xref: /linux/drivers/net/tun.c (revision 12871a0bd67dd4db4418e1daafcd46e9d329ef10)
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 random_ether_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/system.h>
73 #include <asm/uaccess.h>
74 
75 /* Uncomment to enable debugging */
76 /* #define TUN_DEBUG 1 */
77 
78 #ifdef TUN_DEBUG
79 static int debug;
80 
81 #define tun_debug(level, tun, fmt, args...)			\
82 do {								\
83 	if (tun->debug)						\
84 		netdev_printk(level, tun->dev, fmt, ##args);	\
85 } while (0)
86 #define DBG1(level, fmt, args...)				\
87 do {								\
88 	if (debug == 2)						\
89 		printk(level fmt, ##args);			\
90 } while (0)
91 #else
92 #define tun_debug(level, tun, fmt, args...)			\
93 do {								\
94 	if (0)							\
95 		netdev_printk(level, tun->dev, fmt, ##args);	\
96 } while (0)
97 #define DBG1(level, fmt, args...)				\
98 do {								\
99 	if (0)							\
100 		printk(level fmt, ##args);			\
101 } while (0)
102 #endif
103 
104 #define FLT_EXACT_COUNT 8
105 struct tap_filter {
106 	unsigned int    count;    /* Number of addrs. Zero means disabled */
107 	u32             mask[2];  /* Mask of the hashed addrs */
108 	unsigned char	addr[FLT_EXACT_COUNT][ETH_ALEN];
109 };
110 
111 struct tun_file {
112 	atomic_t count;
113 	struct tun_struct *tun;
114 	struct net *net;
115 };
116 
117 struct tun_sock;
118 
119 struct tun_struct {
120 	struct tun_file		*tfile;
121 	unsigned int 		flags;
122 	uid_t			owner;
123 	gid_t			group;
124 
125 	struct net_device	*dev;
126 	u32			set_features;
127 #define TUN_USER_FEATURES (NETIF_F_HW_CSUM|NETIF_F_TSO_ECN|NETIF_F_TSO| \
128 			  NETIF_F_TSO6|NETIF_F_UFO)
129 	struct fasync_struct	*fasync;
130 
131 	struct tap_filter       txflt;
132 	struct socket		socket;
133 	struct socket_wq	wq;
134 
135 	int			vnet_hdr_sz;
136 
137 #ifdef TUN_DEBUG
138 	int debug;
139 #endif
140 };
141 
142 struct tun_sock {
143 	struct sock		sk;
144 	struct tun_struct	*tun;
145 };
146 
147 static inline struct tun_sock *tun_sk(struct sock *sk)
148 {
149 	return container_of(sk, struct tun_sock, sk);
150 }
151 
152 static int tun_attach(struct tun_struct *tun, struct file *file)
153 {
154 	struct tun_file *tfile = file->private_data;
155 	int err;
156 
157 	ASSERT_RTNL();
158 
159 	netif_tx_lock_bh(tun->dev);
160 
161 	err = -EINVAL;
162 	if (tfile->tun)
163 		goto out;
164 
165 	err = -EBUSY;
166 	if (tun->tfile)
167 		goto out;
168 
169 	err = 0;
170 	tfile->tun = tun;
171 	tun->tfile = tfile;
172 	tun->socket.file = file;
173 	netif_carrier_on(tun->dev);
174 	dev_hold(tun->dev);
175 	sock_hold(tun->socket.sk);
176 	atomic_inc(&tfile->count);
177 
178 out:
179 	netif_tx_unlock_bh(tun->dev);
180 	return err;
181 }
182 
183 static void __tun_detach(struct tun_struct *tun)
184 {
185 	/* Detach from net device */
186 	netif_tx_lock_bh(tun->dev);
187 	netif_carrier_off(tun->dev);
188 	tun->tfile = NULL;
189 	tun->socket.file = NULL;
190 	netif_tx_unlock_bh(tun->dev);
191 
192 	/* Drop read queue */
193 	skb_queue_purge(&tun->socket.sk->sk_receive_queue);
194 
195 	/* Drop the extra count on the net device */
196 	dev_put(tun->dev);
197 }
198 
199 static void tun_detach(struct tun_struct *tun)
200 {
201 	rtnl_lock();
202 	__tun_detach(tun);
203 	rtnl_unlock();
204 }
205 
206 static struct tun_struct *__tun_get(struct tun_file *tfile)
207 {
208 	struct tun_struct *tun = NULL;
209 
210 	if (atomic_inc_not_zero(&tfile->count))
211 		tun = tfile->tun;
212 
213 	return tun;
214 }
215 
216 static struct tun_struct *tun_get(struct file *file)
217 {
218 	return __tun_get(file->private_data);
219 }
220 
221 static void tun_put(struct tun_struct *tun)
222 {
223 	struct tun_file *tfile = tun->tfile;
224 
225 	if (atomic_dec_and_test(&tfile->count))
226 		tun_detach(tfile->tun);
227 }
228 
229 /* TAP filtering */
230 static void addr_hash_set(u32 *mask, const u8 *addr)
231 {
232 	int n = ether_crc(ETH_ALEN, addr) >> 26;
233 	mask[n >> 5] |= (1 << (n & 31));
234 }
235 
236 static unsigned int addr_hash_test(const u32 *mask, const u8 *addr)
237 {
238 	int n = ether_crc(ETH_ALEN, addr) >> 26;
239 	return mask[n >> 5] & (1 << (n & 31));
240 }
241 
242 static int update_filter(struct tap_filter *filter, void __user *arg)
243 {
244 	struct { u8 u[ETH_ALEN]; } *addr;
245 	struct tun_filter uf;
246 	int err, alen, n, nexact;
247 
248 	if (copy_from_user(&uf, arg, sizeof(uf)))
249 		return -EFAULT;
250 
251 	if (!uf.count) {
252 		/* Disabled */
253 		filter->count = 0;
254 		return 0;
255 	}
256 
257 	alen = ETH_ALEN * uf.count;
258 	addr = kmalloc(alen, GFP_KERNEL);
259 	if (!addr)
260 		return -ENOMEM;
261 
262 	if (copy_from_user(addr, arg + sizeof(uf), alen)) {
263 		err = -EFAULT;
264 		goto done;
265 	}
266 
267 	/* The filter is updated without holding any locks. Which is
268 	 * perfectly safe. We disable it first and in the worst
269 	 * case we'll accept a few undesired packets. */
270 	filter->count = 0;
271 	wmb();
272 
273 	/* Use first set of addresses as an exact filter */
274 	for (n = 0; n < uf.count && n < FLT_EXACT_COUNT; n++)
275 		memcpy(filter->addr[n], addr[n].u, ETH_ALEN);
276 
277 	nexact = n;
278 
279 	/* Remaining multicast addresses are hashed,
280 	 * unicast will leave the filter disabled. */
281 	memset(filter->mask, 0, sizeof(filter->mask));
282 	for (; n < uf.count; n++) {
283 		if (!is_multicast_ether_addr(addr[n].u)) {
284 			err = 0; /* no filter */
285 			goto done;
286 		}
287 		addr_hash_set(filter->mask, addr[n].u);
288 	}
289 
290 	/* For ALLMULTI just set the mask to all ones.
291 	 * This overrides the mask populated above. */
292 	if ((uf.flags & TUN_FLT_ALLMULTI))
293 		memset(filter->mask, ~0, sizeof(filter->mask));
294 
295 	/* Now enable the filter */
296 	wmb();
297 	filter->count = nexact;
298 
299 	/* Return the number of exact filters */
300 	err = nexact;
301 
302 done:
303 	kfree(addr);
304 	return err;
305 }
306 
307 /* Returns: 0 - drop, !=0 - accept */
308 static int run_filter(struct tap_filter *filter, const struct sk_buff *skb)
309 {
310 	/* Cannot use eth_hdr(skb) here because skb_mac_hdr() is incorrect
311 	 * at this point. */
312 	struct ethhdr *eh = (struct ethhdr *) skb->data;
313 	int i;
314 
315 	/* Exact match */
316 	for (i = 0; i < filter->count; i++)
317 		if (!compare_ether_addr(eh->h_dest, filter->addr[i]))
318 			return 1;
319 
320 	/* Inexact match (multicast only) */
321 	if (is_multicast_ether_addr(eh->h_dest))
322 		return addr_hash_test(filter->mask, eh->h_dest);
323 
324 	return 0;
325 }
326 
327 /*
328  * Checks whether the packet is accepted or not.
329  * Returns: 0 - drop, !=0 - accept
330  */
331 static int check_filter(struct tap_filter *filter, const struct sk_buff *skb)
332 {
333 	if (!filter->count)
334 		return 1;
335 
336 	return run_filter(filter, skb);
337 }
338 
339 /* Network device part of the driver */
340 
341 static const struct ethtool_ops tun_ethtool_ops;
342 
343 /* Net device detach from fd. */
344 static void tun_net_uninit(struct net_device *dev)
345 {
346 	struct tun_struct *tun = netdev_priv(dev);
347 	struct tun_file *tfile = tun->tfile;
348 
349 	/* Inform the methods they need to stop using the dev.
350 	 */
351 	if (tfile) {
352 		wake_up_all(&tun->wq.wait);
353 		if (atomic_dec_and_test(&tfile->count))
354 			__tun_detach(tun);
355 	}
356 }
357 
358 static void tun_free_netdev(struct net_device *dev)
359 {
360 	struct tun_struct *tun = netdev_priv(dev);
361 
362 	sock_put(tun->socket.sk);
363 }
364 
365 /* Net device open. */
366 static int tun_net_open(struct net_device *dev)
367 {
368 	netif_start_queue(dev);
369 	return 0;
370 }
371 
372 /* Net device close. */
373 static int tun_net_close(struct net_device *dev)
374 {
375 	netif_stop_queue(dev);
376 	return 0;
377 }
378 
379 /* Net device start xmit */
380 static netdev_tx_t tun_net_xmit(struct sk_buff *skb, struct net_device *dev)
381 {
382 	struct tun_struct *tun = netdev_priv(dev);
383 
384 	tun_debug(KERN_INFO, tun, "tun_net_xmit %d\n", skb->len);
385 
386 	/* Drop packet if interface is not attached */
387 	if (!tun->tfile)
388 		goto drop;
389 
390 	/* Drop if the filter does not like it.
391 	 * This is a noop if the filter is disabled.
392 	 * Filter can be enabled only for the TAP devices. */
393 	if (!check_filter(&tun->txflt, skb))
394 		goto drop;
395 
396 	if (tun->socket.sk->sk_filter &&
397 	    sk_filter(tun->socket.sk, skb))
398 		goto drop;
399 
400 	if (skb_queue_len(&tun->socket.sk->sk_receive_queue) >= dev->tx_queue_len) {
401 		if (!(tun->flags & TUN_ONE_QUEUE)) {
402 			/* Normal queueing mode. */
403 			/* Packet scheduler handles dropping of further packets. */
404 			netif_stop_queue(dev);
405 
406 			/* We won't see all dropped packets individually, so overrun
407 			 * error is more appropriate. */
408 			dev->stats.tx_fifo_errors++;
409 		} else {
410 			/* Single queue mode.
411 			 * Driver handles dropping of all packets itself. */
412 			goto drop;
413 		}
414 	}
415 
416 	/* Orphan the skb - required as we might hang on to it
417 	 * for indefinite time. */
418 	skb_orphan(skb);
419 
420 	/* Enqueue packet */
421 	skb_queue_tail(&tun->socket.sk->sk_receive_queue, skb);
422 
423 	/* Notify and wake up reader process */
424 	if (tun->flags & TUN_FASYNC)
425 		kill_fasync(&tun->fasync, SIGIO, POLL_IN);
426 	wake_up_interruptible_poll(&tun->wq.wait, POLLIN |
427 				   POLLRDNORM | POLLRDBAND);
428 	return NETDEV_TX_OK;
429 
430 drop:
431 	dev->stats.tx_dropped++;
432 	kfree_skb(skb);
433 	return NETDEV_TX_OK;
434 }
435 
436 static void tun_net_mclist(struct net_device *dev)
437 {
438 	/*
439 	 * This callback is supposed to deal with mc filter in
440 	 * _rx_ path and has nothing to do with the _tx_ path.
441 	 * In rx path we always accept everything userspace gives us.
442 	 */
443 }
444 
445 #define MIN_MTU 68
446 #define MAX_MTU 65535
447 
448 static int
449 tun_net_change_mtu(struct net_device *dev, int new_mtu)
450 {
451 	if (new_mtu < MIN_MTU || new_mtu + dev->hard_header_len > MAX_MTU)
452 		return -EINVAL;
453 	dev->mtu = new_mtu;
454 	return 0;
455 }
456 
457 static u32 tun_net_fix_features(struct net_device *dev, u32 features)
458 {
459 	struct tun_struct *tun = netdev_priv(dev);
460 
461 	return (features & tun->set_features) | (features & ~TUN_USER_FEATURES);
462 }
463 
464 static const struct net_device_ops tun_netdev_ops = {
465 	.ndo_uninit		= tun_net_uninit,
466 	.ndo_open		= tun_net_open,
467 	.ndo_stop		= tun_net_close,
468 	.ndo_start_xmit		= tun_net_xmit,
469 	.ndo_change_mtu		= tun_net_change_mtu,
470 	.ndo_fix_features	= tun_net_fix_features,
471 };
472 
473 static const struct net_device_ops tap_netdev_ops = {
474 	.ndo_uninit		= tun_net_uninit,
475 	.ndo_open		= tun_net_open,
476 	.ndo_stop		= tun_net_close,
477 	.ndo_start_xmit		= tun_net_xmit,
478 	.ndo_change_mtu		= tun_net_change_mtu,
479 	.ndo_fix_features	= tun_net_fix_features,
480 	.ndo_set_multicast_list	= tun_net_mclist,
481 	.ndo_set_mac_address	= eth_mac_addr,
482 	.ndo_validate_addr	= eth_validate_addr,
483 };
484 
485 /* Initialize net device. */
486 static void tun_net_init(struct net_device *dev)
487 {
488 	struct tun_struct *tun = netdev_priv(dev);
489 
490 	switch (tun->flags & TUN_TYPE_MASK) {
491 	case TUN_TUN_DEV:
492 		dev->netdev_ops = &tun_netdev_ops;
493 
494 		/* Point-to-Point TUN Device */
495 		dev->hard_header_len = 0;
496 		dev->addr_len = 0;
497 		dev->mtu = 1500;
498 
499 		/* Zero header length */
500 		dev->type = ARPHRD_NONE;
501 		dev->flags = IFF_POINTOPOINT | IFF_NOARP | IFF_MULTICAST;
502 		dev->tx_queue_len = TUN_READQ_SIZE;  /* We prefer our own queue length */
503 		break;
504 
505 	case TUN_TAP_DEV:
506 		dev->netdev_ops = &tap_netdev_ops;
507 		/* Ethernet TAP Device */
508 		ether_setup(dev);
509 
510 		random_ether_addr(dev->dev_addr);
511 
512 		dev->tx_queue_len = TUN_READQ_SIZE;  /* We prefer our own queue length */
513 		break;
514 	}
515 }
516 
517 /* Character device part */
518 
519 /* Poll */
520 static unsigned int tun_chr_poll(struct file *file, poll_table * wait)
521 {
522 	struct tun_file *tfile = file->private_data;
523 	struct tun_struct *tun = __tun_get(tfile);
524 	struct sock *sk;
525 	unsigned int mask = 0;
526 
527 	if (!tun)
528 		return POLLERR;
529 
530 	sk = tun->socket.sk;
531 
532 	tun_debug(KERN_INFO, tun, "tun_chr_poll\n");
533 
534 	poll_wait(file, &tun->wq.wait, wait);
535 
536 	if (!skb_queue_empty(&sk->sk_receive_queue))
537 		mask |= POLLIN | POLLRDNORM;
538 
539 	if (sock_writeable(sk) ||
540 	    (!test_and_set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags) &&
541 	     sock_writeable(sk)))
542 		mask |= POLLOUT | POLLWRNORM;
543 
544 	if (tun->dev->reg_state != NETREG_REGISTERED)
545 		mask = POLLERR;
546 
547 	tun_put(tun);
548 	return mask;
549 }
550 
551 /* prepad is the amount to reserve at front.  len is length after that.
552  * linear is a hint as to how much to copy (usually headers). */
553 static inline struct sk_buff *tun_alloc_skb(struct tun_struct *tun,
554 					    size_t prepad, size_t len,
555 					    size_t linear, int noblock)
556 {
557 	struct sock *sk = tun->socket.sk;
558 	struct sk_buff *skb;
559 	int err;
560 
561 	sock_update_classid(sk);
562 
563 	/* Under a page?  Don't bother with paged skb. */
564 	if (prepad + len < PAGE_SIZE || !linear)
565 		linear = len;
566 
567 	skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
568 				   &err);
569 	if (!skb)
570 		return ERR_PTR(err);
571 
572 	skb_reserve(skb, prepad);
573 	skb_put(skb, linear);
574 	skb->data_len = len - linear;
575 	skb->len += len - linear;
576 
577 	return skb;
578 }
579 
580 /* Get packet from user space buffer */
581 static __inline__ ssize_t tun_get_user(struct tun_struct *tun,
582 				       const struct iovec *iv, size_t count,
583 				       int noblock)
584 {
585 	struct tun_pi pi = { 0, cpu_to_be16(ETH_P_IP) };
586 	struct sk_buff *skb;
587 	size_t len = count, align = 0;
588 	struct virtio_net_hdr gso = { 0 };
589 	int offset = 0;
590 
591 	if (!(tun->flags & TUN_NO_PI)) {
592 		if ((len -= sizeof(pi)) > count)
593 			return -EINVAL;
594 
595 		if (memcpy_fromiovecend((void *)&pi, iv, 0, sizeof(pi)))
596 			return -EFAULT;
597 		offset += sizeof(pi);
598 	}
599 
600 	if (tun->flags & TUN_VNET_HDR) {
601 		if ((len -= tun->vnet_hdr_sz) > count)
602 			return -EINVAL;
603 
604 		if (memcpy_fromiovecend((void *)&gso, iv, offset, sizeof(gso)))
605 			return -EFAULT;
606 
607 		if ((gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
608 		    gso.csum_start + gso.csum_offset + 2 > gso.hdr_len)
609 			gso.hdr_len = gso.csum_start + gso.csum_offset + 2;
610 
611 		if (gso.hdr_len > len)
612 			return -EINVAL;
613 		offset += tun->vnet_hdr_sz;
614 	}
615 
616 	if ((tun->flags & TUN_TYPE_MASK) == TUN_TAP_DEV) {
617 		align = NET_IP_ALIGN;
618 		if (unlikely(len < ETH_HLEN ||
619 			     (gso.hdr_len && gso.hdr_len < ETH_HLEN)))
620 			return -EINVAL;
621 	}
622 
623 	skb = tun_alloc_skb(tun, align, len, gso.hdr_len, noblock);
624 	if (IS_ERR(skb)) {
625 		if (PTR_ERR(skb) != -EAGAIN)
626 			tun->dev->stats.rx_dropped++;
627 		return PTR_ERR(skb);
628 	}
629 
630 	if (skb_copy_datagram_from_iovec(skb, 0, iv, offset, len)) {
631 		tun->dev->stats.rx_dropped++;
632 		kfree_skb(skb);
633 		return -EFAULT;
634 	}
635 
636 	if (gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
637 		if (!skb_partial_csum_set(skb, gso.csum_start,
638 					  gso.csum_offset)) {
639 			tun->dev->stats.rx_frame_errors++;
640 			kfree_skb(skb);
641 			return -EINVAL;
642 		}
643 	}
644 
645 	switch (tun->flags & TUN_TYPE_MASK) {
646 	case TUN_TUN_DEV:
647 		if (tun->flags & TUN_NO_PI) {
648 			switch (skb->data[0] & 0xf0) {
649 			case 0x40:
650 				pi.proto = htons(ETH_P_IP);
651 				break;
652 			case 0x60:
653 				pi.proto = htons(ETH_P_IPV6);
654 				break;
655 			default:
656 				tun->dev->stats.rx_dropped++;
657 				kfree_skb(skb);
658 				return -EINVAL;
659 			}
660 		}
661 
662 		skb_reset_mac_header(skb);
663 		skb->protocol = pi.proto;
664 		skb->dev = tun->dev;
665 		break;
666 	case TUN_TAP_DEV:
667 		skb->protocol = eth_type_trans(skb, tun->dev);
668 		break;
669 	};
670 
671 	if (gso.gso_type != VIRTIO_NET_HDR_GSO_NONE) {
672 		pr_debug("GSO!\n");
673 		switch (gso.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
674 		case VIRTIO_NET_HDR_GSO_TCPV4:
675 			skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
676 			break;
677 		case VIRTIO_NET_HDR_GSO_TCPV6:
678 			skb_shinfo(skb)->gso_type = SKB_GSO_TCPV6;
679 			break;
680 		case VIRTIO_NET_HDR_GSO_UDP:
681 			skb_shinfo(skb)->gso_type = SKB_GSO_UDP;
682 			break;
683 		default:
684 			tun->dev->stats.rx_frame_errors++;
685 			kfree_skb(skb);
686 			return -EINVAL;
687 		}
688 
689 		if (gso.gso_type & VIRTIO_NET_HDR_GSO_ECN)
690 			skb_shinfo(skb)->gso_type |= SKB_GSO_TCP_ECN;
691 
692 		skb_shinfo(skb)->gso_size = gso.gso_size;
693 		if (skb_shinfo(skb)->gso_size == 0) {
694 			tun->dev->stats.rx_frame_errors++;
695 			kfree_skb(skb);
696 			return -EINVAL;
697 		}
698 
699 		/* Header must be checked, and gso_segs computed. */
700 		skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
701 		skb_shinfo(skb)->gso_segs = 0;
702 	}
703 
704 	netif_rx_ni(skb);
705 
706 	tun->dev->stats.rx_packets++;
707 	tun->dev->stats.rx_bytes += len;
708 
709 	return count;
710 }
711 
712 static ssize_t tun_chr_aio_write(struct kiocb *iocb, const struct iovec *iv,
713 			      unsigned long count, loff_t pos)
714 {
715 	struct file *file = iocb->ki_filp;
716 	struct tun_struct *tun = tun_get(file);
717 	ssize_t result;
718 
719 	if (!tun)
720 		return -EBADFD;
721 
722 	tun_debug(KERN_INFO, tun, "tun_chr_write %ld\n", count);
723 
724 	result = tun_get_user(tun, iv, iov_length(iv, count),
725 			      file->f_flags & O_NONBLOCK);
726 
727 	tun_put(tun);
728 	return result;
729 }
730 
731 /* Put packet to the user space buffer */
732 static __inline__ ssize_t tun_put_user(struct tun_struct *tun,
733 				       struct sk_buff *skb,
734 				       const struct iovec *iv, int len)
735 {
736 	struct tun_pi pi = { 0, skb->protocol };
737 	ssize_t total = 0;
738 
739 	if (!(tun->flags & TUN_NO_PI)) {
740 		if ((len -= sizeof(pi)) < 0)
741 			return -EINVAL;
742 
743 		if (len < skb->len) {
744 			/* Packet will be striped */
745 			pi.flags |= TUN_PKT_STRIP;
746 		}
747 
748 		if (memcpy_toiovecend(iv, (void *) &pi, 0, sizeof(pi)))
749 			return -EFAULT;
750 		total += sizeof(pi);
751 	}
752 
753 	if (tun->flags & TUN_VNET_HDR) {
754 		struct virtio_net_hdr gso = { 0 }; /* no info leak */
755 		if ((len -= tun->vnet_hdr_sz) < 0)
756 			return -EINVAL;
757 
758 		if (skb_is_gso(skb)) {
759 			struct skb_shared_info *sinfo = skb_shinfo(skb);
760 
761 			/* This is a hint as to how much should be linear. */
762 			gso.hdr_len = skb_headlen(skb);
763 			gso.gso_size = sinfo->gso_size;
764 			if (sinfo->gso_type & SKB_GSO_TCPV4)
765 				gso.gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
766 			else if (sinfo->gso_type & SKB_GSO_TCPV6)
767 				gso.gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
768 			else if (sinfo->gso_type & SKB_GSO_UDP)
769 				gso.gso_type = VIRTIO_NET_HDR_GSO_UDP;
770 			else {
771 				pr_err("unexpected GSO type: "
772 				       "0x%x, gso_size %d, hdr_len %d\n",
773 				       sinfo->gso_type, gso.gso_size,
774 				       gso.hdr_len);
775 				print_hex_dump(KERN_ERR, "tun: ",
776 					       DUMP_PREFIX_NONE,
777 					       16, 1, skb->head,
778 					       min((int)gso.hdr_len, 64), true);
779 				WARN_ON_ONCE(1);
780 				return -EINVAL;
781 			}
782 			if (sinfo->gso_type & SKB_GSO_TCP_ECN)
783 				gso.gso_type |= VIRTIO_NET_HDR_GSO_ECN;
784 		} else
785 			gso.gso_type = VIRTIO_NET_HDR_GSO_NONE;
786 
787 		if (skb->ip_summed == CHECKSUM_PARTIAL) {
788 			gso.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
789 			gso.csum_start = skb_checksum_start_offset(skb);
790 			gso.csum_offset = skb->csum_offset;
791 		} /* else everything is zero */
792 
793 		if (unlikely(memcpy_toiovecend(iv, (void *)&gso, total,
794 					       sizeof(gso))))
795 			return -EFAULT;
796 		total += tun->vnet_hdr_sz;
797 	}
798 
799 	len = min_t(int, skb->len, len);
800 
801 	skb_copy_datagram_const_iovec(skb, 0, iv, total, len);
802 	total += skb->len;
803 
804 	tun->dev->stats.tx_packets++;
805 	tun->dev->stats.tx_bytes += len;
806 
807 	return total;
808 }
809 
810 static ssize_t tun_do_read(struct tun_struct *tun,
811 			   struct kiocb *iocb, const struct iovec *iv,
812 			   ssize_t len, int noblock)
813 {
814 	DECLARE_WAITQUEUE(wait, current);
815 	struct sk_buff *skb;
816 	ssize_t ret = 0;
817 
818 	tun_debug(KERN_INFO, tun, "tun_chr_read\n");
819 
820 	add_wait_queue(&tun->wq.wait, &wait);
821 	while (len) {
822 		current->state = TASK_INTERRUPTIBLE;
823 
824 		/* Read frames from the queue */
825 		if (!(skb=skb_dequeue(&tun->socket.sk->sk_receive_queue))) {
826 			if (noblock) {
827 				ret = -EAGAIN;
828 				break;
829 			}
830 			if (signal_pending(current)) {
831 				ret = -ERESTARTSYS;
832 				break;
833 			}
834 			if (tun->dev->reg_state != NETREG_REGISTERED) {
835 				ret = -EIO;
836 				break;
837 			}
838 
839 			/* Nothing to read, let's sleep */
840 			schedule();
841 			continue;
842 		}
843 		netif_wake_queue(tun->dev);
844 
845 		ret = tun_put_user(tun, skb, iv, len);
846 		kfree_skb(skb);
847 		break;
848 	}
849 
850 	current->state = TASK_RUNNING;
851 	remove_wait_queue(&tun->wq.wait, &wait);
852 
853 	return ret;
854 }
855 
856 static ssize_t tun_chr_aio_read(struct kiocb *iocb, const struct iovec *iv,
857 			    unsigned long count, loff_t pos)
858 {
859 	struct file *file = iocb->ki_filp;
860 	struct tun_file *tfile = file->private_data;
861 	struct tun_struct *tun = __tun_get(tfile);
862 	ssize_t len, ret;
863 
864 	if (!tun)
865 		return -EBADFD;
866 	len = iov_length(iv, count);
867 	if (len < 0) {
868 		ret = -EINVAL;
869 		goto out;
870 	}
871 
872 	ret = tun_do_read(tun, iocb, iv, len, file->f_flags & O_NONBLOCK);
873 	ret = min_t(ssize_t, ret, len);
874 out:
875 	tun_put(tun);
876 	return ret;
877 }
878 
879 static void tun_setup(struct net_device *dev)
880 {
881 	struct tun_struct *tun = netdev_priv(dev);
882 
883 	tun->owner = -1;
884 	tun->group = -1;
885 
886 	dev->ethtool_ops = &tun_ethtool_ops;
887 	dev->destructor = tun_free_netdev;
888 }
889 
890 /* Trivial set of netlink ops to allow deleting tun or tap
891  * device with netlink.
892  */
893 static int tun_validate(struct nlattr *tb[], struct nlattr *data[])
894 {
895 	return -EINVAL;
896 }
897 
898 static struct rtnl_link_ops tun_link_ops __read_mostly = {
899 	.kind		= DRV_NAME,
900 	.priv_size	= sizeof(struct tun_struct),
901 	.setup		= tun_setup,
902 	.validate	= tun_validate,
903 };
904 
905 static void tun_sock_write_space(struct sock *sk)
906 {
907 	struct tun_struct *tun;
908 	wait_queue_head_t *wqueue;
909 
910 	if (!sock_writeable(sk))
911 		return;
912 
913 	if (!test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags))
914 		return;
915 
916 	wqueue = sk_sleep(sk);
917 	if (wqueue && waitqueue_active(wqueue))
918 		wake_up_interruptible_sync_poll(wqueue, POLLOUT |
919 						POLLWRNORM | POLLWRBAND);
920 
921 	tun = tun_sk(sk)->tun;
922 	kill_fasync(&tun->fasync, SIGIO, POLL_OUT);
923 }
924 
925 static void tun_sock_destruct(struct sock *sk)
926 {
927 	free_netdev(tun_sk(sk)->tun->dev);
928 }
929 
930 static int tun_sendmsg(struct kiocb *iocb, struct socket *sock,
931 		       struct msghdr *m, size_t total_len)
932 {
933 	struct tun_struct *tun = container_of(sock, struct tun_struct, socket);
934 	return tun_get_user(tun, m->msg_iov, total_len,
935 			    m->msg_flags & MSG_DONTWAIT);
936 }
937 
938 static int tun_recvmsg(struct kiocb *iocb, struct socket *sock,
939 		       struct msghdr *m, size_t total_len,
940 		       int flags)
941 {
942 	struct tun_struct *tun = container_of(sock, struct tun_struct, socket);
943 	int ret;
944 	if (flags & ~(MSG_DONTWAIT|MSG_TRUNC))
945 		return -EINVAL;
946 	ret = tun_do_read(tun, iocb, m->msg_iov, total_len,
947 			  flags & MSG_DONTWAIT);
948 	if (ret > total_len) {
949 		m->msg_flags |= MSG_TRUNC;
950 		ret = flags & MSG_TRUNC ? ret : total_len;
951 	}
952 	return ret;
953 }
954 
955 /* Ops structure to mimic raw sockets with tun */
956 static const struct proto_ops tun_socket_ops = {
957 	.sendmsg = tun_sendmsg,
958 	.recvmsg = tun_recvmsg,
959 };
960 
961 static struct proto tun_proto = {
962 	.name		= "tun",
963 	.owner		= THIS_MODULE,
964 	.obj_size	= sizeof(struct tun_sock),
965 };
966 
967 static int tun_flags(struct tun_struct *tun)
968 {
969 	int flags = 0;
970 
971 	if (tun->flags & TUN_TUN_DEV)
972 		flags |= IFF_TUN;
973 	else
974 		flags |= IFF_TAP;
975 
976 	if (tun->flags & TUN_NO_PI)
977 		flags |= IFF_NO_PI;
978 
979 	if (tun->flags & TUN_ONE_QUEUE)
980 		flags |= IFF_ONE_QUEUE;
981 
982 	if (tun->flags & TUN_VNET_HDR)
983 		flags |= IFF_VNET_HDR;
984 
985 	return flags;
986 }
987 
988 static ssize_t tun_show_flags(struct device *dev, struct device_attribute *attr,
989 			      char *buf)
990 {
991 	struct tun_struct *tun = netdev_priv(to_net_dev(dev));
992 	return sprintf(buf, "0x%x\n", tun_flags(tun));
993 }
994 
995 static ssize_t tun_show_owner(struct device *dev, struct device_attribute *attr,
996 			      char *buf)
997 {
998 	struct tun_struct *tun = netdev_priv(to_net_dev(dev));
999 	return sprintf(buf, "%d\n", tun->owner);
1000 }
1001 
1002 static ssize_t tun_show_group(struct device *dev, struct device_attribute *attr,
1003 			      char *buf)
1004 {
1005 	struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1006 	return sprintf(buf, "%d\n", tun->group);
1007 }
1008 
1009 static DEVICE_ATTR(tun_flags, 0444, tun_show_flags, NULL);
1010 static DEVICE_ATTR(owner, 0444, tun_show_owner, NULL);
1011 static DEVICE_ATTR(group, 0444, tun_show_group, NULL);
1012 
1013 static int tun_set_iff(struct net *net, struct file *file, struct ifreq *ifr)
1014 {
1015 	struct sock *sk;
1016 	struct tun_struct *tun;
1017 	struct net_device *dev;
1018 	int err;
1019 
1020 	dev = __dev_get_by_name(net, ifr->ifr_name);
1021 	if (dev) {
1022 		const struct cred *cred = current_cred();
1023 
1024 		if (ifr->ifr_flags & IFF_TUN_EXCL)
1025 			return -EBUSY;
1026 		if ((ifr->ifr_flags & IFF_TUN) && dev->netdev_ops == &tun_netdev_ops)
1027 			tun = netdev_priv(dev);
1028 		else if ((ifr->ifr_flags & IFF_TAP) && dev->netdev_ops == &tap_netdev_ops)
1029 			tun = netdev_priv(dev);
1030 		else
1031 			return -EINVAL;
1032 
1033 		if (((tun->owner != -1 && cred->euid != tun->owner) ||
1034 		     (tun->group != -1 && !in_egroup_p(tun->group))) &&
1035 		    !capable(CAP_NET_ADMIN))
1036 			return -EPERM;
1037 		err = security_tun_dev_attach(tun->socket.sk);
1038 		if (err < 0)
1039 			return err;
1040 
1041 		err = tun_attach(tun, file);
1042 		if (err < 0)
1043 			return err;
1044 	}
1045 	else {
1046 		char *name;
1047 		unsigned long flags = 0;
1048 
1049 		if (!capable(CAP_NET_ADMIN))
1050 			return -EPERM;
1051 		err = security_tun_dev_create();
1052 		if (err < 0)
1053 			return err;
1054 
1055 		/* Set dev type */
1056 		if (ifr->ifr_flags & IFF_TUN) {
1057 			/* TUN device */
1058 			flags |= TUN_TUN_DEV;
1059 			name = "tun%d";
1060 		} else if (ifr->ifr_flags & IFF_TAP) {
1061 			/* TAP device */
1062 			flags |= TUN_TAP_DEV;
1063 			name = "tap%d";
1064 		} else
1065 			return -EINVAL;
1066 
1067 		if (*ifr->ifr_name)
1068 			name = ifr->ifr_name;
1069 
1070 		dev = alloc_netdev(sizeof(struct tun_struct), name,
1071 				   tun_setup);
1072 		if (!dev)
1073 			return -ENOMEM;
1074 
1075 		dev_net_set(dev, net);
1076 		dev->rtnl_link_ops = &tun_link_ops;
1077 
1078 		tun = netdev_priv(dev);
1079 		tun->dev = dev;
1080 		tun->flags = flags;
1081 		tun->txflt.count = 0;
1082 		tun->vnet_hdr_sz = sizeof(struct virtio_net_hdr);
1083 
1084 		err = -ENOMEM;
1085 		sk = sk_alloc(net, AF_UNSPEC, GFP_KERNEL, &tun_proto);
1086 		if (!sk)
1087 			goto err_free_dev;
1088 
1089 		tun->socket.wq = &tun->wq;
1090 		init_waitqueue_head(&tun->wq.wait);
1091 		tun->socket.ops = &tun_socket_ops;
1092 		sock_init_data(&tun->socket, sk);
1093 		sk->sk_write_space = tun_sock_write_space;
1094 		sk->sk_sndbuf = INT_MAX;
1095 
1096 		tun_sk(sk)->tun = tun;
1097 
1098 		security_tun_dev_post_create(sk);
1099 
1100 		tun_net_init(dev);
1101 
1102 		dev->hw_features = NETIF_F_SG | NETIF_F_FRAGLIST |
1103 			TUN_USER_FEATURES;
1104 		dev->features = dev->hw_features;
1105 
1106 		err = register_netdevice(tun->dev);
1107 		if (err < 0)
1108 			goto err_free_sk;
1109 
1110 		if (device_create_file(&tun->dev->dev, &dev_attr_tun_flags) ||
1111 		    device_create_file(&tun->dev->dev, &dev_attr_owner) ||
1112 		    device_create_file(&tun->dev->dev, &dev_attr_group))
1113 			pr_err("Failed to create tun sysfs files\n");
1114 
1115 		sk->sk_destruct = tun_sock_destruct;
1116 
1117 		err = tun_attach(tun, file);
1118 		if (err < 0)
1119 			goto failed;
1120 	}
1121 
1122 	tun_debug(KERN_INFO, tun, "tun_set_iff\n");
1123 
1124 	if (ifr->ifr_flags & IFF_NO_PI)
1125 		tun->flags |= TUN_NO_PI;
1126 	else
1127 		tun->flags &= ~TUN_NO_PI;
1128 
1129 	if (ifr->ifr_flags & IFF_ONE_QUEUE)
1130 		tun->flags |= TUN_ONE_QUEUE;
1131 	else
1132 		tun->flags &= ~TUN_ONE_QUEUE;
1133 
1134 	if (ifr->ifr_flags & IFF_VNET_HDR)
1135 		tun->flags |= TUN_VNET_HDR;
1136 	else
1137 		tun->flags &= ~TUN_VNET_HDR;
1138 
1139 	/* Make sure persistent devices do not get stuck in
1140 	 * xoff state.
1141 	 */
1142 	if (netif_running(tun->dev))
1143 		netif_wake_queue(tun->dev);
1144 
1145 	strcpy(ifr->ifr_name, tun->dev->name);
1146 	return 0;
1147 
1148  err_free_sk:
1149 	sock_put(sk);
1150  err_free_dev:
1151 	free_netdev(dev);
1152  failed:
1153 	return err;
1154 }
1155 
1156 static int tun_get_iff(struct net *net, struct tun_struct *tun,
1157 		       struct ifreq *ifr)
1158 {
1159 	tun_debug(KERN_INFO, tun, "tun_get_iff\n");
1160 
1161 	strcpy(ifr->ifr_name, tun->dev->name);
1162 
1163 	ifr->ifr_flags = tun_flags(tun);
1164 
1165 	return 0;
1166 }
1167 
1168 /* This is like a cut-down ethtool ops, except done via tun fd so no
1169  * privs required. */
1170 static int set_offload(struct tun_struct *tun, unsigned long arg)
1171 {
1172 	u32 features = 0;
1173 
1174 	if (arg & TUN_F_CSUM) {
1175 		features |= NETIF_F_HW_CSUM;
1176 		arg &= ~TUN_F_CSUM;
1177 
1178 		if (arg & (TUN_F_TSO4|TUN_F_TSO6)) {
1179 			if (arg & TUN_F_TSO_ECN) {
1180 				features |= NETIF_F_TSO_ECN;
1181 				arg &= ~TUN_F_TSO_ECN;
1182 			}
1183 			if (arg & TUN_F_TSO4)
1184 				features |= NETIF_F_TSO;
1185 			if (arg & TUN_F_TSO6)
1186 				features |= NETIF_F_TSO6;
1187 			arg &= ~(TUN_F_TSO4|TUN_F_TSO6);
1188 		}
1189 
1190 		if (arg & TUN_F_UFO) {
1191 			features |= NETIF_F_UFO;
1192 			arg &= ~TUN_F_UFO;
1193 		}
1194 	}
1195 
1196 	/* This gives the user a way to test for new features in future by
1197 	 * trying to set them. */
1198 	if (arg)
1199 		return -EINVAL;
1200 
1201 	tun->set_features = features;
1202 	netdev_update_features(tun->dev);
1203 
1204 	return 0;
1205 }
1206 
1207 static long __tun_chr_ioctl(struct file *file, unsigned int cmd,
1208 			    unsigned long arg, int ifreq_len)
1209 {
1210 	struct tun_file *tfile = file->private_data;
1211 	struct tun_struct *tun;
1212 	void __user* argp = (void __user*)arg;
1213 	struct sock_fprog fprog;
1214 	struct ifreq ifr;
1215 	int sndbuf;
1216 	int vnet_hdr_sz;
1217 	int ret;
1218 
1219 	if (cmd == TUNSETIFF || _IOC_TYPE(cmd) == 0x89)
1220 		if (copy_from_user(&ifr, argp, ifreq_len))
1221 			return -EFAULT;
1222 
1223 	if (cmd == TUNGETFEATURES) {
1224 		/* Currently this just means: "what IFF flags are valid?".
1225 		 * This is needed because we never checked for invalid flags on
1226 		 * TUNSETIFF. */
1227 		return put_user(IFF_TUN | IFF_TAP | IFF_NO_PI | IFF_ONE_QUEUE |
1228 				IFF_VNET_HDR,
1229 				(unsigned int __user*)argp);
1230 	}
1231 
1232 	rtnl_lock();
1233 
1234 	tun = __tun_get(tfile);
1235 	if (cmd == TUNSETIFF && !tun) {
1236 		ifr.ifr_name[IFNAMSIZ-1] = '\0';
1237 
1238 		ret = tun_set_iff(tfile->net, file, &ifr);
1239 
1240 		if (ret)
1241 			goto unlock;
1242 
1243 		if (copy_to_user(argp, &ifr, ifreq_len))
1244 			ret = -EFAULT;
1245 		goto unlock;
1246 	}
1247 
1248 	ret = -EBADFD;
1249 	if (!tun)
1250 		goto unlock;
1251 
1252 	tun_debug(KERN_INFO, tun, "tun_chr_ioctl cmd %d\n", cmd);
1253 
1254 	ret = 0;
1255 	switch (cmd) {
1256 	case TUNGETIFF:
1257 		ret = tun_get_iff(current->nsproxy->net_ns, tun, &ifr);
1258 		if (ret)
1259 			break;
1260 
1261 		if (copy_to_user(argp, &ifr, ifreq_len))
1262 			ret = -EFAULT;
1263 		break;
1264 
1265 	case TUNSETNOCSUM:
1266 		/* Disable/Enable checksum */
1267 
1268 		/* [unimplemented] */
1269 		tun_debug(KERN_INFO, tun, "ignored: set checksum %s\n",
1270 			  arg ? "disabled" : "enabled");
1271 		break;
1272 
1273 	case TUNSETPERSIST:
1274 		/* Disable/Enable persist mode */
1275 		if (arg)
1276 			tun->flags |= TUN_PERSIST;
1277 		else
1278 			tun->flags &= ~TUN_PERSIST;
1279 
1280 		tun_debug(KERN_INFO, tun, "persist %s\n",
1281 			  arg ? "enabled" : "disabled");
1282 		break;
1283 
1284 	case TUNSETOWNER:
1285 		/* Set owner of the device */
1286 		tun->owner = (uid_t) arg;
1287 
1288 		tun_debug(KERN_INFO, tun, "owner set to %d\n", tun->owner);
1289 		break;
1290 
1291 	case TUNSETGROUP:
1292 		/* Set group of the device */
1293 		tun->group= (gid_t) arg;
1294 
1295 		tun_debug(KERN_INFO, tun, "group set to %d\n", tun->group);
1296 		break;
1297 
1298 	case TUNSETLINK:
1299 		/* Only allow setting the type when the interface is down */
1300 		if (tun->dev->flags & IFF_UP) {
1301 			tun_debug(KERN_INFO, tun,
1302 				  "Linktype set failed because interface is up\n");
1303 			ret = -EBUSY;
1304 		} else {
1305 			tun->dev->type = (int) arg;
1306 			tun_debug(KERN_INFO, tun, "linktype set to %d\n",
1307 				  tun->dev->type);
1308 			ret = 0;
1309 		}
1310 		break;
1311 
1312 #ifdef TUN_DEBUG
1313 	case TUNSETDEBUG:
1314 		tun->debug = arg;
1315 		break;
1316 #endif
1317 	case TUNSETOFFLOAD:
1318 		ret = set_offload(tun, arg);
1319 		break;
1320 
1321 	case TUNSETTXFILTER:
1322 		/* Can be set only for TAPs */
1323 		ret = -EINVAL;
1324 		if ((tun->flags & TUN_TYPE_MASK) != TUN_TAP_DEV)
1325 			break;
1326 		ret = update_filter(&tun->txflt, (void __user *)arg);
1327 		break;
1328 
1329 	case SIOCGIFHWADDR:
1330 		/* Get hw address */
1331 		memcpy(ifr.ifr_hwaddr.sa_data, tun->dev->dev_addr, ETH_ALEN);
1332 		ifr.ifr_hwaddr.sa_family = tun->dev->type;
1333 		if (copy_to_user(argp, &ifr, ifreq_len))
1334 			ret = -EFAULT;
1335 		break;
1336 
1337 	case SIOCSIFHWADDR:
1338 		/* Set hw address */
1339 		tun_debug(KERN_DEBUG, tun, "set hw address: %pM\n",
1340 			  ifr.ifr_hwaddr.sa_data);
1341 
1342 		ret = dev_set_mac_address(tun->dev, &ifr.ifr_hwaddr);
1343 		break;
1344 
1345 	case TUNGETSNDBUF:
1346 		sndbuf = tun->socket.sk->sk_sndbuf;
1347 		if (copy_to_user(argp, &sndbuf, sizeof(sndbuf)))
1348 			ret = -EFAULT;
1349 		break;
1350 
1351 	case TUNSETSNDBUF:
1352 		if (copy_from_user(&sndbuf, argp, sizeof(sndbuf))) {
1353 			ret = -EFAULT;
1354 			break;
1355 		}
1356 
1357 		tun->socket.sk->sk_sndbuf = sndbuf;
1358 		break;
1359 
1360 	case TUNGETVNETHDRSZ:
1361 		vnet_hdr_sz = tun->vnet_hdr_sz;
1362 		if (copy_to_user(argp, &vnet_hdr_sz, sizeof(vnet_hdr_sz)))
1363 			ret = -EFAULT;
1364 		break;
1365 
1366 	case TUNSETVNETHDRSZ:
1367 		if (copy_from_user(&vnet_hdr_sz, argp, sizeof(vnet_hdr_sz))) {
1368 			ret = -EFAULT;
1369 			break;
1370 		}
1371 		if (vnet_hdr_sz < (int)sizeof(struct virtio_net_hdr)) {
1372 			ret = -EINVAL;
1373 			break;
1374 		}
1375 
1376 		tun->vnet_hdr_sz = vnet_hdr_sz;
1377 		break;
1378 
1379 	case TUNATTACHFILTER:
1380 		/* Can be set only for TAPs */
1381 		ret = -EINVAL;
1382 		if ((tun->flags & TUN_TYPE_MASK) != TUN_TAP_DEV)
1383 			break;
1384 		ret = -EFAULT;
1385 		if (copy_from_user(&fprog, argp, sizeof(fprog)))
1386 			break;
1387 
1388 		ret = sk_attach_filter(&fprog, tun->socket.sk);
1389 		break;
1390 
1391 	case TUNDETACHFILTER:
1392 		/* Can be set only for TAPs */
1393 		ret = -EINVAL;
1394 		if ((tun->flags & TUN_TYPE_MASK) != TUN_TAP_DEV)
1395 			break;
1396 		ret = sk_detach_filter(tun->socket.sk);
1397 		break;
1398 
1399 	default:
1400 		ret = -EINVAL;
1401 		break;
1402 	}
1403 
1404 unlock:
1405 	rtnl_unlock();
1406 	if (tun)
1407 		tun_put(tun);
1408 	return ret;
1409 }
1410 
1411 static long tun_chr_ioctl(struct file *file,
1412 			  unsigned int cmd, unsigned long arg)
1413 {
1414 	return __tun_chr_ioctl(file, cmd, arg, sizeof (struct ifreq));
1415 }
1416 
1417 #ifdef CONFIG_COMPAT
1418 static long tun_chr_compat_ioctl(struct file *file,
1419 			 unsigned int cmd, unsigned long arg)
1420 {
1421 	switch (cmd) {
1422 	case TUNSETIFF:
1423 	case TUNGETIFF:
1424 	case TUNSETTXFILTER:
1425 	case TUNGETSNDBUF:
1426 	case TUNSETSNDBUF:
1427 	case SIOCGIFHWADDR:
1428 	case SIOCSIFHWADDR:
1429 		arg = (unsigned long)compat_ptr(arg);
1430 		break;
1431 	default:
1432 		arg = (compat_ulong_t)arg;
1433 		break;
1434 	}
1435 
1436 	/*
1437 	 * compat_ifreq is shorter than ifreq, so we must not access beyond
1438 	 * the end of that structure. All fields that are used in this
1439 	 * driver are compatible though, we don't need to convert the
1440 	 * contents.
1441 	 */
1442 	return __tun_chr_ioctl(file, cmd, arg, sizeof(struct compat_ifreq));
1443 }
1444 #endif /* CONFIG_COMPAT */
1445 
1446 static int tun_chr_fasync(int fd, struct file *file, int on)
1447 {
1448 	struct tun_struct *tun = tun_get(file);
1449 	int ret;
1450 
1451 	if (!tun)
1452 		return -EBADFD;
1453 
1454 	tun_debug(KERN_INFO, tun, "tun_chr_fasync %d\n", on);
1455 
1456 	if ((ret = fasync_helper(fd, file, on, &tun->fasync)) < 0)
1457 		goto out;
1458 
1459 	if (on) {
1460 		ret = __f_setown(file, task_pid(current), PIDTYPE_PID, 0);
1461 		if (ret)
1462 			goto out;
1463 		tun->flags |= TUN_FASYNC;
1464 	} else
1465 		tun->flags &= ~TUN_FASYNC;
1466 	ret = 0;
1467 out:
1468 	tun_put(tun);
1469 	return ret;
1470 }
1471 
1472 static int tun_chr_open(struct inode *inode, struct file * file)
1473 {
1474 	struct tun_file *tfile;
1475 
1476 	DBG1(KERN_INFO, "tunX: tun_chr_open\n");
1477 
1478 	tfile = kmalloc(sizeof(*tfile), GFP_KERNEL);
1479 	if (!tfile)
1480 		return -ENOMEM;
1481 	atomic_set(&tfile->count, 0);
1482 	tfile->tun = NULL;
1483 	tfile->net = get_net(current->nsproxy->net_ns);
1484 	file->private_data = tfile;
1485 	return 0;
1486 }
1487 
1488 static int tun_chr_close(struct inode *inode, struct file *file)
1489 {
1490 	struct tun_file *tfile = file->private_data;
1491 	struct tun_struct *tun;
1492 
1493 	tun = __tun_get(tfile);
1494 	if (tun) {
1495 		struct net_device *dev = tun->dev;
1496 
1497 		tun_debug(KERN_INFO, tun, "tun_chr_close\n");
1498 
1499 		__tun_detach(tun);
1500 
1501 		/* If desirable, unregister the netdevice. */
1502 		if (!(tun->flags & TUN_PERSIST)) {
1503 			rtnl_lock();
1504 			if (dev->reg_state == NETREG_REGISTERED)
1505 				unregister_netdevice(dev);
1506 			rtnl_unlock();
1507 		}
1508 	}
1509 
1510 	tun = tfile->tun;
1511 	if (tun)
1512 		sock_put(tun->socket.sk);
1513 
1514 	put_net(tfile->net);
1515 	kfree(tfile);
1516 
1517 	return 0;
1518 }
1519 
1520 static const struct file_operations tun_fops = {
1521 	.owner	= THIS_MODULE,
1522 	.llseek = no_llseek,
1523 	.read  = do_sync_read,
1524 	.aio_read  = tun_chr_aio_read,
1525 	.write = do_sync_write,
1526 	.aio_write = tun_chr_aio_write,
1527 	.poll	= tun_chr_poll,
1528 	.unlocked_ioctl	= tun_chr_ioctl,
1529 #ifdef CONFIG_COMPAT
1530 	.compat_ioctl = tun_chr_compat_ioctl,
1531 #endif
1532 	.open	= tun_chr_open,
1533 	.release = tun_chr_close,
1534 	.fasync = tun_chr_fasync
1535 };
1536 
1537 static struct miscdevice tun_miscdev = {
1538 	.minor = TUN_MINOR,
1539 	.name = "tun",
1540 	.nodename = "net/tun",
1541 	.fops = &tun_fops,
1542 };
1543 
1544 /* ethtool interface */
1545 
1546 static int tun_get_settings(struct net_device *dev, struct ethtool_cmd *cmd)
1547 {
1548 	cmd->supported		= 0;
1549 	cmd->advertising	= 0;
1550 	ethtool_cmd_speed_set(cmd, SPEED_10);
1551 	cmd->duplex		= DUPLEX_FULL;
1552 	cmd->port		= PORT_TP;
1553 	cmd->phy_address	= 0;
1554 	cmd->transceiver	= XCVR_INTERNAL;
1555 	cmd->autoneg		= AUTONEG_DISABLE;
1556 	cmd->maxtxpkt		= 0;
1557 	cmd->maxrxpkt		= 0;
1558 	return 0;
1559 }
1560 
1561 static void tun_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
1562 {
1563 	struct tun_struct *tun = netdev_priv(dev);
1564 
1565 	strcpy(info->driver, DRV_NAME);
1566 	strcpy(info->version, DRV_VERSION);
1567 	strcpy(info->fw_version, "N/A");
1568 
1569 	switch (tun->flags & TUN_TYPE_MASK) {
1570 	case TUN_TUN_DEV:
1571 		strcpy(info->bus_info, "tun");
1572 		break;
1573 	case TUN_TAP_DEV:
1574 		strcpy(info->bus_info, "tap");
1575 		break;
1576 	}
1577 }
1578 
1579 static u32 tun_get_msglevel(struct net_device *dev)
1580 {
1581 #ifdef TUN_DEBUG
1582 	struct tun_struct *tun = netdev_priv(dev);
1583 	return tun->debug;
1584 #else
1585 	return -EOPNOTSUPP;
1586 #endif
1587 }
1588 
1589 static void tun_set_msglevel(struct net_device *dev, u32 value)
1590 {
1591 #ifdef TUN_DEBUG
1592 	struct tun_struct *tun = netdev_priv(dev);
1593 	tun->debug = value;
1594 #endif
1595 }
1596 
1597 static const struct ethtool_ops tun_ethtool_ops = {
1598 	.get_settings	= tun_get_settings,
1599 	.get_drvinfo	= tun_get_drvinfo,
1600 	.get_msglevel	= tun_get_msglevel,
1601 	.set_msglevel	= tun_set_msglevel,
1602 	.get_link	= ethtool_op_get_link,
1603 };
1604 
1605 
1606 static int __init tun_init(void)
1607 {
1608 	int ret = 0;
1609 
1610 	pr_info("%s, %s\n", DRV_DESCRIPTION, DRV_VERSION);
1611 	pr_info("%s\n", DRV_COPYRIGHT);
1612 
1613 	ret = rtnl_link_register(&tun_link_ops);
1614 	if (ret) {
1615 		pr_err("Can't register link_ops\n");
1616 		goto err_linkops;
1617 	}
1618 
1619 	ret = misc_register(&tun_miscdev);
1620 	if (ret) {
1621 		pr_err("Can't register misc device %d\n", TUN_MINOR);
1622 		goto err_misc;
1623 	}
1624 	return  0;
1625 err_misc:
1626 	rtnl_link_unregister(&tun_link_ops);
1627 err_linkops:
1628 	return ret;
1629 }
1630 
1631 static void tun_cleanup(void)
1632 {
1633 	misc_deregister(&tun_miscdev);
1634 	rtnl_link_unregister(&tun_link_ops);
1635 }
1636 
1637 /* Get an underlying socket object from tun file.  Returns error unless file is
1638  * attached to a device.  The returned object works like a packet socket, it
1639  * can be used for sock_sendmsg/sock_recvmsg.  The caller is responsible for
1640  * holding a reference to the file for as long as the socket is in use. */
1641 struct socket *tun_get_socket(struct file *file)
1642 {
1643 	struct tun_struct *tun;
1644 	if (file->f_op != &tun_fops)
1645 		return ERR_PTR(-EINVAL);
1646 	tun = tun_get(file);
1647 	if (!tun)
1648 		return ERR_PTR(-EBADFD);
1649 	tun_put(tun);
1650 	return &tun->socket;
1651 }
1652 EXPORT_SYMBOL_GPL(tun_get_socket);
1653 
1654 module_init(tun_init);
1655 module_exit(tun_cleanup);
1656 MODULE_DESCRIPTION(DRV_DESCRIPTION);
1657 MODULE_AUTHOR(DRV_COPYRIGHT);
1658 MODULE_LICENSE("GPL");
1659 MODULE_ALIAS_MISCDEV(TUN_MINOR);
1660 MODULE_ALIAS("devname:net/tun");
1661