xref: /openbmc/linux/drivers/net/tun.c (revision 5d097109257c03a71845729f8db6b5770c4bbedc)
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 
853 		eth_hw_addr_random(dev);
854 
855 		dev->tx_queue_len = TUN_READQ_SIZE;  /* We prefer our own queue length */
856 		break;
857 	}
858 }
859 
860 /* Character device part */
861 
862 /* Poll */
863 static unsigned int tun_chr_poll(struct file *file, poll_table *wait)
864 {
865 	struct tun_file *tfile = file->private_data;
866 	struct tun_struct *tun = __tun_get(tfile);
867 	struct sock *sk;
868 	unsigned int mask = 0;
869 
870 	if (!tun)
871 		return POLLERR;
872 
873 	sk = tfile->socket.sk;
874 
875 	tun_debug(KERN_INFO, tun, "tun_chr_poll\n");
876 
877 	poll_wait(file, &tfile->wq.wait, wait);
878 
879 	if (!skb_queue_empty(&sk->sk_receive_queue))
880 		mask |= POLLIN | POLLRDNORM;
881 
882 	if (sock_writeable(sk) ||
883 	    (!test_and_set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags) &&
884 	     sock_writeable(sk)))
885 		mask |= POLLOUT | POLLWRNORM;
886 
887 	if (tun->dev->reg_state != NETREG_REGISTERED)
888 		mask = POLLERR;
889 
890 	tun_put(tun);
891 	return mask;
892 }
893 
894 /* prepad is the amount to reserve at front.  len is length after that.
895  * linear is a hint as to how much to copy (usually headers). */
896 static struct sk_buff *tun_alloc_skb(struct tun_file *tfile,
897 				     size_t prepad, size_t len,
898 				     size_t linear, int noblock)
899 {
900 	struct sock *sk = tfile->socket.sk;
901 	struct sk_buff *skb;
902 	int err;
903 
904 	/* Under a page?  Don't bother with paged skb. */
905 	if (prepad + len < PAGE_SIZE || !linear)
906 		linear = len;
907 
908 	skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
909 				   &err);
910 	if (!skb)
911 		return ERR_PTR(err);
912 
913 	skb_reserve(skb, prepad);
914 	skb_put(skb, linear);
915 	skb->data_len = len - linear;
916 	skb->len += len - linear;
917 
918 	return skb;
919 }
920 
921 /* set skb frags from iovec, this can move to core network code for reuse */
922 static int zerocopy_sg_from_iovec(struct sk_buff *skb, const struct iovec *from,
923 				  int offset, size_t count)
924 {
925 	int len = iov_length(from, count) - offset;
926 	int copy = skb_headlen(skb);
927 	int size, offset1 = 0;
928 	int i = 0;
929 
930 	/* Skip over from offset */
931 	while (count && (offset >= from->iov_len)) {
932 		offset -= from->iov_len;
933 		++from;
934 		--count;
935 	}
936 
937 	/* copy up to skb headlen */
938 	while (count && (copy > 0)) {
939 		size = min_t(unsigned int, copy, from->iov_len - offset);
940 		if (copy_from_user(skb->data + offset1, from->iov_base + offset,
941 				   size))
942 			return -EFAULT;
943 		if (copy > size) {
944 			++from;
945 			--count;
946 			offset = 0;
947 		} else
948 			offset += size;
949 		copy -= size;
950 		offset1 += size;
951 	}
952 
953 	if (len == offset1)
954 		return 0;
955 
956 	while (count--) {
957 		struct page *page[MAX_SKB_FRAGS];
958 		int num_pages;
959 		unsigned long base;
960 		unsigned long truesize;
961 
962 		len = from->iov_len - offset;
963 		if (!len) {
964 			offset = 0;
965 			++from;
966 			continue;
967 		}
968 		base = (unsigned long)from->iov_base + offset;
969 		size = ((base & ~PAGE_MASK) + len + ~PAGE_MASK) >> PAGE_SHIFT;
970 		if (i + size > MAX_SKB_FRAGS)
971 			return -EMSGSIZE;
972 		num_pages = get_user_pages_fast(base, size, 0, &page[i]);
973 		if (num_pages != size) {
974 			for (i = 0; i < num_pages; i++)
975 				put_page(page[i]);
976 			return -EFAULT;
977 		}
978 		truesize = size * PAGE_SIZE;
979 		skb->data_len += len;
980 		skb->len += len;
981 		skb->truesize += truesize;
982 		atomic_add(truesize, &skb->sk->sk_wmem_alloc);
983 		while (len) {
984 			int off = base & ~PAGE_MASK;
985 			int size = min_t(int, len, PAGE_SIZE - off);
986 			__skb_fill_page_desc(skb, i, page[i], off, size);
987 			skb_shinfo(skb)->nr_frags++;
988 			/* increase sk_wmem_alloc */
989 			base += size;
990 			len -= size;
991 			i++;
992 		}
993 		offset = 0;
994 		++from;
995 	}
996 	return 0;
997 }
998 
999 /* Get packet from user space buffer */
1000 static ssize_t tun_get_user(struct tun_struct *tun, struct tun_file *tfile,
1001 			    void *msg_control, const struct iovec *iv,
1002 			    size_t total_len, size_t count, int noblock)
1003 {
1004 	struct tun_pi pi = { 0, cpu_to_be16(ETH_P_IP) };
1005 	struct sk_buff *skb;
1006 	size_t len = total_len, align = NET_SKB_PAD;
1007 	struct virtio_net_hdr gso = { 0 };
1008 	int offset = 0;
1009 	int copylen;
1010 	bool zerocopy = false;
1011 	int err;
1012 
1013 	if (!(tun->flags & TUN_NO_PI)) {
1014 		if ((len -= sizeof(pi)) > total_len)
1015 			return -EINVAL;
1016 
1017 		if (memcpy_fromiovecend((void *)&pi, iv, 0, sizeof(pi)))
1018 			return -EFAULT;
1019 		offset += sizeof(pi);
1020 	}
1021 
1022 	if (tun->flags & TUN_VNET_HDR) {
1023 		if ((len -= tun->vnet_hdr_sz) > total_len)
1024 			return -EINVAL;
1025 
1026 		if (memcpy_fromiovecend((void *)&gso, iv, offset, sizeof(gso)))
1027 			return -EFAULT;
1028 
1029 		if ((gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
1030 		    gso.csum_start + gso.csum_offset + 2 > gso.hdr_len)
1031 			gso.hdr_len = gso.csum_start + gso.csum_offset + 2;
1032 
1033 		if (gso.hdr_len > len)
1034 			return -EINVAL;
1035 		offset += tun->vnet_hdr_sz;
1036 	}
1037 
1038 	if ((tun->flags & TUN_TYPE_MASK) == TUN_TAP_DEV) {
1039 		align += NET_IP_ALIGN;
1040 		if (unlikely(len < ETH_HLEN ||
1041 			     (gso.hdr_len && gso.hdr_len < ETH_HLEN)))
1042 			return -EINVAL;
1043 	}
1044 
1045 	if (msg_control)
1046 		zerocopy = true;
1047 
1048 	if (zerocopy) {
1049 		/* Userspace may produce vectors with count greater than
1050 		 * MAX_SKB_FRAGS, so we need to linearize parts of the skb
1051 		 * to let the rest of data to be fit in the frags.
1052 		 */
1053 		if (count > MAX_SKB_FRAGS) {
1054 			copylen = iov_length(iv, count - MAX_SKB_FRAGS);
1055 			if (copylen < offset)
1056 				copylen = 0;
1057 			else
1058 				copylen -= offset;
1059 		} else
1060 				copylen = 0;
1061 		/* There are 256 bytes to be copied in skb, so there is enough
1062 		 * room for skb expand head in case it is used.
1063 		 * The rest of the buffer is mapped from userspace.
1064 		 */
1065 		if (copylen < gso.hdr_len)
1066 			copylen = gso.hdr_len;
1067 		if (!copylen)
1068 			copylen = GOODCOPY_LEN;
1069 	} else
1070 		copylen = len;
1071 
1072 	skb = tun_alloc_skb(tfile, align, copylen, gso.hdr_len, noblock);
1073 	if (IS_ERR(skb)) {
1074 		if (PTR_ERR(skb) != -EAGAIN)
1075 			tun->dev->stats.rx_dropped++;
1076 		return PTR_ERR(skb);
1077 	}
1078 
1079 	if (zerocopy)
1080 		err = zerocopy_sg_from_iovec(skb, iv, offset, count);
1081 	else
1082 		err = skb_copy_datagram_from_iovec(skb, 0, iv, offset, len);
1083 
1084 	if (err) {
1085 		tun->dev->stats.rx_dropped++;
1086 		kfree_skb(skb);
1087 		return -EFAULT;
1088 	}
1089 
1090 	if (gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
1091 		if (!skb_partial_csum_set(skb, gso.csum_start,
1092 					  gso.csum_offset)) {
1093 			tun->dev->stats.rx_frame_errors++;
1094 			kfree_skb(skb);
1095 			return -EINVAL;
1096 		}
1097 	}
1098 
1099 	switch (tun->flags & TUN_TYPE_MASK) {
1100 	case TUN_TUN_DEV:
1101 		if (tun->flags & TUN_NO_PI) {
1102 			switch (skb->data[0] & 0xf0) {
1103 			case 0x40:
1104 				pi.proto = htons(ETH_P_IP);
1105 				break;
1106 			case 0x60:
1107 				pi.proto = htons(ETH_P_IPV6);
1108 				break;
1109 			default:
1110 				tun->dev->stats.rx_dropped++;
1111 				kfree_skb(skb);
1112 				return -EINVAL;
1113 			}
1114 		}
1115 
1116 		skb_reset_mac_header(skb);
1117 		skb->protocol = pi.proto;
1118 		skb->dev = tun->dev;
1119 		break;
1120 	case TUN_TAP_DEV:
1121 		skb->protocol = eth_type_trans(skb, tun->dev);
1122 		break;
1123 	}
1124 
1125 	if (gso.gso_type != VIRTIO_NET_HDR_GSO_NONE) {
1126 		pr_debug("GSO!\n");
1127 		switch (gso.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
1128 		case VIRTIO_NET_HDR_GSO_TCPV4:
1129 			skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
1130 			break;
1131 		case VIRTIO_NET_HDR_GSO_TCPV6:
1132 			skb_shinfo(skb)->gso_type = SKB_GSO_TCPV6;
1133 			break;
1134 		case VIRTIO_NET_HDR_GSO_UDP:
1135 			skb_shinfo(skb)->gso_type = SKB_GSO_UDP;
1136 			break;
1137 		default:
1138 			tun->dev->stats.rx_frame_errors++;
1139 			kfree_skb(skb);
1140 			return -EINVAL;
1141 		}
1142 
1143 		if (gso.gso_type & VIRTIO_NET_HDR_GSO_ECN)
1144 			skb_shinfo(skb)->gso_type |= SKB_GSO_TCP_ECN;
1145 
1146 		skb_shinfo(skb)->gso_size = gso.gso_size;
1147 		if (skb_shinfo(skb)->gso_size == 0) {
1148 			tun->dev->stats.rx_frame_errors++;
1149 			kfree_skb(skb);
1150 			return -EINVAL;
1151 		}
1152 
1153 		/* Header must be checked, and gso_segs computed. */
1154 		skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
1155 		skb_shinfo(skb)->gso_segs = 0;
1156 	}
1157 
1158 	/* copy skb_ubuf_info for callback when skb has no error */
1159 	if (zerocopy) {
1160 		skb_shinfo(skb)->destructor_arg = msg_control;
1161 		skb_shinfo(skb)->tx_flags |= SKBTX_DEV_ZEROCOPY;
1162 	}
1163 
1164 	netif_rx_ni(skb);
1165 
1166 	tun->dev->stats.rx_packets++;
1167 	tun->dev->stats.rx_bytes += len;
1168 
1169 	tun_flow_update(tun, skb, tfile->queue_index);
1170 	return total_len;
1171 }
1172 
1173 static ssize_t tun_chr_aio_write(struct kiocb *iocb, const struct iovec *iv,
1174 			      unsigned long count, loff_t pos)
1175 {
1176 	struct file *file = iocb->ki_filp;
1177 	struct tun_struct *tun = tun_get(file);
1178 	struct tun_file *tfile = file->private_data;
1179 	ssize_t result;
1180 
1181 	if (!tun)
1182 		return -EBADFD;
1183 
1184 	tun_debug(KERN_INFO, tun, "tun_chr_write %ld\n", count);
1185 
1186 	result = tun_get_user(tun, tfile, NULL, iv, iov_length(iv, count),
1187 			      count, file->f_flags & O_NONBLOCK);
1188 
1189 	tun_put(tun);
1190 	return result;
1191 }
1192 
1193 /* Put packet to the user space buffer */
1194 static ssize_t tun_put_user(struct tun_struct *tun,
1195 			    struct tun_file *tfile,
1196 			    struct sk_buff *skb,
1197 			    const struct iovec *iv, int len)
1198 {
1199 	struct tun_pi pi = { 0, skb->protocol };
1200 	ssize_t total = 0;
1201 
1202 	if (!(tun->flags & TUN_NO_PI)) {
1203 		if ((len -= sizeof(pi)) < 0)
1204 			return -EINVAL;
1205 
1206 		if (len < skb->len) {
1207 			/* Packet will be striped */
1208 			pi.flags |= TUN_PKT_STRIP;
1209 		}
1210 
1211 		if (memcpy_toiovecend(iv, (void *) &pi, 0, sizeof(pi)))
1212 			return -EFAULT;
1213 		total += sizeof(pi);
1214 	}
1215 
1216 	if (tun->flags & TUN_VNET_HDR) {
1217 		struct virtio_net_hdr gso = { 0 }; /* no info leak */
1218 		if ((len -= tun->vnet_hdr_sz) < 0)
1219 			return -EINVAL;
1220 
1221 		if (skb_is_gso(skb)) {
1222 			struct skb_shared_info *sinfo = skb_shinfo(skb);
1223 
1224 			/* This is a hint as to how much should be linear. */
1225 			gso.hdr_len = skb_headlen(skb);
1226 			gso.gso_size = sinfo->gso_size;
1227 			if (sinfo->gso_type & SKB_GSO_TCPV4)
1228 				gso.gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
1229 			else if (sinfo->gso_type & SKB_GSO_TCPV6)
1230 				gso.gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
1231 			else if (sinfo->gso_type & SKB_GSO_UDP)
1232 				gso.gso_type = VIRTIO_NET_HDR_GSO_UDP;
1233 			else {
1234 				pr_err("unexpected GSO type: "
1235 				       "0x%x, gso_size %d, hdr_len %d\n",
1236 				       sinfo->gso_type, gso.gso_size,
1237 				       gso.hdr_len);
1238 				print_hex_dump(KERN_ERR, "tun: ",
1239 					       DUMP_PREFIX_NONE,
1240 					       16, 1, skb->head,
1241 					       min((int)gso.hdr_len, 64), true);
1242 				WARN_ON_ONCE(1);
1243 				return -EINVAL;
1244 			}
1245 			if (sinfo->gso_type & SKB_GSO_TCP_ECN)
1246 				gso.gso_type |= VIRTIO_NET_HDR_GSO_ECN;
1247 		} else
1248 			gso.gso_type = VIRTIO_NET_HDR_GSO_NONE;
1249 
1250 		if (skb->ip_summed == CHECKSUM_PARTIAL) {
1251 			gso.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
1252 			gso.csum_start = skb_checksum_start_offset(skb);
1253 			gso.csum_offset = skb->csum_offset;
1254 		} else if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
1255 			gso.flags = VIRTIO_NET_HDR_F_DATA_VALID;
1256 		} /* else everything is zero */
1257 
1258 		if (unlikely(memcpy_toiovecend(iv, (void *)&gso, total,
1259 					       sizeof(gso))))
1260 			return -EFAULT;
1261 		total += tun->vnet_hdr_sz;
1262 	}
1263 
1264 	len = min_t(int, skb->len, len);
1265 
1266 	skb_copy_datagram_const_iovec(skb, 0, iv, total, len);
1267 	total += skb->len;
1268 
1269 	tun->dev->stats.tx_packets++;
1270 	tun->dev->stats.tx_bytes += len;
1271 
1272 	return total;
1273 }
1274 
1275 static ssize_t tun_do_read(struct tun_struct *tun, struct tun_file *tfile,
1276 			   struct kiocb *iocb, const struct iovec *iv,
1277 			   ssize_t len, int noblock)
1278 {
1279 	DECLARE_WAITQUEUE(wait, current);
1280 	struct sk_buff *skb;
1281 	ssize_t ret = 0;
1282 
1283 	tun_debug(KERN_INFO, tun, "tun_do_read\n");
1284 
1285 	if (unlikely(!noblock))
1286 		add_wait_queue(&tfile->wq.wait, &wait);
1287 	while (len) {
1288 		current->state = TASK_INTERRUPTIBLE;
1289 
1290 		/* Read frames from the queue */
1291 		if (!(skb = skb_dequeue(&tfile->socket.sk->sk_receive_queue))) {
1292 			if (noblock) {
1293 				ret = -EAGAIN;
1294 				break;
1295 			}
1296 			if (signal_pending(current)) {
1297 				ret = -ERESTARTSYS;
1298 				break;
1299 			}
1300 			if (tun->dev->reg_state != NETREG_REGISTERED) {
1301 				ret = -EIO;
1302 				break;
1303 			}
1304 
1305 			/* Nothing to read, let's sleep */
1306 			schedule();
1307 			continue;
1308 		}
1309 
1310 		ret = tun_put_user(tun, tfile, skb, iv, len);
1311 		kfree_skb(skb);
1312 		break;
1313 	}
1314 
1315 	current->state = TASK_RUNNING;
1316 	if (unlikely(!noblock))
1317 		remove_wait_queue(&tfile->wq.wait, &wait);
1318 
1319 	return ret;
1320 }
1321 
1322 static ssize_t tun_chr_aio_read(struct kiocb *iocb, const struct iovec *iv,
1323 			    unsigned long count, loff_t pos)
1324 {
1325 	struct file *file = iocb->ki_filp;
1326 	struct tun_file *tfile = file->private_data;
1327 	struct tun_struct *tun = __tun_get(tfile);
1328 	ssize_t len, ret;
1329 
1330 	if (!tun)
1331 		return -EBADFD;
1332 	len = iov_length(iv, count);
1333 	if (len < 0) {
1334 		ret = -EINVAL;
1335 		goto out;
1336 	}
1337 
1338 	ret = tun_do_read(tun, tfile, iocb, iv, len,
1339 			  file->f_flags & O_NONBLOCK);
1340 	ret = min_t(ssize_t, ret, len);
1341 out:
1342 	tun_put(tun);
1343 	return ret;
1344 }
1345 
1346 static void tun_free_netdev(struct net_device *dev)
1347 {
1348 	struct tun_struct *tun = netdev_priv(dev);
1349 
1350 	tun_flow_uninit(tun);
1351 	free_netdev(dev);
1352 }
1353 
1354 static void tun_setup(struct net_device *dev)
1355 {
1356 	struct tun_struct *tun = netdev_priv(dev);
1357 
1358 	tun->owner = INVALID_UID;
1359 	tun->group = INVALID_GID;
1360 
1361 	dev->ethtool_ops = &tun_ethtool_ops;
1362 	dev->destructor = tun_free_netdev;
1363 }
1364 
1365 /* Trivial set of netlink ops to allow deleting tun or tap
1366  * device with netlink.
1367  */
1368 static int tun_validate(struct nlattr *tb[], struct nlattr *data[])
1369 {
1370 	return -EINVAL;
1371 }
1372 
1373 static struct rtnl_link_ops tun_link_ops __read_mostly = {
1374 	.kind		= DRV_NAME,
1375 	.priv_size	= sizeof(struct tun_struct),
1376 	.setup		= tun_setup,
1377 	.validate	= tun_validate,
1378 };
1379 
1380 static void tun_sock_write_space(struct sock *sk)
1381 {
1382 	struct tun_file *tfile;
1383 	wait_queue_head_t *wqueue;
1384 
1385 	if (!sock_writeable(sk))
1386 		return;
1387 
1388 	if (!test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags))
1389 		return;
1390 
1391 	wqueue = sk_sleep(sk);
1392 	if (wqueue && waitqueue_active(wqueue))
1393 		wake_up_interruptible_sync_poll(wqueue, POLLOUT |
1394 						POLLWRNORM | POLLWRBAND);
1395 
1396 	tfile = container_of(sk, struct tun_file, sk);
1397 	kill_fasync(&tfile->fasync, SIGIO, POLL_OUT);
1398 }
1399 
1400 static int tun_sendmsg(struct kiocb *iocb, struct socket *sock,
1401 		       struct msghdr *m, size_t total_len)
1402 {
1403 	int ret;
1404 	struct tun_file *tfile = container_of(sock, struct tun_file, socket);
1405 	struct tun_struct *tun = __tun_get(tfile);
1406 
1407 	if (!tun)
1408 		return -EBADFD;
1409 	ret = tun_get_user(tun, tfile, m->msg_control, m->msg_iov, total_len,
1410 			   m->msg_iovlen, m->msg_flags & MSG_DONTWAIT);
1411 	tun_put(tun);
1412 	return ret;
1413 }
1414 
1415 
1416 static int tun_recvmsg(struct kiocb *iocb, struct socket *sock,
1417 		       struct msghdr *m, size_t total_len,
1418 		       int flags)
1419 {
1420 	struct tun_file *tfile = container_of(sock, struct tun_file, socket);
1421 	struct tun_struct *tun = __tun_get(tfile);
1422 	int ret;
1423 
1424 	if (!tun)
1425 		return -EBADFD;
1426 
1427 	if (flags & ~(MSG_DONTWAIT|MSG_TRUNC))
1428 		return -EINVAL;
1429 	ret = tun_do_read(tun, tfile, iocb, m->msg_iov, total_len,
1430 			  flags & MSG_DONTWAIT);
1431 	if (ret > total_len) {
1432 		m->msg_flags |= MSG_TRUNC;
1433 		ret = flags & MSG_TRUNC ? ret : total_len;
1434 	}
1435 	tun_put(tun);
1436 	return ret;
1437 }
1438 
1439 static int tun_release(struct socket *sock)
1440 {
1441 	if (sock->sk)
1442 		sock_put(sock->sk);
1443 	return 0;
1444 }
1445 
1446 /* Ops structure to mimic raw sockets with tun */
1447 static const struct proto_ops tun_socket_ops = {
1448 	.sendmsg = tun_sendmsg,
1449 	.recvmsg = tun_recvmsg,
1450 	.release = tun_release,
1451 };
1452 
1453 static struct proto tun_proto = {
1454 	.name		= "tun",
1455 	.owner		= THIS_MODULE,
1456 	.obj_size	= sizeof(struct tun_file),
1457 };
1458 
1459 static int tun_flags(struct tun_struct *tun)
1460 {
1461 	int flags = 0;
1462 
1463 	if (tun->flags & TUN_TUN_DEV)
1464 		flags |= IFF_TUN;
1465 	else
1466 		flags |= IFF_TAP;
1467 
1468 	if (tun->flags & TUN_NO_PI)
1469 		flags |= IFF_NO_PI;
1470 
1471 	/* This flag has no real effect.  We track the value for backwards
1472 	 * compatibility.
1473 	 */
1474 	if (tun->flags & TUN_ONE_QUEUE)
1475 		flags |= IFF_ONE_QUEUE;
1476 
1477 	if (tun->flags & TUN_VNET_HDR)
1478 		flags |= IFF_VNET_HDR;
1479 
1480 	if (tun->flags & TUN_TAP_MQ)
1481 		flags |= IFF_MULTI_QUEUE;
1482 
1483 	return flags;
1484 }
1485 
1486 static ssize_t tun_show_flags(struct device *dev, struct device_attribute *attr,
1487 			      char *buf)
1488 {
1489 	struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1490 	return sprintf(buf, "0x%x\n", tun_flags(tun));
1491 }
1492 
1493 static ssize_t tun_show_owner(struct device *dev, struct device_attribute *attr,
1494 			      char *buf)
1495 {
1496 	struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1497 	return uid_valid(tun->owner)?
1498 		sprintf(buf, "%u\n",
1499 			from_kuid_munged(current_user_ns(), tun->owner)):
1500 		sprintf(buf, "-1\n");
1501 }
1502 
1503 static ssize_t tun_show_group(struct device *dev, struct device_attribute *attr,
1504 			      char *buf)
1505 {
1506 	struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1507 	return gid_valid(tun->group) ?
1508 		sprintf(buf, "%u\n",
1509 			from_kgid_munged(current_user_ns(), tun->group)):
1510 		sprintf(buf, "-1\n");
1511 }
1512 
1513 static DEVICE_ATTR(tun_flags, 0444, tun_show_flags, NULL);
1514 static DEVICE_ATTR(owner, 0444, tun_show_owner, NULL);
1515 static DEVICE_ATTR(group, 0444, tun_show_group, NULL);
1516 
1517 static int tun_set_iff(struct net *net, struct file *file, struct ifreq *ifr)
1518 {
1519 	struct tun_struct *tun;
1520 	struct tun_file *tfile = file->private_data;
1521 	struct net_device *dev;
1522 	int err;
1523 
1524 	dev = __dev_get_by_name(net, ifr->ifr_name);
1525 	if (dev) {
1526 		if (ifr->ifr_flags & IFF_TUN_EXCL)
1527 			return -EBUSY;
1528 		if ((ifr->ifr_flags & IFF_TUN) && dev->netdev_ops == &tun_netdev_ops)
1529 			tun = netdev_priv(dev);
1530 		else if ((ifr->ifr_flags & IFF_TAP) && dev->netdev_ops == &tap_netdev_ops)
1531 			tun = netdev_priv(dev);
1532 		else
1533 			return -EINVAL;
1534 
1535 		if (tun_not_capable(tun))
1536 			return -EPERM;
1537 		err = security_tun_dev_attach(tfile->socket.sk);
1538 		if (err < 0)
1539 			return err;
1540 
1541 		err = tun_attach(tun, file);
1542 		if (err < 0)
1543 			return err;
1544 	}
1545 	else {
1546 		char *name;
1547 		unsigned long flags = 0;
1548 
1549 		if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
1550 			return -EPERM;
1551 		err = security_tun_dev_create();
1552 		if (err < 0)
1553 			return err;
1554 
1555 		/* Set dev type */
1556 		if (ifr->ifr_flags & IFF_TUN) {
1557 			/* TUN device */
1558 			flags |= TUN_TUN_DEV;
1559 			name = "tun%d";
1560 		} else if (ifr->ifr_flags & IFF_TAP) {
1561 			/* TAP device */
1562 			flags |= TUN_TAP_DEV;
1563 			name = "tap%d";
1564 		} else
1565 			return -EINVAL;
1566 
1567 		if (*ifr->ifr_name)
1568 			name = ifr->ifr_name;
1569 
1570 		dev = alloc_netdev_mqs(sizeof(struct tun_struct), name,
1571 				       tun_setup,
1572 				       MAX_TAP_QUEUES, MAX_TAP_QUEUES);
1573 		if (!dev)
1574 			return -ENOMEM;
1575 
1576 		dev_net_set(dev, net);
1577 		dev->rtnl_link_ops = &tun_link_ops;
1578 
1579 		tun = netdev_priv(dev);
1580 		tun->dev = dev;
1581 		tun->flags = flags;
1582 		tun->txflt.count = 0;
1583 		tun->vnet_hdr_sz = sizeof(struct virtio_net_hdr);
1584 
1585 		tun->filter_attached = false;
1586 		tun->sndbuf = tfile->socket.sk->sk_sndbuf;
1587 
1588 		spin_lock_init(&tun->lock);
1589 
1590 		security_tun_dev_post_create(&tfile->sk);
1591 
1592 		tun_net_init(dev);
1593 
1594 		if (tun_flow_init(tun))
1595 			goto err_free_dev;
1596 
1597 		dev->hw_features = NETIF_F_SG | NETIF_F_FRAGLIST |
1598 			TUN_USER_FEATURES;
1599 		dev->features = dev->hw_features;
1600 
1601 		err = tun_attach(tun, file);
1602 		if (err < 0)
1603 			goto err_free_dev;
1604 
1605 		err = register_netdevice(tun->dev);
1606 		if (err < 0)
1607 			goto err_free_dev;
1608 
1609 		if (device_create_file(&tun->dev->dev, &dev_attr_tun_flags) ||
1610 		    device_create_file(&tun->dev->dev, &dev_attr_owner) ||
1611 		    device_create_file(&tun->dev->dev, &dev_attr_group))
1612 			pr_err("Failed to create tun sysfs files\n");
1613 
1614 		netif_carrier_on(tun->dev);
1615 	}
1616 
1617 	tun_debug(KERN_INFO, tun, "tun_set_iff\n");
1618 
1619 	if (ifr->ifr_flags & IFF_NO_PI)
1620 		tun->flags |= TUN_NO_PI;
1621 	else
1622 		tun->flags &= ~TUN_NO_PI;
1623 
1624 	/* This flag has no real effect.  We track the value for backwards
1625 	 * compatibility.
1626 	 */
1627 	if (ifr->ifr_flags & IFF_ONE_QUEUE)
1628 		tun->flags |= TUN_ONE_QUEUE;
1629 	else
1630 		tun->flags &= ~TUN_ONE_QUEUE;
1631 
1632 	if (ifr->ifr_flags & IFF_VNET_HDR)
1633 		tun->flags |= TUN_VNET_HDR;
1634 	else
1635 		tun->flags &= ~TUN_VNET_HDR;
1636 
1637 	if (ifr->ifr_flags & IFF_MULTI_QUEUE)
1638 		tun->flags |= TUN_TAP_MQ;
1639 	else
1640 		tun->flags &= ~TUN_TAP_MQ;
1641 
1642 	/* Make sure persistent devices do not get stuck in
1643 	 * xoff state.
1644 	 */
1645 	if (netif_running(tun->dev))
1646 		netif_tx_wake_all_queues(tun->dev);
1647 
1648 	strcpy(ifr->ifr_name, tun->dev->name);
1649 	return 0;
1650 
1651  err_free_dev:
1652 	free_netdev(dev);
1653 	return err;
1654 }
1655 
1656 static void tun_get_iff(struct net *net, struct tun_struct *tun,
1657 		       struct ifreq *ifr)
1658 {
1659 	tun_debug(KERN_INFO, tun, "tun_get_iff\n");
1660 
1661 	strcpy(ifr->ifr_name, tun->dev->name);
1662 
1663 	ifr->ifr_flags = tun_flags(tun);
1664 
1665 }
1666 
1667 /* This is like a cut-down ethtool ops, except done via tun fd so no
1668  * privs required. */
1669 static int set_offload(struct tun_struct *tun, unsigned long arg)
1670 {
1671 	netdev_features_t features = 0;
1672 
1673 	if (arg & TUN_F_CSUM) {
1674 		features |= NETIF_F_HW_CSUM;
1675 		arg &= ~TUN_F_CSUM;
1676 
1677 		if (arg & (TUN_F_TSO4|TUN_F_TSO6)) {
1678 			if (arg & TUN_F_TSO_ECN) {
1679 				features |= NETIF_F_TSO_ECN;
1680 				arg &= ~TUN_F_TSO_ECN;
1681 			}
1682 			if (arg & TUN_F_TSO4)
1683 				features |= NETIF_F_TSO;
1684 			if (arg & TUN_F_TSO6)
1685 				features |= NETIF_F_TSO6;
1686 			arg &= ~(TUN_F_TSO4|TUN_F_TSO6);
1687 		}
1688 
1689 		if (arg & TUN_F_UFO) {
1690 			features |= NETIF_F_UFO;
1691 			arg &= ~TUN_F_UFO;
1692 		}
1693 	}
1694 
1695 	/* This gives the user a way to test for new features in future by
1696 	 * trying to set them. */
1697 	if (arg)
1698 		return -EINVAL;
1699 
1700 	tun->set_features = features;
1701 	netdev_update_features(tun->dev);
1702 
1703 	return 0;
1704 }
1705 
1706 static void tun_detach_filter(struct tun_struct *tun, int n)
1707 {
1708 	int i;
1709 	struct tun_file *tfile;
1710 
1711 	for (i = 0; i < n; i++) {
1712 		tfile = rcu_dereference_protected(tun->tfiles[i],
1713 						  lockdep_rtnl_is_held());
1714 		sk_detach_filter(tfile->socket.sk);
1715 	}
1716 
1717 	tun->filter_attached = false;
1718 }
1719 
1720 static int tun_attach_filter(struct tun_struct *tun)
1721 {
1722 	int i, ret = 0;
1723 	struct tun_file *tfile;
1724 
1725 	for (i = 0; i < tun->numqueues; i++) {
1726 		tfile = rcu_dereference_protected(tun->tfiles[i],
1727 						  lockdep_rtnl_is_held());
1728 		ret = sk_attach_filter(&tun->fprog, tfile->socket.sk);
1729 		if (ret) {
1730 			tun_detach_filter(tun, i);
1731 			return ret;
1732 		}
1733 	}
1734 
1735 	tun->filter_attached = true;
1736 	return ret;
1737 }
1738 
1739 static void tun_set_sndbuf(struct tun_struct *tun)
1740 {
1741 	struct tun_file *tfile;
1742 	int i;
1743 
1744 	for (i = 0; i < tun->numqueues; i++) {
1745 		tfile = rcu_dereference_protected(tun->tfiles[i],
1746 						lockdep_rtnl_is_held());
1747 		tfile->socket.sk->sk_sndbuf = tun->sndbuf;
1748 	}
1749 }
1750 
1751 static int tun_set_queue(struct file *file, struct ifreq *ifr)
1752 {
1753 	struct tun_file *tfile = file->private_data;
1754 	struct tun_struct *tun;
1755 	struct net_device *dev;
1756 	int ret = 0;
1757 
1758 	rtnl_lock();
1759 
1760 	if (ifr->ifr_flags & IFF_ATTACH_QUEUE) {
1761 		dev = __dev_get_by_name(tfile->net, ifr->ifr_name);
1762 		if (!dev) {
1763 			ret = -EINVAL;
1764 			goto unlock;
1765 		}
1766 
1767 		tun = netdev_priv(dev);
1768 		if (dev->netdev_ops != &tap_netdev_ops &&
1769 			dev->netdev_ops != &tun_netdev_ops)
1770 			ret = -EINVAL;
1771 		else if (tun_not_capable(tun))
1772 			ret = -EPERM;
1773 		else
1774 			ret = tun_attach(tun, file);
1775 	} else if (ifr->ifr_flags & IFF_DETACH_QUEUE)
1776 		__tun_detach(tfile, false);
1777 	else
1778 		ret = -EINVAL;
1779 
1780 unlock:
1781 	rtnl_unlock();
1782 	return ret;
1783 }
1784 
1785 static long __tun_chr_ioctl(struct file *file, unsigned int cmd,
1786 			    unsigned long arg, int ifreq_len)
1787 {
1788 	struct tun_file *tfile = file->private_data;
1789 	struct tun_struct *tun;
1790 	void __user* argp = (void __user*)arg;
1791 	struct ifreq ifr;
1792 	kuid_t owner;
1793 	kgid_t group;
1794 	int sndbuf;
1795 	int vnet_hdr_sz;
1796 	int ret;
1797 
1798 	if (cmd == TUNSETIFF || cmd == TUNSETQUEUE || _IOC_TYPE(cmd) == 0x89) {
1799 		if (copy_from_user(&ifr, argp, ifreq_len))
1800 			return -EFAULT;
1801 	} else {
1802 		memset(&ifr, 0, sizeof(ifr));
1803 	}
1804 	if (cmd == TUNGETFEATURES) {
1805 		/* Currently this just means: "what IFF flags are valid?".
1806 		 * This is needed because we never checked for invalid flags on
1807 		 * TUNSETIFF. */
1808 		return put_user(IFF_TUN | IFF_TAP | IFF_NO_PI | IFF_ONE_QUEUE |
1809 				IFF_VNET_HDR | IFF_MULTI_QUEUE,
1810 				(unsigned int __user*)argp);
1811 	} else if (cmd == TUNSETQUEUE)
1812 		return tun_set_queue(file, &ifr);
1813 
1814 	ret = 0;
1815 	rtnl_lock();
1816 
1817 	tun = __tun_get(tfile);
1818 	if (cmd == TUNSETIFF && !tun) {
1819 		ifr.ifr_name[IFNAMSIZ-1] = '\0';
1820 
1821 		ret = tun_set_iff(tfile->net, file, &ifr);
1822 
1823 		if (ret)
1824 			goto unlock;
1825 
1826 		if (copy_to_user(argp, &ifr, ifreq_len))
1827 			ret = -EFAULT;
1828 		goto unlock;
1829 	}
1830 
1831 	ret = -EBADFD;
1832 	if (!tun)
1833 		goto unlock;
1834 
1835 	tun_debug(KERN_INFO, tun, "tun_chr_ioctl cmd %u\n", cmd);
1836 
1837 	ret = 0;
1838 	switch (cmd) {
1839 	case TUNGETIFF:
1840 		tun_get_iff(current->nsproxy->net_ns, tun, &ifr);
1841 
1842 		if (copy_to_user(argp, &ifr, ifreq_len))
1843 			ret = -EFAULT;
1844 		break;
1845 
1846 	case TUNSETNOCSUM:
1847 		/* Disable/Enable checksum */
1848 
1849 		/* [unimplemented] */
1850 		tun_debug(KERN_INFO, tun, "ignored: set checksum %s\n",
1851 			  arg ? "disabled" : "enabled");
1852 		break;
1853 
1854 	case TUNSETPERSIST:
1855 		/* Disable/Enable persist mode. Keep an extra reference to the
1856 		 * module to prevent the module being unprobed.
1857 		 */
1858 		if (arg) {
1859 			tun->flags |= TUN_PERSIST;
1860 			__module_get(THIS_MODULE);
1861 		} else {
1862 			tun->flags &= ~TUN_PERSIST;
1863 			module_put(THIS_MODULE);
1864 		}
1865 
1866 		tun_debug(KERN_INFO, tun, "persist %s\n",
1867 			  arg ? "enabled" : "disabled");
1868 		break;
1869 
1870 	case TUNSETOWNER:
1871 		/* Set owner of the device */
1872 		owner = make_kuid(current_user_ns(), arg);
1873 		if (!uid_valid(owner)) {
1874 			ret = -EINVAL;
1875 			break;
1876 		}
1877 		tun->owner = owner;
1878 		tun_debug(KERN_INFO, tun, "owner set to %u\n",
1879 			  from_kuid(&init_user_ns, tun->owner));
1880 		break;
1881 
1882 	case TUNSETGROUP:
1883 		/* Set group of the device */
1884 		group = make_kgid(current_user_ns(), arg);
1885 		if (!gid_valid(group)) {
1886 			ret = -EINVAL;
1887 			break;
1888 		}
1889 		tun->group = group;
1890 		tun_debug(KERN_INFO, tun, "group set to %u\n",
1891 			  from_kgid(&init_user_ns, tun->group));
1892 		break;
1893 
1894 	case TUNSETLINK:
1895 		/* Only allow setting the type when the interface is down */
1896 		if (tun->dev->flags & IFF_UP) {
1897 			tun_debug(KERN_INFO, tun,
1898 				  "Linktype set failed because interface is up\n");
1899 			ret = -EBUSY;
1900 		} else {
1901 			tun->dev->type = (int) arg;
1902 			tun_debug(KERN_INFO, tun, "linktype set to %d\n",
1903 				  tun->dev->type);
1904 			ret = 0;
1905 		}
1906 		break;
1907 
1908 #ifdef TUN_DEBUG
1909 	case TUNSETDEBUG:
1910 		tun->debug = arg;
1911 		break;
1912 #endif
1913 	case TUNSETOFFLOAD:
1914 		ret = set_offload(tun, arg);
1915 		break;
1916 
1917 	case TUNSETTXFILTER:
1918 		/* Can be set only for TAPs */
1919 		ret = -EINVAL;
1920 		if ((tun->flags & TUN_TYPE_MASK) != TUN_TAP_DEV)
1921 			break;
1922 		ret = update_filter(&tun->txflt, (void __user *)arg);
1923 		break;
1924 
1925 	case SIOCGIFHWADDR:
1926 		/* Get hw address */
1927 		memcpy(ifr.ifr_hwaddr.sa_data, tun->dev->dev_addr, ETH_ALEN);
1928 		ifr.ifr_hwaddr.sa_family = tun->dev->type;
1929 		if (copy_to_user(argp, &ifr, ifreq_len))
1930 			ret = -EFAULT;
1931 		break;
1932 
1933 	case SIOCSIFHWADDR:
1934 		/* Set hw address */
1935 		tun_debug(KERN_DEBUG, tun, "set hw address: %pM\n",
1936 			  ifr.ifr_hwaddr.sa_data);
1937 
1938 		ret = dev_set_mac_address(tun->dev, &ifr.ifr_hwaddr);
1939 		break;
1940 
1941 	case TUNGETSNDBUF:
1942 		sndbuf = tfile->socket.sk->sk_sndbuf;
1943 		if (copy_to_user(argp, &sndbuf, sizeof(sndbuf)))
1944 			ret = -EFAULT;
1945 		break;
1946 
1947 	case TUNSETSNDBUF:
1948 		if (copy_from_user(&sndbuf, argp, sizeof(sndbuf))) {
1949 			ret = -EFAULT;
1950 			break;
1951 		}
1952 
1953 		tun->sndbuf = sndbuf;
1954 		tun_set_sndbuf(tun);
1955 		break;
1956 
1957 	case TUNGETVNETHDRSZ:
1958 		vnet_hdr_sz = tun->vnet_hdr_sz;
1959 		if (copy_to_user(argp, &vnet_hdr_sz, sizeof(vnet_hdr_sz)))
1960 			ret = -EFAULT;
1961 		break;
1962 
1963 	case TUNSETVNETHDRSZ:
1964 		if (copy_from_user(&vnet_hdr_sz, argp, sizeof(vnet_hdr_sz))) {
1965 			ret = -EFAULT;
1966 			break;
1967 		}
1968 		if (vnet_hdr_sz < (int)sizeof(struct virtio_net_hdr)) {
1969 			ret = -EINVAL;
1970 			break;
1971 		}
1972 
1973 		tun->vnet_hdr_sz = vnet_hdr_sz;
1974 		break;
1975 
1976 	case TUNATTACHFILTER:
1977 		/* Can be set only for TAPs */
1978 		ret = -EINVAL;
1979 		if ((tun->flags & TUN_TYPE_MASK) != TUN_TAP_DEV)
1980 			break;
1981 		ret = -EFAULT;
1982 		if (copy_from_user(&tun->fprog, argp, sizeof(tun->fprog)))
1983 			break;
1984 
1985 		ret = tun_attach_filter(tun);
1986 		break;
1987 
1988 	case TUNDETACHFILTER:
1989 		/* Can be set only for TAPs */
1990 		ret = -EINVAL;
1991 		if ((tun->flags & TUN_TYPE_MASK) != TUN_TAP_DEV)
1992 			break;
1993 		ret = 0;
1994 		tun_detach_filter(tun, tun->numqueues);
1995 		break;
1996 
1997 	default:
1998 		ret = -EINVAL;
1999 		break;
2000 	}
2001 
2002 unlock:
2003 	rtnl_unlock();
2004 	if (tun)
2005 		tun_put(tun);
2006 	return ret;
2007 }
2008 
2009 static long tun_chr_ioctl(struct file *file,
2010 			  unsigned int cmd, unsigned long arg)
2011 {
2012 	return __tun_chr_ioctl(file, cmd, arg, sizeof (struct ifreq));
2013 }
2014 
2015 #ifdef CONFIG_COMPAT
2016 static long tun_chr_compat_ioctl(struct file *file,
2017 			 unsigned int cmd, unsigned long arg)
2018 {
2019 	switch (cmd) {
2020 	case TUNSETIFF:
2021 	case TUNGETIFF:
2022 	case TUNSETTXFILTER:
2023 	case TUNGETSNDBUF:
2024 	case TUNSETSNDBUF:
2025 	case SIOCGIFHWADDR:
2026 	case SIOCSIFHWADDR:
2027 		arg = (unsigned long)compat_ptr(arg);
2028 		break;
2029 	default:
2030 		arg = (compat_ulong_t)arg;
2031 		break;
2032 	}
2033 
2034 	/*
2035 	 * compat_ifreq is shorter than ifreq, so we must not access beyond
2036 	 * the end of that structure. All fields that are used in this
2037 	 * driver are compatible though, we don't need to convert the
2038 	 * contents.
2039 	 */
2040 	return __tun_chr_ioctl(file, cmd, arg, sizeof(struct compat_ifreq));
2041 }
2042 #endif /* CONFIG_COMPAT */
2043 
2044 static int tun_chr_fasync(int fd, struct file *file, int on)
2045 {
2046 	struct tun_file *tfile = file->private_data;
2047 	int ret;
2048 
2049 	if ((ret = fasync_helper(fd, file, on, &tfile->fasync)) < 0)
2050 		goto out;
2051 
2052 	if (on) {
2053 		ret = __f_setown(file, task_pid(current), PIDTYPE_PID, 0);
2054 		if (ret)
2055 			goto out;
2056 		tfile->flags |= TUN_FASYNC;
2057 	} else
2058 		tfile->flags &= ~TUN_FASYNC;
2059 	ret = 0;
2060 out:
2061 	return ret;
2062 }
2063 
2064 static int tun_chr_open(struct inode *inode, struct file * file)
2065 {
2066 	struct tun_file *tfile;
2067 
2068 	DBG1(KERN_INFO, "tunX: tun_chr_open\n");
2069 
2070 	tfile = (struct tun_file *)sk_alloc(&init_net, AF_UNSPEC, GFP_KERNEL,
2071 					    &tun_proto);
2072 	if (!tfile)
2073 		return -ENOMEM;
2074 	rcu_assign_pointer(tfile->tun, NULL);
2075 	tfile->net = get_net(current->nsproxy->net_ns);
2076 	tfile->flags = 0;
2077 
2078 	rcu_assign_pointer(tfile->socket.wq, &tfile->wq);
2079 	init_waitqueue_head(&tfile->wq.wait);
2080 
2081 	tfile->socket.file = file;
2082 	tfile->socket.ops = &tun_socket_ops;
2083 
2084 	sock_init_data(&tfile->socket, &tfile->sk);
2085 	sk_change_net(&tfile->sk, tfile->net);
2086 
2087 	tfile->sk.sk_write_space = tun_sock_write_space;
2088 	tfile->sk.sk_sndbuf = INT_MAX;
2089 
2090 	file->private_data = tfile;
2091 	set_bit(SOCK_EXTERNALLY_ALLOCATED, &tfile->socket.flags);
2092 
2093 	return 0;
2094 }
2095 
2096 static int tun_chr_close(struct inode *inode, struct file *file)
2097 {
2098 	struct tun_file *tfile = file->private_data;
2099 	struct net *net = tfile->net;
2100 
2101 	tun_detach(tfile, true);
2102 	put_net(net);
2103 
2104 	return 0;
2105 }
2106 
2107 static const struct file_operations tun_fops = {
2108 	.owner	= THIS_MODULE,
2109 	.llseek = no_llseek,
2110 	.read  = do_sync_read,
2111 	.aio_read  = tun_chr_aio_read,
2112 	.write = do_sync_write,
2113 	.aio_write = tun_chr_aio_write,
2114 	.poll	= tun_chr_poll,
2115 	.unlocked_ioctl	= tun_chr_ioctl,
2116 #ifdef CONFIG_COMPAT
2117 	.compat_ioctl = tun_chr_compat_ioctl,
2118 #endif
2119 	.open	= tun_chr_open,
2120 	.release = tun_chr_close,
2121 	.fasync = tun_chr_fasync
2122 };
2123 
2124 static struct miscdevice tun_miscdev = {
2125 	.minor = TUN_MINOR,
2126 	.name = "tun",
2127 	.nodename = "net/tun",
2128 	.fops = &tun_fops,
2129 };
2130 
2131 /* ethtool interface */
2132 
2133 static int tun_get_settings(struct net_device *dev, struct ethtool_cmd *cmd)
2134 {
2135 	cmd->supported		= 0;
2136 	cmd->advertising	= 0;
2137 	ethtool_cmd_speed_set(cmd, SPEED_10);
2138 	cmd->duplex		= DUPLEX_FULL;
2139 	cmd->port		= PORT_TP;
2140 	cmd->phy_address	= 0;
2141 	cmd->transceiver	= XCVR_INTERNAL;
2142 	cmd->autoneg		= AUTONEG_DISABLE;
2143 	cmd->maxtxpkt		= 0;
2144 	cmd->maxrxpkt		= 0;
2145 	return 0;
2146 }
2147 
2148 static void tun_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
2149 {
2150 	struct tun_struct *tun = netdev_priv(dev);
2151 
2152 	strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
2153 	strlcpy(info->version, DRV_VERSION, sizeof(info->version));
2154 
2155 	switch (tun->flags & TUN_TYPE_MASK) {
2156 	case TUN_TUN_DEV:
2157 		strlcpy(info->bus_info, "tun", sizeof(info->bus_info));
2158 		break;
2159 	case TUN_TAP_DEV:
2160 		strlcpy(info->bus_info, "tap", sizeof(info->bus_info));
2161 		break;
2162 	}
2163 }
2164 
2165 static u32 tun_get_msglevel(struct net_device *dev)
2166 {
2167 #ifdef TUN_DEBUG
2168 	struct tun_struct *tun = netdev_priv(dev);
2169 	return tun->debug;
2170 #else
2171 	return -EOPNOTSUPP;
2172 #endif
2173 }
2174 
2175 static void tun_set_msglevel(struct net_device *dev, u32 value)
2176 {
2177 #ifdef TUN_DEBUG
2178 	struct tun_struct *tun = netdev_priv(dev);
2179 	tun->debug = value;
2180 #endif
2181 }
2182 
2183 static const struct ethtool_ops tun_ethtool_ops = {
2184 	.get_settings	= tun_get_settings,
2185 	.get_drvinfo	= tun_get_drvinfo,
2186 	.get_msglevel	= tun_get_msglevel,
2187 	.set_msglevel	= tun_set_msglevel,
2188 	.get_link	= ethtool_op_get_link,
2189 };
2190 
2191 
2192 static int __init tun_init(void)
2193 {
2194 	int ret = 0;
2195 
2196 	pr_info("%s, %s\n", DRV_DESCRIPTION, DRV_VERSION);
2197 	pr_info("%s\n", DRV_COPYRIGHT);
2198 
2199 	ret = rtnl_link_register(&tun_link_ops);
2200 	if (ret) {
2201 		pr_err("Can't register link_ops\n");
2202 		goto err_linkops;
2203 	}
2204 
2205 	ret = misc_register(&tun_miscdev);
2206 	if (ret) {
2207 		pr_err("Can't register misc device %d\n", TUN_MINOR);
2208 		goto err_misc;
2209 	}
2210 	return  0;
2211 err_misc:
2212 	rtnl_link_unregister(&tun_link_ops);
2213 err_linkops:
2214 	return ret;
2215 }
2216 
2217 static void tun_cleanup(void)
2218 {
2219 	misc_deregister(&tun_miscdev);
2220 	rtnl_link_unregister(&tun_link_ops);
2221 }
2222 
2223 /* Get an underlying socket object from tun file.  Returns error unless file is
2224  * attached to a device.  The returned object works like a packet socket, it
2225  * can be used for sock_sendmsg/sock_recvmsg.  The caller is responsible for
2226  * holding a reference to the file for as long as the socket is in use. */
2227 struct socket *tun_get_socket(struct file *file)
2228 {
2229 	struct tun_file *tfile;
2230 	if (file->f_op != &tun_fops)
2231 		return ERR_PTR(-EINVAL);
2232 	tfile = file->private_data;
2233 	if (!tfile)
2234 		return ERR_PTR(-EBADFD);
2235 	return &tfile->socket;
2236 }
2237 EXPORT_SYMBOL_GPL(tun_get_socket);
2238 
2239 module_init(tun_init);
2240 module_exit(tun_cleanup);
2241 MODULE_DESCRIPTION(DRV_DESCRIPTION);
2242 MODULE_AUTHOR(DRV_COPYRIGHT);
2243 MODULE_LICENSE("GPL");
2244 MODULE_ALIAS_MISCDEV(TUN_MINOR);
2245 MODULE_ALIAS("devname:net/tun");
2246