xref: /openbmc/linux/drivers/net/tun.c (revision 25bd55bbabd8957693c1ffc2fe1bf16cb4fdabd4)
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 }
213 
214 static inline u16 tun16_to_cpu(struct tun_struct *tun, __virtio16 val)
215 {
216 	return __virtio16_to_cpu(tun_is_little_endian(tun), val);
217 }
218 
219 static inline __virtio16 cpu_to_tun16(struct tun_struct *tun, u16 val)
220 {
221 	return __cpu_to_virtio16(tun_is_little_endian(tun), val);
222 }
223 
224 static inline u32 tun_hashfn(u32 rxhash)
225 {
226 	return rxhash & 0x3ff;
227 }
228 
229 static struct tun_flow_entry *tun_flow_find(struct hlist_head *head, u32 rxhash)
230 {
231 	struct tun_flow_entry *e;
232 
233 	hlist_for_each_entry_rcu(e, head, hash_link) {
234 		if (e->rxhash == rxhash)
235 			return e;
236 	}
237 	return NULL;
238 }
239 
240 static struct tun_flow_entry *tun_flow_create(struct tun_struct *tun,
241 					      struct hlist_head *head,
242 					      u32 rxhash, u16 queue_index)
243 {
244 	struct tun_flow_entry *e = kmalloc(sizeof(*e), GFP_ATOMIC);
245 
246 	if (e) {
247 		tun_debug(KERN_INFO, tun, "create flow: hash %u index %u\n",
248 			  rxhash, queue_index);
249 		e->updated = jiffies;
250 		e->rxhash = rxhash;
251 		e->rps_rxhash = 0;
252 		e->queue_index = queue_index;
253 		e->tun = tun;
254 		hlist_add_head_rcu(&e->hash_link, head);
255 		++tun->flow_count;
256 	}
257 	return e;
258 }
259 
260 static void tun_flow_delete(struct tun_struct *tun, struct tun_flow_entry *e)
261 {
262 	tun_debug(KERN_INFO, tun, "delete flow: hash %u index %u\n",
263 		  e->rxhash, e->queue_index);
264 	hlist_del_rcu(&e->hash_link);
265 	kfree_rcu(e, rcu);
266 	--tun->flow_count;
267 }
268 
269 static void tun_flow_flush(struct tun_struct *tun)
270 {
271 	int i;
272 
273 	spin_lock_bh(&tun->lock);
274 	for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
275 		struct tun_flow_entry *e;
276 		struct hlist_node *n;
277 
278 		hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link)
279 			tun_flow_delete(tun, e);
280 	}
281 	spin_unlock_bh(&tun->lock);
282 }
283 
284 static void tun_flow_delete_by_queue(struct tun_struct *tun, u16 queue_index)
285 {
286 	int i;
287 
288 	spin_lock_bh(&tun->lock);
289 	for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
290 		struct tun_flow_entry *e;
291 		struct hlist_node *n;
292 
293 		hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
294 			if (e->queue_index == queue_index)
295 				tun_flow_delete(tun, e);
296 		}
297 	}
298 	spin_unlock_bh(&tun->lock);
299 }
300 
301 static void tun_flow_cleanup(unsigned long data)
302 {
303 	struct tun_struct *tun = (struct tun_struct *)data;
304 	unsigned long delay = tun->ageing_time;
305 	unsigned long next_timer = jiffies + delay;
306 	unsigned long count = 0;
307 	int i;
308 
309 	tun_debug(KERN_INFO, tun, "tun_flow_cleanup\n");
310 
311 	spin_lock_bh(&tun->lock);
312 	for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
313 		struct tun_flow_entry *e;
314 		struct hlist_node *n;
315 
316 		hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
317 			unsigned long this_timer;
318 			count++;
319 			this_timer = e->updated + delay;
320 			if (time_before_eq(this_timer, jiffies))
321 				tun_flow_delete(tun, e);
322 			else if (time_before(this_timer, next_timer))
323 				next_timer = this_timer;
324 		}
325 	}
326 
327 	if (count)
328 		mod_timer(&tun->flow_gc_timer, round_jiffies_up(next_timer));
329 	spin_unlock_bh(&tun->lock);
330 }
331 
332 static void tun_flow_update(struct tun_struct *tun, u32 rxhash,
333 			    struct tun_file *tfile)
334 {
335 	struct hlist_head *head;
336 	struct tun_flow_entry *e;
337 	unsigned long delay = tun->ageing_time;
338 	u16 queue_index = tfile->queue_index;
339 
340 	if (!rxhash)
341 		return;
342 	else
343 		head = &tun->flows[tun_hashfn(rxhash)];
344 
345 	rcu_read_lock();
346 
347 	/* We may get a very small possibility of OOO during switching, not
348 	 * worth to optimize.*/
349 	if (tun->numqueues == 1 || tfile->detached)
350 		goto unlock;
351 
352 	e = tun_flow_find(head, rxhash);
353 	if (likely(e)) {
354 		/* TODO: keep queueing to old queue until it's empty? */
355 		e->queue_index = queue_index;
356 		e->updated = jiffies;
357 		sock_rps_record_flow_hash(e->rps_rxhash);
358 	} else {
359 		spin_lock_bh(&tun->lock);
360 		if (!tun_flow_find(head, rxhash) &&
361 		    tun->flow_count < MAX_TAP_FLOWS)
362 			tun_flow_create(tun, head, rxhash, queue_index);
363 
364 		if (!timer_pending(&tun->flow_gc_timer))
365 			mod_timer(&tun->flow_gc_timer,
366 				  round_jiffies_up(jiffies + delay));
367 		spin_unlock_bh(&tun->lock);
368 	}
369 
370 unlock:
371 	rcu_read_unlock();
372 }
373 
374 /**
375  * Save the hash received in the stack receive path and update the
376  * flow_hash table accordingly.
377  */
378 static inline void tun_flow_save_rps_rxhash(struct tun_flow_entry *e, u32 hash)
379 {
380 	if (unlikely(e->rps_rxhash != hash))
381 		e->rps_rxhash = hash;
382 }
383 
384 /* We try to identify a flow through its rxhash first. The reason that
385  * we do not check rxq no. is because some cards(e.g 82599), chooses
386  * the rxq based on the txq where the last packet of the flow comes. As
387  * the userspace application move between processors, we may get a
388  * different rxq no. here. If we could not get rxhash, then we would
389  * hope the rxq no. may help here.
390  */
391 static u16 tun_select_queue(struct net_device *dev, struct sk_buff *skb,
392 			    void *accel_priv, select_queue_fallback_t fallback)
393 {
394 	struct tun_struct *tun = netdev_priv(dev);
395 	struct tun_flow_entry *e;
396 	u32 txq = 0;
397 	u32 numqueues = 0;
398 
399 	rcu_read_lock();
400 	numqueues = ACCESS_ONCE(tun->numqueues);
401 
402 	txq = skb_get_hash(skb);
403 	if (txq) {
404 		e = tun_flow_find(&tun->flows[tun_hashfn(txq)], txq);
405 		if (e) {
406 			tun_flow_save_rps_rxhash(e, txq);
407 			txq = e->queue_index;
408 		} else
409 			/* use multiply and shift instead of expensive divide */
410 			txq = ((u64)txq * numqueues) >> 32;
411 	} else if (likely(skb_rx_queue_recorded(skb))) {
412 		txq = skb_get_rx_queue(skb);
413 		while (unlikely(txq >= numqueues))
414 			txq -= numqueues;
415 	}
416 
417 	rcu_read_unlock();
418 	return txq;
419 }
420 
421 static inline bool tun_not_capable(struct tun_struct *tun)
422 {
423 	const struct cred *cred = current_cred();
424 	struct net *net = dev_net(tun->dev);
425 
426 	return ((uid_valid(tun->owner) && !uid_eq(cred->euid, tun->owner)) ||
427 		  (gid_valid(tun->group) && !in_egroup_p(tun->group))) &&
428 		!ns_capable(net->user_ns, CAP_NET_ADMIN);
429 }
430 
431 static void tun_set_real_num_queues(struct tun_struct *tun)
432 {
433 	netif_set_real_num_tx_queues(tun->dev, tun->numqueues);
434 	netif_set_real_num_rx_queues(tun->dev, tun->numqueues);
435 }
436 
437 static void tun_disable_queue(struct tun_struct *tun, struct tun_file *tfile)
438 {
439 	tfile->detached = tun;
440 	list_add_tail(&tfile->next, &tun->disabled);
441 	++tun->numdisabled;
442 }
443 
444 static struct tun_struct *tun_enable_queue(struct tun_file *tfile)
445 {
446 	struct tun_struct *tun = tfile->detached;
447 
448 	tfile->detached = NULL;
449 	list_del_init(&tfile->next);
450 	--tun->numdisabled;
451 	return tun;
452 }
453 
454 static void tun_queue_purge(struct tun_file *tfile)
455 {
456 	skb_queue_purge(&tfile->sk.sk_receive_queue);
457 	skb_queue_purge(&tfile->sk.sk_error_queue);
458 }
459 
460 static void __tun_detach(struct tun_file *tfile, bool clean)
461 {
462 	struct tun_file *ntfile;
463 	struct tun_struct *tun;
464 
465 	tun = rtnl_dereference(tfile->tun);
466 
467 	if (tun && !tfile->detached) {
468 		u16 index = tfile->queue_index;
469 		BUG_ON(index >= tun->numqueues);
470 
471 		rcu_assign_pointer(tun->tfiles[index],
472 				   tun->tfiles[tun->numqueues - 1]);
473 		ntfile = rtnl_dereference(tun->tfiles[index]);
474 		ntfile->queue_index = index;
475 
476 		--tun->numqueues;
477 		if (clean) {
478 			RCU_INIT_POINTER(tfile->tun, NULL);
479 			sock_put(&tfile->sk);
480 		} else
481 			tun_disable_queue(tun, tfile);
482 
483 		synchronize_net();
484 		tun_flow_delete_by_queue(tun, tun->numqueues + 1);
485 		/* Drop read queue */
486 		tun_queue_purge(tfile);
487 		tun_set_real_num_queues(tun);
488 	} else if (tfile->detached && clean) {
489 		tun = tun_enable_queue(tfile);
490 		sock_put(&tfile->sk);
491 	}
492 
493 	if (clean) {
494 		if (tun && tun->numqueues == 0 && tun->numdisabled == 0) {
495 			netif_carrier_off(tun->dev);
496 
497 			if (!(tun->flags & IFF_PERSIST) &&
498 			    tun->dev->reg_state == NETREG_REGISTERED)
499 				unregister_netdevice(tun->dev);
500 		}
501 
502 		BUG_ON(!test_bit(SOCK_EXTERNALLY_ALLOCATED,
503 				 &tfile->socket.flags));
504 		sk_release_kernel(&tfile->sk);
505 	}
506 }
507 
508 static void tun_detach(struct tun_file *tfile, bool clean)
509 {
510 	rtnl_lock();
511 	__tun_detach(tfile, clean);
512 	rtnl_unlock();
513 }
514 
515 static void tun_detach_all(struct net_device *dev)
516 {
517 	struct tun_struct *tun = netdev_priv(dev);
518 	struct tun_file *tfile, *tmp;
519 	int i, n = tun->numqueues;
520 
521 	for (i = 0; i < n; i++) {
522 		tfile = rtnl_dereference(tun->tfiles[i]);
523 		BUG_ON(!tfile);
524 		tfile->socket.sk->sk_data_ready(tfile->socket.sk);
525 		RCU_INIT_POINTER(tfile->tun, NULL);
526 		--tun->numqueues;
527 	}
528 	list_for_each_entry(tfile, &tun->disabled, next) {
529 		tfile->socket.sk->sk_data_ready(tfile->socket.sk);
530 		RCU_INIT_POINTER(tfile->tun, NULL);
531 	}
532 	BUG_ON(tun->numqueues != 0);
533 
534 	synchronize_net();
535 	for (i = 0; i < n; i++) {
536 		tfile = rtnl_dereference(tun->tfiles[i]);
537 		/* Drop read queue */
538 		tun_queue_purge(tfile);
539 		sock_put(&tfile->sk);
540 	}
541 	list_for_each_entry_safe(tfile, tmp, &tun->disabled, next) {
542 		tun_enable_queue(tfile);
543 		tun_queue_purge(tfile);
544 		sock_put(&tfile->sk);
545 	}
546 	BUG_ON(tun->numdisabled != 0);
547 
548 	if (tun->flags & IFF_PERSIST)
549 		module_put(THIS_MODULE);
550 }
551 
552 static int tun_attach(struct tun_struct *tun, struct file *file, bool skip_filter)
553 {
554 	struct tun_file *tfile = file->private_data;
555 	int err;
556 
557 	err = security_tun_dev_attach(tfile->socket.sk, tun->security);
558 	if (err < 0)
559 		goto out;
560 
561 	err = -EINVAL;
562 	if (rtnl_dereference(tfile->tun) && !tfile->detached)
563 		goto out;
564 
565 	err = -EBUSY;
566 	if (!(tun->flags & IFF_MULTI_QUEUE) && tun->numqueues == 1)
567 		goto out;
568 
569 	err = -E2BIG;
570 	if (!tfile->detached &&
571 	    tun->numqueues + tun->numdisabled == MAX_TAP_QUEUES)
572 		goto out;
573 
574 	err = 0;
575 
576 	/* Re-attach the filter to persist device */
577 	if (!skip_filter && (tun->filter_attached == true)) {
578 		err = sk_attach_filter(&tun->fprog, tfile->socket.sk);
579 		if (!err)
580 			goto out;
581 	}
582 	tfile->queue_index = tun->numqueues;
583 	rcu_assign_pointer(tfile->tun, tun);
584 	rcu_assign_pointer(tun->tfiles[tun->numqueues], tfile);
585 	tun->numqueues++;
586 
587 	if (tfile->detached)
588 		tun_enable_queue(tfile);
589 	else
590 		sock_hold(&tfile->sk);
591 
592 	tun_set_real_num_queues(tun);
593 
594 	/* device is allowed to go away first, so no need to hold extra
595 	 * refcnt.
596 	 */
597 
598 out:
599 	return err;
600 }
601 
602 static struct tun_struct *__tun_get(struct tun_file *tfile)
603 {
604 	struct tun_struct *tun;
605 
606 	rcu_read_lock();
607 	tun = rcu_dereference(tfile->tun);
608 	if (tun)
609 		dev_hold(tun->dev);
610 	rcu_read_unlock();
611 
612 	return tun;
613 }
614 
615 static struct tun_struct *tun_get(struct file *file)
616 {
617 	return __tun_get(file->private_data);
618 }
619 
620 static void tun_put(struct tun_struct *tun)
621 {
622 	dev_put(tun->dev);
623 }
624 
625 /* TAP filtering */
626 static void addr_hash_set(u32 *mask, const u8 *addr)
627 {
628 	int n = ether_crc(ETH_ALEN, addr) >> 26;
629 	mask[n >> 5] |= (1 << (n & 31));
630 }
631 
632 static unsigned int addr_hash_test(const u32 *mask, const u8 *addr)
633 {
634 	int n = ether_crc(ETH_ALEN, addr) >> 26;
635 	return mask[n >> 5] & (1 << (n & 31));
636 }
637 
638 static int update_filter(struct tap_filter *filter, void __user *arg)
639 {
640 	struct { u8 u[ETH_ALEN]; } *addr;
641 	struct tun_filter uf;
642 	int err, alen, n, nexact;
643 
644 	if (copy_from_user(&uf, arg, sizeof(uf)))
645 		return -EFAULT;
646 
647 	if (!uf.count) {
648 		/* Disabled */
649 		filter->count = 0;
650 		return 0;
651 	}
652 
653 	alen = ETH_ALEN * uf.count;
654 	addr = kmalloc(alen, GFP_KERNEL);
655 	if (!addr)
656 		return -ENOMEM;
657 
658 	if (copy_from_user(addr, arg + sizeof(uf), alen)) {
659 		err = -EFAULT;
660 		goto done;
661 	}
662 
663 	/* The filter is updated without holding any locks. Which is
664 	 * perfectly safe. We disable it first and in the worst
665 	 * case we'll accept a few undesired packets. */
666 	filter->count = 0;
667 	wmb();
668 
669 	/* Use first set of addresses as an exact filter */
670 	for (n = 0; n < uf.count && n < FLT_EXACT_COUNT; n++)
671 		memcpy(filter->addr[n], addr[n].u, ETH_ALEN);
672 
673 	nexact = n;
674 
675 	/* Remaining multicast addresses are hashed,
676 	 * unicast will leave the filter disabled. */
677 	memset(filter->mask, 0, sizeof(filter->mask));
678 	for (; n < uf.count; n++) {
679 		if (!is_multicast_ether_addr(addr[n].u)) {
680 			err = 0; /* no filter */
681 			goto done;
682 		}
683 		addr_hash_set(filter->mask, addr[n].u);
684 	}
685 
686 	/* For ALLMULTI just set the mask to all ones.
687 	 * This overrides the mask populated above. */
688 	if ((uf.flags & TUN_FLT_ALLMULTI))
689 		memset(filter->mask, ~0, sizeof(filter->mask));
690 
691 	/* Now enable the filter */
692 	wmb();
693 	filter->count = nexact;
694 
695 	/* Return the number of exact filters */
696 	err = nexact;
697 
698 done:
699 	kfree(addr);
700 	return err;
701 }
702 
703 /* Returns: 0 - drop, !=0 - accept */
704 static int run_filter(struct tap_filter *filter, const struct sk_buff *skb)
705 {
706 	/* Cannot use eth_hdr(skb) here because skb_mac_hdr() is incorrect
707 	 * at this point. */
708 	struct ethhdr *eh = (struct ethhdr *) skb->data;
709 	int i;
710 
711 	/* Exact match */
712 	for (i = 0; i < filter->count; i++)
713 		if (ether_addr_equal(eh->h_dest, filter->addr[i]))
714 			return 1;
715 
716 	/* Inexact match (multicast only) */
717 	if (is_multicast_ether_addr(eh->h_dest))
718 		return addr_hash_test(filter->mask, eh->h_dest);
719 
720 	return 0;
721 }
722 
723 /*
724  * Checks whether the packet is accepted or not.
725  * Returns: 0 - drop, !=0 - accept
726  */
727 static int check_filter(struct tap_filter *filter, const struct sk_buff *skb)
728 {
729 	if (!filter->count)
730 		return 1;
731 
732 	return run_filter(filter, skb);
733 }
734 
735 /* Network device part of the driver */
736 
737 static const struct ethtool_ops tun_ethtool_ops;
738 
739 /* Net device detach from fd. */
740 static void tun_net_uninit(struct net_device *dev)
741 {
742 	tun_detach_all(dev);
743 }
744 
745 /* Net device open. */
746 static int tun_net_open(struct net_device *dev)
747 {
748 	netif_tx_start_all_queues(dev);
749 	return 0;
750 }
751 
752 /* Net device close. */
753 static int tun_net_close(struct net_device *dev)
754 {
755 	netif_tx_stop_all_queues(dev);
756 	return 0;
757 }
758 
759 /* Net device start xmit */
760 static netdev_tx_t tun_net_xmit(struct sk_buff *skb, struct net_device *dev)
761 {
762 	struct tun_struct *tun = netdev_priv(dev);
763 	int txq = skb->queue_mapping;
764 	struct tun_file *tfile;
765 	u32 numqueues = 0;
766 
767 	rcu_read_lock();
768 	tfile = rcu_dereference(tun->tfiles[txq]);
769 	numqueues = ACCESS_ONCE(tun->numqueues);
770 
771 	/* Drop packet if interface is not attached */
772 	if (txq >= numqueues)
773 		goto drop;
774 
775 	if (numqueues == 1) {
776 		/* Select queue was not called for the skbuff, so we extract the
777 		 * RPS hash and save it into the flow_table here.
778 		 */
779 		__u32 rxhash;
780 
781 		rxhash = skb_get_hash(skb);
782 		if (rxhash) {
783 			struct tun_flow_entry *e;
784 			e = tun_flow_find(&tun->flows[tun_hashfn(rxhash)],
785 					rxhash);
786 			if (e)
787 				tun_flow_save_rps_rxhash(e, rxhash);
788 		}
789 	}
790 
791 	tun_debug(KERN_INFO, tun, "tun_net_xmit %d\n", skb->len);
792 
793 	BUG_ON(!tfile);
794 
795 	/* Drop if the filter does not like it.
796 	 * This is a noop if the filter is disabled.
797 	 * Filter can be enabled only for the TAP devices. */
798 	if (!check_filter(&tun->txflt, skb))
799 		goto drop;
800 
801 	if (tfile->socket.sk->sk_filter &&
802 	    sk_filter(tfile->socket.sk, skb))
803 		goto drop;
804 
805 	/* Limit the number of packets queued by dividing txq length with the
806 	 * number of queues.
807 	 */
808 	if (skb_queue_len(&tfile->socket.sk->sk_receive_queue) * numqueues
809 			  >= dev->tx_queue_len)
810 		goto drop;
811 
812 	if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC)))
813 		goto drop;
814 
815 	if (skb->sk) {
816 		sock_tx_timestamp(skb->sk, &skb_shinfo(skb)->tx_flags);
817 		sw_tx_timestamp(skb);
818 	}
819 
820 	/* Orphan the skb - required as we might hang on to it
821 	 * for indefinite time.
822 	 */
823 	skb_orphan(skb);
824 
825 	nf_reset(skb);
826 
827 	/* Enqueue packet */
828 	skb_queue_tail(&tfile->socket.sk->sk_receive_queue, skb);
829 
830 	/* Notify and wake up reader process */
831 	if (tfile->flags & TUN_FASYNC)
832 		kill_fasync(&tfile->fasync, SIGIO, POLL_IN);
833 	tfile->socket.sk->sk_data_ready(tfile->socket.sk);
834 
835 	rcu_read_unlock();
836 	return NETDEV_TX_OK;
837 
838 drop:
839 	dev->stats.tx_dropped++;
840 	skb_tx_error(skb);
841 	kfree_skb(skb);
842 	rcu_read_unlock();
843 	return NET_XMIT_DROP;
844 }
845 
846 static void tun_net_mclist(struct net_device *dev)
847 {
848 	/*
849 	 * This callback is supposed to deal with mc filter in
850 	 * _rx_ path and has nothing to do with the _tx_ path.
851 	 * In rx path we always accept everything userspace gives us.
852 	 */
853 }
854 
855 #define MIN_MTU 68
856 #define MAX_MTU 65535
857 
858 static int
859 tun_net_change_mtu(struct net_device *dev, int new_mtu)
860 {
861 	if (new_mtu < MIN_MTU || new_mtu + dev->hard_header_len > MAX_MTU)
862 		return -EINVAL;
863 	dev->mtu = new_mtu;
864 	return 0;
865 }
866 
867 static netdev_features_t tun_net_fix_features(struct net_device *dev,
868 	netdev_features_t features)
869 {
870 	struct tun_struct *tun = netdev_priv(dev);
871 
872 	return (features & tun->set_features) | (features & ~TUN_USER_FEATURES);
873 }
874 #ifdef CONFIG_NET_POLL_CONTROLLER
875 static void tun_poll_controller(struct net_device *dev)
876 {
877 	/*
878 	 * Tun only receives frames when:
879 	 * 1) the char device endpoint gets data from user space
880 	 * 2) the tun socket gets a sendmsg call from user space
881 	 * Since both of those are synchronous operations, we are guaranteed
882 	 * never to have pending data when we poll for it
883 	 * so there is nothing to do here but return.
884 	 * We need this though so netpoll recognizes us as an interface that
885 	 * supports polling, which enables bridge devices in virt setups to
886 	 * still use netconsole
887 	 */
888 	return;
889 }
890 #endif
891 static const struct net_device_ops tun_netdev_ops = {
892 	.ndo_uninit		= tun_net_uninit,
893 	.ndo_open		= tun_net_open,
894 	.ndo_stop		= tun_net_close,
895 	.ndo_start_xmit		= tun_net_xmit,
896 	.ndo_change_mtu		= tun_net_change_mtu,
897 	.ndo_fix_features	= tun_net_fix_features,
898 	.ndo_select_queue	= tun_select_queue,
899 #ifdef CONFIG_NET_POLL_CONTROLLER
900 	.ndo_poll_controller	= tun_poll_controller,
901 #endif
902 };
903 
904 static const struct net_device_ops tap_netdev_ops = {
905 	.ndo_uninit		= tun_net_uninit,
906 	.ndo_open		= tun_net_open,
907 	.ndo_stop		= tun_net_close,
908 	.ndo_start_xmit		= tun_net_xmit,
909 	.ndo_change_mtu		= tun_net_change_mtu,
910 	.ndo_fix_features	= tun_net_fix_features,
911 	.ndo_set_rx_mode	= tun_net_mclist,
912 	.ndo_set_mac_address	= eth_mac_addr,
913 	.ndo_validate_addr	= eth_validate_addr,
914 	.ndo_select_queue	= tun_select_queue,
915 #ifdef CONFIG_NET_POLL_CONTROLLER
916 	.ndo_poll_controller	= tun_poll_controller,
917 #endif
918 };
919 
920 static void tun_flow_init(struct tun_struct *tun)
921 {
922 	int i;
923 
924 	for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++)
925 		INIT_HLIST_HEAD(&tun->flows[i]);
926 
927 	tun->ageing_time = TUN_FLOW_EXPIRE;
928 	setup_timer(&tun->flow_gc_timer, tun_flow_cleanup, (unsigned long)tun);
929 	mod_timer(&tun->flow_gc_timer,
930 		  round_jiffies_up(jiffies + tun->ageing_time));
931 }
932 
933 static void tun_flow_uninit(struct tun_struct *tun)
934 {
935 	del_timer_sync(&tun->flow_gc_timer);
936 	tun_flow_flush(tun);
937 }
938 
939 /* Initialize net device. */
940 static void tun_net_init(struct net_device *dev)
941 {
942 	struct tun_struct *tun = netdev_priv(dev);
943 
944 	switch (tun->flags & TUN_TYPE_MASK) {
945 	case IFF_TUN:
946 		dev->netdev_ops = &tun_netdev_ops;
947 
948 		/* Point-to-Point TUN Device */
949 		dev->hard_header_len = 0;
950 		dev->addr_len = 0;
951 		dev->mtu = 1500;
952 
953 		/* Zero header length */
954 		dev->type = ARPHRD_NONE;
955 		dev->flags = IFF_POINTOPOINT | IFF_NOARP | IFF_MULTICAST;
956 		dev->tx_queue_len = TUN_READQ_SIZE;  /* We prefer our own queue length */
957 		break;
958 
959 	case IFF_TAP:
960 		dev->netdev_ops = &tap_netdev_ops;
961 		/* Ethernet TAP Device */
962 		ether_setup(dev);
963 		dev->priv_flags &= ~IFF_TX_SKB_SHARING;
964 		dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
965 
966 		eth_hw_addr_random(dev);
967 
968 		dev->tx_queue_len = TUN_READQ_SIZE;  /* We prefer our own queue length */
969 		break;
970 	}
971 }
972 
973 /* Character device part */
974 
975 /* Poll */
976 static unsigned int tun_chr_poll(struct file *file, poll_table *wait)
977 {
978 	struct tun_file *tfile = file->private_data;
979 	struct tun_struct *tun = __tun_get(tfile);
980 	struct sock *sk;
981 	unsigned int mask = 0;
982 
983 	if (!tun)
984 		return POLLERR;
985 
986 	sk = tfile->socket.sk;
987 
988 	tun_debug(KERN_INFO, tun, "tun_chr_poll\n");
989 
990 	poll_wait(file, sk_sleep(sk), wait);
991 
992 	if (!skb_queue_empty(&sk->sk_receive_queue))
993 		mask |= POLLIN | POLLRDNORM;
994 
995 	if (sock_writeable(sk) ||
996 	    (!test_and_set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags) &&
997 	     sock_writeable(sk)))
998 		mask |= POLLOUT | POLLWRNORM;
999 
1000 	if (tun->dev->reg_state != NETREG_REGISTERED)
1001 		mask = POLLERR;
1002 
1003 	tun_put(tun);
1004 	return mask;
1005 }
1006 
1007 /* prepad is the amount to reserve at front.  len is length after that.
1008  * linear is a hint as to how much to copy (usually headers). */
1009 static struct sk_buff *tun_alloc_skb(struct tun_file *tfile,
1010 				     size_t prepad, size_t len,
1011 				     size_t linear, int noblock)
1012 {
1013 	struct sock *sk = tfile->socket.sk;
1014 	struct sk_buff *skb;
1015 	int err;
1016 
1017 	/* Under a page?  Don't bother with paged skb. */
1018 	if (prepad + len < PAGE_SIZE || !linear)
1019 		linear = len;
1020 
1021 	skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
1022 				   &err, 0);
1023 	if (!skb)
1024 		return ERR_PTR(err);
1025 
1026 	skb_reserve(skb, prepad);
1027 	skb_put(skb, linear);
1028 	skb->data_len = len - linear;
1029 	skb->len += len - linear;
1030 
1031 	return skb;
1032 }
1033 
1034 /* Get packet from user space buffer */
1035 static ssize_t tun_get_user(struct tun_struct *tun, struct tun_file *tfile,
1036 			    void *msg_control, struct iov_iter *from,
1037 			    int noblock)
1038 {
1039 	struct tun_pi pi = { 0, cpu_to_be16(ETH_P_IP) };
1040 	struct sk_buff *skb;
1041 	size_t total_len = iov_iter_count(from);
1042 	size_t len = total_len, align = NET_SKB_PAD, linear;
1043 	struct virtio_net_hdr gso = { 0 };
1044 	int good_linear;
1045 	int copylen;
1046 	bool zerocopy = false;
1047 	int err;
1048 	u32 rxhash;
1049 	ssize_t n;
1050 
1051 	if (!(tun->flags & IFF_NO_PI)) {
1052 		if (len < sizeof(pi))
1053 			return -EINVAL;
1054 		len -= sizeof(pi);
1055 
1056 		n = copy_from_iter(&pi, sizeof(pi), from);
1057 		if (n != sizeof(pi))
1058 			return -EFAULT;
1059 	}
1060 
1061 	if (tun->flags & IFF_VNET_HDR) {
1062 		if (len < tun->vnet_hdr_sz)
1063 			return -EINVAL;
1064 		len -= tun->vnet_hdr_sz;
1065 
1066 		n = copy_from_iter(&gso, sizeof(gso), from);
1067 		if (n != sizeof(gso))
1068 			return -EFAULT;
1069 
1070 		if ((gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
1071 		    tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2 > tun16_to_cpu(tun, gso.hdr_len))
1072 			gso.hdr_len = cpu_to_tun16(tun, tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2);
1073 
1074 		if (tun16_to_cpu(tun, gso.hdr_len) > len)
1075 			return -EINVAL;
1076 		iov_iter_advance(from, tun->vnet_hdr_sz - sizeof(gso));
1077 	}
1078 
1079 	if ((tun->flags & TUN_TYPE_MASK) == IFF_TAP) {
1080 		align += NET_IP_ALIGN;
1081 		if (unlikely(len < ETH_HLEN ||
1082 			     (gso.hdr_len && tun16_to_cpu(tun, gso.hdr_len) < ETH_HLEN)))
1083 			return -EINVAL;
1084 	}
1085 
1086 	good_linear = SKB_MAX_HEAD(align);
1087 
1088 	if (msg_control) {
1089 		struct iov_iter i = *from;
1090 
1091 		/* There are 256 bytes to be copied in skb, so there is
1092 		 * enough room for skb expand head in case it is used.
1093 		 * The rest of the buffer is mapped from userspace.
1094 		 */
1095 		copylen = gso.hdr_len ? tun16_to_cpu(tun, gso.hdr_len) : GOODCOPY_LEN;
1096 		if (copylen > good_linear)
1097 			copylen = good_linear;
1098 		linear = copylen;
1099 		iov_iter_advance(&i, copylen);
1100 		if (iov_iter_npages(&i, INT_MAX) <= MAX_SKB_FRAGS)
1101 			zerocopy = true;
1102 	}
1103 
1104 	if (!zerocopy) {
1105 		copylen = len;
1106 		if (tun16_to_cpu(tun, gso.hdr_len) > good_linear)
1107 			linear = good_linear;
1108 		else
1109 			linear = tun16_to_cpu(tun, gso.hdr_len);
1110 	}
1111 
1112 	skb = tun_alloc_skb(tfile, align, copylen, linear, noblock);
1113 	if (IS_ERR(skb)) {
1114 		if (PTR_ERR(skb) != -EAGAIN)
1115 			tun->dev->stats.rx_dropped++;
1116 		return PTR_ERR(skb);
1117 	}
1118 
1119 	if (zerocopy)
1120 		err = zerocopy_sg_from_iter(skb, from);
1121 	else {
1122 		err = skb_copy_datagram_from_iter(skb, 0, from, len);
1123 		if (!err && msg_control) {
1124 			struct ubuf_info *uarg = msg_control;
1125 			uarg->callback(uarg, false);
1126 		}
1127 	}
1128 
1129 	if (err) {
1130 		tun->dev->stats.rx_dropped++;
1131 		kfree_skb(skb);
1132 		return -EFAULT;
1133 	}
1134 
1135 	if (gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
1136 		if (!skb_partial_csum_set(skb, tun16_to_cpu(tun, gso.csum_start),
1137 					  tun16_to_cpu(tun, gso.csum_offset))) {
1138 			tun->dev->stats.rx_frame_errors++;
1139 			kfree_skb(skb);
1140 			return -EINVAL;
1141 		}
1142 	}
1143 
1144 	switch (tun->flags & TUN_TYPE_MASK) {
1145 	case IFF_TUN:
1146 		if (tun->flags & IFF_NO_PI) {
1147 			switch (skb->data[0] & 0xf0) {
1148 			case 0x40:
1149 				pi.proto = htons(ETH_P_IP);
1150 				break;
1151 			case 0x60:
1152 				pi.proto = htons(ETH_P_IPV6);
1153 				break;
1154 			default:
1155 				tun->dev->stats.rx_dropped++;
1156 				kfree_skb(skb);
1157 				return -EINVAL;
1158 			}
1159 		}
1160 
1161 		skb_reset_mac_header(skb);
1162 		skb->protocol = pi.proto;
1163 		skb->dev = tun->dev;
1164 		break;
1165 	case IFF_TAP:
1166 		skb->protocol = eth_type_trans(skb, tun->dev);
1167 		break;
1168 	}
1169 
1170 	if (gso.gso_type != VIRTIO_NET_HDR_GSO_NONE) {
1171 		pr_debug("GSO!\n");
1172 		switch (gso.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
1173 		case VIRTIO_NET_HDR_GSO_TCPV4:
1174 			skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
1175 			break;
1176 		case VIRTIO_NET_HDR_GSO_TCPV6:
1177 			skb_shinfo(skb)->gso_type = SKB_GSO_TCPV6;
1178 			break;
1179 		case VIRTIO_NET_HDR_GSO_UDP:
1180 			skb_shinfo(skb)->gso_type = SKB_GSO_UDP;
1181 			break;
1182 		default:
1183 			tun->dev->stats.rx_frame_errors++;
1184 			kfree_skb(skb);
1185 			return -EINVAL;
1186 		}
1187 
1188 		if (gso.gso_type & VIRTIO_NET_HDR_GSO_ECN)
1189 			skb_shinfo(skb)->gso_type |= SKB_GSO_TCP_ECN;
1190 
1191 		skb_shinfo(skb)->gso_size = tun16_to_cpu(tun, gso.gso_size);
1192 		if (skb_shinfo(skb)->gso_size == 0) {
1193 			tun->dev->stats.rx_frame_errors++;
1194 			kfree_skb(skb);
1195 			return -EINVAL;
1196 		}
1197 
1198 		/* Header must be checked, and gso_segs computed. */
1199 		skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
1200 		skb_shinfo(skb)->gso_segs = 0;
1201 	}
1202 
1203 	/* copy skb_ubuf_info for callback when skb has no error */
1204 	if (zerocopy) {
1205 		skb_shinfo(skb)->destructor_arg = msg_control;
1206 		skb_shinfo(skb)->tx_flags |= SKBTX_DEV_ZEROCOPY;
1207 		skb_shinfo(skb)->tx_flags |= SKBTX_SHARED_FRAG;
1208 	}
1209 
1210 	skb_reset_network_header(skb);
1211 	skb_probe_transport_header(skb, 0);
1212 
1213 	rxhash = skb_get_hash(skb);
1214 	netif_rx_ni(skb);
1215 
1216 	tun->dev->stats.rx_packets++;
1217 	tun->dev->stats.rx_bytes += len;
1218 
1219 	tun_flow_update(tun, rxhash, tfile);
1220 	return total_len;
1221 }
1222 
1223 static ssize_t tun_chr_write_iter(struct kiocb *iocb, struct iov_iter *from)
1224 {
1225 	struct file *file = iocb->ki_filp;
1226 	struct tun_struct *tun = tun_get(file);
1227 	struct tun_file *tfile = file->private_data;
1228 	ssize_t result;
1229 
1230 	if (!tun)
1231 		return -EBADFD;
1232 
1233 	result = tun_get_user(tun, tfile, NULL, from, file->f_flags & O_NONBLOCK);
1234 
1235 	tun_put(tun);
1236 	return result;
1237 }
1238 
1239 /* Put packet to the user space buffer */
1240 static ssize_t tun_put_user(struct tun_struct *tun,
1241 			    struct tun_file *tfile,
1242 			    struct sk_buff *skb,
1243 			    struct iov_iter *iter)
1244 {
1245 	struct tun_pi pi = { 0, skb->protocol };
1246 	ssize_t total;
1247 	int vlan_offset = 0;
1248 	int vlan_hlen = 0;
1249 	int vnet_hdr_sz = 0;
1250 
1251 	if (skb_vlan_tag_present(skb))
1252 		vlan_hlen = VLAN_HLEN;
1253 
1254 	if (tun->flags & IFF_VNET_HDR)
1255 		vnet_hdr_sz = tun->vnet_hdr_sz;
1256 
1257 	total = skb->len + vlan_hlen + vnet_hdr_sz;
1258 
1259 	if (!(tun->flags & IFF_NO_PI)) {
1260 		if (iov_iter_count(iter) < sizeof(pi))
1261 			return -EINVAL;
1262 
1263 		total += sizeof(pi);
1264 		if (iov_iter_count(iter) < total) {
1265 			/* Packet will be striped */
1266 			pi.flags |= TUN_PKT_STRIP;
1267 		}
1268 
1269 		if (copy_to_iter(&pi, sizeof(pi), iter) != sizeof(pi))
1270 			return -EFAULT;
1271 	}
1272 
1273 	if (vnet_hdr_sz) {
1274 		struct virtio_net_hdr gso = { 0 }; /* no info leak */
1275 		if (iov_iter_count(iter) < vnet_hdr_sz)
1276 			return -EINVAL;
1277 
1278 		if (skb_is_gso(skb)) {
1279 			struct skb_shared_info *sinfo = skb_shinfo(skb);
1280 
1281 			/* This is a hint as to how much should be linear. */
1282 			gso.hdr_len = cpu_to_tun16(tun, skb_headlen(skb));
1283 			gso.gso_size = cpu_to_tun16(tun, sinfo->gso_size);
1284 			if (sinfo->gso_type & SKB_GSO_TCPV4)
1285 				gso.gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
1286 			else if (sinfo->gso_type & SKB_GSO_TCPV6)
1287 				gso.gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
1288 			else if (sinfo->gso_type & SKB_GSO_UDP)
1289 				gso.gso_type = VIRTIO_NET_HDR_GSO_UDP;
1290 			else {
1291 				pr_err("unexpected GSO type: "
1292 				       "0x%x, gso_size %d, hdr_len %d\n",
1293 				       sinfo->gso_type, tun16_to_cpu(tun, gso.gso_size),
1294 				       tun16_to_cpu(tun, gso.hdr_len));
1295 				print_hex_dump(KERN_ERR, "tun: ",
1296 					       DUMP_PREFIX_NONE,
1297 					       16, 1, skb->head,
1298 					       min((int)tun16_to_cpu(tun, gso.hdr_len), 64), true);
1299 				WARN_ON_ONCE(1);
1300 				return -EINVAL;
1301 			}
1302 			if (sinfo->gso_type & SKB_GSO_TCP_ECN)
1303 				gso.gso_type |= VIRTIO_NET_HDR_GSO_ECN;
1304 		} else
1305 			gso.gso_type = VIRTIO_NET_HDR_GSO_NONE;
1306 
1307 		if (skb->ip_summed == CHECKSUM_PARTIAL) {
1308 			gso.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
1309 			gso.csum_start = cpu_to_tun16(tun, skb_checksum_start_offset(skb) +
1310 						      vlan_hlen);
1311 			gso.csum_offset = cpu_to_tun16(tun, skb->csum_offset);
1312 		} else if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
1313 			gso.flags = VIRTIO_NET_HDR_F_DATA_VALID;
1314 		} /* else everything is zero */
1315 
1316 		if (copy_to_iter(&gso, sizeof(gso), iter) != sizeof(gso))
1317 			return -EFAULT;
1318 
1319 		iov_iter_advance(iter, vnet_hdr_sz - sizeof(gso));
1320 	}
1321 
1322 	if (vlan_hlen) {
1323 		int ret;
1324 		struct {
1325 			__be16 h_vlan_proto;
1326 			__be16 h_vlan_TCI;
1327 		} veth;
1328 
1329 		veth.h_vlan_proto = skb->vlan_proto;
1330 		veth.h_vlan_TCI = htons(skb_vlan_tag_get(skb));
1331 
1332 		vlan_offset = offsetof(struct vlan_ethhdr, h_vlan_proto);
1333 
1334 		ret = skb_copy_datagram_iter(skb, 0, iter, vlan_offset);
1335 		if (ret || !iov_iter_count(iter))
1336 			goto done;
1337 
1338 		ret = copy_to_iter(&veth, sizeof(veth), iter);
1339 		if (ret != sizeof(veth) || !iov_iter_count(iter))
1340 			goto done;
1341 	}
1342 
1343 	skb_copy_datagram_iter(skb, vlan_offset, iter, skb->len - vlan_offset);
1344 
1345 done:
1346 	tun->dev->stats.tx_packets++;
1347 	tun->dev->stats.tx_bytes += skb->len + vlan_hlen;
1348 
1349 	return total;
1350 }
1351 
1352 static ssize_t tun_do_read(struct tun_struct *tun, struct tun_file *tfile,
1353 			   struct iov_iter *to,
1354 			   int noblock)
1355 {
1356 	struct sk_buff *skb;
1357 	ssize_t ret;
1358 	int peeked, err, off = 0;
1359 
1360 	tun_debug(KERN_INFO, tun, "tun_do_read\n");
1361 
1362 	if (!iov_iter_count(to))
1363 		return 0;
1364 
1365 	if (tun->dev->reg_state != NETREG_REGISTERED)
1366 		return -EIO;
1367 
1368 	/* Read frames from queue */
1369 	skb = __skb_recv_datagram(tfile->socket.sk, noblock ? MSG_DONTWAIT : 0,
1370 				  &peeked, &off, &err);
1371 	if (!skb)
1372 		return err;
1373 
1374 	ret = tun_put_user(tun, tfile, skb, to);
1375 	if (unlikely(ret < 0))
1376 		kfree_skb(skb);
1377 	else
1378 		consume_skb(skb);
1379 
1380 	return ret;
1381 }
1382 
1383 static ssize_t tun_chr_read_iter(struct kiocb *iocb, struct iov_iter *to)
1384 {
1385 	struct file *file = iocb->ki_filp;
1386 	struct tun_file *tfile = file->private_data;
1387 	struct tun_struct *tun = __tun_get(tfile);
1388 	ssize_t len = iov_iter_count(to), ret;
1389 
1390 	if (!tun)
1391 		return -EBADFD;
1392 	ret = tun_do_read(tun, tfile, to, file->f_flags & O_NONBLOCK);
1393 	ret = min_t(ssize_t, ret, len);
1394 	if (ret > 0)
1395 		iocb->ki_pos = ret;
1396 	tun_put(tun);
1397 	return ret;
1398 }
1399 
1400 static void tun_free_netdev(struct net_device *dev)
1401 {
1402 	struct tun_struct *tun = netdev_priv(dev);
1403 
1404 	BUG_ON(!(list_empty(&tun->disabled)));
1405 	tun_flow_uninit(tun);
1406 	security_tun_dev_free_security(tun->security);
1407 	free_netdev(dev);
1408 }
1409 
1410 static void tun_setup(struct net_device *dev)
1411 {
1412 	struct tun_struct *tun = netdev_priv(dev);
1413 
1414 	tun->owner = INVALID_UID;
1415 	tun->group = INVALID_GID;
1416 
1417 	dev->ethtool_ops = &tun_ethtool_ops;
1418 	dev->destructor = tun_free_netdev;
1419 }
1420 
1421 /* Trivial set of netlink ops to allow deleting tun or tap
1422  * device with netlink.
1423  */
1424 static int tun_validate(struct nlattr *tb[], struct nlattr *data[])
1425 {
1426 	return -EINVAL;
1427 }
1428 
1429 static struct rtnl_link_ops tun_link_ops __read_mostly = {
1430 	.kind		= DRV_NAME,
1431 	.priv_size	= sizeof(struct tun_struct),
1432 	.setup		= tun_setup,
1433 	.validate	= tun_validate,
1434 };
1435 
1436 static void tun_sock_write_space(struct sock *sk)
1437 {
1438 	struct tun_file *tfile;
1439 	wait_queue_head_t *wqueue;
1440 
1441 	if (!sock_writeable(sk))
1442 		return;
1443 
1444 	if (!test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags))
1445 		return;
1446 
1447 	wqueue = sk_sleep(sk);
1448 	if (wqueue && waitqueue_active(wqueue))
1449 		wake_up_interruptible_sync_poll(wqueue, POLLOUT |
1450 						POLLWRNORM | POLLWRBAND);
1451 
1452 	tfile = container_of(sk, struct tun_file, sk);
1453 	kill_fasync(&tfile->fasync, SIGIO, POLL_OUT);
1454 }
1455 
1456 static int tun_sendmsg(struct socket *sock, struct msghdr *m, size_t total_len)
1457 {
1458 	int ret;
1459 	struct tun_file *tfile = container_of(sock, struct tun_file, socket);
1460 	struct tun_struct *tun = __tun_get(tfile);
1461 
1462 	if (!tun)
1463 		return -EBADFD;
1464 
1465 	ret = tun_get_user(tun, tfile, m->msg_control, &m->msg_iter,
1466 			   m->msg_flags & MSG_DONTWAIT);
1467 	tun_put(tun);
1468 	return ret;
1469 }
1470 
1471 static int tun_recvmsg(struct socket *sock, struct msghdr *m, size_t total_len,
1472 		       int flags)
1473 {
1474 	struct tun_file *tfile = container_of(sock, struct tun_file, socket);
1475 	struct tun_struct *tun = __tun_get(tfile);
1476 	int ret;
1477 
1478 	if (!tun)
1479 		return -EBADFD;
1480 
1481 	if (flags & ~(MSG_DONTWAIT|MSG_TRUNC|MSG_ERRQUEUE)) {
1482 		ret = -EINVAL;
1483 		goto out;
1484 	}
1485 	if (flags & MSG_ERRQUEUE) {
1486 		ret = sock_recv_errqueue(sock->sk, m, total_len,
1487 					 SOL_PACKET, TUN_TX_TIMESTAMP);
1488 		goto out;
1489 	}
1490 	ret = tun_do_read(tun, tfile, &m->msg_iter, flags & MSG_DONTWAIT);
1491 	if (ret > (ssize_t)total_len) {
1492 		m->msg_flags |= MSG_TRUNC;
1493 		ret = flags & MSG_TRUNC ? ret : total_len;
1494 	}
1495 out:
1496 	tun_put(tun);
1497 	return ret;
1498 }
1499 
1500 static int tun_release(struct socket *sock)
1501 {
1502 	if (sock->sk)
1503 		sock_put(sock->sk);
1504 	return 0;
1505 }
1506 
1507 /* Ops structure to mimic raw sockets with tun */
1508 static const struct proto_ops tun_socket_ops = {
1509 	.sendmsg = tun_sendmsg,
1510 	.recvmsg = tun_recvmsg,
1511 	.release = tun_release,
1512 };
1513 
1514 static struct proto tun_proto = {
1515 	.name		= "tun",
1516 	.owner		= THIS_MODULE,
1517 	.obj_size	= sizeof(struct tun_file),
1518 };
1519 
1520 static int tun_flags(struct tun_struct *tun)
1521 {
1522 	return tun->flags & (TUN_FEATURES | IFF_PERSIST | IFF_TUN | IFF_TAP);
1523 }
1524 
1525 static ssize_t tun_show_flags(struct device *dev, struct device_attribute *attr,
1526 			      char *buf)
1527 {
1528 	struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1529 	return sprintf(buf, "0x%x\n", tun_flags(tun));
1530 }
1531 
1532 static ssize_t tun_show_owner(struct device *dev, struct device_attribute *attr,
1533 			      char *buf)
1534 {
1535 	struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1536 	return uid_valid(tun->owner)?
1537 		sprintf(buf, "%u\n",
1538 			from_kuid_munged(current_user_ns(), tun->owner)):
1539 		sprintf(buf, "-1\n");
1540 }
1541 
1542 static ssize_t tun_show_group(struct device *dev, struct device_attribute *attr,
1543 			      char *buf)
1544 {
1545 	struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1546 	return gid_valid(tun->group) ?
1547 		sprintf(buf, "%u\n",
1548 			from_kgid_munged(current_user_ns(), tun->group)):
1549 		sprintf(buf, "-1\n");
1550 }
1551 
1552 static DEVICE_ATTR(tun_flags, 0444, tun_show_flags, NULL);
1553 static DEVICE_ATTR(owner, 0444, tun_show_owner, NULL);
1554 static DEVICE_ATTR(group, 0444, tun_show_group, NULL);
1555 
1556 static struct attribute *tun_dev_attrs[] = {
1557 	&dev_attr_tun_flags.attr,
1558 	&dev_attr_owner.attr,
1559 	&dev_attr_group.attr,
1560 	NULL
1561 };
1562 
1563 static const struct attribute_group tun_attr_group = {
1564 	.attrs = tun_dev_attrs
1565 };
1566 
1567 static int tun_set_iff(struct net *net, struct file *file, struct ifreq *ifr)
1568 {
1569 	struct tun_struct *tun;
1570 	struct tun_file *tfile = file->private_data;
1571 	struct net_device *dev;
1572 	int err;
1573 
1574 	if (tfile->detached)
1575 		return -EINVAL;
1576 
1577 	dev = __dev_get_by_name(net, ifr->ifr_name);
1578 	if (dev) {
1579 		if (ifr->ifr_flags & IFF_TUN_EXCL)
1580 			return -EBUSY;
1581 		if ((ifr->ifr_flags & IFF_TUN) && dev->netdev_ops == &tun_netdev_ops)
1582 			tun = netdev_priv(dev);
1583 		else if ((ifr->ifr_flags & IFF_TAP) && dev->netdev_ops == &tap_netdev_ops)
1584 			tun = netdev_priv(dev);
1585 		else
1586 			return -EINVAL;
1587 
1588 		if (!!(ifr->ifr_flags & IFF_MULTI_QUEUE) !=
1589 		    !!(tun->flags & IFF_MULTI_QUEUE))
1590 			return -EINVAL;
1591 
1592 		if (tun_not_capable(tun))
1593 			return -EPERM;
1594 		err = security_tun_dev_open(tun->security);
1595 		if (err < 0)
1596 			return err;
1597 
1598 		err = tun_attach(tun, file, ifr->ifr_flags & IFF_NOFILTER);
1599 		if (err < 0)
1600 			return err;
1601 
1602 		if (tun->flags & IFF_MULTI_QUEUE &&
1603 		    (tun->numqueues + tun->numdisabled > 1)) {
1604 			/* One or more queue has already been attached, no need
1605 			 * to initialize the device again.
1606 			 */
1607 			return 0;
1608 		}
1609 	}
1610 	else {
1611 		char *name;
1612 		unsigned long flags = 0;
1613 		int queues = ifr->ifr_flags & IFF_MULTI_QUEUE ?
1614 			     MAX_TAP_QUEUES : 1;
1615 
1616 		if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
1617 			return -EPERM;
1618 		err = security_tun_dev_create();
1619 		if (err < 0)
1620 			return err;
1621 
1622 		/* Set dev type */
1623 		if (ifr->ifr_flags & IFF_TUN) {
1624 			/* TUN device */
1625 			flags |= IFF_TUN;
1626 			name = "tun%d";
1627 		} else if (ifr->ifr_flags & IFF_TAP) {
1628 			/* TAP device */
1629 			flags |= IFF_TAP;
1630 			name = "tap%d";
1631 		} else
1632 			return -EINVAL;
1633 
1634 		if (*ifr->ifr_name)
1635 			name = ifr->ifr_name;
1636 
1637 		dev = alloc_netdev_mqs(sizeof(struct tun_struct), name,
1638 				       NET_NAME_UNKNOWN, tun_setup, queues,
1639 				       queues);
1640 
1641 		if (!dev)
1642 			return -ENOMEM;
1643 
1644 		dev_net_set(dev, net);
1645 		dev->rtnl_link_ops = &tun_link_ops;
1646 		dev->ifindex = tfile->ifindex;
1647 		dev->sysfs_groups[0] = &tun_attr_group;
1648 
1649 		tun = netdev_priv(dev);
1650 		tun->dev = dev;
1651 		tun->flags = flags;
1652 		tun->txflt.count = 0;
1653 		tun->vnet_hdr_sz = sizeof(struct virtio_net_hdr);
1654 
1655 		tun->filter_attached = false;
1656 		tun->sndbuf = tfile->socket.sk->sk_sndbuf;
1657 
1658 		spin_lock_init(&tun->lock);
1659 
1660 		err = security_tun_dev_alloc_security(&tun->security);
1661 		if (err < 0)
1662 			goto err_free_dev;
1663 
1664 		tun_net_init(dev);
1665 		tun_flow_init(tun);
1666 
1667 		dev->hw_features = NETIF_F_SG | NETIF_F_FRAGLIST |
1668 				   TUN_USER_FEATURES | NETIF_F_HW_VLAN_CTAG_TX |
1669 				   NETIF_F_HW_VLAN_STAG_TX;
1670 		dev->features = dev->hw_features;
1671 		dev->vlan_features = dev->features &
1672 				     ~(NETIF_F_HW_VLAN_CTAG_TX |
1673 				       NETIF_F_HW_VLAN_STAG_TX);
1674 
1675 		INIT_LIST_HEAD(&tun->disabled);
1676 		err = tun_attach(tun, file, false);
1677 		if (err < 0)
1678 			goto err_free_flow;
1679 
1680 		err = register_netdevice(tun->dev);
1681 		if (err < 0)
1682 			goto err_detach;
1683 	}
1684 
1685 	netif_carrier_on(tun->dev);
1686 
1687 	tun_debug(KERN_INFO, tun, "tun_set_iff\n");
1688 
1689 	tun->flags = (tun->flags & ~TUN_FEATURES) |
1690 		(ifr->ifr_flags & TUN_FEATURES);
1691 
1692 	/* Make sure persistent devices do not get stuck in
1693 	 * xoff state.
1694 	 */
1695 	if (netif_running(tun->dev))
1696 		netif_tx_wake_all_queues(tun->dev);
1697 
1698 	strcpy(ifr->ifr_name, tun->dev->name);
1699 	return 0;
1700 
1701 err_detach:
1702 	tun_detach_all(dev);
1703 err_free_flow:
1704 	tun_flow_uninit(tun);
1705 	security_tun_dev_free_security(tun->security);
1706 err_free_dev:
1707 	free_netdev(dev);
1708 	return err;
1709 }
1710 
1711 static void tun_get_iff(struct net *net, struct tun_struct *tun,
1712 		       struct ifreq *ifr)
1713 {
1714 	tun_debug(KERN_INFO, tun, "tun_get_iff\n");
1715 
1716 	strcpy(ifr->ifr_name, tun->dev->name);
1717 
1718 	ifr->ifr_flags = tun_flags(tun);
1719 
1720 }
1721 
1722 /* This is like a cut-down ethtool ops, except done via tun fd so no
1723  * privs required. */
1724 static int set_offload(struct tun_struct *tun, unsigned long arg)
1725 {
1726 	netdev_features_t features = 0;
1727 
1728 	if (arg & TUN_F_CSUM) {
1729 		features |= NETIF_F_HW_CSUM;
1730 		arg &= ~TUN_F_CSUM;
1731 
1732 		if (arg & (TUN_F_TSO4|TUN_F_TSO6)) {
1733 			if (arg & TUN_F_TSO_ECN) {
1734 				features |= NETIF_F_TSO_ECN;
1735 				arg &= ~TUN_F_TSO_ECN;
1736 			}
1737 			if (arg & TUN_F_TSO4)
1738 				features |= NETIF_F_TSO;
1739 			if (arg & TUN_F_TSO6)
1740 				features |= NETIF_F_TSO6;
1741 			arg &= ~(TUN_F_TSO4|TUN_F_TSO6);
1742 		}
1743 
1744 		if (arg & TUN_F_UFO) {
1745 			features |= NETIF_F_UFO;
1746 			arg &= ~TUN_F_UFO;
1747 		}
1748 	}
1749 
1750 	/* This gives the user a way to test for new features in future by
1751 	 * trying to set them. */
1752 	if (arg)
1753 		return -EINVAL;
1754 
1755 	tun->set_features = features;
1756 	netdev_update_features(tun->dev);
1757 
1758 	return 0;
1759 }
1760 
1761 static void tun_detach_filter(struct tun_struct *tun, int n)
1762 {
1763 	int i;
1764 	struct tun_file *tfile;
1765 
1766 	for (i = 0; i < n; i++) {
1767 		tfile = rtnl_dereference(tun->tfiles[i]);
1768 		sk_detach_filter(tfile->socket.sk);
1769 	}
1770 
1771 	tun->filter_attached = false;
1772 }
1773 
1774 static int tun_attach_filter(struct tun_struct *tun)
1775 {
1776 	int i, ret = 0;
1777 	struct tun_file *tfile;
1778 
1779 	for (i = 0; i < tun->numqueues; i++) {
1780 		tfile = rtnl_dereference(tun->tfiles[i]);
1781 		ret = sk_attach_filter(&tun->fprog, tfile->socket.sk);
1782 		if (ret) {
1783 			tun_detach_filter(tun, i);
1784 			return ret;
1785 		}
1786 	}
1787 
1788 	tun->filter_attached = true;
1789 	return ret;
1790 }
1791 
1792 static void tun_set_sndbuf(struct tun_struct *tun)
1793 {
1794 	struct tun_file *tfile;
1795 	int i;
1796 
1797 	for (i = 0; i < tun->numqueues; i++) {
1798 		tfile = rtnl_dereference(tun->tfiles[i]);
1799 		tfile->socket.sk->sk_sndbuf = tun->sndbuf;
1800 	}
1801 }
1802 
1803 static int tun_set_queue(struct file *file, struct ifreq *ifr)
1804 {
1805 	struct tun_file *tfile = file->private_data;
1806 	struct tun_struct *tun;
1807 	int ret = 0;
1808 
1809 	rtnl_lock();
1810 
1811 	if (ifr->ifr_flags & IFF_ATTACH_QUEUE) {
1812 		tun = tfile->detached;
1813 		if (!tun) {
1814 			ret = -EINVAL;
1815 			goto unlock;
1816 		}
1817 		ret = security_tun_dev_attach_queue(tun->security);
1818 		if (ret < 0)
1819 			goto unlock;
1820 		ret = tun_attach(tun, file, false);
1821 	} else if (ifr->ifr_flags & IFF_DETACH_QUEUE) {
1822 		tun = rtnl_dereference(tfile->tun);
1823 		if (!tun || !(tun->flags & IFF_MULTI_QUEUE) || tfile->detached)
1824 			ret = -EINVAL;
1825 		else
1826 			__tun_detach(tfile, false);
1827 	} else
1828 		ret = -EINVAL;
1829 
1830 unlock:
1831 	rtnl_unlock();
1832 	return ret;
1833 }
1834 
1835 static long __tun_chr_ioctl(struct file *file, unsigned int cmd,
1836 			    unsigned long arg, int ifreq_len)
1837 {
1838 	struct tun_file *tfile = file->private_data;
1839 	struct tun_struct *tun;
1840 	void __user* argp = (void __user*)arg;
1841 	struct ifreq ifr;
1842 	kuid_t owner;
1843 	kgid_t group;
1844 	int sndbuf;
1845 	int vnet_hdr_sz;
1846 	unsigned int ifindex;
1847 	int le;
1848 	int ret;
1849 
1850 	if (cmd == TUNSETIFF || cmd == TUNSETQUEUE || _IOC_TYPE(cmd) == 0x89) {
1851 		if (copy_from_user(&ifr, argp, ifreq_len))
1852 			return -EFAULT;
1853 	} else {
1854 		memset(&ifr, 0, sizeof(ifr));
1855 	}
1856 	if (cmd == TUNGETFEATURES) {
1857 		/* Currently this just means: "what IFF flags are valid?".
1858 		 * This is needed because we never checked for invalid flags on
1859 		 * TUNSETIFF.
1860 		 */
1861 		return put_user(IFF_TUN | IFF_TAP | TUN_FEATURES,
1862 				(unsigned int __user*)argp);
1863 	} else if (cmd == TUNSETQUEUE)
1864 		return tun_set_queue(file, &ifr);
1865 
1866 	ret = 0;
1867 	rtnl_lock();
1868 
1869 	tun = __tun_get(tfile);
1870 	if (cmd == TUNSETIFF && !tun) {
1871 		ifr.ifr_name[IFNAMSIZ-1] = '\0';
1872 
1873 		ret = tun_set_iff(tfile->net, file, &ifr);
1874 
1875 		if (ret)
1876 			goto unlock;
1877 
1878 		if (copy_to_user(argp, &ifr, ifreq_len))
1879 			ret = -EFAULT;
1880 		goto unlock;
1881 	}
1882 	if (cmd == TUNSETIFINDEX) {
1883 		ret = -EPERM;
1884 		if (tun)
1885 			goto unlock;
1886 
1887 		ret = -EFAULT;
1888 		if (copy_from_user(&ifindex, argp, sizeof(ifindex)))
1889 			goto unlock;
1890 
1891 		ret = 0;
1892 		tfile->ifindex = ifindex;
1893 		goto unlock;
1894 	}
1895 
1896 	ret = -EBADFD;
1897 	if (!tun)
1898 		goto unlock;
1899 
1900 	tun_debug(KERN_INFO, tun, "tun_chr_ioctl cmd %u\n", cmd);
1901 
1902 	ret = 0;
1903 	switch (cmd) {
1904 	case TUNGETIFF:
1905 		tun_get_iff(current->nsproxy->net_ns, tun, &ifr);
1906 
1907 		if (tfile->detached)
1908 			ifr.ifr_flags |= IFF_DETACH_QUEUE;
1909 		if (!tfile->socket.sk->sk_filter)
1910 			ifr.ifr_flags |= IFF_NOFILTER;
1911 
1912 		if (copy_to_user(argp, &ifr, ifreq_len))
1913 			ret = -EFAULT;
1914 		break;
1915 
1916 	case TUNSETNOCSUM:
1917 		/* Disable/Enable checksum */
1918 
1919 		/* [unimplemented] */
1920 		tun_debug(KERN_INFO, tun, "ignored: set checksum %s\n",
1921 			  arg ? "disabled" : "enabled");
1922 		break;
1923 
1924 	case TUNSETPERSIST:
1925 		/* Disable/Enable persist mode. Keep an extra reference to the
1926 		 * module to prevent the module being unprobed.
1927 		 */
1928 		if (arg && !(tun->flags & IFF_PERSIST)) {
1929 			tun->flags |= IFF_PERSIST;
1930 			__module_get(THIS_MODULE);
1931 		}
1932 		if (!arg && (tun->flags & IFF_PERSIST)) {
1933 			tun->flags &= ~IFF_PERSIST;
1934 			module_put(THIS_MODULE);
1935 		}
1936 
1937 		tun_debug(KERN_INFO, tun, "persist %s\n",
1938 			  arg ? "enabled" : "disabled");
1939 		break;
1940 
1941 	case TUNSETOWNER:
1942 		/* Set owner of the device */
1943 		owner = make_kuid(current_user_ns(), arg);
1944 		if (!uid_valid(owner)) {
1945 			ret = -EINVAL;
1946 			break;
1947 		}
1948 		tun->owner = owner;
1949 		tun_debug(KERN_INFO, tun, "owner set to %u\n",
1950 			  from_kuid(&init_user_ns, tun->owner));
1951 		break;
1952 
1953 	case TUNSETGROUP:
1954 		/* Set group of the device */
1955 		group = make_kgid(current_user_ns(), arg);
1956 		if (!gid_valid(group)) {
1957 			ret = -EINVAL;
1958 			break;
1959 		}
1960 		tun->group = group;
1961 		tun_debug(KERN_INFO, tun, "group set to %u\n",
1962 			  from_kgid(&init_user_ns, tun->group));
1963 		break;
1964 
1965 	case TUNSETLINK:
1966 		/* Only allow setting the type when the interface is down */
1967 		if (tun->dev->flags & IFF_UP) {
1968 			tun_debug(KERN_INFO, tun,
1969 				  "Linktype set failed because interface is up\n");
1970 			ret = -EBUSY;
1971 		} else {
1972 			tun->dev->type = (int) arg;
1973 			tun_debug(KERN_INFO, tun, "linktype set to %d\n",
1974 				  tun->dev->type);
1975 			ret = 0;
1976 		}
1977 		break;
1978 
1979 #ifdef TUN_DEBUG
1980 	case TUNSETDEBUG:
1981 		tun->debug = arg;
1982 		break;
1983 #endif
1984 	case TUNSETOFFLOAD:
1985 		ret = set_offload(tun, arg);
1986 		break;
1987 
1988 	case TUNSETTXFILTER:
1989 		/* Can be set only for TAPs */
1990 		ret = -EINVAL;
1991 		if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
1992 			break;
1993 		ret = update_filter(&tun->txflt, (void __user *)arg);
1994 		break;
1995 
1996 	case SIOCGIFHWADDR:
1997 		/* Get hw address */
1998 		memcpy(ifr.ifr_hwaddr.sa_data, tun->dev->dev_addr, ETH_ALEN);
1999 		ifr.ifr_hwaddr.sa_family = tun->dev->type;
2000 		if (copy_to_user(argp, &ifr, ifreq_len))
2001 			ret = -EFAULT;
2002 		break;
2003 
2004 	case SIOCSIFHWADDR:
2005 		/* Set hw address */
2006 		tun_debug(KERN_DEBUG, tun, "set hw address: %pM\n",
2007 			  ifr.ifr_hwaddr.sa_data);
2008 
2009 		ret = dev_set_mac_address(tun->dev, &ifr.ifr_hwaddr);
2010 		break;
2011 
2012 	case TUNGETSNDBUF:
2013 		sndbuf = tfile->socket.sk->sk_sndbuf;
2014 		if (copy_to_user(argp, &sndbuf, sizeof(sndbuf)))
2015 			ret = -EFAULT;
2016 		break;
2017 
2018 	case TUNSETSNDBUF:
2019 		if (copy_from_user(&sndbuf, argp, sizeof(sndbuf))) {
2020 			ret = -EFAULT;
2021 			break;
2022 		}
2023 
2024 		tun->sndbuf = sndbuf;
2025 		tun_set_sndbuf(tun);
2026 		break;
2027 
2028 	case TUNGETVNETHDRSZ:
2029 		vnet_hdr_sz = tun->vnet_hdr_sz;
2030 		if (copy_to_user(argp, &vnet_hdr_sz, sizeof(vnet_hdr_sz)))
2031 			ret = -EFAULT;
2032 		break;
2033 
2034 	case TUNSETVNETHDRSZ:
2035 		if (copy_from_user(&vnet_hdr_sz, argp, sizeof(vnet_hdr_sz))) {
2036 			ret = -EFAULT;
2037 			break;
2038 		}
2039 		if (vnet_hdr_sz < (int)sizeof(struct virtio_net_hdr)) {
2040 			ret = -EINVAL;
2041 			break;
2042 		}
2043 
2044 		tun->vnet_hdr_sz = vnet_hdr_sz;
2045 		break;
2046 
2047 	case TUNGETVNETLE:
2048 		le = !!(tun->flags & TUN_VNET_LE);
2049 		if (put_user(le, (int __user *)argp))
2050 			ret = -EFAULT;
2051 		break;
2052 
2053 	case TUNSETVNETLE:
2054 		if (get_user(le, (int __user *)argp)) {
2055 			ret = -EFAULT;
2056 			break;
2057 		}
2058 		if (le)
2059 			tun->flags |= TUN_VNET_LE;
2060 		else
2061 			tun->flags &= ~TUN_VNET_LE;
2062 		break;
2063 
2064 	case TUNATTACHFILTER:
2065 		/* Can be set only for TAPs */
2066 		ret = -EINVAL;
2067 		if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
2068 			break;
2069 		ret = -EFAULT;
2070 		if (copy_from_user(&tun->fprog, argp, sizeof(tun->fprog)))
2071 			break;
2072 
2073 		ret = tun_attach_filter(tun);
2074 		break;
2075 
2076 	case TUNDETACHFILTER:
2077 		/* Can be set only for TAPs */
2078 		ret = -EINVAL;
2079 		if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
2080 			break;
2081 		ret = 0;
2082 		tun_detach_filter(tun, tun->numqueues);
2083 		break;
2084 
2085 	case TUNGETFILTER:
2086 		ret = -EINVAL;
2087 		if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
2088 			break;
2089 		ret = -EFAULT;
2090 		if (copy_to_user(argp, &tun->fprog, sizeof(tun->fprog)))
2091 			break;
2092 		ret = 0;
2093 		break;
2094 
2095 	default:
2096 		ret = -EINVAL;
2097 		break;
2098 	}
2099 
2100 unlock:
2101 	rtnl_unlock();
2102 	if (tun)
2103 		tun_put(tun);
2104 	return ret;
2105 }
2106 
2107 static long tun_chr_ioctl(struct file *file,
2108 			  unsigned int cmd, unsigned long arg)
2109 {
2110 	return __tun_chr_ioctl(file, cmd, arg, sizeof (struct ifreq));
2111 }
2112 
2113 #ifdef CONFIG_COMPAT
2114 static long tun_chr_compat_ioctl(struct file *file,
2115 			 unsigned int cmd, unsigned long arg)
2116 {
2117 	switch (cmd) {
2118 	case TUNSETIFF:
2119 	case TUNGETIFF:
2120 	case TUNSETTXFILTER:
2121 	case TUNGETSNDBUF:
2122 	case TUNSETSNDBUF:
2123 	case SIOCGIFHWADDR:
2124 	case SIOCSIFHWADDR:
2125 		arg = (unsigned long)compat_ptr(arg);
2126 		break;
2127 	default:
2128 		arg = (compat_ulong_t)arg;
2129 		break;
2130 	}
2131 
2132 	/*
2133 	 * compat_ifreq is shorter than ifreq, so we must not access beyond
2134 	 * the end of that structure. All fields that are used in this
2135 	 * driver are compatible though, we don't need to convert the
2136 	 * contents.
2137 	 */
2138 	return __tun_chr_ioctl(file, cmd, arg, sizeof(struct compat_ifreq));
2139 }
2140 #endif /* CONFIG_COMPAT */
2141 
2142 static int tun_chr_fasync(int fd, struct file *file, int on)
2143 {
2144 	struct tun_file *tfile = file->private_data;
2145 	int ret;
2146 
2147 	if ((ret = fasync_helper(fd, file, on, &tfile->fasync)) < 0)
2148 		goto out;
2149 
2150 	if (on) {
2151 		__f_setown(file, task_pid(current), PIDTYPE_PID, 0);
2152 		tfile->flags |= TUN_FASYNC;
2153 	} else
2154 		tfile->flags &= ~TUN_FASYNC;
2155 	ret = 0;
2156 out:
2157 	return ret;
2158 }
2159 
2160 static int tun_chr_open(struct inode *inode, struct file * file)
2161 {
2162 	struct tun_file *tfile;
2163 
2164 	DBG1(KERN_INFO, "tunX: tun_chr_open\n");
2165 
2166 	tfile = (struct tun_file *)sk_alloc(&init_net, AF_UNSPEC, GFP_KERNEL,
2167 					    &tun_proto);
2168 	if (!tfile)
2169 		return -ENOMEM;
2170 	RCU_INIT_POINTER(tfile->tun, NULL);
2171 	tfile->net = get_net(current->nsproxy->net_ns);
2172 	tfile->flags = 0;
2173 	tfile->ifindex = 0;
2174 
2175 	init_waitqueue_head(&tfile->wq.wait);
2176 	RCU_INIT_POINTER(tfile->socket.wq, &tfile->wq);
2177 
2178 	tfile->socket.file = file;
2179 	tfile->socket.ops = &tun_socket_ops;
2180 
2181 	sock_init_data(&tfile->socket, &tfile->sk);
2182 	sk_change_net(&tfile->sk, tfile->net);
2183 
2184 	tfile->sk.sk_write_space = tun_sock_write_space;
2185 	tfile->sk.sk_sndbuf = INT_MAX;
2186 
2187 	file->private_data = tfile;
2188 	set_bit(SOCK_EXTERNALLY_ALLOCATED, &tfile->socket.flags);
2189 	INIT_LIST_HEAD(&tfile->next);
2190 
2191 	sock_set_flag(&tfile->sk, SOCK_ZEROCOPY);
2192 
2193 	return 0;
2194 }
2195 
2196 static int tun_chr_close(struct inode *inode, struct file *file)
2197 {
2198 	struct tun_file *tfile = file->private_data;
2199 	struct net *net = tfile->net;
2200 
2201 	tun_detach(tfile, true);
2202 	put_net(net);
2203 
2204 	return 0;
2205 }
2206 
2207 #ifdef CONFIG_PROC_FS
2208 static void tun_chr_show_fdinfo(struct seq_file *m, struct file *f)
2209 {
2210 	struct tun_struct *tun;
2211 	struct ifreq ifr;
2212 
2213 	memset(&ifr, 0, sizeof(ifr));
2214 
2215 	rtnl_lock();
2216 	tun = tun_get(f);
2217 	if (tun)
2218 		tun_get_iff(current->nsproxy->net_ns, tun, &ifr);
2219 	rtnl_unlock();
2220 
2221 	if (tun)
2222 		tun_put(tun);
2223 
2224 	seq_printf(m, "iff:\t%s\n", ifr.ifr_name);
2225 }
2226 #endif
2227 
2228 static const struct file_operations tun_fops = {
2229 	.owner	= THIS_MODULE,
2230 	.llseek = no_llseek,
2231 	.read_iter  = tun_chr_read_iter,
2232 	.write_iter = tun_chr_write_iter,
2233 	.poll	= tun_chr_poll,
2234 	.unlocked_ioctl	= tun_chr_ioctl,
2235 #ifdef CONFIG_COMPAT
2236 	.compat_ioctl = tun_chr_compat_ioctl,
2237 #endif
2238 	.open	= tun_chr_open,
2239 	.release = tun_chr_close,
2240 	.fasync = tun_chr_fasync,
2241 #ifdef CONFIG_PROC_FS
2242 	.show_fdinfo = tun_chr_show_fdinfo,
2243 #endif
2244 };
2245 
2246 static struct miscdevice tun_miscdev = {
2247 	.minor = TUN_MINOR,
2248 	.name = "tun",
2249 	.nodename = "net/tun",
2250 	.fops = &tun_fops,
2251 };
2252 
2253 /* ethtool interface */
2254 
2255 static int tun_get_settings(struct net_device *dev, struct ethtool_cmd *cmd)
2256 {
2257 	cmd->supported		= 0;
2258 	cmd->advertising	= 0;
2259 	ethtool_cmd_speed_set(cmd, SPEED_10);
2260 	cmd->duplex		= DUPLEX_FULL;
2261 	cmd->port		= PORT_TP;
2262 	cmd->phy_address	= 0;
2263 	cmd->transceiver	= XCVR_INTERNAL;
2264 	cmd->autoneg		= AUTONEG_DISABLE;
2265 	cmd->maxtxpkt		= 0;
2266 	cmd->maxrxpkt		= 0;
2267 	return 0;
2268 }
2269 
2270 static void tun_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
2271 {
2272 	struct tun_struct *tun = netdev_priv(dev);
2273 
2274 	strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
2275 	strlcpy(info->version, DRV_VERSION, sizeof(info->version));
2276 
2277 	switch (tun->flags & TUN_TYPE_MASK) {
2278 	case IFF_TUN:
2279 		strlcpy(info->bus_info, "tun", sizeof(info->bus_info));
2280 		break;
2281 	case IFF_TAP:
2282 		strlcpy(info->bus_info, "tap", sizeof(info->bus_info));
2283 		break;
2284 	}
2285 }
2286 
2287 static u32 tun_get_msglevel(struct net_device *dev)
2288 {
2289 #ifdef TUN_DEBUG
2290 	struct tun_struct *tun = netdev_priv(dev);
2291 	return tun->debug;
2292 #else
2293 	return -EOPNOTSUPP;
2294 #endif
2295 }
2296 
2297 static void tun_set_msglevel(struct net_device *dev, u32 value)
2298 {
2299 #ifdef TUN_DEBUG
2300 	struct tun_struct *tun = netdev_priv(dev);
2301 	tun->debug = value;
2302 #endif
2303 }
2304 
2305 static const struct ethtool_ops tun_ethtool_ops = {
2306 	.get_settings	= tun_get_settings,
2307 	.get_drvinfo	= tun_get_drvinfo,
2308 	.get_msglevel	= tun_get_msglevel,
2309 	.set_msglevel	= tun_set_msglevel,
2310 	.get_link	= ethtool_op_get_link,
2311 	.get_ts_info	= ethtool_op_get_ts_info,
2312 };
2313 
2314 
2315 static int __init tun_init(void)
2316 {
2317 	int ret = 0;
2318 
2319 	pr_info("%s, %s\n", DRV_DESCRIPTION, DRV_VERSION);
2320 	pr_info("%s\n", DRV_COPYRIGHT);
2321 
2322 	ret = rtnl_link_register(&tun_link_ops);
2323 	if (ret) {
2324 		pr_err("Can't register link_ops\n");
2325 		goto err_linkops;
2326 	}
2327 
2328 	ret = misc_register(&tun_miscdev);
2329 	if (ret) {
2330 		pr_err("Can't register misc device %d\n", TUN_MINOR);
2331 		goto err_misc;
2332 	}
2333 	return  0;
2334 err_misc:
2335 	rtnl_link_unregister(&tun_link_ops);
2336 err_linkops:
2337 	return ret;
2338 }
2339 
2340 static void tun_cleanup(void)
2341 {
2342 	misc_deregister(&tun_miscdev);
2343 	rtnl_link_unregister(&tun_link_ops);
2344 }
2345 
2346 /* Get an underlying socket object from tun file.  Returns error unless file is
2347  * attached to a device.  The returned object works like a packet socket, it
2348  * can be used for sock_sendmsg/sock_recvmsg.  The caller is responsible for
2349  * holding a reference to the file for as long as the socket is in use. */
2350 struct socket *tun_get_socket(struct file *file)
2351 {
2352 	struct tun_file *tfile;
2353 	if (file->f_op != &tun_fops)
2354 		return ERR_PTR(-EINVAL);
2355 	tfile = file->private_data;
2356 	if (!tfile)
2357 		return ERR_PTR(-EBADFD);
2358 	return &tfile->socket;
2359 }
2360 EXPORT_SYMBOL_GPL(tun_get_socket);
2361 
2362 module_init(tun_init);
2363 module_exit(tun_cleanup);
2364 MODULE_DESCRIPTION(DRV_DESCRIPTION);
2365 MODULE_AUTHOR(DRV_COPYRIGHT);
2366 MODULE_LICENSE("GPL");
2367 MODULE_ALIAS_MISCDEV(TUN_MINOR);
2368 MODULE_ALIAS("devname:net/tun");
2369