xref: /linux/drivers/net/tun.c (revision a671de086874b9d8155369319b2bd989cf55d77c)
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 /* 1024 is probably a high enough limit: modern hypervisors seem to support on
113  * the order of 100-200 CPUs so this leaves us some breathing space if we want
114  * to match a queue per guest CPU.
115  */
116 #define MAX_TAP_QUEUES 1024
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 };
142 
143 struct tun_flow_entry {
144 	struct hlist_node hash_link;
145 	struct rcu_head rcu;
146 	struct tun_struct *tun;
147 
148 	u32 rxhash;
149 	int queue_index;
150 	unsigned long updated;
151 };
152 
153 #define TUN_NUM_FLOW_ENTRIES 1024
154 
155 /* Since the socket were moved to tun_file, to preserve the behavior of persist
156  * device, socket filter, sndbuf and vnet header size were restore when the
157  * file were attached to a persist device.
158  */
159 struct tun_struct {
160 	struct tun_file __rcu	*tfiles[MAX_TAP_QUEUES];
161 	unsigned int            numqueues;
162 	unsigned int 		flags;
163 	kuid_t			owner;
164 	kgid_t			group;
165 
166 	struct net_device	*dev;
167 	netdev_features_t	set_features;
168 #define TUN_USER_FEATURES (NETIF_F_HW_CSUM|NETIF_F_TSO_ECN|NETIF_F_TSO| \
169 			  NETIF_F_TSO6|NETIF_F_UFO)
170 
171 	int			vnet_hdr_sz;
172 	int			sndbuf;
173 	struct tap_filter	txflt;
174 	struct sock_fprog	fprog;
175 	/* protected by rtnl lock */
176 	bool			filter_attached;
177 #ifdef TUN_DEBUG
178 	int debug;
179 #endif
180 	spinlock_t lock;
181 	struct kmem_cache *flow_cache;
182 	struct hlist_head flows[TUN_NUM_FLOW_ENTRIES];
183 	struct timer_list flow_gc_timer;
184 	unsigned long ageing_time;
185 };
186 
187 static inline u32 tun_hashfn(u32 rxhash)
188 {
189 	return rxhash & 0x3ff;
190 }
191 
192 static struct tun_flow_entry *tun_flow_find(struct hlist_head *head, u32 rxhash)
193 {
194 	struct tun_flow_entry *e;
195 	struct hlist_node *n;
196 
197 	hlist_for_each_entry_rcu(e, n, head, hash_link) {
198 		if (e->rxhash == rxhash)
199 			return e;
200 	}
201 	return NULL;
202 }
203 
204 static struct tun_flow_entry *tun_flow_create(struct tun_struct *tun,
205 					      struct hlist_head *head,
206 					      u32 rxhash, u16 queue_index)
207 {
208 	struct tun_flow_entry *e = kmem_cache_alloc(tun->flow_cache,
209 						    GFP_ATOMIC);
210 	if (e) {
211 		tun_debug(KERN_INFO, tun, "create flow: hash %u index %u\n",
212 			  rxhash, queue_index);
213 		e->updated = jiffies;
214 		e->rxhash = rxhash;
215 		e->queue_index = queue_index;
216 		e->tun = tun;
217 		hlist_add_head_rcu(&e->hash_link, head);
218 	}
219 	return e;
220 }
221 
222 static void tun_flow_free(struct rcu_head *head)
223 {
224 	struct tun_flow_entry *e
225 		= container_of(head, struct tun_flow_entry, rcu);
226 	kmem_cache_free(e->tun->flow_cache, e);
227 }
228 
229 static void tun_flow_delete(struct tun_struct *tun, struct tun_flow_entry *e)
230 {
231 	tun_debug(KERN_INFO, tun, "delete flow: hash %u index %u\n",
232 		  e->rxhash, e->queue_index);
233 	hlist_del_rcu(&e->hash_link);
234 	call_rcu(&e->rcu, tun_flow_free);
235 }
236 
237 static void tun_flow_flush(struct tun_struct *tun)
238 {
239 	int i;
240 
241 	spin_lock_bh(&tun->lock);
242 	for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
243 		struct tun_flow_entry *e;
244 		struct hlist_node *h, *n;
245 
246 		hlist_for_each_entry_safe(e, h, n, &tun->flows[i], hash_link)
247 			tun_flow_delete(tun, e);
248 	}
249 	spin_unlock_bh(&tun->lock);
250 }
251 
252 static void tun_flow_delete_by_queue(struct tun_struct *tun, u16 queue_index)
253 {
254 	int i;
255 
256 	spin_lock_bh(&tun->lock);
257 	for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
258 		struct tun_flow_entry *e;
259 		struct hlist_node *h, *n;
260 
261 		hlist_for_each_entry_safe(e, h, n, &tun->flows[i], hash_link) {
262 			if (e->queue_index == queue_index)
263 				tun_flow_delete(tun, e);
264 		}
265 	}
266 	spin_unlock_bh(&tun->lock);
267 }
268 
269 static void tun_flow_cleanup(unsigned long data)
270 {
271 	struct tun_struct *tun = (struct tun_struct *)data;
272 	unsigned long delay = tun->ageing_time;
273 	unsigned long next_timer = jiffies + delay;
274 	unsigned long count = 0;
275 	int i;
276 
277 	tun_debug(KERN_INFO, tun, "tun_flow_cleanup\n");
278 
279 	spin_lock_bh(&tun->lock);
280 	for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
281 		struct tun_flow_entry *e;
282 		struct hlist_node *h, *n;
283 
284 		hlist_for_each_entry_safe(e, h, n, &tun->flows[i], hash_link) {
285 			unsigned long this_timer;
286 			count++;
287 			this_timer = e->updated + delay;
288 			if (time_before_eq(this_timer, jiffies))
289 				tun_flow_delete(tun, e);
290 			else if (time_before(this_timer, next_timer))
291 				next_timer = this_timer;
292 		}
293 	}
294 
295 	if (count)
296 		mod_timer(&tun->flow_gc_timer, round_jiffies_up(next_timer));
297 	spin_unlock_bh(&tun->lock);
298 }
299 
300 static void tun_flow_update(struct tun_struct *tun, struct sk_buff *skb,
301 			    u16 queue_index)
302 {
303 	struct hlist_head *head;
304 	struct tun_flow_entry *e;
305 	unsigned long delay = tun->ageing_time;
306 	u32 rxhash = skb_get_rxhash(skb);
307 
308 	if (!rxhash)
309 		return;
310 	else
311 		head = &tun->flows[tun_hashfn(rxhash)];
312 
313 	rcu_read_lock();
314 
315 	if (tun->numqueues == 1)
316 		goto unlock;
317 
318 	e = tun_flow_find(head, rxhash);
319 	if (likely(e)) {
320 		/* TODO: keep queueing to old queue until it's empty? */
321 		e->queue_index = queue_index;
322 		e->updated = jiffies;
323 	} else {
324 		spin_lock_bh(&tun->lock);
325 		if (!tun_flow_find(head, rxhash))
326 			tun_flow_create(tun, head, rxhash, queue_index);
327 
328 		if (!timer_pending(&tun->flow_gc_timer))
329 			mod_timer(&tun->flow_gc_timer,
330 				  round_jiffies_up(jiffies + delay));
331 		spin_unlock_bh(&tun->lock);
332 	}
333 
334 unlock:
335 	rcu_read_unlock();
336 }
337 
338 /* We try to identify a flow through its rxhash first. The reason that
339  * we do not check rxq no. is becuase some cards(e.g 82599), chooses
340  * the rxq based on the txq where the last packet of the flow comes. As
341  * the userspace application move between processors, we may get a
342  * different rxq no. here. If we could not get rxhash, then we would
343  * hope the rxq no. may help here.
344  */
345 static u16 tun_select_queue(struct net_device *dev, struct sk_buff *skb)
346 {
347 	struct tun_struct *tun = netdev_priv(dev);
348 	struct tun_flow_entry *e;
349 	u32 txq = 0;
350 	u32 numqueues = 0;
351 
352 	rcu_read_lock();
353 	numqueues = tun->numqueues;
354 
355 	txq = skb_get_rxhash(skb);
356 	if (txq) {
357 		e = tun_flow_find(&tun->flows[tun_hashfn(txq)], txq);
358 		if (e)
359 			txq = e->queue_index;
360 		else
361 			/* use multiply and shift instead of expensive divide */
362 			txq = ((u64)txq * numqueues) >> 32;
363 	} else if (likely(skb_rx_queue_recorded(skb))) {
364 		txq = skb_get_rx_queue(skb);
365 		while (unlikely(txq >= numqueues))
366 			txq -= numqueues;
367 	}
368 
369 	rcu_read_unlock();
370 	return txq;
371 }
372 
373 static inline bool tun_not_capable(struct tun_struct *tun)
374 {
375 	const struct cred *cred = current_cred();
376 	struct net *net = dev_net(tun->dev);
377 
378 	return ((uid_valid(tun->owner) && !uid_eq(cred->euid, tun->owner)) ||
379 		  (gid_valid(tun->group) && !in_egroup_p(tun->group))) &&
380 		!ns_capable(net->user_ns, CAP_NET_ADMIN);
381 }
382 
383 static void tun_set_real_num_queues(struct tun_struct *tun)
384 {
385 	netif_set_real_num_tx_queues(tun->dev, tun->numqueues);
386 	netif_set_real_num_rx_queues(tun->dev, tun->numqueues);
387 }
388 
389 static void __tun_detach(struct tun_file *tfile, bool clean)
390 {
391 	struct tun_file *ntfile;
392 	struct tun_struct *tun;
393 	struct net_device *dev;
394 
395 	tun = rcu_dereference_protected(tfile->tun,
396 					lockdep_rtnl_is_held());
397 	if (tun) {
398 		u16 index = tfile->queue_index;
399 		BUG_ON(index >= tun->numqueues);
400 		dev = tun->dev;
401 
402 		rcu_assign_pointer(tun->tfiles[index],
403 				   tun->tfiles[tun->numqueues - 1]);
404 		rcu_assign_pointer(tfile->tun, NULL);
405 		ntfile = rcu_dereference_protected(tun->tfiles[index],
406 						   lockdep_rtnl_is_held());
407 		ntfile->queue_index = index;
408 
409 		--tun->numqueues;
410 		sock_put(&tfile->sk);
411 
412 		synchronize_net();
413 		tun_flow_delete_by_queue(tun, tun->numqueues + 1);
414 		/* Drop read queue */
415 		skb_queue_purge(&tfile->sk.sk_receive_queue);
416 		tun_set_real_num_queues(tun);
417 
418 		if (tun->numqueues == 0 && !(tun->flags & TUN_PERSIST))
419 			if (dev->reg_state == NETREG_REGISTERED)
420 				unregister_netdevice(dev);
421 	}
422 
423 	if (clean) {
424 		BUG_ON(!test_bit(SOCK_EXTERNALLY_ALLOCATED,
425 				 &tfile->socket.flags));
426 		sk_release_kernel(&tfile->sk);
427 	}
428 }
429 
430 static void tun_detach(struct tun_file *tfile, bool clean)
431 {
432 	rtnl_lock();
433 	__tun_detach(tfile, clean);
434 	rtnl_unlock();
435 }
436 
437 static void tun_detach_all(struct net_device *dev)
438 {
439 	struct tun_struct *tun = netdev_priv(dev);
440 	struct tun_file *tfile;
441 	int i, n = tun->numqueues;
442 
443 	for (i = 0; i < n; i++) {
444 		tfile = rcu_dereference_protected(tun->tfiles[i],
445 						  lockdep_rtnl_is_held());
446 		BUG_ON(!tfile);
447 		wake_up_all(&tfile->wq.wait);
448 		rcu_assign_pointer(tfile->tun, NULL);
449 		--tun->numqueues;
450 	}
451 	BUG_ON(tun->numqueues != 0);
452 
453 	synchronize_net();
454 	for (i = 0; i < n; i++) {
455 		tfile = rcu_dereference_protected(tun->tfiles[i],
456 						  lockdep_rtnl_is_held());
457 		/* Drop read queue */
458 		skb_queue_purge(&tfile->sk.sk_receive_queue);
459 		sock_put(&tfile->sk);
460 	}
461 }
462 
463 static int tun_attach(struct tun_struct *tun, struct file *file)
464 {
465 	struct tun_file *tfile = file->private_data;
466 	int err;
467 
468 	err = -EINVAL;
469 	if (rcu_dereference_protected(tfile->tun, lockdep_rtnl_is_held()))
470 		goto out;
471 
472 	err = -EBUSY;
473 	if (!(tun->flags & TUN_TAP_MQ) && tun->numqueues == 1)
474 		goto out;
475 
476 	err = -E2BIG;
477 	if (tun->numqueues == MAX_TAP_QUEUES)
478 		goto out;
479 
480 	err = 0;
481 
482 	/* Re-attach the filter to presist device */
483 	if (tun->filter_attached == true) {
484 		err = sk_attach_filter(&tun->fprog, tfile->socket.sk);
485 		if (!err)
486 			goto out;
487 	}
488 	tfile->queue_index = tun->numqueues;
489 	rcu_assign_pointer(tfile->tun, tun);
490 	rcu_assign_pointer(tun->tfiles[tun->numqueues], tfile);
491 	sock_hold(&tfile->sk);
492 	tun->numqueues++;
493 
494 	tun_set_real_num_queues(tun);
495 
496 	/* device is allowed to go away first, so no need to hold extra
497 	 * refcnt.
498 	 */
499 
500 out:
501 	return err;
502 }
503 
504 static struct tun_struct *__tun_get(struct tun_file *tfile)
505 {
506 	struct tun_struct *tun;
507 
508 	rcu_read_lock();
509 	tun = rcu_dereference(tfile->tun);
510 	if (tun)
511 		dev_hold(tun->dev);
512 	rcu_read_unlock();
513 
514 	return tun;
515 }
516 
517 static struct tun_struct *tun_get(struct file *file)
518 {
519 	return __tun_get(file->private_data);
520 }
521 
522 static void tun_put(struct tun_struct *tun)
523 {
524 	dev_put(tun->dev);
525 }
526 
527 /* TAP filtering */
528 static void addr_hash_set(u32 *mask, const u8 *addr)
529 {
530 	int n = ether_crc(ETH_ALEN, addr) >> 26;
531 	mask[n >> 5] |= (1 << (n & 31));
532 }
533 
534 static unsigned int addr_hash_test(const u32 *mask, const u8 *addr)
535 {
536 	int n = ether_crc(ETH_ALEN, addr) >> 26;
537 	return mask[n >> 5] & (1 << (n & 31));
538 }
539 
540 static int update_filter(struct tap_filter *filter, void __user *arg)
541 {
542 	struct { u8 u[ETH_ALEN]; } *addr;
543 	struct tun_filter uf;
544 	int err, alen, n, nexact;
545 
546 	if (copy_from_user(&uf, arg, sizeof(uf)))
547 		return -EFAULT;
548 
549 	if (!uf.count) {
550 		/* Disabled */
551 		filter->count = 0;
552 		return 0;
553 	}
554 
555 	alen = ETH_ALEN * uf.count;
556 	addr = kmalloc(alen, GFP_KERNEL);
557 	if (!addr)
558 		return -ENOMEM;
559 
560 	if (copy_from_user(addr, arg + sizeof(uf), alen)) {
561 		err = -EFAULT;
562 		goto done;
563 	}
564 
565 	/* The filter is updated without holding any locks. Which is
566 	 * perfectly safe. We disable it first and in the worst
567 	 * case we'll accept a few undesired packets. */
568 	filter->count = 0;
569 	wmb();
570 
571 	/* Use first set of addresses as an exact filter */
572 	for (n = 0; n < uf.count && n < FLT_EXACT_COUNT; n++)
573 		memcpy(filter->addr[n], addr[n].u, ETH_ALEN);
574 
575 	nexact = n;
576 
577 	/* Remaining multicast addresses are hashed,
578 	 * unicast will leave the filter disabled. */
579 	memset(filter->mask, 0, sizeof(filter->mask));
580 	for (; n < uf.count; n++) {
581 		if (!is_multicast_ether_addr(addr[n].u)) {
582 			err = 0; /* no filter */
583 			goto done;
584 		}
585 		addr_hash_set(filter->mask, addr[n].u);
586 	}
587 
588 	/* For ALLMULTI just set the mask to all ones.
589 	 * This overrides the mask populated above. */
590 	if ((uf.flags & TUN_FLT_ALLMULTI))
591 		memset(filter->mask, ~0, sizeof(filter->mask));
592 
593 	/* Now enable the filter */
594 	wmb();
595 	filter->count = nexact;
596 
597 	/* Return the number of exact filters */
598 	err = nexact;
599 
600 done:
601 	kfree(addr);
602 	return err;
603 }
604 
605 /* Returns: 0 - drop, !=0 - accept */
606 static int run_filter(struct tap_filter *filter, const struct sk_buff *skb)
607 {
608 	/* Cannot use eth_hdr(skb) here because skb_mac_hdr() is incorrect
609 	 * at this point. */
610 	struct ethhdr *eh = (struct ethhdr *) skb->data;
611 	int i;
612 
613 	/* Exact match */
614 	for (i = 0; i < filter->count; i++)
615 		if (ether_addr_equal(eh->h_dest, filter->addr[i]))
616 			return 1;
617 
618 	/* Inexact match (multicast only) */
619 	if (is_multicast_ether_addr(eh->h_dest))
620 		return addr_hash_test(filter->mask, eh->h_dest);
621 
622 	return 0;
623 }
624 
625 /*
626  * Checks whether the packet is accepted or not.
627  * Returns: 0 - drop, !=0 - accept
628  */
629 static int check_filter(struct tap_filter *filter, const struct sk_buff *skb)
630 {
631 	if (!filter->count)
632 		return 1;
633 
634 	return run_filter(filter, skb);
635 }
636 
637 /* Network device part of the driver */
638 
639 static const struct ethtool_ops tun_ethtool_ops;
640 
641 /* Net device detach from fd. */
642 static void tun_net_uninit(struct net_device *dev)
643 {
644 	tun_detach_all(dev);
645 }
646 
647 /* Net device open. */
648 static int tun_net_open(struct net_device *dev)
649 {
650 	netif_tx_start_all_queues(dev);
651 	return 0;
652 }
653 
654 /* Net device close. */
655 static int tun_net_close(struct net_device *dev)
656 {
657 	netif_tx_stop_all_queues(dev);
658 	return 0;
659 }
660 
661 /* Net device start xmit */
662 static netdev_tx_t tun_net_xmit(struct sk_buff *skb, struct net_device *dev)
663 {
664 	struct tun_struct *tun = netdev_priv(dev);
665 	int txq = skb->queue_mapping;
666 	struct tun_file *tfile;
667 
668 	rcu_read_lock();
669 	tfile = rcu_dereference(tun->tfiles[txq]);
670 
671 	/* Drop packet if interface is not attached */
672 	if (txq >= tun->numqueues)
673 		goto drop;
674 
675 	tun_debug(KERN_INFO, tun, "tun_net_xmit %d\n", skb->len);
676 
677 	BUG_ON(!tfile);
678 
679 	/* Drop if the filter does not like it.
680 	 * This is a noop if the filter is disabled.
681 	 * Filter can be enabled only for the TAP devices. */
682 	if (!check_filter(&tun->txflt, skb))
683 		goto drop;
684 
685 	if (tfile->socket.sk->sk_filter &&
686 	    sk_filter(tfile->socket.sk, skb))
687 		goto drop;
688 
689 	/* Limit the number of packets queued by dividing txq length with the
690 	 * number of queues.
691 	 */
692 	if (skb_queue_len(&tfile->socket.sk->sk_receive_queue)
693 			  >= dev->tx_queue_len / tun->numqueues)
694 		goto drop;
695 
696 	/* Orphan the skb - required as we might hang on to it
697 	 * for indefinite time. */
698 	if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC)))
699 		goto drop;
700 	skb_orphan(skb);
701 
702 	/* Enqueue packet */
703 	skb_queue_tail(&tfile->socket.sk->sk_receive_queue, skb);
704 
705 	/* Notify and wake up reader process */
706 	if (tfile->flags & TUN_FASYNC)
707 		kill_fasync(&tfile->fasync, SIGIO, POLL_IN);
708 	wake_up_interruptible_poll(&tfile->wq.wait, POLLIN |
709 				   POLLRDNORM | POLLRDBAND);
710 
711 	rcu_read_unlock();
712 	return NETDEV_TX_OK;
713 
714 drop:
715 	dev->stats.tx_dropped++;
716 	skb_tx_error(skb);
717 	kfree_skb(skb);
718 	rcu_read_unlock();
719 	return NETDEV_TX_OK;
720 }
721 
722 static void tun_net_mclist(struct net_device *dev)
723 {
724 	/*
725 	 * This callback is supposed to deal with mc filter in
726 	 * _rx_ path and has nothing to do with the _tx_ path.
727 	 * In rx path we always accept everything userspace gives us.
728 	 */
729 }
730 
731 #define MIN_MTU 68
732 #define MAX_MTU 65535
733 
734 static int
735 tun_net_change_mtu(struct net_device *dev, int new_mtu)
736 {
737 	if (new_mtu < MIN_MTU || new_mtu + dev->hard_header_len > MAX_MTU)
738 		return -EINVAL;
739 	dev->mtu = new_mtu;
740 	return 0;
741 }
742 
743 static netdev_features_t tun_net_fix_features(struct net_device *dev,
744 	netdev_features_t features)
745 {
746 	struct tun_struct *tun = netdev_priv(dev);
747 
748 	return (features & tun->set_features) | (features & ~TUN_USER_FEATURES);
749 }
750 #ifdef CONFIG_NET_POLL_CONTROLLER
751 static void tun_poll_controller(struct net_device *dev)
752 {
753 	/*
754 	 * Tun only receives frames when:
755 	 * 1) the char device endpoint gets data from user space
756 	 * 2) the tun socket gets a sendmsg call from user space
757 	 * Since both of those are syncronous operations, we are guaranteed
758 	 * never to have pending data when we poll for it
759 	 * so theres nothing to do here but return.
760 	 * We need this though so netpoll recognizes us as an interface that
761 	 * supports polling, which enables bridge devices in virt setups to
762 	 * still use netconsole
763 	 */
764 	return;
765 }
766 #endif
767 static const struct net_device_ops tun_netdev_ops = {
768 	.ndo_uninit		= tun_net_uninit,
769 	.ndo_open		= tun_net_open,
770 	.ndo_stop		= tun_net_close,
771 	.ndo_start_xmit		= tun_net_xmit,
772 	.ndo_change_mtu		= tun_net_change_mtu,
773 	.ndo_fix_features	= tun_net_fix_features,
774 	.ndo_select_queue	= tun_select_queue,
775 #ifdef CONFIG_NET_POLL_CONTROLLER
776 	.ndo_poll_controller	= tun_poll_controller,
777 #endif
778 };
779 
780 static const struct net_device_ops tap_netdev_ops = {
781 	.ndo_uninit		= tun_net_uninit,
782 	.ndo_open		= tun_net_open,
783 	.ndo_stop		= tun_net_close,
784 	.ndo_start_xmit		= tun_net_xmit,
785 	.ndo_change_mtu		= tun_net_change_mtu,
786 	.ndo_fix_features	= tun_net_fix_features,
787 	.ndo_set_rx_mode	= tun_net_mclist,
788 	.ndo_set_mac_address	= eth_mac_addr,
789 	.ndo_validate_addr	= eth_validate_addr,
790 	.ndo_select_queue	= tun_select_queue,
791 #ifdef CONFIG_NET_POLL_CONTROLLER
792 	.ndo_poll_controller	= tun_poll_controller,
793 #endif
794 };
795 
796 static int tun_flow_init(struct tun_struct *tun)
797 {
798 	int i;
799 
800 	tun->flow_cache = kmem_cache_create("tun_flow_cache",
801 					    sizeof(struct tun_flow_entry), 0, 0,
802 					    NULL);
803 	if (!tun->flow_cache)
804 		return -ENOMEM;
805 
806 	for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++)
807 		INIT_HLIST_HEAD(&tun->flows[i]);
808 
809 	tun->ageing_time = TUN_FLOW_EXPIRE;
810 	setup_timer(&tun->flow_gc_timer, tun_flow_cleanup, (unsigned long)tun);
811 	mod_timer(&tun->flow_gc_timer,
812 		  round_jiffies_up(jiffies + tun->ageing_time));
813 
814 	return 0;
815 }
816 
817 static void tun_flow_uninit(struct tun_struct *tun)
818 {
819 	del_timer_sync(&tun->flow_gc_timer);
820 	tun_flow_flush(tun);
821 
822 	/* Wait for completion of call_rcu()'s */
823 	rcu_barrier();
824 	kmem_cache_destroy(tun->flow_cache);
825 }
826 
827 /* Initialize net device. */
828 static void tun_net_init(struct net_device *dev)
829 {
830 	struct tun_struct *tun = netdev_priv(dev);
831 
832 	switch (tun->flags & TUN_TYPE_MASK) {
833 	case TUN_TUN_DEV:
834 		dev->netdev_ops = &tun_netdev_ops;
835 
836 		/* Point-to-Point TUN Device */
837 		dev->hard_header_len = 0;
838 		dev->addr_len = 0;
839 		dev->mtu = 1500;
840 
841 		/* Zero header length */
842 		dev->type = ARPHRD_NONE;
843 		dev->flags = IFF_POINTOPOINT | IFF_NOARP | IFF_MULTICAST;
844 		dev->tx_queue_len = TUN_READQ_SIZE;  /* We prefer our own queue length */
845 		break;
846 
847 	case TUN_TAP_DEV:
848 		dev->netdev_ops = &tap_netdev_ops;
849 		/* Ethernet TAP Device */
850 		ether_setup(dev);
851 		dev->priv_flags &= ~IFF_TX_SKB_SHARING;
852 		dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
853 
854 		eth_hw_addr_random(dev);
855 
856 		dev->tx_queue_len = TUN_READQ_SIZE;  /* We prefer our own queue length */
857 		break;
858 	}
859 }
860 
861 /* Character device part */
862 
863 /* Poll */
864 static unsigned int tun_chr_poll(struct file *file, poll_table *wait)
865 {
866 	struct tun_file *tfile = file->private_data;
867 	struct tun_struct *tun = __tun_get(tfile);
868 	struct sock *sk;
869 	unsigned int mask = 0;
870 
871 	if (!tun)
872 		return POLLERR;
873 
874 	sk = tfile->socket.sk;
875 
876 	tun_debug(KERN_INFO, tun, "tun_chr_poll\n");
877 
878 	poll_wait(file, &tfile->wq.wait, wait);
879 
880 	if (!skb_queue_empty(&sk->sk_receive_queue))
881 		mask |= POLLIN | POLLRDNORM;
882 
883 	if (sock_writeable(sk) ||
884 	    (!test_and_set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags) &&
885 	     sock_writeable(sk)))
886 		mask |= POLLOUT | POLLWRNORM;
887 
888 	if (tun->dev->reg_state != NETREG_REGISTERED)
889 		mask = POLLERR;
890 
891 	tun_put(tun);
892 	return mask;
893 }
894 
895 /* prepad is the amount to reserve at front.  len is length after that.
896  * linear is a hint as to how much to copy (usually headers). */
897 static struct sk_buff *tun_alloc_skb(struct tun_file *tfile,
898 				     size_t prepad, size_t len,
899 				     size_t linear, int noblock)
900 {
901 	struct sock *sk = tfile->socket.sk;
902 	struct sk_buff *skb;
903 	int err;
904 
905 	/* Under a page?  Don't bother with paged skb. */
906 	if (prepad + len < PAGE_SIZE || !linear)
907 		linear = len;
908 
909 	skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
910 				   &err);
911 	if (!skb)
912 		return ERR_PTR(err);
913 
914 	skb_reserve(skb, prepad);
915 	skb_put(skb, linear);
916 	skb->data_len = len - linear;
917 	skb->len += len - linear;
918 
919 	return skb;
920 }
921 
922 /* set skb frags from iovec, this can move to core network code for reuse */
923 static int zerocopy_sg_from_iovec(struct sk_buff *skb, const struct iovec *from,
924 				  int offset, size_t count)
925 {
926 	int len = iov_length(from, count) - offset;
927 	int copy = skb_headlen(skb);
928 	int size, offset1 = 0;
929 	int i = 0;
930 
931 	/* Skip over from offset */
932 	while (count && (offset >= from->iov_len)) {
933 		offset -= from->iov_len;
934 		++from;
935 		--count;
936 	}
937 
938 	/* copy up to skb headlen */
939 	while (count && (copy > 0)) {
940 		size = min_t(unsigned int, copy, from->iov_len - offset);
941 		if (copy_from_user(skb->data + offset1, from->iov_base + offset,
942 				   size))
943 			return -EFAULT;
944 		if (copy > size) {
945 			++from;
946 			--count;
947 			offset = 0;
948 		} else
949 			offset += size;
950 		copy -= size;
951 		offset1 += size;
952 	}
953 
954 	if (len == offset1)
955 		return 0;
956 
957 	while (count--) {
958 		struct page *page[MAX_SKB_FRAGS];
959 		int num_pages;
960 		unsigned long base;
961 		unsigned long truesize;
962 
963 		len = from->iov_len - offset;
964 		if (!len) {
965 			offset = 0;
966 			++from;
967 			continue;
968 		}
969 		base = (unsigned long)from->iov_base + offset;
970 		size = ((base & ~PAGE_MASK) + len + ~PAGE_MASK) >> PAGE_SHIFT;
971 		if (i + size > MAX_SKB_FRAGS)
972 			return -EMSGSIZE;
973 		num_pages = get_user_pages_fast(base, size, 0, &page[i]);
974 		if (num_pages != size) {
975 			for (i = 0; i < num_pages; i++)
976 				put_page(page[i]);
977 			return -EFAULT;
978 		}
979 		truesize = size * PAGE_SIZE;
980 		skb->data_len += len;
981 		skb->len += len;
982 		skb->truesize += truesize;
983 		atomic_add(truesize, &skb->sk->sk_wmem_alloc);
984 		while (len) {
985 			int off = base & ~PAGE_MASK;
986 			int size = min_t(int, len, PAGE_SIZE - off);
987 			__skb_fill_page_desc(skb, i, page[i], off, size);
988 			skb_shinfo(skb)->nr_frags++;
989 			/* increase sk_wmem_alloc */
990 			base += size;
991 			len -= size;
992 			i++;
993 		}
994 		offset = 0;
995 		++from;
996 	}
997 	return 0;
998 }
999 
1000 /* Get packet from user space buffer */
1001 static ssize_t tun_get_user(struct tun_struct *tun, struct tun_file *tfile,
1002 			    void *msg_control, const struct iovec *iv,
1003 			    size_t total_len, size_t count, int noblock)
1004 {
1005 	struct tun_pi pi = { 0, cpu_to_be16(ETH_P_IP) };
1006 	struct sk_buff *skb;
1007 	size_t len = total_len, align = NET_SKB_PAD;
1008 	struct virtio_net_hdr gso = { 0 };
1009 	int offset = 0;
1010 	int copylen;
1011 	bool zerocopy = false;
1012 	int err;
1013 
1014 	if (!(tun->flags & TUN_NO_PI)) {
1015 		if ((len -= sizeof(pi)) > total_len)
1016 			return -EINVAL;
1017 
1018 		if (memcpy_fromiovecend((void *)&pi, iv, 0, sizeof(pi)))
1019 			return -EFAULT;
1020 		offset += sizeof(pi);
1021 	}
1022 
1023 	if (tun->flags & TUN_VNET_HDR) {
1024 		if ((len -= tun->vnet_hdr_sz) > total_len)
1025 			return -EINVAL;
1026 
1027 		if (memcpy_fromiovecend((void *)&gso, iv, offset, sizeof(gso)))
1028 			return -EFAULT;
1029 
1030 		if ((gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
1031 		    gso.csum_start + gso.csum_offset + 2 > gso.hdr_len)
1032 			gso.hdr_len = gso.csum_start + gso.csum_offset + 2;
1033 
1034 		if (gso.hdr_len > len)
1035 			return -EINVAL;
1036 		offset += tun->vnet_hdr_sz;
1037 	}
1038 
1039 	if ((tun->flags & TUN_TYPE_MASK) == TUN_TAP_DEV) {
1040 		align += NET_IP_ALIGN;
1041 		if (unlikely(len < ETH_HLEN ||
1042 			     (gso.hdr_len && gso.hdr_len < ETH_HLEN)))
1043 			return -EINVAL;
1044 	}
1045 
1046 	if (msg_control)
1047 		zerocopy = true;
1048 
1049 	if (zerocopy) {
1050 		/* Userspace may produce vectors with count greater than
1051 		 * MAX_SKB_FRAGS, so we need to linearize parts of the skb
1052 		 * to let the rest of data to be fit in the frags.
1053 		 */
1054 		if (count > MAX_SKB_FRAGS) {
1055 			copylen = iov_length(iv, count - MAX_SKB_FRAGS);
1056 			if (copylen < offset)
1057 				copylen = 0;
1058 			else
1059 				copylen -= offset;
1060 		} else
1061 				copylen = 0;
1062 		/* There are 256 bytes to be copied in skb, so there is enough
1063 		 * room for skb expand head in case it is used.
1064 		 * The rest of the buffer is mapped from userspace.
1065 		 */
1066 		if (copylen < gso.hdr_len)
1067 			copylen = gso.hdr_len;
1068 		if (!copylen)
1069 			copylen = GOODCOPY_LEN;
1070 	} else
1071 		copylen = len;
1072 
1073 	skb = tun_alloc_skb(tfile, align, copylen, gso.hdr_len, noblock);
1074 	if (IS_ERR(skb)) {
1075 		if (PTR_ERR(skb) != -EAGAIN)
1076 			tun->dev->stats.rx_dropped++;
1077 		return PTR_ERR(skb);
1078 	}
1079 
1080 	if (zerocopy)
1081 		err = zerocopy_sg_from_iovec(skb, iv, offset, count);
1082 	else
1083 		err = skb_copy_datagram_from_iovec(skb, 0, iv, offset, len);
1084 
1085 	if (err) {
1086 		tun->dev->stats.rx_dropped++;
1087 		kfree_skb(skb);
1088 		return -EFAULT;
1089 	}
1090 
1091 	if (gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
1092 		if (!skb_partial_csum_set(skb, gso.csum_start,
1093 					  gso.csum_offset)) {
1094 			tun->dev->stats.rx_frame_errors++;
1095 			kfree_skb(skb);
1096 			return -EINVAL;
1097 		}
1098 	}
1099 
1100 	switch (tun->flags & TUN_TYPE_MASK) {
1101 	case TUN_TUN_DEV:
1102 		if (tun->flags & TUN_NO_PI) {
1103 			switch (skb->data[0] & 0xf0) {
1104 			case 0x40:
1105 				pi.proto = htons(ETH_P_IP);
1106 				break;
1107 			case 0x60:
1108 				pi.proto = htons(ETH_P_IPV6);
1109 				break;
1110 			default:
1111 				tun->dev->stats.rx_dropped++;
1112 				kfree_skb(skb);
1113 				return -EINVAL;
1114 			}
1115 		}
1116 
1117 		skb_reset_mac_header(skb);
1118 		skb->protocol = pi.proto;
1119 		skb->dev = tun->dev;
1120 		break;
1121 	case TUN_TAP_DEV:
1122 		skb->protocol = eth_type_trans(skb, tun->dev);
1123 		break;
1124 	}
1125 
1126 	if (gso.gso_type != VIRTIO_NET_HDR_GSO_NONE) {
1127 		pr_debug("GSO!\n");
1128 		switch (gso.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
1129 		case VIRTIO_NET_HDR_GSO_TCPV4:
1130 			skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
1131 			break;
1132 		case VIRTIO_NET_HDR_GSO_TCPV6:
1133 			skb_shinfo(skb)->gso_type = SKB_GSO_TCPV6;
1134 			break;
1135 		case VIRTIO_NET_HDR_GSO_UDP:
1136 			skb_shinfo(skb)->gso_type = SKB_GSO_UDP;
1137 			break;
1138 		default:
1139 			tun->dev->stats.rx_frame_errors++;
1140 			kfree_skb(skb);
1141 			return -EINVAL;
1142 		}
1143 
1144 		if (gso.gso_type & VIRTIO_NET_HDR_GSO_ECN)
1145 			skb_shinfo(skb)->gso_type |= SKB_GSO_TCP_ECN;
1146 
1147 		skb_shinfo(skb)->gso_size = gso.gso_size;
1148 		if (skb_shinfo(skb)->gso_size == 0) {
1149 			tun->dev->stats.rx_frame_errors++;
1150 			kfree_skb(skb);
1151 			return -EINVAL;
1152 		}
1153 
1154 		/* Header must be checked, and gso_segs computed. */
1155 		skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
1156 		skb_shinfo(skb)->gso_segs = 0;
1157 	}
1158 
1159 	/* copy skb_ubuf_info for callback when skb has no error */
1160 	if (zerocopy) {
1161 		skb_shinfo(skb)->destructor_arg = msg_control;
1162 		skb_shinfo(skb)->tx_flags |= SKBTX_DEV_ZEROCOPY;
1163 	}
1164 
1165 	netif_rx_ni(skb);
1166 
1167 	tun->dev->stats.rx_packets++;
1168 	tun->dev->stats.rx_bytes += len;
1169 
1170 	tun_flow_update(tun, skb, tfile->queue_index);
1171 	return total_len;
1172 }
1173 
1174 static ssize_t tun_chr_aio_write(struct kiocb *iocb, const struct iovec *iv,
1175 			      unsigned long count, loff_t pos)
1176 {
1177 	struct file *file = iocb->ki_filp;
1178 	struct tun_struct *tun = tun_get(file);
1179 	struct tun_file *tfile = file->private_data;
1180 	ssize_t result;
1181 
1182 	if (!tun)
1183 		return -EBADFD;
1184 
1185 	tun_debug(KERN_INFO, tun, "tun_chr_write %ld\n", count);
1186 
1187 	result = tun_get_user(tun, tfile, NULL, iv, iov_length(iv, count),
1188 			      count, file->f_flags & O_NONBLOCK);
1189 
1190 	tun_put(tun);
1191 	return result;
1192 }
1193 
1194 /* Put packet to the user space buffer */
1195 static ssize_t tun_put_user(struct tun_struct *tun,
1196 			    struct tun_file *tfile,
1197 			    struct sk_buff *skb,
1198 			    const struct iovec *iv, int len)
1199 {
1200 	struct tun_pi pi = { 0, skb->protocol };
1201 	ssize_t total = 0;
1202 
1203 	if (!(tun->flags & TUN_NO_PI)) {
1204 		if ((len -= sizeof(pi)) < 0)
1205 			return -EINVAL;
1206 
1207 		if (len < skb->len) {
1208 			/* Packet will be striped */
1209 			pi.flags |= TUN_PKT_STRIP;
1210 		}
1211 
1212 		if (memcpy_toiovecend(iv, (void *) &pi, 0, sizeof(pi)))
1213 			return -EFAULT;
1214 		total += sizeof(pi);
1215 	}
1216 
1217 	if (tun->flags & TUN_VNET_HDR) {
1218 		struct virtio_net_hdr gso = { 0 }; /* no info leak */
1219 		if ((len -= tun->vnet_hdr_sz) < 0)
1220 			return -EINVAL;
1221 
1222 		if (skb_is_gso(skb)) {
1223 			struct skb_shared_info *sinfo = skb_shinfo(skb);
1224 
1225 			/* This is a hint as to how much should be linear. */
1226 			gso.hdr_len = skb_headlen(skb);
1227 			gso.gso_size = sinfo->gso_size;
1228 			if (sinfo->gso_type & SKB_GSO_TCPV4)
1229 				gso.gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
1230 			else if (sinfo->gso_type & SKB_GSO_TCPV6)
1231 				gso.gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
1232 			else if (sinfo->gso_type & SKB_GSO_UDP)
1233 				gso.gso_type = VIRTIO_NET_HDR_GSO_UDP;
1234 			else {
1235 				pr_err("unexpected GSO type: "
1236 				       "0x%x, gso_size %d, hdr_len %d\n",
1237 				       sinfo->gso_type, gso.gso_size,
1238 				       gso.hdr_len);
1239 				print_hex_dump(KERN_ERR, "tun: ",
1240 					       DUMP_PREFIX_NONE,
1241 					       16, 1, skb->head,
1242 					       min((int)gso.hdr_len, 64), true);
1243 				WARN_ON_ONCE(1);
1244 				return -EINVAL;
1245 			}
1246 			if (sinfo->gso_type & SKB_GSO_TCP_ECN)
1247 				gso.gso_type |= VIRTIO_NET_HDR_GSO_ECN;
1248 		} else
1249 			gso.gso_type = VIRTIO_NET_HDR_GSO_NONE;
1250 
1251 		if (skb->ip_summed == CHECKSUM_PARTIAL) {
1252 			gso.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
1253 			gso.csum_start = skb_checksum_start_offset(skb);
1254 			gso.csum_offset = skb->csum_offset;
1255 		} else if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
1256 			gso.flags = VIRTIO_NET_HDR_F_DATA_VALID;
1257 		} /* else everything is zero */
1258 
1259 		if (unlikely(memcpy_toiovecend(iv, (void *)&gso, total,
1260 					       sizeof(gso))))
1261 			return -EFAULT;
1262 		total += tun->vnet_hdr_sz;
1263 	}
1264 
1265 	len = min_t(int, skb->len, len);
1266 
1267 	skb_copy_datagram_const_iovec(skb, 0, iv, total, len);
1268 	total += skb->len;
1269 
1270 	tun->dev->stats.tx_packets++;
1271 	tun->dev->stats.tx_bytes += len;
1272 
1273 	return total;
1274 }
1275 
1276 static ssize_t tun_do_read(struct tun_struct *tun, struct tun_file *tfile,
1277 			   struct kiocb *iocb, const struct iovec *iv,
1278 			   ssize_t len, int noblock)
1279 {
1280 	DECLARE_WAITQUEUE(wait, current);
1281 	struct sk_buff *skb;
1282 	ssize_t ret = 0;
1283 
1284 	tun_debug(KERN_INFO, tun, "tun_do_read\n");
1285 
1286 	if (unlikely(!noblock))
1287 		add_wait_queue(&tfile->wq.wait, &wait);
1288 	while (len) {
1289 		current->state = TASK_INTERRUPTIBLE;
1290 
1291 		/* Read frames from the queue */
1292 		if (!(skb = skb_dequeue(&tfile->socket.sk->sk_receive_queue))) {
1293 			if (noblock) {
1294 				ret = -EAGAIN;
1295 				break;
1296 			}
1297 			if (signal_pending(current)) {
1298 				ret = -ERESTARTSYS;
1299 				break;
1300 			}
1301 			if (tun->dev->reg_state != NETREG_REGISTERED) {
1302 				ret = -EIO;
1303 				break;
1304 			}
1305 
1306 			/* Nothing to read, let's sleep */
1307 			schedule();
1308 			continue;
1309 		}
1310 
1311 		ret = tun_put_user(tun, tfile, skb, iv, len);
1312 		kfree_skb(skb);
1313 		break;
1314 	}
1315 
1316 	current->state = TASK_RUNNING;
1317 	if (unlikely(!noblock))
1318 		remove_wait_queue(&tfile->wq.wait, &wait);
1319 
1320 	return ret;
1321 }
1322 
1323 static ssize_t tun_chr_aio_read(struct kiocb *iocb, const struct iovec *iv,
1324 			    unsigned long count, loff_t pos)
1325 {
1326 	struct file *file = iocb->ki_filp;
1327 	struct tun_file *tfile = file->private_data;
1328 	struct tun_struct *tun = __tun_get(tfile);
1329 	ssize_t len, ret;
1330 
1331 	if (!tun)
1332 		return -EBADFD;
1333 	len = iov_length(iv, count);
1334 	if (len < 0) {
1335 		ret = -EINVAL;
1336 		goto out;
1337 	}
1338 
1339 	ret = tun_do_read(tun, tfile, iocb, iv, len,
1340 			  file->f_flags & O_NONBLOCK);
1341 	ret = min_t(ssize_t, ret, len);
1342 out:
1343 	tun_put(tun);
1344 	return ret;
1345 }
1346 
1347 static void tun_free_netdev(struct net_device *dev)
1348 {
1349 	struct tun_struct *tun = netdev_priv(dev);
1350 
1351 	tun_flow_uninit(tun);
1352 	free_netdev(dev);
1353 }
1354 
1355 static void tun_setup(struct net_device *dev)
1356 {
1357 	struct tun_struct *tun = netdev_priv(dev);
1358 
1359 	tun->owner = INVALID_UID;
1360 	tun->group = INVALID_GID;
1361 
1362 	dev->ethtool_ops = &tun_ethtool_ops;
1363 	dev->destructor = tun_free_netdev;
1364 }
1365 
1366 /* Trivial set of netlink ops to allow deleting tun or tap
1367  * device with netlink.
1368  */
1369 static int tun_validate(struct nlattr *tb[], struct nlattr *data[])
1370 {
1371 	return -EINVAL;
1372 }
1373 
1374 static struct rtnl_link_ops tun_link_ops __read_mostly = {
1375 	.kind		= DRV_NAME,
1376 	.priv_size	= sizeof(struct tun_struct),
1377 	.setup		= tun_setup,
1378 	.validate	= tun_validate,
1379 };
1380 
1381 static void tun_sock_write_space(struct sock *sk)
1382 {
1383 	struct tun_file *tfile;
1384 	wait_queue_head_t *wqueue;
1385 
1386 	if (!sock_writeable(sk))
1387 		return;
1388 
1389 	if (!test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags))
1390 		return;
1391 
1392 	wqueue = sk_sleep(sk);
1393 	if (wqueue && waitqueue_active(wqueue))
1394 		wake_up_interruptible_sync_poll(wqueue, POLLOUT |
1395 						POLLWRNORM | POLLWRBAND);
1396 
1397 	tfile = container_of(sk, struct tun_file, sk);
1398 	kill_fasync(&tfile->fasync, SIGIO, POLL_OUT);
1399 }
1400 
1401 static int tun_sendmsg(struct kiocb *iocb, struct socket *sock,
1402 		       struct msghdr *m, size_t total_len)
1403 {
1404 	int ret;
1405 	struct tun_file *tfile = container_of(sock, struct tun_file, socket);
1406 	struct tun_struct *tun = __tun_get(tfile);
1407 
1408 	if (!tun)
1409 		return -EBADFD;
1410 	ret = tun_get_user(tun, tfile, m->msg_control, m->msg_iov, total_len,
1411 			   m->msg_iovlen, m->msg_flags & MSG_DONTWAIT);
1412 	tun_put(tun);
1413 	return ret;
1414 }
1415 
1416 
1417 static int tun_recvmsg(struct kiocb *iocb, struct socket *sock,
1418 		       struct msghdr *m, size_t total_len,
1419 		       int flags)
1420 {
1421 	struct tun_file *tfile = container_of(sock, struct tun_file, socket);
1422 	struct tun_struct *tun = __tun_get(tfile);
1423 	int ret;
1424 
1425 	if (!tun)
1426 		return -EBADFD;
1427 
1428 	if (flags & ~(MSG_DONTWAIT|MSG_TRUNC))
1429 		return -EINVAL;
1430 	ret = tun_do_read(tun, tfile, iocb, m->msg_iov, total_len,
1431 			  flags & MSG_DONTWAIT);
1432 	if (ret > total_len) {
1433 		m->msg_flags |= MSG_TRUNC;
1434 		ret = flags & MSG_TRUNC ? ret : total_len;
1435 	}
1436 	tun_put(tun);
1437 	return ret;
1438 }
1439 
1440 static int tun_release(struct socket *sock)
1441 {
1442 	if (sock->sk)
1443 		sock_put(sock->sk);
1444 	return 0;
1445 }
1446 
1447 /* Ops structure to mimic raw sockets with tun */
1448 static const struct proto_ops tun_socket_ops = {
1449 	.sendmsg = tun_sendmsg,
1450 	.recvmsg = tun_recvmsg,
1451 	.release = tun_release,
1452 };
1453 
1454 static struct proto tun_proto = {
1455 	.name		= "tun",
1456 	.owner		= THIS_MODULE,
1457 	.obj_size	= sizeof(struct tun_file),
1458 };
1459 
1460 static int tun_flags(struct tun_struct *tun)
1461 {
1462 	int flags = 0;
1463 
1464 	if (tun->flags & TUN_TUN_DEV)
1465 		flags |= IFF_TUN;
1466 	else
1467 		flags |= IFF_TAP;
1468 
1469 	if (tun->flags & TUN_NO_PI)
1470 		flags |= IFF_NO_PI;
1471 
1472 	/* This flag has no real effect.  We track the value for backwards
1473 	 * compatibility.
1474 	 */
1475 	if (tun->flags & TUN_ONE_QUEUE)
1476 		flags |= IFF_ONE_QUEUE;
1477 
1478 	if (tun->flags & TUN_VNET_HDR)
1479 		flags |= IFF_VNET_HDR;
1480 
1481 	if (tun->flags & TUN_TAP_MQ)
1482 		flags |= IFF_MULTI_QUEUE;
1483 
1484 	return flags;
1485 }
1486 
1487 static ssize_t tun_show_flags(struct device *dev, struct device_attribute *attr,
1488 			      char *buf)
1489 {
1490 	struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1491 	return sprintf(buf, "0x%x\n", tun_flags(tun));
1492 }
1493 
1494 static ssize_t tun_show_owner(struct device *dev, struct device_attribute *attr,
1495 			      char *buf)
1496 {
1497 	struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1498 	return uid_valid(tun->owner)?
1499 		sprintf(buf, "%u\n",
1500 			from_kuid_munged(current_user_ns(), tun->owner)):
1501 		sprintf(buf, "-1\n");
1502 }
1503 
1504 static ssize_t tun_show_group(struct device *dev, struct device_attribute *attr,
1505 			      char *buf)
1506 {
1507 	struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1508 	return gid_valid(tun->group) ?
1509 		sprintf(buf, "%u\n",
1510 			from_kgid_munged(current_user_ns(), tun->group)):
1511 		sprintf(buf, "-1\n");
1512 }
1513 
1514 static DEVICE_ATTR(tun_flags, 0444, tun_show_flags, NULL);
1515 static DEVICE_ATTR(owner, 0444, tun_show_owner, NULL);
1516 static DEVICE_ATTR(group, 0444, tun_show_group, NULL);
1517 
1518 static int tun_set_iff(struct net *net, struct file *file, struct ifreq *ifr)
1519 {
1520 	struct tun_struct *tun;
1521 	struct tun_file *tfile = file->private_data;
1522 	struct net_device *dev;
1523 	int err;
1524 
1525 	dev = __dev_get_by_name(net, ifr->ifr_name);
1526 	if (dev) {
1527 		if (ifr->ifr_flags & IFF_TUN_EXCL)
1528 			return -EBUSY;
1529 		if ((ifr->ifr_flags & IFF_TUN) && dev->netdev_ops == &tun_netdev_ops)
1530 			tun = netdev_priv(dev);
1531 		else if ((ifr->ifr_flags & IFF_TAP) && dev->netdev_ops == &tap_netdev_ops)
1532 			tun = netdev_priv(dev);
1533 		else
1534 			return -EINVAL;
1535 
1536 		if (tun_not_capable(tun))
1537 			return -EPERM;
1538 		err = security_tun_dev_attach(tfile->socket.sk);
1539 		if (err < 0)
1540 			return err;
1541 
1542 		err = tun_attach(tun, file);
1543 		if (err < 0)
1544 			return err;
1545 	}
1546 	else {
1547 		char *name;
1548 		unsigned long flags = 0;
1549 
1550 		if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
1551 			return -EPERM;
1552 		err = security_tun_dev_create();
1553 		if (err < 0)
1554 			return err;
1555 
1556 		/* Set dev type */
1557 		if (ifr->ifr_flags & IFF_TUN) {
1558 			/* TUN device */
1559 			flags |= TUN_TUN_DEV;
1560 			name = "tun%d";
1561 		} else if (ifr->ifr_flags & IFF_TAP) {
1562 			/* TAP device */
1563 			flags |= TUN_TAP_DEV;
1564 			name = "tap%d";
1565 		} else
1566 			return -EINVAL;
1567 
1568 		if (*ifr->ifr_name)
1569 			name = ifr->ifr_name;
1570 
1571 		dev = alloc_netdev_mqs(sizeof(struct tun_struct), name,
1572 				       tun_setup,
1573 				       MAX_TAP_QUEUES, MAX_TAP_QUEUES);
1574 		if (!dev)
1575 			return -ENOMEM;
1576 
1577 		dev_net_set(dev, net);
1578 		dev->rtnl_link_ops = &tun_link_ops;
1579 
1580 		tun = netdev_priv(dev);
1581 		tun->dev = dev;
1582 		tun->flags = flags;
1583 		tun->txflt.count = 0;
1584 		tun->vnet_hdr_sz = sizeof(struct virtio_net_hdr);
1585 
1586 		tun->filter_attached = false;
1587 		tun->sndbuf = tfile->socket.sk->sk_sndbuf;
1588 
1589 		spin_lock_init(&tun->lock);
1590 
1591 		security_tun_dev_post_create(&tfile->sk);
1592 
1593 		tun_net_init(dev);
1594 
1595 		err = tun_flow_init(tun);
1596 		if (err < 0)
1597 			goto err_free_dev;
1598 
1599 		dev->hw_features = NETIF_F_SG | NETIF_F_FRAGLIST |
1600 			TUN_USER_FEATURES;
1601 		dev->features = dev->hw_features;
1602 
1603 		err = tun_attach(tun, file);
1604 		if (err < 0)
1605 			goto err_free_dev;
1606 
1607 		err = register_netdevice(tun->dev);
1608 		if (err < 0)
1609 			goto err_free_dev;
1610 
1611 		if (device_create_file(&tun->dev->dev, &dev_attr_tun_flags) ||
1612 		    device_create_file(&tun->dev->dev, &dev_attr_owner) ||
1613 		    device_create_file(&tun->dev->dev, &dev_attr_group))
1614 			pr_err("Failed to create tun sysfs files\n");
1615 
1616 		netif_carrier_on(tun->dev);
1617 	}
1618 
1619 	tun_debug(KERN_INFO, tun, "tun_set_iff\n");
1620 
1621 	if (ifr->ifr_flags & IFF_NO_PI)
1622 		tun->flags |= TUN_NO_PI;
1623 	else
1624 		tun->flags &= ~TUN_NO_PI;
1625 
1626 	/* This flag has no real effect.  We track the value for backwards
1627 	 * compatibility.
1628 	 */
1629 	if (ifr->ifr_flags & IFF_ONE_QUEUE)
1630 		tun->flags |= TUN_ONE_QUEUE;
1631 	else
1632 		tun->flags &= ~TUN_ONE_QUEUE;
1633 
1634 	if (ifr->ifr_flags & IFF_VNET_HDR)
1635 		tun->flags |= TUN_VNET_HDR;
1636 	else
1637 		tun->flags &= ~TUN_VNET_HDR;
1638 
1639 	if (ifr->ifr_flags & IFF_MULTI_QUEUE)
1640 		tun->flags |= TUN_TAP_MQ;
1641 	else
1642 		tun->flags &= ~TUN_TAP_MQ;
1643 
1644 	/* Make sure persistent devices do not get stuck in
1645 	 * xoff state.
1646 	 */
1647 	if (netif_running(tun->dev))
1648 		netif_tx_wake_all_queues(tun->dev);
1649 
1650 	strcpy(ifr->ifr_name, tun->dev->name);
1651 	return 0;
1652 
1653  err_free_dev:
1654 	free_netdev(dev);
1655 	return err;
1656 }
1657 
1658 static void tun_get_iff(struct net *net, struct tun_struct *tun,
1659 		       struct ifreq *ifr)
1660 {
1661 	tun_debug(KERN_INFO, tun, "tun_get_iff\n");
1662 
1663 	strcpy(ifr->ifr_name, tun->dev->name);
1664 
1665 	ifr->ifr_flags = tun_flags(tun);
1666 
1667 }
1668 
1669 /* This is like a cut-down ethtool ops, except done via tun fd so no
1670  * privs required. */
1671 static int set_offload(struct tun_struct *tun, unsigned long arg)
1672 {
1673 	netdev_features_t features = 0;
1674 
1675 	if (arg & TUN_F_CSUM) {
1676 		features |= NETIF_F_HW_CSUM;
1677 		arg &= ~TUN_F_CSUM;
1678 
1679 		if (arg & (TUN_F_TSO4|TUN_F_TSO6)) {
1680 			if (arg & TUN_F_TSO_ECN) {
1681 				features |= NETIF_F_TSO_ECN;
1682 				arg &= ~TUN_F_TSO_ECN;
1683 			}
1684 			if (arg & TUN_F_TSO4)
1685 				features |= NETIF_F_TSO;
1686 			if (arg & TUN_F_TSO6)
1687 				features |= NETIF_F_TSO6;
1688 			arg &= ~(TUN_F_TSO4|TUN_F_TSO6);
1689 		}
1690 
1691 		if (arg & TUN_F_UFO) {
1692 			features |= NETIF_F_UFO;
1693 			arg &= ~TUN_F_UFO;
1694 		}
1695 	}
1696 
1697 	/* This gives the user a way to test for new features in future by
1698 	 * trying to set them. */
1699 	if (arg)
1700 		return -EINVAL;
1701 
1702 	tun->set_features = features;
1703 	netdev_update_features(tun->dev);
1704 
1705 	return 0;
1706 }
1707 
1708 static void tun_detach_filter(struct tun_struct *tun, int n)
1709 {
1710 	int i;
1711 	struct tun_file *tfile;
1712 
1713 	for (i = 0; i < n; i++) {
1714 		tfile = rcu_dereference_protected(tun->tfiles[i],
1715 						  lockdep_rtnl_is_held());
1716 		sk_detach_filter(tfile->socket.sk);
1717 	}
1718 
1719 	tun->filter_attached = false;
1720 }
1721 
1722 static int tun_attach_filter(struct tun_struct *tun)
1723 {
1724 	int i, ret = 0;
1725 	struct tun_file *tfile;
1726 
1727 	for (i = 0; i < tun->numqueues; i++) {
1728 		tfile = rcu_dereference_protected(tun->tfiles[i],
1729 						  lockdep_rtnl_is_held());
1730 		ret = sk_attach_filter(&tun->fprog, tfile->socket.sk);
1731 		if (ret) {
1732 			tun_detach_filter(tun, i);
1733 			return ret;
1734 		}
1735 	}
1736 
1737 	tun->filter_attached = true;
1738 	return ret;
1739 }
1740 
1741 static void tun_set_sndbuf(struct tun_struct *tun)
1742 {
1743 	struct tun_file *tfile;
1744 	int i;
1745 
1746 	for (i = 0; i < tun->numqueues; i++) {
1747 		tfile = rcu_dereference_protected(tun->tfiles[i],
1748 						lockdep_rtnl_is_held());
1749 		tfile->socket.sk->sk_sndbuf = tun->sndbuf;
1750 	}
1751 }
1752 
1753 static int tun_set_queue(struct file *file, struct ifreq *ifr)
1754 {
1755 	struct tun_file *tfile = file->private_data;
1756 	struct tun_struct *tun;
1757 	struct net_device *dev;
1758 	int ret = 0;
1759 
1760 	rtnl_lock();
1761 
1762 	if (ifr->ifr_flags & IFF_ATTACH_QUEUE) {
1763 		dev = __dev_get_by_name(tfile->net, ifr->ifr_name);
1764 		if (!dev) {
1765 			ret = -EINVAL;
1766 			goto unlock;
1767 		}
1768 
1769 		tun = netdev_priv(dev);
1770 		if (dev->netdev_ops != &tap_netdev_ops &&
1771 			dev->netdev_ops != &tun_netdev_ops)
1772 			ret = -EINVAL;
1773 		else if (tun_not_capable(tun))
1774 			ret = -EPERM;
1775 		else
1776 			ret = tun_attach(tun, file);
1777 	} else if (ifr->ifr_flags & IFF_DETACH_QUEUE)
1778 		__tun_detach(tfile, false);
1779 	else
1780 		ret = -EINVAL;
1781 
1782 unlock:
1783 	rtnl_unlock();
1784 	return ret;
1785 }
1786 
1787 static long __tun_chr_ioctl(struct file *file, unsigned int cmd,
1788 			    unsigned long arg, int ifreq_len)
1789 {
1790 	struct tun_file *tfile = file->private_data;
1791 	struct tun_struct *tun;
1792 	void __user* argp = (void __user*)arg;
1793 	struct ifreq ifr;
1794 	kuid_t owner;
1795 	kgid_t group;
1796 	int sndbuf;
1797 	int vnet_hdr_sz;
1798 	int ret;
1799 
1800 	if (cmd == TUNSETIFF || cmd == TUNSETQUEUE || _IOC_TYPE(cmd) == 0x89) {
1801 		if (copy_from_user(&ifr, argp, ifreq_len))
1802 			return -EFAULT;
1803 	} else {
1804 		memset(&ifr, 0, sizeof(ifr));
1805 	}
1806 	if (cmd == TUNGETFEATURES) {
1807 		/* Currently this just means: "what IFF flags are valid?".
1808 		 * This is needed because we never checked for invalid flags on
1809 		 * TUNSETIFF. */
1810 		return put_user(IFF_TUN | IFF_TAP | IFF_NO_PI | IFF_ONE_QUEUE |
1811 				IFF_VNET_HDR | IFF_MULTI_QUEUE,
1812 				(unsigned int __user*)argp);
1813 	} else if (cmd == TUNSETQUEUE)
1814 		return tun_set_queue(file, &ifr);
1815 
1816 	ret = 0;
1817 	rtnl_lock();
1818 
1819 	tun = __tun_get(tfile);
1820 	if (cmd == TUNSETIFF && !tun) {
1821 		ifr.ifr_name[IFNAMSIZ-1] = '\0';
1822 
1823 		ret = tun_set_iff(tfile->net, file, &ifr);
1824 
1825 		if (ret)
1826 			goto unlock;
1827 
1828 		if (copy_to_user(argp, &ifr, ifreq_len))
1829 			ret = -EFAULT;
1830 		goto unlock;
1831 	}
1832 
1833 	ret = -EBADFD;
1834 	if (!tun)
1835 		goto unlock;
1836 
1837 	tun_debug(KERN_INFO, tun, "tun_chr_ioctl cmd %u\n", cmd);
1838 
1839 	ret = 0;
1840 	switch (cmd) {
1841 	case TUNGETIFF:
1842 		tun_get_iff(current->nsproxy->net_ns, tun, &ifr);
1843 
1844 		if (copy_to_user(argp, &ifr, ifreq_len))
1845 			ret = -EFAULT;
1846 		break;
1847 
1848 	case TUNSETNOCSUM:
1849 		/* Disable/Enable checksum */
1850 
1851 		/* [unimplemented] */
1852 		tun_debug(KERN_INFO, tun, "ignored: set checksum %s\n",
1853 			  arg ? "disabled" : "enabled");
1854 		break;
1855 
1856 	case TUNSETPERSIST:
1857 		/* Disable/Enable persist mode. Keep an extra reference to the
1858 		 * module to prevent the module being unprobed.
1859 		 */
1860 		if (arg) {
1861 			tun->flags |= TUN_PERSIST;
1862 			__module_get(THIS_MODULE);
1863 		} else {
1864 			tun->flags &= ~TUN_PERSIST;
1865 			module_put(THIS_MODULE);
1866 		}
1867 
1868 		tun_debug(KERN_INFO, tun, "persist %s\n",
1869 			  arg ? "enabled" : "disabled");
1870 		break;
1871 
1872 	case TUNSETOWNER:
1873 		/* Set owner of the device */
1874 		owner = make_kuid(current_user_ns(), arg);
1875 		if (!uid_valid(owner)) {
1876 			ret = -EINVAL;
1877 			break;
1878 		}
1879 		tun->owner = owner;
1880 		tun_debug(KERN_INFO, tun, "owner set to %u\n",
1881 			  from_kuid(&init_user_ns, tun->owner));
1882 		break;
1883 
1884 	case TUNSETGROUP:
1885 		/* Set group of the device */
1886 		group = make_kgid(current_user_ns(), arg);
1887 		if (!gid_valid(group)) {
1888 			ret = -EINVAL;
1889 			break;
1890 		}
1891 		tun->group = group;
1892 		tun_debug(KERN_INFO, tun, "group set to %u\n",
1893 			  from_kgid(&init_user_ns, tun->group));
1894 		break;
1895 
1896 	case TUNSETLINK:
1897 		/* Only allow setting the type when the interface is down */
1898 		if (tun->dev->flags & IFF_UP) {
1899 			tun_debug(KERN_INFO, tun,
1900 				  "Linktype set failed because interface is up\n");
1901 			ret = -EBUSY;
1902 		} else {
1903 			tun->dev->type = (int) arg;
1904 			tun_debug(KERN_INFO, tun, "linktype set to %d\n",
1905 				  tun->dev->type);
1906 			ret = 0;
1907 		}
1908 		break;
1909 
1910 #ifdef TUN_DEBUG
1911 	case TUNSETDEBUG:
1912 		tun->debug = arg;
1913 		break;
1914 #endif
1915 	case TUNSETOFFLOAD:
1916 		ret = set_offload(tun, arg);
1917 		break;
1918 
1919 	case TUNSETTXFILTER:
1920 		/* Can be set only for TAPs */
1921 		ret = -EINVAL;
1922 		if ((tun->flags & TUN_TYPE_MASK) != TUN_TAP_DEV)
1923 			break;
1924 		ret = update_filter(&tun->txflt, (void __user *)arg);
1925 		break;
1926 
1927 	case SIOCGIFHWADDR:
1928 		/* Get hw address */
1929 		memcpy(ifr.ifr_hwaddr.sa_data, tun->dev->dev_addr, ETH_ALEN);
1930 		ifr.ifr_hwaddr.sa_family = tun->dev->type;
1931 		if (copy_to_user(argp, &ifr, ifreq_len))
1932 			ret = -EFAULT;
1933 		break;
1934 
1935 	case SIOCSIFHWADDR:
1936 		/* Set hw address */
1937 		tun_debug(KERN_DEBUG, tun, "set hw address: %pM\n",
1938 			  ifr.ifr_hwaddr.sa_data);
1939 
1940 		ret = dev_set_mac_address(tun->dev, &ifr.ifr_hwaddr);
1941 		break;
1942 
1943 	case TUNGETSNDBUF:
1944 		sndbuf = tfile->socket.sk->sk_sndbuf;
1945 		if (copy_to_user(argp, &sndbuf, sizeof(sndbuf)))
1946 			ret = -EFAULT;
1947 		break;
1948 
1949 	case TUNSETSNDBUF:
1950 		if (copy_from_user(&sndbuf, argp, sizeof(sndbuf))) {
1951 			ret = -EFAULT;
1952 			break;
1953 		}
1954 
1955 		tun->sndbuf = sndbuf;
1956 		tun_set_sndbuf(tun);
1957 		break;
1958 
1959 	case TUNGETVNETHDRSZ:
1960 		vnet_hdr_sz = tun->vnet_hdr_sz;
1961 		if (copy_to_user(argp, &vnet_hdr_sz, sizeof(vnet_hdr_sz)))
1962 			ret = -EFAULT;
1963 		break;
1964 
1965 	case TUNSETVNETHDRSZ:
1966 		if (copy_from_user(&vnet_hdr_sz, argp, sizeof(vnet_hdr_sz))) {
1967 			ret = -EFAULT;
1968 			break;
1969 		}
1970 		if (vnet_hdr_sz < (int)sizeof(struct virtio_net_hdr)) {
1971 			ret = -EINVAL;
1972 			break;
1973 		}
1974 
1975 		tun->vnet_hdr_sz = vnet_hdr_sz;
1976 		break;
1977 
1978 	case TUNATTACHFILTER:
1979 		/* Can be set only for TAPs */
1980 		ret = -EINVAL;
1981 		if ((tun->flags & TUN_TYPE_MASK) != TUN_TAP_DEV)
1982 			break;
1983 		ret = -EFAULT;
1984 		if (copy_from_user(&tun->fprog, argp, sizeof(tun->fprog)))
1985 			break;
1986 
1987 		ret = tun_attach_filter(tun);
1988 		break;
1989 
1990 	case TUNDETACHFILTER:
1991 		/* Can be set only for TAPs */
1992 		ret = -EINVAL;
1993 		if ((tun->flags & TUN_TYPE_MASK) != TUN_TAP_DEV)
1994 			break;
1995 		ret = 0;
1996 		tun_detach_filter(tun, tun->numqueues);
1997 		break;
1998 
1999 	default:
2000 		ret = -EINVAL;
2001 		break;
2002 	}
2003 
2004 unlock:
2005 	rtnl_unlock();
2006 	if (tun)
2007 		tun_put(tun);
2008 	return ret;
2009 }
2010 
2011 static long tun_chr_ioctl(struct file *file,
2012 			  unsigned int cmd, unsigned long arg)
2013 {
2014 	return __tun_chr_ioctl(file, cmd, arg, sizeof (struct ifreq));
2015 }
2016 
2017 #ifdef CONFIG_COMPAT
2018 static long tun_chr_compat_ioctl(struct file *file,
2019 			 unsigned int cmd, unsigned long arg)
2020 {
2021 	switch (cmd) {
2022 	case TUNSETIFF:
2023 	case TUNGETIFF:
2024 	case TUNSETTXFILTER:
2025 	case TUNGETSNDBUF:
2026 	case TUNSETSNDBUF:
2027 	case SIOCGIFHWADDR:
2028 	case SIOCSIFHWADDR:
2029 		arg = (unsigned long)compat_ptr(arg);
2030 		break;
2031 	default:
2032 		arg = (compat_ulong_t)arg;
2033 		break;
2034 	}
2035 
2036 	/*
2037 	 * compat_ifreq is shorter than ifreq, so we must not access beyond
2038 	 * the end of that structure. All fields that are used in this
2039 	 * driver are compatible though, we don't need to convert the
2040 	 * contents.
2041 	 */
2042 	return __tun_chr_ioctl(file, cmd, arg, sizeof(struct compat_ifreq));
2043 }
2044 #endif /* CONFIG_COMPAT */
2045 
2046 static int tun_chr_fasync(int fd, struct file *file, int on)
2047 {
2048 	struct tun_file *tfile = file->private_data;
2049 	int ret;
2050 
2051 	if ((ret = fasync_helper(fd, file, on, &tfile->fasync)) < 0)
2052 		goto out;
2053 
2054 	if (on) {
2055 		ret = __f_setown(file, task_pid(current), PIDTYPE_PID, 0);
2056 		if (ret)
2057 			goto out;
2058 		tfile->flags |= TUN_FASYNC;
2059 	} else
2060 		tfile->flags &= ~TUN_FASYNC;
2061 	ret = 0;
2062 out:
2063 	return ret;
2064 }
2065 
2066 static int tun_chr_open(struct inode *inode, struct file * file)
2067 {
2068 	struct tun_file *tfile;
2069 
2070 	DBG1(KERN_INFO, "tunX: tun_chr_open\n");
2071 
2072 	tfile = (struct tun_file *)sk_alloc(&init_net, AF_UNSPEC, GFP_KERNEL,
2073 					    &tun_proto);
2074 	if (!tfile)
2075 		return -ENOMEM;
2076 	rcu_assign_pointer(tfile->tun, NULL);
2077 	tfile->net = get_net(current->nsproxy->net_ns);
2078 	tfile->flags = 0;
2079 
2080 	rcu_assign_pointer(tfile->socket.wq, &tfile->wq);
2081 	init_waitqueue_head(&tfile->wq.wait);
2082 
2083 	tfile->socket.file = file;
2084 	tfile->socket.ops = &tun_socket_ops;
2085 
2086 	sock_init_data(&tfile->socket, &tfile->sk);
2087 	sk_change_net(&tfile->sk, tfile->net);
2088 
2089 	tfile->sk.sk_write_space = tun_sock_write_space;
2090 	tfile->sk.sk_sndbuf = INT_MAX;
2091 
2092 	file->private_data = tfile;
2093 	set_bit(SOCK_EXTERNALLY_ALLOCATED, &tfile->socket.flags);
2094 
2095 	return 0;
2096 }
2097 
2098 static int tun_chr_close(struct inode *inode, struct file *file)
2099 {
2100 	struct tun_file *tfile = file->private_data;
2101 	struct net *net = tfile->net;
2102 
2103 	tun_detach(tfile, true);
2104 	put_net(net);
2105 
2106 	return 0;
2107 }
2108 
2109 static const struct file_operations tun_fops = {
2110 	.owner	= THIS_MODULE,
2111 	.llseek = no_llseek,
2112 	.read  = do_sync_read,
2113 	.aio_read  = tun_chr_aio_read,
2114 	.write = do_sync_write,
2115 	.aio_write = tun_chr_aio_write,
2116 	.poll	= tun_chr_poll,
2117 	.unlocked_ioctl	= tun_chr_ioctl,
2118 #ifdef CONFIG_COMPAT
2119 	.compat_ioctl = tun_chr_compat_ioctl,
2120 #endif
2121 	.open	= tun_chr_open,
2122 	.release = tun_chr_close,
2123 	.fasync = tun_chr_fasync
2124 };
2125 
2126 static struct miscdevice tun_miscdev = {
2127 	.minor = TUN_MINOR,
2128 	.name = "tun",
2129 	.nodename = "net/tun",
2130 	.fops = &tun_fops,
2131 };
2132 
2133 /* ethtool interface */
2134 
2135 static int tun_get_settings(struct net_device *dev, struct ethtool_cmd *cmd)
2136 {
2137 	cmd->supported		= 0;
2138 	cmd->advertising	= 0;
2139 	ethtool_cmd_speed_set(cmd, SPEED_10);
2140 	cmd->duplex		= DUPLEX_FULL;
2141 	cmd->port		= PORT_TP;
2142 	cmd->phy_address	= 0;
2143 	cmd->transceiver	= XCVR_INTERNAL;
2144 	cmd->autoneg		= AUTONEG_DISABLE;
2145 	cmd->maxtxpkt		= 0;
2146 	cmd->maxrxpkt		= 0;
2147 	return 0;
2148 }
2149 
2150 static void tun_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
2151 {
2152 	struct tun_struct *tun = netdev_priv(dev);
2153 
2154 	strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
2155 	strlcpy(info->version, DRV_VERSION, sizeof(info->version));
2156 
2157 	switch (tun->flags & TUN_TYPE_MASK) {
2158 	case TUN_TUN_DEV:
2159 		strlcpy(info->bus_info, "tun", sizeof(info->bus_info));
2160 		break;
2161 	case TUN_TAP_DEV:
2162 		strlcpy(info->bus_info, "tap", sizeof(info->bus_info));
2163 		break;
2164 	}
2165 }
2166 
2167 static u32 tun_get_msglevel(struct net_device *dev)
2168 {
2169 #ifdef TUN_DEBUG
2170 	struct tun_struct *tun = netdev_priv(dev);
2171 	return tun->debug;
2172 #else
2173 	return -EOPNOTSUPP;
2174 #endif
2175 }
2176 
2177 static void tun_set_msglevel(struct net_device *dev, u32 value)
2178 {
2179 #ifdef TUN_DEBUG
2180 	struct tun_struct *tun = netdev_priv(dev);
2181 	tun->debug = value;
2182 #endif
2183 }
2184 
2185 static const struct ethtool_ops tun_ethtool_ops = {
2186 	.get_settings	= tun_get_settings,
2187 	.get_drvinfo	= tun_get_drvinfo,
2188 	.get_msglevel	= tun_get_msglevel,
2189 	.set_msglevel	= tun_set_msglevel,
2190 	.get_link	= ethtool_op_get_link,
2191 };
2192 
2193 
2194 static int __init tun_init(void)
2195 {
2196 	int ret = 0;
2197 
2198 	pr_info("%s, %s\n", DRV_DESCRIPTION, DRV_VERSION);
2199 	pr_info("%s\n", DRV_COPYRIGHT);
2200 
2201 	ret = rtnl_link_register(&tun_link_ops);
2202 	if (ret) {
2203 		pr_err("Can't register link_ops\n");
2204 		goto err_linkops;
2205 	}
2206 
2207 	ret = misc_register(&tun_miscdev);
2208 	if (ret) {
2209 		pr_err("Can't register misc device %d\n", TUN_MINOR);
2210 		goto err_misc;
2211 	}
2212 	return  0;
2213 err_misc:
2214 	rtnl_link_unregister(&tun_link_ops);
2215 err_linkops:
2216 	return ret;
2217 }
2218 
2219 static void tun_cleanup(void)
2220 {
2221 	misc_deregister(&tun_miscdev);
2222 	rtnl_link_unregister(&tun_link_ops);
2223 }
2224 
2225 /* Get an underlying socket object from tun file.  Returns error unless file is
2226  * attached to a device.  The returned object works like a packet socket, it
2227  * can be used for sock_sendmsg/sock_recvmsg.  The caller is responsible for
2228  * holding a reference to the file for as long as the socket is in use. */
2229 struct socket *tun_get_socket(struct file *file)
2230 {
2231 	struct tun_file *tfile;
2232 	if (file->f_op != &tun_fops)
2233 		return ERR_PTR(-EINVAL);
2234 	tfile = file->private_data;
2235 	if (!tfile)
2236 		return ERR_PTR(-EBADFD);
2237 	return &tfile->socket;
2238 }
2239 EXPORT_SYMBOL_GPL(tun_get_socket);
2240 
2241 module_init(tun_init);
2242 module_exit(tun_cleanup);
2243 MODULE_DESCRIPTION(DRV_DESCRIPTION);
2244 MODULE_AUTHOR(DRV_COPYRIGHT);
2245 MODULE_LICENSE("GPL");
2246 MODULE_ALIAS_MISCDEV(TUN_MINOR);
2247 MODULE_ALIAS("devname:net/tun");
2248