xref: /openbmc/linux/drivers/net/tun.c (revision 39b6f3aa)
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 = 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 			for (i = 0; i < num_pages; i++)
1014 				put_page(page[i]);
1015 			return -EFAULT;
1016 		}
1017 		truesize = size * PAGE_SIZE;
1018 		skb->data_len += len;
1019 		skb->len += len;
1020 		skb->truesize += truesize;
1021 		atomic_add(truesize, &skb->sk->sk_wmem_alloc);
1022 		while (len) {
1023 			int off = base & ~PAGE_MASK;
1024 			int size = min_t(int, len, PAGE_SIZE - off);
1025 			__skb_fill_page_desc(skb, i, page[i], off, size);
1026 			skb_shinfo(skb)->nr_frags++;
1027 			/* increase sk_wmem_alloc */
1028 			base += size;
1029 			len -= size;
1030 			i++;
1031 		}
1032 		offset = 0;
1033 		++from;
1034 	}
1035 	return 0;
1036 }
1037 
1038 /* Get packet from user space buffer */
1039 static ssize_t tun_get_user(struct tun_struct *tun, struct tun_file *tfile,
1040 			    void *msg_control, const struct iovec *iv,
1041 			    size_t total_len, size_t count, int noblock)
1042 {
1043 	struct tun_pi pi = { 0, cpu_to_be16(ETH_P_IP) };
1044 	struct sk_buff *skb;
1045 	size_t len = total_len, align = NET_SKB_PAD;
1046 	struct virtio_net_hdr gso = { 0 };
1047 	int offset = 0;
1048 	int copylen;
1049 	bool zerocopy = false;
1050 	int err;
1051 	u32 rxhash;
1052 
1053 	if (!(tun->flags & TUN_NO_PI)) {
1054 		if ((len -= sizeof(pi)) > total_len)
1055 			return -EINVAL;
1056 
1057 		if (memcpy_fromiovecend((void *)&pi, iv, 0, sizeof(pi)))
1058 			return -EFAULT;
1059 		offset += sizeof(pi);
1060 	}
1061 
1062 	if (tun->flags & TUN_VNET_HDR) {
1063 		if ((len -= tun->vnet_hdr_sz) > total_len)
1064 			return -EINVAL;
1065 
1066 		if (memcpy_fromiovecend((void *)&gso, iv, offset, sizeof(gso)))
1067 			return -EFAULT;
1068 
1069 		if ((gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
1070 		    gso.csum_start + gso.csum_offset + 2 > gso.hdr_len)
1071 			gso.hdr_len = gso.csum_start + gso.csum_offset + 2;
1072 
1073 		if (gso.hdr_len > len)
1074 			return -EINVAL;
1075 		offset += tun->vnet_hdr_sz;
1076 	}
1077 
1078 	if ((tun->flags & TUN_TYPE_MASK) == TUN_TAP_DEV) {
1079 		align += NET_IP_ALIGN;
1080 		if (unlikely(len < ETH_HLEN ||
1081 			     (gso.hdr_len && gso.hdr_len < ETH_HLEN)))
1082 			return -EINVAL;
1083 	}
1084 
1085 	if (msg_control)
1086 		zerocopy = true;
1087 
1088 	if (zerocopy) {
1089 		/* Userspace may produce vectors with count greater than
1090 		 * MAX_SKB_FRAGS, so we need to linearize parts of the skb
1091 		 * to let the rest of data to be fit in the frags.
1092 		 */
1093 		if (count > MAX_SKB_FRAGS) {
1094 			copylen = iov_length(iv, count - MAX_SKB_FRAGS);
1095 			if (copylen < offset)
1096 				copylen = 0;
1097 			else
1098 				copylen -= offset;
1099 		} else
1100 				copylen = 0;
1101 		/* There are 256 bytes to be copied in skb, so there is enough
1102 		 * room for skb expand head in case it is used.
1103 		 * The rest of the buffer is mapped from userspace.
1104 		 */
1105 		if (copylen < gso.hdr_len)
1106 			copylen = gso.hdr_len;
1107 		if (!copylen)
1108 			copylen = GOODCOPY_LEN;
1109 	} else
1110 		copylen = len;
1111 
1112 	skb = tun_alloc_skb(tfile, align, copylen, gso.hdr_len, 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_iovec(skb, iv, offset, count);
1121 	else
1122 		err = skb_copy_datagram_from_iovec(skb, 0, iv, offset, len);
1123 
1124 	if (err) {
1125 		tun->dev->stats.rx_dropped++;
1126 		kfree_skb(skb);
1127 		return -EFAULT;
1128 	}
1129 
1130 	if (gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
1131 		if (!skb_partial_csum_set(skb, gso.csum_start,
1132 					  gso.csum_offset)) {
1133 			tun->dev->stats.rx_frame_errors++;
1134 			kfree_skb(skb);
1135 			return -EINVAL;
1136 		}
1137 	}
1138 
1139 	switch (tun->flags & TUN_TYPE_MASK) {
1140 	case TUN_TUN_DEV:
1141 		if (tun->flags & TUN_NO_PI) {
1142 			switch (skb->data[0] & 0xf0) {
1143 			case 0x40:
1144 				pi.proto = htons(ETH_P_IP);
1145 				break;
1146 			case 0x60:
1147 				pi.proto = htons(ETH_P_IPV6);
1148 				break;
1149 			default:
1150 				tun->dev->stats.rx_dropped++;
1151 				kfree_skb(skb);
1152 				return -EINVAL;
1153 			}
1154 		}
1155 
1156 		skb_reset_mac_header(skb);
1157 		skb->protocol = pi.proto;
1158 		skb->dev = tun->dev;
1159 		break;
1160 	case TUN_TAP_DEV:
1161 		skb->protocol = eth_type_trans(skb, tun->dev);
1162 		break;
1163 	}
1164 
1165 	if (gso.gso_type != VIRTIO_NET_HDR_GSO_NONE) {
1166 		pr_debug("GSO!\n");
1167 		switch (gso.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
1168 		case VIRTIO_NET_HDR_GSO_TCPV4:
1169 			skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
1170 			break;
1171 		case VIRTIO_NET_HDR_GSO_TCPV6:
1172 			skb_shinfo(skb)->gso_type = SKB_GSO_TCPV6;
1173 			break;
1174 		case VIRTIO_NET_HDR_GSO_UDP:
1175 			skb_shinfo(skb)->gso_type = SKB_GSO_UDP;
1176 			break;
1177 		default:
1178 			tun->dev->stats.rx_frame_errors++;
1179 			kfree_skb(skb);
1180 			return -EINVAL;
1181 		}
1182 
1183 		if (gso.gso_type & VIRTIO_NET_HDR_GSO_ECN)
1184 			skb_shinfo(skb)->gso_type |= SKB_GSO_TCP_ECN;
1185 
1186 		skb_shinfo(skb)->gso_size = gso.gso_size;
1187 		if (skb_shinfo(skb)->gso_size == 0) {
1188 			tun->dev->stats.rx_frame_errors++;
1189 			kfree_skb(skb);
1190 			return -EINVAL;
1191 		}
1192 
1193 		/* Header must be checked, and gso_segs computed. */
1194 		skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
1195 		skb_shinfo(skb)->gso_segs = 0;
1196 	}
1197 
1198 	/* copy skb_ubuf_info for callback when skb has no error */
1199 	if (zerocopy) {
1200 		skb_shinfo(skb)->destructor_arg = msg_control;
1201 		skb_shinfo(skb)->tx_flags |= SKBTX_DEV_ZEROCOPY;
1202 		skb_shinfo(skb)->tx_flags |= SKBTX_SHARED_FRAG;
1203 	}
1204 
1205 	skb_reset_network_header(skb);
1206 	skb_probe_transport_header(skb, 0);
1207 
1208 	rxhash = skb_get_rxhash(skb);
1209 	netif_rx_ni(skb);
1210 
1211 	tun->dev->stats.rx_packets++;
1212 	tun->dev->stats.rx_bytes += len;
1213 
1214 	tun_flow_update(tun, rxhash, tfile);
1215 	return total_len;
1216 }
1217 
1218 static ssize_t tun_chr_aio_write(struct kiocb *iocb, const struct iovec *iv,
1219 			      unsigned long count, loff_t pos)
1220 {
1221 	struct file *file = iocb->ki_filp;
1222 	struct tun_struct *tun = tun_get(file);
1223 	struct tun_file *tfile = file->private_data;
1224 	ssize_t result;
1225 
1226 	if (!tun)
1227 		return -EBADFD;
1228 
1229 	tun_debug(KERN_INFO, tun, "tun_chr_write %ld\n", count);
1230 
1231 	result = tun_get_user(tun, tfile, NULL, iv, iov_length(iv, count),
1232 			      count, file->f_flags & O_NONBLOCK);
1233 
1234 	tun_put(tun);
1235 	return result;
1236 }
1237 
1238 /* Put packet to the user space buffer */
1239 static ssize_t tun_put_user(struct tun_struct *tun,
1240 			    struct tun_file *tfile,
1241 			    struct sk_buff *skb,
1242 			    const struct iovec *iv, int len)
1243 {
1244 	struct tun_pi pi = { 0, skb->protocol };
1245 	ssize_t total = 0;
1246 
1247 	if (!(tun->flags & TUN_NO_PI)) {
1248 		if ((len -= sizeof(pi)) < 0)
1249 			return -EINVAL;
1250 
1251 		if (len < skb->len) {
1252 			/* Packet will be striped */
1253 			pi.flags |= TUN_PKT_STRIP;
1254 		}
1255 
1256 		if (memcpy_toiovecend(iv, (void *) &pi, 0, sizeof(pi)))
1257 			return -EFAULT;
1258 		total += sizeof(pi);
1259 	}
1260 
1261 	if (tun->flags & TUN_VNET_HDR) {
1262 		struct virtio_net_hdr gso = { 0 }; /* no info leak */
1263 		if ((len -= tun->vnet_hdr_sz) < 0)
1264 			return -EINVAL;
1265 
1266 		if (skb_is_gso(skb)) {
1267 			struct skb_shared_info *sinfo = skb_shinfo(skb);
1268 
1269 			/* This is a hint as to how much should be linear. */
1270 			gso.hdr_len = skb_headlen(skb);
1271 			gso.gso_size = sinfo->gso_size;
1272 			if (sinfo->gso_type & SKB_GSO_TCPV4)
1273 				gso.gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
1274 			else if (sinfo->gso_type & SKB_GSO_TCPV6)
1275 				gso.gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
1276 			else if (sinfo->gso_type & SKB_GSO_UDP)
1277 				gso.gso_type = VIRTIO_NET_HDR_GSO_UDP;
1278 			else {
1279 				pr_err("unexpected GSO type: "
1280 				       "0x%x, gso_size %d, hdr_len %d\n",
1281 				       sinfo->gso_type, gso.gso_size,
1282 				       gso.hdr_len);
1283 				print_hex_dump(KERN_ERR, "tun: ",
1284 					       DUMP_PREFIX_NONE,
1285 					       16, 1, skb->head,
1286 					       min((int)gso.hdr_len, 64), true);
1287 				WARN_ON_ONCE(1);
1288 				return -EINVAL;
1289 			}
1290 			if (sinfo->gso_type & SKB_GSO_TCP_ECN)
1291 				gso.gso_type |= VIRTIO_NET_HDR_GSO_ECN;
1292 		} else
1293 			gso.gso_type = VIRTIO_NET_HDR_GSO_NONE;
1294 
1295 		if (skb->ip_summed == CHECKSUM_PARTIAL) {
1296 			gso.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
1297 			gso.csum_start = skb_checksum_start_offset(skb);
1298 			gso.csum_offset = skb->csum_offset;
1299 		} else if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
1300 			gso.flags = VIRTIO_NET_HDR_F_DATA_VALID;
1301 		} /* else everything is zero */
1302 
1303 		if (unlikely(memcpy_toiovecend(iv, (void *)&gso, total,
1304 					       sizeof(gso))))
1305 			return -EFAULT;
1306 		total += tun->vnet_hdr_sz;
1307 	}
1308 
1309 	len = min_t(int, skb->len, len);
1310 
1311 	skb_copy_datagram_const_iovec(skb, 0, iv, total, len);
1312 	total += skb->len;
1313 
1314 	tun->dev->stats.tx_packets++;
1315 	tun->dev->stats.tx_bytes += len;
1316 
1317 	return total;
1318 }
1319 
1320 static ssize_t tun_do_read(struct tun_struct *tun, struct tun_file *tfile,
1321 			   struct kiocb *iocb, const struct iovec *iv,
1322 			   ssize_t len, int noblock)
1323 {
1324 	DECLARE_WAITQUEUE(wait, current);
1325 	struct sk_buff *skb;
1326 	ssize_t ret = 0;
1327 
1328 	tun_debug(KERN_INFO, tun, "tun_do_read\n");
1329 
1330 	if (unlikely(!noblock))
1331 		add_wait_queue(&tfile->wq.wait, &wait);
1332 	while (len) {
1333 		current->state = TASK_INTERRUPTIBLE;
1334 
1335 		/* Read frames from the queue */
1336 		if (!(skb = skb_dequeue(&tfile->socket.sk->sk_receive_queue))) {
1337 			if (noblock) {
1338 				ret = -EAGAIN;
1339 				break;
1340 			}
1341 			if (signal_pending(current)) {
1342 				ret = -ERESTARTSYS;
1343 				break;
1344 			}
1345 			if (tun->dev->reg_state != NETREG_REGISTERED) {
1346 				ret = -EIO;
1347 				break;
1348 			}
1349 
1350 			/* Nothing to read, let's sleep */
1351 			schedule();
1352 			continue;
1353 		}
1354 
1355 		ret = tun_put_user(tun, tfile, skb, iv, len);
1356 		kfree_skb(skb);
1357 		break;
1358 	}
1359 
1360 	current->state = TASK_RUNNING;
1361 	if (unlikely(!noblock))
1362 		remove_wait_queue(&tfile->wq.wait, &wait);
1363 
1364 	return ret;
1365 }
1366 
1367 static ssize_t tun_chr_aio_read(struct kiocb *iocb, const struct iovec *iv,
1368 			    unsigned long count, loff_t pos)
1369 {
1370 	struct file *file = iocb->ki_filp;
1371 	struct tun_file *tfile = file->private_data;
1372 	struct tun_struct *tun = __tun_get(tfile);
1373 	ssize_t len, ret;
1374 
1375 	if (!tun)
1376 		return -EBADFD;
1377 	len = iov_length(iv, count);
1378 	if (len < 0) {
1379 		ret = -EINVAL;
1380 		goto out;
1381 	}
1382 
1383 	ret = tun_do_read(tun, tfile, iocb, iv, len,
1384 			  file->f_flags & O_NONBLOCK);
1385 	ret = min_t(ssize_t, ret, len);
1386 out:
1387 	tun_put(tun);
1388 	return ret;
1389 }
1390 
1391 static void tun_free_netdev(struct net_device *dev)
1392 {
1393 	struct tun_struct *tun = netdev_priv(dev);
1394 
1395 	BUG_ON(!(list_empty(&tun->disabled)));
1396 	tun_flow_uninit(tun);
1397 	security_tun_dev_free_security(tun->security);
1398 	free_netdev(dev);
1399 }
1400 
1401 static void tun_setup(struct net_device *dev)
1402 {
1403 	struct tun_struct *tun = netdev_priv(dev);
1404 
1405 	tun->owner = INVALID_UID;
1406 	tun->group = INVALID_GID;
1407 
1408 	dev->ethtool_ops = &tun_ethtool_ops;
1409 	dev->destructor = tun_free_netdev;
1410 }
1411 
1412 /* Trivial set of netlink ops to allow deleting tun or tap
1413  * device with netlink.
1414  */
1415 static int tun_validate(struct nlattr *tb[], struct nlattr *data[])
1416 {
1417 	return -EINVAL;
1418 }
1419 
1420 static struct rtnl_link_ops tun_link_ops __read_mostly = {
1421 	.kind		= DRV_NAME,
1422 	.priv_size	= sizeof(struct tun_struct),
1423 	.setup		= tun_setup,
1424 	.validate	= tun_validate,
1425 };
1426 
1427 static void tun_sock_write_space(struct sock *sk)
1428 {
1429 	struct tun_file *tfile;
1430 	wait_queue_head_t *wqueue;
1431 
1432 	if (!sock_writeable(sk))
1433 		return;
1434 
1435 	if (!test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags))
1436 		return;
1437 
1438 	wqueue = sk_sleep(sk);
1439 	if (wqueue && waitqueue_active(wqueue))
1440 		wake_up_interruptible_sync_poll(wqueue, POLLOUT |
1441 						POLLWRNORM | POLLWRBAND);
1442 
1443 	tfile = container_of(sk, struct tun_file, sk);
1444 	kill_fasync(&tfile->fasync, SIGIO, POLL_OUT);
1445 }
1446 
1447 static int tun_sendmsg(struct kiocb *iocb, struct socket *sock,
1448 		       struct msghdr *m, size_t total_len)
1449 {
1450 	int ret;
1451 	struct tun_file *tfile = container_of(sock, struct tun_file, socket);
1452 	struct tun_struct *tun = __tun_get(tfile);
1453 
1454 	if (!tun)
1455 		return -EBADFD;
1456 	ret = tun_get_user(tun, tfile, m->msg_control, m->msg_iov, total_len,
1457 			   m->msg_iovlen, m->msg_flags & MSG_DONTWAIT);
1458 	tun_put(tun);
1459 	return ret;
1460 }
1461 
1462 
1463 static int tun_recvmsg(struct kiocb *iocb, struct socket *sock,
1464 		       struct msghdr *m, size_t total_len,
1465 		       int flags)
1466 {
1467 	struct tun_file *tfile = container_of(sock, struct tun_file, socket);
1468 	struct tun_struct *tun = __tun_get(tfile);
1469 	int ret;
1470 
1471 	if (!tun)
1472 		return -EBADFD;
1473 
1474 	if (flags & ~(MSG_DONTWAIT|MSG_TRUNC)) {
1475 		ret = -EINVAL;
1476 		goto out;
1477 	}
1478 	ret = tun_do_read(tun, tfile, iocb, m->msg_iov, total_len,
1479 			  flags & MSG_DONTWAIT);
1480 	if (ret > total_len) {
1481 		m->msg_flags |= MSG_TRUNC;
1482 		ret = flags & MSG_TRUNC ? ret : total_len;
1483 	}
1484 out:
1485 	tun_put(tun);
1486 	return ret;
1487 }
1488 
1489 static int tun_release(struct socket *sock)
1490 {
1491 	if (sock->sk)
1492 		sock_put(sock->sk);
1493 	return 0;
1494 }
1495 
1496 /* Ops structure to mimic raw sockets with tun */
1497 static const struct proto_ops tun_socket_ops = {
1498 	.sendmsg = tun_sendmsg,
1499 	.recvmsg = tun_recvmsg,
1500 	.release = tun_release,
1501 };
1502 
1503 static struct proto tun_proto = {
1504 	.name		= "tun",
1505 	.owner		= THIS_MODULE,
1506 	.obj_size	= sizeof(struct tun_file),
1507 };
1508 
1509 static int tun_flags(struct tun_struct *tun)
1510 {
1511 	int flags = 0;
1512 
1513 	if (tun->flags & TUN_TUN_DEV)
1514 		flags |= IFF_TUN;
1515 	else
1516 		flags |= IFF_TAP;
1517 
1518 	if (tun->flags & TUN_NO_PI)
1519 		flags |= IFF_NO_PI;
1520 
1521 	/* This flag has no real effect.  We track the value for backwards
1522 	 * compatibility.
1523 	 */
1524 	if (tun->flags & TUN_ONE_QUEUE)
1525 		flags |= IFF_ONE_QUEUE;
1526 
1527 	if (tun->flags & TUN_VNET_HDR)
1528 		flags |= IFF_VNET_HDR;
1529 
1530 	if (tun->flags & TUN_TAP_MQ)
1531 		flags |= IFF_MULTI_QUEUE;
1532 
1533 	return flags;
1534 }
1535 
1536 static ssize_t tun_show_flags(struct device *dev, struct device_attribute *attr,
1537 			      char *buf)
1538 {
1539 	struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1540 	return sprintf(buf, "0x%x\n", tun_flags(tun));
1541 }
1542 
1543 static ssize_t tun_show_owner(struct device *dev, struct device_attribute *attr,
1544 			      char *buf)
1545 {
1546 	struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1547 	return uid_valid(tun->owner)?
1548 		sprintf(buf, "%u\n",
1549 			from_kuid_munged(current_user_ns(), tun->owner)):
1550 		sprintf(buf, "-1\n");
1551 }
1552 
1553 static ssize_t tun_show_group(struct device *dev, struct device_attribute *attr,
1554 			      char *buf)
1555 {
1556 	struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1557 	return gid_valid(tun->group) ?
1558 		sprintf(buf, "%u\n",
1559 			from_kgid_munged(current_user_ns(), tun->group)):
1560 		sprintf(buf, "-1\n");
1561 }
1562 
1563 static DEVICE_ATTR(tun_flags, 0444, tun_show_flags, NULL);
1564 static DEVICE_ATTR(owner, 0444, tun_show_owner, NULL);
1565 static DEVICE_ATTR(group, 0444, tun_show_group, NULL);
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 & TUN_TAP_MQ))
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);
1599 		if (err < 0)
1600 			return err;
1601 
1602 		if (tun->flags & TUN_TAP_MQ &&
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 |= TUN_TUN_DEV;
1626 			name = "tun%d";
1627 		} else if (ifr->ifr_flags & IFF_TAP) {
1628 			/* TAP device */
1629 			flags |= TUN_TAP_DEV;
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 				       tun_setup, queues, queues);
1639 
1640 		if (!dev)
1641 			return -ENOMEM;
1642 
1643 		dev_net_set(dev, net);
1644 		dev->rtnl_link_ops = &tun_link_ops;
1645 
1646 		tun = netdev_priv(dev);
1647 		tun->dev = dev;
1648 		tun->flags = flags;
1649 		tun->txflt.count = 0;
1650 		tun->vnet_hdr_sz = sizeof(struct virtio_net_hdr);
1651 
1652 		tun->filter_attached = false;
1653 		tun->sndbuf = tfile->socket.sk->sk_sndbuf;
1654 
1655 		spin_lock_init(&tun->lock);
1656 
1657 		err = security_tun_dev_alloc_security(&tun->security);
1658 		if (err < 0)
1659 			goto err_free_dev;
1660 
1661 		tun_net_init(dev);
1662 
1663 		err = tun_flow_init(tun);
1664 		if (err < 0)
1665 			goto err_free_dev;
1666 
1667 		dev->hw_features = NETIF_F_SG | NETIF_F_FRAGLIST |
1668 			TUN_USER_FEATURES;
1669 		dev->features = dev->hw_features;
1670 		dev->vlan_features = dev->features;
1671 
1672 		INIT_LIST_HEAD(&tun->disabled);
1673 		err = tun_attach(tun, file);
1674 		if (err < 0)
1675 			goto err_free_dev;
1676 
1677 		err = register_netdevice(tun->dev);
1678 		if (err < 0)
1679 			goto err_free_dev;
1680 
1681 		if (device_create_file(&tun->dev->dev, &dev_attr_tun_flags) ||
1682 		    device_create_file(&tun->dev->dev, &dev_attr_owner) ||
1683 		    device_create_file(&tun->dev->dev, &dev_attr_group))
1684 			pr_err("Failed to create tun sysfs files\n");
1685 	}
1686 
1687 	netif_carrier_on(tun->dev);
1688 
1689 	tun_debug(KERN_INFO, tun, "tun_set_iff\n");
1690 
1691 	if (ifr->ifr_flags & IFF_NO_PI)
1692 		tun->flags |= TUN_NO_PI;
1693 	else
1694 		tun->flags &= ~TUN_NO_PI;
1695 
1696 	/* This flag has no real effect.  We track the value for backwards
1697 	 * compatibility.
1698 	 */
1699 	if (ifr->ifr_flags & IFF_ONE_QUEUE)
1700 		tun->flags |= TUN_ONE_QUEUE;
1701 	else
1702 		tun->flags &= ~TUN_ONE_QUEUE;
1703 
1704 	if (ifr->ifr_flags & IFF_VNET_HDR)
1705 		tun->flags |= TUN_VNET_HDR;
1706 	else
1707 		tun->flags &= ~TUN_VNET_HDR;
1708 
1709 	if (ifr->ifr_flags & IFF_MULTI_QUEUE)
1710 		tun->flags |= TUN_TAP_MQ;
1711 	else
1712 		tun->flags &= ~TUN_TAP_MQ;
1713 
1714 	/* Make sure persistent devices do not get stuck in
1715 	 * xoff state.
1716 	 */
1717 	if (netif_running(tun->dev))
1718 		netif_tx_wake_all_queues(tun->dev);
1719 
1720 	strcpy(ifr->ifr_name, tun->dev->name);
1721 	return 0;
1722 
1723  err_free_dev:
1724 	free_netdev(dev);
1725 	return err;
1726 }
1727 
1728 static void tun_get_iff(struct net *net, struct tun_struct *tun,
1729 		       struct ifreq *ifr)
1730 {
1731 	tun_debug(KERN_INFO, tun, "tun_get_iff\n");
1732 
1733 	strcpy(ifr->ifr_name, tun->dev->name);
1734 
1735 	ifr->ifr_flags = tun_flags(tun);
1736 
1737 }
1738 
1739 /* This is like a cut-down ethtool ops, except done via tun fd so no
1740  * privs required. */
1741 static int set_offload(struct tun_struct *tun, unsigned long arg)
1742 {
1743 	netdev_features_t features = 0;
1744 
1745 	if (arg & TUN_F_CSUM) {
1746 		features |= NETIF_F_HW_CSUM;
1747 		arg &= ~TUN_F_CSUM;
1748 
1749 		if (arg & (TUN_F_TSO4|TUN_F_TSO6)) {
1750 			if (arg & TUN_F_TSO_ECN) {
1751 				features |= NETIF_F_TSO_ECN;
1752 				arg &= ~TUN_F_TSO_ECN;
1753 			}
1754 			if (arg & TUN_F_TSO4)
1755 				features |= NETIF_F_TSO;
1756 			if (arg & TUN_F_TSO6)
1757 				features |= NETIF_F_TSO6;
1758 			arg &= ~(TUN_F_TSO4|TUN_F_TSO6);
1759 		}
1760 
1761 		if (arg & TUN_F_UFO) {
1762 			features |= NETIF_F_UFO;
1763 			arg &= ~TUN_F_UFO;
1764 		}
1765 	}
1766 
1767 	/* This gives the user a way to test for new features in future by
1768 	 * trying to set them. */
1769 	if (arg)
1770 		return -EINVAL;
1771 
1772 	tun->set_features = features;
1773 	netdev_update_features(tun->dev);
1774 
1775 	return 0;
1776 }
1777 
1778 static void tun_detach_filter(struct tun_struct *tun, int n)
1779 {
1780 	int i;
1781 	struct tun_file *tfile;
1782 
1783 	for (i = 0; i < n; i++) {
1784 		tfile = rtnl_dereference(tun->tfiles[i]);
1785 		sk_detach_filter(tfile->socket.sk);
1786 	}
1787 
1788 	tun->filter_attached = false;
1789 }
1790 
1791 static int tun_attach_filter(struct tun_struct *tun)
1792 {
1793 	int i, ret = 0;
1794 	struct tun_file *tfile;
1795 
1796 	for (i = 0; i < tun->numqueues; i++) {
1797 		tfile = rtnl_dereference(tun->tfiles[i]);
1798 		ret = sk_attach_filter(&tun->fprog, tfile->socket.sk);
1799 		if (ret) {
1800 			tun_detach_filter(tun, i);
1801 			return ret;
1802 		}
1803 	}
1804 
1805 	tun->filter_attached = true;
1806 	return ret;
1807 }
1808 
1809 static void tun_set_sndbuf(struct tun_struct *tun)
1810 {
1811 	struct tun_file *tfile;
1812 	int i;
1813 
1814 	for (i = 0; i < tun->numqueues; i++) {
1815 		tfile = rtnl_dereference(tun->tfiles[i]);
1816 		tfile->socket.sk->sk_sndbuf = tun->sndbuf;
1817 	}
1818 }
1819 
1820 static int tun_set_queue(struct file *file, struct ifreq *ifr)
1821 {
1822 	struct tun_file *tfile = file->private_data;
1823 	struct tun_struct *tun;
1824 	int ret = 0;
1825 
1826 	rtnl_lock();
1827 
1828 	if (ifr->ifr_flags & IFF_ATTACH_QUEUE) {
1829 		tun = tfile->detached;
1830 		if (!tun) {
1831 			ret = -EINVAL;
1832 			goto unlock;
1833 		}
1834 		ret = security_tun_dev_attach_queue(tun->security);
1835 		if (ret < 0)
1836 			goto unlock;
1837 		ret = tun_attach(tun, file);
1838 	} else if (ifr->ifr_flags & IFF_DETACH_QUEUE) {
1839 		tun = rtnl_dereference(tfile->tun);
1840 		if (!tun || !(tun->flags & TUN_TAP_MQ) || tfile->detached)
1841 			ret = -EINVAL;
1842 		else
1843 			__tun_detach(tfile, false);
1844 	} else
1845 		ret = -EINVAL;
1846 
1847 unlock:
1848 	rtnl_unlock();
1849 	return ret;
1850 }
1851 
1852 static long __tun_chr_ioctl(struct file *file, unsigned int cmd,
1853 			    unsigned long arg, int ifreq_len)
1854 {
1855 	struct tun_file *tfile = file->private_data;
1856 	struct tun_struct *tun;
1857 	void __user* argp = (void __user*)arg;
1858 	struct ifreq ifr;
1859 	kuid_t owner;
1860 	kgid_t group;
1861 	int sndbuf;
1862 	int vnet_hdr_sz;
1863 	int ret;
1864 
1865 	if (cmd == TUNSETIFF || cmd == TUNSETQUEUE || _IOC_TYPE(cmd) == 0x89) {
1866 		if (copy_from_user(&ifr, argp, ifreq_len))
1867 			return -EFAULT;
1868 	} else {
1869 		memset(&ifr, 0, sizeof(ifr));
1870 	}
1871 	if (cmd == TUNGETFEATURES) {
1872 		/* Currently this just means: "what IFF flags are valid?".
1873 		 * This is needed because we never checked for invalid flags on
1874 		 * TUNSETIFF. */
1875 		return put_user(IFF_TUN | IFF_TAP | IFF_NO_PI | IFF_ONE_QUEUE |
1876 				IFF_VNET_HDR | IFF_MULTI_QUEUE,
1877 				(unsigned int __user*)argp);
1878 	} else if (cmd == TUNSETQUEUE)
1879 		return tun_set_queue(file, &ifr);
1880 
1881 	ret = 0;
1882 	rtnl_lock();
1883 
1884 	tun = __tun_get(tfile);
1885 	if (cmd == TUNSETIFF && !tun) {
1886 		ifr.ifr_name[IFNAMSIZ-1] = '\0';
1887 
1888 		ret = tun_set_iff(tfile->net, file, &ifr);
1889 
1890 		if (ret)
1891 			goto unlock;
1892 
1893 		if (copy_to_user(argp, &ifr, ifreq_len))
1894 			ret = -EFAULT;
1895 		goto unlock;
1896 	}
1897 
1898 	ret = -EBADFD;
1899 	if (!tun)
1900 		goto unlock;
1901 
1902 	tun_debug(KERN_INFO, tun, "tun_chr_ioctl cmd %u\n", cmd);
1903 
1904 	ret = 0;
1905 	switch (cmd) {
1906 	case TUNGETIFF:
1907 		tun_get_iff(current->nsproxy->net_ns, tun, &ifr);
1908 
1909 		if (copy_to_user(argp, &ifr, ifreq_len))
1910 			ret = -EFAULT;
1911 		break;
1912 
1913 	case TUNSETNOCSUM:
1914 		/* Disable/Enable checksum */
1915 
1916 		/* [unimplemented] */
1917 		tun_debug(KERN_INFO, tun, "ignored: set checksum %s\n",
1918 			  arg ? "disabled" : "enabled");
1919 		break;
1920 
1921 	case TUNSETPERSIST:
1922 		/* Disable/Enable persist mode. Keep an extra reference to the
1923 		 * module to prevent the module being unprobed.
1924 		 */
1925 		if (arg && !(tun->flags & TUN_PERSIST)) {
1926 			tun->flags |= TUN_PERSIST;
1927 			__module_get(THIS_MODULE);
1928 		}
1929 		if (!arg && (tun->flags & TUN_PERSIST)) {
1930 			tun->flags &= ~TUN_PERSIST;
1931 			module_put(THIS_MODULE);
1932 		}
1933 
1934 		tun_debug(KERN_INFO, tun, "persist %s\n",
1935 			  arg ? "enabled" : "disabled");
1936 		break;
1937 
1938 	case TUNSETOWNER:
1939 		/* Set owner of the device */
1940 		owner = make_kuid(current_user_ns(), arg);
1941 		if (!uid_valid(owner)) {
1942 			ret = -EINVAL;
1943 			break;
1944 		}
1945 		tun->owner = owner;
1946 		tun_debug(KERN_INFO, tun, "owner set to %u\n",
1947 			  from_kuid(&init_user_ns, tun->owner));
1948 		break;
1949 
1950 	case TUNSETGROUP:
1951 		/* Set group of the device */
1952 		group = make_kgid(current_user_ns(), arg);
1953 		if (!gid_valid(group)) {
1954 			ret = -EINVAL;
1955 			break;
1956 		}
1957 		tun->group = group;
1958 		tun_debug(KERN_INFO, tun, "group set to %u\n",
1959 			  from_kgid(&init_user_ns, tun->group));
1960 		break;
1961 
1962 	case TUNSETLINK:
1963 		/* Only allow setting the type when the interface is down */
1964 		if (tun->dev->flags & IFF_UP) {
1965 			tun_debug(KERN_INFO, tun,
1966 				  "Linktype set failed because interface is up\n");
1967 			ret = -EBUSY;
1968 		} else {
1969 			tun->dev->type = (int) arg;
1970 			tun_debug(KERN_INFO, tun, "linktype set to %d\n",
1971 				  tun->dev->type);
1972 			ret = 0;
1973 		}
1974 		break;
1975 
1976 #ifdef TUN_DEBUG
1977 	case TUNSETDEBUG:
1978 		tun->debug = arg;
1979 		break;
1980 #endif
1981 	case TUNSETOFFLOAD:
1982 		ret = set_offload(tun, arg);
1983 		break;
1984 
1985 	case TUNSETTXFILTER:
1986 		/* Can be set only for TAPs */
1987 		ret = -EINVAL;
1988 		if ((tun->flags & TUN_TYPE_MASK) != TUN_TAP_DEV)
1989 			break;
1990 		ret = update_filter(&tun->txflt, (void __user *)arg);
1991 		break;
1992 
1993 	case SIOCGIFHWADDR:
1994 		/* Get hw address */
1995 		memcpy(ifr.ifr_hwaddr.sa_data, tun->dev->dev_addr, ETH_ALEN);
1996 		ifr.ifr_hwaddr.sa_family = tun->dev->type;
1997 		if (copy_to_user(argp, &ifr, ifreq_len))
1998 			ret = -EFAULT;
1999 		break;
2000 
2001 	case SIOCSIFHWADDR:
2002 		/* Set hw address */
2003 		tun_debug(KERN_DEBUG, tun, "set hw address: %pM\n",
2004 			  ifr.ifr_hwaddr.sa_data);
2005 
2006 		ret = dev_set_mac_address(tun->dev, &ifr.ifr_hwaddr);
2007 		break;
2008 
2009 	case TUNGETSNDBUF:
2010 		sndbuf = tfile->socket.sk->sk_sndbuf;
2011 		if (copy_to_user(argp, &sndbuf, sizeof(sndbuf)))
2012 			ret = -EFAULT;
2013 		break;
2014 
2015 	case TUNSETSNDBUF:
2016 		if (copy_from_user(&sndbuf, argp, sizeof(sndbuf))) {
2017 			ret = -EFAULT;
2018 			break;
2019 		}
2020 
2021 		tun->sndbuf = sndbuf;
2022 		tun_set_sndbuf(tun);
2023 		break;
2024 
2025 	case TUNGETVNETHDRSZ:
2026 		vnet_hdr_sz = tun->vnet_hdr_sz;
2027 		if (copy_to_user(argp, &vnet_hdr_sz, sizeof(vnet_hdr_sz)))
2028 			ret = -EFAULT;
2029 		break;
2030 
2031 	case TUNSETVNETHDRSZ:
2032 		if (copy_from_user(&vnet_hdr_sz, argp, sizeof(vnet_hdr_sz))) {
2033 			ret = -EFAULT;
2034 			break;
2035 		}
2036 		if (vnet_hdr_sz < (int)sizeof(struct virtio_net_hdr)) {
2037 			ret = -EINVAL;
2038 			break;
2039 		}
2040 
2041 		tun->vnet_hdr_sz = vnet_hdr_sz;
2042 		break;
2043 
2044 	case TUNATTACHFILTER:
2045 		/* Can be set only for TAPs */
2046 		ret = -EINVAL;
2047 		if ((tun->flags & TUN_TYPE_MASK) != TUN_TAP_DEV)
2048 			break;
2049 		ret = -EFAULT;
2050 		if (copy_from_user(&tun->fprog, argp, sizeof(tun->fprog)))
2051 			break;
2052 
2053 		ret = tun_attach_filter(tun);
2054 		break;
2055 
2056 	case TUNDETACHFILTER:
2057 		/* Can be set only for TAPs */
2058 		ret = -EINVAL;
2059 		if ((tun->flags & TUN_TYPE_MASK) != TUN_TAP_DEV)
2060 			break;
2061 		ret = 0;
2062 		tun_detach_filter(tun, tun->numqueues);
2063 		break;
2064 
2065 	default:
2066 		ret = -EINVAL;
2067 		break;
2068 	}
2069 
2070 unlock:
2071 	rtnl_unlock();
2072 	if (tun)
2073 		tun_put(tun);
2074 	return ret;
2075 }
2076 
2077 static long tun_chr_ioctl(struct file *file,
2078 			  unsigned int cmd, unsigned long arg)
2079 {
2080 	return __tun_chr_ioctl(file, cmd, arg, sizeof (struct ifreq));
2081 }
2082 
2083 #ifdef CONFIG_COMPAT
2084 static long tun_chr_compat_ioctl(struct file *file,
2085 			 unsigned int cmd, unsigned long arg)
2086 {
2087 	switch (cmd) {
2088 	case TUNSETIFF:
2089 	case TUNGETIFF:
2090 	case TUNSETTXFILTER:
2091 	case TUNGETSNDBUF:
2092 	case TUNSETSNDBUF:
2093 	case SIOCGIFHWADDR:
2094 	case SIOCSIFHWADDR:
2095 		arg = (unsigned long)compat_ptr(arg);
2096 		break;
2097 	default:
2098 		arg = (compat_ulong_t)arg;
2099 		break;
2100 	}
2101 
2102 	/*
2103 	 * compat_ifreq is shorter than ifreq, so we must not access beyond
2104 	 * the end of that structure. All fields that are used in this
2105 	 * driver are compatible though, we don't need to convert the
2106 	 * contents.
2107 	 */
2108 	return __tun_chr_ioctl(file, cmd, arg, sizeof(struct compat_ifreq));
2109 }
2110 #endif /* CONFIG_COMPAT */
2111 
2112 static int tun_chr_fasync(int fd, struct file *file, int on)
2113 {
2114 	struct tun_file *tfile = file->private_data;
2115 	int ret;
2116 
2117 	if ((ret = fasync_helper(fd, file, on, &tfile->fasync)) < 0)
2118 		goto out;
2119 
2120 	if (on) {
2121 		ret = __f_setown(file, task_pid(current), PIDTYPE_PID, 0);
2122 		if (ret)
2123 			goto out;
2124 		tfile->flags |= TUN_FASYNC;
2125 	} else
2126 		tfile->flags &= ~TUN_FASYNC;
2127 	ret = 0;
2128 out:
2129 	return ret;
2130 }
2131 
2132 static int tun_chr_open(struct inode *inode, struct file * file)
2133 {
2134 	struct tun_file *tfile;
2135 
2136 	DBG1(KERN_INFO, "tunX: tun_chr_open\n");
2137 
2138 	tfile = (struct tun_file *)sk_alloc(&init_net, AF_UNSPEC, GFP_KERNEL,
2139 					    &tun_proto);
2140 	if (!tfile)
2141 		return -ENOMEM;
2142 	rcu_assign_pointer(tfile->tun, NULL);
2143 	tfile->net = get_net(current->nsproxy->net_ns);
2144 	tfile->flags = 0;
2145 
2146 	rcu_assign_pointer(tfile->socket.wq, &tfile->wq);
2147 	init_waitqueue_head(&tfile->wq.wait);
2148 
2149 	tfile->socket.file = file;
2150 	tfile->socket.ops = &tun_socket_ops;
2151 
2152 	sock_init_data(&tfile->socket, &tfile->sk);
2153 	sk_change_net(&tfile->sk, tfile->net);
2154 
2155 	tfile->sk.sk_write_space = tun_sock_write_space;
2156 	tfile->sk.sk_sndbuf = INT_MAX;
2157 
2158 	file->private_data = tfile;
2159 	set_bit(SOCK_EXTERNALLY_ALLOCATED, &tfile->socket.flags);
2160 	INIT_LIST_HEAD(&tfile->next);
2161 
2162 	return 0;
2163 }
2164 
2165 static int tun_chr_close(struct inode *inode, struct file *file)
2166 {
2167 	struct tun_file *tfile = file->private_data;
2168 	struct net *net = tfile->net;
2169 
2170 	tun_detach(tfile, true);
2171 	put_net(net);
2172 
2173 	return 0;
2174 }
2175 
2176 static const struct file_operations tun_fops = {
2177 	.owner	= THIS_MODULE,
2178 	.llseek = no_llseek,
2179 	.read  = do_sync_read,
2180 	.aio_read  = tun_chr_aio_read,
2181 	.write = do_sync_write,
2182 	.aio_write = tun_chr_aio_write,
2183 	.poll	= tun_chr_poll,
2184 	.unlocked_ioctl	= tun_chr_ioctl,
2185 #ifdef CONFIG_COMPAT
2186 	.compat_ioctl = tun_chr_compat_ioctl,
2187 #endif
2188 	.open	= tun_chr_open,
2189 	.release = tun_chr_close,
2190 	.fasync = tun_chr_fasync
2191 };
2192 
2193 static struct miscdevice tun_miscdev = {
2194 	.minor = TUN_MINOR,
2195 	.name = "tun",
2196 	.nodename = "net/tun",
2197 	.fops = &tun_fops,
2198 };
2199 
2200 /* ethtool interface */
2201 
2202 static int tun_get_settings(struct net_device *dev, struct ethtool_cmd *cmd)
2203 {
2204 	cmd->supported		= 0;
2205 	cmd->advertising	= 0;
2206 	ethtool_cmd_speed_set(cmd, SPEED_10);
2207 	cmd->duplex		= DUPLEX_FULL;
2208 	cmd->port		= PORT_TP;
2209 	cmd->phy_address	= 0;
2210 	cmd->transceiver	= XCVR_INTERNAL;
2211 	cmd->autoneg		= AUTONEG_DISABLE;
2212 	cmd->maxtxpkt		= 0;
2213 	cmd->maxrxpkt		= 0;
2214 	return 0;
2215 }
2216 
2217 static void tun_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
2218 {
2219 	struct tun_struct *tun = netdev_priv(dev);
2220 
2221 	strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
2222 	strlcpy(info->version, DRV_VERSION, sizeof(info->version));
2223 
2224 	switch (tun->flags & TUN_TYPE_MASK) {
2225 	case TUN_TUN_DEV:
2226 		strlcpy(info->bus_info, "tun", sizeof(info->bus_info));
2227 		break;
2228 	case TUN_TAP_DEV:
2229 		strlcpy(info->bus_info, "tap", sizeof(info->bus_info));
2230 		break;
2231 	}
2232 }
2233 
2234 static u32 tun_get_msglevel(struct net_device *dev)
2235 {
2236 #ifdef TUN_DEBUG
2237 	struct tun_struct *tun = netdev_priv(dev);
2238 	return tun->debug;
2239 #else
2240 	return -EOPNOTSUPP;
2241 #endif
2242 }
2243 
2244 static void tun_set_msglevel(struct net_device *dev, u32 value)
2245 {
2246 #ifdef TUN_DEBUG
2247 	struct tun_struct *tun = netdev_priv(dev);
2248 	tun->debug = value;
2249 #endif
2250 }
2251 
2252 static const struct ethtool_ops tun_ethtool_ops = {
2253 	.get_settings	= tun_get_settings,
2254 	.get_drvinfo	= tun_get_drvinfo,
2255 	.get_msglevel	= tun_get_msglevel,
2256 	.set_msglevel	= tun_set_msglevel,
2257 	.get_link	= ethtool_op_get_link,
2258 };
2259 
2260 
2261 static int __init tun_init(void)
2262 {
2263 	int ret = 0;
2264 
2265 	pr_info("%s, %s\n", DRV_DESCRIPTION, DRV_VERSION);
2266 	pr_info("%s\n", DRV_COPYRIGHT);
2267 
2268 	ret = rtnl_link_register(&tun_link_ops);
2269 	if (ret) {
2270 		pr_err("Can't register link_ops\n");
2271 		goto err_linkops;
2272 	}
2273 
2274 	ret = misc_register(&tun_miscdev);
2275 	if (ret) {
2276 		pr_err("Can't register misc device %d\n", TUN_MINOR);
2277 		goto err_misc;
2278 	}
2279 	return  0;
2280 err_misc:
2281 	rtnl_link_unregister(&tun_link_ops);
2282 err_linkops:
2283 	return ret;
2284 }
2285 
2286 static void tun_cleanup(void)
2287 {
2288 	misc_deregister(&tun_miscdev);
2289 	rtnl_link_unregister(&tun_link_ops);
2290 }
2291 
2292 /* Get an underlying socket object from tun file.  Returns error unless file is
2293  * attached to a device.  The returned object works like a packet socket, it
2294  * can be used for sock_sendmsg/sock_recvmsg.  The caller is responsible for
2295  * holding a reference to the file for as long as the socket is in use. */
2296 struct socket *tun_get_socket(struct file *file)
2297 {
2298 	struct tun_file *tfile;
2299 	if (file->f_op != &tun_fops)
2300 		return ERR_PTR(-EINVAL);
2301 	tfile = file->private_data;
2302 	if (!tfile)
2303 		return ERR_PTR(-EBADFD);
2304 	return &tfile->socket;
2305 }
2306 EXPORT_SYMBOL_GPL(tun_get_socket);
2307 
2308 module_init(tun_init);
2309 module_exit(tun_cleanup);
2310 MODULE_DESCRIPTION(DRV_DESCRIPTION);
2311 MODULE_AUTHOR(DRV_COPYRIGHT);
2312 MODULE_LICENSE("GPL");
2313 MODULE_ALIAS_MISCDEV(TUN_MINOR);
2314 MODULE_ALIAS("devname:net/tun");
2315