xref: /openbmc/linux/drivers/net/tun.c (revision 7b6d864b)
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/crc32.h>
64 #include <linux/nsproxy.h>
65 #include <linux/virtio_net.h>
66 #include <linux/rcupdate.h>
67 #include <net/net_namespace.h>
68 #include <net/netns/generic.h>
69 #include <net/rtnetlink.h>
70 #include <net/sock.h>
71 
72 #include <asm/uaccess.h>
73 
74 /* Uncomment to enable debugging */
75 /* #define TUN_DEBUG 1 */
76 
77 #ifdef TUN_DEBUG
78 static int debug;
79 
80 #define tun_debug(level, tun, fmt, args...)			\
81 do {								\
82 	if (tun->debug)						\
83 		netdev_printk(level, tun->dev, fmt, ##args);	\
84 } while (0)
85 #define DBG1(level, fmt, args...)				\
86 do {								\
87 	if (debug == 2)						\
88 		printk(level fmt, ##args);			\
89 } while (0)
90 #else
91 #define tun_debug(level, tun, fmt, args...)			\
92 do {								\
93 	if (0)							\
94 		netdev_printk(level, tun->dev, fmt, ##args);	\
95 } while (0)
96 #define DBG1(level, fmt, args...)				\
97 do {								\
98 	if (0)							\
99 		printk(level fmt, ##args);			\
100 } while (0)
101 #endif
102 
103 #define GOODCOPY_LEN 128
104 
105 #define FLT_EXACT_COUNT 8
106 struct tap_filter {
107 	unsigned int    count;    /* Number of addrs. Zero means disabled */
108 	u32             mask[2];  /* Mask of the hashed addrs */
109 	unsigned char	addr[FLT_EXACT_COUNT][ETH_ALEN];
110 };
111 
112 /* DEFAULT_MAX_NUM_RSS_QUEUES were choosed to let the rx/tx queues allocated for
113  * the netdevice to be fit in one page. So we can make sure the success of
114  * memory allocation. TODO: increase the limit. */
115 #define MAX_TAP_QUEUES DEFAULT_MAX_NUM_RSS_QUEUES
116 #define MAX_TAP_FLOWS  4096
117 
118 #define TUN_FLOW_EXPIRE (3 * HZ)
119 
120 /* A tun_file connects an open character device to a tuntap netdevice. It
121  * also contains all socket related strctures (except sock_fprog and tap_filter)
122  * to serve as one transmit queue for tuntap device. The sock_fprog and
123  * tap_filter were kept in tun_struct since they were used for filtering for the
124  * netdevice not for a specific queue (at least I didn't see the requirement for
125  * this).
126  *
127  * RCU usage:
128  * The tun_file and tun_struct are loosely coupled, the pointer from one to the
129  * other can only be read while rcu_read_lock or rtnl_lock is held.
130  */
131 struct tun_file {
132 	struct sock sk;
133 	struct socket socket;
134 	struct socket_wq wq;
135 	struct tun_struct __rcu *tun;
136 	struct net *net;
137 	struct fasync_struct *fasync;
138 	/* only used for fasnyc */
139 	unsigned int flags;
140 	u16 queue_index;
141 	struct list_head next;
142 	struct tun_struct *detached;
143 };
144 
145 struct tun_flow_entry {
146 	struct hlist_node hash_link;
147 	struct rcu_head rcu;
148 	struct tun_struct *tun;
149 
150 	u32 rxhash;
151 	int queue_index;
152 	unsigned long updated;
153 };
154 
155 #define TUN_NUM_FLOW_ENTRIES 1024
156 
157 /* Since the socket were moved to tun_file, to preserve the behavior of persist
158  * device, socket filter, sndbuf and vnet header size were restore when the
159  * file were attached to a persist device.
160  */
161 struct tun_struct {
162 	struct tun_file __rcu	*tfiles[MAX_TAP_QUEUES];
163 	unsigned int            numqueues;
164 	unsigned int 		flags;
165 	kuid_t			owner;
166 	kgid_t			group;
167 
168 	struct net_device	*dev;
169 	netdev_features_t	set_features;
170 #define TUN_USER_FEATURES (NETIF_F_HW_CSUM|NETIF_F_TSO_ECN|NETIF_F_TSO| \
171 			  NETIF_F_TSO6|NETIF_F_UFO)
172 
173 	int			vnet_hdr_sz;
174 	int			sndbuf;
175 	struct tap_filter	txflt;
176 	struct sock_fprog	fprog;
177 	/* protected by rtnl lock */
178 	bool			filter_attached;
179 #ifdef TUN_DEBUG
180 	int debug;
181 #endif
182 	spinlock_t lock;
183 	struct hlist_head flows[TUN_NUM_FLOW_ENTRIES];
184 	struct timer_list flow_gc_timer;
185 	unsigned long ageing_time;
186 	unsigned int numdisabled;
187 	struct list_head disabled;
188 	void *security;
189 	u32 flow_count;
190 };
191 
192 static inline u32 tun_hashfn(u32 rxhash)
193 {
194 	return rxhash & 0x3ff;
195 }
196 
197 static struct tun_flow_entry *tun_flow_find(struct hlist_head *head, u32 rxhash)
198 {
199 	struct tun_flow_entry *e;
200 
201 	hlist_for_each_entry_rcu(e, head, hash_link) {
202 		if (e->rxhash == rxhash)
203 			return e;
204 	}
205 	return NULL;
206 }
207 
208 static struct tun_flow_entry *tun_flow_create(struct tun_struct *tun,
209 					      struct hlist_head *head,
210 					      u32 rxhash, u16 queue_index)
211 {
212 	struct tun_flow_entry *e = kmalloc(sizeof(*e), GFP_ATOMIC);
213 
214 	if (e) {
215 		tun_debug(KERN_INFO, tun, "create flow: hash %u index %u\n",
216 			  rxhash, queue_index);
217 		e->updated = jiffies;
218 		e->rxhash = rxhash;
219 		e->queue_index = queue_index;
220 		e->tun = tun;
221 		hlist_add_head_rcu(&e->hash_link, head);
222 		++tun->flow_count;
223 	}
224 	return e;
225 }
226 
227 static void tun_flow_delete(struct tun_struct *tun, struct tun_flow_entry *e)
228 {
229 	tun_debug(KERN_INFO, tun, "delete flow: hash %u index %u\n",
230 		  e->rxhash, e->queue_index);
231 	hlist_del_rcu(&e->hash_link);
232 	kfree_rcu(e, rcu);
233 	--tun->flow_count;
234 }
235 
236 static void tun_flow_flush(struct tun_struct *tun)
237 {
238 	int i;
239 
240 	spin_lock_bh(&tun->lock);
241 	for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
242 		struct tun_flow_entry *e;
243 		struct hlist_node *n;
244 
245 		hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link)
246 			tun_flow_delete(tun, e);
247 	}
248 	spin_unlock_bh(&tun->lock);
249 }
250 
251 static void tun_flow_delete_by_queue(struct tun_struct *tun, u16 queue_index)
252 {
253 	int i;
254 
255 	spin_lock_bh(&tun->lock);
256 	for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
257 		struct tun_flow_entry *e;
258 		struct hlist_node *n;
259 
260 		hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
261 			if (e->queue_index == queue_index)
262 				tun_flow_delete(tun, e);
263 		}
264 	}
265 	spin_unlock_bh(&tun->lock);
266 }
267 
268 static void tun_flow_cleanup(unsigned long data)
269 {
270 	struct tun_struct *tun = (struct tun_struct *)data;
271 	unsigned long delay = tun->ageing_time;
272 	unsigned long next_timer = jiffies + delay;
273 	unsigned long count = 0;
274 	int i;
275 
276 	tun_debug(KERN_INFO, tun, "tun_flow_cleanup\n");
277 
278 	spin_lock_bh(&tun->lock);
279 	for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
280 		struct tun_flow_entry *e;
281 		struct hlist_node *n;
282 
283 		hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
284 			unsigned long this_timer;
285 			count++;
286 			this_timer = e->updated + delay;
287 			if (time_before_eq(this_timer, jiffies))
288 				tun_flow_delete(tun, e);
289 			else if (time_before(this_timer, next_timer))
290 				next_timer = this_timer;
291 		}
292 	}
293 
294 	if (count)
295 		mod_timer(&tun->flow_gc_timer, round_jiffies_up(next_timer));
296 	spin_unlock_bh(&tun->lock);
297 }
298 
299 static void tun_flow_update(struct tun_struct *tun, u32 rxhash,
300 			    struct tun_file *tfile)
301 {
302 	struct hlist_head *head;
303 	struct tun_flow_entry *e;
304 	unsigned long delay = tun->ageing_time;
305 	u16 queue_index = tfile->queue_index;
306 
307 	if (!rxhash)
308 		return;
309 	else
310 		head = &tun->flows[tun_hashfn(rxhash)];
311 
312 	rcu_read_lock();
313 
314 	/* We may get a very small possibility of OOO during switching, not
315 	 * worth to optimize.*/
316 	if (tun->numqueues == 1 || tfile->detached)
317 		goto unlock;
318 
319 	e = tun_flow_find(head, rxhash);
320 	if (likely(e)) {
321 		/* TODO: keep queueing to old queue until it's empty? */
322 		e->queue_index = queue_index;
323 		e->updated = jiffies;
324 	} else {
325 		spin_lock_bh(&tun->lock);
326 		if (!tun_flow_find(head, rxhash) &&
327 		    tun->flow_count < MAX_TAP_FLOWS)
328 			tun_flow_create(tun, head, rxhash, queue_index);
329 
330 		if (!timer_pending(&tun->flow_gc_timer))
331 			mod_timer(&tun->flow_gc_timer,
332 				  round_jiffies_up(jiffies + delay));
333 		spin_unlock_bh(&tun->lock);
334 	}
335 
336 unlock:
337 	rcu_read_unlock();
338 }
339 
340 /* We try to identify a flow through its rxhash first. The reason that
341  * we do not check rxq no. is becuase some cards(e.g 82599), chooses
342  * the rxq based on the txq where the last packet of the flow comes. As
343  * the userspace application move between processors, we may get a
344  * different rxq no. here. If we could not get rxhash, then we would
345  * hope the rxq no. may help here.
346  */
347 static u16 tun_select_queue(struct net_device *dev, struct sk_buff *skb)
348 {
349 	struct tun_struct *tun = netdev_priv(dev);
350 	struct tun_flow_entry *e;
351 	u32 txq = 0;
352 	u32 numqueues = 0;
353 
354 	rcu_read_lock();
355 	numqueues = ACCESS_ONCE(tun->numqueues);
356 
357 	txq = skb_get_rxhash(skb);
358 	if (txq) {
359 		e = tun_flow_find(&tun->flows[tun_hashfn(txq)], txq);
360 		if (e)
361 			txq = e->queue_index;
362 		else
363 			/* use multiply and shift instead of expensive divide */
364 			txq = ((u64)txq * numqueues) >> 32;
365 	} else if (likely(skb_rx_queue_recorded(skb))) {
366 		txq = skb_get_rx_queue(skb);
367 		while (unlikely(txq >= numqueues))
368 			txq -= numqueues;
369 	}
370 
371 	rcu_read_unlock();
372 	return txq;
373 }
374 
375 static inline bool tun_not_capable(struct tun_struct *tun)
376 {
377 	const struct cred *cred = current_cred();
378 	struct net *net = dev_net(tun->dev);
379 
380 	return ((uid_valid(tun->owner) && !uid_eq(cred->euid, tun->owner)) ||
381 		  (gid_valid(tun->group) && !in_egroup_p(tun->group))) &&
382 		!ns_capable(net->user_ns, CAP_NET_ADMIN);
383 }
384 
385 static void tun_set_real_num_queues(struct tun_struct *tun)
386 {
387 	netif_set_real_num_tx_queues(tun->dev, tun->numqueues);
388 	netif_set_real_num_rx_queues(tun->dev, tun->numqueues);
389 }
390 
391 static void tun_disable_queue(struct tun_struct *tun, struct tun_file *tfile)
392 {
393 	tfile->detached = tun;
394 	list_add_tail(&tfile->next, &tun->disabled);
395 	++tun->numdisabled;
396 }
397 
398 static struct tun_struct *tun_enable_queue(struct tun_file *tfile)
399 {
400 	struct tun_struct *tun = tfile->detached;
401 
402 	tfile->detached = NULL;
403 	list_del_init(&tfile->next);
404 	--tun->numdisabled;
405 	return tun;
406 }
407 
408 static void __tun_detach(struct tun_file *tfile, bool clean)
409 {
410 	struct tun_file *ntfile;
411 	struct tun_struct *tun;
412 
413 	tun = rtnl_dereference(tfile->tun);
414 
415 	if (tun && !tfile->detached) {
416 		u16 index = tfile->queue_index;
417 		BUG_ON(index >= tun->numqueues);
418 
419 		rcu_assign_pointer(tun->tfiles[index],
420 				   tun->tfiles[tun->numqueues - 1]);
421 		ntfile = rtnl_dereference(tun->tfiles[index]);
422 		ntfile->queue_index = index;
423 
424 		--tun->numqueues;
425 		if (clean) {
426 			rcu_assign_pointer(tfile->tun, NULL);
427 			sock_put(&tfile->sk);
428 		} else
429 			tun_disable_queue(tun, tfile);
430 
431 		synchronize_net();
432 		tun_flow_delete_by_queue(tun, tun->numqueues + 1);
433 		/* Drop read queue */
434 		skb_queue_purge(&tfile->sk.sk_receive_queue);
435 		tun_set_real_num_queues(tun);
436 	} else if (tfile->detached && clean) {
437 		tun = tun_enable_queue(tfile);
438 		sock_put(&tfile->sk);
439 	}
440 
441 	if (clean) {
442 		if (tun && tun->numqueues == 0 && tun->numdisabled == 0) {
443 			netif_carrier_off(tun->dev);
444 
445 			if (!(tun->flags & TUN_PERSIST) &&
446 			    tun->dev->reg_state == NETREG_REGISTERED)
447 				unregister_netdevice(tun->dev);
448 		}
449 
450 		BUG_ON(!test_bit(SOCK_EXTERNALLY_ALLOCATED,
451 				 &tfile->socket.flags));
452 		sk_release_kernel(&tfile->sk);
453 	}
454 }
455 
456 static void tun_detach(struct tun_file *tfile, bool clean)
457 {
458 	rtnl_lock();
459 	__tun_detach(tfile, clean);
460 	rtnl_unlock();
461 }
462 
463 static void tun_detach_all(struct net_device *dev)
464 {
465 	struct tun_struct *tun = netdev_priv(dev);
466 	struct tun_file *tfile, *tmp;
467 	int i, n = tun->numqueues;
468 
469 	for (i = 0; i < n; i++) {
470 		tfile = rtnl_dereference(tun->tfiles[i]);
471 		BUG_ON(!tfile);
472 		wake_up_all(&tfile->wq.wait);
473 		rcu_assign_pointer(tfile->tun, NULL);
474 		--tun->numqueues;
475 	}
476 	list_for_each_entry(tfile, &tun->disabled, next) {
477 		wake_up_all(&tfile->wq.wait);
478 		rcu_assign_pointer(tfile->tun, NULL);
479 	}
480 	BUG_ON(tun->numqueues != 0);
481 
482 	synchronize_net();
483 	for (i = 0; i < n; i++) {
484 		tfile = rtnl_dereference(tun->tfiles[i]);
485 		/* Drop read queue */
486 		skb_queue_purge(&tfile->sk.sk_receive_queue);
487 		sock_put(&tfile->sk);
488 	}
489 	list_for_each_entry_safe(tfile, tmp, &tun->disabled, next) {
490 		tun_enable_queue(tfile);
491 		skb_queue_purge(&tfile->sk.sk_receive_queue);
492 		sock_put(&tfile->sk);
493 	}
494 	BUG_ON(tun->numdisabled != 0);
495 
496 	if (tun->flags & TUN_PERSIST)
497 		module_put(THIS_MODULE);
498 }
499 
500 static int tun_attach(struct tun_struct *tun, struct file *file)
501 {
502 	struct tun_file *tfile = file->private_data;
503 	int err;
504 
505 	err = security_tun_dev_attach(tfile->socket.sk, tun->security);
506 	if (err < 0)
507 		goto out;
508 
509 	err = -EINVAL;
510 	if (rtnl_dereference(tfile->tun) && !tfile->detached)
511 		goto out;
512 
513 	err = -EBUSY;
514 	if (!(tun->flags & TUN_TAP_MQ) && tun->numqueues == 1)
515 		goto out;
516 
517 	err = -E2BIG;
518 	if (!tfile->detached &&
519 	    tun->numqueues + tun->numdisabled == MAX_TAP_QUEUES)
520 		goto out;
521 
522 	err = 0;
523 
524 	/* Re-attach the filter to presist device */
525 	if (tun->filter_attached == true) {
526 		err = sk_attach_filter(&tun->fprog, tfile->socket.sk);
527 		if (!err)
528 			goto out;
529 	}
530 	tfile->queue_index = tun->numqueues;
531 	rcu_assign_pointer(tfile->tun, tun);
532 	rcu_assign_pointer(tun->tfiles[tun->numqueues], tfile);
533 	tun->numqueues++;
534 
535 	if (tfile->detached)
536 		tun_enable_queue(tfile);
537 	else
538 		sock_hold(&tfile->sk);
539 
540 	tun_set_real_num_queues(tun);
541 
542 	/* device is allowed to go away first, so no need to hold extra
543 	 * refcnt.
544 	 */
545 
546 out:
547 	return err;
548 }
549 
550 static struct tun_struct *__tun_get(struct tun_file *tfile)
551 {
552 	struct tun_struct *tun;
553 
554 	rcu_read_lock();
555 	tun = rcu_dereference(tfile->tun);
556 	if (tun)
557 		dev_hold(tun->dev);
558 	rcu_read_unlock();
559 
560 	return tun;
561 }
562 
563 static struct tun_struct *tun_get(struct file *file)
564 {
565 	return __tun_get(file->private_data);
566 }
567 
568 static void tun_put(struct tun_struct *tun)
569 {
570 	dev_put(tun->dev);
571 }
572 
573 /* TAP filtering */
574 static void addr_hash_set(u32 *mask, const u8 *addr)
575 {
576 	int n = ether_crc(ETH_ALEN, addr) >> 26;
577 	mask[n >> 5] |= (1 << (n & 31));
578 }
579 
580 static unsigned int addr_hash_test(const u32 *mask, const u8 *addr)
581 {
582 	int n = ether_crc(ETH_ALEN, addr) >> 26;
583 	return mask[n >> 5] & (1 << (n & 31));
584 }
585 
586 static int update_filter(struct tap_filter *filter, void __user *arg)
587 {
588 	struct { u8 u[ETH_ALEN]; } *addr;
589 	struct tun_filter uf;
590 	int err, alen, n, nexact;
591 
592 	if (copy_from_user(&uf, arg, sizeof(uf)))
593 		return -EFAULT;
594 
595 	if (!uf.count) {
596 		/* Disabled */
597 		filter->count = 0;
598 		return 0;
599 	}
600 
601 	alen = ETH_ALEN * uf.count;
602 	addr = kmalloc(alen, GFP_KERNEL);
603 	if (!addr)
604 		return -ENOMEM;
605 
606 	if (copy_from_user(addr, arg + sizeof(uf), alen)) {
607 		err = -EFAULT;
608 		goto done;
609 	}
610 
611 	/* The filter is updated without holding any locks. Which is
612 	 * perfectly safe. We disable it first and in the worst
613 	 * case we'll accept a few undesired packets. */
614 	filter->count = 0;
615 	wmb();
616 
617 	/* Use first set of addresses as an exact filter */
618 	for (n = 0; n < uf.count && n < FLT_EXACT_COUNT; n++)
619 		memcpy(filter->addr[n], addr[n].u, ETH_ALEN);
620 
621 	nexact = n;
622 
623 	/* Remaining multicast addresses are hashed,
624 	 * unicast will leave the filter disabled. */
625 	memset(filter->mask, 0, sizeof(filter->mask));
626 	for (; n < uf.count; n++) {
627 		if (!is_multicast_ether_addr(addr[n].u)) {
628 			err = 0; /* no filter */
629 			goto done;
630 		}
631 		addr_hash_set(filter->mask, addr[n].u);
632 	}
633 
634 	/* For ALLMULTI just set the mask to all ones.
635 	 * This overrides the mask populated above. */
636 	if ((uf.flags & TUN_FLT_ALLMULTI))
637 		memset(filter->mask, ~0, sizeof(filter->mask));
638 
639 	/* Now enable the filter */
640 	wmb();
641 	filter->count = nexact;
642 
643 	/* Return the number of exact filters */
644 	err = nexact;
645 
646 done:
647 	kfree(addr);
648 	return err;
649 }
650 
651 /* Returns: 0 - drop, !=0 - accept */
652 static int run_filter(struct tap_filter *filter, const struct sk_buff *skb)
653 {
654 	/* Cannot use eth_hdr(skb) here because skb_mac_hdr() is incorrect
655 	 * at this point. */
656 	struct ethhdr *eh = (struct ethhdr *) skb->data;
657 	int i;
658 
659 	/* Exact match */
660 	for (i = 0; i < filter->count; i++)
661 		if (ether_addr_equal(eh->h_dest, filter->addr[i]))
662 			return 1;
663 
664 	/* Inexact match (multicast only) */
665 	if (is_multicast_ether_addr(eh->h_dest))
666 		return addr_hash_test(filter->mask, eh->h_dest);
667 
668 	return 0;
669 }
670 
671 /*
672  * Checks whether the packet is accepted or not.
673  * Returns: 0 - drop, !=0 - accept
674  */
675 static int check_filter(struct tap_filter *filter, const struct sk_buff *skb)
676 {
677 	if (!filter->count)
678 		return 1;
679 
680 	return run_filter(filter, skb);
681 }
682 
683 /* Network device part of the driver */
684 
685 static const struct ethtool_ops tun_ethtool_ops;
686 
687 /* Net device detach from fd. */
688 static void tun_net_uninit(struct net_device *dev)
689 {
690 	tun_detach_all(dev);
691 }
692 
693 /* Net device open. */
694 static int tun_net_open(struct net_device *dev)
695 {
696 	netif_tx_start_all_queues(dev);
697 	return 0;
698 }
699 
700 /* Net device close. */
701 static int tun_net_close(struct net_device *dev)
702 {
703 	netif_tx_stop_all_queues(dev);
704 	return 0;
705 }
706 
707 /* Net device start xmit */
708 static netdev_tx_t tun_net_xmit(struct sk_buff *skb, struct net_device *dev)
709 {
710 	struct tun_struct *tun = netdev_priv(dev);
711 	int txq = skb->queue_mapping;
712 	struct tun_file *tfile;
713 
714 	rcu_read_lock();
715 	tfile = rcu_dereference(tun->tfiles[txq]);
716 
717 	/* Drop packet if interface is not attached */
718 	if (txq >= tun->numqueues)
719 		goto drop;
720 
721 	tun_debug(KERN_INFO, tun, "tun_net_xmit %d\n", skb->len);
722 
723 	BUG_ON(!tfile);
724 
725 	/* Drop if the filter does not like it.
726 	 * This is a noop if the filter is disabled.
727 	 * Filter can be enabled only for the TAP devices. */
728 	if (!check_filter(&tun->txflt, skb))
729 		goto drop;
730 
731 	if (tfile->socket.sk->sk_filter &&
732 	    sk_filter(tfile->socket.sk, skb))
733 		goto drop;
734 
735 	/* Limit the number of packets queued by dividing txq length with the
736 	 * number of queues.
737 	 */
738 	if (skb_queue_len(&tfile->socket.sk->sk_receive_queue)
739 			  >= dev->tx_queue_len / tun->numqueues)
740 		goto drop;
741 
742 	/* Orphan the skb - required as we might hang on to it
743 	 * for indefinite time. */
744 	if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC)))
745 		goto drop;
746 	skb_orphan(skb);
747 
748 	nf_reset(skb);
749 
750 	/* Enqueue packet */
751 	skb_queue_tail(&tfile->socket.sk->sk_receive_queue, skb);
752 
753 	/* Notify and wake up reader process */
754 	if (tfile->flags & TUN_FASYNC)
755 		kill_fasync(&tfile->fasync, SIGIO, POLL_IN);
756 	wake_up_interruptible_poll(&tfile->wq.wait, POLLIN |
757 				   POLLRDNORM | POLLRDBAND);
758 
759 	rcu_read_unlock();
760 	return NETDEV_TX_OK;
761 
762 drop:
763 	dev->stats.tx_dropped++;
764 	skb_tx_error(skb);
765 	kfree_skb(skb);
766 	rcu_read_unlock();
767 	return NETDEV_TX_OK;
768 }
769 
770 static void tun_net_mclist(struct net_device *dev)
771 {
772 	/*
773 	 * This callback is supposed to deal with mc filter in
774 	 * _rx_ path and has nothing to do with the _tx_ path.
775 	 * In rx path we always accept everything userspace gives us.
776 	 */
777 }
778 
779 #define MIN_MTU 68
780 #define MAX_MTU 65535
781 
782 static int
783 tun_net_change_mtu(struct net_device *dev, int new_mtu)
784 {
785 	if (new_mtu < MIN_MTU || new_mtu + dev->hard_header_len > MAX_MTU)
786 		return -EINVAL;
787 	dev->mtu = new_mtu;
788 	return 0;
789 }
790 
791 static netdev_features_t tun_net_fix_features(struct net_device *dev,
792 	netdev_features_t features)
793 {
794 	struct tun_struct *tun = netdev_priv(dev);
795 
796 	return (features & tun->set_features) | (features & ~TUN_USER_FEATURES);
797 }
798 #ifdef CONFIG_NET_POLL_CONTROLLER
799 static void tun_poll_controller(struct net_device *dev)
800 {
801 	/*
802 	 * Tun only receives frames when:
803 	 * 1) the char device endpoint gets data from user space
804 	 * 2) the tun socket gets a sendmsg call from user space
805 	 * Since both of those are syncronous operations, we are guaranteed
806 	 * never to have pending data when we poll for it
807 	 * so theres nothing to do here but return.
808 	 * We need this though so netpoll recognizes us as an interface that
809 	 * supports polling, which enables bridge devices in virt setups to
810 	 * still use netconsole
811 	 */
812 	return;
813 }
814 #endif
815 static const struct net_device_ops tun_netdev_ops = {
816 	.ndo_uninit		= tun_net_uninit,
817 	.ndo_open		= tun_net_open,
818 	.ndo_stop		= tun_net_close,
819 	.ndo_start_xmit		= tun_net_xmit,
820 	.ndo_change_mtu		= tun_net_change_mtu,
821 	.ndo_fix_features	= tun_net_fix_features,
822 	.ndo_select_queue	= tun_select_queue,
823 #ifdef CONFIG_NET_POLL_CONTROLLER
824 	.ndo_poll_controller	= tun_poll_controller,
825 #endif
826 };
827 
828 static const struct net_device_ops tap_netdev_ops = {
829 	.ndo_uninit		= tun_net_uninit,
830 	.ndo_open		= tun_net_open,
831 	.ndo_stop		= tun_net_close,
832 	.ndo_start_xmit		= tun_net_xmit,
833 	.ndo_change_mtu		= tun_net_change_mtu,
834 	.ndo_fix_features	= tun_net_fix_features,
835 	.ndo_set_rx_mode	= tun_net_mclist,
836 	.ndo_set_mac_address	= eth_mac_addr,
837 	.ndo_validate_addr	= eth_validate_addr,
838 	.ndo_select_queue	= tun_select_queue,
839 #ifdef CONFIG_NET_POLL_CONTROLLER
840 	.ndo_poll_controller	= tun_poll_controller,
841 #endif
842 };
843 
844 static int tun_flow_init(struct tun_struct *tun)
845 {
846 	int i;
847 
848 	for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++)
849 		INIT_HLIST_HEAD(&tun->flows[i]);
850 
851 	tun->ageing_time = TUN_FLOW_EXPIRE;
852 	setup_timer(&tun->flow_gc_timer, tun_flow_cleanup, (unsigned long)tun);
853 	mod_timer(&tun->flow_gc_timer,
854 		  round_jiffies_up(jiffies + tun->ageing_time));
855 
856 	return 0;
857 }
858 
859 static void tun_flow_uninit(struct tun_struct *tun)
860 {
861 	del_timer_sync(&tun->flow_gc_timer);
862 	tun_flow_flush(tun);
863 }
864 
865 /* Initialize net device. */
866 static void tun_net_init(struct net_device *dev)
867 {
868 	struct tun_struct *tun = netdev_priv(dev);
869 
870 	switch (tun->flags & TUN_TYPE_MASK) {
871 	case TUN_TUN_DEV:
872 		dev->netdev_ops = &tun_netdev_ops;
873 
874 		/* Point-to-Point TUN Device */
875 		dev->hard_header_len = 0;
876 		dev->addr_len = 0;
877 		dev->mtu = 1500;
878 
879 		/* Zero header length */
880 		dev->type = ARPHRD_NONE;
881 		dev->flags = IFF_POINTOPOINT | IFF_NOARP | IFF_MULTICAST;
882 		dev->tx_queue_len = TUN_READQ_SIZE;  /* We prefer our own queue length */
883 		break;
884 
885 	case TUN_TAP_DEV:
886 		dev->netdev_ops = &tap_netdev_ops;
887 		/* Ethernet TAP Device */
888 		ether_setup(dev);
889 		dev->priv_flags &= ~IFF_TX_SKB_SHARING;
890 		dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
891 
892 		eth_hw_addr_random(dev);
893 
894 		dev->tx_queue_len = TUN_READQ_SIZE;  /* We prefer our own queue length */
895 		break;
896 	}
897 }
898 
899 /* Character device part */
900 
901 /* Poll */
902 static unsigned int tun_chr_poll(struct file *file, poll_table *wait)
903 {
904 	struct tun_file *tfile = file->private_data;
905 	struct tun_struct *tun = __tun_get(tfile);
906 	struct sock *sk;
907 	unsigned int mask = 0;
908 
909 	if (!tun)
910 		return POLLERR;
911 
912 	sk = tfile->socket.sk;
913 
914 	tun_debug(KERN_INFO, tun, "tun_chr_poll\n");
915 
916 	poll_wait(file, &tfile->wq.wait, wait);
917 
918 	if (!skb_queue_empty(&sk->sk_receive_queue))
919 		mask |= POLLIN | POLLRDNORM;
920 
921 	if (sock_writeable(sk) ||
922 	    (!test_and_set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags) &&
923 	     sock_writeable(sk)))
924 		mask |= POLLOUT | POLLWRNORM;
925 
926 	if (tun->dev->reg_state != NETREG_REGISTERED)
927 		mask = POLLERR;
928 
929 	tun_put(tun);
930 	return mask;
931 }
932 
933 /* prepad is the amount to reserve at front.  len is length after that.
934  * linear is a hint as to how much to copy (usually headers). */
935 static struct sk_buff *tun_alloc_skb(struct tun_file *tfile,
936 				     size_t prepad, size_t len,
937 				     size_t linear, int noblock)
938 {
939 	struct sock *sk = tfile->socket.sk;
940 	struct sk_buff *skb;
941 	int err;
942 
943 	/* Under a page?  Don't bother with paged skb. */
944 	if (prepad + len < PAGE_SIZE || !linear)
945 		linear = len;
946 
947 	skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
948 				   &err);
949 	if (!skb)
950 		return ERR_PTR(err);
951 
952 	skb_reserve(skb, prepad);
953 	skb_put(skb, linear);
954 	skb->data_len = len - linear;
955 	skb->len += len - linear;
956 
957 	return skb;
958 }
959 
960 /* set skb frags from iovec, this can move to core network code for reuse */
961 static int zerocopy_sg_from_iovec(struct sk_buff *skb, const struct iovec *from,
962 				  int offset, size_t count)
963 {
964 	int len = iov_length(from, count) - offset;
965 	int copy = skb_headlen(skb);
966 	int size, offset1 = 0;
967 	int i = 0;
968 
969 	/* Skip over from offset */
970 	while (count && (offset >= from->iov_len)) {
971 		offset -= from->iov_len;
972 		++from;
973 		--count;
974 	}
975 
976 	/* copy up to skb headlen */
977 	while (count && (copy > 0)) {
978 		size = min_t(unsigned int, copy, from->iov_len - offset);
979 		if (copy_from_user(skb->data + offset1, from->iov_base + offset,
980 				   size))
981 			return -EFAULT;
982 		if (copy > size) {
983 			++from;
984 			--count;
985 			offset = 0;
986 		} else
987 			offset += size;
988 		copy -= size;
989 		offset1 += size;
990 	}
991 
992 	if (len == offset1)
993 		return 0;
994 
995 	while (count--) {
996 		struct page *page[MAX_SKB_FRAGS];
997 		int num_pages;
998 		unsigned long base;
999 		unsigned long truesize;
1000 
1001 		len = from->iov_len - offset;
1002 		if (!len) {
1003 			offset = 0;
1004 			++from;
1005 			continue;
1006 		}
1007 		base = (unsigned long)from->iov_base + offset;
1008 		size = ((base & ~PAGE_MASK) + len + ~PAGE_MASK) >> PAGE_SHIFT;
1009 		if (i + size > MAX_SKB_FRAGS)
1010 			return -EMSGSIZE;
1011 		num_pages = get_user_pages_fast(base, size, 0, &page[i]);
1012 		if (num_pages != size) {
1013 			int j;
1014 
1015 			for (j = 0; j < num_pages; j++)
1016 				put_page(page[i + j]);
1017 			return -EFAULT;
1018 		}
1019 		truesize = size * PAGE_SIZE;
1020 		skb->data_len += len;
1021 		skb->len += len;
1022 		skb->truesize += truesize;
1023 		atomic_add(truesize, &skb->sk->sk_wmem_alloc);
1024 		while (len) {
1025 			int off = base & ~PAGE_MASK;
1026 			int size = min_t(int, len, PAGE_SIZE - off);
1027 			__skb_fill_page_desc(skb, i, page[i], off, size);
1028 			skb_shinfo(skb)->nr_frags++;
1029 			/* increase sk_wmem_alloc */
1030 			base += size;
1031 			len -= size;
1032 			i++;
1033 		}
1034 		offset = 0;
1035 		++from;
1036 	}
1037 	return 0;
1038 }
1039 
1040 /* Get packet from user space buffer */
1041 static ssize_t tun_get_user(struct tun_struct *tun, struct tun_file *tfile,
1042 			    void *msg_control, const struct iovec *iv,
1043 			    size_t total_len, size_t count, int noblock)
1044 {
1045 	struct tun_pi pi = { 0, cpu_to_be16(ETH_P_IP) };
1046 	struct sk_buff *skb;
1047 	size_t len = total_len, align = NET_SKB_PAD;
1048 	struct virtio_net_hdr gso = { 0 };
1049 	int offset = 0;
1050 	int copylen;
1051 	bool zerocopy = false;
1052 	int err;
1053 	u32 rxhash;
1054 
1055 	if (!(tun->flags & TUN_NO_PI)) {
1056 		if ((len -= sizeof(pi)) > total_len)
1057 			return -EINVAL;
1058 
1059 		if (memcpy_fromiovecend((void *)&pi, iv, 0, sizeof(pi)))
1060 			return -EFAULT;
1061 		offset += sizeof(pi);
1062 	}
1063 
1064 	if (tun->flags & TUN_VNET_HDR) {
1065 		if ((len -= tun->vnet_hdr_sz) > total_len)
1066 			return -EINVAL;
1067 
1068 		if (memcpy_fromiovecend((void *)&gso, iv, offset, sizeof(gso)))
1069 			return -EFAULT;
1070 
1071 		if ((gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
1072 		    gso.csum_start + gso.csum_offset + 2 > gso.hdr_len)
1073 			gso.hdr_len = gso.csum_start + gso.csum_offset + 2;
1074 
1075 		if (gso.hdr_len > len)
1076 			return -EINVAL;
1077 		offset += tun->vnet_hdr_sz;
1078 	}
1079 
1080 	if ((tun->flags & TUN_TYPE_MASK) == TUN_TAP_DEV) {
1081 		align += NET_IP_ALIGN;
1082 		if (unlikely(len < ETH_HLEN ||
1083 			     (gso.hdr_len && gso.hdr_len < ETH_HLEN)))
1084 			return -EINVAL;
1085 	}
1086 
1087 	if (msg_control)
1088 		zerocopy = true;
1089 
1090 	if (zerocopy) {
1091 		/* Userspace may produce vectors with count greater than
1092 		 * MAX_SKB_FRAGS, so we need to linearize parts of the skb
1093 		 * to let the rest of data to be fit in the frags.
1094 		 */
1095 		if (count > MAX_SKB_FRAGS) {
1096 			copylen = iov_length(iv, count - MAX_SKB_FRAGS);
1097 			if (copylen < offset)
1098 				copylen = 0;
1099 			else
1100 				copylen -= offset;
1101 		} else
1102 				copylen = 0;
1103 		/* There are 256 bytes to be copied in skb, so there is enough
1104 		 * room for skb expand head in case it is used.
1105 		 * The rest of the buffer is mapped from userspace.
1106 		 */
1107 		if (copylen < gso.hdr_len)
1108 			copylen = gso.hdr_len;
1109 		if (!copylen)
1110 			copylen = GOODCOPY_LEN;
1111 	} else
1112 		copylen = len;
1113 
1114 	skb = tun_alloc_skb(tfile, align, copylen, gso.hdr_len, noblock);
1115 	if (IS_ERR(skb)) {
1116 		if (PTR_ERR(skb) != -EAGAIN)
1117 			tun->dev->stats.rx_dropped++;
1118 		return PTR_ERR(skb);
1119 	}
1120 
1121 	if (zerocopy)
1122 		err = zerocopy_sg_from_iovec(skb, iv, offset, count);
1123 	else
1124 		err = skb_copy_datagram_from_iovec(skb, 0, iv, offset, len);
1125 
1126 	if (err) {
1127 		tun->dev->stats.rx_dropped++;
1128 		kfree_skb(skb);
1129 		return -EFAULT;
1130 	}
1131 
1132 	if (gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
1133 		if (!skb_partial_csum_set(skb, gso.csum_start,
1134 					  gso.csum_offset)) {
1135 			tun->dev->stats.rx_frame_errors++;
1136 			kfree_skb(skb);
1137 			return -EINVAL;
1138 		}
1139 	}
1140 
1141 	switch (tun->flags & TUN_TYPE_MASK) {
1142 	case TUN_TUN_DEV:
1143 		if (tun->flags & TUN_NO_PI) {
1144 			switch (skb->data[0] & 0xf0) {
1145 			case 0x40:
1146 				pi.proto = htons(ETH_P_IP);
1147 				break;
1148 			case 0x60:
1149 				pi.proto = htons(ETH_P_IPV6);
1150 				break;
1151 			default:
1152 				tun->dev->stats.rx_dropped++;
1153 				kfree_skb(skb);
1154 				return -EINVAL;
1155 			}
1156 		}
1157 
1158 		skb_reset_mac_header(skb);
1159 		skb->protocol = pi.proto;
1160 		skb->dev = tun->dev;
1161 		break;
1162 	case TUN_TAP_DEV:
1163 		skb->protocol = eth_type_trans(skb, tun->dev);
1164 		break;
1165 	}
1166 
1167 	if (gso.gso_type != VIRTIO_NET_HDR_GSO_NONE) {
1168 		pr_debug("GSO!\n");
1169 		switch (gso.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
1170 		case VIRTIO_NET_HDR_GSO_TCPV4:
1171 			skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
1172 			break;
1173 		case VIRTIO_NET_HDR_GSO_TCPV6:
1174 			skb_shinfo(skb)->gso_type = SKB_GSO_TCPV6;
1175 			break;
1176 		case VIRTIO_NET_HDR_GSO_UDP:
1177 			skb_shinfo(skb)->gso_type = SKB_GSO_UDP;
1178 			break;
1179 		default:
1180 			tun->dev->stats.rx_frame_errors++;
1181 			kfree_skb(skb);
1182 			return -EINVAL;
1183 		}
1184 
1185 		if (gso.gso_type & VIRTIO_NET_HDR_GSO_ECN)
1186 			skb_shinfo(skb)->gso_type |= SKB_GSO_TCP_ECN;
1187 
1188 		skb_shinfo(skb)->gso_size = gso.gso_size;
1189 		if (skb_shinfo(skb)->gso_size == 0) {
1190 			tun->dev->stats.rx_frame_errors++;
1191 			kfree_skb(skb);
1192 			return -EINVAL;
1193 		}
1194 
1195 		/* Header must be checked, and gso_segs computed. */
1196 		skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
1197 		skb_shinfo(skb)->gso_segs = 0;
1198 	}
1199 
1200 	/* copy skb_ubuf_info for callback when skb has no error */
1201 	if (zerocopy) {
1202 		skb_shinfo(skb)->destructor_arg = msg_control;
1203 		skb_shinfo(skb)->tx_flags |= SKBTX_DEV_ZEROCOPY;
1204 		skb_shinfo(skb)->tx_flags |= SKBTX_SHARED_FRAG;
1205 	}
1206 
1207 	skb_reset_network_header(skb);
1208 	skb_probe_transport_header(skb, 0);
1209 
1210 	rxhash = skb_get_rxhash(skb);
1211 	netif_rx_ni(skb);
1212 
1213 	tun->dev->stats.rx_packets++;
1214 	tun->dev->stats.rx_bytes += len;
1215 
1216 	tun_flow_update(tun, rxhash, tfile);
1217 	return total_len;
1218 }
1219 
1220 static ssize_t tun_chr_aio_write(struct kiocb *iocb, const struct iovec *iv,
1221 			      unsigned long count, loff_t pos)
1222 {
1223 	struct file *file = iocb->ki_filp;
1224 	struct tun_struct *tun = tun_get(file);
1225 	struct tun_file *tfile = file->private_data;
1226 	ssize_t result;
1227 
1228 	if (!tun)
1229 		return -EBADFD;
1230 
1231 	tun_debug(KERN_INFO, tun, "tun_chr_write %ld\n", count);
1232 
1233 	result = tun_get_user(tun, tfile, NULL, iv, iov_length(iv, count),
1234 			      count, 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 			    const struct iovec *iv, int len)
1245 {
1246 	struct tun_pi pi = { 0, skb->protocol };
1247 	ssize_t total = 0;
1248 
1249 	if (!(tun->flags & TUN_NO_PI)) {
1250 		if ((len -= sizeof(pi)) < 0)
1251 			return -EINVAL;
1252 
1253 		if (len < skb->len) {
1254 			/* Packet will be striped */
1255 			pi.flags |= TUN_PKT_STRIP;
1256 		}
1257 
1258 		if (memcpy_toiovecend(iv, (void *) &pi, 0, sizeof(pi)))
1259 			return -EFAULT;
1260 		total += sizeof(pi);
1261 	}
1262 
1263 	if (tun->flags & TUN_VNET_HDR) {
1264 		struct virtio_net_hdr gso = { 0 }; /* no info leak */
1265 		if ((len -= tun->vnet_hdr_sz) < 0)
1266 			return -EINVAL;
1267 
1268 		if (skb_is_gso(skb)) {
1269 			struct skb_shared_info *sinfo = skb_shinfo(skb);
1270 
1271 			/* This is a hint as to how much should be linear. */
1272 			gso.hdr_len = skb_headlen(skb);
1273 			gso.gso_size = sinfo->gso_size;
1274 			if (sinfo->gso_type & SKB_GSO_TCPV4)
1275 				gso.gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
1276 			else if (sinfo->gso_type & SKB_GSO_TCPV6)
1277 				gso.gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
1278 			else if (sinfo->gso_type & SKB_GSO_UDP)
1279 				gso.gso_type = VIRTIO_NET_HDR_GSO_UDP;
1280 			else {
1281 				pr_err("unexpected GSO type: "
1282 				       "0x%x, gso_size %d, hdr_len %d\n",
1283 				       sinfo->gso_type, gso.gso_size,
1284 				       gso.hdr_len);
1285 				print_hex_dump(KERN_ERR, "tun: ",
1286 					       DUMP_PREFIX_NONE,
1287 					       16, 1, skb->head,
1288 					       min((int)gso.hdr_len, 64), true);
1289 				WARN_ON_ONCE(1);
1290 				return -EINVAL;
1291 			}
1292 			if (sinfo->gso_type & SKB_GSO_TCP_ECN)
1293 				gso.gso_type |= VIRTIO_NET_HDR_GSO_ECN;
1294 		} else
1295 			gso.gso_type = VIRTIO_NET_HDR_GSO_NONE;
1296 
1297 		if (skb->ip_summed == CHECKSUM_PARTIAL) {
1298 			gso.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
1299 			gso.csum_start = skb_checksum_start_offset(skb);
1300 			gso.csum_offset = skb->csum_offset;
1301 		} else if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
1302 			gso.flags = VIRTIO_NET_HDR_F_DATA_VALID;
1303 		} /* else everything is zero */
1304 
1305 		if (unlikely(memcpy_toiovecend(iv, (void *)&gso, total,
1306 					       sizeof(gso))))
1307 			return -EFAULT;
1308 		total += tun->vnet_hdr_sz;
1309 	}
1310 
1311 	len = min_t(int, skb->len, len);
1312 
1313 	skb_copy_datagram_const_iovec(skb, 0, iv, total, len);
1314 	total += skb->len;
1315 
1316 	tun->dev->stats.tx_packets++;
1317 	tun->dev->stats.tx_bytes += len;
1318 
1319 	return total;
1320 }
1321 
1322 static ssize_t tun_do_read(struct tun_struct *tun, struct tun_file *tfile,
1323 			   struct kiocb *iocb, const struct iovec *iv,
1324 			   ssize_t len, int noblock)
1325 {
1326 	DECLARE_WAITQUEUE(wait, current);
1327 	struct sk_buff *skb;
1328 	ssize_t ret = 0;
1329 
1330 	tun_debug(KERN_INFO, tun, "tun_do_read\n");
1331 
1332 	if (unlikely(!noblock))
1333 		add_wait_queue(&tfile->wq.wait, &wait);
1334 	while (len) {
1335 		current->state = TASK_INTERRUPTIBLE;
1336 
1337 		/* Read frames from the queue */
1338 		if (!(skb = skb_dequeue(&tfile->socket.sk->sk_receive_queue))) {
1339 			if (noblock) {
1340 				ret = -EAGAIN;
1341 				break;
1342 			}
1343 			if (signal_pending(current)) {
1344 				ret = -ERESTARTSYS;
1345 				break;
1346 			}
1347 			if (tun->dev->reg_state != NETREG_REGISTERED) {
1348 				ret = -EIO;
1349 				break;
1350 			}
1351 
1352 			/* Nothing to read, let's sleep */
1353 			schedule();
1354 			continue;
1355 		}
1356 
1357 		ret = tun_put_user(tun, tfile, skb, iv, len);
1358 		kfree_skb(skb);
1359 		break;
1360 	}
1361 
1362 	current->state = TASK_RUNNING;
1363 	if (unlikely(!noblock))
1364 		remove_wait_queue(&tfile->wq.wait, &wait);
1365 
1366 	return ret;
1367 }
1368 
1369 static ssize_t tun_chr_aio_read(struct kiocb *iocb, const struct iovec *iv,
1370 			    unsigned long count, loff_t pos)
1371 {
1372 	struct file *file = iocb->ki_filp;
1373 	struct tun_file *tfile = file->private_data;
1374 	struct tun_struct *tun = __tun_get(tfile);
1375 	ssize_t len, ret;
1376 
1377 	if (!tun)
1378 		return -EBADFD;
1379 	len = iov_length(iv, count);
1380 	if (len < 0) {
1381 		ret = -EINVAL;
1382 		goto out;
1383 	}
1384 
1385 	ret = tun_do_read(tun, tfile, iocb, iv, len,
1386 			  file->f_flags & O_NONBLOCK);
1387 	ret = min_t(ssize_t, ret, len);
1388 out:
1389 	tun_put(tun);
1390 	return ret;
1391 }
1392 
1393 static void tun_free_netdev(struct net_device *dev)
1394 {
1395 	struct tun_struct *tun = netdev_priv(dev);
1396 
1397 	BUG_ON(!(list_empty(&tun->disabled)));
1398 	tun_flow_uninit(tun);
1399 	security_tun_dev_free_security(tun->security);
1400 	free_netdev(dev);
1401 }
1402 
1403 static void tun_setup(struct net_device *dev)
1404 {
1405 	struct tun_struct *tun = netdev_priv(dev);
1406 
1407 	tun->owner = INVALID_UID;
1408 	tun->group = INVALID_GID;
1409 
1410 	dev->ethtool_ops = &tun_ethtool_ops;
1411 	dev->destructor = tun_free_netdev;
1412 }
1413 
1414 /* Trivial set of netlink ops to allow deleting tun or tap
1415  * device with netlink.
1416  */
1417 static int tun_validate(struct nlattr *tb[], struct nlattr *data[])
1418 {
1419 	return -EINVAL;
1420 }
1421 
1422 static struct rtnl_link_ops tun_link_ops __read_mostly = {
1423 	.kind		= DRV_NAME,
1424 	.priv_size	= sizeof(struct tun_struct),
1425 	.setup		= tun_setup,
1426 	.validate	= tun_validate,
1427 };
1428 
1429 static void tun_sock_write_space(struct sock *sk)
1430 {
1431 	struct tun_file *tfile;
1432 	wait_queue_head_t *wqueue;
1433 
1434 	if (!sock_writeable(sk))
1435 		return;
1436 
1437 	if (!test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags))
1438 		return;
1439 
1440 	wqueue = sk_sleep(sk);
1441 	if (wqueue && waitqueue_active(wqueue))
1442 		wake_up_interruptible_sync_poll(wqueue, POLLOUT |
1443 						POLLWRNORM | POLLWRBAND);
1444 
1445 	tfile = container_of(sk, struct tun_file, sk);
1446 	kill_fasync(&tfile->fasync, SIGIO, POLL_OUT);
1447 }
1448 
1449 static int tun_sendmsg(struct kiocb *iocb, struct socket *sock,
1450 		       struct msghdr *m, size_t total_len)
1451 {
1452 	int ret;
1453 	struct tun_file *tfile = container_of(sock, struct tun_file, socket);
1454 	struct tun_struct *tun = __tun_get(tfile);
1455 
1456 	if (!tun)
1457 		return -EBADFD;
1458 	ret = tun_get_user(tun, tfile, m->msg_control, m->msg_iov, total_len,
1459 			   m->msg_iovlen, m->msg_flags & MSG_DONTWAIT);
1460 	tun_put(tun);
1461 	return ret;
1462 }
1463 
1464 
1465 static int tun_recvmsg(struct kiocb *iocb, struct socket *sock,
1466 		       struct msghdr *m, size_t total_len,
1467 		       int flags)
1468 {
1469 	struct tun_file *tfile = container_of(sock, struct tun_file, socket);
1470 	struct tun_struct *tun = __tun_get(tfile);
1471 	int ret;
1472 
1473 	if (!tun)
1474 		return -EBADFD;
1475 
1476 	if (flags & ~(MSG_DONTWAIT|MSG_TRUNC)) {
1477 		ret = -EINVAL;
1478 		goto out;
1479 	}
1480 	ret = tun_do_read(tun, tfile, iocb, m->msg_iov, total_len,
1481 			  flags & MSG_DONTWAIT);
1482 	if (ret > total_len) {
1483 		m->msg_flags |= MSG_TRUNC;
1484 		ret = flags & MSG_TRUNC ? ret : total_len;
1485 	}
1486 out:
1487 	tun_put(tun);
1488 	return ret;
1489 }
1490 
1491 static int tun_release(struct socket *sock)
1492 {
1493 	if (sock->sk)
1494 		sock_put(sock->sk);
1495 	return 0;
1496 }
1497 
1498 /* Ops structure to mimic raw sockets with tun */
1499 static const struct proto_ops tun_socket_ops = {
1500 	.sendmsg = tun_sendmsg,
1501 	.recvmsg = tun_recvmsg,
1502 	.release = tun_release,
1503 };
1504 
1505 static struct proto tun_proto = {
1506 	.name		= "tun",
1507 	.owner		= THIS_MODULE,
1508 	.obj_size	= sizeof(struct tun_file),
1509 };
1510 
1511 static int tun_flags(struct tun_struct *tun)
1512 {
1513 	int flags = 0;
1514 
1515 	if (tun->flags & TUN_TUN_DEV)
1516 		flags |= IFF_TUN;
1517 	else
1518 		flags |= IFF_TAP;
1519 
1520 	if (tun->flags & TUN_NO_PI)
1521 		flags |= IFF_NO_PI;
1522 
1523 	/* This flag has no real effect.  We track the value for backwards
1524 	 * compatibility.
1525 	 */
1526 	if (tun->flags & TUN_ONE_QUEUE)
1527 		flags |= IFF_ONE_QUEUE;
1528 
1529 	if (tun->flags & TUN_VNET_HDR)
1530 		flags |= IFF_VNET_HDR;
1531 
1532 	if (tun->flags & TUN_TAP_MQ)
1533 		flags |= IFF_MULTI_QUEUE;
1534 
1535 	return flags;
1536 }
1537 
1538 static ssize_t tun_show_flags(struct device *dev, struct device_attribute *attr,
1539 			      char *buf)
1540 {
1541 	struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1542 	return sprintf(buf, "0x%x\n", tun_flags(tun));
1543 }
1544 
1545 static ssize_t tun_show_owner(struct device *dev, struct device_attribute *attr,
1546 			      char *buf)
1547 {
1548 	struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1549 	return uid_valid(tun->owner)?
1550 		sprintf(buf, "%u\n",
1551 			from_kuid_munged(current_user_ns(), tun->owner)):
1552 		sprintf(buf, "-1\n");
1553 }
1554 
1555 static ssize_t tun_show_group(struct device *dev, struct device_attribute *attr,
1556 			      char *buf)
1557 {
1558 	struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1559 	return gid_valid(tun->group) ?
1560 		sprintf(buf, "%u\n",
1561 			from_kgid_munged(current_user_ns(), tun->group)):
1562 		sprintf(buf, "-1\n");
1563 }
1564 
1565 static DEVICE_ATTR(tun_flags, 0444, tun_show_flags, NULL);
1566 static DEVICE_ATTR(owner, 0444, tun_show_owner, NULL);
1567 static DEVICE_ATTR(group, 0444, tun_show_group, NULL);
1568 
1569 static int tun_set_iff(struct net *net, struct file *file, struct ifreq *ifr)
1570 {
1571 	struct tun_struct *tun;
1572 	struct tun_file *tfile = file->private_data;
1573 	struct net_device *dev;
1574 	int err;
1575 
1576 	if (tfile->detached)
1577 		return -EINVAL;
1578 
1579 	dev = __dev_get_by_name(net, ifr->ifr_name);
1580 	if (dev) {
1581 		if (ifr->ifr_flags & IFF_TUN_EXCL)
1582 			return -EBUSY;
1583 		if ((ifr->ifr_flags & IFF_TUN) && dev->netdev_ops == &tun_netdev_ops)
1584 			tun = netdev_priv(dev);
1585 		else if ((ifr->ifr_flags & IFF_TAP) && dev->netdev_ops == &tap_netdev_ops)
1586 			tun = netdev_priv(dev);
1587 		else
1588 			return -EINVAL;
1589 
1590 		if (!!(ifr->ifr_flags & IFF_MULTI_QUEUE) !=
1591 		    !!(tun->flags & TUN_TAP_MQ))
1592 			return -EINVAL;
1593 
1594 		if (tun_not_capable(tun))
1595 			return -EPERM;
1596 		err = security_tun_dev_open(tun->security);
1597 		if (err < 0)
1598 			return err;
1599 
1600 		err = tun_attach(tun, file);
1601 		if (err < 0)
1602 			return err;
1603 
1604 		if (tun->flags & TUN_TAP_MQ &&
1605 		    (tun->numqueues + tun->numdisabled > 1)) {
1606 			/* One or more queue has already been attached, no need
1607 			 * to initialize the device again.
1608 			 */
1609 			return 0;
1610 		}
1611 	}
1612 	else {
1613 		char *name;
1614 		unsigned long flags = 0;
1615 		int queues = ifr->ifr_flags & IFF_MULTI_QUEUE ?
1616 			     MAX_TAP_QUEUES : 1;
1617 
1618 		if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
1619 			return -EPERM;
1620 		err = security_tun_dev_create();
1621 		if (err < 0)
1622 			return err;
1623 
1624 		/* Set dev type */
1625 		if (ifr->ifr_flags & IFF_TUN) {
1626 			/* TUN device */
1627 			flags |= TUN_TUN_DEV;
1628 			name = "tun%d";
1629 		} else if (ifr->ifr_flags & IFF_TAP) {
1630 			/* TAP device */
1631 			flags |= TUN_TAP_DEV;
1632 			name = "tap%d";
1633 		} else
1634 			return -EINVAL;
1635 
1636 		if (*ifr->ifr_name)
1637 			name = ifr->ifr_name;
1638 
1639 		dev = alloc_netdev_mqs(sizeof(struct tun_struct), name,
1640 				       tun_setup, queues, queues);
1641 
1642 		if (!dev)
1643 			return -ENOMEM;
1644 
1645 		dev_net_set(dev, net);
1646 		dev->rtnl_link_ops = &tun_link_ops;
1647 
1648 		tun = netdev_priv(dev);
1649 		tun->dev = dev;
1650 		tun->flags = flags;
1651 		tun->txflt.count = 0;
1652 		tun->vnet_hdr_sz = sizeof(struct virtio_net_hdr);
1653 
1654 		tun->filter_attached = false;
1655 		tun->sndbuf = tfile->socket.sk->sk_sndbuf;
1656 
1657 		spin_lock_init(&tun->lock);
1658 
1659 		err = security_tun_dev_alloc_security(&tun->security);
1660 		if (err < 0)
1661 			goto err_free_dev;
1662 
1663 		tun_net_init(dev);
1664 
1665 		err = tun_flow_init(tun);
1666 		if (err < 0)
1667 			goto err_free_dev;
1668 
1669 		dev->hw_features = NETIF_F_SG | NETIF_F_FRAGLIST |
1670 			TUN_USER_FEATURES;
1671 		dev->features = dev->hw_features;
1672 		dev->vlan_features = dev->features;
1673 
1674 		INIT_LIST_HEAD(&tun->disabled);
1675 		err = tun_attach(tun, file);
1676 		if (err < 0)
1677 			goto err_free_dev;
1678 
1679 		err = register_netdevice(tun->dev);
1680 		if (err < 0)
1681 			goto err_free_dev;
1682 
1683 		if (device_create_file(&tun->dev->dev, &dev_attr_tun_flags) ||
1684 		    device_create_file(&tun->dev->dev, &dev_attr_owner) ||
1685 		    device_create_file(&tun->dev->dev, &dev_attr_group))
1686 			pr_err("Failed to create tun sysfs files\n");
1687 	}
1688 
1689 	netif_carrier_on(tun->dev);
1690 
1691 	tun_debug(KERN_INFO, tun, "tun_set_iff\n");
1692 
1693 	if (ifr->ifr_flags & IFF_NO_PI)
1694 		tun->flags |= TUN_NO_PI;
1695 	else
1696 		tun->flags &= ~TUN_NO_PI;
1697 
1698 	/* This flag has no real effect.  We track the value for backwards
1699 	 * compatibility.
1700 	 */
1701 	if (ifr->ifr_flags & IFF_ONE_QUEUE)
1702 		tun->flags |= TUN_ONE_QUEUE;
1703 	else
1704 		tun->flags &= ~TUN_ONE_QUEUE;
1705 
1706 	if (ifr->ifr_flags & IFF_VNET_HDR)
1707 		tun->flags |= TUN_VNET_HDR;
1708 	else
1709 		tun->flags &= ~TUN_VNET_HDR;
1710 
1711 	if (ifr->ifr_flags & IFF_MULTI_QUEUE)
1712 		tun->flags |= TUN_TAP_MQ;
1713 	else
1714 		tun->flags &= ~TUN_TAP_MQ;
1715 
1716 	/* Make sure persistent devices do not get stuck in
1717 	 * xoff state.
1718 	 */
1719 	if (netif_running(tun->dev))
1720 		netif_tx_wake_all_queues(tun->dev);
1721 
1722 	strcpy(ifr->ifr_name, tun->dev->name);
1723 	return 0;
1724 
1725  err_free_dev:
1726 	free_netdev(dev);
1727 	return err;
1728 }
1729 
1730 static void tun_get_iff(struct net *net, struct tun_struct *tun,
1731 		       struct ifreq *ifr)
1732 {
1733 	tun_debug(KERN_INFO, tun, "tun_get_iff\n");
1734 
1735 	strcpy(ifr->ifr_name, tun->dev->name);
1736 
1737 	ifr->ifr_flags = tun_flags(tun);
1738 
1739 }
1740 
1741 /* This is like a cut-down ethtool ops, except done via tun fd so no
1742  * privs required. */
1743 static int set_offload(struct tun_struct *tun, unsigned long arg)
1744 {
1745 	netdev_features_t features = 0;
1746 
1747 	if (arg & TUN_F_CSUM) {
1748 		features |= NETIF_F_HW_CSUM;
1749 		arg &= ~TUN_F_CSUM;
1750 
1751 		if (arg & (TUN_F_TSO4|TUN_F_TSO6)) {
1752 			if (arg & TUN_F_TSO_ECN) {
1753 				features |= NETIF_F_TSO_ECN;
1754 				arg &= ~TUN_F_TSO_ECN;
1755 			}
1756 			if (arg & TUN_F_TSO4)
1757 				features |= NETIF_F_TSO;
1758 			if (arg & TUN_F_TSO6)
1759 				features |= NETIF_F_TSO6;
1760 			arg &= ~(TUN_F_TSO4|TUN_F_TSO6);
1761 		}
1762 
1763 		if (arg & TUN_F_UFO) {
1764 			features |= NETIF_F_UFO;
1765 			arg &= ~TUN_F_UFO;
1766 		}
1767 	}
1768 
1769 	/* This gives the user a way to test for new features in future by
1770 	 * trying to set them. */
1771 	if (arg)
1772 		return -EINVAL;
1773 
1774 	tun->set_features = features;
1775 	netdev_update_features(tun->dev);
1776 
1777 	return 0;
1778 }
1779 
1780 static void tun_detach_filter(struct tun_struct *tun, int n)
1781 {
1782 	int i;
1783 	struct tun_file *tfile;
1784 
1785 	for (i = 0; i < n; i++) {
1786 		tfile = rtnl_dereference(tun->tfiles[i]);
1787 		sk_detach_filter(tfile->socket.sk);
1788 	}
1789 
1790 	tun->filter_attached = false;
1791 }
1792 
1793 static int tun_attach_filter(struct tun_struct *tun)
1794 {
1795 	int i, ret = 0;
1796 	struct tun_file *tfile;
1797 
1798 	for (i = 0; i < tun->numqueues; i++) {
1799 		tfile = rtnl_dereference(tun->tfiles[i]);
1800 		ret = sk_attach_filter(&tun->fprog, tfile->socket.sk);
1801 		if (ret) {
1802 			tun_detach_filter(tun, i);
1803 			return ret;
1804 		}
1805 	}
1806 
1807 	tun->filter_attached = true;
1808 	return ret;
1809 }
1810 
1811 static void tun_set_sndbuf(struct tun_struct *tun)
1812 {
1813 	struct tun_file *tfile;
1814 	int i;
1815 
1816 	for (i = 0; i < tun->numqueues; i++) {
1817 		tfile = rtnl_dereference(tun->tfiles[i]);
1818 		tfile->socket.sk->sk_sndbuf = tun->sndbuf;
1819 	}
1820 }
1821 
1822 static int tun_set_queue(struct file *file, struct ifreq *ifr)
1823 {
1824 	struct tun_file *tfile = file->private_data;
1825 	struct tun_struct *tun;
1826 	int ret = 0;
1827 
1828 	rtnl_lock();
1829 
1830 	if (ifr->ifr_flags & IFF_ATTACH_QUEUE) {
1831 		tun = tfile->detached;
1832 		if (!tun) {
1833 			ret = -EINVAL;
1834 			goto unlock;
1835 		}
1836 		ret = security_tun_dev_attach_queue(tun->security);
1837 		if (ret < 0)
1838 			goto unlock;
1839 		ret = tun_attach(tun, file);
1840 	} else if (ifr->ifr_flags & IFF_DETACH_QUEUE) {
1841 		tun = rtnl_dereference(tfile->tun);
1842 		if (!tun || !(tun->flags & TUN_TAP_MQ) || tfile->detached)
1843 			ret = -EINVAL;
1844 		else
1845 			__tun_detach(tfile, false);
1846 	} else
1847 		ret = -EINVAL;
1848 
1849 unlock:
1850 	rtnl_unlock();
1851 	return ret;
1852 }
1853 
1854 static long __tun_chr_ioctl(struct file *file, unsigned int cmd,
1855 			    unsigned long arg, int ifreq_len)
1856 {
1857 	struct tun_file *tfile = file->private_data;
1858 	struct tun_struct *tun;
1859 	void __user* argp = (void __user*)arg;
1860 	struct ifreq ifr;
1861 	kuid_t owner;
1862 	kgid_t group;
1863 	int sndbuf;
1864 	int vnet_hdr_sz;
1865 	int ret;
1866 
1867 	if (cmd == TUNSETIFF || cmd == TUNSETQUEUE || _IOC_TYPE(cmd) == 0x89) {
1868 		if (copy_from_user(&ifr, argp, ifreq_len))
1869 			return -EFAULT;
1870 	} else {
1871 		memset(&ifr, 0, sizeof(ifr));
1872 	}
1873 	if (cmd == TUNGETFEATURES) {
1874 		/* Currently this just means: "what IFF flags are valid?".
1875 		 * This is needed because we never checked for invalid flags on
1876 		 * TUNSETIFF. */
1877 		return put_user(IFF_TUN | IFF_TAP | IFF_NO_PI | IFF_ONE_QUEUE |
1878 				IFF_VNET_HDR | IFF_MULTI_QUEUE,
1879 				(unsigned int __user*)argp);
1880 	} else if (cmd == TUNSETQUEUE)
1881 		return tun_set_queue(file, &ifr);
1882 
1883 	ret = 0;
1884 	rtnl_lock();
1885 
1886 	tun = __tun_get(tfile);
1887 	if (cmd == TUNSETIFF && !tun) {
1888 		ifr.ifr_name[IFNAMSIZ-1] = '\0';
1889 
1890 		ret = tun_set_iff(tfile->net, file, &ifr);
1891 
1892 		if (ret)
1893 			goto unlock;
1894 
1895 		if (copy_to_user(argp, &ifr, ifreq_len))
1896 			ret = -EFAULT;
1897 		goto unlock;
1898 	}
1899 
1900 	ret = -EBADFD;
1901 	if (!tun)
1902 		goto unlock;
1903 
1904 	tun_debug(KERN_INFO, tun, "tun_chr_ioctl cmd %u\n", cmd);
1905 
1906 	ret = 0;
1907 	switch (cmd) {
1908 	case TUNGETIFF:
1909 		tun_get_iff(current->nsproxy->net_ns, tun, &ifr);
1910 
1911 		if (copy_to_user(argp, &ifr, ifreq_len))
1912 			ret = -EFAULT;
1913 		break;
1914 
1915 	case TUNSETNOCSUM:
1916 		/* Disable/Enable checksum */
1917 
1918 		/* [unimplemented] */
1919 		tun_debug(KERN_INFO, tun, "ignored: set checksum %s\n",
1920 			  arg ? "disabled" : "enabled");
1921 		break;
1922 
1923 	case TUNSETPERSIST:
1924 		/* Disable/Enable persist mode. Keep an extra reference to the
1925 		 * module to prevent the module being unprobed.
1926 		 */
1927 		if (arg && !(tun->flags & TUN_PERSIST)) {
1928 			tun->flags |= TUN_PERSIST;
1929 			__module_get(THIS_MODULE);
1930 		}
1931 		if (!arg && (tun->flags & TUN_PERSIST)) {
1932 			tun->flags &= ~TUN_PERSIST;
1933 			module_put(THIS_MODULE);
1934 		}
1935 
1936 		tun_debug(KERN_INFO, tun, "persist %s\n",
1937 			  arg ? "enabled" : "disabled");
1938 		break;
1939 
1940 	case TUNSETOWNER:
1941 		/* Set owner of the device */
1942 		owner = make_kuid(current_user_ns(), arg);
1943 		if (!uid_valid(owner)) {
1944 			ret = -EINVAL;
1945 			break;
1946 		}
1947 		tun->owner = owner;
1948 		tun_debug(KERN_INFO, tun, "owner set to %u\n",
1949 			  from_kuid(&init_user_ns, tun->owner));
1950 		break;
1951 
1952 	case TUNSETGROUP:
1953 		/* Set group of the device */
1954 		group = make_kgid(current_user_ns(), arg);
1955 		if (!gid_valid(group)) {
1956 			ret = -EINVAL;
1957 			break;
1958 		}
1959 		tun->group = group;
1960 		tun_debug(KERN_INFO, tun, "group set to %u\n",
1961 			  from_kgid(&init_user_ns, tun->group));
1962 		break;
1963 
1964 	case TUNSETLINK:
1965 		/* Only allow setting the type when the interface is down */
1966 		if (tun->dev->flags & IFF_UP) {
1967 			tun_debug(KERN_INFO, tun,
1968 				  "Linktype set failed because interface is up\n");
1969 			ret = -EBUSY;
1970 		} else {
1971 			tun->dev->type = (int) arg;
1972 			tun_debug(KERN_INFO, tun, "linktype set to %d\n",
1973 				  tun->dev->type);
1974 			ret = 0;
1975 		}
1976 		break;
1977 
1978 #ifdef TUN_DEBUG
1979 	case TUNSETDEBUG:
1980 		tun->debug = arg;
1981 		break;
1982 #endif
1983 	case TUNSETOFFLOAD:
1984 		ret = set_offload(tun, arg);
1985 		break;
1986 
1987 	case TUNSETTXFILTER:
1988 		/* Can be set only for TAPs */
1989 		ret = -EINVAL;
1990 		if ((tun->flags & TUN_TYPE_MASK) != TUN_TAP_DEV)
1991 			break;
1992 		ret = update_filter(&tun->txflt, (void __user *)arg);
1993 		break;
1994 
1995 	case SIOCGIFHWADDR:
1996 		/* Get hw address */
1997 		memcpy(ifr.ifr_hwaddr.sa_data, tun->dev->dev_addr, ETH_ALEN);
1998 		ifr.ifr_hwaddr.sa_family = tun->dev->type;
1999 		if (copy_to_user(argp, &ifr, ifreq_len))
2000 			ret = -EFAULT;
2001 		break;
2002 
2003 	case SIOCSIFHWADDR:
2004 		/* Set hw address */
2005 		tun_debug(KERN_DEBUG, tun, "set hw address: %pM\n",
2006 			  ifr.ifr_hwaddr.sa_data);
2007 
2008 		ret = dev_set_mac_address(tun->dev, &ifr.ifr_hwaddr);
2009 		break;
2010 
2011 	case TUNGETSNDBUF:
2012 		sndbuf = tfile->socket.sk->sk_sndbuf;
2013 		if (copy_to_user(argp, &sndbuf, sizeof(sndbuf)))
2014 			ret = -EFAULT;
2015 		break;
2016 
2017 	case TUNSETSNDBUF:
2018 		if (copy_from_user(&sndbuf, argp, sizeof(sndbuf))) {
2019 			ret = -EFAULT;
2020 			break;
2021 		}
2022 
2023 		tun->sndbuf = sndbuf;
2024 		tun_set_sndbuf(tun);
2025 		break;
2026 
2027 	case TUNGETVNETHDRSZ:
2028 		vnet_hdr_sz = tun->vnet_hdr_sz;
2029 		if (copy_to_user(argp, &vnet_hdr_sz, sizeof(vnet_hdr_sz)))
2030 			ret = -EFAULT;
2031 		break;
2032 
2033 	case TUNSETVNETHDRSZ:
2034 		if (copy_from_user(&vnet_hdr_sz, argp, sizeof(vnet_hdr_sz))) {
2035 			ret = -EFAULT;
2036 			break;
2037 		}
2038 		if (vnet_hdr_sz < (int)sizeof(struct virtio_net_hdr)) {
2039 			ret = -EINVAL;
2040 			break;
2041 		}
2042 
2043 		tun->vnet_hdr_sz = vnet_hdr_sz;
2044 		break;
2045 
2046 	case TUNATTACHFILTER:
2047 		/* Can be set only for TAPs */
2048 		ret = -EINVAL;
2049 		if ((tun->flags & TUN_TYPE_MASK) != TUN_TAP_DEV)
2050 			break;
2051 		ret = -EFAULT;
2052 		if (copy_from_user(&tun->fprog, argp, sizeof(tun->fprog)))
2053 			break;
2054 
2055 		ret = tun_attach_filter(tun);
2056 		break;
2057 
2058 	case TUNDETACHFILTER:
2059 		/* Can be set only for TAPs */
2060 		ret = -EINVAL;
2061 		if ((tun->flags & TUN_TYPE_MASK) != TUN_TAP_DEV)
2062 			break;
2063 		ret = 0;
2064 		tun_detach_filter(tun, tun->numqueues);
2065 		break;
2066 
2067 	default:
2068 		ret = -EINVAL;
2069 		break;
2070 	}
2071 
2072 unlock:
2073 	rtnl_unlock();
2074 	if (tun)
2075 		tun_put(tun);
2076 	return ret;
2077 }
2078 
2079 static long tun_chr_ioctl(struct file *file,
2080 			  unsigned int cmd, unsigned long arg)
2081 {
2082 	return __tun_chr_ioctl(file, cmd, arg, sizeof (struct ifreq));
2083 }
2084 
2085 #ifdef CONFIG_COMPAT
2086 static long tun_chr_compat_ioctl(struct file *file,
2087 			 unsigned int cmd, unsigned long arg)
2088 {
2089 	switch (cmd) {
2090 	case TUNSETIFF:
2091 	case TUNGETIFF:
2092 	case TUNSETTXFILTER:
2093 	case TUNGETSNDBUF:
2094 	case TUNSETSNDBUF:
2095 	case SIOCGIFHWADDR:
2096 	case SIOCSIFHWADDR:
2097 		arg = (unsigned long)compat_ptr(arg);
2098 		break;
2099 	default:
2100 		arg = (compat_ulong_t)arg;
2101 		break;
2102 	}
2103 
2104 	/*
2105 	 * compat_ifreq is shorter than ifreq, so we must not access beyond
2106 	 * the end of that structure. All fields that are used in this
2107 	 * driver are compatible though, we don't need to convert the
2108 	 * contents.
2109 	 */
2110 	return __tun_chr_ioctl(file, cmd, arg, sizeof(struct compat_ifreq));
2111 }
2112 #endif /* CONFIG_COMPAT */
2113 
2114 static int tun_chr_fasync(int fd, struct file *file, int on)
2115 {
2116 	struct tun_file *tfile = file->private_data;
2117 	int ret;
2118 
2119 	if ((ret = fasync_helper(fd, file, on, &tfile->fasync)) < 0)
2120 		goto out;
2121 
2122 	if (on) {
2123 		ret = __f_setown(file, task_pid(current), PIDTYPE_PID, 0);
2124 		if (ret)
2125 			goto out;
2126 		tfile->flags |= TUN_FASYNC;
2127 	} else
2128 		tfile->flags &= ~TUN_FASYNC;
2129 	ret = 0;
2130 out:
2131 	return ret;
2132 }
2133 
2134 static int tun_chr_open(struct inode *inode, struct file * file)
2135 {
2136 	struct tun_file *tfile;
2137 
2138 	DBG1(KERN_INFO, "tunX: tun_chr_open\n");
2139 
2140 	tfile = (struct tun_file *)sk_alloc(&init_net, AF_UNSPEC, GFP_KERNEL,
2141 					    &tun_proto);
2142 	if (!tfile)
2143 		return -ENOMEM;
2144 	rcu_assign_pointer(tfile->tun, NULL);
2145 	tfile->net = get_net(current->nsproxy->net_ns);
2146 	tfile->flags = 0;
2147 
2148 	rcu_assign_pointer(tfile->socket.wq, &tfile->wq);
2149 	init_waitqueue_head(&tfile->wq.wait);
2150 
2151 	tfile->socket.file = file;
2152 	tfile->socket.ops = &tun_socket_ops;
2153 
2154 	sock_init_data(&tfile->socket, &tfile->sk);
2155 	sk_change_net(&tfile->sk, tfile->net);
2156 
2157 	tfile->sk.sk_write_space = tun_sock_write_space;
2158 	tfile->sk.sk_sndbuf = INT_MAX;
2159 
2160 	file->private_data = tfile;
2161 	set_bit(SOCK_EXTERNALLY_ALLOCATED, &tfile->socket.flags);
2162 	INIT_LIST_HEAD(&tfile->next);
2163 
2164 	sock_set_flag(&tfile->sk, SOCK_ZEROCOPY);
2165 
2166 	return 0;
2167 }
2168 
2169 static int tun_chr_close(struct inode *inode, struct file *file)
2170 {
2171 	struct tun_file *tfile = file->private_data;
2172 	struct net *net = tfile->net;
2173 
2174 	tun_detach(tfile, true);
2175 	put_net(net);
2176 
2177 	return 0;
2178 }
2179 
2180 static const struct file_operations tun_fops = {
2181 	.owner	= THIS_MODULE,
2182 	.llseek = no_llseek,
2183 	.read  = do_sync_read,
2184 	.aio_read  = tun_chr_aio_read,
2185 	.write = do_sync_write,
2186 	.aio_write = tun_chr_aio_write,
2187 	.poll	= tun_chr_poll,
2188 	.unlocked_ioctl	= tun_chr_ioctl,
2189 #ifdef CONFIG_COMPAT
2190 	.compat_ioctl = tun_chr_compat_ioctl,
2191 #endif
2192 	.open	= tun_chr_open,
2193 	.release = tun_chr_close,
2194 	.fasync = tun_chr_fasync
2195 };
2196 
2197 static struct miscdevice tun_miscdev = {
2198 	.minor = TUN_MINOR,
2199 	.name = "tun",
2200 	.nodename = "net/tun",
2201 	.fops = &tun_fops,
2202 };
2203 
2204 /* ethtool interface */
2205 
2206 static int tun_get_settings(struct net_device *dev, struct ethtool_cmd *cmd)
2207 {
2208 	cmd->supported		= 0;
2209 	cmd->advertising	= 0;
2210 	ethtool_cmd_speed_set(cmd, SPEED_10);
2211 	cmd->duplex		= DUPLEX_FULL;
2212 	cmd->port		= PORT_TP;
2213 	cmd->phy_address	= 0;
2214 	cmd->transceiver	= XCVR_INTERNAL;
2215 	cmd->autoneg		= AUTONEG_DISABLE;
2216 	cmd->maxtxpkt		= 0;
2217 	cmd->maxrxpkt		= 0;
2218 	return 0;
2219 }
2220 
2221 static void tun_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
2222 {
2223 	struct tun_struct *tun = netdev_priv(dev);
2224 
2225 	strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
2226 	strlcpy(info->version, DRV_VERSION, sizeof(info->version));
2227 
2228 	switch (tun->flags & TUN_TYPE_MASK) {
2229 	case TUN_TUN_DEV:
2230 		strlcpy(info->bus_info, "tun", sizeof(info->bus_info));
2231 		break;
2232 	case TUN_TAP_DEV:
2233 		strlcpy(info->bus_info, "tap", sizeof(info->bus_info));
2234 		break;
2235 	}
2236 }
2237 
2238 static u32 tun_get_msglevel(struct net_device *dev)
2239 {
2240 #ifdef TUN_DEBUG
2241 	struct tun_struct *tun = netdev_priv(dev);
2242 	return tun->debug;
2243 #else
2244 	return -EOPNOTSUPP;
2245 #endif
2246 }
2247 
2248 static void tun_set_msglevel(struct net_device *dev, u32 value)
2249 {
2250 #ifdef TUN_DEBUG
2251 	struct tun_struct *tun = netdev_priv(dev);
2252 	tun->debug = value;
2253 #endif
2254 }
2255 
2256 static const struct ethtool_ops tun_ethtool_ops = {
2257 	.get_settings	= tun_get_settings,
2258 	.get_drvinfo	= tun_get_drvinfo,
2259 	.get_msglevel	= tun_get_msglevel,
2260 	.set_msglevel	= tun_set_msglevel,
2261 	.get_link	= ethtool_op_get_link,
2262 };
2263 
2264 
2265 static int __init tun_init(void)
2266 {
2267 	int ret = 0;
2268 
2269 	pr_info("%s, %s\n", DRV_DESCRIPTION, DRV_VERSION);
2270 	pr_info("%s\n", DRV_COPYRIGHT);
2271 
2272 	ret = rtnl_link_register(&tun_link_ops);
2273 	if (ret) {
2274 		pr_err("Can't register link_ops\n");
2275 		goto err_linkops;
2276 	}
2277 
2278 	ret = misc_register(&tun_miscdev);
2279 	if (ret) {
2280 		pr_err("Can't register misc device %d\n", TUN_MINOR);
2281 		goto err_misc;
2282 	}
2283 	return  0;
2284 err_misc:
2285 	rtnl_link_unregister(&tun_link_ops);
2286 err_linkops:
2287 	return ret;
2288 }
2289 
2290 static void tun_cleanup(void)
2291 {
2292 	misc_deregister(&tun_miscdev);
2293 	rtnl_link_unregister(&tun_link_ops);
2294 }
2295 
2296 /* Get an underlying socket object from tun file.  Returns error unless file is
2297  * attached to a device.  The returned object works like a packet socket, it
2298  * can be used for sock_sendmsg/sock_recvmsg.  The caller is responsible for
2299  * holding a reference to the file for as long as the socket is in use. */
2300 struct socket *tun_get_socket(struct file *file)
2301 {
2302 	struct tun_file *tfile;
2303 	if (file->f_op != &tun_fops)
2304 		return ERR_PTR(-EINVAL);
2305 	tfile = file->private_data;
2306 	if (!tfile)
2307 		return ERR_PTR(-EBADFD);
2308 	return &tfile->socket;
2309 }
2310 EXPORT_SYMBOL_GPL(tun_get_socket);
2311 
2312 module_init(tun_init);
2313 module_exit(tun_cleanup);
2314 MODULE_DESCRIPTION(DRV_DESCRIPTION);
2315 MODULE_AUTHOR(DRV_COPYRIGHT);
2316 MODULE_LICENSE("GPL");
2317 MODULE_ALIAS_MISCDEV(TUN_MINOR);
2318 MODULE_ALIAS("devname:net/tun");
2319