xref: /openbmc/linux/drivers/net/tun.c (revision 565d76cb)
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 	struct fasync_struct	*fasync;
127 
128 	struct tap_filter       txflt;
129 	struct socket		socket;
130 	struct socket_wq	wq;
131 
132 	int			vnet_hdr_sz;
133 
134 #ifdef TUN_DEBUG
135 	int debug;
136 #endif
137 };
138 
139 struct tun_sock {
140 	struct sock		sk;
141 	struct tun_struct	*tun;
142 };
143 
144 static inline struct tun_sock *tun_sk(struct sock *sk)
145 {
146 	return container_of(sk, struct tun_sock, sk);
147 }
148 
149 static int tun_attach(struct tun_struct *tun, struct file *file)
150 {
151 	struct tun_file *tfile = file->private_data;
152 	int err;
153 
154 	ASSERT_RTNL();
155 
156 	netif_tx_lock_bh(tun->dev);
157 
158 	err = -EINVAL;
159 	if (tfile->tun)
160 		goto out;
161 
162 	err = -EBUSY;
163 	if (tun->tfile)
164 		goto out;
165 
166 	err = 0;
167 	tfile->tun = tun;
168 	tun->tfile = tfile;
169 	tun->socket.file = file;
170 	netif_carrier_on(tun->dev);
171 	dev_hold(tun->dev);
172 	sock_hold(tun->socket.sk);
173 	atomic_inc(&tfile->count);
174 
175 out:
176 	netif_tx_unlock_bh(tun->dev);
177 	return err;
178 }
179 
180 static void __tun_detach(struct tun_struct *tun)
181 {
182 	/* Detach from net device */
183 	netif_tx_lock_bh(tun->dev);
184 	netif_carrier_off(tun->dev);
185 	tun->tfile = NULL;
186 	tun->socket.file = NULL;
187 	netif_tx_unlock_bh(tun->dev);
188 
189 	/* Drop read queue */
190 	skb_queue_purge(&tun->socket.sk->sk_receive_queue);
191 
192 	/* Drop the extra count on the net device */
193 	dev_put(tun->dev);
194 }
195 
196 static void tun_detach(struct tun_struct *tun)
197 {
198 	rtnl_lock();
199 	__tun_detach(tun);
200 	rtnl_unlock();
201 }
202 
203 static struct tun_struct *__tun_get(struct tun_file *tfile)
204 {
205 	struct tun_struct *tun = NULL;
206 
207 	if (atomic_inc_not_zero(&tfile->count))
208 		tun = tfile->tun;
209 
210 	return tun;
211 }
212 
213 static struct tun_struct *tun_get(struct file *file)
214 {
215 	return __tun_get(file->private_data);
216 }
217 
218 static void tun_put(struct tun_struct *tun)
219 {
220 	struct tun_file *tfile = tun->tfile;
221 
222 	if (atomic_dec_and_test(&tfile->count))
223 		tun_detach(tfile->tun);
224 }
225 
226 /* TAP filtering */
227 static void addr_hash_set(u32 *mask, const u8 *addr)
228 {
229 	int n = ether_crc(ETH_ALEN, addr) >> 26;
230 	mask[n >> 5] |= (1 << (n & 31));
231 }
232 
233 static unsigned int addr_hash_test(const u32 *mask, const u8 *addr)
234 {
235 	int n = ether_crc(ETH_ALEN, addr) >> 26;
236 	return mask[n >> 5] & (1 << (n & 31));
237 }
238 
239 static int update_filter(struct tap_filter *filter, void __user *arg)
240 {
241 	struct { u8 u[ETH_ALEN]; } *addr;
242 	struct tun_filter uf;
243 	int err, alen, n, nexact;
244 
245 	if (copy_from_user(&uf, arg, sizeof(uf)))
246 		return -EFAULT;
247 
248 	if (!uf.count) {
249 		/* Disabled */
250 		filter->count = 0;
251 		return 0;
252 	}
253 
254 	alen = ETH_ALEN * uf.count;
255 	addr = kmalloc(alen, GFP_KERNEL);
256 	if (!addr)
257 		return -ENOMEM;
258 
259 	if (copy_from_user(addr, arg + sizeof(uf), alen)) {
260 		err = -EFAULT;
261 		goto done;
262 	}
263 
264 	/* The filter is updated without holding any locks. Which is
265 	 * perfectly safe. We disable it first and in the worst
266 	 * case we'll accept a few undesired packets. */
267 	filter->count = 0;
268 	wmb();
269 
270 	/* Use first set of addresses as an exact filter */
271 	for (n = 0; n < uf.count && n < FLT_EXACT_COUNT; n++)
272 		memcpy(filter->addr[n], addr[n].u, ETH_ALEN);
273 
274 	nexact = n;
275 
276 	/* Remaining multicast addresses are hashed,
277 	 * unicast will leave the filter disabled. */
278 	memset(filter->mask, 0, sizeof(filter->mask));
279 	for (; n < uf.count; n++) {
280 		if (!is_multicast_ether_addr(addr[n].u)) {
281 			err = 0; /* no filter */
282 			goto done;
283 		}
284 		addr_hash_set(filter->mask, addr[n].u);
285 	}
286 
287 	/* For ALLMULTI just set the mask to all ones.
288 	 * This overrides the mask populated above. */
289 	if ((uf.flags & TUN_FLT_ALLMULTI))
290 		memset(filter->mask, ~0, sizeof(filter->mask));
291 
292 	/* Now enable the filter */
293 	wmb();
294 	filter->count = nexact;
295 
296 	/* Return the number of exact filters */
297 	err = nexact;
298 
299 done:
300 	kfree(addr);
301 	return err;
302 }
303 
304 /* Returns: 0 - drop, !=0 - accept */
305 static int run_filter(struct tap_filter *filter, const struct sk_buff *skb)
306 {
307 	/* Cannot use eth_hdr(skb) here because skb_mac_hdr() is incorrect
308 	 * at this point. */
309 	struct ethhdr *eh = (struct ethhdr *) skb->data;
310 	int i;
311 
312 	/* Exact match */
313 	for (i = 0; i < filter->count; i++)
314 		if (!compare_ether_addr(eh->h_dest, filter->addr[i]))
315 			return 1;
316 
317 	/* Inexact match (multicast only) */
318 	if (is_multicast_ether_addr(eh->h_dest))
319 		return addr_hash_test(filter->mask, eh->h_dest);
320 
321 	return 0;
322 }
323 
324 /*
325  * Checks whether the packet is accepted or not.
326  * Returns: 0 - drop, !=0 - accept
327  */
328 static int check_filter(struct tap_filter *filter, const struct sk_buff *skb)
329 {
330 	if (!filter->count)
331 		return 1;
332 
333 	return run_filter(filter, skb);
334 }
335 
336 /* Network device part of the driver */
337 
338 static const struct ethtool_ops tun_ethtool_ops;
339 
340 /* Net device detach from fd. */
341 static void tun_net_uninit(struct net_device *dev)
342 {
343 	struct tun_struct *tun = netdev_priv(dev);
344 	struct tun_file *tfile = tun->tfile;
345 
346 	/* Inform the methods they need to stop using the dev.
347 	 */
348 	if (tfile) {
349 		wake_up_all(&tun->wq.wait);
350 		if (atomic_dec_and_test(&tfile->count))
351 			__tun_detach(tun);
352 	}
353 }
354 
355 static void tun_free_netdev(struct net_device *dev)
356 {
357 	struct tun_struct *tun = netdev_priv(dev);
358 
359 	sock_put(tun->socket.sk);
360 }
361 
362 /* Net device open. */
363 static int tun_net_open(struct net_device *dev)
364 {
365 	netif_start_queue(dev);
366 	return 0;
367 }
368 
369 /* Net device close. */
370 static int tun_net_close(struct net_device *dev)
371 {
372 	netif_stop_queue(dev);
373 	return 0;
374 }
375 
376 /* Net device start xmit */
377 static netdev_tx_t tun_net_xmit(struct sk_buff *skb, struct net_device *dev)
378 {
379 	struct tun_struct *tun = netdev_priv(dev);
380 
381 	tun_debug(KERN_INFO, tun, "tun_net_xmit %d\n", skb->len);
382 
383 	/* Drop packet if interface is not attached */
384 	if (!tun->tfile)
385 		goto drop;
386 
387 	/* Drop if the filter does not like it.
388 	 * This is a noop if the filter is disabled.
389 	 * Filter can be enabled only for the TAP devices. */
390 	if (!check_filter(&tun->txflt, skb))
391 		goto drop;
392 
393 	if (tun->socket.sk->sk_filter &&
394 	    sk_filter(tun->socket.sk, skb))
395 		goto drop;
396 
397 	if (skb_queue_len(&tun->socket.sk->sk_receive_queue) >= dev->tx_queue_len) {
398 		if (!(tun->flags & TUN_ONE_QUEUE)) {
399 			/* Normal queueing mode. */
400 			/* Packet scheduler handles dropping of further packets. */
401 			netif_stop_queue(dev);
402 
403 			/* We won't see all dropped packets individually, so overrun
404 			 * error is more appropriate. */
405 			dev->stats.tx_fifo_errors++;
406 		} else {
407 			/* Single queue mode.
408 			 * Driver handles dropping of all packets itself. */
409 			goto drop;
410 		}
411 	}
412 
413 	/* Orphan the skb - required as we might hang on to it
414 	 * for indefinite time. */
415 	skb_orphan(skb);
416 
417 	/* Enqueue packet */
418 	skb_queue_tail(&tun->socket.sk->sk_receive_queue, skb);
419 
420 	/* Notify and wake up reader process */
421 	if (tun->flags & TUN_FASYNC)
422 		kill_fasync(&tun->fasync, SIGIO, POLL_IN);
423 	wake_up_interruptible_poll(&tun->wq.wait, POLLIN |
424 				   POLLRDNORM | POLLRDBAND);
425 	return NETDEV_TX_OK;
426 
427 drop:
428 	dev->stats.tx_dropped++;
429 	kfree_skb(skb);
430 	return NETDEV_TX_OK;
431 }
432 
433 static void tun_net_mclist(struct net_device *dev)
434 {
435 	/*
436 	 * This callback is supposed to deal with mc filter in
437 	 * _rx_ path and has nothing to do with the _tx_ path.
438 	 * In rx path we always accept everything userspace gives us.
439 	 */
440 }
441 
442 #define MIN_MTU 68
443 #define MAX_MTU 65535
444 
445 static int
446 tun_net_change_mtu(struct net_device *dev, int new_mtu)
447 {
448 	if (new_mtu < MIN_MTU || new_mtu + dev->hard_header_len > MAX_MTU)
449 		return -EINVAL;
450 	dev->mtu = new_mtu;
451 	return 0;
452 }
453 
454 static const struct net_device_ops tun_netdev_ops = {
455 	.ndo_uninit		= tun_net_uninit,
456 	.ndo_open		= tun_net_open,
457 	.ndo_stop		= tun_net_close,
458 	.ndo_start_xmit		= tun_net_xmit,
459 	.ndo_change_mtu		= tun_net_change_mtu,
460 };
461 
462 static const struct net_device_ops tap_netdev_ops = {
463 	.ndo_uninit		= tun_net_uninit,
464 	.ndo_open		= tun_net_open,
465 	.ndo_stop		= tun_net_close,
466 	.ndo_start_xmit		= tun_net_xmit,
467 	.ndo_change_mtu		= tun_net_change_mtu,
468 	.ndo_set_multicast_list	= tun_net_mclist,
469 	.ndo_set_mac_address	= eth_mac_addr,
470 	.ndo_validate_addr	= eth_validate_addr,
471 };
472 
473 /* Initialize net device. */
474 static void tun_net_init(struct net_device *dev)
475 {
476 	struct tun_struct *tun = netdev_priv(dev);
477 
478 	switch (tun->flags & TUN_TYPE_MASK) {
479 	case TUN_TUN_DEV:
480 		dev->netdev_ops = &tun_netdev_ops;
481 
482 		/* Point-to-Point TUN Device */
483 		dev->hard_header_len = 0;
484 		dev->addr_len = 0;
485 		dev->mtu = 1500;
486 
487 		/* Zero header length */
488 		dev->type = ARPHRD_NONE;
489 		dev->flags = IFF_POINTOPOINT | IFF_NOARP | IFF_MULTICAST;
490 		dev->tx_queue_len = TUN_READQ_SIZE;  /* We prefer our own queue length */
491 		break;
492 
493 	case TUN_TAP_DEV:
494 		dev->netdev_ops = &tap_netdev_ops;
495 		/* Ethernet TAP Device */
496 		ether_setup(dev);
497 
498 		random_ether_addr(dev->dev_addr);
499 
500 		dev->tx_queue_len = TUN_READQ_SIZE;  /* We prefer our own queue length */
501 		break;
502 	}
503 }
504 
505 /* Character device part */
506 
507 /* Poll */
508 static unsigned int tun_chr_poll(struct file *file, poll_table * wait)
509 {
510 	struct tun_file *tfile = file->private_data;
511 	struct tun_struct *tun = __tun_get(tfile);
512 	struct sock *sk;
513 	unsigned int mask = 0;
514 
515 	if (!tun)
516 		return POLLERR;
517 
518 	sk = tun->socket.sk;
519 
520 	tun_debug(KERN_INFO, tun, "tun_chr_poll\n");
521 
522 	poll_wait(file, &tun->wq.wait, wait);
523 
524 	if (!skb_queue_empty(&sk->sk_receive_queue))
525 		mask |= POLLIN | POLLRDNORM;
526 
527 	if (sock_writeable(sk) ||
528 	    (!test_and_set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags) &&
529 	     sock_writeable(sk)))
530 		mask |= POLLOUT | POLLWRNORM;
531 
532 	if (tun->dev->reg_state != NETREG_REGISTERED)
533 		mask = POLLERR;
534 
535 	tun_put(tun);
536 	return mask;
537 }
538 
539 /* prepad is the amount to reserve at front.  len is length after that.
540  * linear is a hint as to how much to copy (usually headers). */
541 static inline struct sk_buff *tun_alloc_skb(struct tun_struct *tun,
542 					    size_t prepad, size_t len,
543 					    size_t linear, int noblock)
544 {
545 	struct sock *sk = tun->socket.sk;
546 	struct sk_buff *skb;
547 	int err;
548 
549 	sock_update_classid(sk);
550 
551 	/* Under a page?  Don't bother with paged skb. */
552 	if (prepad + len < PAGE_SIZE || !linear)
553 		linear = len;
554 
555 	skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
556 				   &err);
557 	if (!skb)
558 		return ERR_PTR(err);
559 
560 	skb_reserve(skb, prepad);
561 	skb_put(skb, linear);
562 	skb->data_len = len - linear;
563 	skb->len += len - linear;
564 
565 	return skb;
566 }
567 
568 /* Get packet from user space buffer */
569 static __inline__ ssize_t tun_get_user(struct tun_struct *tun,
570 				       const struct iovec *iv, size_t count,
571 				       int noblock)
572 {
573 	struct tun_pi pi = { 0, cpu_to_be16(ETH_P_IP) };
574 	struct sk_buff *skb;
575 	size_t len = count, align = 0;
576 	struct virtio_net_hdr gso = { 0 };
577 	int offset = 0;
578 
579 	if (!(tun->flags & TUN_NO_PI)) {
580 		if ((len -= sizeof(pi)) > count)
581 			return -EINVAL;
582 
583 		if (memcpy_fromiovecend((void *)&pi, iv, 0, sizeof(pi)))
584 			return -EFAULT;
585 		offset += sizeof(pi);
586 	}
587 
588 	if (tun->flags & TUN_VNET_HDR) {
589 		if ((len -= tun->vnet_hdr_sz) > count)
590 			return -EINVAL;
591 
592 		if (memcpy_fromiovecend((void *)&gso, iv, offset, sizeof(gso)))
593 			return -EFAULT;
594 
595 		if ((gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
596 		    gso.csum_start + gso.csum_offset + 2 > gso.hdr_len)
597 			gso.hdr_len = gso.csum_start + gso.csum_offset + 2;
598 
599 		if (gso.hdr_len > len)
600 			return -EINVAL;
601 		offset += tun->vnet_hdr_sz;
602 	}
603 
604 	if ((tun->flags & TUN_TYPE_MASK) == TUN_TAP_DEV) {
605 		align = NET_IP_ALIGN;
606 		if (unlikely(len < ETH_HLEN ||
607 			     (gso.hdr_len && gso.hdr_len < ETH_HLEN)))
608 			return -EINVAL;
609 	}
610 
611 	skb = tun_alloc_skb(tun, align, len, gso.hdr_len, noblock);
612 	if (IS_ERR(skb)) {
613 		if (PTR_ERR(skb) != -EAGAIN)
614 			tun->dev->stats.rx_dropped++;
615 		return PTR_ERR(skb);
616 	}
617 
618 	if (skb_copy_datagram_from_iovec(skb, 0, iv, offset, len)) {
619 		tun->dev->stats.rx_dropped++;
620 		kfree_skb(skb);
621 		return -EFAULT;
622 	}
623 
624 	if (gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
625 		if (!skb_partial_csum_set(skb, gso.csum_start,
626 					  gso.csum_offset)) {
627 			tun->dev->stats.rx_frame_errors++;
628 			kfree_skb(skb);
629 			return -EINVAL;
630 		}
631 	} else if (tun->flags & TUN_NOCHECKSUM)
632 		skb->ip_summed = CHECKSUM_UNNECESSARY;
633 
634 	switch (tun->flags & TUN_TYPE_MASK) {
635 	case TUN_TUN_DEV:
636 		if (tun->flags & TUN_NO_PI) {
637 			switch (skb->data[0] & 0xf0) {
638 			case 0x40:
639 				pi.proto = htons(ETH_P_IP);
640 				break;
641 			case 0x60:
642 				pi.proto = htons(ETH_P_IPV6);
643 				break;
644 			default:
645 				tun->dev->stats.rx_dropped++;
646 				kfree_skb(skb);
647 				return -EINVAL;
648 			}
649 		}
650 
651 		skb_reset_mac_header(skb);
652 		skb->protocol = pi.proto;
653 		skb->dev = tun->dev;
654 		break;
655 	case TUN_TAP_DEV:
656 		skb->protocol = eth_type_trans(skb, tun->dev);
657 		break;
658 	};
659 
660 	if (gso.gso_type != VIRTIO_NET_HDR_GSO_NONE) {
661 		pr_debug("GSO!\n");
662 		switch (gso.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
663 		case VIRTIO_NET_HDR_GSO_TCPV4:
664 			skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
665 			break;
666 		case VIRTIO_NET_HDR_GSO_TCPV6:
667 			skb_shinfo(skb)->gso_type = SKB_GSO_TCPV6;
668 			break;
669 		case VIRTIO_NET_HDR_GSO_UDP:
670 			skb_shinfo(skb)->gso_type = SKB_GSO_UDP;
671 			break;
672 		default:
673 			tun->dev->stats.rx_frame_errors++;
674 			kfree_skb(skb);
675 			return -EINVAL;
676 		}
677 
678 		if (gso.gso_type & VIRTIO_NET_HDR_GSO_ECN)
679 			skb_shinfo(skb)->gso_type |= SKB_GSO_TCP_ECN;
680 
681 		skb_shinfo(skb)->gso_size = gso.gso_size;
682 		if (skb_shinfo(skb)->gso_size == 0) {
683 			tun->dev->stats.rx_frame_errors++;
684 			kfree_skb(skb);
685 			return -EINVAL;
686 		}
687 
688 		/* Header must be checked, and gso_segs computed. */
689 		skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
690 		skb_shinfo(skb)->gso_segs = 0;
691 	}
692 
693 	netif_rx_ni(skb);
694 
695 	tun->dev->stats.rx_packets++;
696 	tun->dev->stats.rx_bytes += len;
697 
698 	return count;
699 }
700 
701 static ssize_t tun_chr_aio_write(struct kiocb *iocb, const struct iovec *iv,
702 			      unsigned long count, loff_t pos)
703 {
704 	struct file *file = iocb->ki_filp;
705 	struct tun_struct *tun = tun_get(file);
706 	ssize_t result;
707 
708 	if (!tun)
709 		return -EBADFD;
710 
711 	tun_debug(KERN_INFO, tun, "tun_chr_write %ld\n", count);
712 
713 	result = tun_get_user(tun, iv, iov_length(iv, count),
714 			      file->f_flags & O_NONBLOCK);
715 
716 	tun_put(tun);
717 	return result;
718 }
719 
720 /* Put packet to the user space buffer */
721 static __inline__ ssize_t tun_put_user(struct tun_struct *tun,
722 				       struct sk_buff *skb,
723 				       const struct iovec *iv, int len)
724 {
725 	struct tun_pi pi = { 0, skb->protocol };
726 	ssize_t total = 0;
727 
728 	if (!(tun->flags & TUN_NO_PI)) {
729 		if ((len -= sizeof(pi)) < 0)
730 			return -EINVAL;
731 
732 		if (len < skb->len) {
733 			/* Packet will be striped */
734 			pi.flags |= TUN_PKT_STRIP;
735 		}
736 
737 		if (memcpy_toiovecend(iv, (void *) &pi, 0, sizeof(pi)))
738 			return -EFAULT;
739 		total += sizeof(pi);
740 	}
741 
742 	if (tun->flags & TUN_VNET_HDR) {
743 		struct virtio_net_hdr gso = { 0 }; /* no info leak */
744 		if ((len -= tun->vnet_hdr_sz) < 0)
745 			return -EINVAL;
746 
747 		if (skb_is_gso(skb)) {
748 			struct skb_shared_info *sinfo = skb_shinfo(skb);
749 
750 			/* This is a hint as to how much should be linear. */
751 			gso.hdr_len = skb_headlen(skb);
752 			gso.gso_size = sinfo->gso_size;
753 			if (sinfo->gso_type & SKB_GSO_TCPV4)
754 				gso.gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
755 			else if (sinfo->gso_type & SKB_GSO_TCPV6)
756 				gso.gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
757 			else if (sinfo->gso_type & SKB_GSO_UDP)
758 				gso.gso_type = VIRTIO_NET_HDR_GSO_UDP;
759 			else {
760 				pr_err("unexpected GSO type: "
761 				       "0x%x, gso_size %d, hdr_len %d\n",
762 				       sinfo->gso_type, gso.gso_size,
763 				       gso.hdr_len);
764 				print_hex_dump(KERN_ERR, "tun: ",
765 					       DUMP_PREFIX_NONE,
766 					       16, 1, skb->head,
767 					       min((int)gso.hdr_len, 64), true);
768 				WARN_ON_ONCE(1);
769 				return -EINVAL;
770 			}
771 			if (sinfo->gso_type & SKB_GSO_TCP_ECN)
772 				gso.gso_type |= VIRTIO_NET_HDR_GSO_ECN;
773 		} else
774 			gso.gso_type = VIRTIO_NET_HDR_GSO_NONE;
775 
776 		if (skb->ip_summed == CHECKSUM_PARTIAL) {
777 			gso.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
778 			gso.csum_start = skb_checksum_start_offset(skb);
779 			gso.csum_offset = skb->csum_offset;
780 		} /* else everything is zero */
781 
782 		if (unlikely(memcpy_toiovecend(iv, (void *)&gso, total,
783 					       sizeof(gso))))
784 			return -EFAULT;
785 		total += tun->vnet_hdr_sz;
786 	}
787 
788 	len = min_t(int, skb->len, len);
789 
790 	skb_copy_datagram_const_iovec(skb, 0, iv, total, len);
791 	total += skb->len;
792 
793 	tun->dev->stats.tx_packets++;
794 	tun->dev->stats.tx_bytes += len;
795 
796 	return total;
797 }
798 
799 static ssize_t tun_do_read(struct tun_struct *tun,
800 			   struct kiocb *iocb, const struct iovec *iv,
801 			   ssize_t len, int noblock)
802 {
803 	DECLARE_WAITQUEUE(wait, current);
804 	struct sk_buff *skb;
805 	ssize_t ret = 0;
806 
807 	tun_debug(KERN_INFO, tun, "tun_chr_read\n");
808 
809 	add_wait_queue(&tun->wq.wait, &wait);
810 	while (len) {
811 		current->state = TASK_INTERRUPTIBLE;
812 
813 		/* Read frames from the queue */
814 		if (!(skb=skb_dequeue(&tun->socket.sk->sk_receive_queue))) {
815 			if (noblock) {
816 				ret = -EAGAIN;
817 				break;
818 			}
819 			if (signal_pending(current)) {
820 				ret = -ERESTARTSYS;
821 				break;
822 			}
823 			if (tun->dev->reg_state != NETREG_REGISTERED) {
824 				ret = -EIO;
825 				break;
826 			}
827 
828 			/* Nothing to read, let's sleep */
829 			schedule();
830 			continue;
831 		}
832 		netif_wake_queue(tun->dev);
833 
834 		ret = tun_put_user(tun, skb, iv, len);
835 		kfree_skb(skb);
836 		break;
837 	}
838 
839 	current->state = TASK_RUNNING;
840 	remove_wait_queue(&tun->wq.wait, &wait);
841 
842 	return ret;
843 }
844 
845 static ssize_t tun_chr_aio_read(struct kiocb *iocb, const struct iovec *iv,
846 			    unsigned long count, loff_t pos)
847 {
848 	struct file *file = iocb->ki_filp;
849 	struct tun_file *tfile = file->private_data;
850 	struct tun_struct *tun = __tun_get(tfile);
851 	ssize_t len, ret;
852 
853 	if (!tun)
854 		return -EBADFD;
855 	len = iov_length(iv, count);
856 	if (len < 0) {
857 		ret = -EINVAL;
858 		goto out;
859 	}
860 
861 	ret = tun_do_read(tun, iocb, iv, len, file->f_flags & O_NONBLOCK);
862 	ret = min_t(ssize_t, ret, len);
863 out:
864 	tun_put(tun);
865 	return ret;
866 }
867 
868 static void tun_setup(struct net_device *dev)
869 {
870 	struct tun_struct *tun = netdev_priv(dev);
871 
872 	tun->owner = -1;
873 	tun->group = -1;
874 
875 	dev->ethtool_ops = &tun_ethtool_ops;
876 	dev->destructor = tun_free_netdev;
877 }
878 
879 /* Trivial set of netlink ops to allow deleting tun or tap
880  * device with netlink.
881  */
882 static int tun_validate(struct nlattr *tb[], struct nlattr *data[])
883 {
884 	return -EINVAL;
885 }
886 
887 static struct rtnl_link_ops tun_link_ops __read_mostly = {
888 	.kind		= DRV_NAME,
889 	.priv_size	= sizeof(struct tun_struct),
890 	.setup		= tun_setup,
891 	.validate	= tun_validate,
892 };
893 
894 static void tun_sock_write_space(struct sock *sk)
895 {
896 	struct tun_struct *tun;
897 	wait_queue_head_t *wqueue;
898 
899 	if (!sock_writeable(sk))
900 		return;
901 
902 	if (!test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags))
903 		return;
904 
905 	wqueue = sk_sleep(sk);
906 	if (wqueue && waitqueue_active(wqueue))
907 		wake_up_interruptible_sync_poll(wqueue, POLLOUT |
908 						POLLWRNORM | POLLWRBAND);
909 
910 	tun = tun_sk(sk)->tun;
911 	kill_fasync(&tun->fasync, SIGIO, POLL_OUT);
912 }
913 
914 static void tun_sock_destruct(struct sock *sk)
915 {
916 	free_netdev(tun_sk(sk)->tun->dev);
917 }
918 
919 static int tun_sendmsg(struct kiocb *iocb, struct socket *sock,
920 		       struct msghdr *m, size_t total_len)
921 {
922 	struct tun_struct *tun = container_of(sock, struct tun_struct, socket);
923 	return tun_get_user(tun, m->msg_iov, total_len,
924 			    m->msg_flags & MSG_DONTWAIT);
925 }
926 
927 static int tun_recvmsg(struct kiocb *iocb, struct socket *sock,
928 		       struct msghdr *m, size_t total_len,
929 		       int flags)
930 {
931 	struct tun_struct *tun = container_of(sock, struct tun_struct, socket);
932 	int ret;
933 	if (flags & ~(MSG_DONTWAIT|MSG_TRUNC))
934 		return -EINVAL;
935 	ret = tun_do_read(tun, iocb, m->msg_iov, total_len,
936 			  flags & MSG_DONTWAIT);
937 	if (ret > total_len) {
938 		m->msg_flags |= MSG_TRUNC;
939 		ret = flags & MSG_TRUNC ? ret : total_len;
940 	}
941 	return ret;
942 }
943 
944 /* Ops structure to mimic raw sockets with tun */
945 static const struct proto_ops tun_socket_ops = {
946 	.sendmsg = tun_sendmsg,
947 	.recvmsg = tun_recvmsg,
948 };
949 
950 static struct proto tun_proto = {
951 	.name		= "tun",
952 	.owner		= THIS_MODULE,
953 	.obj_size	= sizeof(struct tun_sock),
954 };
955 
956 static int tun_flags(struct tun_struct *tun)
957 {
958 	int flags = 0;
959 
960 	if (tun->flags & TUN_TUN_DEV)
961 		flags |= IFF_TUN;
962 	else
963 		flags |= IFF_TAP;
964 
965 	if (tun->flags & TUN_NO_PI)
966 		flags |= IFF_NO_PI;
967 
968 	if (tun->flags & TUN_ONE_QUEUE)
969 		flags |= IFF_ONE_QUEUE;
970 
971 	if (tun->flags & TUN_VNET_HDR)
972 		flags |= IFF_VNET_HDR;
973 
974 	return flags;
975 }
976 
977 static ssize_t tun_show_flags(struct device *dev, struct device_attribute *attr,
978 			      char *buf)
979 {
980 	struct tun_struct *tun = netdev_priv(to_net_dev(dev));
981 	return sprintf(buf, "0x%x\n", tun_flags(tun));
982 }
983 
984 static ssize_t tun_show_owner(struct device *dev, struct device_attribute *attr,
985 			      char *buf)
986 {
987 	struct tun_struct *tun = netdev_priv(to_net_dev(dev));
988 	return sprintf(buf, "%d\n", tun->owner);
989 }
990 
991 static ssize_t tun_show_group(struct device *dev, struct device_attribute *attr,
992 			      char *buf)
993 {
994 	struct tun_struct *tun = netdev_priv(to_net_dev(dev));
995 	return sprintf(buf, "%d\n", tun->group);
996 }
997 
998 static DEVICE_ATTR(tun_flags, 0444, tun_show_flags, NULL);
999 static DEVICE_ATTR(owner, 0444, tun_show_owner, NULL);
1000 static DEVICE_ATTR(group, 0444, tun_show_group, NULL);
1001 
1002 static int tun_set_iff(struct net *net, struct file *file, struct ifreq *ifr)
1003 {
1004 	struct sock *sk;
1005 	struct tun_struct *tun;
1006 	struct net_device *dev;
1007 	int err;
1008 
1009 	dev = __dev_get_by_name(net, ifr->ifr_name);
1010 	if (dev) {
1011 		const struct cred *cred = current_cred();
1012 
1013 		if (ifr->ifr_flags & IFF_TUN_EXCL)
1014 			return -EBUSY;
1015 		if ((ifr->ifr_flags & IFF_TUN) && dev->netdev_ops == &tun_netdev_ops)
1016 			tun = netdev_priv(dev);
1017 		else if ((ifr->ifr_flags & IFF_TAP) && dev->netdev_ops == &tap_netdev_ops)
1018 			tun = netdev_priv(dev);
1019 		else
1020 			return -EINVAL;
1021 
1022 		if (((tun->owner != -1 && cred->euid != tun->owner) ||
1023 		     (tun->group != -1 && !in_egroup_p(tun->group))) &&
1024 		    !capable(CAP_NET_ADMIN))
1025 			return -EPERM;
1026 		err = security_tun_dev_attach(tun->socket.sk);
1027 		if (err < 0)
1028 			return err;
1029 
1030 		err = tun_attach(tun, file);
1031 		if (err < 0)
1032 			return err;
1033 	}
1034 	else {
1035 		char *name;
1036 		unsigned long flags = 0;
1037 
1038 		if (!capable(CAP_NET_ADMIN))
1039 			return -EPERM;
1040 		err = security_tun_dev_create();
1041 		if (err < 0)
1042 			return err;
1043 
1044 		/* Set dev type */
1045 		if (ifr->ifr_flags & IFF_TUN) {
1046 			/* TUN device */
1047 			flags |= TUN_TUN_DEV;
1048 			name = "tun%d";
1049 		} else if (ifr->ifr_flags & IFF_TAP) {
1050 			/* TAP device */
1051 			flags |= TUN_TAP_DEV;
1052 			name = "tap%d";
1053 		} else
1054 			return -EINVAL;
1055 
1056 		if (*ifr->ifr_name)
1057 			name = ifr->ifr_name;
1058 
1059 		dev = alloc_netdev(sizeof(struct tun_struct), name,
1060 				   tun_setup);
1061 		if (!dev)
1062 			return -ENOMEM;
1063 
1064 		dev_net_set(dev, net);
1065 		dev->rtnl_link_ops = &tun_link_ops;
1066 
1067 		tun = netdev_priv(dev);
1068 		tun->dev = dev;
1069 		tun->flags = flags;
1070 		tun->txflt.count = 0;
1071 		tun->vnet_hdr_sz = sizeof(struct virtio_net_hdr);
1072 
1073 		err = -ENOMEM;
1074 		sk = sk_alloc(net, AF_UNSPEC, GFP_KERNEL, &tun_proto);
1075 		if (!sk)
1076 			goto err_free_dev;
1077 
1078 		tun->socket.wq = &tun->wq;
1079 		init_waitqueue_head(&tun->wq.wait);
1080 		tun->socket.ops = &tun_socket_ops;
1081 		sock_init_data(&tun->socket, sk);
1082 		sk->sk_write_space = tun_sock_write_space;
1083 		sk->sk_sndbuf = INT_MAX;
1084 
1085 		tun_sk(sk)->tun = tun;
1086 
1087 		security_tun_dev_post_create(sk);
1088 
1089 		tun_net_init(dev);
1090 
1091 		if (strchr(dev->name, '%')) {
1092 			err = dev_alloc_name(dev, dev->name);
1093 			if (err < 0)
1094 				goto err_free_sk;
1095 		}
1096 
1097 		err = register_netdevice(tun->dev);
1098 		if (err < 0)
1099 			goto err_free_sk;
1100 
1101 		if (device_create_file(&tun->dev->dev, &dev_attr_tun_flags) ||
1102 		    device_create_file(&tun->dev->dev, &dev_attr_owner) ||
1103 		    device_create_file(&tun->dev->dev, &dev_attr_group))
1104 			pr_err("Failed to create tun sysfs files\n");
1105 
1106 		sk->sk_destruct = tun_sock_destruct;
1107 
1108 		err = tun_attach(tun, file);
1109 		if (err < 0)
1110 			goto failed;
1111 	}
1112 
1113 	tun_debug(KERN_INFO, tun, "tun_set_iff\n");
1114 
1115 	if (ifr->ifr_flags & IFF_NO_PI)
1116 		tun->flags |= TUN_NO_PI;
1117 	else
1118 		tun->flags &= ~TUN_NO_PI;
1119 
1120 	if (ifr->ifr_flags & IFF_ONE_QUEUE)
1121 		tun->flags |= TUN_ONE_QUEUE;
1122 	else
1123 		tun->flags &= ~TUN_ONE_QUEUE;
1124 
1125 	if (ifr->ifr_flags & IFF_VNET_HDR)
1126 		tun->flags |= TUN_VNET_HDR;
1127 	else
1128 		tun->flags &= ~TUN_VNET_HDR;
1129 
1130 	/* Make sure persistent devices do not get stuck in
1131 	 * xoff state.
1132 	 */
1133 	if (netif_running(tun->dev))
1134 		netif_wake_queue(tun->dev);
1135 
1136 	strcpy(ifr->ifr_name, tun->dev->name);
1137 	return 0;
1138 
1139  err_free_sk:
1140 	sock_put(sk);
1141  err_free_dev:
1142 	free_netdev(dev);
1143  failed:
1144 	return err;
1145 }
1146 
1147 static int tun_get_iff(struct net *net, struct tun_struct *tun,
1148 		       struct ifreq *ifr)
1149 {
1150 	tun_debug(KERN_INFO, tun, "tun_get_iff\n");
1151 
1152 	strcpy(ifr->ifr_name, tun->dev->name);
1153 
1154 	ifr->ifr_flags = tun_flags(tun);
1155 
1156 	return 0;
1157 }
1158 
1159 /* This is like a cut-down ethtool ops, except done via tun fd so no
1160  * privs required. */
1161 static int set_offload(struct net_device *dev, unsigned long arg)
1162 {
1163 	u32 old_features, features;
1164 
1165 	old_features = dev->features;
1166 	/* Unset features, set them as we chew on the arg. */
1167 	features = (old_features & ~(NETIF_F_HW_CSUM|NETIF_F_SG|NETIF_F_FRAGLIST
1168 				    |NETIF_F_TSO_ECN|NETIF_F_TSO|NETIF_F_TSO6
1169 				    |NETIF_F_UFO));
1170 
1171 	if (arg & TUN_F_CSUM) {
1172 		features |= NETIF_F_HW_CSUM|NETIF_F_SG|NETIF_F_FRAGLIST;
1173 		arg &= ~TUN_F_CSUM;
1174 
1175 		if (arg & (TUN_F_TSO4|TUN_F_TSO6)) {
1176 			if (arg & TUN_F_TSO_ECN) {
1177 				features |= NETIF_F_TSO_ECN;
1178 				arg &= ~TUN_F_TSO_ECN;
1179 			}
1180 			if (arg & TUN_F_TSO4)
1181 				features |= NETIF_F_TSO;
1182 			if (arg & TUN_F_TSO6)
1183 				features |= NETIF_F_TSO6;
1184 			arg &= ~(TUN_F_TSO4|TUN_F_TSO6);
1185 		}
1186 
1187 		if (arg & TUN_F_UFO) {
1188 			features |= NETIF_F_UFO;
1189 			arg &= ~TUN_F_UFO;
1190 		}
1191 	}
1192 
1193 	/* This gives the user a way to test for new features in future by
1194 	 * trying to set them. */
1195 	if (arg)
1196 		return -EINVAL;
1197 
1198 	dev->features = features;
1199 	if (old_features != dev->features)
1200 		netdev_features_change(dev);
1201 
1202 	return 0;
1203 }
1204 
1205 static long __tun_chr_ioctl(struct file *file, unsigned int cmd,
1206 			    unsigned long arg, int ifreq_len)
1207 {
1208 	struct tun_file *tfile = file->private_data;
1209 	struct tun_struct *tun;
1210 	void __user* argp = (void __user*)arg;
1211 	struct sock_fprog fprog;
1212 	struct ifreq ifr;
1213 	int sndbuf;
1214 	int vnet_hdr_sz;
1215 	int ret;
1216 
1217 	if (cmd == TUNSETIFF || _IOC_TYPE(cmd) == 0x89)
1218 		if (copy_from_user(&ifr, argp, ifreq_len))
1219 			return -EFAULT;
1220 
1221 	if (cmd == TUNGETFEATURES) {
1222 		/* Currently this just means: "what IFF flags are valid?".
1223 		 * This is needed because we never checked for invalid flags on
1224 		 * TUNSETIFF. */
1225 		return put_user(IFF_TUN | IFF_TAP | IFF_NO_PI | IFF_ONE_QUEUE |
1226 				IFF_VNET_HDR,
1227 				(unsigned int __user*)argp);
1228 	}
1229 
1230 	rtnl_lock();
1231 
1232 	tun = __tun_get(tfile);
1233 	if (cmd == TUNSETIFF && !tun) {
1234 		ifr.ifr_name[IFNAMSIZ-1] = '\0';
1235 
1236 		ret = tun_set_iff(tfile->net, file, &ifr);
1237 
1238 		if (ret)
1239 			goto unlock;
1240 
1241 		if (copy_to_user(argp, &ifr, ifreq_len))
1242 			ret = -EFAULT;
1243 		goto unlock;
1244 	}
1245 
1246 	ret = -EBADFD;
1247 	if (!tun)
1248 		goto unlock;
1249 
1250 	tun_debug(KERN_INFO, tun, "tun_chr_ioctl cmd %d\n", cmd);
1251 
1252 	ret = 0;
1253 	switch (cmd) {
1254 	case TUNGETIFF:
1255 		ret = tun_get_iff(current->nsproxy->net_ns, tun, &ifr);
1256 		if (ret)
1257 			break;
1258 
1259 		if (copy_to_user(argp, &ifr, ifreq_len))
1260 			ret = -EFAULT;
1261 		break;
1262 
1263 	case TUNSETNOCSUM:
1264 		/* Disable/Enable checksum */
1265 		if (arg)
1266 			tun->flags |= TUN_NOCHECKSUM;
1267 		else
1268 			tun->flags &= ~TUN_NOCHECKSUM;
1269 
1270 		tun_debug(KERN_INFO, tun, "checksum %s\n",
1271 			  arg ? "disabled" : "enabled");
1272 		break;
1273 
1274 	case TUNSETPERSIST:
1275 		/* Disable/Enable persist mode */
1276 		if (arg)
1277 			tun->flags |= TUN_PERSIST;
1278 		else
1279 			tun->flags &= ~TUN_PERSIST;
1280 
1281 		tun_debug(KERN_INFO, tun, "persist %s\n",
1282 			  arg ? "enabled" : "disabled");
1283 		break;
1284 
1285 	case TUNSETOWNER:
1286 		/* Set owner of the device */
1287 		tun->owner = (uid_t) arg;
1288 
1289 		tun_debug(KERN_INFO, tun, "owner set to %d\n", tun->owner);
1290 		break;
1291 
1292 	case TUNSETGROUP:
1293 		/* Set group of the device */
1294 		tun->group= (gid_t) arg;
1295 
1296 		tun_debug(KERN_INFO, tun, "group set to %d\n", tun->group);
1297 		break;
1298 
1299 	case TUNSETLINK:
1300 		/* Only allow setting the type when the interface is down */
1301 		if (tun->dev->flags & IFF_UP) {
1302 			tun_debug(KERN_INFO, tun,
1303 				  "Linktype set failed because interface is up\n");
1304 			ret = -EBUSY;
1305 		} else {
1306 			tun->dev->type = (int) arg;
1307 			tun_debug(KERN_INFO, tun, "linktype set to %d\n",
1308 				  tun->dev->type);
1309 			ret = 0;
1310 		}
1311 		break;
1312 
1313 #ifdef TUN_DEBUG
1314 	case TUNSETDEBUG:
1315 		tun->debug = arg;
1316 		break;
1317 #endif
1318 	case TUNSETOFFLOAD:
1319 		ret = set_offload(tun->dev, arg);
1320 		break;
1321 
1322 	case TUNSETTXFILTER:
1323 		/* Can be set only for TAPs */
1324 		ret = -EINVAL;
1325 		if ((tun->flags & TUN_TYPE_MASK) != TUN_TAP_DEV)
1326 			break;
1327 		ret = update_filter(&tun->txflt, (void __user *)arg);
1328 		break;
1329 
1330 	case SIOCGIFHWADDR:
1331 		/* Get hw address */
1332 		memcpy(ifr.ifr_hwaddr.sa_data, tun->dev->dev_addr, ETH_ALEN);
1333 		ifr.ifr_hwaddr.sa_family = tun->dev->type;
1334 		if (copy_to_user(argp, &ifr, ifreq_len))
1335 			ret = -EFAULT;
1336 		break;
1337 
1338 	case SIOCSIFHWADDR:
1339 		/* Set hw address */
1340 		tun_debug(KERN_DEBUG, tun, "set hw address: %pM\n",
1341 			  ifr.ifr_hwaddr.sa_data);
1342 
1343 		ret = dev_set_mac_address(tun->dev, &ifr.ifr_hwaddr);
1344 		break;
1345 
1346 	case TUNGETSNDBUF:
1347 		sndbuf = tun->socket.sk->sk_sndbuf;
1348 		if (copy_to_user(argp, &sndbuf, sizeof(sndbuf)))
1349 			ret = -EFAULT;
1350 		break;
1351 
1352 	case TUNSETSNDBUF:
1353 		if (copy_from_user(&sndbuf, argp, sizeof(sndbuf))) {
1354 			ret = -EFAULT;
1355 			break;
1356 		}
1357 
1358 		tun->socket.sk->sk_sndbuf = sndbuf;
1359 		break;
1360 
1361 	case TUNGETVNETHDRSZ:
1362 		vnet_hdr_sz = tun->vnet_hdr_sz;
1363 		if (copy_to_user(argp, &vnet_hdr_sz, sizeof(vnet_hdr_sz)))
1364 			ret = -EFAULT;
1365 		break;
1366 
1367 	case TUNSETVNETHDRSZ:
1368 		if (copy_from_user(&vnet_hdr_sz, argp, sizeof(vnet_hdr_sz))) {
1369 			ret = -EFAULT;
1370 			break;
1371 		}
1372 		if (vnet_hdr_sz < (int)sizeof(struct virtio_net_hdr)) {
1373 			ret = -EINVAL;
1374 			break;
1375 		}
1376 
1377 		tun->vnet_hdr_sz = vnet_hdr_sz;
1378 		break;
1379 
1380 	case TUNATTACHFILTER:
1381 		/* Can be set only for TAPs */
1382 		ret = -EINVAL;
1383 		if ((tun->flags & TUN_TYPE_MASK) != TUN_TAP_DEV)
1384 			break;
1385 		ret = -EFAULT;
1386 		if (copy_from_user(&fprog, argp, sizeof(fprog)))
1387 			break;
1388 
1389 		ret = sk_attach_filter(&fprog, tun->socket.sk);
1390 		break;
1391 
1392 	case TUNDETACHFILTER:
1393 		/* Can be set only for TAPs */
1394 		ret = -EINVAL;
1395 		if ((tun->flags & TUN_TYPE_MASK) != TUN_TAP_DEV)
1396 			break;
1397 		ret = sk_detach_filter(tun->socket.sk);
1398 		break;
1399 
1400 	default:
1401 		ret = -EINVAL;
1402 		break;
1403 	}
1404 
1405 unlock:
1406 	rtnl_unlock();
1407 	if (tun)
1408 		tun_put(tun);
1409 	return ret;
1410 }
1411 
1412 static long tun_chr_ioctl(struct file *file,
1413 			  unsigned int cmd, unsigned long arg)
1414 {
1415 	return __tun_chr_ioctl(file, cmd, arg, sizeof (struct ifreq));
1416 }
1417 
1418 #ifdef CONFIG_COMPAT
1419 static long tun_chr_compat_ioctl(struct file *file,
1420 			 unsigned int cmd, unsigned long arg)
1421 {
1422 	switch (cmd) {
1423 	case TUNSETIFF:
1424 	case TUNGETIFF:
1425 	case TUNSETTXFILTER:
1426 	case TUNGETSNDBUF:
1427 	case TUNSETSNDBUF:
1428 	case SIOCGIFHWADDR:
1429 	case SIOCSIFHWADDR:
1430 		arg = (unsigned long)compat_ptr(arg);
1431 		break;
1432 	default:
1433 		arg = (compat_ulong_t)arg;
1434 		break;
1435 	}
1436 
1437 	/*
1438 	 * compat_ifreq is shorter than ifreq, so we must not access beyond
1439 	 * the end of that structure. All fields that are used in this
1440 	 * driver are compatible though, we don't need to convert the
1441 	 * contents.
1442 	 */
1443 	return __tun_chr_ioctl(file, cmd, arg, sizeof(struct compat_ifreq));
1444 }
1445 #endif /* CONFIG_COMPAT */
1446 
1447 static int tun_chr_fasync(int fd, struct file *file, int on)
1448 {
1449 	struct tun_struct *tun = tun_get(file);
1450 	int ret;
1451 
1452 	if (!tun)
1453 		return -EBADFD;
1454 
1455 	tun_debug(KERN_INFO, tun, "tun_chr_fasync %d\n", on);
1456 
1457 	if ((ret = fasync_helper(fd, file, on, &tun->fasync)) < 0)
1458 		goto out;
1459 
1460 	if (on) {
1461 		ret = __f_setown(file, task_pid(current), PIDTYPE_PID, 0);
1462 		if (ret)
1463 			goto out;
1464 		tun->flags |= TUN_FASYNC;
1465 	} else
1466 		tun->flags &= ~TUN_FASYNC;
1467 	ret = 0;
1468 out:
1469 	tun_put(tun);
1470 	return ret;
1471 }
1472 
1473 static int tun_chr_open(struct inode *inode, struct file * file)
1474 {
1475 	struct tun_file *tfile;
1476 
1477 	DBG1(KERN_INFO, "tunX: tun_chr_open\n");
1478 
1479 	tfile = kmalloc(sizeof(*tfile), GFP_KERNEL);
1480 	if (!tfile)
1481 		return -ENOMEM;
1482 	atomic_set(&tfile->count, 0);
1483 	tfile->tun = NULL;
1484 	tfile->net = get_net(current->nsproxy->net_ns);
1485 	file->private_data = tfile;
1486 	return 0;
1487 }
1488 
1489 static int tun_chr_close(struct inode *inode, struct file *file)
1490 {
1491 	struct tun_file *tfile = file->private_data;
1492 	struct tun_struct *tun;
1493 
1494 	tun = __tun_get(tfile);
1495 	if (tun) {
1496 		struct net_device *dev = tun->dev;
1497 
1498 		tun_debug(KERN_INFO, tun, "tun_chr_close\n");
1499 
1500 		__tun_detach(tun);
1501 
1502 		/* If desirable, unregister the netdevice. */
1503 		if (!(tun->flags & TUN_PERSIST)) {
1504 			rtnl_lock();
1505 			if (dev->reg_state == NETREG_REGISTERED)
1506 				unregister_netdevice(dev);
1507 			rtnl_unlock();
1508 		}
1509 	}
1510 
1511 	tun = tfile->tun;
1512 	if (tun)
1513 		sock_put(tun->socket.sk);
1514 
1515 	put_net(tfile->net);
1516 	kfree(tfile);
1517 
1518 	return 0;
1519 }
1520 
1521 static const struct file_operations tun_fops = {
1522 	.owner	= THIS_MODULE,
1523 	.llseek = no_llseek,
1524 	.read  = do_sync_read,
1525 	.aio_read  = tun_chr_aio_read,
1526 	.write = do_sync_write,
1527 	.aio_write = tun_chr_aio_write,
1528 	.poll	= tun_chr_poll,
1529 	.unlocked_ioctl	= tun_chr_ioctl,
1530 #ifdef CONFIG_COMPAT
1531 	.compat_ioctl = tun_chr_compat_ioctl,
1532 #endif
1533 	.open	= tun_chr_open,
1534 	.release = tun_chr_close,
1535 	.fasync = tun_chr_fasync
1536 };
1537 
1538 static struct miscdevice tun_miscdev = {
1539 	.minor = TUN_MINOR,
1540 	.name = "tun",
1541 	.nodename = "net/tun",
1542 	.fops = &tun_fops,
1543 };
1544 
1545 /* ethtool interface */
1546 
1547 static int tun_get_settings(struct net_device *dev, struct ethtool_cmd *cmd)
1548 {
1549 	cmd->supported		= 0;
1550 	cmd->advertising	= 0;
1551 	cmd->speed		= SPEED_10;
1552 	cmd->duplex		= DUPLEX_FULL;
1553 	cmd->port		= PORT_TP;
1554 	cmd->phy_address	= 0;
1555 	cmd->transceiver	= XCVR_INTERNAL;
1556 	cmd->autoneg		= AUTONEG_DISABLE;
1557 	cmd->maxtxpkt		= 0;
1558 	cmd->maxrxpkt		= 0;
1559 	return 0;
1560 }
1561 
1562 static void tun_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
1563 {
1564 	struct tun_struct *tun = netdev_priv(dev);
1565 
1566 	strcpy(info->driver, DRV_NAME);
1567 	strcpy(info->version, DRV_VERSION);
1568 	strcpy(info->fw_version, "N/A");
1569 
1570 	switch (tun->flags & TUN_TYPE_MASK) {
1571 	case TUN_TUN_DEV:
1572 		strcpy(info->bus_info, "tun");
1573 		break;
1574 	case TUN_TAP_DEV:
1575 		strcpy(info->bus_info, "tap");
1576 		break;
1577 	}
1578 }
1579 
1580 static u32 tun_get_msglevel(struct net_device *dev)
1581 {
1582 #ifdef TUN_DEBUG
1583 	struct tun_struct *tun = netdev_priv(dev);
1584 	return tun->debug;
1585 #else
1586 	return -EOPNOTSUPP;
1587 #endif
1588 }
1589 
1590 static void tun_set_msglevel(struct net_device *dev, u32 value)
1591 {
1592 #ifdef TUN_DEBUG
1593 	struct tun_struct *tun = netdev_priv(dev);
1594 	tun->debug = value;
1595 #endif
1596 }
1597 
1598 static u32 tun_get_rx_csum(struct net_device *dev)
1599 {
1600 	struct tun_struct *tun = netdev_priv(dev);
1601 	return (tun->flags & TUN_NOCHECKSUM) == 0;
1602 }
1603 
1604 static int tun_set_rx_csum(struct net_device *dev, u32 data)
1605 {
1606 	struct tun_struct *tun = netdev_priv(dev);
1607 	if (data)
1608 		tun->flags &= ~TUN_NOCHECKSUM;
1609 	else
1610 		tun->flags |= TUN_NOCHECKSUM;
1611 	return 0;
1612 }
1613 
1614 static const struct ethtool_ops tun_ethtool_ops = {
1615 	.get_settings	= tun_get_settings,
1616 	.get_drvinfo	= tun_get_drvinfo,
1617 	.get_msglevel	= tun_get_msglevel,
1618 	.set_msglevel	= tun_set_msglevel,
1619 	.get_link	= ethtool_op_get_link,
1620 	.get_rx_csum	= tun_get_rx_csum,
1621 	.set_rx_csum	= tun_set_rx_csum
1622 };
1623 
1624 
1625 static int __init tun_init(void)
1626 {
1627 	int ret = 0;
1628 
1629 	pr_info("%s, %s\n", DRV_DESCRIPTION, DRV_VERSION);
1630 	pr_info("%s\n", DRV_COPYRIGHT);
1631 
1632 	ret = rtnl_link_register(&tun_link_ops);
1633 	if (ret) {
1634 		pr_err("Can't register link_ops\n");
1635 		goto err_linkops;
1636 	}
1637 
1638 	ret = misc_register(&tun_miscdev);
1639 	if (ret) {
1640 		pr_err("Can't register misc device %d\n", TUN_MINOR);
1641 		goto err_misc;
1642 	}
1643 	return  0;
1644 err_misc:
1645 	rtnl_link_unregister(&tun_link_ops);
1646 err_linkops:
1647 	return ret;
1648 }
1649 
1650 static void tun_cleanup(void)
1651 {
1652 	misc_deregister(&tun_miscdev);
1653 	rtnl_link_unregister(&tun_link_ops);
1654 }
1655 
1656 /* Get an underlying socket object from tun file.  Returns error unless file is
1657  * attached to a device.  The returned object works like a packet socket, it
1658  * can be used for sock_sendmsg/sock_recvmsg.  The caller is responsible for
1659  * holding a reference to the file for as long as the socket is in use. */
1660 struct socket *tun_get_socket(struct file *file)
1661 {
1662 	struct tun_struct *tun;
1663 	if (file->f_op != &tun_fops)
1664 		return ERR_PTR(-EINVAL);
1665 	tun = tun_get(file);
1666 	if (!tun)
1667 		return ERR_PTR(-EBADFD);
1668 	tun_put(tun);
1669 	return &tun->socket;
1670 }
1671 EXPORT_SYMBOL_GPL(tun_get_socket);
1672 
1673 module_init(tun_init);
1674 module_exit(tun_cleanup);
1675 MODULE_DESCRIPTION(DRV_DESCRIPTION);
1676 MODULE_AUTHOR(DRV_COPYRIGHT);
1677 MODULE_LICENSE("GPL");
1678 MODULE_ALIAS_MISCDEV(TUN_MINOR);
1679 MODULE_ALIAS("devname:net/tun");
1680