xref: /linux/drivers/net/tun.c (revision 25aee3debe0464f6c680173041fa3de30ec9ff54)
1 /*
2  *  TUN - Universal TUN/TAP device driver.
3  *  Copyright (C) 1999-2002 Maxim Krasnyansky <maxk@qualcomm.com>
4  *
5  *  This program is free software; you can redistribute it and/or modify
6  *  it under the terms of the GNU General Public License as published by
7  *  the Free Software Foundation; either version 2 of the License, or
8  *  (at your option) any later version.
9  *
10  *  This program is distributed in the hope that it will be useful,
11  *  but WITHOUT ANY WARRANTY; without even the implied warranty of
12  *  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
13  *  GNU General Public License for more details.
14  *
15  *  $Id: tun.c,v 1.15 2002/03/01 02:44:24 maxk Exp $
16  */
17 
18 /*
19  *  Changes:
20  *
21  *  Mike Kershaw <dragorn@kismetwireless.net> 2005/08/14
22  *    Add TUNSETLINK ioctl to set the link encapsulation
23  *
24  *  Mark Smith <markzzzsmith@yahoo.com.au>
25  *    Use eth_random_addr() for tap MAC address.
26  *
27  *  Harald Roelle <harald.roelle@ifi.lmu.de>  2004/04/20
28  *    Fixes in packet dropping, queue length setting and queue wakeup.
29  *    Increased default tx queue length.
30  *    Added ethtool API.
31  *    Minor cleanups
32  *
33  *  Daniel Podlejski <underley@underley.eu.org>
34  *    Modifications for 2.3.99-pre5 kernel.
35  */
36 
37 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
38 
39 #define DRV_NAME	"tun"
40 #define DRV_VERSION	"1.6"
41 #define DRV_DESCRIPTION	"Universal TUN/TAP device driver"
42 #define DRV_COPYRIGHT	"(C) 1999-2004 Max Krasnyansky <maxk@qualcomm.com>"
43 
44 #include <linux/module.h>
45 #include <linux/errno.h>
46 #include <linux/kernel.h>
47 #include <linux/major.h>
48 #include <linux/slab.h>
49 #include <linux/poll.h>
50 #include <linux/fcntl.h>
51 #include <linux/init.h>
52 #include <linux/skbuff.h>
53 #include <linux/netdevice.h>
54 #include <linux/etherdevice.h>
55 #include <linux/miscdevice.h>
56 #include <linux/ethtool.h>
57 #include <linux/rtnetlink.h>
58 #include <linux/compat.h>
59 #include <linux/if.h>
60 #include <linux/if_arp.h>
61 #include <linux/if_ether.h>
62 #include <linux/if_tun.h>
63 #include <linux/crc32.h>
64 #include <linux/nsproxy.h>
65 #include <linux/virtio_net.h>
66 #include <linux/rcupdate.h>
67 #include <net/net_namespace.h>
68 #include <net/netns/generic.h>
69 #include <net/rtnetlink.h>
70 #include <net/sock.h>
71 
72 #include <asm/uaccess.h>
73 
74 /* Uncomment to enable debugging */
75 /* #define TUN_DEBUG 1 */
76 
77 #ifdef TUN_DEBUG
78 static int debug;
79 
80 #define tun_debug(level, tun, fmt, args...)			\
81 do {								\
82 	if (tun->debug)						\
83 		netdev_printk(level, tun->dev, fmt, ##args);	\
84 } while (0)
85 #define DBG1(level, fmt, args...)				\
86 do {								\
87 	if (debug == 2)						\
88 		printk(level fmt, ##args);			\
89 } while (0)
90 #else
91 #define tun_debug(level, tun, fmt, args...)			\
92 do {								\
93 	if (0)							\
94 		netdev_printk(level, tun->dev, fmt, ##args);	\
95 } while (0)
96 #define DBG1(level, fmt, args...)				\
97 do {								\
98 	if (0)							\
99 		printk(level fmt, ##args);			\
100 } while (0)
101 #endif
102 
103 #define GOODCOPY_LEN 128
104 
105 #define FLT_EXACT_COUNT 8
106 struct tap_filter {
107 	unsigned int    count;    /* Number of addrs. Zero means disabled */
108 	u32             mask[2];  /* Mask of the hashed addrs */
109 	unsigned char	addr[FLT_EXACT_COUNT][ETH_ALEN];
110 };
111 
112 struct tun_file {
113 	atomic_t count;
114 	struct tun_struct *tun;
115 	struct net *net;
116 };
117 
118 struct tun_sock;
119 
120 struct tun_struct {
121 	struct tun_file		*tfile;
122 	unsigned int 		flags;
123 	uid_t			owner;
124 	gid_t			group;
125 
126 	struct net_device	*dev;
127 	netdev_features_t	set_features;
128 #define TUN_USER_FEATURES (NETIF_F_HW_CSUM|NETIF_F_TSO_ECN|NETIF_F_TSO| \
129 			  NETIF_F_TSO6|NETIF_F_UFO)
130 	struct fasync_struct	*fasync;
131 
132 	struct tap_filter       txflt;
133 	struct socket		socket;
134 	struct socket_wq	wq;
135 
136 	int			vnet_hdr_sz;
137 
138 #ifdef TUN_DEBUG
139 	int debug;
140 #endif
141 };
142 
143 struct tun_sock {
144 	struct sock		sk;
145 	struct tun_struct	*tun;
146 };
147 
148 static inline struct tun_sock *tun_sk(struct sock *sk)
149 {
150 	return container_of(sk, struct tun_sock, sk);
151 }
152 
153 static int tun_attach(struct tun_struct *tun, struct file *file)
154 {
155 	struct tun_file *tfile = file->private_data;
156 	int err;
157 
158 	ASSERT_RTNL();
159 
160 	netif_tx_lock_bh(tun->dev);
161 
162 	err = -EINVAL;
163 	if (tfile->tun)
164 		goto out;
165 
166 	err = -EBUSY;
167 	if (tun->tfile)
168 		goto out;
169 
170 	err = 0;
171 	tfile->tun = tun;
172 	tun->tfile = tfile;
173 	tun->socket.file = file;
174 	netif_carrier_on(tun->dev);
175 	dev_hold(tun->dev);
176 	sock_hold(tun->socket.sk);
177 	atomic_inc(&tfile->count);
178 
179 out:
180 	netif_tx_unlock_bh(tun->dev);
181 	return err;
182 }
183 
184 static void __tun_detach(struct tun_struct *tun)
185 {
186 	/* Detach from net device */
187 	netif_tx_lock_bh(tun->dev);
188 	netif_carrier_off(tun->dev);
189 	tun->tfile = NULL;
190 	tun->socket.file = NULL;
191 	netif_tx_unlock_bh(tun->dev);
192 
193 	/* Drop read queue */
194 	skb_queue_purge(&tun->socket.sk->sk_receive_queue);
195 
196 	/* Drop the extra count on the net device */
197 	dev_put(tun->dev);
198 }
199 
200 static void tun_detach(struct tun_struct *tun)
201 {
202 	rtnl_lock();
203 	__tun_detach(tun);
204 	rtnl_unlock();
205 }
206 
207 static struct tun_struct *__tun_get(struct tun_file *tfile)
208 {
209 	struct tun_struct *tun = NULL;
210 
211 	if (atomic_inc_not_zero(&tfile->count))
212 		tun = tfile->tun;
213 
214 	return tun;
215 }
216 
217 static struct tun_struct *tun_get(struct file *file)
218 {
219 	return __tun_get(file->private_data);
220 }
221 
222 static void tun_put(struct tun_struct *tun)
223 {
224 	struct tun_file *tfile = tun->tfile;
225 
226 	if (atomic_dec_and_test(&tfile->count))
227 		tun_detach(tfile->tun);
228 }
229 
230 /* TAP filtering */
231 static void addr_hash_set(u32 *mask, const u8 *addr)
232 {
233 	int n = ether_crc(ETH_ALEN, addr) >> 26;
234 	mask[n >> 5] |= (1 << (n & 31));
235 }
236 
237 static unsigned int addr_hash_test(const u32 *mask, const u8 *addr)
238 {
239 	int n = ether_crc(ETH_ALEN, addr) >> 26;
240 	return mask[n >> 5] & (1 << (n & 31));
241 }
242 
243 static int update_filter(struct tap_filter *filter, void __user *arg)
244 {
245 	struct { u8 u[ETH_ALEN]; } *addr;
246 	struct tun_filter uf;
247 	int err, alen, n, nexact;
248 
249 	if (copy_from_user(&uf, arg, sizeof(uf)))
250 		return -EFAULT;
251 
252 	if (!uf.count) {
253 		/* Disabled */
254 		filter->count = 0;
255 		return 0;
256 	}
257 
258 	alen = ETH_ALEN * uf.count;
259 	addr = kmalloc(alen, GFP_KERNEL);
260 	if (!addr)
261 		return -ENOMEM;
262 
263 	if (copy_from_user(addr, arg + sizeof(uf), alen)) {
264 		err = -EFAULT;
265 		goto done;
266 	}
267 
268 	/* The filter is updated without holding any locks. Which is
269 	 * perfectly safe. We disable it first and in the worst
270 	 * case we'll accept a few undesired packets. */
271 	filter->count = 0;
272 	wmb();
273 
274 	/* Use first set of addresses as an exact filter */
275 	for (n = 0; n < uf.count && n < FLT_EXACT_COUNT; n++)
276 		memcpy(filter->addr[n], addr[n].u, ETH_ALEN);
277 
278 	nexact = n;
279 
280 	/* Remaining multicast addresses are hashed,
281 	 * unicast will leave the filter disabled. */
282 	memset(filter->mask, 0, sizeof(filter->mask));
283 	for (; n < uf.count; n++) {
284 		if (!is_multicast_ether_addr(addr[n].u)) {
285 			err = 0; /* no filter */
286 			goto done;
287 		}
288 		addr_hash_set(filter->mask, addr[n].u);
289 	}
290 
291 	/* For ALLMULTI just set the mask to all ones.
292 	 * This overrides the mask populated above. */
293 	if ((uf.flags & TUN_FLT_ALLMULTI))
294 		memset(filter->mask, ~0, sizeof(filter->mask));
295 
296 	/* Now enable the filter */
297 	wmb();
298 	filter->count = nexact;
299 
300 	/* Return the number of exact filters */
301 	err = nexact;
302 
303 done:
304 	kfree(addr);
305 	return err;
306 }
307 
308 /* Returns: 0 - drop, !=0 - accept */
309 static int run_filter(struct tap_filter *filter, const struct sk_buff *skb)
310 {
311 	/* Cannot use eth_hdr(skb) here because skb_mac_hdr() is incorrect
312 	 * at this point. */
313 	struct ethhdr *eh = (struct ethhdr *) skb->data;
314 	int i;
315 
316 	/* Exact match */
317 	for (i = 0; i < filter->count; i++)
318 		if (ether_addr_equal(eh->h_dest, filter->addr[i]))
319 			return 1;
320 
321 	/* Inexact match (multicast only) */
322 	if (is_multicast_ether_addr(eh->h_dest))
323 		return addr_hash_test(filter->mask, eh->h_dest);
324 
325 	return 0;
326 }
327 
328 /*
329  * Checks whether the packet is accepted or not.
330  * Returns: 0 - drop, !=0 - accept
331  */
332 static int check_filter(struct tap_filter *filter, const struct sk_buff *skb)
333 {
334 	if (!filter->count)
335 		return 1;
336 
337 	return run_filter(filter, skb);
338 }
339 
340 /* Network device part of the driver */
341 
342 static const struct ethtool_ops tun_ethtool_ops;
343 
344 /* Net device detach from fd. */
345 static void tun_net_uninit(struct net_device *dev)
346 {
347 	struct tun_struct *tun = netdev_priv(dev);
348 	struct tun_file *tfile = tun->tfile;
349 
350 	/* Inform the methods they need to stop using the dev.
351 	 */
352 	if (tfile) {
353 		wake_up_all(&tun->wq.wait);
354 		if (atomic_dec_and_test(&tfile->count))
355 			__tun_detach(tun);
356 	}
357 }
358 
359 static void tun_free_netdev(struct net_device *dev)
360 {
361 	struct tun_struct *tun = netdev_priv(dev);
362 
363 	BUG_ON(!test_bit(SOCK_EXTERNALLY_ALLOCATED, &tun->socket.flags));
364 
365 	sk_release_kernel(tun->socket.sk);
366 }
367 
368 /* Net device open. */
369 static int tun_net_open(struct net_device *dev)
370 {
371 	netif_start_queue(dev);
372 	return 0;
373 }
374 
375 /* Net device close. */
376 static int tun_net_close(struct net_device *dev)
377 {
378 	netif_stop_queue(dev);
379 	return 0;
380 }
381 
382 /* Net device start xmit */
383 static netdev_tx_t tun_net_xmit(struct sk_buff *skb, struct net_device *dev)
384 {
385 	struct tun_struct *tun = netdev_priv(dev);
386 
387 	tun_debug(KERN_INFO, tun, "tun_net_xmit %d\n", skb->len);
388 
389 	/* Drop packet if interface is not attached */
390 	if (!tun->tfile)
391 		goto drop;
392 
393 	/* Drop if the filter does not like it.
394 	 * This is a noop if the filter is disabled.
395 	 * Filter can be enabled only for the TAP devices. */
396 	if (!check_filter(&tun->txflt, skb))
397 		goto drop;
398 
399 	if (tun->socket.sk->sk_filter &&
400 	    sk_filter(tun->socket.sk, skb))
401 		goto drop;
402 
403 	if (skb_queue_len(&tun->socket.sk->sk_receive_queue) >= dev->tx_queue_len) {
404 		if (!(tun->flags & TUN_ONE_QUEUE)) {
405 			/* Normal queueing mode. */
406 			/* Packet scheduler handles dropping of further packets. */
407 			netif_stop_queue(dev);
408 
409 			/* We won't see all dropped packets individually, so overrun
410 			 * error is more appropriate. */
411 			dev->stats.tx_fifo_errors++;
412 		} else {
413 			/* Single queue mode.
414 			 * Driver handles dropping of all packets itself. */
415 			goto drop;
416 		}
417 	}
418 
419 	/* Orphan the skb - required as we might hang on to it
420 	 * for indefinite time. */
421 	if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC)))
422 		goto drop;
423 	skb_orphan(skb);
424 
425 	/* Enqueue packet */
426 	skb_queue_tail(&tun->socket.sk->sk_receive_queue, skb);
427 
428 	/* Notify and wake up reader process */
429 	if (tun->flags & TUN_FASYNC)
430 		kill_fasync(&tun->fasync, SIGIO, POLL_IN);
431 	wake_up_interruptible_poll(&tun->wq.wait, POLLIN |
432 				   POLLRDNORM | POLLRDBAND);
433 	return NETDEV_TX_OK;
434 
435 drop:
436 	dev->stats.tx_dropped++;
437 	kfree_skb(skb);
438 	return NETDEV_TX_OK;
439 }
440 
441 static void tun_net_mclist(struct net_device *dev)
442 {
443 	/*
444 	 * This callback is supposed to deal with mc filter in
445 	 * _rx_ path and has nothing to do with the _tx_ path.
446 	 * In rx path we always accept everything userspace gives us.
447 	 */
448 }
449 
450 #define MIN_MTU 68
451 #define MAX_MTU 65535
452 
453 static int
454 tun_net_change_mtu(struct net_device *dev, int new_mtu)
455 {
456 	if (new_mtu < MIN_MTU || new_mtu + dev->hard_header_len > MAX_MTU)
457 		return -EINVAL;
458 	dev->mtu = new_mtu;
459 	return 0;
460 }
461 
462 static netdev_features_t tun_net_fix_features(struct net_device *dev,
463 	netdev_features_t features)
464 {
465 	struct tun_struct *tun = netdev_priv(dev);
466 
467 	return (features & tun->set_features) | (features & ~TUN_USER_FEATURES);
468 }
469 #ifdef CONFIG_NET_POLL_CONTROLLER
470 static void tun_poll_controller(struct net_device *dev)
471 {
472 	/*
473 	 * Tun only receives frames when:
474 	 * 1) the char device endpoint gets data from user space
475 	 * 2) the tun socket gets a sendmsg call from user space
476 	 * Since both of those are syncronous operations, we are guaranteed
477 	 * never to have pending data when we poll for it
478 	 * so theres nothing to do here but return.
479 	 * We need this though so netpoll recognizes us as an interface that
480 	 * supports polling, which enables bridge devices in virt setups to
481 	 * still use netconsole
482 	 */
483 	return;
484 }
485 #endif
486 static const struct net_device_ops tun_netdev_ops = {
487 	.ndo_uninit		= tun_net_uninit,
488 	.ndo_open		= tun_net_open,
489 	.ndo_stop		= tun_net_close,
490 	.ndo_start_xmit		= tun_net_xmit,
491 	.ndo_change_mtu		= tun_net_change_mtu,
492 	.ndo_fix_features	= tun_net_fix_features,
493 #ifdef CONFIG_NET_POLL_CONTROLLER
494 	.ndo_poll_controller	= tun_poll_controller,
495 #endif
496 };
497 
498 static const struct net_device_ops tap_netdev_ops = {
499 	.ndo_uninit		= tun_net_uninit,
500 	.ndo_open		= tun_net_open,
501 	.ndo_stop		= tun_net_close,
502 	.ndo_start_xmit		= tun_net_xmit,
503 	.ndo_change_mtu		= tun_net_change_mtu,
504 	.ndo_fix_features	= tun_net_fix_features,
505 	.ndo_set_rx_mode	= tun_net_mclist,
506 	.ndo_set_mac_address	= eth_mac_addr,
507 	.ndo_validate_addr	= eth_validate_addr,
508 #ifdef CONFIG_NET_POLL_CONTROLLER
509 	.ndo_poll_controller	= tun_poll_controller,
510 #endif
511 };
512 
513 /* Initialize net device. */
514 static void tun_net_init(struct net_device *dev)
515 {
516 	struct tun_struct *tun = netdev_priv(dev);
517 
518 	switch (tun->flags & TUN_TYPE_MASK) {
519 	case TUN_TUN_DEV:
520 		dev->netdev_ops = &tun_netdev_ops;
521 
522 		/* Point-to-Point TUN Device */
523 		dev->hard_header_len = 0;
524 		dev->addr_len = 0;
525 		dev->mtu = 1500;
526 
527 		/* Zero header length */
528 		dev->type = ARPHRD_NONE;
529 		dev->flags = IFF_POINTOPOINT | IFF_NOARP | IFF_MULTICAST;
530 		dev->tx_queue_len = TUN_READQ_SIZE;  /* We prefer our own queue length */
531 		break;
532 
533 	case TUN_TAP_DEV:
534 		dev->netdev_ops = &tap_netdev_ops;
535 		/* Ethernet TAP Device */
536 		ether_setup(dev);
537 		dev->priv_flags &= ~IFF_TX_SKB_SHARING;
538 
539 		eth_hw_addr_random(dev);
540 
541 		dev->tx_queue_len = TUN_READQ_SIZE;  /* We prefer our own queue length */
542 		break;
543 	}
544 }
545 
546 /* Character device part */
547 
548 /* Poll */
549 static unsigned int tun_chr_poll(struct file *file, poll_table * wait)
550 {
551 	struct tun_file *tfile = file->private_data;
552 	struct tun_struct *tun = __tun_get(tfile);
553 	struct sock *sk;
554 	unsigned int mask = 0;
555 
556 	if (!tun)
557 		return POLLERR;
558 
559 	sk = tun->socket.sk;
560 
561 	tun_debug(KERN_INFO, tun, "tun_chr_poll\n");
562 
563 	poll_wait(file, &tun->wq.wait, wait);
564 
565 	if (!skb_queue_empty(&sk->sk_receive_queue))
566 		mask |= POLLIN | POLLRDNORM;
567 
568 	if (sock_writeable(sk) ||
569 	    (!test_and_set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags) &&
570 	     sock_writeable(sk)))
571 		mask |= POLLOUT | POLLWRNORM;
572 
573 	if (tun->dev->reg_state != NETREG_REGISTERED)
574 		mask = POLLERR;
575 
576 	tun_put(tun);
577 	return mask;
578 }
579 
580 /* prepad is the amount to reserve at front.  len is length after that.
581  * linear is a hint as to how much to copy (usually headers). */
582 static struct sk_buff *tun_alloc_skb(struct tun_struct *tun,
583 				     size_t prepad, size_t len,
584 				     size_t linear, int noblock)
585 {
586 	struct sock *sk = tun->socket.sk;
587 	struct sk_buff *skb;
588 	int err;
589 
590 	sock_update_classid(sk);
591 
592 	/* Under a page?  Don't bother with paged skb. */
593 	if (prepad + len < PAGE_SIZE || !linear)
594 		linear = len;
595 
596 	skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
597 				   &err);
598 	if (!skb)
599 		return ERR_PTR(err);
600 
601 	skb_reserve(skb, prepad);
602 	skb_put(skb, linear);
603 	skb->data_len = len - linear;
604 	skb->len += len - linear;
605 
606 	return skb;
607 }
608 
609 /* set skb frags from iovec, this can move to core network code for reuse */
610 static int zerocopy_sg_from_iovec(struct sk_buff *skb, const struct iovec *from,
611 				  int offset, size_t count)
612 {
613 	int len = iov_length(from, count) - offset;
614 	int copy = skb_headlen(skb);
615 	int size, offset1 = 0;
616 	int i = 0;
617 
618 	/* Skip over from offset */
619 	while (count && (offset >= from->iov_len)) {
620 		offset -= from->iov_len;
621 		++from;
622 		--count;
623 	}
624 
625 	/* copy up to skb headlen */
626 	while (count && (copy > 0)) {
627 		size = min_t(unsigned int, copy, from->iov_len - offset);
628 		if (copy_from_user(skb->data + offset1, from->iov_base + offset,
629 				   size))
630 			return -EFAULT;
631 		if (copy > size) {
632 			++from;
633 			--count;
634 			offset = 0;
635 		} else
636 			offset += size;
637 		copy -= size;
638 		offset1 += size;
639 	}
640 
641 	if (len == offset1)
642 		return 0;
643 
644 	while (count--) {
645 		struct page *page[MAX_SKB_FRAGS];
646 		int num_pages;
647 		unsigned long base;
648 		unsigned long truesize;
649 
650 		len = from->iov_len - offset;
651 		if (!len) {
652 			offset = 0;
653 			++from;
654 			continue;
655 		}
656 		base = (unsigned long)from->iov_base + offset;
657 		size = ((base & ~PAGE_MASK) + len + ~PAGE_MASK) >> PAGE_SHIFT;
658 		if (i + size > MAX_SKB_FRAGS)
659 			return -EMSGSIZE;
660 		num_pages = get_user_pages_fast(base, size, 0, &page[i]);
661 		if (num_pages != size) {
662 			for (i = 0; i < num_pages; i++)
663 				put_page(page[i]);
664 			return -EFAULT;
665 		}
666 		truesize = size * PAGE_SIZE;
667 		skb->data_len += len;
668 		skb->len += len;
669 		skb->truesize += truesize;
670 		atomic_add(truesize, &skb->sk->sk_wmem_alloc);
671 		while (len) {
672 			int off = base & ~PAGE_MASK;
673 			int size = min_t(int, len, PAGE_SIZE - off);
674 			__skb_fill_page_desc(skb, i, page[i], off, size);
675 			skb_shinfo(skb)->nr_frags++;
676 			/* increase sk_wmem_alloc */
677 			base += size;
678 			len -= size;
679 			i++;
680 		}
681 		offset = 0;
682 		++from;
683 	}
684 	return 0;
685 }
686 
687 /* Get packet from user space buffer */
688 static ssize_t tun_get_user(struct tun_struct *tun, void *msg_control,
689 			    const struct iovec *iv, size_t total_len,
690 			    size_t count, int noblock)
691 {
692 	struct tun_pi pi = { 0, cpu_to_be16(ETH_P_IP) };
693 	struct sk_buff *skb;
694 	size_t len = total_len, align = NET_SKB_PAD;
695 	struct virtio_net_hdr gso = { 0 };
696 	int offset = 0;
697 	int copylen;
698 	bool zerocopy = false;
699 	int err;
700 
701 	if (!(tun->flags & TUN_NO_PI)) {
702 		if ((len -= sizeof(pi)) > total_len)
703 			return -EINVAL;
704 
705 		if (memcpy_fromiovecend((void *)&pi, iv, 0, sizeof(pi)))
706 			return -EFAULT;
707 		offset += sizeof(pi);
708 	}
709 
710 	if (tun->flags & TUN_VNET_HDR) {
711 		if ((len -= tun->vnet_hdr_sz) > total_len)
712 			return -EINVAL;
713 
714 		if (memcpy_fromiovecend((void *)&gso, iv, offset, sizeof(gso)))
715 			return -EFAULT;
716 
717 		if ((gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
718 		    gso.csum_start + gso.csum_offset + 2 > gso.hdr_len)
719 			gso.hdr_len = gso.csum_start + gso.csum_offset + 2;
720 
721 		if (gso.hdr_len > len)
722 			return -EINVAL;
723 		offset += tun->vnet_hdr_sz;
724 	}
725 
726 	if ((tun->flags & TUN_TYPE_MASK) == TUN_TAP_DEV) {
727 		align += NET_IP_ALIGN;
728 		if (unlikely(len < ETH_HLEN ||
729 			     (gso.hdr_len && gso.hdr_len < ETH_HLEN)))
730 			return -EINVAL;
731 	}
732 
733 	if (msg_control)
734 		zerocopy = true;
735 
736 	if (zerocopy) {
737 		/* Userspace may produce vectors with count greater than
738 		 * MAX_SKB_FRAGS, so we need to linearize parts of the skb
739 		 * to let the rest of data to be fit in the frags.
740 		 */
741 		if (count > MAX_SKB_FRAGS) {
742 			copylen = iov_length(iv, count - MAX_SKB_FRAGS);
743 			if (copylen < offset)
744 				copylen = 0;
745 			else
746 				copylen -= offset;
747 		} else
748 				copylen = 0;
749 		/* There are 256 bytes to be copied in skb, so there is enough
750 		 * room for skb expand head in case it is used.
751 		 * The rest of the buffer is mapped from userspace.
752 		 */
753 		if (copylen < gso.hdr_len)
754 			copylen = gso.hdr_len;
755 		if (!copylen)
756 			copylen = GOODCOPY_LEN;
757 	} else
758 		copylen = len;
759 
760 	skb = tun_alloc_skb(tun, align, copylen, gso.hdr_len, noblock);
761 	if (IS_ERR(skb)) {
762 		if (PTR_ERR(skb) != -EAGAIN)
763 			tun->dev->stats.rx_dropped++;
764 		return PTR_ERR(skb);
765 	}
766 
767 	if (zerocopy)
768 		err = zerocopy_sg_from_iovec(skb, iv, offset, count);
769 	else
770 		err = skb_copy_datagram_from_iovec(skb, 0, iv, offset, len);
771 
772 	if (err) {
773 		tun->dev->stats.rx_dropped++;
774 		kfree_skb(skb);
775 		return -EFAULT;
776 	}
777 
778 	if (gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
779 		if (!skb_partial_csum_set(skb, gso.csum_start,
780 					  gso.csum_offset)) {
781 			tun->dev->stats.rx_frame_errors++;
782 			kfree_skb(skb);
783 			return -EINVAL;
784 		}
785 	}
786 
787 	switch (tun->flags & TUN_TYPE_MASK) {
788 	case TUN_TUN_DEV:
789 		if (tun->flags & TUN_NO_PI) {
790 			switch (skb->data[0] & 0xf0) {
791 			case 0x40:
792 				pi.proto = htons(ETH_P_IP);
793 				break;
794 			case 0x60:
795 				pi.proto = htons(ETH_P_IPV6);
796 				break;
797 			default:
798 				tun->dev->stats.rx_dropped++;
799 				kfree_skb(skb);
800 				return -EINVAL;
801 			}
802 		}
803 
804 		skb_reset_mac_header(skb);
805 		skb->protocol = pi.proto;
806 		skb->dev = tun->dev;
807 		break;
808 	case TUN_TAP_DEV:
809 		skb->protocol = eth_type_trans(skb, tun->dev);
810 		break;
811 	}
812 
813 	if (gso.gso_type != VIRTIO_NET_HDR_GSO_NONE) {
814 		pr_debug("GSO!\n");
815 		switch (gso.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
816 		case VIRTIO_NET_HDR_GSO_TCPV4:
817 			skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
818 			break;
819 		case VIRTIO_NET_HDR_GSO_TCPV6:
820 			skb_shinfo(skb)->gso_type = SKB_GSO_TCPV6;
821 			break;
822 		case VIRTIO_NET_HDR_GSO_UDP:
823 			skb_shinfo(skb)->gso_type = SKB_GSO_UDP;
824 			break;
825 		default:
826 			tun->dev->stats.rx_frame_errors++;
827 			kfree_skb(skb);
828 			return -EINVAL;
829 		}
830 
831 		if (gso.gso_type & VIRTIO_NET_HDR_GSO_ECN)
832 			skb_shinfo(skb)->gso_type |= SKB_GSO_TCP_ECN;
833 
834 		skb_shinfo(skb)->gso_size = gso.gso_size;
835 		if (skb_shinfo(skb)->gso_size == 0) {
836 			tun->dev->stats.rx_frame_errors++;
837 			kfree_skb(skb);
838 			return -EINVAL;
839 		}
840 
841 		/* Header must be checked, and gso_segs computed. */
842 		skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
843 		skb_shinfo(skb)->gso_segs = 0;
844 	}
845 
846 	/* copy skb_ubuf_info for callback when skb has no error */
847 	if (zerocopy) {
848 		skb_shinfo(skb)->destructor_arg = msg_control;
849 		skb_shinfo(skb)->tx_flags |= SKBTX_DEV_ZEROCOPY;
850 	}
851 
852 	netif_rx_ni(skb);
853 
854 	tun->dev->stats.rx_packets++;
855 	tun->dev->stats.rx_bytes += len;
856 
857 	return total_len;
858 }
859 
860 static ssize_t tun_chr_aio_write(struct kiocb *iocb, const struct iovec *iv,
861 			      unsigned long count, loff_t pos)
862 {
863 	struct file *file = iocb->ki_filp;
864 	struct tun_struct *tun = tun_get(file);
865 	ssize_t result;
866 
867 	if (!tun)
868 		return -EBADFD;
869 
870 	tun_debug(KERN_INFO, tun, "tun_chr_write %ld\n", count);
871 
872 	result = tun_get_user(tun, NULL, iv, iov_length(iv, count), count,
873 			      file->f_flags & O_NONBLOCK);
874 
875 	tun_put(tun);
876 	return result;
877 }
878 
879 /* Put packet to the user space buffer */
880 static ssize_t tun_put_user(struct tun_struct *tun,
881 			    struct sk_buff *skb,
882 			    const struct iovec *iv, int len)
883 {
884 	struct tun_pi pi = { 0, skb->protocol };
885 	ssize_t total = 0;
886 
887 	if (!(tun->flags & TUN_NO_PI)) {
888 		if ((len -= sizeof(pi)) < 0)
889 			return -EINVAL;
890 
891 		if (len < skb->len) {
892 			/* Packet will be striped */
893 			pi.flags |= TUN_PKT_STRIP;
894 		}
895 
896 		if (memcpy_toiovecend(iv, (void *) &pi, 0, sizeof(pi)))
897 			return -EFAULT;
898 		total += sizeof(pi);
899 	}
900 
901 	if (tun->flags & TUN_VNET_HDR) {
902 		struct virtio_net_hdr gso = { 0 }; /* no info leak */
903 		if ((len -= tun->vnet_hdr_sz) < 0)
904 			return -EINVAL;
905 
906 		if (skb_is_gso(skb)) {
907 			struct skb_shared_info *sinfo = skb_shinfo(skb);
908 
909 			/* This is a hint as to how much should be linear. */
910 			gso.hdr_len = skb_headlen(skb);
911 			gso.gso_size = sinfo->gso_size;
912 			if (sinfo->gso_type & SKB_GSO_TCPV4)
913 				gso.gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
914 			else if (sinfo->gso_type & SKB_GSO_TCPV6)
915 				gso.gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
916 			else if (sinfo->gso_type & SKB_GSO_UDP)
917 				gso.gso_type = VIRTIO_NET_HDR_GSO_UDP;
918 			else {
919 				pr_err("unexpected GSO type: "
920 				       "0x%x, gso_size %d, hdr_len %d\n",
921 				       sinfo->gso_type, gso.gso_size,
922 				       gso.hdr_len);
923 				print_hex_dump(KERN_ERR, "tun: ",
924 					       DUMP_PREFIX_NONE,
925 					       16, 1, skb->head,
926 					       min((int)gso.hdr_len, 64), true);
927 				WARN_ON_ONCE(1);
928 				return -EINVAL;
929 			}
930 			if (sinfo->gso_type & SKB_GSO_TCP_ECN)
931 				gso.gso_type |= VIRTIO_NET_HDR_GSO_ECN;
932 		} else
933 			gso.gso_type = VIRTIO_NET_HDR_GSO_NONE;
934 
935 		if (skb->ip_summed == CHECKSUM_PARTIAL) {
936 			gso.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
937 			gso.csum_start = skb_checksum_start_offset(skb);
938 			gso.csum_offset = skb->csum_offset;
939 		} else if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
940 			gso.flags = VIRTIO_NET_HDR_F_DATA_VALID;
941 		} /* else everything is zero */
942 
943 		if (unlikely(memcpy_toiovecend(iv, (void *)&gso, total,
944 					       sizeof(gso))))
945 			return -EFAULT;
946 		total += tun->vnet_hdr_sz;
947 	}
948 
949 	len = min_t(int, skb->len, len);
950 
951 	skb_copy_datagram_const_iovec(skb, 0, iv, total, len);
952 	total += skb->len;
953 
954 	tun->dev->stats.tx_packets++;
955 	tun->dev->stats.tx_bytes += len;
956 
957 	return total;
958 }
959 
960 static ssize_t tun_do_read(struct tun_struct *tun,
961 			   struct kiocb *iocb, const struct iovec *iv,
962 			   ssize_t len, int noblock)
963 {
964 	DECLARE_WAITQUEUE(wait, current);
965 	struct sk_buff *skb;
966 	ssize_t ret = 0;
967 
968 	tun_debug(KERN_INFO, tun, "tun_chr_read\n");
969 
970 	if (unlikely(!noblock))
971 		add_wait_queue(&tun->wq.wait, &wait);
972 	while (len) {
973 		current->state = TASK_INTERRUPTIBLE;
974 
975 		/* Read frames from the queue */
976 		if (!(skb=skb_dequeue(&tun->socket.sk->sk_receive_queue))) {
977 			if (noblock) {
978 				ret = -EAGAIN;
979 				break;
980 			}
981 			if (signal_pending(current)) {
982 				ret = -ERESTARTSYS;
983 				break;
984 			}
985 			if (tun->dev->reg_state != NETREG_REGISTERED) {
986 				ret = -EIO;
987 				break;
988 			}
989 
990 			/* Nothing to read, let's sleep */
991 			schedule();
992 			continue;
993 		}
994 		netif_wake_queue(tun->dev);
995 
996 		ret = tun_put_user(tun, skb, iv, len);
997 		kfree_skb(skb);
998 		break;
999 	}
1000 
1001 	current->state = TASK_RUNNING;
1002 	if (unlikely(!noblock))
1003 		remove_wait_queue(&tun->wq.wait, &wait);
1004 
1005 	return ret;
1006 }
1007 
1008 static ssize_t tun_chr_aio_read(struct kiocb *iocb, const struct iovec *iv,
1009 			    unsigned long count, loff_t pos)
1010 {
1011 	struct file *file = iocb->ki_filp;
1012 	struct tun_file *tfile = file->private_data;
1013 	struct tun_struct *tun = __tun_get(tfile);
1014 	ssize_t len, ret;
1015 
1016 	if (!tun)
1017 		return -EBADFD;
1018 	len = iov_length(iv, count);
1019 	if (len < 0) {
1020 		ret = -EINVAL;
1021 		goto out;
1022 	}
1023 
1024 	ret = tun_do_read(tun, iocb, iv, len, file->f_flags & O_NONBLOCK);
1025 	ret = min_t(ssize_t, ret, len);
1026 out:
1027 	tun_put(tun);
1028 	return ret;
1029 }
1030 
1031 static void tun_setup(struct net_device *dev)
1032 {
1033 	struct tun_struct *tun = netdev_priv(dev);
1034 
1035 	tun->owner = -1;
1036 	tun->group = -1;
1037 
1038 	dev->ethtool_ops = &tun_ethtool_ops;
1039 	dev->destructor = tun_free_netdev;
1040 }
1041 
1042 /* Trivial set of netlink ops to allow deleting tun or tap
1043  * device with netlink.
1044  */
1045 static int tun_validate(struct nlattr *tb[], struct nlattr *data[])
1046 {
1047 	return -EINVAL;
1048 }
1049 
1050 static struct rtnl_link_ops tun_link_ops __read_mostly = {
1051 	.kind		= DRV_NAME,
1052 	.priv_size	= sizeof(struct tun_struct),
1053 	.setup		= tun_setup,
1054 	.validate	= tun_validate,
1055 };
1056 
1057 static void tun_sock_write_space(struct sock *sk)
1058 {
1059 	struct tun_struct *tun;
1060 	wait_queue_head_t *wqueue;
1061 
1062 	if (!sock_writeable(sk))
1063 		return;
1064 
1065 	if (!test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags))
1066 		return;
1067 
1068 	wqueue = sk_sleep(sk);
1069 	if (wqueue && waitqueue_active(wqueue))
1070 		wake_up_interruptible_sync_poll(wqueue, POLLOUT |
1071 						POLLWRNORM | POLLWRBAND);
1072 
1073 	tun = tun_sk(sk)->tun;
1074 	kill_fasync(&tun->fasync, SIGIO, POLL_OUT);
1075 }
1076 
1077 static void tun_sock_destruct(struct sock *sk)
1078 {
1079 	free_netdev(tun_sk(sk)->tun->dev);
1080 }
1081 
1082 static int tun_sendmsg(struct kiocb *iocb, struct socket *sock,
1083 		       struct msghdr *m, size_t total_len)
1084 {
1085 	struct tun_struct *tun = container_of(sock, struct tun_struct, socket);
1086 	return tun_get_user(tun, m->msg_control, m->msg_iov, total_len,
1087 			    m->msg_iovlen, m->msg_flags & MSG_DONTWAIT);
1088 }
1089 
1090 static int tun_recvmsg(struct kiocb *iocb, struct socket *sock,
1091 		       struct msghdr *m, size_t total_len,
1092 		       int flags)
1093 {
1094 	struct tun_struct *tun = container_of(sock, struct tun_struct, socket);
1095 	int ret;
1096 	if (flags & ~(MSG_DONTWAIT|MSG_TRUNC))
1097 		return -EINVAL;
1098 	ret = tun_do_read(tun, iocb, m->msg_iov, total_len,
1099 			  flags & MSG_DONTWAIT);
1100 	if (ret > total_len) {
1101 		m->msg_flags |= MSG_TRUNC;
1102 		ret = flags & MSG_TRUNC ? ret : total_len;
1103 	}
1104 	return ret;
1105 }
1106 
1107 static int tun_release(struct socket *sock)
1108 {
1109 	if (sock->sk)
1110 		sock_put(sock->sk);
1111 	return 0;
1112 }
1113 
1114 /* Ops structure to mimic raw sockets with tun */
1115 static const struct proto_ops tun_socket_ops = {
1116 	.sendmsg = tun_sendmsg,
1117 	.recvmsg = tun_recvmsg,
1118 	.release = tun_release,
1119 };
1120 
1121 static struct proto tun_proto = {
1122 	.name		= "tun",
1123 	.owner		= THIS_MODULE,
1124 	.obj_size	= sizeof(struct tun_sock),
1125 };
1126 
1127 static int tun_flags(struct tun_struct *tun)
1128 {
1129 	int flags = 0;
1130 
1131 	if (tun->flags & TUN_TUN_DEV)
1132 		flags |= IFF_TUN;
1133 	else
1134 		flags |= IFF_TAP;
1135 
1136 	if (tun->flags & TUN_NO_PI)
1137 		flags |= IFF_NO_PI;
1138 
1139 	if (tun->flags & TUN_ONE_QUEUE)
1140 		flags |= IFF_ONE_QUEUE;
1141 
1142 	if (tun->flags & TUN_VNET_HDR)
1143 		flags |= IFF_VNET_HDR;
1144 
1145 	return flags;
1146 }
1147 
1148 static ssize_t tun_show_flags(struct device *dev, struct device_attribute *attr,
1149 			      char *buf)
1150 {
1151 	struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1152 	return sprintf(buf, "0x%x\n", tun_flags(tun));
1153 }
1154 
1155 static ssize_t tun_show_owner(struct device *dev, struct device_attribute *attr,
1156 			      char *buf)
1157 {
1158 	struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1159 	return sprintf(buf, "%d\n", tun->owner);
1160 }
1161 
1162 static ssize_t tun_show_group(struct device *dev, struct device_attribute *attr,
1163 			      char *buf)
1164 {
1165 	struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1166 	return sprintf(buf, "%d\n", tun->group);
1167 }
1168 
1169 static DEVICE_ATTR(tun_flags, 0444, tun_show_flags, NULL);
1170 static DEVICE_ATTR(owner, 0444, tun_show_owner, NULL);
1171 static DEVICE_ATTR(group, 0444, tun_show_group, NULL);
1172 
1173 static int tun_set_iff(struct net *net, struct file *file, struct ifreq *ifr)
1174 {
1175 	struct sock *sk;
1176 	struct tun_struct *tun;
1177 	struct net_device *dev;
1178 	int err;
1179 
1180 	dev = __dev_get_by_name(net, ifr->ifr_name);
1181 	if (dev) {
1182 		const struct cred *cred = current_cred();
1183 
1184 		if (ifr->ifr_flags & IFF_TUN_EXCL)
1185 			return -EBUSY;
1186 		if ((ifr->ifr_flags & IFF_TUN) && dev->netdev_ops == &tun_netdev_ops)
1187 			tun = netdev_priv(dev);
1188 		else if ((ifr->ifr_flags & IFF_TAP) && dev->netdev_ops == &tap_netdev_ops)
1189 			tun = netdev_priv(dev);
1190 		else
1191 			return -EINVAL;
1192 
1193 		if (((tun->owner != -1 && cred->euid != tun->owner) ||
1194 		     (tun->group != -1 && !in_egroup_p(tun->group))) &&
1195 		    !capable(CAP_NET_ADMIN))
1196 			return -EPERM;
1197 		err = security_tun_dev_attach(tun->socket.sk);
1198 		if (err < 0)
1199 			return err;
1200 
1201 		err = tun_attach(tun, file);
1202 		if (err < 0)
1203 			return err;
1204 	}
1205 	else {
1206 		char *name;
1207 		unsigned long flags = 0;
1208 
1209 		if (!capable(CAP_NET_ADMIN))
1210 			return -EPERM;
1211 		err = security_tun_dev_create();
1212 		if (err < 0)
1213 			return err;
1214 
1215 		/* Set dev type */
1216 		if (ifr->ifr_flags & IFF_TUN) {
1217 			/* TUN device */
1218 			flags |= TUN_TUN_DEV;
1219 			name = "tun%d";
1220 		} else if (ifr->ifr_flags & IFF_TAP) {
1221 			/* TAP device */
1222 			flags |= TUN_TAP_DEV;
1223 			name = "tap%d";
1224 		} else
1225 			return -EINVAL;
1226 
1227 		if (*ifr->ifr_name)
1228 			name = ifr->ifr_name;
1229 
1230 		dev = alloc_netdev(sizeof(struct tun_struct), name,
1231 				   tun_setup);
1232 		if (!dev)
1233 			return -ENOMEM;
1234 
1235 		dev_net_set(dev, net);
1236 		dev->rtnl_link_ops = &tun_link_ops;
1237 
1238 		tun = netdev_priv(dev);
1239 		tun->dev = dev;
1240 		tun->flags = flags;
1241 		tun->txflt.count = 0;
1242 		tun->vnet_hdr_sz = sizeof(struct virtio_net_hdr);
1243 		set_bit(SOCK_EXTERNALLY_ALLOCATED, &tun->socket.flags);
1244 
1245 		err = -ENOMEM;
1246 		sk = sk_alloc(&init_net, AF_UNSPEC, GFP_KERNEL, &tun_proto);
1247 		if (!sk)
1248 			goto err_free_dev;
1249 
1250 		sk_change_net(sk, net);
1251 		tun->socket.wq = &tun->wq;
1252 		init_waitqueue_head(&tun->wq.wait);
1253 		tun->socket.ops = &tun_socket_ops;
1254 		sock_init_data(&tun->socket, sk);
1255 		sk->sk_write_space = tun_sock_write_space;
1256 		sk->sk_sndbuf = INT_MAX;
1257 		sock_set_flag(sk, SOCK_ZEROCOPY);
1258 
1259 		tun_sk(sk)->tun = tun;
1260 
1261 		security_tun_dev_post_create(sk);
1262 
1263 		tun_net_init(dev);
1264 
1265 		dev->hw_features = NETIF_F_SG | NETIF_F_FRAGLIST |
1266 			TUN_USER_FEATURES;
1267 		dev->features = dev->hw_features;
1268 
1269 		err = register_netdevice(tun->dev);
1270 		if (err < 0)
1271 			goto err_free_sk;
1272 
1273 		if (device_create_file(&tun->dev->dev, &dev_attr_tun_flags) ||
1274 		    device_create_file(&tun->dev->dev, &dev_attr_owner) ||
1275 		    device_create_file(&tun->dev->dev, &dev_attr_group))
1276 			pr_err("Failed to create tun sysfs files\n");
1277 
1278 		sk->sk_destruct = tun_sock_destruct;
1279 
1280 		err = tun_attach(tun, file);
1281 		if (err < 0)
1282 			goto failed;
1283 	}
1284 
1285 	tun_debug(KERN_INFO, tun, "tun_set_iff\n");
1286 
1287 	if (ifr->ifr_flags & IFF_NO_PI)
1288 		tun->flags |= TUN_NO_PI;
1289 	else
1290 		tun->flags &= ~TUN_NO_PI;
1291 
1292 	if (ifr->ifr_flags & IFF_ONE_QUEUE)
1293 		tun->flags |= TUN_ONE_QUEUE;
1294 	else
1295 		tun->flags &= ~TUN_ONE_QUEUE;
1296 
1297 	if (ifr->ifr_flags & IFF_VNET_HDR)
1298 		tun->flags |= TUN_VNET_HDR;
1299 	else
1300 		tun->flags &= ~TUN_VNET_HDR;
1301 
1302 	/* Make sure persistent devices do not get stuck in
1303 	 * xoff state.
1304 	 */
1305 	if (netif_running(tun->dev))
1306 		netif_wake_queue(tun->dev);
1307 
1308 	strcpy(ifr->ifr_name, tun->dev->name);
1309 	return 0;
1310 
1311  err_free_sk:
1312 	tun_free_netdev(dev);
1313  err_free_dev:
1314 	free_netdev(dev);
1315  failed:
1316 	return err;
1317 }
1318 
1319 static int tun_get_iff(struct net *net, struct tun_struct *tun,
1320 		       struct ifreq *ifr)
1321 {
1322 	tun_debug(KERN_INFO, tun, "tun_get_iff\n");
1323 
1324 	strcpy(ifr->ifr_name, tun->dev->name);
1325 
1326 	ifr->ifr_flags = tun_flags(tun);
1327 
1328 	return 0;
1329 }
1330 
1331 /* This is like a cut-down ethtool ops, except done via tun fd so no
1332  * privs required. */
1333 static int set_offload(struct tun_struct *tun, unsigned long arg)
1334 {
1335 	netdev_features_t features = 0;
1336 
1337 	if (arg & TUN_F_CSUM) {
1338 		features |= NETIF_F_HW_CSUM;
1339 		arg &= ~TUN_F_CSUM;
1340 
1341 		if (arg & (TUN_F_TSO4|TUN_F_TSO6)) {
1342 			if (arg & TUN_F_TSO_ECN) {
1343 				features |= NETIF_F_TSO_ECN;
1344 				arg &= ~TUN_F_TSO_ECN;
1345 			}
1346 			if (arg & TUN_F_TSO4)
1347 				features |= NETIF_F_TSO;
1348 			if (arg & TUN_F_TSO6)
1349 				features |= NETIF_F_TSO6;
1350 			arg &= ~(TUN_F_TSO4|TUN_F_TSO6);
1351 		}
1352 
1353 		if (arg & TUN_F_UFO) {
1354 			features |= NETIF_F_UFO;
1355 			arg &= ~TUN_F_UFO;
1356 		}
1357 	}
1358 
1359 	/* This gives the user a way to test for new features in future by
1360 	 * trying to set them. */
1361 	if (arg)
1362 		return -EINVAL;
1363 
1364 	tun->set_features = features;
1365 	netdev_update_features(tun->dev);
1366 
1367 	return 0;
1368 }
1369 
1370 static long __tun_chr_ioctl(struct file *file, unsigned int cmd,
1371 			    unsigned long arg, int ifreq_len)
1372 {
1373 	struct tun_file *tfile = file->private_data;
1374 	struct tun_struct *tun;
1375 	void __user* argp = (void __user*)arg;
1376 	struct sock_fprog fprog;
1377 	struct ifreq ifr;
1378 	int sndbuf;
1379 	int vnet_hdr_sz;
1380 	int ret;
1381 
1382 	if (cmd == TUNSETIFF || _IOC_TYPE(cmd) == 0x89) {
1383 		if (copy_from_user(&ifr, argp, ifreq_len))
1384 			return -EFAULT;
1385 	} else {
1386 		memset(&ifr, 0, sizeof(ifr));
1387 	}
1388 	if (cmd == TUNGETFEATURES) {
1389 		/* Currently this just means: "what IFF flags are valid?".
1390 		 * This is needed because we never checked for invalid flags on
1391 		 * TUNSETIFF. */
1392 		return put_user(IFF_TUN | IFF_TAP | IFF_NO_PI | IFF_ONE_QUEUE |
1393 				IFF_VNET_HDR,
1394 				(unsigned int __user*)argp);
1395 	}
1396 
1397 	rtnl_lock();
1398 
1399 	tun = __tun_get(tfile);
1400 	if (cmd == TUNSETIFF && !tun) {
1401 		ifr.ifr_name[IFNAMSIZ-1] = '\0';
1402 
1403 		ret = tun_set_iff(tfile->net, file, &ifr);
1404 
1405 		if (ret)
1406 			goto unlock;
1407 
1408 		if (copy_to_user(argp, &ifr, ifreq_len))
1409 			ret = -EFAULT;
1410 		goto unlock;
1411 	}
1412 
1413 	ret = -EBADFD;
1414 	if (!tun)
1415 		goto unlock;
1416 
1417 	tun_debug(KERN_INFO, tun, "tun_chr_ioctl cmd %d\n", cmd);
1418 
1419 	ret = 0;
1420 	switch (cmd) {
1421 	case TUNGETIFF:
1422 		ret = tun_get_iff(current->nsproxy->net_ns, tun, &ifr);
1423 		if (ret)
1424 			break;
1425 
1426 		if (copy_to_user(argp, &ifr, ifreq_len))
1427 			ret = -EFAULT;
1428 		break;
1429 
1430 	case TUNSETNOCSUM:
1431 		/* Disable/Enable checksum */
1432 
1433 		/* [unimplemented] */
1434 		tun_debug(KERN_INFO, tun, "ignored: set checksum %s\n",
1435 			  arg ? "disabled" : "enabled");
1436 		break;
1437 
1438 	case TUNSETPERSIST:
1439 		/* Disable/Enable persist mode */
1440 		if (arg)
1441 			tun->flags |= TUN_PERSIST;
1442 		else
1443 			tun->flags &= ~TUN_PERSIST;
1444 
1445 		tun_debug(KERN_INFO, tun, "persist %s\n",
1446 			  arg ? "enabled" : "disabled");
1447 		break;
1448 
1449 	case TUNSETOWNER:
1450 		/* Set owner of the device */
1451 		tun->owner = (uid_t) arg;
1452 
1453 		tun_debug(KERN_INFO, tun, "owner set to %d\n", tun->owner);
1454 		break;
1455 
1456 	case TUNSETGROUP:
1457 		/* Set group of the device */
1458 		tun->group= (gid_t) arg;
1459 
1460 		tun_debug(KERN_INFO, tun, "group set to %d\n", tun->group);
1461 		break;
1462 
1463 	case TUNSETLINK:
1464 		/* Only allow setting the type when the interface is down */
1465 		if (tun->dev->flags & IFF_UP) {
1466 			tun_debug(KERN_INFO, tun,
1467 				  "Linktype set failed because interface is up\n");
1468 			ret = -EBUSY;
1469 		} else {
1470 			tun->dev->type = (int) arg;
1471 			tun_debug(KERN_INFO, tun, "linktype set to %d\n",
1472 				  tun->dev->type);
1473 			ret = 0;
1474 		}
1475 		break;
1476 
1477 #ifdef TUN_DEBUG
1478 	case TUNSETDEBUG:
1479 		tun->debug = arg;
1480 		break;
1481 #endif
1482 	case TUNSETOFFLOAD:
1483 		ret = set_offload(tun, arg);
1484 		break;
1485 
1486 	case TUNSETTXFILTER:
1487 		/* Can be set only for TAPs */
1488 		ret = -EINVAL;
1489 		if ((tun->flags & TUN_TYPE_MASK) != TUN_TAP_DEV)
1490 			break;
1491 		ret = update_filter(&tun->txflt, (void __user *)arg);
1492 		break;
1493 
1494 	case SIOCGIFHWADDR:
1495 		/* Get hw address */
1496 		memcpy(ifr.ifr_hwaddr.sa_data, tun->dev->dev_addr, ETH_ALEN);
1497 		ifr.ifr_hwaddr.sa_family = tun->dev->type;
1498 		if (copy_to_user(argp, &ifr, ifreq_len))
1499 			ret = -EFAULT;
1500 		break;
1501 
1502 	case SIOCSIFHWADDR:
1503 		/* Set hw address */
1504 		tun_debug(KERN_DEBUG, tun, "set hw address: %pM\n",
1505 			  ifr.ifr_hwaddr.sa_data);
1506 
1507 		ret = dev_set_mac_address(tun->dev, &ifr.ifr_hwaddr);
1508 		break;
1509 
1510 	case TUNGETSNDBUF:
1511 		sndbuf = tun->socket.sk->sk_sndbuf;
1512 		if (copy_to_user(argp, &sndbuf, sizeof(sndbuf)))
1513 			ret = -EFAULT;
1514 		break;
1515 
1516 	case TUNSETSNDBUF:
1517 		if (copy_from_user(&sndbuf, argp, sizeof(sndbuf))) {
1518 			ret = -EFAULT;
1519 			break;
1520 		}
1521 
1522 		tun->socket.sk->sk_sndbuf = sndbuf;
1523 		break;
1524 
1525 	case TUNGETVNETHDRSZ:
1526 		vnet_hdr_sz = tun->vnet_hdr_sz;
1527 		if (copy_to_user(argp, &vnet_hdr_sz, sizeof(vnet_hdr_sz)))
1528 			ret = -EFAULT;
1529 		break;
1530 
1531 	case TUNSETVNETHDRSZ:
1532 		if (copy_from_user(&vnet_hdr_sz, argp, sizeof(vnet_hdr_sz))) {
1533 			ret = -EFAULT;
1534 			break;
1535 		}
1536 		if (vnet_hdr_sz < (int)sizeof(struct virtio_net_hdr)) {
1537 			ret = -EINVAL;
1538 			break;
1539 		}
1540 
1541 		tun->vnet_hdr_sz = vnet_hdr_sz;
1542 		break;
1543 
1544 	case TUNATTACHFILTER:
1545 		/* Can be set only for TAPs */
1546 		ret = -EINVAL;
1547 		if ((tun->flags & TUN_TYPE_MASK) != TUN_TAP_DEV)
1548 			break;
1549 		ret = -EFAULT;
1550 		if (copy_from_user(&fprog, argp, sizeof(fprog)))
1551 			break;
1552 
1553 		ret = sk_attach_filter(&fprog, tun->socket.sk);
1554 		break;
1555 
1556 	case TUNDETACHFILTER:
1557 		/* Can be set only for TAPs */
1558 		ret = -EINVAL;
1559 		if ((tun->flags & TUN_TYPE_MASK) != TUN_TAP_DEV)
1560 			break;
1561 		ret = sk_detach_filter(tun->socket.sk);
1562 		break;
1563 
1564 	default:
1565 		ret = -EINVAL;
1566 		break;
1567 	}
1568 
1569 unlock:
1570 	rtnl_unlock();
1571 	if (tun)
1572 		tun_put(tun);
1573 	return ret;
1574 }
1575 
1576 static long tun_chr_ioctl(struct file *file,
1577 			  unsigned int cmd, unsigned long arg)
1578 {
1579 	return __tun_chr_ioctl(file, cmd, arg, sizeof (struct ifreq));
1580 }
1581 
1582 #ifdef CONFIG_COMPAT
1583 static long tun_chr_compat_ioctl(struct file *file,
1584 			 unsigned int cmd, unsigned long arg)
1585 {
1586 	switch (cmd) {
1587 	case TUNSETIFF:
1588 	case TUNGETIFF:
1589 	case TUNSETTXFILTER:
1590 	case TUNGETSNDBUF:
1591 	case TUNSETSNDBUF:
1592 	case SIOCGIFHWADDR:
1593 	case SIOCSIFHWADDR:
1594 		arg = (unsigned long)compat_ptr(arg);
1595 		break;
1596 	default:
1597 		arg = (compat_ulong_t)arg;
1598 		break;
1599 	}
1600 
1601 	/*
1602 	 * compat_ifreq is shorter than ifreq, so we must not access beyond
1603 	 * the end of that structure. All fields that are used in this
1604 	 * driver are compatible though, we don't need to convert the
1605 	 * contents.
1606 	 */
1607 	return __tun_chr_ioctl(file, cmd, arg, sizeof(struct compat_ifreq));
1608 }
1609 #endif /* CONFIG_COMPAT */
1610 
1611 static int tun_chr_fasync(int fd, struct file *file, int on)
1612 {
1613 	struct tun_struct *tun = tun_get(file);
1614 	int ret;
1615 
1616 	if (!tun)
1617 		return -EBADFD;
1618 
1619 	tun_debug(KERN_INFO, tun, "tun_chr_fasync %d\n", on);
1620 
1621 	if ((ret = fasync_helper(fd, file, on, &tun->fasync)) < 0)
1622 		goto out;
1623 
1624 	if (on) {
1625 		ret = __f_setown(file, task_pid(current), PIDTYPE_PID, 0);
1626 		if (ret)
1627 			goto out;
1628 		tun->flags |= TUN_FASYNC;
1629 	} else
1630 		tun->flags &= ~TUN_FASYNC;
1631 	ret = 0;
1632 out:
1633 	tun_put(tun);
1634 	return ret;
1635 }
1636 
1637 static int tun_chr_open(struct inode *inode, struct file * file)
1638 {
1639 	struct tun_file *tfile;
1640 
1641 	DBG1(KERN_INFO, "tunX: tun_chr_open\n");
1642 
1643 	tfile = kmalloc(sizeof(*tfile), GFP_KERNEL);
1644 	if (!tfile)
1645 		return -ENOMEM;
1646 	atomic_set(&tfile->count, 0);
1647 	tfile->tun = NULL;
1648 	tfile->net = get_net(current->nsproxy->net_ns);
1649 	file->private_data = tfile;
1650 	return 0;
1651 }
1652 
1653 static int tun_chr_close(struct inode *inode, struct file *file)
1654 {
1655 	struct tun_file *tfile = file->private_data;
1656 	struct tun_struct *tun;
1657 
1658 	tun = __tun_get(tfile);
1659 	if (tun) {
1660 		struct net_device *dev = tun->dev;
1661 
1662 		tun_debug(KERN_INFO, tun, "tun_chr_close\n");
1663 
1664 		__tun_detach(tun);
1665 
1666 		/* If desirable, unregister the netdevice. */
1667 		if (!(tun->flags & TUN_PERSIST)) {
1668 			rtnl_lock();
1669 			if (dev->reg_state == NETREG_REGISTERED)
1670 				unregister_netdevice(dev);
1671 			rtnl_unlock();
1672 		}
1673 	}
1674 
1675 	tun = tfile->tun;
1676 	if (tun)
1677 		sock_put(tun->socket.sk);
1678 
1679 	put_net(tfile->net);
1680 	kfree(tfile);
1681 
1682 	return 0;
1683 }
1684 
1685 static const struct file_operations tun_fops = {
1686 	.owner	= THIS_MODULE,
1687 	.llseek = no_llseek,
1688 	.read  = do_sync_read,
1689 	.aio_read  = tun_chr_aio_read,
1690 	.write = do_sync_write,
1691 	.aio_write = tun_chr_aio_write,
1692 	.poll	= tun_chr_poll,
1693 	.unlocked_ioctl	= tun_chr_ioctl,
1694 #ifdef CONFIG_COMPAT
1695 	.compat_ioctl = tun_chr_compat_ioctl,
1696 #endif
1697 	.open	= tun_chr_open,
1698 	.release = tun_chr_close,
1699 	.fasync = tun_chr_fasync
1700 };
1701 
1702 static struct miscdevice tun_miscdev = {
1703 	.minor = TUN_MINOR,
1704 	.name = "tun",
1705 	.nodename = "net/tun",
1706 	.fops = &tun_fops,
1707 };
1708 
1709 /* ethtool interface */
1710 
1711 static int tun_get_settings(struct net_device *dev, struct ethtool_cmd *cmd)
1712 {
1713 	cmd->supported		= 0;
1714 	cmd->advertising	= 0;
1715 	ethtool_cmd_speed_set(cmd, SPEED_10);
1716 	cmd->duplex		= DUPLEX_FULL;
1717 	cmd->port		= PORT_TP;
1718 	cmd->phy_address	= 0;
1719 	cmd->transceiver	= XCVR_INTERNAL;
1720 	cmd->autoneg		= AUTONEG_DISABLE;
1721 	cmd->maxtxpkt		= 0;
1722 	cmd->maxrxpkt		= 0;
1723 	return 0;
1724 }
1725 
1726 static void tun_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
1727 {
1728 	struct tun_struct *tun = netdev_priv(dev);
1729 
1730 	strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
1731 	strlcpy(info->version, DRV_VERSION, sizeof(info->version));
1732 
1733 	switch (tun->flags & TUN_TYPE_MASK) {
1734 	case TUN_TUN_DEV:
1735 		strlcpy(info->bus_info, "tun", sizeof(info->bus_info));
1736 		break;
1737 	case TUN_TAP_DEV:
1738 		strlcpy(info->bus_info, "tap", sizeof(info->bus_info));
1739 		break;
1740 	}
1741 }
1742 
1743 static u32 tun_get_msglevel(struct net_device *dev)
1744 {
1745 #ifdef TUN_DEBUG
1746 	struct tun_struct *tun = netdev_priv(dev);
1747 	return tun->debug;
1748 #else
1749 	return -EOPNOTSUPP;
1750 #endif
1751 }
1752 
1753 static void tun_set_msglevel(struct net_device *dev, u32 value)
1754 {
1755 #ifdef TUN_DEBUG
1756 	struct tun_struct *tun = netdev_priv(dev);
1757 	tun->debug = value;
1758 #endif
1759 }
1760 
1761 static const struct ethtool_ops tun_ethtool_ops = {
1762 	.get_settings	= tun_get_settings,
1763 	.get_drvinfo	= tun_get_drvinfo,
1764 	.get_msglevel	= tun_get_msglevel,
1765 	.set_msglevel	= tun_set_msglevel,
1766 	.get_link	= ethtool_op_get_link,
1767 };
1768 
1769 
1770 static int __init tun_init(void)
1771 {
1772 	int ret = 0;
1773 
1774 	pr_info("%s, %s\n", DRV_DESCRIPTION, DRV_VERSION);
1775 	pr_info("%s\n", DRV_COPYRIGHT);
1776 
1777 	ret = rtnl_link_register(&tun_link_ops);
1778 	if (ret) {
1779 		pr_err("Can't register link_ops\n");
1780 		goto err_linkops;
1781 	}
1782 
1783 	ret = misc_register(&tun_miscdev);
1784 	if (ret) {
1785 		pr_err("Can't register misc device %d\n", TUN_MINOR);
1786 		goto err_misc;
1787 	}
1788 	return  0;
1789 err_misc:
1790 	rtnl_link_unregister(&tun_link_ops);
1791 err_linkops:
1792 	return ret;
1793 }
1794 
1795 static void tun_cleanup(void)
1796 {
1797 	misc_deregister(&tun_miscdev);
1798 	rtnl_link_unregister(&tun_link_ops);
1799 }
1800 
1801 /* Get an underlying socket object from tun file.  Returns error unless file is
1802  * attached to a device.  The returned object works like a packet socket, it
1803  * can be used for sock_sendmsg/sock_recvmsg.  The caller is responsible for
1804  * holding a reference to the file for as long as the socket is in use. */
1805 struct socket *tun_get_socket(struct file *file)
1806 {
1807 	struct tun_struct *tun;
1808 	if (file->f_op != &tun_fops)
1809 		return ERR_PTR(-EINVAL);
1810 	tun = tun_get(file);
1811 	if (!tun)
1812 		return ERR_PTR(-EBADFD);
1813 	tun_put(tun);
1814 	return &tun->socket;
1815 }
1816 EXPORT_SYMBOL_GPL(tun_get_socket);
1817 
1818 module_init(tun_init);
1819 module_exit(tun_cleanup);
1820 MODULE_DESCRIPTION(DRV_DESCRIPTION);
1821 MODULE_AUTHOR(DRV_COPYRIGHT);
1822 MODULE_LICENSE("GPL");
1823 MODULE_ALIAS_MISCDEV(TUN_MINOR);
1824 MODULE_ALIAS("devname:net/tun");
1825