xref: /linux/drivers/net/tun.c (revision 95e9fd10f06cb5642028b6b851e32b8c8afb4571)
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 	netif_tx_unlock_bh(tun->dev);
191 
192 	/* Drop read queue */
193 	skb_queue_purge(&tun->socket.sk->sk_receive_queue);
194 
195 	/* Drop the extra count on the net device */
196 	dev_put(tun->dev);
197 }
198 
199 static void tun_detach(struct tun_struct *tun)
200 {
201 	rtnl_lock();
202 	__tun_detach(tun);
203 	rtnl_unlock();
204 }
205 
206 static struct tun_struct *__tun_get(struct tun_file *tfile)
207 {
208 	struct tun_struct *tun = NULL;
209 
210 	if (atomic_inc_not_zero(&tfile->count))
211 		tun = tfile->tun;
212 
213 	return tun;
214 }
215 
216 static struct tun_struct *tun_get(struct file *file)
217 {
218 	return __tun_get(file->private_data);
219 }
220 
221 static void tun_put(struct tun_struct *tun)
222 {
223 	struct tun_file *tfile = tun->tfile;
224 
225 	if (atomic_dec_and_test(&tfile->count))
226 		tun_detach(tfile->tun);
227 }
228 
229 /* TAP filtering */
230 static void addr_hash_set(u32 *mask, const u8 *addr)
231 {
232 	int n = ether_crc(ETH_ALEN, addr) >> 26;
233 	mask[n >> 5] |= (1 << (n & 31));
234 }
235 
236 static unsigned int addr_hash_test(const u32 *mask, const u8 *addr)
237 {
238 	int n = ether_crc(ETH_ALEN, addr) >> 26;
239 	return mask[n >> 5] & (1 << (n & 31));
240 }
241 
242 static int update_filter(struct tap_filter *filter, void __user *arg)
243 {
244 	struct { u8 u[ETH_ALEN]; } *addr;
245 	struct tun_filter uf;
246 	int err, alen, n, nexact;
247 
248 	if (copy_from_user(&uf, arg, sizeof(uf)))
249 		return -EFAULT;
250 
251 	if (!uf.count) {
252 		/* Disabled */
253 		filter->count = 0;
254 		return 0;
255 	}
256 
257 	alen = ETH_ALEN * uf.count;
258 	addr = kmalloc(alen, GFP_KERNEL);
259 	if (!addr)
260 		return -ENOMEM;
261 
262 	if (copy_from_user(addr, arg + sizeof(uf), alen)) {
263 		err = -EFAULT;
264 		goto done;
265 	}
266 
267 	/* The filter is updated without holding any locks. Which is
268 	 * perfectly safe. We disable it first and in the worst
269 	 * case we'll accept a few undesired packets. */
270 	filter->count = 0;
271 	wmb();
272 
273 	/* Use first set of addresses as an exact filter */
274 	for (n = 0; n < uf.count && n < FLT_EXACT_COUNT; n++)
275 		memcpy(filter->addr[n], addr[n].u, ETH_ALEN);
276 
277 	nexact = n;
278 
279 	/* Remaining multicast addresses are hashed,
280 	 * unicast will leave the filter disabled. */
281 	memset(filter->mask, 0, sizeof(filter->mask));
282 	for (; n < uf.count; n++) {
283 		if (!is_multicast_ether_addr(addr[n].u)) {
284 			err = 0; /* no filter */
285 			goto done;
286 		}
287 		addr_hash_set(filter->mask, addr[n].u);
288 	}
289 
290 	/* For ALLMULTI just set the mask to all ones.
291 	 * This overrides the mask populated above. */
292 	if ((uf.flags & TUN_FLT_ALLMULTI))
293 		memset(filter->mask, ~0, sizeof(filter->mask));
294 
295 	/* Now enable the filter */
296 	wmb();
297 	filter->count = nexact;
298 
299 	/* Return the number of exact filters */
300 	err = nexact;
301 
302 done:
303 	kfree(addr);
304 	return err;
305 }
306 
307 /* Returns: 0 - drop, !=0 - accept */
308 static int run_filter(struct tap_filter *filter, const struct sk_buff *skb)
309 {
310 	/* Cannot use eth_hdr(skb) here because skb_mac_hdr() is incorrect
311 	 * at this point. */
312 	struct ethhdr *eh = (struct ethhdr *) skb->data;
313 	int i;
314 
315 	/* Exact match */
316 	for (i = 0; i < filter->count; i++)
317 		if (ether_addr_equal(eh->h_dest, filter->addr[i]))
318 			return 1;
319 
320 	/* Inexact match (multicast only) */
321 	if (is_multicast_ether_addr(eh->h_dest))
322 		return addr_hash_test(filter->mask, eh->h_dest);
323 
324 	return 0;
325 }
326 
327 /*
328  * Checks whether the packet is accepted or not.
329  * Returns: 0 - drop, !=0 - accept
330  */
331 static int check_filter(struct tap_filter *filter, const struct sk_buff *skb)
332 {
333 	if (!filter->count)
334 		return 1;
335 
336 	return run_filter(filter, skb);
337 }
338 
339 /* Network device part of the driver */
340 
341 static const struct ethtool_ops tun_ethtool_ops;
342 
343 /* Net device detach from fd. */
344 static void tun_net_uninit(struct net_device *dev)
345 {
346 	struct tun_struct *tun = netdev_priv(dev);
347 	struct tun_file *tfile = tun->tfile;
348 
349 	/* Inform the methods they need to stop using the dev.
350 	 */
351 	if (tfile) {
352 		wake_up_all(&tun->wq.wait);
353 		if (atomic_dec_and_test(&tfile->count))
354 			__tun_detach(tun);
355 	}
356 }
357 
358 static void tun_free_netdev(struct net_device *dev)
359 {
360 	struct tun_struct *tun = netdev_priv(dev);
361 
362 	BUG_ON(!test_bit(SOCK_EXTERNALLY_ALLOCATED, &tun->socket.flags));
363 
364 	sk_release_kernel(tun->socket.sk);
365 }
366 
367 /* Net device open. */
368 static int tun_net_open(struct net_device *dev)
369 {
370 	netif_start_queue(dev);
371 	return 0;
372 }
373 
374 /* Net device close. */
375 static int tun_net_close(struct net_device *dev)
376 {
377 	netif_stop_queue(dev);
378 	return 0;
379 }
380 
381 /* Net device start xmit */
382 static netdev_tx_t tun_net_xmit(struct sk_buff *skb, struct net_device *dev)
383 {
384 	struct tun_struct *tun = netdev_priv(dev);
385 
386 	tun_debug(KERN_INFO, tun, "tun_net_xmit %d\n", skb->len);
387 
388 	/* Drop packet if interface is not attached */
389 	if (!tun->tfile)
390 		goto drop;
391 
392 	/* Drop if the filter does not like it.
393 	 * This is a noop if the filter is disabled.
394 	 * Filter can be enabled only for the TAP devices. */
395 	if (!check_filter(&tun->txflt, skb))
396 		goto drop;
397 
398 	if (tun->socket.sk->sk_filter &&
399 	    sk_filter(tun->socket.sk, skb))
400 		goto drop;
401 
402 	if (skb_queue_len(&tun->socket.sk->sk_receive_queue) >= dev->tx_queue_len) {
403 		if (!(tun->flags & TUN_ONE_QUEUE)) {
404 			/* Normal queueing mode. */
405 			/* Packet scheduler handles dropping of further packets. */
406 			netif_stop_queue(dev);
407 
408 			/* We won't see all dropped packets individually, so overrun
409 			 * error is more appropriate. */
410 			dev->stats.tx_fifo_errors++;
411 		} else {
412 			/* Single queue mode.
413 			 * Driver handles dropping of all packets itself. */
414 			goto drop;
415 		}
416 	}
417 
418 	/* Orphan the skb - required as we might hang on to it
419 	 * for indefinite time. */
420 	if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC)))
421 		goto drop;
422 	skb_orphan(skb);
423 
424 	/* Enqueue packet */
425 	skb_queue_tail(&tun->socket.sk->sk_receive_queue, skb);
426 
427 	/* Notify and wake up reader process */
428 	if (tun->flags & TUN_FASYNC)
429 		kill_fasync(&tun->fasync, SIGIO, POLL_IN);
430 	wake_up_interruptible_poll(&tun->wq.wait, POLLIN |
431 				   POLLRDNORM | POLLRDBAND);
432 	return NETDEV_TX_OK;
433 
434 drop:
435 	dev->stats.tx_dropped++;
436 	kfree_skb(skb);
437 	return NETDEV_TX_OK;
438 }
439 
440 static void tun_net_mclist(struct net_device *dev)
441 {
442 	/*
443 	 * This callback is supposed to deal with mc filter in
444 	 * _rx_ path and has nothing to do with the _tx_ path.
445 	 * In rx path we always accept everything userspace gives us.
446 	 */
447 }
448 
449 #define MIN_MTU 68
450 #define MAX_MTU 65535
451 
452 static int
453 tun_net_change_mtu(struct net_device *dev, int new_mtu)
454 {
455 	if (new_mtu < MIN_MTU || new_mtu + dev->hard_header_len > MAX_MTU)
456 		return -EINVAL;
457 	dev->mtu = new_mtu;
458 	return 0;
459 }
460 
461 static netdev_features_t tun_net_fix_features(struct net_device *dev,
462 	netdev_features_t features)
463 {
464 	struct tun_struct *tun = netdev_priv(dev);
465 
466 	return (features & tun->set_features) | (features & ~TUN_USER_FEATURES);
467 }
468 #ifdef CONFIG_NET_POLL_CONTROLLER
469 static void tun_poll_controller(struct net_device *dev)
470 {
471 	/*
472 	 * Tun only receives frames when:
473 	 * 1) the char device endpoint gets data from user space
474 	 * 2) the tun socket gets a sendmsg call from user space
475 	 * Since both of those are syncronous operations, we are guaranteed
476 	 * never to have pending data when we poll for it
477 	 * so theres nothing to do here but return.
478 	 * We need this though so netpoll recognizes us as an interface that
479 	 * supports polling, which enables bridge devices in virt setups to
480 	 * still use netconsole
481 	 */
482 	return;
483 }
484 #endif
485 static const struct net_device_ops tun_netdev_ops = {
486 	.ndo_uninit		= tun_net_uninit,
487 	.ndo_open		= tun_net_open,
488 	.ndo_stop		= tun_net_close,
489 	.ndo_start_xmit		= tun_net_xmit,
490 	.ndo_change_mtu		= tun_net_change_mtu,
491 	.ndo_fix_features	= tun_net_fix_features,
492 #ifdef CONFIG_NET_POLL_CONTROLLER
493 	.ndo_poll_controller	= tun_poll_controller,
494 #endif
495 };
496 
497 static const struct net_device_ops tap_netdev_ops = {
498 	.ndo_uninit		= tun_net_uninit,
499 	.ndo_open		= tun_net_open,
500 	.ndo_stop		= tun_net_close,
501 	.ndo_start_xmit		= tun_net_xmit,
502 	.ndo_change_mtu		= tun_net_change_mtu,
503 	.ndo_fix_features	= tun_net_fix_features,
504 	.ndo_set_rx_mode	= tun_net_mclist,
505 	.ndo_set_mac_address	= eth_mac_addr,
506 	.ndo_validate_addr	= eth_validate_addr,
507 #ifdef CONFIG_NET_POLL_CONTROLLER
508 	.ndo_poll_controller	= tun_poll_controller,
509 #endif
510 };
511 
512 /* Initialize net device. */
513 static void tun_net_init(struct net_device *dev)
514 {
515 	struct tun_struct *tun = netdev_priv(dev);
516 
517 	switch (tun->flags & TUN_TYPE_MASK) {
518 	case TUN_TUN_DEV:
519 		dev->netdev_ops = &tun_netdev_ops;
520 
521 		/* Point-to-Point TUN Device */
522 		dev->hard_header_len = 0;
523 		dev->addr_len = 0;
524 		dev->mtu = 1500;
525 
526 		/* Zero header length */
527 		dev->type = ARPHRD_NONE;
528 		dev->flags = IFF_POINTOPOINT | IFF_NOARP | IFF_MULTICAST;
529 		dev->tx_queue_len = TUN_READQ_SIZE;  /* We prefer our own queue length */
530 		break;
531 
532 	case TUN_TAP_DEV:
533 		dev->netdev_ops = &tap_netdev_ops;
534 		/* Ethernet TAP Device */
535 		ether_setup(dev);
536 		dev->priv_flags &= ~IFF_TX_SKB_SHARING;
537 
538 		eth_hw_addr_random(dev);
539 
540 		dev->tx_queue_len = TUN_READQ_SIZE;  /* We prefer our own queue length */
541 		break;
542 	}
543 }
544 
545 /* Character device part */
546 
547 /* Poll */
548 static unsigned int tun_chr_poll(struct file *file, poll_table * wait)
549 {
550 	struct tun_file *tfile = file->private_data;
551 	struct tun_struct *tun = __tun_get(tfile);
552 	struct sock *sk;
553 	unsigned int mask = 0;
554 
555 	if (!tun)
556 		return POLLERR;
557 
558 	sk = tun->socket.sk;
559 
560 	tun_debug(KERN_INFO, tun, "tun_chr_poll\n");
561 
562 	poll_wait(file, &tun->wq.wait, wait);
563 
564 	if (!skb_queue_empty(&sk->sk_receive_queue))
565 		mask |= POLLIN | POLLRDNORM;
566 
567 	if (sock_writeable(sk) ||
568 	    (!test_and_set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags) &&
569 	     sock_writeable(sk)))
570 		mask |= POLLOUT | POLLWRNORM;
571 
572 	if (tun->dev->reg_state != NETREG_REGISTERED)
573 		mask = POLLERR;
574 
575 	tun_put(tun);
576 	return mask;
577 }
578 
579 /* prepad is the amount to reserve at front.  len is length after that.
580  * linear is a hint as to how much to copy (usually headers). */
581 static struct sk_buff *tun_alloc_skb(struct tun_struct *tun,
582 				     size_t prepad, size_t len,
583 				     size_t linear, int noblock)
584 {
585 	struct sock *sk = tun->socket.sk;
586 	struct sk_buff *skb;
587 	int err;
588 
589 	sock_update_classid(sk);
590 
591 	/* Under a page?  Don't bother with paged skb. */
592 	if (prepad + len < PAGE_SIZE || !linear)
593 		linear = len;
594 
595 	skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
596 				   &err);
597 	if (!skb)
598 		return ERR_PTR(err);
599 
600 	skb_reserve(skb, prepad);
601 	skb_put(skb, linear);
602 	skb->data_len = len - linear;
603 	skb->len += len - linear;
604 
605 	return skb;
606 }
607 
608 /* set skb frags from iovec, this can move to core network code for reuse */
609 static int zerocopy_sg_from_iovec(struct sk_buff *skb, const struct iovec *from,
610 				  int offset, size_t count)
611 {
612 	int len = iov_length(from, count) - offset;
613 	int copy = skb_headlen(skb);
614 	int size, offset1 = 0;
615 	int i = 0;
616 
617 	/* Skip over from offset */
618 	while (count && (offset >= from->iov_len)) {
619 		offset -= from->iov_len;
620 		++from;
621 		--count;
622 	}
623 
624 	/* copy up to skb headlen */
625 	while (count && (copy > 0)) {
626 		size = min_t(unsigned int, copy, from->iov_len - offset);
627 		if (copy_from_user(skb->data + offset1, from->iov_base + offset,
628 				   size))
629 			return -EFAULT;
630 		if (copy > size) {
631 			++from;
632 			--count;
633 			offset = 0;
634 		} else
635 			offset += size;
636 		copy -= size;
637 		offset1 += size;
638 	}
639 
640 	if (len == offset1)
641 		return 0;
642 
643 	while (count--) {
644 		struct page *page[MAX_SKB_FRAGS];
645 		int num_pages;
646 		unsigned long base;
647 		unsigned long truesize;
648 
649 		len = from->iov_len - offset;
650 		if (!len) {
651 			offset = 0;
652 			++from;
653 			continue;
654 		}
655 		base = (unsigned long)from->iov_base + offset;
656 		size = ((base & ~PAGE_MASK) + len + ~PAGE_MASK) >> PAGE_SHIFT;
657 		if (i + size > MAX_SKB_FRAGS)
658 			return -EMSGSIZE;
659 		num_pages = get_user_pages_fast(base, size, 0, &page[i]);
660 		if (num_pages != size) {
661 			for (i = 0; i < num_pages; i++)
662 				put_page(page[i]);
663 			return -EFAULT;
664 		}
665 		truesize = size * PAGE_SIZE;
666 		skb->data_len += len;
667 		skb->len += len;
668 		skb->truesize += truesize;
669 		atomic_add(truesize, &skb->sk->sk_wmem_alloc);
670 		while (len) {
671 			int off = base & ~PAGE_MASK;
672 			int size = min_t(int, len, PAGE_SIZE - off);
673 			__skb_fill_page_desc(skb, i, page[i], off, size);
674 			skb_shinfo(skb)->nr_frags++;
675 			/* increase sk_wmem_alloc */
676 			base += size;
677 			len -= size;
678 			i++;
679 		}
680 		offset = 0;
681 		++from;
682 	}
683 	return 0;
684 }
685 
686 /* Get packet from user space buffer */
687 static ssize_t tun_get_user(struct tun_struct *tun, void *msg_control,
688 			    const struct iovec *iv, size_t total_len,
689 			    size_t count, int noblock)
690 {
691 	struct tun_pi pi = { 0, cpu_to_be16(ETH_P_IP) };
692 	struct sk_buff *skb;
693 	size_t len = total_len, align = NET_SKB_PAD;
694 	struct virtio_net_hdr gso = { 0 };
695 	int offset = 0;
696 	int copylen;
697 	bool zerocopy = false;
698 	int err;
699 
700 	if (!(tun->flags & TUN_NO_PI)) {
701 		if ((len -= sizeof(pi)) > total_len)
702 			return -EINVAL;
703 
704 		if (memcpy_fromiovecend((void *)&pi, iv, 0, sizeof(pi)))
705 			return -EFAULT;
706 		offset += sizeof(pi);
707 	}
708 
709 	if (tun->flags & TUN_VNET_HDR) {
710 		if ((len -= tun->vnet_hdr_sz) > total_len)
711 			return -EINVAL;
712 
713 		if (memcpy_fromiovecend((void *)&gso, iv, offset, sizeof(gso)))
714 			return -EFAULT;
715 
716 		if ((gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
717 		    gso.csum_start + gso.csum_offset + 2 > gso.hdr_len)
718 			gso.hdr_len = gso.csum_start + gso.csum_offset + 2;
719 
720 		if (gso.hdr_len > len)
721 			return -EINVAL;
722 		offset += tun->vnet_hdr_sz;
723 	}
724 
725 	if ((tun->flags & TUN_TYPE_MASK) == TUN_TAP_DEV) {
726 		align += NET_IP_ALIGN;
727 		if (unlikely(len < ETH_HLEN ||
728 			     (gso.hdr_len && gso.hdr_len < ETH_HLEN)))
729 			return -EINVAL;
730 	}
731 
732 	if (msg_control)
733 		zerocopy = true;
734 
735 	if (zerocopy) {
736 		/* Userspace may produce vectors with count greater than
737 		 * MAX_SKB_FRAGS, so we need to linearize parts of the skb
738 		 * to let the rest of data to be fit in the frags.
739 		 */
740 		if (count > MAX_SKB_FRAGS) {
741 			copylen = iov_length(iv, count - MAX_SKB_FRAGS);
742 			if (copylen < offset)
743 				copylen = 0;
744 			else
745 				copylen -= offset;
746 		} else
747 				copylen = 0;
748 		/* There are 256 bytes to be copied in skb, so there is enough
749 		 * room for skb expand head in case it is used.
750 		 * The rest of the buffer is mapped from userspace.
751 		 */
752 		if (copylen < gso.hdr_len)
753 			copylen = gso.hdr_len;
754 		if (!copylen)
755 			copylen = GOODCOPY_LEN;
756 	} else
757 		copylen = len;
758 
759 	skb = tun_alloc_skb(tun, align, copylen, gso.hdr_len, noblock);
760 	if (IS_ERR(skb)) {
761 		if (PTR_ERR(skb) != -EAGAIN)
762 			tun->dev->stats.rx_dropped++;
763 		return PTR_ERR(skb);
764 	}
765 
766 	if (zerocopy)
767 		err = zerocopy_sg_from_iovec(skb, iv, offset, count);
768 	else
769 		err = skb_copy_datagram_from_iovec(skb, 0, iv, offset, len);
770 
771 	if (err) {
772 		tun->dev->stats.rx_dropped++;
773 		kfree_skb(skb);
774 		return -EFAULT;
775 	}
776 
777 	if (gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
778 		if (!skb_partial_csum_set(skb, gso.csum_start,
779 					  gso.csum_offset)) {
780 			tun->dev->stats.rx_frame_errors++;
781 			kfree_skb(skb);
782 			return -EINVAL;
783 		}
784 	}
785 
786 	switch (tun->flags & TUN_TYPE_MASK) {
787 	case TUN_TUN_DEV:
788 		if (tun->flags & TUN_NO_PI) {
789 			switch (skb->data[0] & 0xf0) {
790 			case 0x40:
791 				pi.proto = htons(ETH_P_IP);
792 				break;
793 			case 0x60:
794 				pi.proto = htons(ETH_P_IPV6);
795 				break;
796 			default:
797 				tun->dev->stats.rx_dropped++;
798 				kfree_skb(skb);
799 				return -EINVAL;
800 			}
801 		}
802 
803 		skb_reset_mac_header(skb);
804 		skb->protocol = pi.proto;
805 		skb->dev = tun->dev;
806 		break;
807 	case TUN_TAP_DEV:
808 		skb->protocol = eth_type_trans(skb, tun->dev);
809 		break;
810 	}
811 
812 	if (gso.gso_type != VIRTIO_NET_HDR_GSO_NONE) {
813 		pr_debug("GSO!\n");
814 		switch (gso.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
815 		case VIRTIO_NET_HDR_GSO_TCPV4:
816 			skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
817 			break;
818 		case VIRTIO_NET_HDR_GSO_TCPV6:
819 			skb_shinfo(skb)->gso_type = SKB_GSO_TCPV6;
820 			break;
821 		case VIRTIO_NET_HDR_GSO_UDP:
822 			skb_shinfo(skb)->gso_type = SKB_GSO_UDP;
823 			break;
824 		default:
825 			tun->dev->stats.rx_frame_errors++;
826 			kfree_skb(skb);
827 			return -EINVAL;
828 		}
829 
830 		if (gso.gso_type & VIRTIO_NET_HDR_GSO_ECN)
831 			skb_shinfo(skb)->gso_type |= SKB_GSO_TCP_ECN;
832 
833 		skb_shinfo(skb)->gso_size = gso.gso_size;
834 		if (skb_shinfo(skb)->gso_size == 0) {
835 			tun->dev->stats.rx_frame_errors++;
836 			kfree_skb(skb);
837 			return -EINVAL;
838 		}
839 
840 		/* Header must be checked, and gso_segs computed. */
841 		skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
842 		skb_shinfo(skb)->gso_segs = 0;
843 	}
844 
845 	/* copy skb_ubuf_info for callback when skb has no error */
846 	if (zerocopy) {
847 		skb_shinfo(skb)->destructor_arg = msg_control;
848 		skb_shinfo(skb)->tx_flags |= SKBTX_DEV_ZEROCOPY;
849 	}
850 
851 	netif_rx_ni(skb);
852 
853 	tun->dev->stats.rx_packets++;
854 	tun->dev->stats.rx_bytes += len;
855 
856 	return total_len;
857 }
858 
859 static ssize_t tun_chr_aio_write(struct kiocb *iocb, const struct iovec *iv,
860 			      unsigned long count, loff_t pos)
861 {
862 	struct file *file = iocb->ki_filp;
863 	struct tun_struct *tun = tun_get(file);
864 	ssize_t result;
865 
866 	if (!tun)
867 		return -EBADFD;
868 
869 	tun_debug(KERN_INFO, tun, "tun_chr_write %ld\n", count);
870 
871 	result = tun_get_user(tun, NULL, iv, iov_length(iv, count), count,
872 			      file->f_flags & O_NONBLOCK);
873 
874 	tun_put(tun);
875 	return result;
876 }
877 
878 /* Put packet to the user space buffer */
879 static ssize_t tun_put_user(struct tun_struct *tun,
880 			    struct sk_buff *skb,
881 			    const struct iovec *iv, int len)
882 {
883 	struct tun_pi pi = { 0, skb->protocol };
884 	ssize_t total = 0;
885 
886 	if (!(tun->flags & TUN_NO_PI)) {
887 		if ((len -= sizeof(pi)) < 0)
888 			return -EINVAL;
889 
890 		if (len < skb->len) {
891 			/* Packet will be striped */
892 			pi.flags |= TUN_PKT_STRIP;
893 		}
894 
895 		if (memcpy_toiovecend(iv, (void *) &pi, 0, sizeof(pi)))
896 			return -EFAULT;
897 		total += sizeof(pi);
898 	}
899 
900 	if (tun->flags & TUN_VNET_HDR) {
901 		struct virtio_net_hdr gso = { 0 }; /* no info leak */
902 		if ((len -= tun->vnet_hdr_sz) < 0)
903 			return -EINVAL;
904 
905 		if (skb_is_gso(skb)) {
906 			struct skb_shared_info *sinfo = skb_shinfo(skb);
907 
908 			/* This is a hint as to how much should be linear. */
909 			gso.hdr_len = skb_headlen(skb);
910 			gso.gso_size = sinfo->gso_size;
911 			if (sinfo->gso_type & SKB_GSO_TCPV4)
912 				gso.gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
913 			else if (sinfo->gso_type & SKB_GSO_TCPV6)
914 				gso.gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
915 			else if (sinfo->gso_type & SKB_GSO_UDP)
916 				gso.gso_type = VIRTIO_NET_HDR_GSO_UDP;
917 			else {
918 				pr_err("unexpected GSO type: "
919 				       "0x%x, gso_size %d, hdr_len %d\n",
920 				       sinfo->gso_type, gso.gso_size,
921 				       gso.hdr_len);
922 				print_hex_dump(KERN_ERR, "tun: ",
923 					       DUMP_PREFIX_NONE,
924 					       16, 1, skb->head,
925 					       min((int)gso.hdr_len, 64), true);
926 				WARN_ON_ONCE(1);
927 				return -EINVAL;
928 			}
929 			if (sinfo->gso_type & SKB_GSO_TCP_ECN)
930 				gso.gso_type |= VIRTIO_NET_HDR_GSO_ECN;
931 		} else
932 			gso.gso_type = VIRTIO_NET_HDR_GSO_NONE;
933 
934 		if (skb->ip_summed == CHECKSUM_PARTIAL) {
935 			gso.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
936 			gso.csum_start = skb_checksum_start_offset(skb);
937 			gso.csum_offset = skb->csum_offset;
938 		} else if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
939 			gso.flags = VIRTIO_NET_HDR_F_DATA_VALID;
940 		} /* else everything is zero */
941 
942 		if (unlikely(memcpy_toiovecend(iv, (void *)&gso, total,
943 					       sizeof(gso))))
944 			return -EFAULT;
945 		total += tun->vnet_hdr_sz;
946 	}
947 
948 	len = min_t(int, skb->len, len);
949 
950 	skb_copy_datagram_const_iovec(skb, 0, iv, total, len);
951 	total += skb->len;
952 
953 	tun->dev->stats.tx_packets++;
954 	tun->dev->stats.tx_bytes += len;
955 
956 	return total;
957 }
958 
959 static ssize_t tun_do_read(struct tun_struct *tun,
960 			   struct kiocb *iocb, const struct iovec *iv,
961 			   ssize_t len, int noblock)
962 {
963 	DECLARE_WAITQUEUE(wait, current);
964 	struct sk_buff *skb;
965 	ssize_t ret = 0;
966 
967 	tun_debug(KERN_INFO, tun, "tun_chr_read\n");
968 
969 	if (unlikely(!noblock))
970 		add_wait_queue(&tun->wq.wait, &wait);
971 	while (len) {
972 		current->state = TASK_INTERRUPTIBLE;
973 
974 		/* Read frames from the queue */
975 		if (!(skb=skb_dequeue(&tun->socket.sk->sk_receive_queue))) {
976 			if (noblock) {
977 				ret = -EAGAIN;
978 				break;
979 			}
980 			if (signal_pending(current)) {
981 				ret = -ERESTARTSYS;
982 				break;
983 			}
984 			if (tun->dev->reg_state != NETREG_REGISTERED) {
985 				ret = -EIO;
986 				break;
987 			}
988 
989 			/* Nothing to read, let's sleep */
990 			schedule();
991 			continue;
992 		}
993 		netif_wake_queue(tun->dev);
994 
995 		ret = tun_put_user(tun, skb, iv, len);
996 		kfree_skb(skb);
997 		break;
998 	}
999 
1000 	current->state = TASK_RUNNING;
1001 	if (unlikely(!noblock))
1002 		remove_wait_queue(&tun->wq.wait, &wait);
1003 
1004 	return ret;
1005 }
1006 
1007 static ssize_t tun_chr_aio_read(struct kiocb *iocb, const struct iovec *iv,
1008 			    unsigned long count, loff_t pos)
1009 {
1010 	struct file *file = iocb->ki_filp;
1011 	struct tun_file *tfile = file->private_data;
1012 	struct tun_struct *tun = __tun_get(tfile);
1013 	ssize_t len, ret;
1014 
1015 	if (!tun)
1016 		return -EBADFD;
1017 	len = iov_length(iv, count);
1018 	if (len < 0) {
1019 		ret = -EINVAL;
1020 		goto out;
1021 	}
1022 
1023 	ret = tun_do_read(tun, iocb, iv, len, file->f_flags & O_NONBLOCK);
1024 	ret = min_t(ssize_t, ret, len);
1025 out:
1026 	tun_put(tun);
1027 	return ret;
1028 }
1029 
1030 static void tun_setup(struct net_device *dev)
1031 {
1032 	struct tun_struct *tun = netdev_priv(dev);
1033 
1034 	tun->owner = -1;
1035 	tun->group = -1;
1036 
1037 	dev->ethtool_ops = &tun_ethtool_ops;
1038 	dev->destructor = tun_free_netdev;
1039 }
1040 
1041 /* Trivial set of netlink ops to allow deleting tun or tap
1042  * device with netlink.
1043  */
1044 static int tun_validate(struct nlattr *tb[], struct nlattr *data[])
1045 {
1046 	return -EINVAL;
1047 }
1048 
1049 static struct rtnl_link_ops tun_link_ops __read_mostly = {
1050 	.kind		= DRV_NAME,
1051 	.priv_size	= sizeof(struct tun_struct),
1052 	.setup		= tun_setup,
1053 	.validate	= tun_validate,
1054 };
1055 
1056 static void tun_sock_write_space(struct sock *sk)
1057 {
1058 	struct tun_struct *tun;
1059 	wait_queue_head_t *wqueue;
1060 
1061 	if (!sock_writeable(sk))
1062 		return;
1063 
1064 	if (!test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags))
1065 		return;
1066 
1067 	wqueue = sk_sleep(sk);
1068 	if (wqueue && waitqueue_active(wqueue))
1069 		wake_up_interruptible_sync_poll(wqueue, POLLOUT |
1070 						POLLWRNORM | POLLWRBAND);
1071 
1072 	tun = tun_sk(sk)->tun;
1073 	kill_fasync(&tun->fasync, SIGIO, POLL_OUT);
1074 }
1075 
1076 static void tun_sock_destruct(struct sock *sk)
1077 {
1078 	free_netdev(tun_sk(sk)->tun->dev);
1079 }
1080 
1081 static int tun_sendmsg(struct kiocb *iocb, struct socket *sock,
1082 		       struct msghdr *m, size_t total_len)
1083 {
1084 	struct tun_struct *tun = container_of(sock, struct tun_struct, socket);
1085 	return tun_get_user(tun, m->msg_control, m->msg_iov, total_len,
1086 			    m->msg_iovlen, m->msg_flags & MSG_DONTWAIT);
1087 }
1088 
1089 static int tun_recvmsg(struct kiocb *iocb, struct socket *sock,
1090 		       struct msghdr *m, size_t total_len,
1091 		       int flags)
1092 {
1093 	struct tun_struct *tun = container_of(sock, struct tun_struct, socket);
1094 	int ret;
1095 	if (flags & ~(MSG_DONTWAIT|MSG_TRUNC))
1096 		return -EINVAL;
1097 	ret = tun_do_read(tun, iocb, m->msg_iov, total_len,
1098 			  flags & MSG_DONTWAIT);
1099 	if (ret > total_len) {
1100 		m->msg_flags |= MSG_TRUNC;
1101 		ret = flags & MSG_TRUNC ? ret : total_len;
1102 	}
1103 	return ret;
1104 }
1105 
1106 static int tun_release(struct socket *sock)
1107 {
1108 	if (sock->sk)
1109 		sock_put(sock->sk);
1110 	return 0;
1111 }
1112 
1113 /* Ops structure to mimic raw sockets with tun */
1114 static const struct proto_ops tun_socket_ops = {
1115 	.sendmsg = tun_sendmsg,
1116 	.recvmsg = tun_recvmsg,
1117 	.release = tun_release,
1118 };
1119 
1120 static struct proto tun_proto = {
1121 	.name		= "tun",
1122 	.owner		= THIS_MODULE,
1123 	.obj_size	= sizeof(struct tun_sock),
1124 };
1125 
1126 static int tun_flags(struct tun_struct *tun)
1127 {
1128 	int flags = 0;
1129 
1130 	if (tun->flags & TUN_TUN_DEV)
1131 		flags |= IFF_TUN;
1132 	else
1133 		flags |= IFF_TAP;
1134 
1135 	if (tun->flags & TUN_NO_PI)
1136 		flags |= IFF_NO_PI;
1137 
1138 	if (tun->flags & TUN_ONE_QUEUE)
1139 		flags |= IFF_ONE_QUEUE;
1140 
1141 	if (tun->flags & TUN_VNET_HDR)
1142 		flags |= IFF_VNET_HDR;
1143 
1144 	return flags;
1145 }
1146 
1147 static ssize_t tun_show_flags(struct device *dev, struct device_attribute *attr,
1148 			      char *buf)
1149 {
1150 	struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1151 	return sprintf(buf, "0x%x\n", tun_flags(tun));
1152 }
1153 
1154 static ssize_t tun_show_owner(struct device *dev, struct device_attribute *attr,
1155 			      char *buf)
1156 {
1157 	struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1158 	return sprintf(buf, "%d\n", tun->owner);
1159 }
1160 
1161 static ssize_t tun_show_group(struct device *dev, struct device_attribute *attr,
1162 			      char *buf)
1163 {
1164 	struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1165 	return sprintf(buf, "%d\n", tun->group);
1166 }
1167 
1168 static DEVICE_ATTR(tun_flags, 0444, tun_show_flags, NULL);
1169 static DEVICE_ATTR(owner, 0444, tun_show_owner, NULL);
1170 static DEVICE_ATTR(group, 0444, tun_show_group, NULL);
1171 
1172 static int tun_set_iff(struct net *net, struct file *file, struct ifreq *ifr)
1173 {
1174 	struct sock *sk;
1175 	struct tun_struct *tun;
1176 	struct net_device *dev;
1177 	int err;
1178 
1179 	dev = __dev_get_by_name(net, ifr->ifr_name);
1180 	if (dev) {
1181 		const struct cred *cred = current_cred();
1182 
1183 		if (ifr->ifr_flags & IFF_TUN_EXCL)
1184 			return -EBUSY;
1185 		if ((ifr->ifr_flags & IFF_TUN) && dev->netdev_ops == &tun_netdev_ops)
1186 			tun = netdev_priv(dev);
1187 		else if ((ifr->ifr_flags & IFF_TAP) && dev->netdev_ops == &tap_netdev_ops)
1188 			tun = netdev_priv(dev);
1189 		else
1190 			return -EINVAL;
1191 
1192 		if (((tun->owner != -1 && cred->euid != tun->owner) ||
1193 		     (tun->group != -1 && !in_egroup_p(tun->group))) &&
1194 		    !capable(CAP_NET_ADMIN))
1195 			return -EPERM;
1196 		err = security_tun_dev_attach(tun->socket.sk);
1197 		if (err < 0)
1198 			return err;
1199 
1200 		err = tun_attach(tun, file);
1201 		if (err < 0)
1202 			return err;
1203 	}
1204 	else {
1205 		char *name;
1206 		unsigned long flags = 0;
1207 
1208 		if (!capable(CAP_NET_ADMIN))
1209 			return -EPERM;
1210 		err = security_tun_dev_create();
1211 		if (err < 0)
1212 			return err;
1213 
1214 		/* Set dev type */
1215 		if (ifr->ifr_flags & IFF_TUN) {
1216 			/* TUN device */
1217 			flags |= TUN_TUN_DEV;
1218 			name = "tun%d";
1219 		} else if (ifr->ifr_flags & IFF_TAP) {
1220 			/* TAP device */
1221 			flags |= TUN_TAP_DEV;
1222 			name = "tap%d";
1223 		} else
1224 			return -EINVAL;
1225 
1226 		if (*ifr->ifr_name)
1227 			name = ifr->ifr_name;
1228 
1229 		dev = alloc_netdev(sizeof(struct tun_struct), name,
1230 				   tun_setup);
1231 		if (!dev)
1232 			return -ENOMEM;
1233 
1234 		dev_net_set(dev, net);
1235 		dev->rtnl_link_ops = &tun_link_ops;
1236 
1237 		tun = netdev_priv(dev);
1238 		tun->dev = dev;
1239 		tun->flags = flags;
1240 		tun->txflt.count = 0;
1241 		tun->vnet_hdr_sz = sizeof(struct virtio_net_hdr);
1242 		set_bit(SOCK_EXTERNALLY_ALLOCATED, &tun->socket.flags);
1243 
1244 		err = -ENOMEM;
1245 		sk = sk_alloc(&init_net, AF_UNSPEC, GFP_KERNEL, &tun_proto);
1246 		if (!sk)
1247 			goto err_free_dev;
1248 
1249 		sk_change_net(sk, net);
1250 		tun->socket.wq = &tun->wq;
1251 		init_waitqueue_head(&tun->wq.wait);
1252 		tun->socket.ops = &tun_socket_ops;
1253 		sock_init_data(&tun->socket, sk);
1254 		sk->sk_write_space = tun_sock_write_space;
1255 		sk->sk_sndbuf = INT_MAX;
1256 		sock_set_flag(sk, SOCK_ZEROCOPY);
1257 
1258 		tun_sk(sk)->tun = tun;
1259 
1260 		security_tun_dev_post_create(sk);
1261 
1262 		tun_net_init(dev);
1263 
1264 		dev->hw_features = NETIF_F_SG | NETIF_F_FRAGLIST |
1265 			TUN_USER_FEATURES;
1266 		dev->features = dev->hw_features;
1267 
1268 		err = register_netdevice(tun->dev);
1269 		if (err < 0)
1270 			goto err_free_sk;
1271 
1272 		if (device_create_file(&tun->dev->dev, &dev_attr_tun_flags) ||
1273 		    device_create_file(&tun->dev->dev, &dev_attr_owner) ||
1274 		    device_create_file(&tun->dev->dev, &dev_attr_group))
1275 			pr_err("Failed to create tun sysfs files\n");
1276 
1277 		sk->sk_destruct = tun_sock_destruct;
1278 
1279 		err = tun_attach(tun, file);
1280 		if (err < 0)
1281 			goto failed;
1282 	}
1283 
1284 	tun_debug(KERN_INFO, tun, "tun_set_iff\n");
1285 
1286 	if (ifr->ifr_flags & IFF_NO_PI)
1287 		tun->flags |= TUN_NO_PI;
1288 	else
1289 		tun->flags &= ~TUN_NO_PI;
1290 
1291 	if (ifr->ifr_flags & IFF_ONE_QUEUE)
1292 		tun->flags |= TUN_ONE_QUEUE;
1293 	else
1294 		tun->flags &= ~TUN_ONE_QUEUE;
1295 
1296 	if (ifr->ifr_flags & IFF_VNET_HDR)
1297 		tun->flags |= TUN_VNET_HDR;
1298 	else
1299 		tun->flags &= ~TUN_VNET_HDR;
1300 
1301 	/* Make sure persistent devices do not get stuck in
1302 	 * xoff state.
1303 	 */
1304 	if (netif_running(tun->dev))
1305 		netif_wake_queue(tun->dev);
1306 
1307 	strcpy(ifr->ifr_name, tun->dev->name);
1308 	return 0;
1309 
1310  err_free_sk:
1311 	tun_free_netdev(dev);
1312  err_free_dev:
1313 	free_netdev(dev);
1314  failed:
1315 	return err;
1316 }
1317 
1318 static int tun_get_iff(struct net *net, struct tun_struct *tun,
1319 		       struct ifreq *ifr)
1320 {
1321 	tun_debug(KERN_INFO, tun, "tun_get_iff\n");
1322 
1323 	strcpy(ifr->ifr_name, tun->dev->name);
1324 
1325 	ifr->ifr_flags = tun_flags(tun);
1326 
1327 	return 0;
1328 }
1329 
1330 /* This is like a cut-down ethtool ops, except done via tun fd so no
1331  * privs required. */
1332 static int set_offload(struct tun_struct *tun, unsigned long arg)
1333 {
1334 	netdev_features_t features = 0;
1335 
1336 	if (arg & TUN_F_CSUM) {
1337 		features |= NETIF_F_HW_CSUM;
1338 		arg &= ~TUN_F_CSUM;
1339 
1340 		if (arg & (TUN_F_TSO4|TUN_F_TSO6)) {
1341 			if (arg & TUN_F_TSO_ECN) {
1342 				features |= NETIF_F_TSO_ECN;
1343 				arg &= ~TUN_F_TSO_ECN;
1344 			}
1345 			if (arg & TUN_F_TSO4)
1346 				features |= NETIF_F_TSO;
1347 			if (arg & TUN_F_TSO6)
1348 				features |= NETIF_F_TSO6;
1349 			arg &= ~(TUN_F_TSO4|TUN_F_TSO6);
1350 		}
1351 
1352 		if (arg & TUN_F_UFO) {
1353 			features |= NETIF_F_UFO;
1354 			arg &= ~TUN_F_UFO;
1355 		}
1356 	}
1357 
1358 	/* This gives the user a way to test for new features in future by
1359 	 * trying to set them. */
1360 	if (arg)
1361 		return -EINVAL;
1362 
1363 	tun->set_features = features;
1364 	netdev_update_features(tun->dev);
1365 
1366 	return 0;
1367 }
1368 
1369 static long __tun_chr_ioctl(struct file *file, unsigned int cmd,
1370 			    unsigned long arg, int ifreq_len)
1371 {
1372 	struct tun_file *tfile = file->private_data;
1373 	struct tun_struct *tun;
1374 	void __user* argp = (void __user*)arg;
1375 	struct sock_fprog fprog;
1376 	struct ifreq ifr;
1377 	int sndbuf;
1378 	int vnet_hdr_sz;
1379 	int ret;
1380 
1381 	if (cmd == TUNSETIFF || _IOC_TYPE(cmd) == 0x89) {
1382 		if (copy_from_user(&ifr, argp, ifreq_len))
1383 			return -EFAULT;
1384 	} else {
1385 		memset(&ifr, 0, sizeof(ifr));
1386 	}
1387 	if (cmd == TUNGETFEATURES) {
1388 		/* Currently this just means: "what IFF flags are valid?".
1389 		 * This is needed because we never checked for invalid flags on
1390 		 * TUNSETIFF. */
1391 		return put_user(IFF_TUN | IFF_TAP | IFF_NO_PI | IFF_ONE_QUEUE |
1392 				IFF_VNET_HDR,
1393 				(unsigned int __user*)argp);
1394 	}
1395 
1396 	rtnl_lock();
1397 
1398 	tun = __tun_get(tfile);
1399 	if (cmd == TUNSETIFF && !tun) {
1400 		ifr.ifr_name[IFNAMSIZ-1] = '\0';
1401 
1402 		ret = tun_set_iff(tfile->net, file, &ifr);
1403 
1404 		if (ret)
1405 			goto unlock;
1406 
1407 		if (copy_to_user(argp, &ifr, ifreq_len))
1408 			ret = -EFAULT;
1409 		goto unlock;
1410 	}
1411 
1412 	ret = -EBADFD;
1413 	if (!tun)
1414 		goto unlock;
1415 
1416 	tun_debug(KERN_INFO, tun, "tun_chr_ioctl cmd %d\n", cmd);
1417 
1418 	ret = 0;
1419 	switch (cmd) {
1420 	case TUNGETIFF:
1421 		ret = tun_get_iff(current->nsproxy->net_ns, tun, &ifr);
1422 		if (ret)
1423 			break;
1424 
1425 		if (copy_to_user(argp, &ifr, ifreq_len))
1426 			ret = -EFAULT;
1427 		break;
1428 
1429 	case TUNSETNOCSUM:
1430 		/* Disable/Enable checksum */
1431 
1432 		/* [unimplemented] */
1433 		tun_debug(KERN_INFO, tun, "ignored: set checksum %s\n",
1434 			  arg ? "disabled" : "enabled");
1435 		break;
1436 
1437 	case TUNSETPERSIST:
1438 		/* Disable/Enable persist mode */
1439 		if (arg)
1440 			tun->flags |= TUN_PERSIST;
1441 		else
1442 			tun->flags &= ~TUN_PERSIST;
1443 
1444 		tun_debug(KERN_INFO, tun, "persist %s\n",
1445 			  arg ? "enabled" : "disabled");
1446 		break;
1447 
1448 	case TUNSETOWNER:
1449 		/* Set owner of the device */
1450 		tun->owner = (uid_t) arg;
1451 
1452 		tun_debug(KERN_INFO, tun, "owner set to %d\n", tun->owner);
1453 		break;
1454 
1455 	case TUNSETGROUP:
1456 		/* Set group of the device */
1457 		tun->group= (gid_t) arg;
1458 
1459 		tun_debug(KERN_INFO, tun, "group set to %d\n", tun->group);
1460 		break;
1461 
1462 	case TUNSETLINK:
1463 		/* Only allow setting the type when the interface is down */
1464 		if (tun->dev->flags & IFF_UP) {
1465 			tun_debug(KERN_INFO, tun,
1466 				  "Linktype set failed because interface is up\n");
1467 			ret = -EBUSY;
1468 		} else {
1469 			tun->dev->type = (int) arg;
1470 			tun_debug(KERN_INFO, tun, "linktype set to %d\n",
1471 				  tun->dev->type);
1472 			ret = 0;
1473 		}
1474 		break;
1475 
1476 #ifdef TUN_DEBUG
1477 	case TUNSETDEBUG:
1478 		tun->debug = arg;
1479 		break;
1480 #endif
1481 	case TUNSETOFFLOAD:
1482 		ret = set_offload(tun, arg);
1483 		break;
1484 
1485 	case TUNSETTXFILTER:
1486 		/* Can be set only for TAPs */
1487 		ret = -EINVAL;
1488 		if ((tun->flags & TUN_TYPE_MASK) != TUN_TAP_DEV)
1489 			break;
1490 		ret = update_filter(&tun->txflt, (void __user *)arg);
1491 		break;
1492 
1493 	case SIOCGIFHWADDR:
1494 		/* Get hw address */
1495 		memcpy(ifr.ifr_hwaddr.sa_data, tun->dev->dev_addr, ETH_ALEN);
1496 		ifr.ifr_hwaddr.sa_family = tun->dev->type;
1497 		if (copy_to_user(argp, &ifr, ifreq_len))
1498 			ret = -EFAULT;
1499 		break;
1500 
1501 	case SIOCSIFHWADDR:
1502 		/* Set hw address */
1503 		tun_debug(KERN_DEBUG, tun, "set hw address: %pM\n",
1504 			  ifr.ifr_hwaddr.sa_data);
1505 
1506 		ret = dev_set_mac_address(tun->dev, &ifr.ifr_hwaddr);
1507 		break;
1508 
1509 	case TUNGETSNDBUF:
1510 		sndbuf = tun->socket.sk->sk_sndbuf;
1511 		if (copy_to_user(argp, &sndbuf, sizeof(sndbuf)))
1512 			ret = -EFAULT;
1513 		break;
1514 
1515 	case TUNSETSNDBUF:
1516 		if (copy_from_user(&sndbuf, argp, sizeof(sndbuf))) {
1517 			ret = -EFAULT;
1518 			break;
1519 		}
1520 
1521 		tun->socket.sk->sk_sndbuf = sndbuf;
1522 		break;
1523 
1524 	case TUNGETVNETHDRSZ:
1525 		vnet_hdr_sz = tun->vnet_hdr_sz;
1526 		if (copy_to_user(argp, &vnet_hdr_sz, sizeof(vnet_hdr_sz)))
1527 			ret = -EFAULT;
1528 		break;
1529 
1530 	case TUNSETVNETHDRSZ:
1531 		if (copy_from_user(&vnet_hdr_sz, argp, sizeof(vnet_hdr_sz))) {
1532 			ret = -EFAULT;
1533 			break;
1534 		}
1535 		if (vnet_hdr_sz < (int)sizeof(struct virtio_net_hdr)) {
1536 			ret = -EINVAL;
1537 			break;
1538 		}
1539 
1540 		tun->vnet_hdr_sz = vnet_hdr_sz;
1541 		break;
1542 
1543 	case TUNATTACHFILTER:
1544 		/* Can be set only for TAPs */
1545 		ret = -EINVAL;
1546 		if ((tun->flags & TUN_TYPE_MASK) != TUN_TAP_DEV)
1547 			break;
1548 		ret = -EFAULT;
1549 		if (copy_from_user(&fprog, argp, sizeof(fprog)))
1550 			break;
1551 
1552 		ret = sk_attach_filter(&fprog, tun->socket.sk);
1553 		break;
1554 
1555 	case TUNDETACHFILTER:
1556 		/* Can be set only for TAPs */
1557 		ret = -EINVAL;
1558 		if ((tun->flags & TUN_TYPE_MASK) != TUN_TAP_DEV)
1559 			break;
1560 		ret = sk_detach_filter(tun->socket.sk);
1561 		break;
1562 
1563 	default:
1564 		ret = -EINVAL;
1565 		break;
1566 	}
1567 
1568 unlock:
1569 	rtnl_unlock();
1570 	if (tun)
1571 		tun_put(tun);
1572 	return ret;
1573 }
1574 
1575 static long tun_chr_ioctl(struct file *file,
1576 			  unsigned int cmd, unsigned long arg)
1577 {
1578 	return __tun_chr_ioctl(file, cmd, arg, sizeof (struct ifreq));
1579 }
1580 
1581 #ifdef CONFIG_COMPAT
1582 static long tun_chr_compat_ioctl(struct file *file,
1583 			 unsigned int cmd, unsigned long arg)
1584 {
1585 	switch (cmd) {
1586 	case TUNSETIFF:
1587 	case TUNGETIFF:
1588 	case TUNSETTXFILTER:
1589 	case TUNGETSNDBUF:
1590 	case TUNSETSNDBUF:
1591 	case SIOCGIFHWADDR:
1592 	case SIOCSIFHWADDR:
1593 		arg = (unsigned long)compat_ptr(arg);
1594 		break;
1595 	default:
1596 		arg = (compat_ulong_t)arg;
1597 		break;
1598 	}
1599 
1600 	/*
1601 	 * compat_ifreq is shorter than ifreq, so we must not access beyond
1602 	 * the end of that structure. All fields that are used in this
1603 	 * driver are compatible though, we don't need to convert the
1604 	 * contents.
1605 	 */
1606 	return __tun_chr_ioctl(file, cmd, arg, sizeof(struct compat_ifreq));
1607 }
1608 #endif /* CONFIG_COMPAT */
1609 
1610 static int tun_chr_fasync(int fd, struct file *file, int on)
1611 {
1612 	struct tun_struct *tun = tun_get(file);
1613 	int ret;
1614 
1615 	if (!tun)
1616 		return -EBADFD;
1617 
1618 	tun_debug(KERN_INFO, tun, "tun_chr_fasync %d\n", on);
1619 
1620 	if ((ret = fasync_helper(fd, file, on, &tun->fasync)) < 0)
1621 		goto out;
1622 
1623 	if (on) {
1624 		ret = __f_setown(file, task_pid(current), PIDTYPE_PID, 0);
1625 		if (ret)
1626 			goto out;
1627 		tun->flags |= TUN_FASYNC;
1628 	} else
1629 		tun->flags &= ~TUN_FASYNC;
1630 	ret = 0;
1631 out:
1632 	tun_put(tun);
1633 	return ret;
1634 }
1635 
1636 static int tun_chr_open(struct inode *inode, struct file * file)
1637 {
1638 	struct tun_file *tfile;
1639 
1640 	DBG1(KERN_INFO, "tunX: tun_chr_open\n");
1641 
1642 	tfile = kmalloc(sizeof(*tfile), GFP_KERNEL);
1643 	if (!tfile)
1644 		return -ENOMEM;
1645 	atomic_set(&tfile->count, 0);
1646 	tfile->tun = NULL;
1647 	tfile->net = get_net(current->nsproxy->net_ns);
1648 	file->private_data = tfile;
1649 	return 0;
1650 }
1651 
1652 static int tun_chr_close(struct inode *inode, struct file *file)
1653 {
1654 	struct tun_file *tfile = file->private_data;
1655 	struct tun_struct *tun;
1656 
1657 	tun = __tun_get(tfile);
1658 	if (tun) {
1659 		struct net_device *dev = tun->dev;
1660 
1661 		tun_debug(KERN_INFO, tun, "tun_chr_close\n");
1662 
1663 		__tun_detach(tun);
1664 
1665 		/* If desirable, unregister the netdevice. */
1666 		if (!(tun->flags & TUN_PERSIST)) {
1667 			rtnl_lock();
1668 			if (dev->reg_state == NETREG_REGISTERED)
1669 				unregister_netdevice(dev);
1670 			rtnl_unlock();
1671 		}
1672 	}
1673 
1674 	tun = tfile->tun;
1675 	if (tun)
1676 		sock_put(tun->socket.sk);
1677 
1678 	put_net(tfile->net);
1679 	kfree(tfile);
1680 
1681 	return 0;
1682 }
1683 
1684 static const struct file_operations tun_fops = {
1685 	.owner	= THIS_MODULE,
1686 	.llseek = no_llseek,
1687 	.read  = do_sync_read,
1688 	.aio_read  = tun_chr_aio_read,
1689 	.write = do_sync_write,
1690 	.aio_write = tun_chr_aio_write,
1691 	.poll	= tun_chr_poll,
1692 	.unlocked_ioctl	= tun_chr_ioctl,
1693 #ifdef CONFIG_COMPAT
1694 	.compat_ioctl = tun_chr_compat_ioctl,
1695 #endif
1696 	.open	= tun_chr_open,
1697 	.release = tun_chr_close,
1698 	.fasync = tun_chr_fasync
1699 };
1700 
1701 static struct miscdevice tun_miscdev = {
1702 	.minor = TUN_MINOR,
1703 	.name = "tun",
1704 	.nodename = "net/tun",
1705 	.fops = &tun_fops,
1706 };
1707 
1708 /* ethtool interface */
1709 
1710 static int tun_get_settings(struct net_device *dev, struct ethtool_cmd *cmd)
1711 {
1712 	cmd->supported		= 0;
1713 	cmd->advertising	= 0;
1714 	ethtool_cmd_speed_set(cmd, SPEED_10);
1715 	cmd->duplex		= DUPLEX_FULL;
1716 	cmd->port		= PORT_TP;
1717 	cmd->phy_address	= 0;
1718 	cmd->transceiver	= XCVR_INTERNAL;
1719 	cmd->autoneg		= AUTONEG_DISABLE;
1720 	cmd->maxtxpkt		= 0;
1721 	cmd->maxrxpkt		= 0;
1722 	return 0;
1723 }
1724 
1725 static void tun_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
1726 {
1727 	struct tun_struct *tun = netdev_priv(dev);
1728 
1729 	strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
1730 	strlcpy(info->version, DRV_VERSION, sizeof(info->version));
1731 
1732 	switch (tun->flags & TUN_TYPE_MASK) {
1733 	case TUN_TUN_DEV:
1734 		strlcpy(info->bus_info, "tun", sizeof(info->bus_info));
1735 		break;
1736 	case TUN_TAP_DEV:
1737 		strlcpy(info->bus_info, "tap", sizeof(info->bus_info));
1738 		break;
1739 	}
1740 }
1741 
1742 static u32 tun_get_msglevel(struct net_device *dev)
1743 {
1744 #ifdef TUN_DEBUG
1745 	struct tun_struct *tun = netdev_priv(dev);
1746 	return tun->debug;
1747 #else
1748 	return -EOPNOTSUPP;
1749 #endif
1750 }
1751 
1752 static void tun_set_msglevel(struct net_device *dev, u32 value)
1753 {
1754 #ifdef TUN_DEBUG
1755 	struct tun_struct *tun = netdev_priv(dev);
1756 	tun->debug = value;
1757 #endif
1758 }
1759 
1760 static const struct ethtool_ops tun_ethtool_ops = {
1761 	.get_settings	= tun_get_settings,
1762 	.get_drvinfo	= tun_get_drvinfo,
1763 	.get_msglevel	= tun_get_msglevel,
1764 	.set_msglevel	= tun_set_msglevel,
1765 	.get_link	= ethtool_op_get_link,
1766 };
1767 
1768 
1769 static int __init tun_init(void)
1770 {
1771 	int ret = 0;
1772 
1773 	pr_info("%s, %s\n", DRV_DESCRIPTION, DRV_VERSION);
1774 	pr_info("%s\n", DRV_COPYRIGHT);
1775 
1776 	ret = rtnl_link_register(&tun_link_ops);
1777 	if (ret) {
1778 		pr_err("Can't register link_ops\n");
1779 		goto err_linkops;
1780 	}
1781 
1782 	ret = misc_register(&tun_miscdev);
1783 	if (ret) {
1784 		pr_err("Can't register misc device %d\n", TUN_MINOR);
1785 		goto err_misc;
1786 	}
1787 	return  0;
1788 err_misc:
1789 	rtnl_link_unregister(&tun_link_ops);
1790 err_linkops:
1791 	return ret;
1792 }
1793 
1794 static void tun_cleanup(void)
1795 {
1796 	misc_deregister(&tun_miscdev);
1797 	rtnl_link_unregister(&tun_link_ops);
1798 }
1799 
1800 /* Get an underlying socket object from tun file.  Returns error unless file is
1801  * attached to a device.  The returned object works like a packet socket, it
1802  * can be used for sock_sendmsg/sock_recvmsg.  The caller is responsible for
1803  * holding a reference to the file for as long as the socket is in use. */
1804 struct socket *tun_get_socket(struct file *file)
1805 {
1806 	struct tun_struct *tun;
1807 	if (file->f_op != &tun_fops)
1808 		return ERR_PTR(-EINVAL);
1809 	tun = tun_get(file);
1810 	if (!tun)
1811 		return ERR_PTR(-EBADFD);
1812 	tun_put(tun);
1813 	return &tun->socket;
1814 }
1815 EXPORT_SYMBOL_GPL(tun_get_socket);
1816 
1817 module_init(tun_init);
1818 module_exit(tun_cleanup);
1819 MODULE_DESCRIPTION(DRV_DESCRIPTION);
1820 MODULE_AUTHOR(DRV_COPYRIGHT);
1821 MODULE_LICENSE("GPL");
1822 MODULE_ALIAS_MISCDEV(TUN_MINOR);
1823 MODULE_ALIAS("devname:net/tun");
1824