xref: /openbmc/linux/drivers/net/tun.c (revision 7d82410950aa74adccf035c332e409af2bb93e92)
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 bool tun_is_little_endian(struct tun_struct *tun)
210 {
211 	return tun->flags & TUN_VNET_LE ||
212 		virtio_legacy_is_little_endian();
213 }
214 
215 static inline u16 tun16_to_cpu(struct tun_struct *tun, __virtio16 val)
216 {
217 	return __virtio16_to_cpu(tun_is_little_endian(tun), val);
218 }
219 
220 static inline __virtio16 cpu_to_tun16(struct tun_struct *tun, u16 val)
221 {
222 	return __cpu_to_virtio16(tun_is_little_endian(tun), val);
223 }
224 
225 static inline u32 tun_hashfn(u32 rxhash)
226 {
227 	return rxhash & 0x3ff;
228 }
229 
230 static struct tun_flow_entry *tun_flow_find(struct hlist_head *head, u32 rxhash)
231 {
232 	struct tun_flow_entry *e;
233 
234 	hlist_for_each_entry_rcu(e, head, hash_link) {
235 		if (e->rxhash == rxhash)
236 			return e;
237 	}
238 	return NULL;
239 }
240 
241 static struct tun_flow_entry *tun_flow_create(struct tun_struct *tun,
242 					      struct hlist_head *head,
243 					      u32 rxhash, u16 queue_index)
244 {
245 	struct tun_flow_entry *e = kmalloc(sizeof(*e), GFP_ATOMIC);
246 
247 	if (e) {
248 		tun_debug(KERN_INFO, tun, "create flow: hash %u index %u\n",
249 			  rxhash, queue_index);
250 		e->updated = jiffies;
251 		e->rxhash = rxhash;
252 		e->rps_rxhash = 0;
253 		e->queue_index = queue_index;
254 		e->tun = tun;
255 		hlist_add_head_rcu(&e->hash_link, head);
256 		++tun->flow_count;
257 	}
258 	return e;
259 }
260 
261 static void tun_flow_delete(struct tun_struct *tun, struct tun_flow_entry *e)
262 {
263 	tun_debug(KERN_INFO, tun, "delete flow: hash %u index %u\n",
264 		  e->rxhash, e->queue_index);
265 	hlist_del_rcu(&e->hash_link);
266 	kfree_rcu(e, rcu);
267 	--tun->flow_count;
268 }
269 
270 static void tun_flow_flush(struct tun_struct *tun)
271 {
272 	int i;
273 
274 	spin_lock_bh(&tun->lock);
275 	for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
276 		struct tun_flow_entry *e;
277 		struct hlist_node *n;
278 
279 		hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link)
280 			tun_flow_delete(tun, e);
281 	}
282 	spin_unlock_bh(&tun->lock);
283 }
284 
285 static void tun_flow_delete_by_queue(struct tun_struct *tun, u16 queue_index)
286 {
287 	int i;
288 
289 	spin_lock_bh(&tun->lock);
290 	for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
291 		struct tun_flow_entry *e;
292 		struct hlist_node *n;
293 
294 		hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
295 			if (e->queue_index == queue_index)
296 				tun_flow_delete(tun, e);
297 		}
298 	}
299 	spin_unlock_bh(&tun->lock);
300 }
301 
302 static void tun_flow_cleanup(unsigned long data)
303 {
304 	struct tun_struct *tun = (struct tun_struct *)data;
305 	unsigned long delay = tun->ageing_time;
306 	unsigned long next_timer = jiffies + delay;
307 	unsigned long count = 0;
308 	int i;
309 
310 	tun_debug(KERN_INFO, tun, "tun_flow_cleanup\n");
311 
312 	spin_lock_bh(&tun->lock);
313 	for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
314 		struct tun_flow_entry *e;
315 		struct hlist_node *n;
316 
317 		hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
318 			unsigned long this_timer;
319 			count++;
320 			this_timer = e->updated + delay;
321 			if (time_before_eq(this_timer, jiffies))
322 				tun_flow_delete(tun, e);
323 			else if (time_before(this_timer, next_timer))
324 				next_timer = this_timer;
325 		}
326 	}
327 
328 	if (count)
329 		mod_timer(&tun->flow_gc_timer, round_jiffies_up(next_timer));
330 	spin_unlock_bh(&tun->lock);
331 }
332 
333 static void tun_flow_update(struct tun_struct *tun, u32 rxhash,
334 			    struct tun_file *tfile)
335 {
336 	struct hlist_head *head;
337 	struct tun_flow_entry *e;
338 	unsigned long delay = tun->ageing_time;
339 	u16 queue_index = tfile->queue_index;
340 
341 	if (!rxhash)
342 		return;
343 	else
344 		head = &tun->flows[tun_hashfn(rxhash)];
345 
346 	rcu_read_lock();
347 
348 	/* We may get a very small possibility of OOO during switching, not
349 	 * worth to optimize.*/
350 	if (tun->numqueues == 1 || tfile->detached)
351 		goto unlock;
352 
353 	e = tun_flow_find(head, rxhash);
354 	if (likely(e)) {
355 		/* TODO: keep queueing to old queue until it's empty? */
356 		e->queue_index = queue_index;
357 		e->updated = jiffies;
358 		sock_rps_record_flow_hash(e->rps_rxhash);
359 	} else {
360 		spin_lock_bh(&tun->lock);
361 		if (!tun_flow_find(head, rxhash) &&
362 		    tun->flow_count < MAX_TAP_FLOWS)
363 			tun_flow_create(tun, head, rxhash, queue_index);
364 
365 		if (!timer_pending(&tun->flow_gc_timer))
366 			mod_timer(&tun->flow_gc_timer,
367 				  round_jiffies_up(jiffies + delay));
368 		spin_unlock_bh(&tun->lock);
369 	}
370 
371 unlock:
372 	rcu_read_unlock();
373 }
374 
375 /**
376  * Save the hash received in the stack receive path and update the
377  * flow_hash table accordingly.
378  */
379 static inline void tun_flow_save_rps_rxhash(struct tun_flow_entry *e, u32 hash)
380 {
381 	if (unlikely(e->rps_rxhash != hash))
382 		e->rps_rxhash = hash;
383 }
384 
385 /* We try to identify a flow through its rxhash first. The reason that
386  * we do not check rxq no. is because some cards(e.g 82599), chooses
387  * the rxq based on the txq where the last packet of the flow comes. As
388  * the userspace application move between processors, we may get a
389  * different rxq no. here. If we could not get rxhash, then we would
390  * hope the rxq no. may help here.
391  */
392 static u16 tun_select_queue(struct net_device *dev, struct sk_buff *skb,
393 			    void *accel_priv, select_queue_fallback_t fallback)
394 {
395 	struct tun_struct *tun = netdev_priv(dev);
396 	struct tun_flow_entry *e;
397 	u32 txq = 0;
398 	u32 numqueues = 0;
399 
400 	rcu_read_lock();
401 	numqueues = ACCESS_ONCE(tun->numqueues);
402 
403 	txq = skb_get_hash(skb);
404 	if (txq) {
405 		e = tun_flow_find(&tun->flows[tun_hashfn(txq)], txq);
406 		if (e) {
407 			tun_flow_save_rps_rxhash(e, txq);
408 			txq = e->queue_index;
409 		} else
410 			/* use multiply and shift instead of expensive divide */
411 			txq = ((u64)txq * numqueues) >> 32;
412 	} else if (likely(skb_rx_queue_recorded(skb))) {
413 		txq = skb_get_rx_queue(skb);
414 		while (unlikely(txq >= numqueues))
415 			txq -= numqueues;
416 	}
417 
418 	rcu_read_unlock();
419 	return txq;
420 }
421 
422 static inline bool tun_not_capable(struct tun_struct *tun)
423 {
424 	const struct cred *cred = current_cred();
425 	struct net *net = dev_net(tun->dev);
426 
427 	return ((uid_valid(tun->owner) && !uid_eq(cred->euid, tun->owner)) ||
428 		  (gid_valid(tun->group) && !in_egroup_p(tun->group))) &&
429 		!ns_capable(net->user_ns, CAP_NET_ADMIN);
430 }
431 
432 static void tun_set_real_num_queues(struct tun_struct *tun)
433 {
434 	netif_set_real_num_tx_queues(tun->dev, tun->numqueues);
435 	netif_set_real_num_rx_queues(tun->dev, tun->numqueues);
436 }
437 
438 static void tun_disable_queue(struct tun_struct *tun, struct tun_file *tfile)
439 {
440 	tfile->detached = tun;
441 	list_add_tail(&tfile->next, &tun->disabled);
442 	++tun->numdisabled;
443 }
444 
445 static struct tun_struct *tun_enable_queue(struct tun_file *tfile)
446 {
447 	struct tun_struct *tun = tfile->detached;
448 
449 	tfile->detached = NULL;
450 	list_del_init(&tfile->next);
451 	--tun->numdisabled;
452 	return tun;
453 }
454 
455 static void tun_queue_purge(struct tun_file *tfile)
456 {
457 	skb_queue_purge(&tfile->sk.sk_receive_queue);
458 	skb_queue_purge(&tfile->sk.sk_error_queue);
459 }
460 
461 static void __tun_detach(struct tun_file *tfile, bool clean)
462 {
463 	struct tun_file *ntfile;
464 	struct tun_struct *tun;
465 
466 	tun = rtnl_dereference(tfile->tun);
467 
468 	if (tun && !tfile->detached) {
469 		u16 index = tfile->queue_index;
470 		BUG_ON(index >= tun->numqueues);
471 
472 		rcu_assign_pointer(tun->tfiles[index],
473 				   tun->tfiles[tun->numqueues - 1]);
474 		ntfile = rtnl_dereference(tun->tfiles[index]);
475 		ntfile->queue_index = index;
476 
477 		--tun->numqueues;
478 		if (clean) {
479 			RCU_INIT_POINTER(tfile->tun, NULL);
480 			sock_put(&tfile->sk);
481 		} else
482 			tun_disable_queue(tun, tfile);
483 
484 		synchronize_net();
485 		tun_flow_delete_by_queue(tun, tun->numqueues + 1);
486 		/* Drop read queue */
487 		tun_queue_purge(tfile);
488 		tun_set_real_num_queues(tun);
489 	} else if (tfile->detached && clean) {
490 		tun = tun_enable_queue(tfile);
491 		sock_put(&tfile->sk);
492 	}
493 
494 	if (clean) {
495 		if (tun && tun->numqueues == 0 && tun->numdisabled == 0) {
496 			netif_carrier_off(tun->dev);
497 
498 			if (!(tun->flags & IFF_PERSIST) &&
499 			    tun->dev->reg_state == NETREG_REGISTERED)
500 				unregister_netdevice(tun->dev);
501 		}
502 
503 		BUG_ON(!test_bit(SOCK_EXTERNALLY_ALLOCATED,
504 				 &tfile->socket.flags));
505 		sk_release_kernel(&tfile->sk);
506 	}
507 }
508 
509 static void tun_detach(struct tun_file *tfile, bool clean)
510 {
511 	rtnl_lock();
512 	__tun_detach(tfile, clean);
513 	rtnl_unlock();
514 }
515 
516 static void tun_detach_all(struct net_device *dev)
517 {
518 	struct tun_struct *tun = netdev_priv(dev);
519 	struct tun_file *tfile, *tmp;
520 	int i, n = tun->numqueues;
521 
522 	for (i = 0; i < n; i++) {
523 		tfile = rtnl_dereference(tun->tfiles[i]);
524 		BUG_ON(!tfile);
525 		tfile->socket.sk->sk_data_ready(tfile->socket.sk);
526 		RCU_INIT_POINTER(tfile->tun, NULL);
527 		--tun->numqueues;
528 	}
529 	list_for_each_entry(tfile, &tun->disabled, next) {
530 		tfile->socket.sk->sk_data_ready(tfile->socket.sk);
531 		RCU_INIT_POINTER(tfile->tun, NULL);
532 	}
533 	BUG_ON(tun->numqueues != 0);
534 
535 	synchronize_net();
536 	for (i = 0; i < n; i++) {
537 		tfile = rtnl_dereference(tun->tfiles[i]);
538 		/* Drop read queue */
539 		tun_queue_purge(tfile);
540 		sock_put(&tfile->sk);
541 	}
542 	list_for_each_entry_safe(tfile, tmp, &tun->disabled, next) {
543 		tun_enable_queue(tfile);
544 		tun_queue_purge(tfile);
545 		sock_put(&tfile->sk);
546 	}
547 	BUG_ON(tun->numdisabled != 0);
548 
549 	if (tun->flags & IFF_PERSIST)
550 		module_put(THIS_MODULE);
551 }
552 
553 static int tun_attach(struct tun_struct *tun, struct file *file, bool skip_filter)
554 {
555 	struct tun_file *tfile = file->private_data;
556 	int err;
557 
558 	err = security_tun_dev_attach(tfile->socket.sk, tun->security);
559 	if (err < 0)
560 		goto out;
561 
562 	err = -EINVAL;
563 	if (rtnl_dereference(tfile->tun) && !tfile->detached)
564 		goto out;
565 
566 	err = -EBUSY;
567 	if (!(tun->flags & IFF_MULTI_QUEUE) && tun->numqueues == 1)
568 		goto out;
569 
570 	err = -E2BIG;
571 	if (!tfile->detached &&
572 	    tun->numqueues + tun->numdisabled == MAX_TAP_QUEUES)
573 		goto out;
574 
575 	err = 0;
576 
577 	/* Re-attach the filter to persist device */
578 	if (!skip_filter && (tun->filter_attached == true)) {
579 		err = sk_attach_filter(&tun->fprog, tfile->socket.sk);
580 		if (!err)
581 			goto out;
582 	}
583 	tfile->queue_index = tun->numqueues;
584 	rcu_assign_pointer(tfile->tun, tun);
585 	rcu_assign_pointer(tun->tfiles[tun->numqueues], tfile);
586 	tun->numqueues++;
587 
588 	if (tfile->detached)
589 		tun_enable_queue(tfile);
590 	else
591 		sock_hold(&tfile->sk);
592 
593 	tun_set_real_num_queues(tun);
594 
595 	/* device is allowed to go away first, so no need to hold extra
596 	 * refcnt.
597 	 */
598 
599 out:
600 	return err;
601 }
602 
603 static struct tun_struct *__tun_get(struct tun_file *tfile)
604 {
605 	struct tun_struct *tun;
606 
607 	rcu_read_lock();
608 	tun = rcu_dereference(tfile->tun);
609 	if (tun)
610 		dev_hold(tun->dev);
611 	rcu_read_unlock();
612 
613 	return tun;
614 }
615 
616 static struct tun_struct *tun_get(struct file *file)
617 {
618 	return __tun_get(file->private_data);
619 }
620 
621 static void tun_put(struct tun_struct *tun)
622 {
623 	dev_put(tun->dev);
624 }
625 
626 /* TAP filtering */
627 static void addr_hash_set(u32 *mask, const u8 *addr)
628 {
629 	int n = ether_crc(ETH_ALEN, addr) >> 26;
630 	mask[n >> 5] |= (1 << (n & 31));
631 }
632 
633 static unsigned int addr_hash_test(const u32 *mask, const u8 *addr)
634 {
635 	int n = ether_crc(ETH_ALEN, addr) >> 26;
636 	return mask[n >> 5] & (1 << (n & 31));
637 }
638 
639 static int update_filter(struct tap_filter *filter, void __user *arg)
640 {
641 	struct { u8 u[ETH_ALEN]; } *addr;
642 	struct tun_filter uf;
643 	int err, alen, n, nexact;
644 
645 	if (copy_from_user(&uf, arg, sizeof(uf)))
646 		return -EFAULT;
647 
648 	if (!uf.count) {
649 		/* Disabled */
650 		filter->count = 0;
651 		return 0;
652 	}
653 
654 	alen = ETH_ALEN * uf.count;
655 	addr = kmalloc(alen, GFP_KERNEL);
656 	if (!addr)
657 		return -ENOMEM;
658 
659 	if (copy_from_user(addr, arg + sizeof(uf), alen)) {
660 		err = -EFAULT;
661 		goto done;
662 	}
663 
664 	/* The filter is updated without holding any locks. Which is
665 	 * perfectly safe. We disable it first and in the worst
666 	 * case we'll accept a few undesired packets. */
667 	filter->count = 0;
668 	wmb();
669 
670 	/* Use first set of addresses as an exact filter */
671 	for (n = 0; n < uf.count && n < FLT_EXACT_COUNT; n++)
672 		memcpy(filter->addr[n], addr[n].u, ETH_ALEN);
673 
674 	nexact = n;
675 
676 	/* Remaining multicast addresses are hashed,
677 	 * unicast will leave the filter disabled. */
678 	memset(filter->mask, 0, sizeof(filter->mask));
679 	for (; n < uf.count; n++) {
680 		if (!is_multicast_ether_addr(addr[n].u)) {
681 			err = 0; /* no filter */
682 			goto done;
683 		}
684 		addr_hash_set(filter->mask, addr[n].u);
685 	}
686 
687 	/* For ALLMULTI just set the mask to all ones.
688 	 * This overrides the mask populated above. */
689 	if ((uf.flags & TUN_FLT_ALLMULTI))
690 		memset(filter->mask, ~0, sizeof(filter->mask));
691 
692 	/* Now enable the filter */
693 	wmb();
694 	filter->count = nexact;
695 
696 	/* Return the number of exact filters */
697 	err = nexact;
698 
699 done:
700 	kfree(addr);
701 	return err;
702 }
703 
704 /* Returns: 0 - drop, !=0 - accept */
705 static int run_filter(struct tap_filter *filter, const struct sk_buff *skb)
706 {
707 	/* Cannot use eth_hdr(skb) here because skb_mac_hdr() is incorrect
708 	 * at this point. */
709 	struct ethhdr *eh = (struct ethhdr *) skb->data;
710 	int i;
711 
712 	/* Exact match */
713 	for (i = 0; i < filter->count; i++)
714 		if (ether_addr_equal(eh->h_dest, filter->addr[i]))
715 			return 1;
716 
717 	/* Inexact match (multicast only) */
718 	if (is_multicast_ether_addr(eh->h_dest))
719 		return addr_hash_test(filter->mask, eh->h_dest);
720 
721 	return 0;
722 }
723 
724 /*
725  * Checks whether the packet is accepted or not.
726  * Returns: 0 - drop, !=0 - accept
727  */
728 static int check_filter(struct tap_filter *filter, const struct sk_buff *skb)
729 {
730 	if (!filter->count)
731 		return 1;
732 
733 	return run_filter(filter, skb);
734 }
735 
736 /* Network device part of the driver */
737 
738 static const struct ethtool_ops tun_ethtool_ops;
739 
740 /* Net device detach from fd. */
741 static void tun_net_uninit(struct net_device *dev)
742 {
743 	tun_detach_all(dev);
744 }
745 
746 /* Net device open. */
747 static int tun_net_open(struct net_device *dev)
748 {
749 	netif_tx_start_all_queues(dev);
750 	return 0;
751 }
752 
753 /* Net device close. */
754 static int tun_net_close(struct net_device *dev)
755 {
756 	netif_tx_stop_all_queues(dev);
757 	return 0;
758 }
759 
760 /* Net device start xmit */
761 static netdev_tx_t tun_net_xmit(struct sk_buff *skb, struct net_device *dev)
762 {
763 	struct tun_struct *tun = netdev_priv(dev);
764 	int txq = skb->queue_mapping;
765 	struct tun_file *tfile;
766 	u32 numqueues = 0;
767 
768 	rcu_read_lock();
769 	tfile = rcu_dereference(tun->tfiles[txq]);
770 	numqueues = ACCESS_ONCE(tun->numqueues);
771 
772 	/* Drop packet if interface is not attached */
773 	if (txq >= numqueues)
774 		goto drop;
775 
776 	if (numqueues == 1) {
777 		/* Select queue was not called for the skbuff, so we extract the
778 		 * RPS hash and save it into the flow_table here.
779 		 */
780 		__u32 rxhash;
781 
782 		rxhash = skb_get_hash(skb);
783 		if (rxhash) {
784 			struct tun_flow_entry *e;
785 			e = tun_flow_find(&tun->flows[tun_hashfn(rxhash)],
786 					rxhash);
787 			if (e)
788 				tun_flow_save_rps_rxhash(e, rxhash);
789 		}
790 	}
791 
792 	tun_debug(KERN_INFO, tun, "tun_net_xmit %d\n", skb->len);
793 
794 	BUG_ON(!tfile);
795 
796 	/* Drop if the filter does not like it.
797 	 * This is a noop if the filter is disabled.
798 	 * Filter can be enabled only for the TAP devices. */
799 	if (!check_filter(&tun->txflt, skb))
800 		goto drop;
801 
802 	if (tfile->socket.sk->sk_filter &&
803 	    sk_filter(tfile->socket.sk, skb))
804 		goto drop;
805 
806 	/* Limit the number of packets queued by dividing txq length with the
807 	 * number of queues.
808 	 */
809 	if (skb_queue_len(&tfile->socket.sk->sk_receive_queue) * numqueues
810 			  >= dev->tx_queue_len)
811 		goto drop;
812 
813 	if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC)))
814 		goto drop;
815 
816 	if (skb->sk) {
817 		sock_tx_timestamp(skb->sk, &skb_shinfo(skb)->tx_flags);
818 		sw_tx_timestamp(skb);
819 	}
820 
821 	/* Orphan the skb - required as we might hang on to it
822 	 * for indefinite time.
823 	 */
824 	skb_orphan(skb);
825 
826 	nf_reset(skb);
827 
828 	/* Enqueue packet */
829 	skb_queue_tail(&tfile->socket.sk->sk_receive_queue, skb);
830 
831 	/* Notify and wake up reader process */
832 	if (tfile->flags & TUN_FASYNC)
833 		kill_fasync(&tfile->fasync, SIGIO, POLL_IN);
834 	tfile->socket.sk->sk_data_ready(tfile->socket.sk);
835 
836 	rcu_read_unlock();
837 	return NETDEV_TX_OK;
838 
839 drop:
840 	dev->stats.tx_dropped++;
841 	skb_tx_error(skb);
842 	kfree_skb(skb);
843 	rcu_read_unlock();
844 	return NET_XMIT_DROP;
845 }
846 
847 static void tun_net_mclist(struct net_device *dev)
848 {
849 	/*
850 	 * This callback is supposed to deal with mc filter in
851 	 * _rx_ path and has nothing to do with the _tx_ path.
852 	 * In rx path we always accept everything userspace gives us.
853 	 */
854 }
855 
856 #define MIN_MTU 68
857 #define MAX_MTU 65535
858 
859 static int
860 tun_net_change_mtu(struct net_device *dev, int new_mtu)
861 {
862 	if (new_mtu < MIN_MTU || new_mtu + dev->hard_header_len > MAX_MTU)
863 		return -EINVAL;
864 	dev->mtu = new_mtu;
865 	return 0;
866 }
867 
868 static netdev_features_t tun_net_fix_features(struct net_device *dev,
869 	netdev_features_t features)
870 {
871 	struct tun_struct *tun = netdev_priv(dev);
872 
873 	return (features & tun->set_features) | (features & ~TUN_USER_FEATURES);
874 }
875 #ifdef CONFIG_NET_POLL_CONTROLLER
876 static void tun_poll_controller(struct net_device *dev)
877 {
878 	/*
879 	 * Tun only receives frames when:
880 	 * 1) the char device endpoint gets data from user space
881 	 * 2) the tun socket gets a sendmsg call from user space
882 	 * Since both of those are synchronous operations, we are guaranteed
883 	 * never to have pending data when we poll for it
884 	 * so there is nothing to do here but return.
885 	 * We need this though so netpoll recognizes us as an interface that
886 	 * supports polling, which enables bridge devices in virt setups to
887 	 * still use netconsole
888 	 */
889 	return;
890 }
891 #endif
892 static const struct net_device_ops tun_netdev_ops = {
893 	.ndo_uninit		= tun_net_uninit,
894 	.ndo_open		= tun_net_open,
895 	.ndo_stop		= tun_net_close,
896 	.ndo_start_xmit		= tun_net_xmit,
897 	.ndo_change_mtu		= tun_net_change_mtu,
898 	.ndo_fix_features	= tun_net_fix_features,
899 	.ndo_select_queue	= tun_select_queue,
900 #ifdef CONFIG_NET_POLL_CONTROLLER
901 	.ndo_poll_controller	= tun_poll_controller,
902 #endif
903 };
904 
905 static const struct net_device_ops tap_netdev_ops = {
906 	.ndo_uninit		= tun_net_uninit,
907 	.ndo_open		= tun_net_open,
908 	.ndo_stop		= tun_net_close,
909 	.ndo_start_xmit		= tun_net_xmit,
910 	.ndo_change_mtu		= tun_net_change_mtu,
911 	.ndo_fix_features	= tun_net_fix_features,
912 	.ndo_set_rx_mode	= tun_net_mclist,
913 	.ndo_set_mac_address	= eth_mac_addr,
914 	.ndo_validate_addr	= eth_validate_addr,
915 	.ndo_select_queue	= tun_select_queue,
916 #ifdef CONFIG_NET_POLL_CONTROLLER
917 	.ndo_poll_controller	= tun_poll_controller,
918 #endif
919 };
920 
921 static void tun_flow_init(struct tun_struct *tun)
922 {
923 	int i;
924 
925 	for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++)
926 		INIT_HLIST_HEAD(&tun->flows[i]);
927 
928 	tun->ageing_time = TUN_FLOW_EXPIRE;
929 	setup_timer(&tun->flow_gc_timer, tun_flow_cleanup, (unsigned long)tun);
930 	mod_timer(&tun->flow_gc_timer,
931 		  round_jiffies_up(jiffies + tun->ageing_time));
932 }
933 
934 static void tun_flow_uninit(struct tun_struct *tun)
935 {
936 	del_timer_sync(&tun->flow_gc_timer);
937 	tun_flow_flush(tun);
938 }
939 
940 /* Initialize net device. */
941 static void tun_net_init(struct net_device *dev)
942 {
943 	struct tun_struct *tun = netdev_priv(dev);
944 
945 	switch (tun->flags & TUN_TYPE_MASK) {
946 	case IFF_TUN:
947 		dev->netdev_ops = &tun_netdev_ops;
948 
949 		/* Point-to-Point TUN Device */
950 		dev->hard_header_len = 0;
951 		dev->addr_len = 0;
952 		dev->mtu = 1500;
953 
954 		/* Zero header length */
955 		dev->type = ARPHRD_NONE;
956 		dev->flags = IFF_POINTOPOINT | IFF_NOARP | IFF_MULTICAST;
957 		dev->tx_queue_len = TUN_READQ_SIZE;  /* We prefer our own queue length */
958 		break;
959 
960 	case IFF_TAP:
961 		dev->netdev_ops = &tap_netdev_ops;
962 		/* Ethernet TAP Device */
963 		ether_setup(dev);
964 		dev->priv_flags &= ~IFF_TX_SKB_SHARING;
965 		dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
966 
967 		eth_hw_addr_random(dev);
968 
969 		dev->tx_queue_len = TUN_READQ_SIZE;  /* We prefer our own queue length */
970 		break;
971 	}
972 }
973 
974 /* Character device part */
975 
976 /* Poll */
977 static unsigned int tun_chr_poll(struct file *file, poll_table *wait)
978 {
979 	struct tun_file *tfile = file->private_data;
980 	struct tun_struct *tun = __tun_get(tfile);
981 	struct sock *sk;
982 	unsigned int mask = 0;
983 
984 	if (!tun)
985 		return POLLERR;
986 
987 	sk = tfile->socket.sk;
988 
989 	tun_debug(KERN_INFO, tun, "tun_chr_poll\n");
990 
991 	poll_wait(file, sk_sleep(sk), wait);
992 
993 	if (!skb_queue_empty(&sk->sk_receive_queue))
994 		mask |= POLLIN | POLLRDNORM;
995 
996 	if (sock_writeable(sk) ||
997 	    (!test_and_set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags) &&
998 	     sock_writeable(sk)))
999 		mask |= POLLOUT | POLLWRNORM;
1000 
1001 	if (tun->dev->reg_state != NETREG_REGISTERED)
1002 		mask = POLLERR;
1003 
1004 	tun_put(tun);
1005 	return mask;
1006 }
1007 
1008 /* prepad is the amount to reserve at front.  len is length after that.
1009  * linear is a hint as to how much to copy (usually headers). */
1010 static struct sk_buff *tun_alloc_skb(struct tun_file *tfile,
1011 				     size_t prepad, size_t len,
1012 				     size_t linear, int noblock)
1013 {
1014 	struct sock *sk = tfile->socket.sk;
1015 	struct sk_buff *skb;
1016 	int err;
1017 
1018 	/* Under a page?  Don't bother with paged skb. */
1019 	if (prepad + len < PAGE_SIZE || !linear)
1020 		linear = len;
1021 
1022 	skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
1023 				   &err, 0);
1024 	if (!skb)
1025 		return ERR_PTR(err);
1026 
1027 	skb_reserve(skb, prepad);
1028 	skb_put(skb, linear);
1029 	skb->data_len = len - linear;
1030 	skb->len += len - linear;
1031 
1032 	return skb;
1033 }
1034 
1035 /* Get packet from user space buffer */
1036 static ssize_t tun_get_user(struct tun_struct *tun, struct tun_file *tfile,
1037 			    void *msg_control, struct iov_iter *from,
1038 			    int noblock)
1039 {
1040 	struct tun_pi pi = { 0, cpu_to_be16(ETH_P_IP) };
1041 	struct sk_buff *skb;
1042 	size_t total_len = iov_iter_count(from);
1043 	size_t len = total_len, align = NET_SKB_PAD, linear;
1044 	struct virtio_net_hdr gso = { 0 };
1045 	int good_linear;
1046 	int copylen;
1047 	bool zerocopy = false;
1048 	int err;
1049 	u32 rxhash;
1050 	ssize_t n;
1051 
1052 	if (!(tun->flags & IFF_NO_PI)) {
1053 		if (len < sizeof(pi))
1054 			return -EINVAL;
1055 		len -= sizeof(pi);
1056 
1057 		n = copy_from_iter(&pi, sizeof(pi), from);
1058 		if (n != sizeof(pi))
1059 			return -EFAULT;
1060 	}
1061 
1062 	if (tun->flags & IFF_VNET_HDR) {
1063 		if (len < tun->vnet_hdr_sz)
1064 			return -EINVAL;
1065 		len -= tun->vnet_hdr_sz;
1066 
1067 		n = copy_from_iter(&gso, sizeof(gso), from);
1068 		if (n != sizeof(gso))
1069 			return -EFAULT;
1070 
1071 		if ((gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
1072 		    tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2 > tun16_to_cpu(tun, gso.hdr_len))
1073 			gso.hdr_len = cpu_to_tun16(tun, tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2);
1074 
1075 		if (tun16_to_cpu(tun, gso.hdr_len) > len)
1076 			return -EINVAL;
1077 		iov_iter_advance(from, tun->vnet_hdr_sz - sizeof(gso));
1078 	}
1079 
1080 	if ((tun->flags & TUN_TYPE_MASK) == IFF_TAP) {
1081 		align += NET_IP_ALIGN;
1082 		if (unlikely(len < ETH_HLEN ||
1083 			     (gso.hdr_len && tun16_to_cpu(tun, gso.hdr_len) < ETH_HLEN)))
1084 			return -EINVAL;
1085 	}
1086 
1087 	good_linear = SKB_MAX_HEAD(align);
1088 
1089 	if (msg_control) {
1090 		struct iov_iter i = *from;
1091 
1092 		/* There are 256 bytes to be copied in skb, so there is
1093 		 * enough room for skb expand head in case it is used.
1094 		 * The rest of the buffer is mapped from userspace.
1095 		 */
1096 		copylen = gso.hdr_len ? tun16_to_cpu(tun, gso.hdr_len) : GOODCOPY_LEN;
1097 		if (copylen > good_linear)
1098 			copylen = good_linear;
1099 		linear = copylen;
1100 		iov_iter_advance(&i, copylen);
1101 		if (iov_iter_npages(&i, INT_MAX) <= MAX_SKB_FRAGS)
1102 			zerocopy = true;
1103 	}
1104 
1105 	if (!zerocopy) {
1106 		copylen = len;
1107 		if (tun16_to_cpu(tun, gso.hdr_len) > good_linear)
1108 			linear = good_linear;
1109 		else
1110 			linear = tun16_to_cpu(tun, gso.hdr_len);
1111 	}
1112 
1113 	skb = tun_alloc_skb(tfile, align, copylen, linear, noblock);
1114 	if (IS_ERR(skb)) {
1115 		if (PTR_ERR(skb) != -EAGAIN)
1116 			tun->dev->stats.rx_dropped++;
1117 		return PTR_ERR(skb);
1118 	}
1119 
1120 	if (zerocopy)
1121 		err = zerocopy_sg_from_iter(skb, from);
1122 	else {
1123 		err = skb_copy_datagram_from_iter(skb, 0, from, len);
1124 		if (!err && msg_control) {
1125 			struct ubuf_info *uarg = msg_control;
1126 			uarg->callback(uarg, false);
1127 		}
1128 	}
1129 
1130 	if (err) {
1131 		tun->dev->stats.rx_dropped++;
1132 		kfree_skb(skb);
1133 		return -EFAULT;
1134 	}
1135 
1136 	if (gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
1137 		if (!skb_partial_csum_set(skb, tun16_to_cpu(tun, gso.csum_start),
1138 					  tun16_to_cpu(tun, gso.csum_offset))) {
1139 			tun->dev->stats.rx_frame_errors++;
1140 			kfree_skb(skb);
1141 			return -EINVAL;
1142 		}
1143 	}
1144 
1145 	switch (tun->flags & TUN_TYPE_MASK) {
1146 	case IFF_TUN:
1147 		if (tun->flags & IFF_NO_PI) {
1148 			switch (skb->data[0] & 0xf0) {
1149 			case 0x40:
1150 				pi.proto = htons(ETH_P_IP);
1151 				break;
1152 			case 0x60:
1153 				pi.proto = htons(ETH_P_IPV6);
1154 				break;
1155 			default:
1156 				tun->dev->stats.rx_dropped++;
1157 				kfree_skb(skb);
1158 				return -EINVAL;
1159 			}
1160 		}
1161 
1162 		skb_reset_mac_header(skb);
1163 		skb->protocol = pi.proto;
1164 		skb->dev = tun->dev;
1165 		break;
1166 	case IFF_TAP:
1167 		skb->protocol = eth_type_trans(skb, tun->dev);
1168 		break;
1169 	}
1170 
1171 	if (gso.gso_type != VIRTIO_NET_HDR_GSO_NONE) {
1172 		pr_debug("GSO!\n");
1173 		switch (gso.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
1174 		case VIRTIO_NET_HDR_GSO_TCPV4:
1175 			skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
1176 			break;
1177 		case VIRTIO_NET_HDR_GSO_TCPV6:
1178 			skb_shinfo(skb)->gso_type = SKB_GSO_TCPV6;
1179 			break;
1180 		case VIRTIO_NET_HDR_GSO_UDP:
1181 			skb_shinfo(skb)->gso_type = SKB_GSO_UDP;
1182 			break;
1183 		default:
1184 			tun->dev->stats.rx_frame_errors++;
1185 			kfree_skb(skb);
1186 			return -EINVAL;
1187 		}
1188 
1189 		if (gso.gso_type & VIRTIO_NET_HDR_GSO_ECN)
1190 			skb_shinfo(skb)->gso_type |= SKB_GSO_TCP_ECN;
1191 
1192 		skb_shinfo(skb)->gso_size = tun16_to_cpu(tun, gso.gso_size);
1193 		if (skb_shinfo(skb)->gso_size == 0) {
1194 			tun->dev->stats.rx_frame_errors++;
1195 			kfree_skb(skb);
1196 			return -EINVAL;
1197 		}
1198 
1199 		/* Header must be checked, and gso_segs computed. */
1200 		skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
1201 		skb_shinfo(skb)->gso_segs = 0;
1202 	}
1203 
1204 	/* copy skb_ubuf_info for callback when skb has no error */
1205 	if (zerocopy) {
1206 		skb_shinfo(skb)->destructor_arg = msg_control;
1207 		skb_shinfo(skb)->tx_flags |= SKBTX_DEV_ZEROCOPY;
1208 		skb_shinfo(skb)->tx_flags |= SKBTX_SHARED_FRAG;
1209 	}
1210 
1211 	skb_reset_network_header(skb);
1212 	skb_probe_transport_header(skb, 0);
1213 
1214 	rxhash = skb_get_hash(skb);
1215 	netif_rx_ni(skb);
1216 
1217 	tun->dev->stats.rx_packets++;
1218 	tun->dev->stats.rx_bytes += len;
1219 
1220 	tun_flow_update(tun, rxhash, tfile);
1221 	return total_len;
1222 }
1223 
1224 static ssize_t tun_chr_write_iter(struct kiocb *iocb, struct iov_iter *from)
1225 {
1226 	struct file *file = iocb->ki_filp;
1227 	struct tun_struct *tun = tun_get(file);
1228 	struct tun_file *tfile = file->private_data;
1229 	ssize_t result;
1230 
1231 	if (!tun)
1232 		return -EBADFD;
1233 
1234 	result = tun_get_user(tun, tfile, NULL, from, file->f_flags & O_NONBLOCK);
1235 
1236 	tun_put(tun);
1237 	return result;
1238 }
1239 
1240 /* Put packet to the user space buffer */
1241 static ssize_t tun_put_user(struct tun_struct *tun,
1242 			    struct tun_file *tfile,
1243 			    struct sk_buff *skb,
1244 			    struct iov_iter *iter)
1245 {
1246 	struct tun_pi pi = { 0, skb->protocol };
1247 	ssize_t total;
1248 	int vlan_offset = 0;
1249 	int vlan_hlen = 0;
1250 	int vnet_hdr_sz = 0;
1251 
1252 	if (skb_vlan_tag_present(skb))
1253 		vlan_hlen = VLAN_HLEN;
1254 
1255 	if (tun->flags & IFF_VNET_HDR)
1256 		vnet_hdr_sz = tun->vnet_hdr_sz;
1257 
1258 	total = skb->len + vlan_hlen + vnet_hdr_sz;
1259 
1260 	if (!(tun->flags & IFF_NO_PI)) {
1261 		if (iov_iter_count(iter) < sizeof(pi))
1262 			return -EINVAL;
1263 
1264 		total += sizeof(pi);
1265 		if (iov_iter_count(iter) < total) {
1266 			/* Packet will be striped */
1267 			pi.flags |= TUN_PKT_STRIP;
1268 		}
1269 
1270 		if (copy_to_iter(&pi, sizeof(pi), iter) != sizeof(pi))
1271 			return -EFAULT;
1272 	}
1273 
1274 	if (vnet_hdr_sz) {
1275 		struct virtio_net_hdr gso = { 0 }; /* no info leak */
1276 		if (iov_iter_count(iter) < vnet_hdr_sz)
1277 			return -EINVAL;
1278 
1279 		if (skb_is_gso(skb)) {
1280 			struct skb_shared_info *sinfo = skb_shinfo(skb);
1281 
1282 			/* This is a hint as to how much should be linear. */
1283 			gso.hdr_len = cpu_to_tun16(tun, skb_headlen(skb));
1284 			gso.gso_size = cpu_to_tun16(tun, sinfo->gso_size);
1285 			if (sinfo->gso_type & SKB_GSO_TCPV4)
1286 				gso.gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
1287 			else if (sinfo->gso_type & SKB_GSO_TCPV6)
1288 				gso.gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
1289 			else if (sinfo->gso_type & SKB_GSO_UDP)
1290 				gso.gso_type = VIRTIO_NET_HDR_GSO_UDP;
1291 			else {
1292 				pr_err("unexpected GSO type: "
1293 				       "0x%x, gso_size %d, hdr_len %d\n",
1294 				       sinfo->gso_type, tun16_to_cpu(tun, gso.gso_size),
1295 				       tun16_to_cpu(tun, gso.hdr_len));
1296 				print_hex_dump(KERN_ERR, "tun: ",
1297 					       DUMP_PREFIX_NONE,
1298 					       16, 1, skb->head,
1299 					       min((int)tun16_to_cpu(tun, gso.hdr_len), 64), true);
1300 				WARN_ON_ONCE(1);
1301 				return -EINVAL;
1302 			}
1303 			if (sinfo->gso_type & SKB_GSO_TCP_ECN)
1304 				gso.gso_type |= VIRTIO_NET_HDR_GSO_ECN;
1305 		} else
1306 			gso.gso_type = VIRTIO_NET_HDR_GSO_NONE;
1307 
1308 		if (skb->ip_summed == CHECKSUM_PARTIAL) {
1309 			gso.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
1310 			gso.csum_start = cpu_to_tun16(tun, skb_checksum_start_offset(skb) +
1311 						      vlan_hlen);
1312 			gso.csum_offset = cpu_to_tun16(tun, skb->csum_offset);
1313 		} else if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
1314 			gso.flags = VIRTIO_NET_HDR_F_DATA_VALID;
1315 		} /* else everything is zero */
1316 
1317 		if (copy_to_iter(&gso, sizeof(gso), iter) != sizeof(gso))
1318 			return -EFAULT;
1319 
1320 		iov_iter_advance(iter, vnet_hdr_sz - sizeof(gso));
1321 	}
1322 
1323 	if (vlan_hlen) {
1324 		int ret;
1325 		struct {
1326 			__be16 h_vlan_proto;
1327 			__be16 h_vlan_TCI;
1328 		} veth;
1329 
1330 		veth.h_vlan_proto = skb->vlan_proto;
1331 		veth.h_vlan_TCI = htons(skb_vlan_tag_get(skb));
1332 
1333 		vlan_offset = offsetof(struct vlan_ethhdr, h_vlan_proto);
1334 
1335 		ret = skb_copy_datagram_iter(skb, 0, iter, vlan_offset);
1336 		if (ret || !iov_iter_count(iter))
1337 			goto done;
1338 
1339 		ret = copy_to_iter(&veth, sizeof(veth), iter);
1340 		if (ret != sizeof(veth) || !iov_iter_count(iter))
1341 			goto done;
1342 	}
1343 
1344 	skb_copy_datagram_iter(skb, vlan_offset, iter, skb->len - vlan_offset);
1345 
1346 done:
1347 	tun->dev->stats.tx_packets++;
1348 	tun->dev->stats.tx_bytes += skb->len + vlan_hlen;
1349 
1350 	return total;
1351 }
1352 
1353 static ssize_t tun_do_read(struct tun_struct *tun, struct tun_file *tfile,
1354 			   struct iov_iter *to,
1355 			   int noblock)
1356 {
1357 	struct sk_buff *skb;
1358 	ssize_t ret;
1359 	int peeked, err, off = 0;
1360 
1361 	tun_debug(KERN_INFO, tun, "tun_do_read\n");
1362 
1363 	if (!iov_iter_count(to))
1364 		return 0;
1365 
1366 	if (tun->dev->reg_state != NETREG_REGISTERED)
1367 		return -EIO;
1368 
1369 	/* Read frames from queue */
1370 	skb = __skb_recv_datagram(tfile->socket.sk, noblock ? MSG_DONTWAIT : 0,
1371 				  &peeked, &off, &err);
1372 	if (!skb)
1373 		return err;
1374 
1375 	ret = tun_put_user(tun, tfile, skb, to);
1376 	if (unlikely(ret < 0))
1377 		kfree_skb(skb);
1378 	else
1379 		consume_skb(skb);
1380 
1381 	return ret;
1382 }
1383 
1384 static ssize_t tun_chr_read_iter(struct kiocb *iocb, struct iov_iter *to)
1385 {
1386 	struct file *file = iocb->ki_filp;
1387 	struct tun_file *tfile = file->private_data;
1388 	struct tun_struct *tun = __tun_get(tfile);
1389 	ssize_t len = iov_iter_count(to), ret;
1390 
1391 	if (!tun)
1392 		return -EBADFD;
1393 	ret = tun_do_read(tun, tfile, to, file->f_flags & O_NONBLOCK);
1394 	ret = min_t(ssize_t, ret, len);
1395 	if (ret > 0)
1396 		iocb->ki_pos = ret;
1397 	tun_put(tun);
1398 	return ret;
1399 }
1400 
1401 static void tun_free_netdev(struct net_device *dev)
1402 {
1403 	struct tun_struct *tun = netdev_priv(dev);
1404 
1405 	BUG_ON(!(list_empty(&tun->disabled)));
1406 	tun_flow_uninit(tun);
1407 	security_tun_dev_free_security(tun->security);
1408 	free_netdev(dev);
1409 }
1410 
1411 static void tun_setup(struct net_device *dev)
1412 {
1413 	struct tun_struct *tun = netdev_priv(dev);
1414 
1415 	tun->owner = INVALID_UID;
1416 	tun->group = INVALID_GID;
1417 
1418 	dev->ethtool_ops = &tun_ethtool_ops;
1419 	dev->destructor = tun_free_netdev;
1420 }
1421 
1422 /* Trivial set of netlink ops to allow deleting tun or tap
1423  * device with netlink.
1424  */
1425 static int tun_validate(struct nlattr *tb[], struct nlattr *data[])
1426 {
1427 	return -EINVAL;
1428 }
1429 
1430 static struct rtnl_link_ops tun_link_ops __read_mostly = {
1431 	.kind		= DRV_NAME,
1432 	.priv_size	= sizeof(struct tun_struct),
1433 	.setup		= tun_setup,
1434 	.validate	= tun_validate,
1435 };
1436 
1437 static void tun_sock_write_space(struct sock *sk)
1438 {
1439 	struct tun_file *tfile;
1440 	wait_queue_head_t *wqueue;
1441 
1442 	if (!sock_writeable(sk))
1443 		return;
1444 
1445 	if (!test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags))
1446 		return;
1447 
1448 	wqueue = sk_sleep(sk);
1449 	if (wqueue && waitqueue_active(wqueue))
1450 		wake_up_interruptible_sync_poll(wqueue, POLLOUT |
1451 						POLLWRNORM | POLLWRBAND);
1452 
1453 	tfile = container_of(sk, struct tun_file, sk);
1454 	kill_fasync(&tfile->fasync, SIGIO, POLL_OUT);
1455 }
1456 
1457 static int tun_sendmsg(struct socket *sock, struct msghdr *m, size_t total_len)
1458 {
1459 	int ret;
1460 	struct tun_file *tfile = container_of(sock, struct tun_file, socket);
1461 	struct tun_struct *tun = __tun_get(tfile);
1462 
1463 	if (!tun)
1464 		return -EBADFD;
1465 
1466 	ret = tun_get_user(tun, tfile, m->msg_control, &m->msg_iter,
1467 			   m->msg_flags & MSG_DONTWAIT);
1468 	tun_put(tun);
1469 	return ret;
1470 }
1471 
1472 static int tun_recvmsg(struct socket *sock, struct msghdr *m, size_t total_len,
1473 		       int flags)
1474 {
1475 	struct tun_file *tfile = container_of(sock, struct tun_file, socket);
1476 	struct tun_struct *tun = __tun_get(tfile);
1477 	int ret;
1478 
1479 	if (!tun)
1480 		return -EBADFD;
1481 
1482 	if (flags & ~(MSG_DONTWAIT|MSG_TRUNC|MSG_ERRQUEUE)) {
1483 		ret = -EINVAL;
1484 		goto out;
1485 	}
1486 	if (flags & MSG_ERRQUEUE) {
1487 		ret = sock_recv_errqueue(sock->sk, m, total_len,
1488 					 SOL_PACKET, TUN_TX_TIMESTAMP);
1489 		goto out;
1490 	}
1491 	ret = tun_do_read(tun, tfile, &m->msg_iter, flags & MSG_DONTWAIT);
1492 	if (ret > (ssize_t)total_len) {
1493 		m->msg_flags |= MSG_TRUNC;
1494 		ret = flags & MSG_TRUNC ? ret : total_len;
1495 	}
1496 out:
1497 	tun_put(tun);
1498 	return ret;
1499 }
1500 
1501 static int tun_release(struct socket *sock)
1502 {
1503 	if (sock->sk)
1504 		sock_put(sock->sk);
1505 	return 0;
1506 }
1507 
1508 /* Ops structure to mimic raw sockets with tun */
1509 static const struct proto_ops tun_socket_ops = {
1510 	.sendmsg = tun_sendmsg,
1511 	.recvmsg = tun_recvmsg,
1512 	.release = tun_release,
1513 };
1514 
1515 static struct proto tun_proto = {
1516 	.name		= "tun",
1517 	.owner		= THIS_MODULE,
1518 	.obj_size	= sizeof(struct tun_file),
1519 };
1520 
1521 static int tun_flags(struct tun_struct *tun)
1522 {
1523 	return tun->flags & (TUN_FEATURES | IFF_PERSIST | IFF_TUN | IFF_TAP);
1524 }
1525 
1526 static ssize_t tun_show_flags(struct device *dev, struct device_attribute *attr,
1527 			      char *buf)
1528 {
1529 	struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1530 	return sprintf(buf, "0x%x\n", tun_flags(tun));
1531 }
1532 
1533 static ssize_t tun_show_owner(struct device *dev, struct device_attribute *attr,
1534 			      char *buf)
1535 {
1536 	struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1537 	return uid_valid(tun->owner)?
1538 		sprintf(buf, "%u\n",
1539 			from_kuid_munged(current_user_ns(), tun->owner)):
1540 		sprintf(buf, "-1\n");
1541 }
1542 
1543 static ssize_t tun_show_group(struct device *dev, struct device_attribute *attr,
1544 			      char *buf)
1545 {
1546 	struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1547 	return gid_valid(tun->group) ?
1548 		sprintf(buf, "%u\n",
1549 			from_kgid_munged(current_user_ns(), tun->group)):
1550 		sprintf(buf, "-1\n");
1551 }
1552 
1553 static DEVICE_ATTR(tun_flags, 0444, tun_show_flags, NULL);
1554 static DEVICE_ATTR(owner, 0444, tun_show_owner, NULL);
1555 static DEVICE_ATTR(group, 0444, tun_show_group, NULL);
1556 
1557 static struct attribute *tun_dev_attrs[] = {
1558 	&dev_attr_tun_flags.attr,
1559 	&dev_attr_owner.attr,
1560 	&dev_attr_group.attr,
1561 	NULL
1562 };
1563 
1564 static const struct attribute_group tun_attr_group = {
1565 	.attrs = tun_dev_attrs
1566 };
1567 
1568 static int tun_set_iff(struct net *net, struct file *file, struct ifreq *ifr)
1569 {
1570 	struct tun_struct *tun;
1571 	struct tun_file *tfile = file->private_data;
1572 	struct net_device *dev;
1573 	int err;
1574 
1575 	if (tfile->detached)
1576 		return -EINVAL;
1577 
1578 	dev = __dev_get_by_name(net, ifr->ifr_name);
1579 	if (dev) {
1580 		if (ifr->ifr_flags & IFF_TUN_EXCL)
1581 			return -EBUSY;
1582 		if ((ifr->ifr_flags & IFF_TUN) && dev->netdev_ops == &tun_netdev_ops)
1583 			tun = netdev_priv(dev);
1584 		else if ((ifr->ifr_flags & IFF_TAP) && dev->netdev_ops == &tap_netdev_ops)
1585 			tun = netdev_priv(dev);
1586 		else
1587 			return -EINVAL;
1588 
1589 		if (!!(ifr->ifr_flags & IFF_MULTI_QUEUE) !=
1590 		    !!(tun->flags & IFF_MULTI_QUEUE))
1591 			return -EINVAL;
1592 
1593 		if (tun_not_capable(tun))
1594 			return -EPERM;
1595 		err = security_tun_dev_open(tun->security);
1596 		if (err < 0)
1597 			return err;
1598 
1599 		err = tun_attach(tun, file, ifr->ifr_flags & IFF_NOFILTER);
1600 		if (err < 0)
1601 			return err;
1602 
1603 		if (tun->flags & IFF_MULTI_QUEUE &&
1604 		    (tun->numqueues + tun->numdisabled > 1)) {
1605 			/* One or more queue has already been attached, no need
1606 			 * to initialize the device again.
1607 			 */
1608 			return 0;
1609 		}
1610 	}
1611 	else {
1612 		char *name;
1613 		unsigned long flags = 0;
1614 		int queues = ifr->ifr_flags & IFF_MULTI_QUEUE ?
1615 			     MAX_TAP_QUEUES : 1;
1616 
1617 		if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
1618 			return -EPERM;
1619 		err = security_tun_dev_create();
1620 		if (err < 0)
1621 			return err;
1622 
1623 		/* Set dev type */
1624 		if (ifr->ifr_flags & IFF_TUN) {
1625 			/* TUN device */
1626 			flags |= IFF_TUN;
1627 			name = "tun%d";
1628 		} else if (ifr->ifr_flags & IFF_TAP) {
1629 			/* TAP device */
1630 			flags |= IFF_TAP;
1631 			name = "tap%d";
1632 		} else
1633 			return -EINVAL;
1634 
1635 		if (*ifr->ifr_name)
1636 			name = ifr->ifr_name;
1637 
1638 		dev = alloc_netdev_mqs(sizeof(struct tun_struct), name,
1639 				       NET_NAME_UNKNOWN, tun_setup, queues,
1640 				       queues);
1641 
1642 		if (!dev)
1643 			return -ENOMEM;
1644 
1645 		dev_net_set(dev, net);
1646 		dev->rtnl_link_ops = &tun_link_ops;
1647 		dev->ifindex = tfile->ifindex;
1648 		dev->sysfs_groups[0] = &tun_attr_group;
1649 
1650 		tun = netdev_priv(dev);
1651 		tun->dev = dev;
1652 		tun->flags = flags;
1653 		tun->txflt.count = 0;
1654 		tun->vnet_hdr_sz = sizeof(struct virtio_net_hdr);
1655 
1656 		tun->filter_attached = false;
1657 		tun->sndbuf = tfile->socket.sk->sk_sndbuf;
1658 
1659 		spin_lock_init(&tun->lock);
1660 
1661 		err = security_tun_dev_alloc_security(&tun->security);
1662 		if (err < 0)
1663 			goto err_free_dev;
1664 
1665 		tun_net_init(dev);
1666 		tun_flow_init(tun);
1667 
1668 		dev->hw_features = NETIF_F_SG | NETIF_F_FRAGLIST |
1669 				   TUN_USER_FEATURES | NETIF_F_HW_VLAN_CTAG_TX |
1670 				   NETIF_F_HW_VLAN_STAG_TX;
1671 		dev->features = dev->hw_features;
1672 		dev->vlan_features = dev->features &
1673 				     ~(NETIF_F_HW_VLAN_CTAG_TX |
1674 				       NETIF_F_HW_VLAN_STAG_TX);
1675 
1676 		INIT_LIST_HEAD(&tun->disabled);
1677 		err = tun_attach(tun, file, false);
1678 		if (err < 0)
1679 			goto err_free_flow;
1680 
1681 		err = register_netdevice(tun->dev);
1682 		if (err < 0)
1683 			goto err_detach;
1684 	}
1685 
1686 	netif_carrier_on(tun->dev);
1687 
1688 	tun_debug(KERN_INFO, tun, "tun_set_iff\n");
1689 
1690 	tun->flags = (tun->flags & ~TUN_FEATURES) |
1691 		(ifr->ifr_flags & TUN_FEATURES);
1692 
1693 	/* Make sure persistent devices do not get stuck in
1694 	 * xoff state.
1695 	 */
1696 	if (netif_running(tun->dev))
1697 		netif_tx_wake_all_queues(tun->dev);
1698 
1699 	strcpy(ifr->ifr_name, tun->dev->name);
1700 	return 0;
1701 
1702 err_detach:
1703 	tun_detach_all(dev);
1704 err_free_flow:
1705 	tun_flow_uninit(tun);
1706 	security_tun_dev_free_security(tun->security);
1707 err_free_dev:
1708 	free_netdev(dev);
1709 	return err;
1710 }
1711 
1712 static void tun_get_iff(struct net *net, struct tun_struct *tun,
1713 		       struct ifreq *ifr)
1714 {
1715 	tun_debug(KERN_INFO, tun, "tun_get_iff\n");
1716 
1717 	strcpy(ifr->ifr_name, tun->dev->name);
1718 
1719 	ifr->ifr_flags = tun_flags(tun);
1720 
1721 }
1722 
1723 /* This is like a cut-down ethtool ops, except done via tun fd so no
1724  * privs required. */
1725 static int set_offload(struct tun_struct *tun, unsigned long arg)
1726 {
1727 	netdev_features_t features = 0;
1728 
1729 	if (arg & TUN_F_CSUM) {
1730 		features |= NETIF_F_HW_CSUM;
1731 		arg &= ~TUN_F_CSUM;
1732 
1733 		if (arg & (TUN_F_TSO4|TUN_F_TSO6)) {
1734 			if (arg & TUN_F_TSO_ECN) {
1735 				features |= NETIF_F_TSO_ECN;
1736 				arg &= ~TUN_F_TSO_ECN;
1737 			}
1738 			if (arg & TUN_F_TSO4)
1739 				features |= NETIF_F_TSO;
1740 			if (arg & TUN_F_TSO6)
1741 				features |= NETIF_F_TSO6;
1742 			arg &= ~(TUN_F_TSO4|TUN_F_TSO6);
1743 		}
1744 
1745 		if (arg & TUN_F_UFO) {
1746 			features |= NETIF_F_UFO;
1747 			arg &= ~TUN_F_UFO;
1748 		}
1749 	}
1750 
1751 	/* This gives the user a way to test for new features in future by
1752 	 * trying to set them. */
1753 	if (arg)
1754 		return -EINVAL;
1755 
1756 	tun->set_features = features;
1757 	netdev_update_features(tun->dev);
1758 
1759 	return 0;
1760 }
1761 
1762 static void tun_detach_filter(struct tun_struct *tun, int n)
1763 {
1764 	int i;
1765 	struct tun_file *tfile;
1766 
1767 	for (i = 0; i < n; i++) {
1768 		tfile = rtnl_dereference(tun->tfiles[i]);
1769 		sk_detach_filter(tfile->socket.sk);
1770 	}
1771 
1772 	tun->filter_attached = false;
1773 }
1774 
1775 static int tun_attach_filter(struct tun_struct *tun)
1776 {
1777 	int i, ret = 0;
1778 	struct tun_file *tfile;
1779 
1780 	for (i = 0; i < tun->numqueues; i++) {
1781 		tfile = rtnl_dereference(tun->tfiles[i]);
1782 		ret = sk_attach_filter(&tun->fprog, tfile->socket.sk);
1783 		if (ret) {
1784 			tun_detach_filter(tun, i);
1785 			return ret;
1786 		}
1787 	}
1788 
1789 	tun->filter_attached = true;
1790 	return ret;
1791 }
1792 
1793 static void tun_set_sndbuf(struct tun_struct *tun)
1794 {
1795 	struct tun_file *tfile;
1796 	int i;
1797 
1798 	for (i = 0; i < tun->numqueues; i++) {
1799 		tfile = rtnl_dereference(tun->tfiles[i]);
1800 		tfile->socket.sk->sk_sndbuf = tun->sndbuf;
1801 	}
1802 }
1803 
1804 static int tun_set_queue(struct file *file, struct ifreq *ifr)
1805 {
1806 	struct tun_file *tfile = file->private_data;
1807 	struct tun_struct *tun;
1808 	int ret = 0;
1809 
1810 	rtnl_lock();
1811 
1812 	if (ifr->ifr_flags & IFF_ATTACH_QUEUE) {
1813 		tun = tfile->detached;
1814 		if (!tun) {
1815 			ret = -EINVAL;
1816 			goto unlock;
1817 		}
1818 		ret = security_tun_dev_attach_queue(tun->security);
1819 		if (ret < 0)
1820 			goto unlock;
1821 		ret = tun_attach(tun, file, false);
1822 	} else if (ifr->ifr_flags & IFF_DETACH_QUEUE) {
1823 		tun = rtnl_dereference(tfile->tun);
1824 		if (!tun || !(tun->flags & IFF_MULTI_QUEUE) || tfile->detached)
1825 			ret = -EINVAL;
1826 		else
1827 			__tun_detach(tfile, false);
1828 	} else
1829 		ret = -EINVAL;
1830 
1831 unlock:
1832 	rtnl_unlock();
1833 	return ret;
1834 }
1835 
1836 static long __tun_chr_ioctl(struct file *file, unsigned int cmd,
1837 			    unsigned long arg, int ifreq_len)
1838 {
1839 	struct tun_file *tfile = file->private_data;
1840 	struct tun_struct *tun;
1841 	void __user* argp = (void __user*)arg;
1842 	struct ifreq ifr;
1843 	kuid_t owner;
1844 	kgid_t group;
1845 	int sndbuf;
1846 	int vnet_hdr_sz;
1847 	unsigned int ifindex;
1848 	int le;
1849 	int ret;
1850 
1851 	if (cmd == TUNSETIFF || cmd == TUNSETQUEUE || _IOC_TYPE(cmd) == 0x89) {
1852 		if (copy_from_user(&ifr, argp, ifreq_len))
1853 			return -EFAULT;
1854 	} else {
1855 		memset(&ifr, 0, sizeof(ifr));
1856 	}
1857 	if (cmd == TUNGETFEATURES) {
1858 		/* Currently this just means: "what IFF flags are valid?".
1859 		 * This is needed because we never checked for invalid flags on
1860 		 * TUNSETIFF.
1861 		 */
1862 		return put_user(IFF_TUN | IFF_TAP | TUN_FEATURES,
1863 				(unsigned int __user*)argp);
1864 	} else if (cmd == TUNSETQUEUE)
1865 		return tun_set_queue(file, &ifr);
1866 
1867 	ret = 0;
1868 	rtnl_lock();
1869 
1870 	tun = __tun_get(tfile);
1871 	if (cmd == TUNSETIFF && !tun) {
1872 		ifr.ifr_name[IFNAMSIZ-1] = '\0';
1873 
1874 		ret = tun_set_iff(tfile->net, file, &ifr);
1875 
1876 		if (ret)
1877 			goto unlock;
1878 
1879 		if (copy_to_user(argp, &ifr, ifreq_len))
1880 			ret = -EFAULT;
1881 		goto unlock;
1882 	}
1883 	if (cmd == TUNSETIFINDEX) {
1884 		ret = -EPERM;
1885 		if (tun)
1886 			goto unlock;
1887 
1888 		ret = -EFAULT;
1889 		if (copy_from_user(&ifindex, argp, sizeof(ifindex)))
1890 			goto unlock;
1891 
1892 		ret = 0;
1893 		tfile->ifindex = ifindex;
1894 		goto unlock;
1895 	}
1896 
1897 	ret = -EBADFD;
1898 	if (!tun)
1899 		goto unlock;
1900 
1901 	tun_debug(KERN_INFO, tun, "tun_chr_ioctl cmd %u\n", cmd);
1902 
1903 	ret = 0;
1904 	switch (cmd) {
1905 	case TUNGETIFF:
1906 		tun_get_iff(current->nsproxy->net_ns, tun, &ifr);
1907 
1908 		if (tfile->detached)
1909 			ifr.ifr_flags |= IFF_DETACH_QUEUE;
1910 		if (!tfile->socket.sk->sk_filter)
1911 			ifr.ifr_flags |= IFF_NOFILTER;
1912 
1913 		if (copy_to_user(argp, &ifr, ifreq_len))
1914 			ret = -EFAULT;
1915 		break;
1916 
1917 	case TUNSETNOCSUM:
1918 		/* Disable/Enable checksum */
1919 
1920 		/* [unimplemented] */
1921 		tun_debug(KERN_INFO, tun, "ignored: set checksum %s\n",
1922 			  arg ? "disabled" : "enabled");
1923 		break;
1924 
1925 	case TUNSETPERSIST:
1926 		/* Disable/Enable persist mode. Keep an extra reference to the
1927 		 * module to prevent the module being unprobed.
1928 		 */
1929 		if (arg && !(tun->flags & IFF_PERSIST)) {
1930 			tun->flags |= IFF_PERSIST;
1931 			__module_get(THIS_MODULE);
1932 		}
1933 		if (!arg && (tun->flags & IFF_PERSIST)) {
1934 			tun->flags &= ~IFF_PERSIST;
1935 			module_put(THIS_MODULE);
1936 		}
1937 
1938 		tun_debug(KERN_INFO, tun, "persist %s\n",
1939 			  arg ? "enabled" : "disabled");
1940 		break;
1941 
1942 	case TUNSETOWNER:
1943 		/* Set owner of the device */
1944 		owner = make_kuid(current_user_ns(), arg);
1945 		if (!uid_valid(owner)) {
1946 			ret = -EINVAL;
1947 			break;
1948 		}
1949 		tun->owner = owner;
1950 		tun_debug(KERN_INFO, tun, "owner set to %u\n",
1951 			  from_kuid(&init_user_ns, tun->owner));
1952 		break;
1953 
1954 	case TUNSETGROUP:
1955 		/* Set group of the device */
1956 		group = make_kgid(current_user_ns(), arg);
1957 		if (!gid_valid(group)) {
1958 			ret = -EINVAL;
1959 			break;
1960 		}
1961 		tun->group = group;
1962 		tun_debug(KERN_INFO, tun, "group set to %u\n",
1963 			  from_kgid(&init_user_ns, tun->group));
1964 		break;
1965 
1966 	case TUNSETLINK:
1967 		/* Only allow setting the type when the interface is down */
1968 		if (tun->dev->flags & IFF_UP) {
1969 			tun_debug(KERN_INFO, tun,
1970 				  "Linktype set failed because interface is up\n");
1971 			ret = -EBUSY;
1972 		} else {
1973 			tun->dev->type = (int) arg;
1974 			tun_debug(KERN_INFO, tun, "linktype set to %d\n",
1975 				  tun->dev->type);
1976 			ret = 0;
1977 		}
1978 		break;
1979 
1980 #ifdef TUN_DEBUG
1981 	case TUNSETDEBUG:
1982 		tun->debug = arg;
1983 		break;
1984 #endif
1985 	case TUNSETOFFLOAD:
1986 		ret = set_offload(tun, arg);
1987 		break;
1988 
1989 	case TUNSETTXFILTER:
1990 		/* Can be set only for TAPs */
1991 		ret = -EINVAL;
1992 		if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
1993 			break;
1994 		ret = update_filter(&tun->txflt, (void __user *)arg);
1995 		break;
1996 
1997 	case SIOCGIFHWADDR:
1998 		/* Get hw address */
1999 		memcpy(ifr.ifr_hwaddr.sa_data, tun->dev->dev_addr, ETH_ALEN);
2000 		ifr.ifr_hwaddr.sa_family = tun->dev->type;
2001 		if (copy_to_user(argp, &ifr, ifreq_len))
2002 			ret = -EFAULT;
2003 		break;
2004 
2005 	case SIOCSIFHWADDR:
2006 		/* Set hw address */
2007 		tun_debug(KERN_DEBUG, tun, "set hw address: %pM\n",
2008 			  ifr.ifr_hwaddr.sa_data);
2009 
2010 		ret = dev_set_mac_address(tun->dev, &ifr.ifr_hwaddr);
2011 		break;
2012 
2013 	case TUNGETSNDBUF:
2014 		sndbuf = tfile->socket.sk->sk_sndbuf;
2015 		if (copy_to_user(argp, &sndbuf, sizeof(sndbuf)))
2016 			ret = -EFAULT;
2017 		break;
2018 
2019 	case TUNSETSNDBUF:
2020 		if (copy_from_user(&sndbuf, argp, sizeof(sndbuf))) {
2021 			ret = -EFAULT;
2022 			break;
2023 		}
2024 
2025 		tun->sndbuf = sndbuf;
2026 		tun_set_sndbuf(tun);
2027 		break;
2028 
2029 	case TUNGETVNETHDRSZ:
2030 		vnet_hdr_sz = tun->vnet_hdr_sz;
2031 		if (copy_to_user(argp, &vnet_hdr_sz, sizeof(vnet_hdr_sz)))
2032 			ret = -EFAULT;
2033 		break;
2034 
2035 	case TUNSETVNETHDRSZ:
2036 		if (copy_from_user(&vnet_hdr_sz, argp, sizeof(vnet_hdr_sz))) {
2037 			ret = -EFAULT;
2038 			break;
2039 		}
2040 		if (vnet_hdr_sz < (int)sizeof(struct virtio_net_hdr)) {
2041 			ret = -EINVAL;
2042 			break;
2043 		}
2044 
2045 		tun->vnet_hdr_sz = vnet_hdr_sz;
2046 		break;
2047 
2048 	case TUNGETVNETLE:
2049 		le = !!(tun->flags & TUN_VNET_LE);
2050 		if (put_user(le, (int __user *)argp))
2051 			ret = -EFAULT;
2052 		break;
2053 
2054 	case TUNSETVNETLE:
2055 		if (get_user(le, (int __user *)argp)) {
2056 			ret = -EFAULT;
2057 			break;
2058 		}
2059 		if (le)
2060 			tun->flags |= TUN_VNET_LE;
2061 		else
2062 			tun->flags &= ~TUN_VNET_LE;
2063 		break;
2064 
2065 	case TUNATTACHFILTER:
2066 		/* Can be set only for TAPs */
2067 		ret = -EINVAL;
2068 		if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
2069 			break;
2070 		ret = -EFAULT;
2071 		if (copy_from_user(&tun->fprog, argp, sizeof(tun->fprog)))
2072 			break;
2073 
2074 		ret = tun_attach_filter(tun);
2075 		break;
2076 
2077 	case TUNDETACHFILTER:
2078 		/* Can be set only for TAPs */
2079 		ret = -EINVAL;
2080 		if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
2081 			break;
2082 		ret = 0;
2083 		tun_detach_filter(tun, tun->numqueues);
2084 		break;
2085 
2086 	case TUNGETFILTER:
2087 		ret = -EINVAL;
2088 		if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
2089 			break;
2090 		ret = -EFAULT;
2091 		if (copy_to_user(argp, &tun->fprog, sizeof(tun->fprog)))
2092 			break;
2093 		ret = 0;
2094 		break;
2095 
2096 	default:
2097 		ret = -EINVAL;
2098 		break;
2099 	}
2100 
2101 unlock:
2102 	rtnl_unlock();
2103 	if (tun)
2104 		tun_put(tun);
2105 	return ret;
2106 }
2107 
2108 static long tun_chr_ioctl(struct file *file,
2109 			  unsigned int cmd, unsigned long arg)
2110 {
2111 	return __tun_chr_ioctl(file, cmd, arg, sizeof (struct ifreq));
2112 }
2113 
2114 #ifdef CONFIG_COMPAT
2115 static long tun_chr_compat_ioctl(struct file *file,
2116 			 unsigned int cmd, unsigned long arg)
2117 {
2118 	switch (cmd) {
2119 	case TUNSETIFF:
2120 	case TUNGETIFF:
2121 	case TUNSETTXFILTER:
2122 	case TUNGETSNDBUF:
2123 	case TUNSETSNDBUF:
2124 	case SIOCGIFHWADDR:
2125 	case SIOCSIFHWADDR:
2126 		arg = (unsigned long)compat_ptr(arg);
2127 		break;
2128 	default:
2129 		arg = (compat_ulong_t)arg;
2130 		break;
2131 	}
2132 
2133 	/*
2134 	 * compat_ifreq is shorter than ifreq, so we must not access beyond
2135 	 * the end of that structure. All fields that are used in this
2136 	 * driver are compatible though, we don't need to convert the
2137 	 * contents.
2138 	 */
2139 	return __tun_chr_ioctl(file, cmd, arg, sizeof(struct compat_ifreq));
2140 }
2141 #endif /* CONFIG_COMPAT */
2142 
2143 static int tun_chr_fasync(int fd, struct file *file, int on)
2144 {
2145 	struct tun_file *tfile = file->private_data;
2146 	int ret;
2147 
2148 	if ((ret = fasync_helper(fd, file, on, &tfile->fasync)) < 0)
2149 		goto out;
2150 
2151 	if (on) {
2152 		__f_setown(file, task_pid(current), PIDTYPE_PID, 0);
2153 		tfile->flags |= TUN_FASYNC;
2154 	} else
2155 		tfile->flags &= ~TUN_FASYNC;
2156 	ret = 0;
2157 out:
2158 	return ret;
2159 }
2160 
2161 static int tun_chr_open(struct inode *inode, struct file * file)
2162 {
2163 	struct tun_file *tfile;
2164 
2165 	DBG1(KERN_INFO, "tunX: tun_chr_open\n");
2166 
2167 	tfile = (struct tun_file *)sk_alloc(&init_net, AF_UNSPEC, GFP_KERNEL,
2168 					    &tun_proto);
2169 	if (!tfile)
2170 		return -ENOMEM;
2171 	RCU_INIT_POINTER(tfile->tun, NULL);
2172 	tfile->net = get_net(current->nsproxy->net_ns);
2173 	tfile->flags = 0;
2174 	tfile->ifindex = 0;
2175 
2176 	init_waitqueue_head(&tfile->wq.wait);
2177 	RCU_INIT_POINTER(tfile->socket.wq, &tfile->wq);
2178 
2179 	tfile->socket.file = file;
2180 	tfile->socket.ops = &tun_socket_ops;
2181 
2182 	sock_init_data(&tfile->socket, &tfile->sk);
2183 	sk_change_net(&tfile->sk, tfile->net);
2184 
2185 	tfile->sk.sk_write_space = tun_sock_write_space;
2186 	tfile->sk.sk_sndbuf = INT_MAX;
2187 
2188 	file->private_data = tfile;
2189 	set_bit(SOCK_EXTERNALLY_ALLOCATED, &tfile->socket.flags);
2190 	INIT_LIST_HEAD(&tfile->next);
2191 
2192 	sock_set_flag(&tfile->sk, SOCK_ZEROCOPY);
2193 
2194 	return 0;
2195 }
2196 
2197 static int tun_chr_close(struct inode *inode, struct file *file)
2198 {
2199 	struct tun_file *tfile = file->private_data;
2200 	struct net *net = tfile->net;
2201 
2202 	tun_detach(tfile, true);
2203 	put_net(net);
2204 
2205 	return 0;
2206 }
2207 
2208 #ifdef CONFIG_PROC_FS
2209 static void tun_chr_show_fdinfo(struct seq_file *m, struct file *f)
2210 {
2211 	struct tun_struct *tun;
2212 	struct ifreq ifr;
2213 
2214 	memset(&ifr, 0, sizeof(ifr));
2215 
2216 	rtnl_lock();
2217 	tun = tun_get(f);
2218 	if (tun)
2219 		tun_get_iff(current->nsproxy->net_ns, tun, &ifr);
2220 	rtnl_unlock();
2221 
2222 	if (tun)
2223 		tun_put(tun);
2224 
2225 	seq_printf(m, "iff:\t%s\n", ifr.ifr_name);
2226 }
2227 #endif
2228 
2229 static const struct file_operations tun_fops = {
2230 	.owner	= THIS_MODULE,
2231 	.llseek = no_llseek,
2232 	.read_iter  = tun_chr_read_iter,
2233 	.write_iter = tun_chr_write_iter,
2234 	.poll	= tun_chr_poll,
2235 	.unlocked_ioctl	= tun_chr_ioctl,
2236 #ifdef CONFIG_COMPAT
2237 	.compat_ioctl = tun_chr_compat_ioctl,
2238 #endif
2239 	.open	= tun_chr_open,
2240 	.release = tun_chr_close,
2241 	.fasync = tun_chr_fasync,
2242 #ifdef CONFIG_PROC_FS
2243 	.show_fdinfo = tun_chr_show_fdinfo,
2244 #endif
2245 };
2246 
2247 static struct miscdevice tun_miscdev = {
2248 	.minor = TUN_MINOR,
2249 	.name = "tun",
2250 	.nodename = "net/tun",
2251 	.fops = &tun_fops,
2252 };
2253 
2254 /* ethtool interface */
2255 
2256 static int tun_get_settings(struct net_device *dev, struct ethtool_cmd *cmd)
2257 {
2258 	cmd->supported		= 0;
2259 	cmd->advertising	= 0;
2260 	ethtool_cmd_speed_set(cmd, SPEED_10);
2261 	cmd->duplex		= DUPLEX_FULL;
2262 	cmd->port		= PORT_TP;
2263 	cmd->phy_address	= 0;
2264 	cmd->transceiver	= XCVR_INTERNAL;
2265 	cmd->autoneg		= AUTONEG_DISABLE;
2266 	cmd->maxtxpkt		= 0;
2267 	cmd->maxrxpkt		= 0;
2268 	return 0;
2269 }
2270 
2271 static void tun_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
2272 {
2273 	struct tun_struct *tun = netdev_priv(dev);
2274 
2275 	strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
2276 	strlcpy(info->version, DRV_VERSION, sizeof(info->version));
2277 
2278 	switch (tun->flags & TUN_TYPE_MASK) {
2279 	case IFF_TUN:
2280 		strlcpy(info->bus_info, "tun", sizeof(info->bus_info));
2281 		break;
2282 	case IFF_TAP:
2283 		strlcpy(info->bus_info, "tap", sizeof(info->bus_info));
2284 		break;
2285 	}
2286 }
2287 
2288 static u32 tun_get_msglevel(struct net_device *dev)
2289 {
2290 #ifdef TUN_DEBUG
2291 	struct tun_struct *tun = netdev_priv(dev);
2292 	return tun->debug;
2293 #else
2294 	return -EOPNOTSUPP;
2295 #endif
2296 }
2297 
2298 static void tun_set_msglevel(struct net_device *dev, u32 value)
2299 {
2300 #ifdef TUN_DEBUG
2301 	struct tun_struct *tun = netdev_priv(dev);
2302 	tun->debug = value;
2303 #endif
2304 }
2305 
2306 static const struct ethtool_ops tun_ethtool_ops = {
2307 	.get_settings	= tun_get_settings,
2308 	.get_drvinfo	= tun_get_drvinfo,
2309 	.get_msglevel	= tun_get_msglevel,
2310 	.set_msglevel	= tun_set_msglevel,
2311 	.get_link	= ethtool_op_get_link,
2312 	.get_ts_info	= ethtool_op_get_ts_info,
2313 };
2314 
2315 
2316 static int __init tun_init(void)
2317 {
2318 	int ret = 0;
2319 
2320 	pr_info("%s, %s\n", DRV_DESCRIPTION, DRV_VERSION);
2321 	pr_info("%s\n", DRV_COPYRIGHT);
2322 
2323 	ret = rtnl_link_register(&tun_link_ops);
2324 	if (ret) {
2325 		pr_err("Can't register link_ops\n");
2326 		goto err_linkops;
2327 	}
2328 
2329 	ret = misc_register(&tun_miscdev);
2330 	if (ret) {
2331 		pr_err("Can't register misc device %d\n", TUN_MINOR);
2332 		goto err_misc;
2333 	}
2334 	return  0;
2335 err_misc:
2336 	rtnl_link_unregister(&tun_link_ops);
2337 err_linkops:
2338 	return ret;
2339 }
2340 
2341 static void tun_cleanup(void)
2342 {
2343 	misc_deregister(&tun_miscdev);
2344 	rtnl_link_unregister(&tun_link_ops);
2345 }
2346 
2347 /* Get an underlying socket object from tun file.  Returns error unless file is
2348  * attached to a device.  The returned object works like a packet socket, it
2349  * can be used for sock_sendmsg/sock_recvmsg.  The caller is responsible for
2350  * holding a reference to the file for as long as the socket is in use. */
2351 struct socket *tun_get_socket(struct file *file)
2352 {
2353 	struct tun_file *tfile;
2354 	if (file->f_op != &tun_fops)
2355 		return ERR_PTR(-EINVAL);
2356 	tfile = file->private_data;
2357 	if (!tfile)
2358 		return ERR_PTR(-EBADFD);
2359 	return &tfile->socket;
2360 }
2361 EXPORT_SYMBOL_GPL(tun_get_socket);
2362 
2363 module_init(tun_init);
2364 module_exit(tun_cleanup);
2365 MODULE_DESCRIPTION(DRV_DESCRIPTION);
2366 MODULE_AUTHOR(DRV_COPYRIGHT);
2367 MODULE_LICENSE("GPL");
2368 MODULE_ALIAS_MISCDEV(TUN_MINOR);
2369 MODULE_ALIAS("devname:net/tun");
2370