_rx_ni(skb);
1201 
1202 	tun->dev->stats.rx_packets++;
1203 	tun->dev->stats.rx_bytes += len;
1204 
1205 	tun_flow_update(tun, rxhash, tfile);
1206 	return total_len;
1207 }
1208 
1209 static ssize_t tun_chr_aio_write(struct kiocb *iocb, const struct iovec *iv,
1210 			      unsigned long count, loff_t pos)
1211 {
1212 	struct file *file = iocb->ki_filp;
1213 	struct tun_struct *tun = tun_get(file);
1214 	struct tun_file *tfile = file->private_data;
1215 	ssize_t result;
1216 
1217 	if (!tun)
1218 		return -EBADFD;
1219 
1220 	tun_debug(KERN_INFO, tun, "tun_chr_write %ld\n", count);
1221 
1222 	result = tun_get_user(tun, tfile, NULL, iv, iov_length(iv, count),
1223 			      count, file->f_flags & O_NONBLOCK);
1224 
1225 	tun_put(tun);
1226 	return result;
1227 }
1228 
1229 /* Put packet to the user space buffer */
1230 static ssize_t tun_put_user(struct tun_struct *tun,
1231 			    struct tun_file *tfile,
1232 			    struct sk_buff *skb,
1233 			    const struct iovec *iv, int len)
1234 {
1235 	struct tun_pi pi = { 0, skb->protocol };
1236 	ssize_t total = 0;
1237 	int vlan_offset = 0, copied;
1238 	int vlan_hlen = 0;
1239 
1240 	if (vlan_tx_tag_present(skb))
1241 		vlan_hlen = VLAN_HLEN;
1242 
1243 	if (!(tun->flags & TUN_NO_PI)) {
1244 		if ((len -= sizeof(pi)) < 0)
1245 			return -EINVAL;
1246 
1247 		if (len < skb->len) {
1248 			/* Packet will be striped */
1249 			pi.flags |= TUN_PKT_STRIP;
1250 		}
1251 
1252 		if (memcpy_toiovecend(iv, (void *) &pi, 0, sizeof(pi)))
1253 			return -EFAULT;
1254 		total += sizeof(pi);
1255 	}
1256 
1257 	if (tun->flags & TUN_VNET_HDR) {
1258 		struct virtio_net_hdr gso = { 0 }; /* no info leak */
1259 		if ((len -= tun->vnet_hdr_sz) < 0)
1260 			return -EINVAL;
1261 
1262 		if (skb_is_gso(skb)) {
1263 			struct skb_shared_info *sinfo = skb_shinfo(skb);
1264 
1265 			/* This is a hint as to how much should be linear. */
1266 			gso.hdr_len = skb_headlen(skb);
1267 			gso.gso_size = sinfo->gso_size;
1268 			if (sinfo->gso_type & SKB_GSO_TCPV4)
1269 				gso.gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
1270 			else if (sinfo->gso_type & SKB_GSO_TCPV6)
1271 				gso.gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
1272 			else {
1273 				pr_err("unexpected GSO type: "
1274 				       "0x%x, gso_size %d, hdr_len %d\n",
1275 				       sinfo->gso_type, gso.gso_size,
1276 				       gso.hdr_len);
1277 				print_hex_dump(KERN_ERR, "tun: ",
1278 					       DUMP_PREFIX_NONE,
1279 					       16, 1, skb->head,
1280 					       min((int)gso.hdr_len, 64), true);
1281 				WARN_ON_ONCE(1);
1282 				return -EINVAL;
1283 			}
1284 			if (sinfo->gso_type & SKB_GSO_TCP_ECN)
1285 				gso.gso_type |= VIRTIO_NET_HDR_GSO_ECN;
1286 		} else
1287 			gso.gso_type = VIRTIO_NET_HDR_GSO_NONE;
1288 
1289 		if (skb->ip_summed == CHECKSUM_PARTIAL) {
1290 			gso.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
1291 			gso.csum_start = skb_checksum_start_offset(skb) +
1292 					 vlan_hlen;
1293 			gso.csum_offset = skb->csum_offset;
1294 		} else if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
1295 			gso.flags = VIRTIO_NET_HDR_F_DATA_VALID;
1296 		} /* else everything is zero */
1297 
1298 		if (unlikely(memcpy_toiovecend(iv, (void *)&gso, total,
1299 					       sizeof(gso))))
1300 			return -EFAULT;
1301 		total += tun->vnet_hdr_sz;
1302 	}
1303 
1304 	copied = total;
1305 	len = min_t(int, skb->len + vlan_hlen, len);
1306 	total += skb->len + vlan_hlen;
1307 	if (vlan_hlen) {
1308 		int copy, ret;
1309 		struct {
1310 			__be16 h_vlan_proto;
1311 			__be16 h_vlan_TCI;
1312 		} veth;
1313 
1314 		veth.h_vlan_proto = skb->vlan_proto;
1315 		veth.h_vlan_TCI = htons(vlan_tx_tag_get(skb));
1316 
1317 		vlan_offset = offsetof(struct vlan_ethhdr, h_vlan_proto);
1318 
1319 		copy = min_t(int, vlan_offset, len);
1320 		ret = skb_copy_datagram_const_iovec(skb, 0, iv, copied, copy);
1321 		len -= copy;
1322 		copied += copy;
1323 		if (ret || !len)
1324 			goto done;
1325 
1326 		copy = min_t(int, sizeof(veth), len);
1327 		ret = memcpy_toiovecend(iv, (void *)&veth, copied, copy);
1328 		len -= copy;
1329 		copied += copy;
1330 		if (ret || !len)
1331 			goto done;
1332 	}
1333 
1334 	skb_copy_datagram_const_iovec(skb, vlan_offset, iv, copied, len);
1335 
1336 done:
1337 	tun->dev->stats.tx_packets++;
1338 	tun->dev->stats.tx_bytes += len;
1339 
1340 	return total;
1341 }
1342 
1343 static ssize_t tun_do_read(struct tun_struct *tun, struct tun_file *tfile,
1344 			   const struct iovec *iv, ssize_t len, int noblock)
1345 {
1346 	struct sk_buff *skb;
1347 	ssize_t ret = 0;
1348 	int peeked, err, off = 0;
1349 
1350 	tun_debug(KERN_INFO, tun, "tun_do_read\n");
1351 
1352 	if (!len)
1353 		return ret;
1354 
1355 	if (tun->dev->reg_state != NETREG_REGISTERED)
1356 		return -EIO;
1357 
1358 	/* Read frames from queue */
1359 	skb = __skb_recv_datagram(tfile->socket.sk, noblock ? MSG_DONTWAIT : 0,
1360 				  &peeked, &off, &err);
1361 	if (skb) {
1362 		ret = tun_put_user(tun, tfile, skb, iv, len);
1363 		kfree_skb(skb);
1364 	} else
1365 		ret = err;
1366 
1367 	return ret;
1368 }
1369 
1370 static ssize_t tun_chr_aio_read(struct kiocb *iocb, const struct iovec *iv,
1371 			    unsigned long count, loff_t pos)
1372 {
1373 	struct file *file = iocb->ki_filp;
1374 	struct tun_file *tfile = file->private_data;
1375 	struct tun_struct *tun = __tun_get(tfile);
1376 	ssize_t len, ret;
1377 
1378 	if (!tun)
1379 		return -EBADFD;
1380 	len = iov_length(iv, count);
1381 	if (len < 0) {
1382 		ret = -EINVAL;
1383 		goto out;
1384 	}
1385 
1386 	ret = tun_do_read(tun, tfile, iv, len,
1387 			  file->f_flags & O_NONBLOCK);
1388 	ret = min_t(ssize_t, ret, len);
1389 	if (ret > 0)
1390 		iocb->ki_pos = ret;
1391 out:
1392 	tun_put(tun);
1393 	return ret;
1394 }
1395 
1396 static void tun_free_netdev(struct net_device *dev)
1397 {
1398 	struct tun_struct *tun = netdev_priv(dev);
1399 
1400 	BUG_ON(!(list_empty(&tun->disabled)));
1401 	tun_flow_uninit(tun);
1402 	security_tun_dev_free_security(tun->security);
1403 	free_netdev(dev);
1404 }
1405 
1406 static void tun_setup(struct net_device *dev)
1407 {
1408 	struct tun_struct *tun = netdev_priv(dev);
1409 
1410 	tun->owner = INVALID_UID;
1411 	tun->group = INVALID_GID;
1412 
1413 	dev->ethtool_ops = &tun_ethtool_ops;
1414 	dev->destructor = tun_free_netdev;
1415 }
1416 
1417 /* Trivial set of netlink ops to allow deleting tun or tap
1418  * device with netlink.
1419  */
1420 static int tun_validate(struct nlattr *tb[], struct nlattr *data[])
1421 {
1422 	return -EINVAL;
1423 }
1424 
1425 static struct rtnl_link_ops tun_link_ops __read_mostly = {
1426 	.kind		= DRV_NAME,
1427 	.priv_size	= sizeof(struct tun_struct),
1428 	.setup		= tun_setup,
1429 	.validate	= tun_validate,
1430 };
1431 
1432 static void tun_sock_write_space(struct sock *sk)
1433 {
1434 	struct tun_file *tfile;
1435 	wait_queue_head_t *wqueue;
1436 
1437 	if (!sock_writeable(sk))
1438 		return;
1439 
1440 	if (!test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags))
1441 		return;
1442 
1443 	wqueue = sk_sleep(sk);
1444 	if (wqueue && waitqueue_active(wqueue))
1445 		wake_up_interruptible_sync_poll(wqueue, POLLOUT |
1446 						POLLWRNORM | POLLWRBAND);
1447 
1448 	tfile = container_of(sk, struct tun_file, sk);
1449 	kill_fasync(&tfile->fasync, SIGIO, POLL_OUT);
1450 }
1451 
1452 static int tun_sendmsg(struct kiocb *iocb, struct socket *sock,
1453 		       struct msghdr *m, size_t total_len)
1454 {
1455 	int ret;
1456 	struct tun_file *tfile = container_of(sock, struct tun_file, socket);
1457 	struct tun_struct *tun = __tun_get(tfile);
1458 
1459 	if (!tun)
1460 		return -EBADFD;
1461 	ret = tun_get_user(tun, tfile, m->msg_control, m->msg_iov, total_len,
1462 			   m->msg_iovlen, m->msg_flags & MSG_DONTWAIT);
1463 	tun_put(tun);
1464 	return ret;
1465 }
1466 
1467 static int tun_recvmsg(struct kiocb *iocb, struct socket *sock,
1468 		       struct msghdr *m, size_t total_len,
1469 		       int flags)
1470 {
1471 	struct tun_file *tfile = container_of(sock, struct tun_file, socket);
1472 	struct tun_struct *tun = __tun_get(tfile);
1473 	int ret;
1474 
1475 	if (!tun)
1476 		return -EBADFD;
1477 
1478 	if (flags & ~(MSG_DONTWAIT|MSG_TRUNC|MSG_ERRQUEUE)) {
1479 		ret = -EINVAL;
1480 		goto out;
1481 	}
1482 	if (flags & MSG_ERRQUEUE) {
1483 		ret = sock_recv_errqueue(sock->sk, m, total_len,
1484 					 SOL_PACKET, TUN_TX_TIMESTAMP);
1485 		goto out;
1486 	}
1487 	ret = tun_do_read(tun, tfile, m->msg_iov, total_len,
1488 			  flags & MSG_DONTWAIT);
1489 	if (ret > total_len) {
1490 		m->msg_flags |= MSG_TRUNC;
1491 		ret = flags & MSG_TRUNC ? ret : total_len;
1492 	}
1493 out:
1494 	tun_put(tun);
1495 	return ret;
1496 }
1497 
1498 static int tun_release(struct socket *sock)
1499 {
1500 	if (sock->sk)
1501 		sock_put(sock->sk);
1502 	return 0;
1503 }
1504 
1505 /* Ops structure to mimic raw sockets with tun */
1506 static const struct proto_ops tun_socket_ops = {
1507 	.sendmsg = tun_sendmsg,
1508 	.recvmsg = tun_recvmsg,
1509 	.release = tun_release,
1510 };
1511 
1512 static struct proto tun_proto = {
1513 	.name		= "tun",
1514 	.owner		= THIS_MODULE,
1515 	.obj_size	= sizeof(struct tun_file),
1516 };
1517 
1518 static int tun_flags(struct tun_struct *tun)
1519 {
1520 	int flags = 0;
1521 
1522 	if (tun->flags & TUN_TUN_DEV)
1523 		flags |= IFF_TUN;
1524 	else
1525 		flags |= IFF_TAP;
1526 
1527 	if (tun->flags & TUN_NO_PI)
1528 		flags |= IFF_NO_PI;
1529 
1530 	/* This flag has no real effect.  We track the value for backwards
1531 	 * compatibility.
1532 	 */
1533 	if (tun->flags & TUN_ONE_QUEUE)
1534 		flags |= IFF_ONE_QUEUE;
1535 
1536 	if (tun->flags & TUN_VNET_HDR)
1537 		flags |= IFF_VNET_HDR;
1538 
1539 	if (tun->flags & TUN_TAP_MQ)
1540 		flags |= IFF_MULTI_QUEUE;
1541 
1542 	if (tun->flags & TUN_PERSIST)
1543 		flags |= IFF_PERSIST;
1544 
1545 	return flags;
1546 }
1547 
1548 static ssize_t tun_show_flags(struct device *dev, struct device_attribute *attr,
1549 			      char *buf)
1550 {
1551 	struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1552 	return sprintf(buf, "0x%x\n", tun_flags(tun));
1553 }
1554 
1555 static ssize_t tun_show_owner(struct device *dev, struct device_attribute *attr,
1556 			      char *buf)
1557 {
1558 	struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1559 	return uid_valid(tun->owner)?
1560 		sprintf(buf, "%u\n",
1561 			from_kuid_munged(current_user_ns(), tun->owner)):
1562 		sprintf(buf, "-1\n");
1563 }
1564 
1565 static ssize_t tun_show_group(struct device *dev, struct device_attribute *attr,
1566 			      char *buf)
1567 {
1568 	struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1569 	return gid_valid(tun->group) ?
1570 		sprintf(buf, "%u\n",
1571 			from_kgid_munged(current_user_ns(), tun->group)):
1572 		sprintf(buf, "-1\n");
1573 }
1574 
1575 static DEVICE_ATTR(tun_flags, 0444, tun_show_flags, NULL);
1576 static DEVICE_ATTR(owner, 0444, tun_show_owner, NULL);
1577 static DEVICE_ATTR(group, 0444, tun_show_group, NULL);
1578 
1579 static int tun_set_iff(struct net *net, struct file *file, struct ifreq *ifr)
1580 {
1581 	struct tun_struct *tun;
1582 	struct tun_file *tfile = file->private_data;
1583 	struct net_device *dev;
1584 	int err;
1585 
1586 	if (tfile->detached)
1587 		return -EINVAL;
1588 
1589 	dev = __dev_get_by_name(net, ifr->ifr_name);
1590 	if (dev) {
1591 		if (ifr->ifr_flags & IFF_TUN_EXCL)
1592 			return -EBUSY;
1593 		if ((ifr->ifr_flags & IFF_TUN) && dev->netdev_ops == &tun_netdev_ops)
1594 			tun = netdev_priv(dev);
1595 		else if ((ifr->ifr_flags & IFF_TAP) && dev->netdev_ops == &tap_netdev_ops)
1596 			tun = netdev_priv(dev);
1597 		else
1598 			return -EINVAL;
1599 
1600 		if (!!(ifr->ifr_flags & IFF_MULTI_QUEUE) !=
1601 		    !!(tun->flags & TUN_TAP_MQ))
1602 			return -EINVAL;
1603 
1604 		if (tun_not_capable(tun))
1605 			return -EPERM;
1606 		err = security_tun_dev_open(tun->security);
1607 		if (err < 0)
1608 			return err;
1609 
1610 		err = tun_attach(tun, file, ifr->ifr_flags & IFF_NOFILTER);
1611 		if (err < 0)
1612 			return err;
1613 
1614 		if (tun->flags & TUN_TAP_MQ &&
1615 		    (tun->numqueues + tun->numdisabled > 1)) {
1616 			/* One or more queue has already been attached, no need
1617 			 * to initialize the device again.
1618 			 */
1619 			return 0;
1620 		}
1621 	}
1622 	else {
1623 		char *name;
1624 		unsigned long flags = 0;
1625 		int queues = ifr->ifr_flags & IFF_MULTI_QUEUE ?
1626 			     MAX_TAP_QUEUES : 1;
1627 
1628 		if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
1629 			return -EPERM;
1630 		err = security_tun_dev_create();
1631 		if (err < 0)
1632 			return err;
1633 
1634 		/* Set dev type */
1635 		if (ifr->ifr_flags & IFF_TUN) {
1636 			/* TUN device */
1637 			flags |= TUN_TUN_DEV;
1638 			name = "tun%d";
1639 		} else if (ifr->ifr_flags & IFF_TAP) {
1640 			/* TAP device */
1641 			flags |= TUN_TAP_DEV;
1642 			name = "tap%d";
1643 		} else
1644 			return -EINVAL;
1645 
1646 		if (*ifr->ifr_name)
1647 			name = ifr->ifr_name;
1648 
1649 		dev = alloc_netdev_mqs(sizeof(struct tun_struct), name,
1650 				       NET_NAME_UNKNOWN, tun_setup, queues,
1651 				       queues);
1652 
1653 		if (!dev)
1654 			return -ENOMEM;
1655 
1656 		dev_net_set(dev, net);
1657 		dev->rtnl_link_ops = &tun_link_ops;
1658 		dev->ifindex = tfile->ifindex;
1659 
1660 		tun = netdev_priv(dev);
1661 		tun->dev = dev;
1662 		tun->flags = flags;
1663 		tun->txflt.count = 0;
1664 		tun->vnet_hdr_sz = sizeof(struct virtio_net_hdr);
1665 
1666 		tun->filter_attached = false;
1667 		tun->sndbuf = tfile->socket.sk->sk_sndbuf;
1668 
1669 		spin_lock_init(&tun->lock);
1670 
1671 		err = security_tun_dev_alloc_security(&tun->security);
1672 		if (err < 0)
1673 			goto err_free_dev;
1674 
1675 		tun_net_init(dev);
1676 		tun_flow_init(tun);
1677 
1678 		dev->hw_features = NETIF_F_SG | NETIF_F_FRAGLIST |
1679 				   TUN_USER_FEATURES | NETIF_F_HW_VLAN_CTAG_TX |
1680 				   NETIF_F_HW_VLAN_STAG_TX;
1681 		dev->features = dev->hw_features;
1682 		dev->vlan_features = dev->features &
1683 				     ~(NETIF_F_HW_VLAN_CTAG_TX |
1684 				       NETIF_F_HW_VLAN_STAG_TX);
1685 
1686 		INIT_LIST_HEAD(&tun->disabled);
1687 		err = tun_attach(tun, file, false);
1688 		if (err < 0)
1689 			goto err_free_flow;
1690 
1691 		err = register_netdevice(tun->dev);
1692 		if (err < 0)
1693 			goto err_detach;
1694 
1695 		if (device_create_file(&tun->dev->dev, &dev_attr_tun_flags) ||
1696 		    device_create_file(&tun->dev->dev, &dev_attr_owner) ||
1697 		    device_create_file(&tun->dev->dev, &dev_attr_group))
1698 			pr_err("Failed to create tun sysfs files\n");
1699 	}
1700 
1701 	netif_carrier_on(tun->dev);
1702 
1703 	tun_debug(KERN_INFO, tun, "tun_set_iff\n");
1704 
1705 	if (ifr->ifr_flags & IFF_NO_PI)
1706 		tun->flags |= TUN_NO_PI;
1707 	else
1708 		tun->flags &= ~TUN_NO_PI;
1709 
1710 	/* This flag has no real effect.  We track the value for backwards
1711 	 * compatibility.
1712 	 */
1713 	if (ifr->ifr_flags & IFF_ONE_QUEUE)
1714 		tun->flags |= TUN_ONE_QUEUE;
1715 	else
1716 		tun->flags &= ~TUN_ONE_QUEUE;
1717 
1718 	if (ifr->ifr_flags & IFF_VNET_HDR)
1719 		tun->flags |= TUN_VNET_HDR;
1720 	else
1721 		tun->flags &= ~TUN_VNET_HDR;
1722 
1723 	if (ifr->ifr_flags & IFF_MULTI_QUEUE)
1724 		tun->flags |= TUN_TAP_MQ;
1725 	else
1726 		tun->flags &= ~TUN_TAP_MQ;
1727 
1728 	/* Make sure persistent devices do not get stuck in
1729 	 * xoff state.
1730 	 */
1731 	if (netif_running(tun->dev))
1732 		netif_tx_wake_all_queues(tun->dev);
1733 
1734 	strcpy(ifr->ifr_name, tun->dev->name);
1735 	return 0;
1736 
1737 err_detach:
1738 	tun_detach_all(dev);
1739 err_free_flow:
1740 	tun_flow_uninit(tun);
1741 	security_tun_dev_free_security(tun->security);
1742 err_free_dev:
1743 	free_netdev(dev);
1744 	return err;
1745 }
1746 
1747 static void tun_get_iff(struct net *net, struct tun_struct *tun,
1748 		       struct ifreq *ifr)
1749 {
1750 	tun_debug(KERN_INFO, tun, "tun_get_iff\n");
1751 
1752 	strcpy(ifr->ifr_name, tun->dev->name);
1753 
1754 	ifr->ifr_flags = tun_flags(tun);
1755 
1756 }
1757 
1758 /* This is like a cut-down ethtool ops, except done via tun fd so no
1759  * privs required. */
1760 static int set_offload(struct tun_struct *tun, unsigned long arg)
1761 {
1762 	netdev_features_t features = 0;
1763 
1764 	if (arg & TUN_F_CSUM) {
1765 		features |= NETIF_F_HW_CSUM;
1766 		arg &= ~TUN_F_CSUM;
1767 
1768 		if (arg & (TUN_F_TSO4|TUN_F_TSO6)) {
1769 			if (arg & TUN_F_TSO_ECN) {
1770 				features |= NETIF_F_TSO_ECN;
1771 				arg &= ~TUN_F_TSO_ECN;
1772 			}
1773 			if (arg & TUN_F_TSO4)
1774 				features |= NETIF_F_TSO;
1775 			if (arg & TUN_F_TSO6)
1776 				features |= NETIF_F_TSO6;
1777 			arg &= ~(TUN_F_TSO4|TUN_F_TSO6);
1778 		}
1779 	}
1780 
1781 	/* This gives the user a way to test for new features in future by
1782 	 * trying to set them. */
1783 	if (arg)
1784 		return -EINVAL;
1785 
1786 	tun->set_features = features;
1787 	netdev_update_features(tun->dev);
1788 
1789 	return 0;
1790 }
1791 
1792 static void tun_detach_filter(struct tun_struct *tun, int n)
1793 {
1794 	int i;
1795 	struct tun_file *tfile;
1796 
1797 	for (i = 0; i < n; i++) {
1798 		tfile = rtnl_dereference(tun->tfiles[i]);
1799 		sk_detach_filter(tfile->socket.sk);
1800 	}
1801 
1802 	tun->filter_attached = false;
1803 }
1804 
1805 static int tun_attach_filter(struct tun_struct *tun)
1806 {
1807 	int i, ret = 0;
1808 	struct tun_file *tfile;
1809 
1810 	for (i = 0; i < tun->numqueues; i++) {
1811 		tfile = rtnl_dereference(tun->tfiles[i]);
1812 		ret = sk_attach_filter(&tun->fprog, tfile->socket.sk);
1813 		if (ret) {
1814 			tun_detach_filter(tun, i);
1815 			return ret;
1816 		}
1817 	}
1818 
1819 	tun->filter_attached = true;
1820 	return ret;
1821 }
1822 
1823 static void tun_set_sndbuf(struct tun_struct *tun)
1824 {
1825 	struct tun_file *tfile;
1826 	int i;
1827 
1828 	for (i = 0; i < tun->numqueues; i++) {
1829 		tfile = rtnl_dereference(tun->tfiles[i]);
1830 		tfile->socket.sk->sk_sndbuf = tun->sndbuf;
1831 	}
1832 }
1833 
1834 static int tun_set_queue(struct file *file, struct ifreq *ifr)
1835 {
1836 	struct tun_file *tfile = file->private_data;
1837 	struct tun_struct *tun;
1838 	int ret = 0;
1839 
1840 	rtnl_lock();
1841 
1842 	if (ifr->ifr_flags & IFF_ATTACH_QUEUE) {
1843 		tun = tfile->detached;
1844 		if (!tun) {
1845 			ret = -EINVAL;
1846 			goto unlock;
1847 		}
1848 		ret = security_tun_dev_attach_queue(tun->security);
1849 		if (ret < 0)
1850 			goto unlock;
1851 		ret = tun_attach(tun, file, false);
1852 	} else if (ifr->ifr_flags & IFF_DETACH_QUEUE) {
1853 		tun = rtnl_dereference(tfile->tun);
1854 		if (!tun || !(tun->flags & TUN_TAP_MQ) || tfile->detached)
1855 			ret = -EINVAL;
1856 		else
1857 			__tun_detach(tfile, false);
1858 	} else
1859 		ret = -EINVAL;
1860 
1861 unlock:
1862 	rtnl_unlock();
1863 	return ret;
1864 }
1865 
1866 static long __tun_chr_ioctl(struct file *file, unsigned int cmd,
1867 			    unsigned long arg, int ifreq_len)
1868 {
1869 	struct tun_file *tfile = file->private_data;
1870 	struct tun_struct *tun;
1871 	void __user* argp = (void __user*)arg;
1872 	struct ifreq ifr;
1873 	kuid_t owner;
1874 	kgid_t group;
1875 	int sndbuf;
1876 	int vnet_hdr_sz;
1877 	unsigned int ifindex;
1878 	int ret;
1879 
1880 	if (cmd == TUNSETIFF || cmd == TUNSETQUEUE || _IOC_TYPE(cmd) == 0x89) {
1881 		if (copy_from_user(&ifr, argp, ifreq_len))
1882 			return -EFAULT;
1883 	} else {
1884 		memset(&ifr, 0, sizeof(ifr));
1885 	}
1886 	if (cmd == TUNGETFEATURES) {
1887 		/* Currently this just means: "what IFF flags are valid?".
1888 		 * This is needed because we never checked for invalid flags on
1889 		 * TUNSETIFF. */
1890 		return put_user(IFF_TUN | IFF_TAP | IFF_NO_PI | IFF_ONE_QUEUE |
1891 				IFF_VNET_HDR | IFF_MULTI_QUEUE,
1892 				(unsigned int __user*)argp);
1893 	} else if (cmd == TUNSETQUEUE)
1894 		return tun_set_queue(file, &ifr);
1895 
1896 	ret = 0;
1897 	rtnl_lock();
1898 
1899 	tun = __tun_get(tfile);
1900 	if (cmd == TUNSETIFF && !tun) {
1901 		ifr.ifr_name[IFNAMSIZ-1] = '\0';
1902 
1903 		ret = tun_set_iff(tfile->net, file, &ifr);
1904 
1905 		if (ret)
1906 			goto unlock;
1907 
1908 		if (copy_to_user(argp, &ifr, ifreq_len))
1909 			ret = -EFAULT;
1910 		goto unlock;
1911 	}
1912 	if (cmd == TUNSETIFINDEX) {
1913 		ret = -EPERM;
1914 		if (tun)
1915 			goto unlock;
1916 
1917 		ret = -EFAULT;
1918 		if (copy_from_user(&ifindex, argp, sizeof(ifindex)))
1919 			goto unlock;
1920 
1921 		ret = 0;
1922 		tfile->ifindex = ifindex;
1923 		goto unlock;
1924 	}
1925 
1926 	ret = -EBADFD;
1927 	if (!tun)
1928 		goto unlock;
1929 
1930 	tun_debug(KERN_INFO, tun, "tun_chr_ioctl cmd %u\n", cmd);
1931 
1932 	ret = 0;
1933 	switch (cmd) {
1934 	case TUNGETIFF:
1935 		tun_get_iff(current->nsproxy->net_ns, tun, &ifr);
1936 
1937 		if (tfile->detached)
1938 			ifr.ifr_flags |= IFF_DETACH_QUEUE;
1939 		if (!tfile->socket.sk->sk_filter)
1940 			ifr.ifr_flags |= IFF_NOFILTER;
1941 
1942 		if (copy_to_user(argp, &ifr, ifreq_len))
1943 			ret = -EFAULT;
1944 		break;
1945 
1946 	case TUNSETNOCSUM:
1947 		/* Disable/Enable checksum */
1948 
1949 		/* [unimplemented] */
1950 		tun_debug(KERN_INFO, tun, "ignored: set checksum %s\n",
1951 			  arg ? "disabled" : "enabled");
1952 		break;
1953 
1954 	case TUNSETPERSIST:
1955 		/* Disable/Enable persist mode. Keep an extra reference to the
1956 		 * module to prevent the module being unprobed.
1957 		 */
1958 		if (arg && !(tun->flags & TUN_PERSIST)) {
1959 			tun->flags |= TUN_PERSIST;
1960 			__module_get(THIS_MODULE);
1961 		}
1962 		if (!arg && (tun->flags & TUN_PERSIST)) {
1963 			tun->flags &= ~TUN_PERSIST;
1964 			module_put(THIS_MODULE);
1965 		}
1966 
1967 		tun_debug(KERN_INFO, tun, "persist %s\n",
1968 			  arg ? "enabled" : "disabled");
1969 		break;
1970 
1971 	case TUNSETOWNER:
1972 		/* Set owner of the device */
1973 		owner = make_kuid(current_user_ns(), arg);
1974 		if (!uid_valid(owner)) {
1975 			ret = -EINVAL;
1976 			break;
1977 		}
1978 		tun->owner = owner;
1979 		tun_debug(KERN_INFO, tun, "owner set to %u\n",
1980 			  from_kuid(&init_user_ns, tun->owner));
1981 		break;
1982 
1983 	case TUNSETGROUP:
1984 		/* Set group of the device */
1985 		group = make_kgid(current_user_ns(), arg);
1986 		if (!gid_valid(group)) {
1987 			ret = -EINVAL;
1988 			break;
1989 		}
1990 		tun->group = group;
1991 		tun_debug(KERN_INFO, tun, "group set to %u\n",
1992 			  from_kgid(&init_user_ns, tun->group));
1993 		break;
1994 
1995 	case TUNSETLINK:
1996 		/* Only allow setting the type when the interface is down */
1997 		if (tun->dev->flags & IFF_UP) {
1998 			tun_debug(KERN_INFO, tun,
1999 				  "Linktype set failed because interface is up\n");
2000 			ret = -EBUSY;
2001 		} else {
2002 			tun->dev->type = (int) arg;
2003 			tun_debug(KERN_INFO, tun, "linktype set to %d\n",
2004 				  tun->dev->type);
2005 			ret = 0;
2006 		}
2007 		break;
2008 
2009 #ifdef TUN_DEBUG
2010 	case TUNSETDEBUG:
2011 		tun->debug = arg;
2012 		break;
2013 #endif
2014 	case TUNSETOFFLOAD:
2015 		ret = set_offload(tun, arg);
2016 		break;
2017 
2018 	case TUNSETTXFILTER:
2019 		/* Can be set only for TAPs */
2020 		ret = -EINVAL;
2021 		if ((tun->flags & TUN_TYPE_MASK) != TUN_TAP_DEV)
2022 			break;
2023 		ret = update_filter(&tun->txflt, (void __user *)arg);
2024 		break;
2025 
2026 	case SIOCGIFHWADDR:
2027 		/* Get hw address */
2028 		memcpy(ifr.ifr_hwaddr.sa_data, tun->dev->dev_addr, ETH_ALEN);
2029 		ifr.ifr_hwaddr.sa_family = tun->dev->type;
2030 		if (copy_to_user(argp, &ifr, ifreq_len))
2031 			ret = -EFAULT;
2032 		break;
2033 
2034 	case SIOCSIFHWADDR:
2035 		/* Set hw address */
2036 		tun_debug(KERN_DEBUG, tun, "set hw address: %pM\n",
2037 			  ifr.ifr_hwaddr.sa_data);
2038 
2039 		ret = dev_set_mac_address(tun->dev, &ifr.ifr_hwaddr);
2040 		break;
2041 
2042 	case TUNGETSNDBUF:
2043 		sndbuf = tfile->socket.sk->sk_sndbuf;
2044 		if (copy_to_user(argp, &sndbuf, sizeof(sndbuf)))
2045 			ret = -EFAULT;
2046 		break;
2047 
2048 	case TUNSETSNDBUF:
2049 		if (copy_from_user(&sndbuf, argp, sizeof(sndbuf))) {
2050 			ret = -EFAULT;
2051 			break;
2052 		}
2053 
2054 		tun->sndbuf = sndbuf;
2055 		tun_set_sndbuf(tun);
2056 		break;
2057 
2058 	case TUNGETVNETHDRSZ:
2059 		vnet_hdr_sz = tun->vnet_hdr_sz;
2060 		if (copy_to_user(argp, &vnet_hdr_sz, sizeof(vnet_hdr_sz)))
2061 			ret = -EFAULT;
2062 		break;
2063 
2064 	case TUNSETVNETHDRSZ:
2065 		if (copy_from_user(&vnet_hdr_sz, argp, sizeof(vnet_hdr_sz))) {
2066 			ret = -EFAULT;
2067 			break;
2068 		}
2069 		if (vnet_hdr_sz < (int)sizeof(struct virtio_net_hdr)) {
2070 			ret = -EINVAL;
2071 			break;
2072 		}
2073 
2074 		tun->vnet_hdr_sz = vnet_hdr_sz;
2075 		break;
2076 
2077 	case TUNATTACHFILTER:
2078 		/* Can be set only for TAPs */
2079 		ret = -EINVAL;
2080 		if ((tun->flags & TUN_TYPE_MASK) != TUN_TAP_DEV)
2081 			break;
2082 		ret = -EFAULT;
2083 		if (copy_from_user(&tun->fprog, argp, sizeof(tun->fprog)))
2084 			break;
2085 
2086 		ret = tun_attach_filter(tun);
2087 		break;
2088 
2089 	case TUNDETACHFILTER:
2090 		/* Can be set only for TAPs */
2091 		ret = -EINVAL;
2092 		if ((tun->flags & TUN_TYPE_MASK) != TUN_TAP_DEV)
2093 			break;
2094 		ret = 0;
2095 		tun_detach_filter(tun, tun->numqueues);
2096 		break;
2097 
2098 	case TUNGETFILTER:
2099 		ret = -EINVAL;
2100 		if ((tun->flags & TUN_TYPE_MASK) != TUN_TAP_DEV)
2101 			break;
2102 		ret = -EFAULT;
2103 		if (copy_to_user(argp, &tun->fprog, sizeof(tun->fprog)))
2104 			break;
2105 		ret = 0;
2106 		break;
2107 
2108 	default:
2109 		ret = -EINVAL;
2110 		break;
2111 	}
2112 
2113 unlock:
2114 	rtnl_unlock();
2115 	if (tun)
2116 		tun_put(tun);
2117 	return ret;
2118 }
2119 
2120 static long tun_chr_ioctl(struct file *file,
2121 			  unsigned int cmd, unsigned long arg)
2122 {
2123 	return __tun_chr_ioctl(file, cmd, arg, sizeof (struct ifreq));
2124 }
2125 
2126 #ifdef CONFIG_COMPAT
2127 static long tun_chr_compat_ioctl(struct file *file,
2128 			 unsigned int cmd, unsigned long arg)
2129 {
2130 	switch (cmd) {
2131 	case TUNSETIFF:
2132 	case TUNGETIFF:
2133 	case TUNSETTXFILTER:
2134 	case TUNGETSNDBUF:
2135 	case TUNSETSNDBUF:
2136 	case SIOCGIFHWADDR:
2137 	case SIOCSIFHWADDR:
2138 		arg = (unsigned long)compat_ptr(arg);
2139 		break;
2140 	default:
2141 		arg = (compat_ulong_t)arg;
2142 		break;
2143 	}
2144 
2145 	/*
2146 	 * compat_ifreq is shorter than ifreq, so we must not access beyond
2147 	 * the end of that structure. All fields that are used in this
2148 	 * driver are compatible though, we don't need to convert the
2149 	 * contents.
2150 	 */
2151 	return __tun_chr_ioctl(file, cmd, arg, sizeof(struct compat_ifreq));
2152 }
2153 #endif /* CONFIG_COMPAT */
2154 
2155 static int tun_chr_fasync(int fd, struct file *file, int on)
2156 {
2157 	struct tun_file *tfile = file->private_data;
2158 	int ret;
2159 
2160 	if ((ret = fasync_helper(fd, file, on, &tfile->fasync)) < 0)
2161 		goto out;
2162 
2163 	if (on) {
2164 		__f_setown(file, task_pid(current), PIDTYPE_PID, 0);
2165 		tfile->flags |= TUN_FASYNC;
2166 	} else
2167 		tfile->flags &= ~TUN_FASYNC;
2168 	ret = 0;
2169 out:
2170 	return ret;
2171 }
2172 
2173 static int tun_chr_open(struct inode *inode, struct file * file)
2174 {
2175 	struct tun_file *tfile;
2176 
2177 	DBG1(KERN_INFO, "tunX: tun_chr_open\n");
2178 
2179 	tfile = (struct tun_file *)sk_alloc(&init_net, AF_UNSPEC, GFP_KERNEL,
2180 					    &tun_proto);
2181 	if (!tfile)
2182 		return -ENOMEM;
2183 	RCU_INIT_POINTER(tfile->tun, NULL);
2184 	tfile->net = get_net(current->nsproxy->net_ns);
2185 	tfile->flags = 0;
2186 	tfile->ifindex = 0;
2187 
2188 	init_waitqueue_head(&tfile->wq.wait);
2189 	RCU_INIT_POINTER(tfile->socket.wq, &tfile->wq);
2190 
2191 	tfile->socket.file = file;
2192 	tfile->socket.ops = &tun_socket_ops;
2193 
2194 	sock_init_data(&tfile->socket, &tfile->sk);
2195 	sk_change_net(&tfile->sk, tfile->net);
2196 
2197 	tfile->sk.sk_write_space = tun_sock_write_space;
2198 	tfile->sk.sk_sndbuf = INT_MAX;
2199 
2200 	file->private_data = tfile;
2201 	set_bit(SOCK_EXTERNALLY_ALLOCATED, &tfile->socket.flags);
2202 	INIT_LIST_HEAD(&tfile->next);
2203 
2204 	sock_set_flag(&tfile->sk, SOCK_ZEROCOPY);
2205 
2206 	return 0;
2207 }
2208 
2209 static int tun_chr_close(struct inode *inode, struct file *file)
2210 {
2211 	struct tun_file *tfile = file->private_data;
2212 	struct net *net = tfile->net;
2213 
2214 	tun_detach(tfile, true);
2215 	put_net(net);
2216 
2217 	return 0;
2218 }
2219 
2220 #ifdef CONFIG_PROC_FS
2221 static int tun_chr_show_fdinfo(struct seq_file *m, struct file *f)
2222 {
2223 	struct tun_struct *tun;
2224 	struct ifreq ifr;
2225 
2226 	memset(&ifr, 0, sizeof(ifr));
2227 
2228 	rtnl_lock();
2229 	tun = tun_get(f);
2230 	if (tun)
2231 		tun_get_iff(current->nsproxy->net_ns, tun, &ifr);
2232 	rtnl_unlock();
2233 
2234 	if (tun)
2235 		tun_put(tun);
2236 
2237 	return seq_printf(m, "iff:\t%s\n", ifr.ifr_name);
2238 }
2239 #endif
2240 
2241 static const struct file_operations tun_fops = {
2242 	.owner	= THIS_MODULE,
2243 	.llseek = no_llseek,
2244 	.read  = do_sync_read,
2245 	.aio_read  = tun_chr_aio_read,
2246 	.write = do_sync_write,
2247 	.aio_write = tun_chr_aio_write,
2248 	.poll	= tun_chr_poll,
2249 	.unlocked_ioctl	= tun_chr_ioctl,
2250 #ifdef CONFIG_COMPAT
2251 	.compat_ioctl = tun_chr_compat_ioctl,
2252 #endif
2253 	.open	= tun_chr_open,
2254 	.release = tun_chr_close,
2255 	.fasync = tun_chr_fasync,
2256 #ifdef CONFIG_PROC_FS
2257 	.show_fdinfo = tun_chr_show_fdinfo,
2258 #endif
2259 };
2260 
2261 static struct miscdevice tun_miscdev = {
2262 	.minor = TUN_MINOR,
2263 	.name = "tun",
2264 	.nodename = "net/tun",
2265 	.fops = &tun_fops,
2266 };
2267 
2268 /* ethtool interface */
2269 
2270 static int tun_get_settings(struct net_device *dev, struct ethtool_cmd *cmd)
2271 {
2272 	cmd->supported		= 0;
2273 	cmd->advertising	= 0;
2274 	ethtool_cmd_speed_set(cmd, SPEED_10);
2275 	cmd->duplex		= DUPLEX_FULL;
2276 	cmd->port		= PORT_TP;
2277 	cmd->phy_address	= 0;
2278 	cmd->transceiver	= XCVR_INTERNAL;
2279 	cmd->autoneg		= AUTONEG_DISABLE;
2280 	cmd->maxtxpkt		= 0;
2281 	cmd->maxrxpkt		= 0;
2282 	return 0;
2283 }
2284 
2285 static void tun_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
2286 {
2287 	struct tun_struct *tun = netdev_priv(dev);
2288 
2289 	strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
2290 	strlcpy(info->version, DRV_VERSION, sizeof(info->version));
2291 
2292 	switch (tun->flags & TUN_TYPE_MASK) {
2293 	case TUN_TUN_DEV:
2294 		strlcpy(info->bus_info, "tun", sizeof(info->bus_info));
2295 		break;
2296 	case TUN_TAP_DEV:
2297 		strlcpy(info->bus_info, "tap", sizeof(info->bus_info));
2298 		break;
2299 	}
2300 }
2301 
2302 static u32 tun_get_msglevel(struct net_device *dev)
2303 {
2304 #ifdef TUN_DEBUG
2305 	struct tun_struct *tun = netdev_priv(dev);
2306 	return tun->debug;
2307 #else
2308 	return -EOPNOTSUPP;
2309 #endif
2310 }
2311 
2312 static void tun_set_msglevel(struct net_device *dev, u32 value)
2313 {
2314 #ifdef TUN_DEBUG
2315 	struct tun_struct *tun = netdev_priv(dev);
2316 	tun->debug = value;
2317 #endif
2318 }
2319 
2320 static const struct ethtool_ops tun_ethtool_ops = {
2321 	.get_settings	= tun_get_settings,
2322 	.get_drvinfo	= tun_get_drvinfo,
2323 	.get_msglevel	= tun_get_msglevel,
2324 	.set_msglevel	= tun_set_msglevel,
2325 	.get_link	= ethtool_op_get_link,
2326 	.get_ts_info	= ethtool_op_get_ts_info,
2327 };
2328 
2329 
2330 static int __init tun_init(void)
2331 {
2332 	int ret = 0;
2333 
2334 	pr_info("%s, %s\n", DRV_DESCRIPTION, DRV_VERSION);
2335 	pr_info("%s\n", DRV_COPYRIGHT);
2336 
2337 	ret = rtnl_link_register(&tun_link_ops);
2338 	if (ret) {
2339 		pr_err("Can't register link_ops\n");
2340 		goto err_linkops;
2341 	}
2342 
2343 	ret = misc_register(&tun_miscdev);
2344 	if (ret) {
2345 		pr_err("Can't register misc device %d\n", TUN_MINOR);
2346 		goto err_misc;
2347 	}
2348 	return  0;
2349 err_misc:
2350 	rtnl_link_unregister(&tun_link_ops);
2351 err_linkops:
2352 	return ret;
2353 }
2354 
2355 static void tun_cleanup(void)
2356 {
2357 	misc_deregister(&tun_miscdev);
2358 	rtnl_link_unregister(&tun_link_ops);
2359 }
2360 
2361 /* Get an underlying socket object from tun file.  Returns error unless file is
2362  * attached to a device.  The returned object works like a packet socket, it
2363  * can be used for sock_sendmsg/sock_recvmsg.  The caller is responsible for
2364  * holding a reference to the file for as long as the socket is in use. */
2365 struct socket *tun_get_socket(struct file *file)
2366 {
2367 	struct tun_file *tfile;
2368 	if (file->f_op != &tun_fops)
2369 		return ERR_PTR(-EINVAL);
2370 	tfile = file->private_data;
2371 	if (!tfile)
2372 		return ERR_PTR(-EBADFD);
2373 	return &tfile->socket;
2374 }
2375 EXPORT_SYMBOL_GPL(tun_get_socket);
2376 
2377 module_init(tun_init);
2378 module_exit(tun_cleanup);
2379 MODULE_DESCRIPTION(DRV_DESCRIPTION);
2380 MODULE_AUTHOR(DRV_COPYRIGHT);
2381 MODULE_LICENSE("GPL");
2382 MODULE_ALIAS_MISCDEV(TUN_MINOR);
2383 MODULE_ALIAS("devname:net/tun");
2384