xref: /openbmc/linux/drivers/net/tun.c (revision d41ced01f21ddd2c3a01531bb9edf6c41064e9fc)
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/uaccess.h>
73 
74 /* Uncomment to enable debugging */
75 /* #define TUN_DEBUG 1 */
76 
77 #ifdef TUN_DEBUG
78 static int debug;
79 
80 #define tun_debug(level, tun, fmt, args...)			\
81 do {								\
82 	if (tun->debug)						\
83 		netdev_printk(level, tun->dev, fmt, ##args);	\
84 } while (0)
85 #define DBG1(level, fmt, args...)				\
86 do {								\
87 	if (debug == 2)						\
88 		printk(level fmt, ##args);			\
89 } while (0)
90 #else
91 #define tun_debug(level, tun, fmt, args...)			\
92 do {								\
93 	if (0)							\
94 		netdev_printk(level, tun->dev, fmt, ##args);	\
95 } while (0)
96 #define DBG1(level, fmt, args...)				\
97 do {								\
98 	if (0)							\
99 		printk(level fmt, ##args);			\
100 } while (0)
101 #endif
102 
103 #define FLT_EXACT_COUNT 8
104 struct tap_filter {
105 	unsigned int    count;    /* Number of addrs. Zero means disabled */
106 	u32             mask[2];  /* Mask of the hashed addrs */
107 	unsigned char	addr[FLT_EXACT_COUNT][ETH_ALEN];
108 };
109 
110 struct tun_file {
111 	atomic_t count;
112 	struct tun_struct *tun;
113 	struct net *net;
114 };
115 
116 struct tun_sock;
117 
118 struct tun_struct {
119 	struct tun_file		*tfile;
120 	unsigned int 		flags;
121 	uid_t			owner;
122 	gid_t			group;
123 
124 	struct net_device	*dev;
125 	netdev_features_t	set_features;
126 #define TUN_USER_FEATURES (NETIF_F_HW_CSUM|NETIF_F_TSO_ECN|NETIF_F_TSO| \
127 			  NETIF_F_TSO6|NETIF_F_UFO)
128 	struct fasync_struct	*fasync;
129 
130 	struct tap_filter       txflt;
131 	struct socket		socket;
132 	struct socket_wq	wq;
133 
134 	int			vnet_hdr_sz;
135 
136 #ifdef TUN_DEBUG
137 	int debug;
138 #endif
139 };
140 
141 struct tun_sock {
142 	struct sock		sk;
143 	struct tun_struct	*tun;
144 };
145 
146 static inline struct tun_sock *tun_sk(struct sock *sk)
147 {
148 	return container_of(sk, struct tun_sock, sk);
149 }
150 
151 static int tun_attach(struct tun_struct *tun, struct file *file)
152 {
153 	struct tun_file *tfile = file->private_data;
154 	int err;
155 
156 	ASSERT_RTNL();
157 
158 	netif_tx_lock_bh(tun->dev);
159 
160 	err = -EINVAL;
161 	if (tfile->tun)
162 		goto out;
163 
164 	err = -EBUSY;
165 	if (tun->tfile)
166 		goto out;
167 
168 	err = 0;
169 	tfile->tun = tun;
170 	tun->tfile = tfile;
171 	tun->socket.file = file;
172 	netif_carrier_on(tun->dev);
173 	dev_hold(tun->dev);
174 	sock_hold(tun->socket.sk);
175 	atomic_inc(&tfile->count);
176 
177 out:
178 	netif_tx_unlock_bh(tun->dev);
179 	return err;
180 }
181 
182 static void __tun_detach(struct tun_struct *tun)
183 {
184 	/* Detach from net device */
185 	netif_tx_lock_bh(tun->dev);
186 	netif_carrier_off(tun->dev);
187 	tun->tfile = NULL;
188 	tun->socket.file = NULL;
189 	netif_tx_unlock_bh(tun->dev);
190 
191 	/* Drop read queue */
192 	skb_queue_purge(&tun->socket.sk->sk_receive_queue);
193 
194 	/* Drop the extra count on the net device */
195 	dev_put(tun->dev);
196 }
197 
198 static void tun_detach(struct tun_struct *tun)
199 {
200 	rtnl_lock();
201 	__tun_detach(tun);
202 	rtnl_unlock();
203 }
204 
205 static struct tun_struct *__tun_get(struct tun_file *tfile)
206 {
207 	struct tun_struct *tun = NULL;
208 
209 	if (atomic_inc_not_zero(&tfile->count))
210 		tun = tfile->tun;
211 
212 	return tun;
213 }
214 
215 static struct tun_struct *tun_get(struct file *file)
216 {
217 	return __tun_get(file->private_data);
218 }
219 
220 static void tun_put(struct tun_struct *tun)
221 {
222 	struct tun_file *tfile = tun->tfile;
223 
224 	if (atomic_dec_and_test(&tfile->count))
225 		tun_detach(tfile->tun);
226 }
227 
228 /* TAP filtering */
229 static void addr_hash_set(u32 *mask, const u8 *addr)
230 {
231 	int n = ether_crc(ETH_ALEN, addr) >> 26;
232 	mask[n >> 5] |= (1 << (n & 31));
233 }
234 
235 static unsigned int addr_hash_test(const u32 *mask, const u8 *addr)
236 {
237 	int n = ether_crc(ETH_ALEN, addr) >> 26;
238 	return mask[n >> 5] & (1 << (n & 31));
239 }
240 
241 static int update_filter(struct tap_filter *filter, void __user *arg)
242 {
243 	struct { u8 u[ETH_ALEN]; } *addr;
244 	struct tun_filter uf;
245 	int err, alen, n, nexact;
246 
247 	if (copy_from_user(&uf, arg, sizeof(uf)))
248 		return -EFAULT;
249 
250 	if (!uf.count) {
251 		/* Disabled */
252 		filter->count = 0;
253 		return 0;
254 	}
255 
256 	alen = ETH_ALEN * uf.count;
257 	addr = kmalloc(alen, GFP_KERNEL);
258 	if (!addr)
259 		return -ENOMEM;
260 
261 	if (copy_from_user(addr, arg + sizeof(uf), alen)) {
262 		err = -EFAULT;
263 		goto done;
264 	}
265 
266 	/* The filter is updated without holding any locks. Which is
267 	 * perfectly safe. We disable it first and in the worst
268 	 * case we'll accept a few undesired packets. */
269 	filter->count = 0;
270 	wmb();
271 
272 	/* Use first set of addresses as an exact filter */
273 	for (n = 0; n < uf.count && n < FLT_EXACT_COUNT; n++)
274 		memcpy(filter->addr[n], addr[n].u, ETH_ALEN);
275 
276 	nexact = n;
277 
278 	/* Remaining multicast addresses are hashed,
279 	 * unicast will leave the filter disabled. */
280 	memset(filter->mask, 0, sizeof(filter->mask));
281 	for (; n < uf.count; n++) {
282 		if (!is_multicast_ether_addr(addr[n].u)) {
283 			err = 0; /* no filter */
284 			goto done;
285 		}
286 		addr_hash_set(filter->mask, addr[n].u);
287 	}
288 
289 	/* For ALLMULTI just set the mask to all ones.
290 	 * This overrides the mask populated above. */
291 	if ((uf.flags & TUN_FLT_ALLMULTI))
292 		memset(filter->mask, ~0, sizeof(filter->mask));
293 
294 	/* Now enable the filter */
295 	wmb();
296 	filter->count = nexact;
297 
298 	/* Return the number of exact filters */
299 	err = nexact;
300 
301 done:
302 	kfree(addr);
303 	return err;
304 }
305 
306 /* Returns: 0 - drop, !=0 - accept */
307 static int run_filter(struct tap_filter *filter, const struct sk_buff *skb)
308 {
309 	/* Cannot use eth_hdr(skb) here because skb_mac_hdr() is incorrect
310 	 * at this point. */
311 	struct ethhdr *eh = (struct ethhdr *) skb->data;
312 	int i;
313 
314 	/* Exact match */
315 	for (i = 0; i < filter->count; i++)
316 		if (!compare_ether_addr(eh->h_dest, filter->addr[i]))
317 			return 1;
318 
319 	/* Inexact match (multicast only) */
320 	if (is_multicast_ether_addr(eh->h_dest))
321 		return addr_hash_test(filter->mask, eh->h_dest);
322 
323 	return 0;
324 }
325 
326 /*
327  * Checks whether the packet is accepted or not.
328  * Returns: 0 - drop, !=0 - accept
329  */
330 static int check_filter(struct tap_filter *filter, const struct sk_buff *skb)
331 {
332 	if (!filter->count)
333 		return 1;
334 
335 	return run_filter(filter, skb);
336 }
337 
338 /* Network device part of the driver */
339 
340 static const struct ethtool_ops tun_ethtool_ops;
341 
342 /* Net device detach from fd. */
343 static void tun_net_uninit(struct net_device *dev)
344 {
345 	struct tun_struct *tun = netdev_priv(dev);
346 	struct tun_file *tfile = tun->tfile;
347 
348 	/* Inform the methods they need to stop using the dev.
349 	 */
350 	if (tfile) {
351 		wake_up_all(&tun->wq.wait);
352 		if (atomic_dec_and_test(&tfile->count))
353 			__tun_detach(tun);
354 	}
355 }
356 
357 static void tun_free_netdev(struct net_device *dev)
358 {
359 	struct tun_struct *tun = netdev_priv(dev);
360 
361 	sk_release_kernel(tun->socket.sk);
362 }
363 
364 /* Net device open. */
365 static int tun_net_open(struct net_device *dev)
366 {
367 	netif_start_queue(dev);
368 	return 0;
369 }
370 
371 /* Net device close. */
372 static int tun_net_close(struct net_device *dev)
373 {
374 	netif_stop_queue(dev);
375 	return 0;
376 }
377 
378 /* Net device start xmit */
379 static netdev_tx_t tun_net_xmit(struct sk_buff *skb, struct net_device *dev)
380 {
381 	struct tun_struct *tun = netdev_priv(dev);
382 
383 	tun_debug(KERN_INFO, tun, "tun_net_xmit %d\n", skb->len);
384 
385 	/* Drop packet if interface is not attached */
386 	if (!tun->tfile)
387 		goto drop;
388 
389 	/* Drop if the filter does not like it.
390 	 * This is a noop if the filter is disabled.
391 	 * Filter can be enabled only for the TAP devices. */
392 	if (!check_filter(&tun->txflt, skb))
393 		goto drop;
394 
395 	if (tun->socket.sk->sk_filter &&
396 	    sk_filter(tun->socket.sk, skb))
397 		goto drop;
398 
399 	if (skb_queue_len(&tun->socket.sk->sk_receive_queue) >= dev->tx_queue_len) {
400 		if (!(tun->flags & TUN_ONE_QUEUE)) {
401 			/* Normal queueing mode. */
402 			/* Packet scheduler handles dropping of further packets. */
403 			netif_stop_queue(dev);
404 
405 			/* We won't see all dropped packets individually, so overrun
406 			 * error is more appropriate. */
407 			dev->stats.tx_fifo_errors++;
408 		} else {
409 			/* Single queue mode.
410 			 * Driver handles dropping of all packets itself. */
411 			goto drop;
412 		}
413 	}
414 
415 	/* Orphan the skb - required as we might hang on to it
416 	 * for indefinite time. */
417 	skb_orphan(skb);
418 
419 	/* Enqueue packet */
420 	skb_queue_tail(&tun->socket.sk->sk_receive_queue, skb);
421 
422 	/* Notify and wake up reader process */
423 	if (tun->flags & TUN_FASYNC)
424 		kill_fasync(&tun->fasync, SIGIO, POLL_IN);
425 	wake_up_interruptible_poll(&tun->wq.wait, POLLIN |
426 				   POLLRDNORM | POLLRDBAND);
427 	return NETDEV_TX_OK;
428 
429 drop:
430 	dev->stats.tx_dropped++;
431 	kfree_skb(skb);
432 	return NETDEV_TX_OK;
433 }
434 
435 static void tun_net_mclist(struct net_device *dev)
436 {
437 	/*
438 	 * This callback is supposed to deal with mc filter in
439 	 * _rx_ path and has nothing to do with the _tx_ path.
440 	 * In rx path we always accept everything userspace gives us.
441 	 */
442 }
443 
444 #define MIN_MTU 68
445 #define MAX_MTU 65535
446 
447 static int
448 tun_net_change_mtu(struct net_device *dev, int new_mtu)
449 {
450 	if (new_mtu < MIN_MTU || new_mtu + dev->hard_header_len > MAX_MTU)
451 		return -EINVAL;
452 	dev->mtu = new_mtu;
453 	return 0;
454 }
455 
456 static netdev_features_t tun_net_fix_features(struct net_device *dev,
457 	netdev_features_t features)
458 {
459 	struct tun_struct *tun = netdev_priv(dev);
460 
461 	return (features & tun->set_features) | (features & ~TUN_USER_FEATURES);
462 }
463 #ifdef CONFIG_NET_POLL_CONTROLLER
464 static void tun_poll_controller(struct net_device *dev)
465 {
466 	/*
467 	 * Tun only receives frames when:
468 	 * 1) the char device endpoint gets data from user space
469 	 * 2) the tun socket gets a sendmsg call from user space
470 	 * Since both of those are syncronous operations, we are guaranteed
471 	 * never to have pending data when we poll for it
472 	 * so theres nothing to do here but return.
473 	 * We need this though so netpoll recognizes us as an interface that
474 	 * supports polling, which enables bridge devices in virt setups to
475 	 * still use netconsole
476 	 */
477 	return;
478 }
479 #endif
480 static const struct net_device_ops tun_netdev_ops = {
481 	.ndo_uninit		= tun_net_uninit,
482 	.ndo_open		= tun_net_open,
483 	.ndo_stop		= tun_net_close,
484 	.ndo_start_xmit		= tun_net_xmit,
485 	.ndo_change_mtu		= tun_net_change_mtu,
486 	.ndo_fix_features	= tun_net_fix_features,
487 #ifdef CONFIG_NET_POLL_CONTROLLER
488 	.ndo_poll_controller	= tun_poll_controller,
489 #endif
490 };
491 
492 static const struct net_device_ops tap_netdev_ops = {
493 	.ndo_uninit		= tun_net_uninit,
494 	.ndo_open		= tun_net_open,
495 	.ndo_stop		= tun_net_close,
496 	.ndo_start_xmit		= tun_net_xmit,
497 	.ndo_change_mtu		= tun_net_change_mtu,
498 	.ndo_fix_features	= tun_net_fix_features,
499 	.ndo_set_rx_mode	= tun_net_mclist,
500 	.ndo_set_mac_address	= eth_mac_addr,
501 	.ndo_validate_addr	= eth_validate_addr,
502 #ifdef CONFIG_NET_POLL_CONTROLLER
503 	.ndo_poll_controller	= tun_poll_controller,
504 #endif
505 };
506 
507 /* Initialize net device. */
508 static void tun_net_init(struct net_device *dev)
509 {
510 	struct tun_struct *tun = netdev_priv(dev);
511 
512 	switch (tun->flags & TUN_TYPE_MASK) {
513 	case TUN_TUN_DEV:
514 		dev->netdev_ops = &tun_netdev_ops;
515 
516 		/* Point-to-Point TUN Device */
517 		dev->hard_header_len = 0;
518 		dev->addr_len = 0;
519 		dev->mtu = 1500;
520 
521 		/* Zero header length */
522 		dev->type = ARPHRD_NONE;
523 		dev->flags = IFF_POINTOPOINT | IFF_NOARP | IFF_MULTICAST;
524 		dev->tx_queue_len = TUN_READQ_SIZE;  /* We prefer our own queue length */
525 		break;
526 
527 	case TUN_TAP_DEV:
528 		dev->netdev_ops = &tap_netdev_ops;
529 		/* Ethernet TAP Device */
530 		ether_setup(dev);
531 		dev->priv_flags &= ~IFF_TX_SKB_SHARING;
532 
533 		eth_hw_addr_random(dev);
534 
535 		dev->tx_queue_len = TUN_READQ_SIZE;  /* We prefer our own queue length */
536 		break;
537 	}
538 }
539 
540 /* Character device part */
541 
542 /* Poll */
543 static unsigned int tun_chr_poll(struct file *file, poll_table * wait)
544 {
545 	struct tun_file *tfile = file->private_data;
546 	struct tun_struct *tun = __tun_get(tfile);
547 	struct sock *sk;
548 	unsigned int mask = 0;
549 
550 	if (!tun)
551 		return POLLERR;
552 
553 	sk = tun->socket.sk;
554 
555 	tun_debug(KERN_INFO, tun, "tun_chr_poll\n");
556 
557 	poll_wait(file, &tun->wq.wait, wait);
558 
559 	if (!skb_queue_empty(&sk->sk_receive_queue))
560 		mask |= POLLIN | POLLRDNORM;
561 
562 	if (sock_writeable(sk) ||
563 	    (!test_and_set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags) &&
564 	     sock_writeable(sk)))
565 		mask |= POLLOUT | POLLWRNORM;
566 
567 	if (tun->dev->reg_state != NETREG_REGISTERED)
568 		mask = POLLERR;
569 
570 	tun_put(tun);
571 	return mask;
572 }
573 
574 /* prepad is the amount to reserve at front.  len is length after that.
575  * linear is a hint as to how much to copy (usually headers). */
576 static struct sk_buff *tun_alloc_skb(struct tun_struct *tun,
577 				     size_t prepad, size_t len,
578 				     size_t linear, int noblock)
579 {
580 	struct sock *sk = tun->socket.sk;
581 	struct sk_buff *skb;
582 	int err;
583 
584 	sock_update_classid(sk);
585 
586 	/* Under a page?  Don't bother with paged skb. */
587 	if (prepad + len < PAGE_SIZE || !linear)
588 		linear = len;
589 
590 	skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
591 				   &err);
592 	if (!skb)
593 		return ERR_PTR(err);
594 
595 	skb_reserve(skb, prepad);
596 	skb_put(skb, linear);
597 	skb->data_len = len - linear;
598 	skb->len += len - linear;
599 
600 	return skb;
601 }
602 
603 /* Get packet from user space buffer */
604 static ssize_t tun_get_user(struct tun_struct *tun,
605 			    const struct iovec *iv, size_t count,
606 			    int noblock)
607 {
608 	struct tun_pi pi = { 0, cpu_to_be16(ETH_P_IP) };
609 	struct sk_buff *skb;
610 	size_t len = count, align = NET_SKB_PAD;
611 	struct virtio_net_hdr gso = { 0 };
612 	int offset = 0;
613 
614 	if (!(tun->flags & TUN_NO_PI)) {
615 		if ((len -= sizeof(pi)) > count)
616 			return -EINVAL;
617 
618 		if (memcpy_fromiovecend((void *)&pi, iv, 0, sizeof(pi)))
619 			return -EFAULT;
620 		offset += sizeof(pi);
621 	}
622 
623 	if (tun->flags & TUN_VNET_HDR) {
624 		if ((len -= tun->vnet_hdr_sz) > count)
625 			return -EINVAL;
626 
627 		if (memcpy_fromiovecend((void *)&gso, iv, offset, sizeof(gso)))
628 			return -EFAULT;
629 
630 		if ((gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
631 		    gso.csum_start + gso.csum_offset + 2 > gso.hdr_len)
632 			gso.hdr_len = gso.csum_start + gso.csum_offset + 2;
633 
634 		if (gso.hdr_len > len)
635 			return -EINVAL;
636 		offset += tun->vnet_hdr_sz;
637 	}
638 
639 	if ((tun->flags & TUN_TYPE_MASK) == TUN_TAP_DEV) {
640 		align += NET_IP_ALIGN;
641 		if (unlikely(len < ETH_HLEN ||
642 			     (gso.hdr_len && gso.hdr_len < ETH_HLEN)))
643 			return -EINVAL;
644 	}
645 
646 	skb = tun_alloc_skb(tun, align, len, gso.hdr_len, noblock);
647 	if (IS_ERR(skb)) {
648 		if (PTR_ERR(skb) != -EAGAIN)
649 			tun->dev->stats.rx_dropped++;
650 		return PTR_ERR(skb);
651 	}
652 
653 	if (skb_copy_datagram_from_iovec(skb, 0, iv, offset, len)) {
654 		tun->dev->stats.rx_dropped++;
655 		kfree_skb(skb);
656 		return -EFAULT;
657 	}
658 
659 	if (gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
660 		if (!skb_partial_csum_set(skb, gso.csum_start,
661 					  gso.csum_offset)) {
662 			tun->dev->stats.rx_frame_errors++;
663 			kfree_skb(skb);
664 			return -EINVAL;
665 		}
666 	}
667 
668 	switch (tun->flags & TUN_TYPE_MASK) {
669 	case TUN_TUN_DEV:
670 		if (tun->flags & TUN_NO_PI) {
671 			switch (skb->data[0] & 0xf0) {
672 			case 0x40:
673 				pi.proto = htons(ETH_P_IP);
674 				break;
675 			case 0x60:
676 				pi.proto = htons(ETH_P_IPV6);
677 				break;
678 			default:
679 				tun->dev->stats.rx_dropped++;
680 				kfree_skb(skb);
681 				return -EINVAL;
682 			}
683 		}
684 
685 		skb_reset_mac_header(skb);
686 		skb->protocol = pi.proto;
687 		skb->dev = tun->dev;
688 		break;
689 	case TUN_TAP_DEV:
690 		skb->protocol = eth_type_trans(skb, tun->dev);
691 		break;
692 	}
693 
694 	if (gso.gso_type != VIRTIO_NET_HDR_GSO_NONE) {
695 		pr_debug("GSO!\n");
696 		switch (gso.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
697 		case VIRTIO_NET_HDR_GSO_TCPV4:
698 			skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
699 			break;
700 		case VIRTIO_NET_HDR_GSO_TCPV6:
701 			skb_shinfo(skb)->gso_type = SKB_GSO_TCPV6;
702 			break;
703 		case VIRTIO_NET_HDR_GSO_UDP:
704 			skb_shinfo(skb)->gso_type = SKB_GSO_UDP;
705 			break;
706 		default:
707 			tun->dev->stats.rx_frame_errors++;
708 			kfree_skb(skb);
709 			return -EINVAL;
710 		}
711 
712 		if (gso.gso_type & VIRTIO_NET_HDR_GSO_ECN)
713 			skb_shinfo(skb)->gso_type |= SKB_GSO_TCP_ECN;
714 
715 		skb_shinfo(skb)->gso_size = gso.gso_size;
716 		if (skb_shinfo(skb)->gso_size == 0) {
717 			tun->dev->stats.rx_frame_errors++;
718 			kfree_skb(skb);
719 			return -EINVAL;
720 		}
721 
722 		/* Header must be checked, and gso_segs computed. */
723 		skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
724 		skb_shinfo(skb)->gso_segs = 0;
725 	}
726 
727 	netif_rx_ni(skb);
728 
729 	tun->dev->stats.rx_packets++;
730 	tun->dev->stats.rx_bytes += len;
731 
732 	return count;
733 }
734 
735 static ssize_t tun_chr_aio_write(struct kiocb *iocb, const struct iovec *iv,
736 			      unsigned long count, loff_t pos)
737 {
738 	struct file *file = iocb->ki_filp;
739 	struct tun_struct *tun = tun_get(file);
740 	ssize_t result;
741 
742 	if (!tun)
743 		return -EBADFD;
744 
745 	tun_debug(KERN_INFO, tun, "tun_chr_write %ld\n", count);
746 
747 	result = tun_get_user(tun, iv, iov_length(iv, count),
748 			      file->f_flags & O_NONBLOCK);
749 
750 	tun_put(tun);
751 	return result;
752 }
753 
754 /* Put packet to the user space buffer */
755 static ssize_t tun_put_user(struct tun_struct *tun,
756 			    struct sk_buff *skb,
757 			    const struct iovec *iv, int len)
758 {
759 	struct tun_pi pi = { 0, skb->protocol };
760 	ssize_t total = 0;
761 
762 	if (!(tun->flags & TUN_NO_PI)) {
763 		if ((len -= sizeof(pi)) < 0)
764 			return -EINVAL;
765 
766 		if (len < skb->len) {
767 			/* Packet will be striped */
768 			pi.flags |= TUN_PKT_STRIP;
769 		}
770 
771 		if (memcpy_toiovecend(iv, (void *) &pi, 0, sizeof(pi)))
772 			return -EFAULT;
773 		total += sizeof(pi);
774 	}
775 
776 	if (tun->flags & TUN_VNET_HDR) {
777 		struct virtio_net_hdr gso = { 0 }; /* no info leak */
778 		if ((len -= tun->vnet_hdr_sz) < 0)
779 			return -EINVAL;
780 
781 		if (skb_is_gso(skb)) {
782 			struct skb_shared_info *sinfo = skb_shinfo(skb);
783 
784 			/* This is a hint as to how much should be linear. */
785 			gso.hdr_len = skb_headlen(skb);
786 			gso.gso_size = sinfo->gso_size;
787 			if (sinfo->gso_type & SKB_GSO_TCPV4)
788 				gso.gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
789 			else if (sinfo->gso_type & SKB_GSO_TCPV6)
790 				gso.gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
791 			else if (sinfo->gso_type & SKB_GSO_UDP)
792 				gso.gso_type = VIRTIO_NET_HDR_GSO_UDP;
793 			else {
794 				pr_err("unexpected GSO type: "
795 				       "0x%x, gso_size %d, hdr_len %d\n",
796 				       sinfo->gso_type, gso.gso_size,
797 				       gso.hdr_len);
798 				print_hex_dump(KERN_ERR, "tun: ",
799 					       DUMP_PREFIX_NONE,
800 					       16, 1, skb->head,
801 					       min((int)gso.hdr_len, 64), true);
802 				WARN_ON_ONCE(1);
803 				return -EINVAL;
804 			}
805 			if (sinfo->gso_type & SKB_GSO_TCP_ECN)
806 				gso.gso_type |= VIRTIO_NET_HDR_GSO_ECN;
807 		} else
808 			gso.gso_type = VIRTIO_NET_HDR_GSO_NONE;
809 
810 		if (skb->ip_summed == CHECKSUM_PARTIAL) {
811 			gso.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
812 			gso.csum_start = skb_checksum_start_offset(skb);
813 			gso.csum_offset = skb->csum_offset;
814 		} else if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
815 			gso.flags = VIRTIO_NET_HDR_F_DATA_VALID;
816 		} /* else everything is zero */
817 
818 		if (unlikely(memcpy_toiovecend(iv, (void *)&gso, total,
819 					       sizeof(gso))))
820 			return -EFAULT;
821 		total += tun->vnet_hdr_sz;
822 	}
823 
824 	len = min_t(int, skb->len, len);
825 
826 	skb_copy_datagram_const_iovec(skb, 0, iv, total, len);
827 	total += skb->len;
828 
829 	tun->dev->stats.tx_packets++;
830 	tun->dev->stats.tx_bytes += len;
831 
832 	return total;
833 }
834 
835 static ssize_t tun_do_read(struct tun_struct *tun,
836 			   struct kiocb *iocb, const struct iovec *iv,
837 			   ssize_t len, int noblock)
838 {
839 	DECLARE_WAITQUEUE(wait, current);
840 	struct sk_buff *skb;
841 	ssize_t ret = 0;
842 
843 	tun_debug(KERN_INFO, tun, "tun_chr_read\n");
844 
845 	if (unlikely(!noblock))
846 		add_wait_queue(&tun->wq.wait, &wait);
847 	while (len) {
848 		current->state = TASK_INTERRUPTIBLE;
849 
850 		/* Read frames from the queue */
851 		if (!(skb=skb_dequeue(&tun->socket.sk->sk_receive_queue))) {
852 			if (noblock) {
853 				ret = -EAGAIN;
854 				break;
855 			}
856 			if (signal_pending(current)) {
857 				ret = -ERESTARTSYS;
858 				break;
859 			}
860 			if (tun->dev->reg_state != NETREG_REGISTERED) {
861 				ret = -EIO;
862 				break;
863 			}
864 
865 			/* Nothing to read, let's sleep */
866 			schedule();
867 			continue;
868 		}
869 		netif_wake_queue(tun->dev);
870 
871 		ret = tun_put_user(tun, skb, iv, len);
872 		kfree_skb(skb);
873 		break;
874 	}
875 
876 	current->state = TASK_RUNNING;
877 	if (unlikely(!noblock))
878 		remove_wait_queue(&tun->wq.wait, &wait);
879 
880 	return ret;
881 }
882 
883 static ssize_t tun_chr_aio_read(struct kiocb *iocb, const struct iovec *iv,
884 			    unsigned long count, loff_t pos)
885 {
886 	struct file *file = iocb->ki_filp;
887 	struct tun_file *tfile = file->private_data;
888 	struct tun_struct *tun = __tun_get(tfile);
889 	ssize_t len, ret;
890 
891 	if (!tun)
892 		return -EBADFD;
893 	len = iov_length(iv, count);
894 	if (len < 0) {
895 		ret = -EINVAL;
896 		goto out;
897 	}
898 
899 	ret = tun_do_read(tun, iocb, iv, len, file->f_flags & O_NONBLOCK);
900 	ret = min_t(ssize_t, ret, len);
901 out:
902 	tun_put(tun);
903 	return ret;
904 }
905 
906 static void tun_setup(struct net_device *dev)
907 {
908 	struct tun_struct *tun = netdev_priv(dev);
909 
910 	tun->owner = -1;
911 	tun->group = -1;
912 
913 	dev->ethtool_ops = &tun_ethtool_ops;
914 	dev->destructor = tun_free_netdev;
915 }
916 
917 /* Trivial set of netlink ops to allow deleting tun or tap
918  * device with netlink.
919  */
920 static int tun_validate(struct nlattr *tb[], struct nlattr *data[])
921 {
922 	return -EINVAL;
923 }
924 
925 static struct rtnl_link_ops tun_link_ops __read_mostly = {
926 	.kind		= DRV_NAME,
927 	.priv_size	= sizeof(struct tun_struct),
928 	.setup		= tun_setup,
929 	.validate	= tun_validate,
930 };
931 
932 static void tun_sock_write_space(struct sock *sk)
933 {
934 	struct tun_struct *tun;
935 	wait_queue_head_t *wqueue;
936 
937 	if (!sock_writeable(sk))
938 		return;
939 
940 	if (!test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags))
941 		return;
942 
943 	wqueue = sk_sleep(sk);
944 	if (wqueue && waitqueue_active(wqueue))
945 		wake_up_interruptible_sync_poll(wqueue, POLLOUT |
946 						POLLWRNORM | POLLWRBAND);
947 
948 	tun = tun_sk(sk)->tun;
949 	kill_fasync(&tun->fasync, SIGIO, POLL_OUT);
950 }
951 
952 static void tun_sock_destruct(struct sock *sk)
953 {
954 	free_netdev(tun_sk(sk)->tun->dev);
955 }
956 
957 static int tun_sendmsg(struct kiocb *iocb, struct socket *sock,
958 		       struct msghdr *m, size_t total_len)
959 {
960 	struct tun_struct *tun = container_of(sock, struct tun_struct, socket);
961 	return tun_get_user(tun, m->msg_iov, total_len,
962 			    m->msg_flags & MSG_DONTWAIT);
963 }
964 
965 static int tun_recvmsg(struct kiocb *iocb, struct socket *sock,
966 		       struct msghdr *m, size_t total_len,
967 		       int flags)
968 {
969 	struct tun_struct *tun = container_of(sock, struct tun_struct, socket);
970 	int ret;
971 	if (flags & ~(MSG_DONTWAIT|MSG_TRUNC))
972 		return -EINVAL;
973 	ret = tun_do_read(tun, iocb, m->msg_iov, total_len,
974 			  flags & MSG_DONTWAIT);
975 	if (ret > total_len) {
976 		m->msg_flags |= MSG_TRUNC;
977 		ret = flags & MSG_TRUNC ? ret : total_len;
978 	}
979 	return ret;
980 }
981 
982 static int tun_release(struct socket *sock)
983 {
984 	if (sock->sk)
985 		sock_put(sock->sk);
986 	return 0;
987 }
988 
989 /* Ops structure to mimic raw sockets with tun */
990 static const struct proto_ops tun_socket_ops = {
991 	.sendmsg = tun_sendmsg,
992 	.recvmsg = tun_recvmsg,
993 	.release = tun_release,
994 };
995 
996 static struct proto tun_proto = {
997 	.name		= "tun",
998 	.owner		= THIS_MODULE,
999 	.obj_size	= sizeof(struct tun_sock),
1000 };
1001 
1002 static int tun_flags(struct tun_struct *tun)
1003 {
1004 	int flags = 0;
1005 
1006 	if (tun->flags & TUN_TUN_DEV)
1007 		flags |= IFF_TUN;
1008 	else
1009 		flags |= IFF_TAP;
1010 
1011 	if (tun->flags & TUN_NO_PI)
1012 		flags |= IFF_NO_PI;
1013 
1014 	if (tun->flags & TUN_ONE_QUEUE)
1015 		flags |= IFF_ONE_QUEUE;
1016 
1017 	if (tun->flags & TUN_VNET_HDR)
1018 		flags |= IFF_VNET_HDR;
1019 
1020 	return flags;
1021 }
1022 
1023 static ssize_t tun_show_flags(struct device *dev, struct device_attribute *attr,
1024 			      char *buf)
1025 {
1026 	struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1027 	return sprintf(buf, "0x%x\n", tun_flags(tun));
1028 }
1029 
1030 static ssize_t tun_show_owner(struct device *dev, struct device_attribute *attr,
1031 			      char *buf)
1032 {
1033 	struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1034 	return sprintf(buf, "%d\n", tun->owner);
1035 }
1036 
1037 static ssize_t tun_show_group(struct device *dev, struct device_attribute *attr,
1038 			      char *buf)
1039 {
1040 	struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1041 	return sprintf(buf, "%d\n", tun->group);
1042 }
1043 
1044 static DEVICE_ATTR(tun_flags, 0444, tun_show_flags, NULL);
1045 static DEVICE_ATTR(owner, 0444, tun_show_owner, NULL);
1046 static DEVICE_ATTR(group, 0444, tun_show_group, NULL);
1047 
1048 static int tun_set_iff(struct net *net, struct file *file, struct ifreq *ifr)
1049 {
1050 	struct sock *sk;
1051 	struct tun_struct *tun;
1052 	struct net_device *dev;
1053 	int err;
1054 
1055 	dev = __dev_get_by_name(net, ifr->ifr_name);
1056 	if (dev) {
1057 		const struct cred *cred = current_cred();
1058 
1059 		if (ifr->ifr_flags & IFF_TUN_EXCL)
1060 			return -EBUSY;
1061 		if ((ifr->ifr_flags & IFF_TUN) && dev->netdev_ops == &tun_netdev_ops)
1062 			tun = netdev_priv(dev);
1063 		else if ((ifr->ifr_flags & IFF_TAP) && dev->netdev_ops == &tap_netdev_ops)
1064 			tun = netdev_priv(dev);
1065 		else
1066 			return -EINVAL;
1067 
1068 		if (((tun->owner != -1 && cred->euid != tun->owner) ||
1069 		     (tun->group != -1 && !in_egroup_p(tun->group))) &&
1070 		    !capable(CAP_NET_ADMIN))
1071 			return -EPERM;
1072 		err = security_tun_dev_attach(tun->socket.sk);
1073 		if (err < 0)
1074 			return err;
1075 
1076 		err = tun_attach(tun, file);
1077 		if (err < 0)
1078 			return err;
1079 	}
1080 	else {
1081 		char *name;
1082 		unsigned long flags = 0;
1083 
1084 		if (!capable(CAP_NET_ADMIN))
1085 			return -EPERM;
1086 		err = security_tun_dev_create();
1087 		if (err < 0)
1088 			return err;
1089 
1090 		/* Set dev type */
1091 		if (ifr->ifr_flags & IFF_TUN) {
1092 			/* TUN device */
1093 			flags |= TUN_TUN_DEV;
1094 			name = "tun%d";
1095 		} else if (ifr->ifr_flags & IFF_TAP) {
1096 			/* TAP device */
1097 			flags |= TUN_TAP_DEV;
1098 			name = "tap%d";
1099 		} else
1100 			return -EINVAL;
1101 
1102 		if (*ifr->ifr_name)
1103 			name = ifr->ifr_name;
1104 
1105 		dev = alloc_netdev(sizeof(struct tun_struct), name,
1106 				   tun_setup);
1107 		if (!dev)
1108 			return -ENOMEM;
1109 
1110 		dev_net_set(dev, net);
1111 		dev->rtnl_link_ops = &tun_link_ops;
1112 
1113 		tun = netdev_priv(dev);
1114 		tun->dev = dev;
1115 		tun->flags = flags;
1116 		tun->txflt.count = 0;
1117 		tun->vnet_hdr_sz = sizeof(struct virtio_net_hdr);
1118 
1119 		err = -ENOMEM;
1120 		sk = sk_alloc(&init_net, AF_UNSPEC, GFP_KERNEL, &tun_proto);
1121 		if (!sk)
1122 			goto err_free_dev;
1123 
1124 		sk_change_net(sk, net);
1125 		tun->socket.wq = &tun->wq;
1126 		init_waitqueue_head(&tun->wq.wait);
1127 		tun->socket.ops = &tun_socket_ops;
1128 		sock_init_data(&tun->socket, sk);
1129 		sk->sk_write_space = tun_sock_write_space;
1130 		sk->sk_sndbuf = INT_MAX;
1131 
1132 		tun_sk(sk)->tun = tun;
1133 
1134 		security_tun_dev_post_create(sk);
1135 
1136 		tun_net_init(dev);
1137 
1138 		dev->hw_features = NETIF_F_SG | NETIF_F_FRAGLIST |
1139 			TUN_USER_FEATURES;
1140 		dev->features = dev->hw_features;
1141 
1142 		err = register_netdevice(tun->dev);
1143 		if (err < 0)
1144 			goto err_free_sk;
1145 
1146 		if (device_create_file(&tun->dev->dev, &dev_attr_tun_flags) ||
1147 		    device_create_file(&tun->dev->dev, &dev_attr_owner) ||
1148 		    device_create_file(&tun->dev->dev, &dev_attr_group))
1149 			pr_err("Failed to create tun sysfs files\n");
1150 
1151 		sk->sk_destruct = tun_sock_destruct;
1152 
1153 		err = tun_attach(tun, file);
1154 		if (err < 0)
1155 			goto failed;
1156 	}
1157 
1158 	tun_debug(KERN_INFO, tun, "tun_set_iff\n");
1159 
1160 	if (ifr->ifr_flags & IFF_NO_PI)
1161 		tun->flags |= TUN_NO_PI;
1162 	else
1163 		tun->flags &= ~TUN_NO_PI;
1164 
1165 	if (ifr->ifr_flags & IFF_ONE_QUEUE)
1166 		tun->flags |= TUN_ONE_QUEUE;
1167 	else
1168 		tun->flags &= ~TUN_ONE_QUEUE;
1169 
1170 	if (ifr->ifr_flags & IFF_VNET_HDR)
1171 		tun->flags |= TUN_VNET_HDR;
1172 	else
1173 		tun->flags &= ~TUN_VNET_HDR;
1174 
1175 	/* Make sure persistent devices do not get stuck in
1176 	 * xoff state.
1177 	 */
1178 	if (netif_running(tun->dev))
1179 		netif_wake_queue(tun->dev);
1180 
1181 	strcpy(ifr->ifr_name, tun->dev->name);
1182 	return 0;
1183 
1184  err_free_sk:
1185 	tun_free_netdev(dev);
1186  err_free_dev:
1187 	free_netdev(dev);
1188  failed:
1189 	return err;
1190 }
1191 
1192 static int tun_get_iff(struct net *net, struct tun_struct *tun,
1193 		       struct ifreq *ifr)
1194 {
1195 	tun_debug(KERN_INFO, tun, "tun_get_iff\n");
1196 
1197 	strcpy(ifr->ifr_name, tun->dev->name);
1198 
1199 	ifr->ifr_flags = tun_flags(tun);
1200 
1201 	return 0;
1202 }
1203 
1204 /* This is like a cut-down ethtool ops, except done via tun fd so no
1205  * privs required. */
1206 static int set_offload(struct tun_struct *tun, unsigned long arg)
1207 {
1208 	netdev_features_t features = 0;
1209 
1210 	if (arg & TUN_F_CSUM) {
1211 		features |= NETIF_F_HW_CSUM;
1212 		arg &= ~TUN_F_CSUM;
1213 
1214 		if (arg & (TUN_F_TSO4|TUN_F_TSO6)) {
1215 			if (arg & TUN_F_TSO_ECN) {
1216 				features |= NETIF_F_TSO_ECN;
1217 				arg &= ~TUN_F_TSO_ECN;
1218 			}
1219 			if (arg & TUN_F_TSO4)
1220 				features |= NETIF_F_TSO;
1221 			if (arg & TUN_F_TSO6)
1222 				features |= NETIF_F_TSO6;
1223 			arg &= ~(TUN_F_TSO4|TUN_F_TSO6);
1224 		}
1225 
1226 		if (arg & TUN_F_UFO) {
1227 			features |= NETIF_F_UFO;
1228 			arg &= ~TUN_F_UFO;
1229 		}
1230 	}
1231 
1232 	/* This gives the user a way to test for new features in future by
1233 	 * trying to set them. */
1234 	if (arg)
1235 		return -EINVAL;
1236 
1237 	tun->set_features = features;
1238 	netdev_update_features(tun->dev);
1239 
1240 	return 0;
1241 }
1242 
1243 static long __tun_chr_ioctl(struct file *file, unsigned int cmd,
1244 			    unsigned long arg, int ifreq_len)
1245 {
1246 	struct tun_file *tfile = file->private_data;
1247 	struct tun_struct *tun;
1248 	void __user* argp = (void __user*)arg;
1249 	struct sock_fprog fprog;
1250 	struct ifreq ifr;
1251 	int sndbuf;
1252 	int vnet_hdr_sz;
1253 	int ret;
1254 
1255 	if (cmd == TUNSETIFF || _IOC_TYPE(cmd) == 0x89)
1256 		if (copy_from_user(&ifr, argp, ifreq_len))
1257 			return -EFAULT;
1258 
1259 	if (cmd == TUNGETFEATURES) {
1260 		/* Currently this just means: "what IFF flags are valid?".
1261 		 * This is needed because we never checked for invalid flags on
1262 		 * TUNSETIFF. */
1263 		return put_user(IFF_TUN | IFF_TAP | IFF_NO_PI | IFF_ONE_QUEUE |
1264 				IFF_VNET_HDR,
1265 				(unsigned int __user*)argp);
1266 	}
1267 
1268 	rtnl_lock();
1269 
1270 	tun = __tun_get(tfile);
1271 	if (cmd == TUNSETIFF && !tun) {
1272 		ifr.ifr_name[IFNAMSIZ-1] = '\0';
1273 
1274 		ret = tun_set_iff(tfile->net, file, &ifr);
1275 
1276 		if (ret)
1277 			goto unlock;
1278 
1279 		if (copy_to_user(argp, &ifr, ifreq_len))
1280 			ret = -EFAULT;
1281 		goto unlock;
1282 	}
1283 
1284 	ret = -EBADFD;
1285 	if (!tun)
1286 		goto unlock;
1287 
1288 	tun_debug(KERN_INFO, tun, "tun_chr_ioctl cmd %d\n", cmd);
1289 
1290 	ret = 0;
1291 	switch (cmd) {
1292 	case TUNGETIFF:
1293 		ret = tun_get_iff(current->nsproxy->net_ns, tun, &ifr);
1294 		if (ret)
1295 			break;
1296 
1297 		if (copy_to_user(argp, &ifr, ifreq_len))
1298 			ret = -EFAULT;
1299 		break;
1300 
1301 	case TUNSETNOCSUM:
1302 		/* Disable/Enable checksum */
1303 
1304 		/* [unimplemented] */
1305 		tun_debug(KERN_INFO, tun, "ignored: set checksum %s\n",
1306 			  arg ? "disabled" : "enabled");
1307 		break;
1308 
1309 	case TUNSETPERSIST:
1310 		/* Disable/Enable persist mode */
1311 		if (arg)
1312 			tun->flags |= TUN_PERSIST;
1313 		else
1314 			tun->flags &= ~TUN_PERSIST;
1315 
1316 		tun_debug(KERN_INFO, tun, "persist %s\n",
1317 			  arg ? "enabled" : "disabled");
1318 		break;
1319 
1320 	case TUNSETOWNER:
1321 		/* Set owner of the device */
1322 		tun->owner = (uid_t) arg;
1323 
1324 		tun_debug(KERN_INFO, tun, "owner set to %d\n", tun->owner);
1325 		break;
1326 
1327 	case TUNSETGROUP:
1328 		/* Set group of the device */
1329 		tun->group= (gid_t) arg;
1330 
1331 		tun_debug(KERN_INFO, tun, "group set to %d\n", tun->group);
1332 		break;
1333 
1334 	case TUNSETLINK:
1335 		/* Only allow setting the type when the interface is down */
1336 		if (tun->dev->flags & IFF_UP) {
1337 			tun_debug(KERN_INFO, tun,
1338 				  "Linktype set failed because interface is up\n");
1339 			ret = -EBUSY;
1340 		} else {
1341 			tun->dev->type = (int) arg;
1342 			tun_debug(KERN_INFO, tun, "linktype set to %d\n",
1343 				  tun->dev->type);
1344 			ret = 0;
1345 		}
1346 		break;
1347 
1348 #ifdef TUN_DEBUG
1349 	case TUNSETDEBUG:
1350 		tun->debug = arg;
1351 		break;
1352 #endif
1353 	case TUNSETOFFLOAD:
1354 		ret = set_offload(tun, arg);
1355 		break;
1356 
1357 	case TUNSETTXFILTER:
1358 		/* Can be set only for TAPs */
1359 		ret = -EINVAL;
1360 		if ((tun->flags & TUN_TYPE_MASK) != TUN_TAP_DEV)
1361 			break;
1362 		ret = update_filter(&tun->txflt, (void __user *)arg);
1363 		break;
1364 
1365 	case SIOCGIFHWADDR:
1366 		/* Get hw address */
1367 		memcpy(ifr.ifr_hwaddr.sa_data, tun->dev->dev_addr, ETH_ALEN);
1368 		ifr.ifr_hwaddr.sa_family = tun->dev->type;
1369 		if (copy_to_user(argp, &ifr, ifreq_len))
1370 			ret = -EFAULT;
1371 		break;
1372 
1373 	case SIOCSIFHWADDR:
1374 		/* Set hw address */
1375 		tun_debug(KERN_DEBUG, tun, "set hw address: %pM\n",
1376 			  ifr.ifr_hwaddr.sa_data);
1377 
1378 		ret = dev_set_mac_address(tun->dev, &ifr.ifr_hwaddr);
1379 		break;
1380 
1381 	case TUNGETSNDBUF:
1382 		sndbuf = tun->socket.sk->sk_sndbuf;
1383 		if (copy_to_user(argp, &sndbuf, sizeof(sndbuf)))
1384 			ret = -EFAULT;
1385 		break;
1386 
1387 	case TUNSETSNDBUF:
1388 		if (copy_from_user(&sndbuf, argp, sizeof(sndbuf))) {
1389 			ret = -EFAULT;
1390 			break;
1391 		}
1392 
1393 		tun->socket.sk->sk_sndbuf = sndbuf;
1394 		break;
1395 
1396 	case TUNGETVNETHDRSZ:
1397 		vnet_hdr_sz = tun->vnet_hdr_sz;
1398 		if (copy_to_user(argp, &vnet_hdr_sz, sizeof(vnet_hdr_sz)))
1399 			ret = -EFAULT;
1400 		break;
1401 
1402 	case TUNSETVNETHDRSZ:
1403 		if (copy_from_user(&vnet_hdr_sz, argp, sizeof(vnet_hdr_sz))) {
1404 			ret = -EFAULT;
1405 			break;
1406 		}
1407 		if (vnet_hdr_sz < (int)sizeof(struct virtio_net_hdr)) {
1408 			ret = -EINVAL;
1409 			break;
1410 		}
1411 
1412 		tun->vnet_hdr_sz = vnet_hdr_sz;
1413 		break;
1414 
1415 	case TUNATTACHFILTER:
1416 		/* Can be set only for TAPs */
1417 		ret = -EINVAL;
1418 		if ((tun->flags & TUN_TYPE_MASK) != TUN_TAP_DEV)
1419 			break;
1420 		ret = -EFAULT;
1421 		if (copy_from_user(&fprog, argp, sizeof(fprog)))
1422 			break;
1423 
1424 		ret = sk_attach_filter(&fprog, tun->socket.sk);
1425 		break;
1426 
1427 	case TUNDETACHFILTER:
1428 		/* Can be set only for TAPs */
1429 		ret = -EINVAL;
1430 		if ((tun->flags & TUN_TYPE_MASK) != TUN_TAP_DEV)
1431 			break;
1432 		ret = sk_detach_filter(tun->socket.sk);
1433 		break;
1434 
1435 	default:
1436 		ret = -EINVAL;
1437 		break;
1438 	}
1439 
1440 unlock:
1441 	rtnl_unlock();
1442 	if (tun)
1443 		tun_put(tun);
1444 	return ret;
1445 }
1446 
1447 static long tun_chr_ioctl(struct file *file,
1448 			  unsigned int cmd, unsigned long arg)
1449 {
1450 	return __tun_chr_ioctl(file, cmd, arg, sizeof (struct ifreq));
1451 }
1452 
1453 #ifdef CONFIG_COMPAT
1454 static long tun_chr_compat_ioctl(struct file *file,
1455 			 unsigned int cmd, unsigned long arg)
1456 {
1457 	switch (cmd) {
1458 	case TUNSETIFF:
1459 	case TUNGETIFF:
1460 	case TUNSETTXFILTER:
1461 	case TUNGETSNDBUF:
1462 	case TUNSETSNDBUF:
1463 	case SIOCGIFHWADDR:
1464 	case SIOCSIFHWADDR:
1465 		arg = (unsigned long)compat_ptr(arg);
1466 		break;
1467 	default:
1468 		arg = (compat_ulong_t)arg;
1469 		break;
1470 	}
1471 
1472 	/*
1473 	 * compat_ifreq is shorter than ifreq, so we must not access beyond
1474 	 * the end of that structure. All fields that are used in this
1475 	 * driver are compatible though, we don't need to convert the
1476 	 * contents.
1477 	 */
1478 	return __tun_chr_ioctl(file, cmd, arg, sizeof(struct compat_ifreq));
1479 }
1480 #endif /* CONFIG_COMPAT */
1481 
1482 static int tun_chr_fasync(int fd, struct file *file, int on)
1483 {
1484 	struct tun_struct *tun = tun_get(file);
1485 	int ret;
1486 
1487 	if (!tun)
1488 		return -EBADFD;
1489 
1490 	tun_debug(KERN_INFO, tun, "tun_chr_fasync %d\n", on);
1491 
1492 	if ((ret = fasync_helper(fd, file, on, &tun->fasync)) < 0)
1493 		goto out;
1494 
1495 	if (on) {
1496 		ret = __f_setown(file, task_pid(current), PIDTYPE_PID, 0);
1497 		if (ret)
1498 			goto out;
1499 		tun->flags |= TUN_FASYNC;
1500 	} else
1501 		tun->flags &= ~TUN_FASYNC;
1502 	ret = 0;
1503 out:
1504 	tun_put(tun);
1505 	return ret;
1506 }
1507 
1508 static int tun_chr_open(struct inode *inode, struct file * file)
1509 {
1510 	struct tun_file *tfile;
1511 
1512 	DBG1(KERN_INFO, "tunX: tun_chr_open\n");
1513 
1514 	tfile = kmalloc(sizeof(*tfile), GFP_KERNEL);
1515 	if (!tfile)
1516 		return -ENOMEM;
1517 	atomic_set(&tfile->count, 0);
1518 	tfile->tun = NULL;
1519 	tfile->net = get_net(current->nsproxy->net_ns);
1520 	file->private_data = tfile;
1521 	return 0;
1522 }
1523 
1524 static int tun_chr_close(struct inode *inode, struct file *file)
1525 {
1526 	struct tun_file *tfile = file->private_data;
1527 	struct tun_struct *tun;
1528 
1529 	tun = __tun_get(tfile);
1530 	if (tun) {
1531 		struct net_device *dev = tun->dev;
1532 
1533 		tun_debug(KERN_INFO, tun, "tun_chr_close\n");
1534 
1535 		__tun_detach(tun);
1536 
1537 		/* If desirable, unregister the netdevice. */
1538 		if (!(tun->flags & TUN_PERSIST)) {
1539 			rtnl_lock();
1540 			if (dev->reg_state == NETREG_REGISTERED)
1541 				unregister_netdevice(dev);
1542 			rtnl_unlock();
1543 		}
1544 	}
1545 
1546 	tun = tfile->tun;
1547 	if (tun)
1548 		sock_put(tun->socket.sk);
1549 
1550 	put_net(tfile->net);
1551 	kfree(tfile);
1552 
1553 	return 0;
1554 }
1555 
1556 static const struct file_operations tun_fops = {
1557 	.owner	= THIS_MODULE,
1558 	.llseek = no_llseek,
1559 	.read  = do_sync_read,
1560 	.aio_read  = tun_chr_aio_read,
1561 	.write = do_sync_write,
1562 	.aio_write = tun_chr_aio_write,
1563 	.poll	= tun_chr_poll,
1564 	.unlocked_ioctl	= tun_chr_ioctl,
1565 #ifdef CONFIG_COMPAT
1566 	.compat_ioctl = tun_chr_compat_ioctl,
1567 #endif
1568 	.open	= tun_chr_open,
1569 	.release = tun_chr_close,
1570 	.fasync = tun_chr_fasync
1571 };
1572 
1573 static struct miscdevice tun_miscdev = {
1574 	.minor = TUN_MINOR,
1575 	.name = "tun",
1576 	.nodename = "net/tun",
1577 	.fops = &tun_fops,
1578 };
1579 
1580 /* ethtool interface */
1581 
1582 static int tun_get_settings(struct net_device *dev, struct ethtool_cmd *cmd)
1583 {
1584 	cmd->supported		= 0;
1585 	cmd->advertising	= 0;
1586 	ethtool_cmd_speed_set(cmd, SPEED_10);
1587 	cmd->duplex		= DUPLEX_FULL;
1588 	cmd->port		= PORT_TP;
1589 	cmd->phy_address	= 0;
1590 	cmd->transceiver	= XCVR_INTERNAL;
1591 	cmd->autoneg		= AUTONEG_DISABLE;
1592 	cmd->maxtxpkt		= 0;
1593 	cmd->maxrxpkt		= 0;
1594 	return 0;
1595 }
1596 
1597 static void tun_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
1598 {
1599 	struct tun_struct *tun = netdev_priv(dev);
1600 
1601 	strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
1602 	strlcpy(info->version, DRV_VERSION, sizeof(info->version));
1603 
1604 	switch (tun->flags & TUN_TYPE_MASK) {
1605 	case TUN_TUN_DEV:
1606 		strlcpy(info->bus_info, "tun", sizeof(info->bus_info));
1607 		break;
1608 	case TUN_TAP_DEV:
1609 		strlcpy(info->bus_info, "tap", sizeof(info->bus_info));
1610 		break;
1611 	}
1612 }
1613 
1614 static u32 tun_get_msglevel(struct net_device *dev)
1615 {
1616 #ifdef TUN_DEBUG
1617 	struct tun_struct *tun = netdev_priv(dev);
1618 	return tun->debug;
1619 #else
1620 	return -EOPNOTSUPP;
1621 #endif
1622 }
1623 
1624 static void tun_set_msglevel(struct net_device *dev, u32 value)
1625 {
1626 #ifdef TUN_DEBUG
1627 	struct tun_struct *tun = netdev_priv(dev);
1628 	tun->debug = value;
1629 #endif
1630 }
1631 
1632 static const struct ethtool_ops tun_ethtool_ops = {
1633 	.get_settings	= tun_get_settings,
1634 	.get_drvinfo	= tun_get_drvinfo,
1635 	.get_msglevel	= tun_get_msglevel,
1636 	.set_msglevel	= tun_set_msglevel,
1637 	.get_link	= ethtool_op_get_link,
1638 };
1639 
1640 
1641 static int __init tun_init(void)
1642 {
1643 	int ret = 0;
1644 
1645 	pr_info("%s, %s\n", DRV_DESCRIPTION, DRV_VERSION);
1646 	pr_info("%s\n", DRV_COPYRIGHT);
1647 
1648 	ret = rtnl_link_register(&tun_link_ops);
1649 	if (ret) {
1650 		pr_err("Can't register link_ops\n");
1651 		goto err_linkops;
1652 	}
1653 
1654 	ret = misc_register(&tun_miscdev);
1655 	if (ret) {
1656 		pr_err("Can't register misc device %d\n", TUN_MINOR);
1657 		goto err_misc;
1658 	}
1659 	return  0;
1660 err_misc:
1661 	rtnl_link_unregister(&tun_link_ops);
1662 err_linkops:
1663 	return ret;
1664 }
1665 
1666 static void tun_cleanup(void)
1667 {
1668 	misc_deregister(&tun_miscdev);
1669 	rtnl_link_unregister(&tun_link_ops);
1670 }
1671 
1672 /* Get an underlying socket object from tun file.  Returns error unless file is
1673  * attached to a device.  The returned object works like a packet socket, it
1674  * can be used for sock_sendmsg/sock_recvmsg.  The caller is responsible for
1675  * holding a reference to the file for as long as the socket is in use. */
1676 struct socket *tun_get_socket(struct file *file)
1677 {
1678 	struct tun_struct *tun;
1679 	if (file->f_op != &tun_fops)
1680 		return ERR_PTR(-EINVAL);
1681 	tun = tun_get(file);
1682 	if (!tun)
1683 		return ERR_PTR(-EBADFD);
1684 	tun_put(tun);
1685 	return &tun->socket;
1686 }
1687 EXPORT_SYMBOL_GPL(tun_get_socket);
1688 
1689 module_init(tun_init);
1690 module_exit(tun_cleanup);
1691 MODULE_DESCRIPTION(DRV_DESCRIPTION);
1692 MODULE_AUTHOR(DRV_COPYRIGHT);
1693 MODULE_LICENSE("GPL");
1694 MODULE_ALIAS_MISCDEV(TUN_MINOR);
1695 MODULE_ALIAS("devname:net/tun");
1696