xref: /linux/drivers/net/tun.c (revision a33f32244d8550da8b4a26e277ce07d5c6d158b5)
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 DRV_NAME	"tun"
38 #define DRV_VERSION	"1.6"
39 #define DRV_DESCRIPTION	"Universal TUN/TAP device driver"
40 #define DRV_COPYRIGHT	"(C) 1999-2004 Max Krasnyansky <maxk@qualcomm.com>"
41 
42 #include <linux/module.h>
43 #include <linux/errno.h>
44 #include <linux/kernel.h>
45 #include <linux/major.h>
46 #include <linux/slab.h>
47 #include <linux/poll.h>
48 #include <linux/fcntl.h>
49 #include <linux/init.h>
50 #include <linux/skbuff.h>
51 #include <linux/netdevice.h>
52 #include <linux/etherdevice.h>
53 #include <linux/miscdevice.h>
54 #include <linux/ethtool.h>
55 #include <linux/rtnetlink.h>
56 #include <linux/compat.h>
57 #include <linux/if.h>
58 #include <linux/if_arp.h>
59 #include <linux/if_ether.h>
60 #include <linux/if_tun.h>
61 #include <linux/crc32.h>
62 #include <linux/nsproxy.h>
63 #include <linux/virtio_net.h>
64 #include <linux/rcupdate.h>
65 #include <net/net_namespace.h>
66 #include <net/netns/generic.h>
67 #include <net/rtnetlink.h>
68 #include <net/sock.h>
69 
70 #include <asm/system.h>
71 #include <asm/uaccess.h>
72 
73 /* Uncomment to enable debugging */
74 /* #define TUN_DEBUG 1 */
75 
76 #ifdef TUN_DEBUG
77 static int debug;
78 
79 #define DBG  if(tun->debug)printk
80 #define DBG1 if(debug==2)printk
81 #else
82 #define DBG( a... )
83 #define DBG1( a... )
84 #endif
85 
86 #define FLT_EXACT_COUNT 8
87 struct tap_filter {
88 	unsigned int    count;    /* Number of addrs. Zero means disabled */
89 	u32             mask[2];  /* Mask of the hashed addrs */
90 	unsigned char	addr[FLT_EXACT_COUNT][ETH_ALEN];
91 };
92 
93 struct tun_file {
94 	atomic_t count;
95 	struct tun_struct *tun;
96 	struct net *net;
97 };
98 
99 struct tun_sock;
100 
101 struct tun_struct {
102 	struct tun_file		*tfile;
103 	unsigned int 		flags;
104 	uid_t			owner;
105 	gid_t			group;
106 
107 	struct net_device	*dev;
108 	struct fasync_struct	*fasync;
109 
110 	struct tap_filter       txflt;
111 	struct socket		socket;
112 
113 #ifdef TUN_DEBUG
114 	int debug;
115 #endif
116 };
117 
118 struct tun_sock {
119 	struct sock		sk;
120 	struct tun_struct	*tun;
121 };
122 
123 static inline struct tun_sock *tun_sk(struct sock *sk)
124 {
125 	return container_of(sk, struct tun_sock, sk);
126 }
127 
128 static int tun_attach(struct tun_struct *tun, struct file *file)
129 {
130 	struct tun_file *tfile = file->private_data;
131 	int err;
132 
133 	ASSERT_RTNL();
134 
135 	netif_tx_lock_bh(tun->dev);
136 
137 	err = -EINVAL;
138 	if (tfile->tun)
139 		goto out;
140 
141 	err = -EBUSY;
142 	if (tun->tfile)
143 		goto out;
144 
145 	err = 0;
146 	tfile->tun = tun;
147 	tun->tfile = tfile;
148 	tun->socket.file = file;
149 	dev_hold(tun->dev);
150 	sock_hold(tun->socket.sk);
151 	atomic_inc(&tfile->count);
152 
153 out:
154 	netif_tx_unlock_bh(tun->dev);
155 	return err;
156 }
157 
158 static void __tun_detach(struct tun_struct *tun)
159 {
160 	/* Detach from net device */
161 	netif_tx_lock_bh(tun->dev);
162 	tun->tfile = NULL;
163 	tun->socket.file = NULL;
164 	netif_tx_unlock_bh(tun->dev);
165 
166 	/* Drop read queue */
167 	skb_queue_purge(&tun->socket.sk->sk_receive_queue);
168 
169 	/* Drop the extra count on the net device */
170 	dev_put(tun->dev);
171 }
172 
173 static void tun_detach(struct tun_struct *tun)
174 {
175 	rtnl_lock();
176 	__tun_detach(tun);
177 	rtnl_unlock();
178 }
179 
180 static struct tun_struct *__tun_get(struct tun_file *tfile)
181 {
182 	struct tun_struct *tun = NULL;
183 
184 	if (atomic_inc_not_zero(&tfile->count))
185 		tun = tfile->tun;
186 
187 	return tun;
188 }
189 
190 static struct tun_struct *tun_get(struct file *file)
191 {
192 	return __tun_get(file->private_data);
193 }
194 
195 static void tun_put(struct tun_struct *tun)
196 {
197 	struct tun_file *tfile = tun->tfile;
198 
199 	if (atomic_dec_and_test(&tfile->count))
200 		tun_detach(tfile->tun);
201 }
202 
203 /* TAP filterting */
204 static void addr_hash_set(u32 *mask, const u8 *addr)
205 {
206 	int n = ether_crc(ETH_ALEN, addr) >> 26;
207 	mask[n >> 5] |= (1 << (n & 31));
208 }
209 
210 static unsigned int addr_hash_test(const u32 *mask, const u8 *addr)
211 {
212 	int n = ether_crc(ETH_ALEN, addr) >> 26;
213 	return mask[n >> 5] & (1 << (n & 31));
214 }
215 
216 static int update_filter(struct tap_filter *filter, void __user *arg)
217 {
218 	struct { u8 u[ETH_ALEN]; } *addr;
219 	struct tun_filter uf;
220 	int err, alen, n, nexact;
221 
222 	if (copy_from_user(&uf, arg, sizeof(uf)))
223 		return -EFAULT;
224 
225 	if (!uf.count) {
226 		/* Disabled */
227 		filter->count = 0;
228 		return 0;
229 	}
230 
231 	alen = ETH_ALEN * uf.count;
232 	addr = kmalloc(alen, GFP_KERNEL);
233 	if (!addr)
234 		return -ENOMEM;
235 
236 	if (copy_from_user(addr, arg + sizeof(uf), alen)) {
237 		err = -EFAULT;
238 		goto done;
239 	}
240 
241 	/* The filter is updated without holding any locks. Which is
242 	 * perfectly safe. We disable it first and in the worst
243 	 * case we'll accept a few undesired packets. */
244 	filter->count = 0;
245 	wmb();
246 
247 	/* Use first set of addresses as an exact filter */
248 	for (n = 0; n < uf.count && n < FLT_EXACT_COUNT; n++)
249 		memcpy(filter->addr[n], addr[n].u, ETH_ALEN);
250 
251 	nexact = n;
252 
253 	/* Remaining multicast addresses are hashed,
254 	 * unicast will leave the filter disabled. */
255 	memset(filter->mask, 0, sizeof(filter->mask));
256 	for (; n < uf.count; n++) {
257 		if (!is_multicast_ether_addr(addr[n].u)) {
258 			err = 0; /* no filter */
259 			goto done;
260 		}
261 		addr_hash_set(filter->mask, addr[n].u);
262 	}
263 
264 	/* For ALLMULTI just set the mask to all ones.
265 	 * This overrides the mask populated above. */
266 	if ((uf.flags & TUN_FLT_ALLMULTI))
267 		memset(filter->mask, ~0, sizeof(filter->mask));
268 
269 	/* Now enable the filter */
270 	wmb();
271 	filter->count = nexact;
272 
273 	/* Return the number of exact filters */
274 	err = nexact;
275 
276 done:
277 	kfree(addr);
278 	return err;
279 }
280 
281 /* Returns: 0 - drop, !=0 - accept */
282 static int run_filter(struct tap_filter *filter, const struct sk_buff *skb)
283 {
284 	/* Cannot use eth_hdr(skb) here because skb_mac_hdr() is incorrect
285 	 * at this point. */
286 	struct ethhdr *eh = (struct ethhdr *) skb->data;
287 	int i;
288 
289 	/* Exact match */
290 	for (i = 0; i < filter->count; i++)
291 		if (!compare_ether_addr(eh->h_dest, filter->addr[i]))
292 			return 1;
293 
294 	/* Inexact match (multicast only) */
295 	if (is_multicast_ether_addr(eh->h_dest))
296 		return addr_hash_test(filter->mask, eh->h_dest);
297 
298 	return 0;
299 }
300 
301 /*
302  * Checks whether the packet is accepted or not.
303  * Returns: 0 - drop, !=0 - accept
304  */
305 static int check_filter(struct tap_filter *filter, const struct sk_buff *skb)
306 {
307 	if (!filter->count)
308 		return 1;
309 
310 	return run_filter(filter, skb);
311 }
312 
313 /* Network device part of the driver */
314 
315 static const struct ethtool_ops tun_ethtool_ops;
316 
317 /* Net device detach from fd. */
318 static void tun_net_uninit(struct net_device *dev)
319 {
320 	struct tun_struct *tun = netdev_priv(dev);
321 	struct tun_file *tfile = tun->tfile;
322 
323 	/* Inform the methods they need to stop using the dev.
324 	 */
325 	if (tfile) {
326 		wake_up_all(&tun->socket.wait);
327 		if (atomic_dec_and_test(&tfile->count))
328 			__tun_detach(tun);
329 	}
330 }
331 
332 static void tun_free_netdev(struct net_device *dev)
333 {
334 	struct tun_struct *tun = netdev_priv(dev);
335 
336 	sock_put(tun->socket.sk);
337 }
338 
339 /* Net device open. */
340 static int tun_net_open(struct net_device *dev)
341 {
342 	netif_start_queue(dev);
343 	return 0;
344 }
345 
346 /* Net device close. */
347 static int tun_net_close(struct net_device *dev)
348 {
349 	netif_stop_queue(dev);
350 	return 0;
351 }
352 
353 /* Net device start xmit */
354 static netdev_tx_t tun_net_xmit(struct sk_buff *skb, struct net_device *dev)
355 {
356 	struct tun_struct *tun = netdev_priv(dev);
357 
358 	DBG(KERN_INFO "%s: tun_net_xmit %d\n", tun->dev->name, skb->len);
359 
360 	/* Drop packet if interface is not attached */
361 	if (!tun->tfile)
362 		goto drop;
363 
364 	/* Drop if the filter does not like it.
365 	 * This is a noop if the filter is disabled.
366 	 * Filter can be enabled only for the TAP devices. */
367 	if (!check_filter(&tun->txflt, skb))
368 		goto drop;
369 
370 	if (tun->socket.sk->sk_filter &&
371 	    sk_filter(tun->socket.sk, skb))
372 		goto drop;
373 
374 	if (skb_queue_len(&tun->socket.sk->sk_receive_queue) >= dev->tx_queue_len) {
375 		if (!(tun->flags & TUN_ONE_QUEUE)) {
376 			/* Normal queueing mode. */
377 			/* Packet scheduler handles dropping of further packets. */
378 			netif_stop_queue(dev);
379 
380 			/* We won't see all dropped packets individually, so overrun
381 			 * error is more appropriate. */
382 			dev->stats.tx_fifo_errors++;
383 		} else {
384 			/* Single queue mode.
385 			 * Driver handles dropping of all packets itself. */
386 			goto drop;
387 		}
388 	}
389 
390 	/* Orphan the skb - required as we might hang on to it
391 	 * for indefinite time. */
392 	skb_orphan(skb);
393 
394 	/* Enqueue packet */
395 	skb_queue_tail(&tun->socket.sk->sk_receive_queue, skb);
396 	dev->trans_start = jiffies;
397 
398 	/* Notify and wake up reader process */
399 	if (tun->flags & TUN_FASYNC)
400 		kill_fasync(&tun->fasync, SIGIO, POLL_IN);
401 	wake_up_interruptible_poll(&tun->socket.wait, POLLIN |
402 				   POLLRDNORM | POLLRDBAND);
403 	return NETDEV_TX_OK;
404 
405 drop:
406 	dev->stats.tx_dropped++;
407 	kfree_skb(skb);
408 	return NETDEV_TX_OK;
409 }
410 
411 static void tun_net_mclist(struct net_device *dev)
412 {
413 	/*
414 	 * This callback is supposed to deal with mc filter in
415 	 * _rx_ path and has nothing to do with the _tx_ path.
416 	 * In rx path we always accept everything userspace gives us.
417 	 */
418 	return;
419 }
420 
421 #define MIN_MTU 68
422 #define MAX_MTU 65535
423 
424 static int
425 tun_net_change_mtu(struct net_device *dev, int new_mtu)
426 {
427 	if (new_mtu < MIN_MTU || new_mtu + dev->hard_header_len > MAX_MTU)
428 		return -EINVAL;
429 	dev->mtu = new_mtu;
430 	return 0;
431 }
432 
433 static const struct net_device_ops tun_netdev_ops = {
434 	.ndo_uninit		= tun_net_uninit,
435 	.ndo_open		= tun_net_open,
436 	.ndo_stop		= tun_net_close,
437 	.ndo_start_xmit		= tun_net_xmit,
438 	.ndo_change_mtu		= tun_net_change_mtu,
439 };
440 
441 static const struct net_device_ops tap_netdev_ops = {
442 	.ndo_uninit		= tun_net_uninit,
443 	.ndo_open		= tun_net_open,
444 	.ndo_stop		= tun_net_close,
445 	.ndo_start_xmit		= tun_net_xmit,
446 	.ndo_change_mtu		= tun_net_change_mtu,
447 	.ndo_set_multicast_list	= tun_net_mclist,
448 	.ndo_set_mac_address	= eth_mac_addr,
449 	.ndo_validate_addr	= eth_validate_addr,
450 };
451 
452 /* Initialize net device. */
453 static void tun_net_init(struct net_device *dev)
454 {
455 	struct tun_struct *tun = netdev_priv(dev);
456 
457 	switch (tun->flags & TUN_TYPE_MASK) {
458 	case TUN_TUN_DEV:
459 		dev->netdev_ops = &tun_netdev_ops;
460 
461 		/* Point-to-Point TUN Device */
462 		dev->hard_header_len = 0;
463 		dev->addr_len = 0;
464 		dev->mtu = 1500;
465 
466 		/* Zero header length */
467 		dev->type = ARPHRD_NONE;
468 		dev->flags = IFF_POINTOPOINT | IFF_NOARP | IFF_MULTICAST;
469 		dev->tx_queue_len = TUN_READQ_SIZE;  /* We prefer our own queue length */
470 		break;
471 
472 	case TUN_TAP_DEV:
473 		dev->netdev_ops = &tap_netdev_ops;
474 		/* Ethernet TAP Device */
475 		ether_setup(dev);
476 
477 		random_ether_addr(dev->dev_addr);
478 
479 		dev->tx_queue_len = TUN_READQ_SIZE;  /* We prefer our own queue length */
480 		break;
481 	}
482 }
483 
484 /* Character device part */
485 
486 /* Poll */
487 static unsigned int tun_chr_poll(struct file *file, poll_table * wait)
488 {
489 	struct tun_file *tfile = file->private_data;
490 	struct tun_struct *tun = __tun_get(tfile);
491 	struct sock *sk;
492 	unsigned int mask = 0;
493 
494 	if (!tun)
495 		return POLLERR;
496 
497 	sk = tun->socket.sk;
498 
499 	DBG(KERN_INFO "%s: tun_chr_poll\n", tun->dev->name);
500 
501 	poll_wait(file, &tun->socket.wait, wait);
502 
503 	if (!skb_queue_empty(&sk->sk_receive_queue))
504 		mask |= POLLIN | POLLRDNORM;
505 
506 	if (sock_writeable(sk) ||
507 	    (!test_and_set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags) &&
508 	     sock_writeable(sk)))
509 		mask |= POLLOUT | POLLWRNORM;
510 
511 	if (tun->dev->reg_state != NETREG_REGISTERED)
512 		mask = POLLERR;
513 
514 	tun_put(tun);
515 	return mask;
516 }
517 
518 /* prepad is the amount to reserve at front.  len is length after that.
519  * linear is a hint as to how much to copy (usually headers). */
520 static inline struct sk_buff *tun_alloc_skb(struct tun_struct *tun,
521 					    size_t prepad, size_t len,
522 					    size_t linear, int noblock)
523 {
524 	struct sock *sk = tun->socket.sk;
525 	struct sk_buff *skb;
526 	int err;
527 
528 	/* Under a page?  Don't bother with paged skb. */
529 	if (prepad + len < PAGE_SIZE || !linear)
530 		linear = len;
531 
532 	skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
533 				   &err);
534 	if (!skb)
535 		return ERR_PTR(err);
536 
537 	skb_reserve(skb, prepad);
538 	skb_put(skb, linear);
539 	skb->data_len = len - linear;
540 	skb->len += len - linear;
541 
542 	return skb;
543 }
544 
545 /* Get packet from user space buffer */
546 static __inline__ ssize_t tun_get_user(struct tun_struct *tun,
547 				       const struct iovec *iv, size_t count,
548 				       int noblock)
549 {
550 	struct tun_pi pi = { 0, cpu_to_be16(ETH_P_IP) };
551 	struct sk_buff *skb;
552 	size_t len = count, align = 0;
553 	struct virtio_net_hdr gso = { 0 };
554 	int offset = 0;
555 
556 	if (!(tun->flags & TUN_NO_PI)) {
557 		if ((len -= sizeof(pi)) > count)
558 			return -EINVAL;
559 
560 		if (memcpy_fromiovecend((void *)&pi, iv, 0, sizeof(pi)))
561 			return -EFAULT;
562 		offset += sizeof(pi);
563 	}
564 
565 	if (tun->flags & TUN_VNET_HDR) {
566 		if ((len -= sizeof(gso)) > count)
567 			return -EINVAL;
568 
569 		if (memcpy_fromiovecend((void *)&gso, iv, offset, sizeof(gso)))
570 			return -EFAULT;
571 
572 		if ((gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
573 		    gso.csum_start + gso.csum_offset + 2 > gso.hdr_len)
574 			gso.hdr_len = gso.csum_start + gso.csum_offset + 2;
575 
576 		if (gso.hdr_len > len)
577 			return -EINVAL;
578 		offset += sizeof(gso);
579 	}
580 
581 	if ((tun->flags & TUN_TYPE_MASK) == TUN_TAP_DEV) {
582 		align = NET_IP_ALIGN;
583 		if (unlikely(len < ETH_HLEN ||
584 			     (gso.hdr_len && gso.hdr_len < ETH_HLEN)))
585 			return -EINVAL;
586 	}
587 
588 	skb = tun_alloc_skb(tun, align, len, gso.hdr_len, noblock);
589 	if (IS_ERR(skb)) {
590 		if (PTR_ERR(skb) != -EAGAIN)
591 			tun->dev->stats.rx_dropped++;
592 		return PTR_ERR(skb);
593 	}
594 
595 	if (skb_copy_datagram_from_iovec(skb, 0, iv, offset, len)) {
596 		tun->dev->stats.rx_dropped++;
597 		kfree_skb(skb);
598 		return -EFAULT;
599 	}
600 
601 	if (gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
602 		if (!skb_partial_csum_set(skb, gso.csum_start,
603 					  gso.csum_offset)) {
604 			tun->dev->stats.rx_frame_errors++;
605 			kfree_skb(skb);
606 			return -EINVAL;
607 		}
608 	} else if (tun->flags & TUN_NOCHECKSUM)
609 		skb->ip_summed = CHECKSUM_UNNECESSARY;
610 
611 	switch (tun->flags & TUN_TYPE_MASK) {
612 	case TUN_TUN_DEV:
613 		if (tun->flags & TUN_NO_PI) {
614 			switch (skb->data[0] & 0xf0) {
615 			case 0x40:
616 				pi.proto = htons(ETH_P_IP);
617 				break;
618 			case 0x60:
619 				pi.proto = htons(ETH_P_IPV6);
620 				break;
621 			default:
622 				tun->dev->stats.rx_dropped++;
623 				kfree_skb(skb);
624 				return -EINVAL;
625 			}
626 		}
627 
628 		skb_reset_mac_header(skb);
629 		skb->protocol = pi.proto;
630 		skb->dev = tun->dev;
631 		break;
632 	case TUN_TAP_DEV:
633 		skb->protocol = eth_type_trans(skb, tun->dev);
634 		break;
635 	};
636 
637 	if (gso.gso_type != VIRTIO_NET_HDR_GSO_NONE) {
638 		pr_debug("GSO!\n");
639 		switch (gso.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
640 		case VIRTIO_NET_HDR_GSO_TCPV4:
641 			skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
642 			break;
643 		case VIRTIO_NET_HDR_GSO_TCPV6:
644 			skb_shinfo(skb)->gso_type = SKB_GSO_TCPV6;
645 			break;
646 		case VIRTIO_NET_HDR_GSO_UDP:
647 			skb_shinfo(skb)->gso_type = SKB_GSO_UDP;
648 			break;
649 		default:
650 			tun->dev->stats.rx_frame_errors++;
651 			kfree_skb(skb);
652 			return -EINVAL;
653 		}
654 
655 		if (gso.gso_type & VIRTIO_NET_HDR_GSO_ECN)
656 			skb_shinfo(skb)->gso_type |= SKB_GSO_TCP_ECN;
657 
658 		skb_shinfo(skb)->gso_size = gso.gso_size;
659 		if (skb_shinfo(skb)->gso_size == 0) {
660 			tun->dev->stats.rx_frame_errors++;
661 			kfree_skb(skb);
662 			return -EINVAL;
663 		}
664 
665 		/* Header must be checked, and gso_segs computed. */
666 		skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
667 		skb_shinfo(skb)->gso_segs = 0;
668 	}
669 
670 	netif_rx_ni(skb);
671 
672 	tun->dev->stats.rx_packets++;
673 	tun->dev->stats.rx_bytes += len;
674 
675 	return count;
676 }
677 
678 static ssize_t tun_chr_aio_write(struct kiocb *iocb, const struct iovec *iv,
679 			      unsigned long count, loff_t pos)
680 {
681 	struct file *file = iocb->ki_filp;
682 	struct tun_struct *tun = tun_get(file);
683 	ssize_t result;
684 
685 	if (!tun)
686 		return -EBADFD;
687 
688 	DBG(KERN_INFO "%s: tun_chr_write %ld\n", tun->dev->name, count);
689 
690 	result = tun_get_user(tun, iv, iov_length(iv, count),
691 			      file->f_flags & O_NONBLOCK);
692 
693 	tun_put(tun);
694 	return result;
695 }
696 
697 /* Put packet to the user space buffer */
698 static __inline__ ssize_t tun_put_user(struct tun_struct *tun,
699 				       struct sk_buff *skb,
700 				       const struct iovec *iv, int len)
701 {
702 	struct tun_pi pi = { 0, skb->protocol };
703 	ssize_t total = 0;
704 
705 	if (!(tun->flags & TUN_NO_PI)) {
706 		if ((len -= sizeof(pi)) < 0)
707 			return -EINVAL;
708 
709 		if (len < skb->len) {
710 			/* Packet will be striped */
711 			pi.flags |= TUN_PKT_STRIP;
712 		}
713 
714 		if (memcpy_toiovecend(iv, (void *) &pi, 0, sizeof(pi)))
715 			return -EFAULT;
716 		total += sizeof(pi);
717 	}
718 
719 	if (tun->flags & TUN_VNET_HDR) {
720 		struct virtio_net_hdr gso = { 0 }; /* no info leak */
721 		if ((len -= sizeof(gso)) < 0)
722 			return -EINVAL;
723 
724 		if (skb_is_gso(skb)) {
725 			struct skb_shared_info *sinfo = skb_shinfo(skb);
726 
727 			/* This is a hint as to how much should be linear. */
728 			gso.hdr_len = skb_headlen(skb);
729 			gso.gso_size = sinfo->gso_size;
730 			if (sinfo->gso_type & SKB_GSO_TCPV4)
731 				gso.gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
732 			else if (sinfo->gso_type & SKB_GSO_TCPV6)
733 				gso.gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
734 			else if (sinfo->gso_type & SKB_GSO_UDP)
735 				gso.gso_type = VIRTIO_NET_HDR_GSO_UDP;
736 			else
737 				BUG();
738 			if (sinfo->gso_type & SKB_GSO_TCP_ECN)
739 				gso.gso_type |= VIRTIO_NET_HDR_GSO_ECN;
740 		} else
741 			gso.gso_type = VIRTIO_NET_HDR_GSO_NONE;
742 
743 		if (skb->ip_summed == CHECKSUM_PARTIAL) {
744 			gso.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
745 			gso.csum_start = skb->csum_start - skb_headroom(skb);
746 			gso.csum_offset = skb->csum_offset;
747 		} /* else everything is zero */
748 
749 		if (unlikely(memcpy_toiovecend(iv, (void *)&gso, total,
750 					       sizeof(gso))))
751 			return -EFAULT;
752 		total += sizeof(gso);
753 	}
754 
755 	len = min_t(int, skb->len, len);
756 
757 	skb_copy_datagram_const_iovec(skb, 0, iv, total, len);
758 	total += skb->len;
759 
760 	tun->dev->stats.tx_packets++;
761 	tun->dev->stats.tx_bytes += len;
762 
763 	return total;
764 }
765 
766 static ssize_t tun_do_read(struct tun_struct *tun,
767 			   struct kiocb *iocb, const struct iovec *iv,
768 			   ssize_t len, int noblock)
769 {
770 	DECLARE_WAITQUEUE(wait, current);
771 	struct sk_buff *skb;
772 	ssize_t ret = 0;
773 
774 	DBG(KERN_INFO "%s: tun_chr_read\n", tun->dev->name);
775 
776 	add_wait_queue(&tun->socket.wait, &wait);
777 	while (len) {
778 		current->state = TASK_INTERRUPTIBLE;
779 
780 		/* Read frames from the queue */
781 		if (!(skb=skb_dequeue(&tun->socket.sk->sk_receive_queue))) {
782 			if (noblock) {
783 				ret = -EAGAIN;
784 				break;
785 			}
786 			if (signal_pending(current)) {
787 				ret = -ERESTARTSYS;
788 				break;
789 			}
790 			if (tun->dev->reg_state != NETREG_REGISTERED) {
791 				ret = -EIO;
792 				break;
793 			}
794 
795 			/* Nothing to read, let's sleep */
796 			schedule();
797 			continue;
798 		}
799 		netif_wake_queue(tun->dev);
800 
801 		ret = tun_put_user(tun, skb, iv, len);
802 		kfree_skb(skb);
803 		break;
804 	}
805 
806 	current->state = TASK_RUNNING;
807 	remove_wait_queue(&tun->socket.wait, &wait);
808 
809 	return ret;
810 }
811 
812 static ssize_t tun_chr_aio_read(struct kiocb *iocb, const struct iovec *iv,
813 			    unsigned long count, loff_t pos)
814 {
815 	struct file *file = iocb->ki_filp;
816 	struct tun_file *tfile = file->private_data;
817 	struct tun_struct *tun = __tun_get(tfile);
818 	ssize_t len, ret;
819 
820 	if (!tun)
821 		return -EBADFD;
822 	len = iov_length(iv, count);
823 	if (len < 0) {
824 		ret = -EINVAL;
825 		goto out;
826 	}
827 
828 	ret = tun_do_read(tun, iocb, iv, len, file->f_flags & O_NONBLOCK);
829 	ret = min_t(ssize_t, ret, len);
830 out:
831 	tun_put(tun);
832 	return ret;
833 }
834 
835 static void tun_setup(struct net_device *dev)
836 {
837 	struct tun_struct *tun = netdev_priv(dev);
838 
839 	tun->owner = -1;
840 	tun->group = -1;
841 
842 	dev->ethtool_ops = &tun_ethtool_ops;
843 	dev->destructor = tun_free_netdev;
844 }
845 
846 /* Trivial set of netlink ops to allow deleting tun or tap
847  * device with netlink.
848  */
849 static int tun_validate(struct nlattr *tb[], struct nlattr *data[])
850 {
851 	return -EINVAL;
852 }
853 
854 static struct rtnl_link_ops tun_link_ops __read_mostly = {
855 	.kind		= DRV_NAME,
856 	.priv_size	= sizeof(struct tun_struct),
857 	.setup		= tun_setup,
858 	.validate	= tun_validate,
859 };
860 
861 static void tun_sock_write_space(struct sock *sk)
862 {
863 	struct tun_struct *tun;
864 
865 	if (!sock_writeable(sk))
866 		return;
867 
868 	if (!test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags))
869 		return;
870 
871 	if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
872 		wake_up_interruptible_sync_poll(sk->sk_sleep, POLLOUT |
873 						POLLWRNORM | POLLWRBAND);
874 
875 	tun = tun_sk(sk)->tun;
876 	kill_fasync(&tun->fasync, SIGIO, POLL_OUT);
877 }
878 
879 static void tun_sock_destruct(struct sock *sk)
880 {
881 	free_netdev(tun_sk(sk)->tun->dev);
882 }
883 
884 static int tun_sendmsg(struct kiocb *iocb, struct socket *sock,
885 		       struct msghdr *m, size_t total_len)
886 {
887 	struct tun_struct *tun = container_of(sock, struct tun_struct, socket);
888 	return tun_get_user(tun, m->msg_iov, total_len,
889 			    m->msg_flags & MSG_DONTWAIT);
890 }
891 
892 static int tun_recvmsg(struct kiocb *iocb, struct socket *sock,
893 		       struct msghdr *m, size_t total_len,
894 		       int flags)
895 {
896 	struct tun_struct *tun = container_of(sock, struct tun_struct, socket);
897 	int ret;
898 	if (flags & ~(MSG_DONTWAIT|MSG_TRUNC))
899 		return -EINVAL;
900 	ret = tun_do_read(tun, iocb, m->msg_iov, total_len,
901 			  flags & MSG_DONTWAIT);
902 	if (ret > total_len) {
903 		m->msg_flags |= MSG_TRUNC;
904 		ret = flags & MSG_TRUNC ? ret : total_len;
905 	}
906 	return ret;
907 }
908 
909 /* Ops structure to mimic raw sockets with tun */
910 static const struct proto_ops tun_socket_ops = {
911 	.sendmsg = tun_sendmsg,
912 	.recvmsg = tun_recvmsg,
913 };
914 
915 static struct proto tun_proto = {
916 	.name		= "tun",
917 	.owner		= THIS_MODULE,
918 	.obj_size	= sizeof(struct tun_sock),
919 };
920 
921 static int tun_flags(struct tun_struct *tun)
922 {
923 	int flags = 0;
924 
925 	if (tun->flags & TUN_TUN_DEV)
926 		flags |= IFF_TUN;
927 	else
928 		flags |= IFF_TAP;
929 
930 	if (tun->flags & TUN_NO_PI)
931 		flags |= IFF_NO_PI;
932 
933 	if (tun->flags & TUN_ONE_QUEUE)
934 		flags |= IFF_ONE_QUEUE;
935 
936 	if (tun->flags & TUN_VNET_HDR)
937 		flags |= IFF_VNET_HDR;
938 
939 	return flags;
940 }
941 
942 static ssize_t tun_show_flags(struct device *dev, struct device_attribute *attr,
943 			      char *buf)
944 {
945 	struct tun_struct *tun = netdev_priv(to_net_dev(dev));
946 	return sprintf(buf, "0x%x\n", tun_flags(tun));
947 }
948 
949 static ssize_t tun_show_owner(struct device *dev, struct device_attribute *attr,
950 			      char *buf)
951 {
952 	struct tun_struct *tun = netdev_priv(to_net_dev(dev));
953 	return sprintf(buf, "%d\n", tun->owner);
954 }
955 
956 static ssize_t tun_show_group(struct device *dev, struct device_attribute *attr,
957 			      char *buf)
958 {
959 	struct tun_struct *tun = netdev_priv(to_net_dev(dev));
960 	return sprintf(buf, "%d\n", tun->group);
961 }
962 
963 static DEVICE_ATTR(tun_flags, 0444, tun_show_flags, NULL);
964 static DEVICE_ATTR(owner, 0444, tun_show_owner, NULL);
965 static DEVICE_ATTR(group, 0444, tun_show_group, NULL);
966 
967 static int tun_set_iff(struct net *net, struct file *file, struct ifreq *ifr)
968 {
969 	struct sock *sk;
970 	struct tun_struct *tun;
971 	struct net_device *dev;
972 	int err;
973 
974 	dev = __dev_get_by_name(net, ifr->ifr_name);
975 	if (dev) {
976 		const struct cred *cred = current_cred();
977 
978 		if (ifr->ifr_flags & IFF_TUN_EXCL)
979 			return -EBUSY;
980 		if ((ifr->ifr_flags & IFF_TUN) && dev->netdev_ops == &tun_netdev_ops)
981 			tun = netdev_priv(dev);
982 		else if ((ifr->ifr_flags & IFF_TAP) && dev->netdev_ops == &tap_netdev_ops)
983 			tun = netdev_priv(dev);
984 		else
985 			return -EINVAL;
986 
987 		if (((tun->owner != -1 && cred->euid != tun->owner) ||
988 		     (tun->group != -1 && !in_egroup_p(tun->group))) &&
989 		    !capable(CAP_NET_ADMIN))
990 			return -EPERM;
991 		err = security_tun_dev_attach(tun->socket.sk);
992 		if (err < 0)
993 			return err;
994 
995 		err = tun_attach(tun, file);
996 		if (err < 0)
997 			return err;
998 	}
999 	else {
1000 		char *name;
1001 		unsigned long flags = 0;
1002 
1003 		if (!capable(CAP_NET_ADMIN))
1004 			return -EPERM;
1005 		err = security_tun_dev_create();
1006 		if (err < 0)
1007 			return err;
1008 
1009 		/* Set dev type */
1010 		if (ifr->ifr_flags & IFF_TUN) {
1011 			/* TUN device */
1012 			flags |= TUN_TUN_DEV;
1013 			name = "tun%d";
1014 		} else if (ifr->ifr_flags & IFF_TAP) {
1015 			/* TAP device */
1016 			flags |= TUN_TAP_DEV;
1017 			name = "tap%d";
1018 		} else
1019 			return -EINVAL;
1020 
1021 		if (*ifr->ifr_name)
1022 			name = ifr->ifr_name;
1023 
1024 		dev = alloc_netdev(sizeof(struct tun_struct), name,
1025 				   tun_setup);
1026 		if (!dev)
1027 			return -ENOMEM;
1028 
1029 		dev_net_set(dev, net);
1030 		dev->rtnl_link_ops = &tun_link_ops;
1031 
1032 		tun = netdev_priv(dev);
1033 		tun->dev = dev;
1034 		tun->flags = flags;
1035 		tun->txflt.count = 0;
1036 
1037 		err = -ENOMEM;
1038 		sk = sk_alloc(net, AF_UNSPEC, GFP_KERNEL, &tun_proto);
1039 		if (!sk)
1040 			goto err_free_dev;
1041 
1042 		init_waitqueue_head(&tun->socket.wait);
1043 		tun->socket.ops = &tun_socket_ops;
1044 		sock_init_data(&tun->socket, sk);
1045 		sk->sk_write_space = tun_sock_write_space;
1046 		sk->sk_sndbuf = INT_MAX;
1047 
1048 		tun_sk(sk)->tun = tun;
1049 
1050 		security_tun_dev_post_create(sk);
1051 
1052 		tun_net_init(dev);
1053 
1054 		if (strchr(dev->name, '%')) {
1055 			err = dev_alloc_name(dev, dev->name);
1056 			if (err < 0)
1057 				goto err_free_sk;
1058 		}
1059 
1060 		err = register_netdevice(tun->dev);
1061 		if (err < 0)
1062 			goto err_free_sk;
1063 
1064 		if (device_create_file(&tun->dev->dev, &dev_attr_tun_flags) ||
1065 		    device_create_file(&tun->dev->dev, &dev_attr_owner) ||
1066 		    device_create_file(&tun->dev->dev, &dev_attr_group))
1067 			printk(KERN_ERR "Failed to create tun sysfs files\n");
1068 
1069 		sk->sk_destruct = tun_sock_destruct;
1070 
1071 		err = tun_attach(tun, file);
1072 		if (err < 0)
1073 			goto failed;
1074 	}
1075 
1076 	DBG(KERN_INFO "%s: tun_set_iff\n", tun->dev->name);
1077 
1078 	if (ifr->ifr_flags & IFF_NO_PI)
1079 		tun->flags |= TUN_NO_PI;
1080 	else
1081 		tun->flags &= ~TUN_NO_PI;
1082 
1083 	if (ifr->ifr_flags & IFF_ONE_QUEUE)
1084 		tun->flags |= TUN_ONE_QUEUE;
1085 	else
1086 		tun->flags &= ~TUN_ONE_QUEUE;
1087 
1088 	if (ifr->ifr_flags & IFF_VNET_HDR)
1089 		tun->flags |= TUN_VNET_HDR;
1090 	else
1091 		tun->flags &= ~TUN_VNET_HDR;
1092 
1093 	/* Make sure persistent devices do not get stuck in
1094 	 * xoff state.
1095 	 */
1096 	if (netif_running(tun->dev))
1097 		netif_wake_queue(tun->dev);
1098 
1099 	strcpy(ifr->ifr_name, tun->dev->name);
1100 	return 0;
1101 
1102  err_free_sk:
1103 	sock_put(sk);
1104  err_free_dev:
1105 	free_netdev(dev);
1106  failed:
1107 	return err;
1108 }
1109 
1110 static int tun_get_iff(struct net *net, struct tun_struct *tun,
1111 		       struct ifreq *ifr)
1112 {
1113 	DBG(KERN_INFO "%s: tun_get_iff\n", tun->dev->name);
1114 
1115 	strcpy(ifr->ifr_name, tun->dev->name);
1116 
1117 	ifr->ifr_flags = tun_flags(tun);
1118 
1119 	return 0;
1120 }
1121 
1122 /* This is like a cut-down ethtool ops, except done via tun fd so no
1123  * privs required. */
1124 static int set_offload(struct net_device *dev, unsigned long arg)
1125 {
1126 	unsigned int old_features, features;
1127 
1128 	old_features = dev->features;
1129 	/* Unset features, set them as we chew on the arg. */
1130 	features = (old_features & ~(NETIF_F_HW_CSUM|NETIF_F_SG|NETIF_F_FRAGLIST
1131 				    |NETIF_F_TSO_ECN|NETIF_F_TSO|NETIF_F_TSO6
1132 				    |NETIF_F_UFO));
1133 
1134 	if (arg & TUN_F_CSUM) {
1135 		features |= NETIF_F_HW_CSUM|NETIF_F_SG|NETIF_F_FRAGLIST;
1136 		arg &= ~TUN_F_CSUM;
1137 
1138 		if (arg & (TUN_F_TSO4|TUN_F_TSO6)) {
1139 			if (arg & TUN_F_TSO_ECN) {
1140 				features |= NETIF_F_TSO_ECN;
1141 				arg &= ~TUN_F_TSO_ECN;
1142 			}
1143 			if (arg & TUN_F_TSO4)
1144 				features |= NETIF_F_TSO;
1145 			if (arg & TUN_F_TSO6)
1146 				features |= NETIF_F_TSO6;
1147 			arg &= ~(TUN_F_TSO4|TUN_F_TSO6);
1148 		}
1149 
1150 		if (arg & TUN_F_UFO) {
1151 			features |= NETIF_F_UFO;
1152 			arg &= ~TUN_F_UFO;
1153 		}
1154 	}
1155 
1156 	/* This gives the user a way to test for new features in future by
1157 	 * trying to set them. */
1158 	if (arg)
1159 		return -EINVAL;
1160 
1161 	dev->features = features;
1162 	if (old_features != dev->features)
1163 		netdev_features_change(dev);
1164 
1165 	return 0;
1166 }
1167 
1168 static long __tun_chr_ioctl(struct file *file, unsigned int cmd,
1169 			    unsigned long arg, int ifreq_len)
1170 {
1171 	struct tun_file *tfile = file->private_data;
1172 	struct tun_struct *tun;
1173 	void __user* argp = (void __user*)arg;
1174 	struct sock_fprog fprog;
1175 	struct ifreq ifr;
1176 	int sndbuf;
1177 	int ret;
1178 
1179 	if (cmd == TUNSETIFF || _IOC_TYPE(cmd) == 0x89)
1180 		if (copy_from_user(&ifr, argp, ifreq_len))
1181 			return -EFAULT;
1182 
1183 	if (cmd == TUNGETFEATURES) {
1184 		/* Currently this just means: "what IFF flags are valid?".
1185 		 * This is needed because we never checked for invalid flags on
1186 		 * TUNSETIFF. */
1187 		return put_user(IFF_TUN | IFF_TAP | IFF_NO_PI | IFF_ONE_QUEUE |
1188 				IFF_VNET_HDR,
1189 				(unsigned int __user*)argp);
1190 	}
1191 
1192 	rtnl_lock();
1193 
1194 	tun = __tun_get(tfile);
1195 	if (cmd == TUNSETIFF && !tun) {
1196 		ifr.ifr_name[IFNAMSIZ-1] = '\0';
1197 
1198 		ret = tun_set_iff(tfile->net, file, &ifr);
1199 
1200 		if (ret)
1201 			goto unlock;
1202 
1203 		if (copy_to_user(argp, &ifr, ifreq_len))
1204 			ret = -EFAULT;
1205 		goto unlock;
1206 	}
1207 
1208 	ret = -EBADFD;
1209 	if (!tun)
1210 		goto unlock;
1211 
1212 	DBG(KERN_INFO "%s: tun_chr_ioctl cmd %d\n", tun->dev->name, cmd);
1213 
1214 	ret = 0;
1215 	switch (cmd) {
1216 	case TUNGETIFF:
1217 		ret = tun_get_iff(current->nsproxy->net_ns, tun, &ifr);
1218 		if (ret)
1219 			break;
1220 
1221 		if (copy_to_user(argp, &ifr, ifreq_len))
1222 			ret = -EFAULT;
1223 		break;
1224 
1225 	case TUNSETNOCSUM:
1226 		/* Disable/Enable checksum */
1227 		if (arg)
1228 			tun->flags |= TUN_NOCHECKSUM;
1229 		else
1230 			tun->flags &= ~TUN_NOCHECKSUM;
1231 
1232 		DBG(KERN_INFO "%s: checksum %s\n",
1233 		    tun->dev->name, arg ? "disabled" : "enabled");
1234 		break;
1235 
1236 	case TUNSETPERSIST:
1237 		/* Disable/Enable persist mode */
1238 		if (arg)
1239 			tun->flags |= TUN_PERSIST;
1240 		else
1241 			tun->flags &= ~TUN_PERSIST;
1242 
1243 		DBG(KERN_INFO "%s: persist %s\n",
1244 		    tun->dev->name, arg ? "enabled" : "disabled");
1245 		break;
1246 
1247 	case TUNSETOWNER:
1248 		/* Set owner of the device */
1249 		tun->owner = (uid_t) arg;
1250 
1251 		DBG(KERN_INFO "%s: owner set to %d\n", tun->dev->name, tun->owner);
1252 		break;
1253 
1254 	case TUNSETGROUP:
1255 		/* Set group of the device */
1256 		tun->group= (gid_t) arg;
1257 
1258 		DBG(KERN_INFO "%s: group set to %d\n", tun->dev->name, tun->group);
1259 		break;
1260 
1261 	case TUNSETLINK:
1262 		/* Only allow setting the type when the interface is down */
1263 		if (tun->dev->flags & IFF_UP) {
1264 			DBG(KERN_INFO "%s: Linktype set failed because interface is up\n",
1265 				tun->dev->name);
1266 			ret = -EBUSY;
1267 		} else {
1268 			tun->dev->type = (int) arg;
1269 			DBG(KERN_INFO "%s: linktype set to %d\n", tun->dev->name, tun->dev->type);
1270 			ret = 0;
1271 		}
1272 		break;
1273 
1274 #ifdef TUN_DEBUG
1275 	case TUNSETDEBUG:
1276 		tun->debug = arg;
1277 		break;
1278 #endif
1279 	case TUNSETOFFLOAD:
1280 		ret = set_offload(tun->dev, arg);
1281 		break;
1282 
1283 	case TUNSETTXFILTER:
1284 		/* Can be set only for TAPs */
1285 		ret = -EINVAL;
1286 		if ((tun->flags & TUN_TYPE_MASK) != TUN_TAP_DEV)
1287 			break;
1288 		ret = update_filter(&tun->txflt, (void __user *)arg);
1289 		break;
1290 
1291 	case SIOCGIFHWADDR:
1292 		/* Get hw addres */
1293 		memcpy(ifr.ifr_hwaddr.sa_data, tun->dev->dev_addr, ETH_ALEN);
1294 		ifr.ifr_hwaddr.sa_family = tun->dev->type;
1295 		if (copy_to_user(argp, &ifr, ifreq_len))
1296 			ret = -EFAULT;
1297 		break;
1298 
1299 	case SIOCSIFHWADDR:
1300 		/* Set hw address */
1301 		DBG(KERN_DEBUG "%s: set hw address: %pM\n",
1302 			tun->dev->name, ifr.ifr_hwaddr.sa_data);
1303 
1304 		ret = dev_set_mac_address(tun->dev, &ifr.ifr_hwaddr);
1305 		break;
1306 
1307 	case TUNGETSNDBUF:
1308 		sndbuf = tun->socket.sk->sk_sndbuf;
1309 		if (copy_to_user(argp, &sndbuf, sizeof(sndbuf)))
1310 			ret = -EFAULT;
1311 		break;
1312 
1313 	case TUNSETSNDBUF:
1314 		if (copy_from_user(&sndbuf, argp, sizeof(sndbuf))) {
1315 			ret = -EFAULT;
1316 			break;
1317 		}
1318 
1319 		tun->socket.sk->sk_sndbuf = sndbuf;
1320 		break;
1321 
1322 	case TUNATTACHFILTER:
1323 		/* Can be set only for TAPs */
1324 		ret = -EINVAL;
1325 		if ((tun->flags & TUN_TYPE_MASK) != TUN_TAP_DEV)
1326 			break;
1327 		ret = -EFAULT;
1328 		if (copy_from_user(&fprog, argp, sizeof(fprog)))
1329 			break;
1330 
1331 		ret = sk_attach_filter(&fprog, tun->socket.sk);
1332 		break;
1333 
1334 	case TUNDETACHFILTER:
1335 		/* Can be set only for TAPs */
1336 		ret = -EINVAL;
1337 		if ((tun->flags & TUN_TYPE_MASK) != TUN_TAP_DEV)
1338 			break;
1339 		ret = sk_detach_filter(tun->socket.sk);
1340 		break;
1341 
1342 	default:
1343 		ret = -EINVAL;
1344 		break;
1345 	};
1346 
1347 unlock:
1348 	rtnl_unlock();
1349 	if (tun)
1350 		tun_put(tun);
1351 	return ret;
1352 }
1353 
1354 static long tun_chr_ioctl(struct file *file,
1355 			  unsigned int cmd, unsigned long arg)
1356 {
1357 	return __tun_chr_ioctl(file, cmd, arg, sizeof (struct ifreq));
1358 }
1359 
1360 #ifdef CONFIG_COMPAT
1361 static long tun_chr_compat_ioctl(struct file *file,
1362 			 unsigned int cmd, unsigned long arg)
1363 {
1364 	switch (cmd) {
1365 	case TUNSETIFF:
1366 	case TUNGETIFF:
1367 	case TUNSETTXFILTER:
1368 	case TUNGETSNDBUF:
1369 	case TUNSETSNDBUF:
1370 	case SIOCGIFHWADDR:
1371 	case SIOCSIFHWADDR:
1372 		arg = (unsigned long)compat_ptr(arg);
1373 		break;
1374 	default:
1375 		arg = (compat_ulong_t)arg;
1376 		break;
1377 	}
1378 
1379 	/*
1380 	 * compat_ifreq is shorter than ifreq, so we must not access beyond
1381 	 * the end of that structure. All fields that are used in this
1382 	 * driver are compatible though, we don't need to convert the
1383 	 * contents.
1384 	 */
1385 	return __tun_chr_ioctl(file, cmd, arg, sizeof(struct compat_ifreq));
1386 }
1387 #endif /* CONFIG_COMPAT */
1388 
1389 static int tun_chr_fasync(int fd, struct file *file, int on)
1390 {
1391 	struct tun_struct *tun = tun_get(file);
1392 	int ret;
1393 
1394 	if (!tun)
1395 		return -EBADFD;
1396 
1397 	DBG(KERN_INFO "%s: tun_chr_fasync %d\n", tun->dev->name, on);
1398 
1399 	if ((ret = fasync_helper(fd, file, on, &tun->fasync)) < 0)
1400 		goto out;
1401 
1402 	if (on) {
1403 		ret = __f_setown(file, task_pid(current), PIDTYPE_PID, 0);
1404 		if (ret)
1405 			goto out;
1406 		tun->flags |= TUN_FASYNC;
1407 	} else
1408 		tun->flags &= ~TUN_FASYNC;
1409 	ret = 0;
1410 out:
1411 	tun_put(tun);
1412 	return ret;
1413 }
1414 
1415 static int tun_chr_open(struct inode *inode, struct file * file)
1416 {
1417 	struct tun_file *tfile;
1418 
1419 	DBG1(KERN_INFO "tunX: tun_chr_open\n");
1420 
1421 	tfile = kmalloc(sizeof(*tfile), GFP_KERNEL);
1422 	if (!tfile)
1423 		return -ENOMEM;
1424 	atomic_set(&tfile->count, 0);
1425 	tfile->tun = NULL;
1426 	tfile->net = get_net(current->nsproxy->net_ns);
1427 	file->private_data = tfile;
1428 	return 0;
1429 }
1430 
1431 static int tun_chr_close(struct inode *inode, struct file *file)
1432 {
1433 	struct tun_file *tfile = file->private_data;
1434 	struct tun_struct *tun;
1435 
1436 	tun = __tun_get(tfile);
1437 	if (tun) {
1438 		struct net_device *dev = tun->dev;
1439 
1440 		DBG(KERN_INFO "%s: tun_chr_close\n", dev->name);
1441 
1442 		__tun_detach(tun);
1443 
1444 		/* If desirable, unregister the netdevice. */
1445 		if (!(tun->flags & TUN_PERSIST)) {
1446 			rtnl_lock();
1447 			if (dev->reg_state == NETREG_REGISTERED)
1448 				unregister_netdevice(dev);
1449 			rtnl_unlock();
1450 		}
1451 	}
1452 
1453 	tun = tfile->tun;
1454 	if (tun)
1455 		sock_put(tun->socket.sk);
1456 
1457 	put_net(tfile->net);
1458 	kfree(tfile);
1459 
1460 	return 0;
1461 }
1462 
1463 static const struct file_operations tun_fops = {
1464 	.owner	= THIS_MODULE,
1465 	.llseek = no_llseek,
1466 	.read  = do_sync_read,
1467 	.aio_read  = tun_chr_aio_read,
1468 	.write = do_sync_write,
1469 	.aio_write = tun_chr_aio_write,
1470 	.poll	= tun_chr_poll,
1471 	.unlocked_ioctl	= tun_chr_ioctl,
1472 #ifdef CONFIG_COMPAT
1473 	.compat_ioctl = tun_chr_compat_ioctl,
1474 #endif
1475 	.open	= tun_chr_open,
1476 	.release = tun_chr_close,
1477 	.fasync = tun_chr_fasync
1478 };
1479 
1480 static struct miscdevice tun_miscdev = {
1481 	.minor = TUN_MINOR,
1482 	.name = "tun",
1483 	.nodename = "net/tun",
1484 	.fops = &tun_fops,
1485 };
1486 
1487 /* ethtool interface */
1488 
1489 static int tun_get_settings(struct net_device *dev, struct ethtool_cmd *cmd)
1490 {
1491 	cmd->supported		= 0;
1492 	cmd->advertising	= 0;
1493 	cmd->speed		= SPEED_10;
1494 	cmd->duplex		= DUPLEX_FULL;
1495 	cmd->port		= PORT_TP;
1496 	cmd->phy_address	= 0;
1497 	cmd->transceiver	= XCVR_INTERNAL;
1498 	cmd->autoneg		= AUTONEG_DISABLE;
1499 	cmd->maxtxpkt		= 0;
1500 	cmd->maxrxpkt		= 0;
1501 	return 0;
1502 }
1503 
1504 static void tun_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
1505 {
1506 	struct tun_struct *tun = netdev_priv(dev);
1507 
1508 	strcpy(info->driver, DRV_NAME);
1509 	strcpy(info->version, DRV_VERSION);
1510 	strcpy(info->fw_version, "N/A");
1511 
1512 	switch (tun->flags & TUN_TYPE_MASK) {
1513 	case TUN_TUN_DEV:
1514 		strcpy(info->bus_info, "tun");
1515 		break;
1516 	case TUN_TAP_DEV:
1517 		strcpy(info->bus_info, "tap");
1518 		break;
1519 	}
1520 }
1521 
1522 static u32 tun_get_msglevel(struct net_device *dev)
1523 {
1524 #ifdef TUN_DEBUG
1525 	struct tun_struct *tun = netdev_priv(dev);
1526 	return tun->debug;
1527 #else
1528 	return -EOPNOTSUPP;
1529 #endif
1530 }
1531 
1532 static void tun_set_msglevel(struct net_device *dev, u32 value)
1533 {
1534 #ifdef TUN_DEBUG
1535 	struct tun_struct *tun = netdev_priv(dev);
1536 	tun->debug = value;
1537 #endif
1538 }
1539 
1540 static u32 tun_get_link(struct net_device *dev)
1541 {
1542 	struct tun_struct *tun = netdev_priv(dev);
1543 	return !!tun->tfile;
1544 }
1545 
1546 static u32 tun_get_rx_csum(struct net_device *dev)
1547 {
1548 	struct tun_struct *tun = netdev_priv(dev);
1549 	return (tun->flags & TUN_NOCHECKSUM) == 0;
1550 }
1551 
1552 static int tun_set_rx_csum(struct net_device *dev, u32 data)
1553 {
1554 	struct tun_struct *tun = netdev_priv(dev);
1555 	if (data)
1556 		tun->flags &= ~TUN_NOCHECKSUM;
1557 	else
1558 		tun->flags |= TUN_NOCHECKSUM;
1559 	return 0;
1560 }
1561 
1562 static const struct ethtool_ops tun_ethtool_ops = {
1563 	.get_settings	= tun_get_settings,
1564 	.get_drvinfo	= tun_get_drvinfo,
1565 	.get_msglevel	= tun_get_msglevel,
1566 	.set_msglevel	= tun_set_msglevel,
1567 	.get_link	= tun_get_link,
1568 	.get_rx_csum	= tun_get_rx_csum,
1569 	.set_rx_csum	= tun_set_rx_csum
1570 };
1571 
1572 
1573 static int __init tun_init(void)
1574 {
1575 	int ret = 0;
1576 
1577 	printk(KERN_INFO "tun: %s, %s\n", DRV_DESCRIPTION, DRV_VERSION);
1578 	printk(KERN_INFO "tun: %s\n", DRV_COPYRIGHT);
1579 
1580 	ret = rtnl_link_register(&tun_link_ops);
1581 	if (ret) {
1582 		printk(KERN_ERR "tun: Can't register link_ops\n");
1583 		goto err_linkops;
1584 	}
1585 
1586 	ret = misc_register(&tun_miscdev);
1587 	if (ret) {
1588 		printk(KERN_ERR "tun: Can't register misc device %d\n", TUN_MINOR);
1589 		goto err_misc;
1590 	}
1591 	return  0;
1592 err_misc:
1593 	rtnl_link_unregister(&tun_link_ops);
1594 err_linkops:
1595 	return ret;
1596 }
1597 
1598 static void tun_cleanup(void)
1599 {
1600 	misc_deregister(&tun_miscdev);
1601 	rtnl_link_unregister(&tun_link_ops);
1602 }
1603 
1604 /* Get an underlying socket object from tun file.  Returns error unless file is
1605  * attached to a device.  The returned object works like a packet socket, it
1606  * can be used for sock_sendmsg/sock_recvmsg.  The caller is responsible for
1607  * holding a reference to the file for as long as the socket is in use. */
1608 struct socket *tun_get_socket(struct file *file)
1609 {
1610 	struct tun_struct *tun;
1611 	if (file->f_op != &tun_fops)
1612 		return ERR_PTR(-EINVAL);
1613 	tun = tun_get(file);
1614 	if (!tun)
1615 		return ERR_PTR(-EBADFD);
1616 	tun_put(tun);
1617 	return &tun->socket;
1618 }
1619 EXPORT_SYMBOL_GPL(tun_get_socket);
1620 
1621 module_init(tun_init);
1622 module_exit(tun_cleanup);
1623 MODULE_DESCRIPTION(DRV_DESCRIPTION);
1624 MODULE_AUTHOR(DRV_COPYRIGHT);
1625 MODULE_LICENSE("GPL");
1626 MODULE_ALIAS_MISCDEV(TUN_MINOR);
1627