xref: /openbmc/linux/drivers/net/tun.c (revision 1ccd4b7b)
1 /*
2  *  TUN - Universal TUN/TAP device driver.
3  *  Copyright (C) 1999-2002 Maxim Krasnyansky <maxk@qualcomm.com>
4  *
5  *  This program is free software; you can redistribute it and/or modify
6  *  it under the terms of the GNU General Public License as published by
7  *  the Free Software Foundation; either version 2 of the License, or
8  *  (at your option) any later version.
9  *
10  *  This program is distributed in the hope that it will be useful,
11  *  but WITHOUT ANY WARRANTY; without even the implied warranty of
12  *  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
13  *  GNU General Public License for more details.
14  *
15  *  $Id: tun.c,v 1.15 2002/03/01 02:44:24 maxk Exp $
16  */
17 
18 /*
19  *  Changes:
20  *
21  *  Mike Kershaw <dragorn@kismetwireless.net> 2005/08/14
22  *    Add TUNSETLINK ioctl to set the link encapsulation
23  *
24  *  Mark Smith <markzzzsmith@yahoo.com.au>
25  *    Use random_ether_addr() for tap MAC address.
26  *
27  *  Harald Roelle <harald.roelle@ifi.lmu.de>  2004/04/20
28  *    Fixes in packet dropping, queue length setting and queue wakeup.
29  *    Increased default tx queue length.
30  *    Added ethtool API.
31  *    Minor cleanups
32  *
33  *  Daniel Podlejski <underley@underley.eu.org>
34  *    Modifications for 2.3.99-pre5 kernel.
35  */
36 
37 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
38 
39 #define DRV_NAME	"tun"
40 #define DRV_VERSION	"1.6"
41 #define DRV_DESCRIPTION	"Universal TUN/TAP device driver"
42 #define DRV_COPYRIGHT	"(C) 1999-2004 Max Krasnyansky <maxk@qualcomm.com>"
43 
44 #include <linux/module.h>
45 #include <linux/errno.h>
46 #include <linux/kernel.h>
47 #include <linux/major.h>
48 #include <linux/slab.h>
49 #include <linux/poll.h>
50 #include <linux/fcntl.h>
51 #include <linux/init.h>
52 #include <linux/skbuff.h>
53 #include <linux/netdevice.h>
54 #include <linux/etherdevice.h>
55 #include <linux/miscdevice.h>
56 #include <linux/ethtool.h>
57 #include <linux/rtnetlink.h>
58 #include <linux/compat.h>
59 #include <linux/if.h>
60 #include <linux/if_arp.h>
61 #include <linux/if_ether.h>
62 #include <linux/if_tun.h>
63 #include <linux/crc32.h>
64 #include <linux/nsproxy.h>
65 #include <linux/virtio_net.h>
66 #include <linux/rcupdate.h>
67 #include <net/net_namespace.h>
68 #include <net/netns/generic.h>
69 #include <net/rtnetlink.h>
70 #include <net/sock.h>
71 
72 #include <asm/system.h>
73 #include <asm/uaccess.h>
74 
75 /* Uncomment to enable debugging */
76 /* #define TUN_DEBUG 1 */
77 
78 #ifdef TUN_DEBUG
79 static int debug;
80 
81 #define tun_debug(level, tun, fmt, args...)			\
82 do {								\
83 	if (tun->debug)						\
84 		netdev_printk(level, tun->dev, fmt, ##args);	\
85 } while (0)
86 #define DBG1(level, fmt, args...)				\
87 do {								\
88 	if (debug == 2)						\
89 		printk(level fmt, ##args);			\
90 } while (0)
91 #else
92 #define tun_debug(level, tun, fmt, args...)			\
93 do {								\
94 	if (0)							\
95 		netdev_printk(level, tun->dev, fmt, ##args);	\
96 } while (0)
97 #define DBG1(level, fmt, args...)				\
98 do {								\
99 	if (0)							\
100 		printk(level fmt, ##args);			\
101 } while (0)
102 #endif
103 
104 #define FLT_EXACT_COUNT 8
105 struct tap_filter {
106 	unsigned int    count;    /* Number of addrs. Zero means disabled */
107 	u32             mask[2];  /* Mask of the hashed addrs */
108 	unsigned char	addr[FLT_EXACT_COUNT][ETH_ALEN];
109 };
110 
111 struct tun_file {
112 	atomic_t count;
113 	struct tun_struct *tun;
114 	struct net *net;
115 };
116 
117 struct tun_sock;
118 
119 struct tun_struct {
120 	struct tun_file		*tfile;
121 	unsigned int 		flags;
122 	uid_t			owner;
123 	gid_t			group;
124 
125 	struct net_device	*dev;
126 	u32			set_features;
127 #define TUN_USER_FEATURES (NETIF_F_HW_CSUM|NETIF_F_TSO_ECN|NETIF_F_TSO| \
128 			  NETIF_F_TSO6|NETIF_F_UFO)
129 	struct fasync_struct	*fasync;
130 
131 	struct tap_filter       txflt;
132 	struct socket		socket;
133 	struct socket_wq	wq;
134 
135 	int			vnet_hdr_sz;
136 
137 #ifdef TUN_DEBUG
138 	int debug;
139 #endif
140 };
141 
142 struct tun_sock {
143 	struct sock		sk;
144 	struct tun_struct	*tun;
145 };
146 
147 static inline struct tun_sock *tun_sk(struct sock *sk)
148 {
149 	return container_of(sk, struct tun_sock, sk);
150 }
151 
152 static int tun_attach(struct tun_struct *tun, struct file *file)
153 {
154 	struct tun_file *tfile = file->private_data;
155 	int err;
156 
157 	ASSERT_RTNL();
158 
159 	netif_tx_lock_bh(tun->dev);
160 
161 	err = -EINVAL;
162 	if (tfile->tun)
163 		goto out;
164 
165 	err = -EBUSY;
166 	if (tun->tfile)
167 		goto out;
168 
169 	err = 0;
170 	tfile->tun = tun;
171 	tun->tfile = tfile;
172 	tun->socket.file = file;
173 	netif_carrier_on(tun->dev);
174 	dev_hold(tun->dev);
175 	sock_hold(tun->socket.sk);
176 	atomic_inc(&tfile->count);
177 
178 out:
179 	netif_tx_unlock_bh(tun->dev);
180 	return err;
181 }
182 
183 static void __tun_detach(struct tun_struct *tun)
184 {
185 	/* Detach from net device */
186 	netif_tx_lock_bh(tun->dev);
187 	netif_carrier_off(tun->dev);
188 	tun->tfile = NULL;
189 	tun->socket.file = 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 (!compare_ether_addr(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 	sock_put(tun->socket.sk);
363 }
364 
365 /* Net device open. */
366 static int tun_net_open(struct net_device *dev)
367 {
368 	netif_start_queue(dev);
369 	return 0;
370 }
371 
372 /* Net device close. */
373 static int tun_net_close(struct net_device *dev)
374 {
375 	netif_stop_queue(dev);
376 	return 0;
377 }
378 
379 /* Net device start xmit */
380 static netdev_tx_t tun_net_xmit(struct sk_buff *skb, struct net_device *dev)
381 {
382 	struct tun_struct *tun = netdev_priv(dev);
383 
384 	tun_debug(KERN_INFO, tun, "tun_net_xmit %d\n", skb->len);
385 
386 	/* Drop packet if interface is not attached */
387 	if (!tun->tfile)
388 		goto drop;
389 
390 	/* Drop if the filter does not like it.
391 	 * This is a noop if the filter is disabled.
392 	 * Filter can be enabled only for the TAP devices. */
393 	if (!check_filter(&tun->txflt, skb))
394 		goto drop;
395 
396 	if (tun->socket.sk->sk_filter &&
397 	    sk_filter(tun->socket.sk, skb))
398 		goto drop;
399 
400 	if (skb_queue_len(&tun->socket.sk->sk_receive_queue) >= dev->tx_queue_len) {
401 		if (!(tun->flags & TUN_ONE_QUEUE)) {
402 			/* Normal queueing mode. */
403 			/* Packet scheduler handles dropping of further packets. */
404 			netif_stop_queue(dev);
405 
406 			/* We won't see all dropped packets individually, so overrun
407 			 * error is more appropriate. */
408 			dev->stats.tx_fifo_errors++;
409 		} else {
410 			/* Single queue mode.
411 			 * Driver handles dropping of all packets itself. */
412 			goto drop;
413 		}
414 	}
415 
416 	/* Orphan the skb - required as we might hang on to it
417 	 * for indefinite time. */
418 	skb_orphan(skb);
419 
420 	/* Enqueue packet */
421 	skb_queue_tail(&tun->socket.sk->sk_receive_queue, skb);
422 
423 	/* Notify and wake up reader process */
424 	if (tun->flags & TUN_FASYNC)
425 		kill_fasync(&tun->fasync, SIGIO, POLL_IN);
426 	wake_up_interruptible_poll(&tun->wq.wait, POLLIN |
427 				   POLLRDNORM | POLLRDBAND);
428 	return NETDEV_TX_OK;
429 
430 drop:
431 	dev->stats.tx_dropped++;
432 	kfree_skb(skb);
433 	return NETDEV_TX_OK;
434 }
435 
436 static void tun_net_mclist(struct net_device *dev)
437 {
438 	/*
439 	 * This callback is supposed to deal with mc filter in
440 	 * _rx_ path and has nothing to do with the _tx_ path.
441 	 * In rx path we always accept everything userspace gives us.
442 	 */
443 }
444 
445 #define MIN_MTU 68
446 #define MAX_MTU 65535
447 
448 static int
449 tun_net_change_mtu(struct net_device *dev, int new_mtu)
450 {
451 	if (new_mtu < MIN_MTU || new_mtu + dev->hard_header_len > MAX_MTU)
452 		return -EINVAL;
453 	dev->mtu = new_mtu;
454 	return 0;
455 }
456 
457 static u32 tun_net_fix_features(struct net_device *dev, u32 features)
458 {
459 	struct tun_struct *tun = netdev_priv(dev);
460 
461 	return (features & tun->set_features) | (features & ~TUN_USER_FEATURES);
462 }
463 #ifdef CONFIG_NET_POLL_CONTROLLER
464 static void tun_poll_controller(struct net_device *dev)
465 {
466 	/*
467 	 * Tun only receives frames when:
468 	 * 1) the char device endpoint gets data from user space
469 	 * 2) the tun socket gets a sendmsg call from user space
470 	 * Since both of those are syncronous operations, we are guaranteed
471 	 * never to have pending data when we poll for it
472 	 * so theres nothing to do here but return.
473 	 * We need this though so netpoll recognizes us as an interface that
474 	 * supports polling, which enables bridge devices in virt setups to
475 	 * still use netconsole
476 	 */
477 	return;
478 }
479 #endif
480 static const struct net_device_ops tun_netdev_ops = {
481 	.ndo_uninit		= tun_net_uninit,
482 	.ndo_open		= tun_net_open,
483 	.ndo_stop		= tun_net_close,
484 	.ndo_start_xmit		= tun_net_xmit,
485 	.ndo_change_mtu		= tun_net_change_mtu,
486 	.ndo_fix_features	= tun_net_fix_features,
487 #ifdef CONFIG_NET_POLL_CONTROLLER
488 	.ndo_poll_controller	= tun_poll_controller,
489 #endif
490 };
491 
492 static const struct net_device_ops tap_netdev_ops = {
493 	.ndo_uninit		= tun_net_uninit,
494 	.ndo_open		= tun_net_open,
495 	.ndo_stop		= tun_net_close,
496 	.ndo_start_xmit		= tun_net_xmit,
497 	.ndo_change_mtu		= tun_net_change_mtu,
498 	.ndo_fix_features	= tun_net_fix_features,
499 	.ndo_set_multicast_list	= tun_net_mclist,
500 	.ndo_set_mac_address	= eth_mac_addr,
501 	.ndo_validate_addr	= eth_validate_addr,
502 #ifdef CONFIG_NET_POLL_CONTROLLER
503 	.ndo_poll_controller	= tun_poll_controller,
504 #endif
505 };
506 
507 /* Initialize net device. */
508 static void tun_net_init(struct net_device *dev)
509 {
510 	struct tun_struct *tun = netdev_priv(dev);
511 
512 	switch (tun->flags & TUN_TYPE_MASK) {
513 	case TUN_TUN_DEV:
514 		dev->netdev_ops = &tun_netdev_ops;
515 
516 		/* Point-to-Point TUN Device */
517 		dev->hard_header_len = 0;
518 		dev->addr_len = 0;
519 		dev->mtu = 1500;
520 
521 		/* Zero header length */
522 		dev->type = ARPHRD_NONE;
523 		dev->flags = IFF_POINTOPOINT | IFF_NOARP | IFF_MULTICAST;
524 		dev->tx_queue_len = TUN_READQ_SIZE;  /* We prefer our own queue length */
525 		break;
526 
527 	case TUN_TAP_DEV:
528 		dev->netdev_ops = &tap_netdev_ops;
529 		/* Ethernet TAP Device */
530 		ether_setup(dev);
531 		dev->priv_flags &= ~IFF_TX_SKB_SHARING;
532 
533 		random_ether_addr(dev->dev_addr);
534 
535 		dev->tx_queue_len = TUN_READQ_SIZE;  /* We prefer our own queue length */
536 		break;
537 	}
538 }
539 
540 /* Character device part */
541 
542 /* Poll */
543 static unsigned int tun_chr_poll(struct file *file, poll_table * wait)
544 {
545 	struct tun_file *tfile = file->private_data;
546 	struct tun_struct *tun = __tun_get(tfile);
547 	struct sock *sk;
548 	unsigned int mask = 0;
549 
550 	if (!tun)
551 		return POLLERR;
552 
553 	sk = tun->socket.sk;
554 
555 	tun_debug(KERN_INFO, tun, "tun_chr_poll\n");
556 
557 	poll_wait(file, &tun->wq.wait, wait);
558 
559 	if (!skb_queue_empty(&sk->sk_receive_queue))
560 		mask |= POLLIN | POLLRDNORM;
561 
562 	if (sock_writeable(sk) ||
563 	    (!test_and_set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags) &&
564 	     sock_writeable(sk)))
565 		mask |= POLLOUT | POLLWRNORM;
566 
567 	if (tun->dev->reg_state != NETREG_REGISTERED)
568 		mask = POLLERR;
569 
570 	tun_put(tun);
571 	return mask;
572 }
573 
574 /* prepad is the amount to reserve at front.  len is length after that.
575  * linear is a hint as to how much to copy (usually headers). */
576 static struct sk_buff *tun_alloc_skb(struct tun_struct *tun,
577 				     size_t prepad, size_t len,
578 				     size_t linear, int noblock)
579 {
580 	struct sock *sk = tun->socket.sk;
581 	struct sk_buff *skb;
582 	int err;
583 
584 	sock_update_classid(sk);
585 
586 	/* Under a page?  Don't bother with paged skb. */
587 	if (prepad + len < PAGE_SIZE || !linear)
588 		linear = len;
589 
590 	skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
591 				   &err);
592 	if (!skb)
593 		return ERR_PTR(err);
594 
595 	skb_reserve(skb, prepad);
596 	skb_put(skb, linear);
597 	skb->data_len = len - linear;
598 	skb->len += len - linear;
599 
600 	return skb;
601 }
602 
603 /* Get packet from user space buffer */
604 static ssize_t tun_get_user(struct tun_struct *tun,
605 			    const struct iovec *iv, size_t count,
606 			    int noblock)
607 {
608 	struct tun_pi pi = { 0, cpu_to_be16(ETH_P_IP) };
609 	struct sk_buff *skb;
610 	size_t len = count, align = NET_SKB_PAD;
611 	struct virtio_net_hdr gso = { 0 };
612 	int offset = 0;
613 
614 	if (!(tun->flags & TUN_NO_PI)) {
615 		if ((len -= sizeof(pi)) > count)
616 			return -EINVAL;
617 
618 		if (memcpy_fromiovecend((void *)&pi, iv, 0, sizeof(pi)))
619 			return -EFAULT;
620 		offset += sizeof(pi);
621 	}
622 
623 	if (tun->flags & TUN_VNET_HDR) {
624 		if ((len -= tun->vnet_hdr_sz) > count)
625 			return -EINVAL;
626 
627 		if (memcpy_fromiovecend((void *)&gso, iv, offset, sizeof(gso)))
628 			return -EFAULT;
629 
630 		if ((gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
631 		    gso.csum_start + gso.csum_offset + 2 > gso.hdr_len)
632 			gso.hdr_len = gso.csum_start + gso.csum_offset + 2;
633 
634 		if (gso.hdr_len > len)
635 			return -EINVAL;
636 		offset += tun->vnet_hdr_sz;
637 	}
638 
639 	if ((tun->flags & TUN_TYPE_MASK) == TUN_TAP_DEV) {
640 		align += NET_IP_ALIGN;
641 		if (unlikely(len < ETH_HLEN ||
642 			     (gso.hdr_len && gso.hdr_len < ETH_HLEN)))
643 			return -EINVAL;
644 	}
645 
646 	skb = tun_alloc_skb(tun, align, len, gso.hdr_len, noblock);
647 	if (IS_ERR(skb)) {
648 		if (PTR_ERR(skb) != -EAGAIN)
649 			tun->dev->stats.rx_dropped++;
650 		return PTR_ERR(skb);
651 	}
652 
653 	if (skb_copy_datagram_from_iovec(skb, 0, iv, offset, len)) {
654 		tun->dev->stats.rx_dropped++;
655 		kfree_skb(skb);
656 		return -EFAULT;
657 	}
658 
659 	if (gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
660 		if (!skb_partial_csum_set(skb, gso.csum_start,
661 					  gso.csum_offset)) {
662 			tun->dev->stats.rx_frame_errors++;
663 			kfree_skb(skb);
664 			return -EINVAL;
665 		}
666 	}
667 
668 	switch (tun->flags & TUN_TYPE_MASK) {
669 	case TUN_TUN_DEV:
670 		if (tun->flags & TUN_NO_PI) {
671 			switch (skb->data[0] & 0xf0) {
672 			case 0x40:
673 				pi.proto = htons(ETH_P_IP);
674 				break;
675 			case 0x60:
676 				pi.proto = htons(ETH_P_IPV6);
677 				break;
678 			default:
679 				tun->dev->stats.rx_dropped++;
680 				kfree_skb(skb);
681 				return -EINVAL;
682 			}
683 		}
684 
685 		skb_reset_mac_header(skb);
686 		skb->protocol = pi.proto;
687 		skb->dev = tun->dev;
688 		break;
689 	case TUN_TAP_DEV:
690 		skb->protocol = eth_type_trans(skb, tun->dev);
691 		break;
692 	}
693 
694 	if (gso.gso_type != VIRTIO_NET_HDR_GSO_NONE) {
695 		pr_debug("GSO!\n");
696 		switch (gso.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
697 		case VIRTIO_NET_HDR_GSO_TCPV4:
698 			skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
699 			break;
700 		case VIRTIO_NET_HDR_GSO_TCPV6:
701 			skb_shinfo(skb)->gso_type = SKB_GSO_TCPV6;
702 			break;
703 		case VIRTIO_NET_HDR_GSO_UDP:
704 			skb_shinfo(skb)->gso_type = SKB_GSO_UDP;
705 			break;
706 		default:
707 			tun->dev->stats.rx_frame_errors++;
708 			kfree_skb(skb);
709 			return -EINVAL;
710 		}
711 
712 		if (gso.gso_type & VIRTIO_NET_HDR_GSO_ECN)
713 			skb_shinfo(skb)->gso_type |= SKB_GSO_TCP_ECN;
714 
715 		skb_shinfo(skb)->gso_size = gso.gso_size;
716 		if (skb_shinfo(skb)->gso_size == 0) {
717 			tun->dev->stats.rx_frame_errors++;
718 			kfree_skb(skb);
719 			return -EINVAL;
720 		}
721 
722 		/* Header must be checked, and gso_segs computed. */
723 		skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
724 		skb_shinfo(skb)->gso_segs = 0;
725 	}
726 
727 	netif_rx_ni(skb);
728 
729 	tun->dev->stats.rx_packets++;
730 	tun->dev->stats.rx_bytes += len;
731 
732 	return count;
733 }
734 
735 static ssize_t tun_chr_aio_write(struct kiocb *iocb, const struct iovec *iv,
736 			      unsigned long count, loff_t pos)
737 {
738 	struct file *file = iocb->ki_filp;
739 	struct tun_struct *tun = tun_get(file);
740 	ssize_t result;
741 
742 	if (!tun)
743 		return -EBADFD;
744 
745 	tun_debug(KERN_INFO, tun, "tun_chr_write %ld\n", count);
746 
747 	result = tun_get_user(tun, iv, iov_length(iv, count),
748 			      file->f_flags & O_NONBLOCK);
749 
750 	tun_put(tun);
751 	return result;
752 }
753 
754 /* Put packet to the user space buffer */
755 static ssize_t tun_put_user(struct tun_struct *tun,
756 			    struct sk_buff *skb,
757 			    const struct iovec *iv, int len)
758 {
759 	struct tun_pi pi = { 0, skb->protocol };
760 	ssize_t total = 0;
761 
762 	if (!(tun->flags & TUN_NO_PI)) {
763 		if ((len -= sizeof(pi)) < 0)
764 			return -EINVAL;
765 
766 		if (len < skb->len) {
767 			/* Packet will be striped */
768 			pi.flags |= TUN_PKT_STRIP;
769 		}
770 
771 		if (memcpy_toiovecend(iv, (void *) &pi, 0, sizeof(pi)))
772 			return -EFAULT;
773 		total += sizeof(pi);
774 	}
775 
776 	if (tun->flags & TUN_VNET_HDR) {
777 		struct virtio_net_hdr gso = { 0 }; /* no info leak */
778 		if ((len -= tun->vnet_hdr_sz) < 0)
779 			return -EINVAL;
780 
781 		if (skb_is_gso(skb)) {
782 			struct skb_shared_info *sinfo = skb_shinfo(skb);
783 
784 			/* This is a hint as to how much should be linear. */
785 			gso.hdr_len = skb_headlen(skb);
786 			gso.gso_size = sinfo->gso_size;
787 			if (sinfo->gso_type & SKB_GSO_TCPV4)
788 				gso.gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
789 			else if (sinfo->gso_type & SKB_GSO_TCPV6)
790 				gso.gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
791 			else if (sinfo->gso_type & SKB_GSO_UDP)
792 				gso.gso_type = VIRTIO_NET_HDR_GSO_UDP;
793 			else {
794 				pr_err("unexpected GSO type: "
795 				       "0x%x, gso_size %d, hdr_len %d\n",
796 				       sinfo->gso_type, gso.gso_size,
797 				       gso.hdr_len);
798 				print_hex_dump(KERN_ERR, "tun: ",
799 					       DUMP_PREFIX_NONE,
800 					       16, 1, skb->head,
801 					       min((int)gso.hdr_len, 64), true);
802 				WARN_ON_ONCE(1);
803 				return -EINVAL;
804 			}
805 			if (sinfo->gso_type & SKB_GSO_TCP_ECN)
806 				gso.gso_type |= VIRTIO_NET_HDR_GSO_ECN;
807 		} else
808 			gso.gso_type = VIRTIO_NET_HDR_GSO_NONE;
809 
810 		if (skb->ip_summed == CHECKSUM_PARTIAL) {
811 			gso.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
812 			gso.csum_start = skb_checksum_start_offset(skb);
813 			gso.csum_offset = skb->csum_offset;
814 		} else if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
815 			gso.flags = VIRTIO_NET_HDR_F_DATA_VALID;
816 		} /* else everything is zero */
817 
818 		if (unlikely(memcpy_toiovecend(iv, (void *)&gso, total,
819 					       sizeof(gso))))
820 			return -EFAULT;
821 		total += tun->vnet_hdr_sz;
822 	}
823 
824 	len = min_t(int, skb->len, len);
825 
826 	skb_copy_datagram_const_iovec(skb, 0, iv, total, len);
827 	total += skb->len;
828 
829 	tun->dev->stats.tx_packets++;
830 	tun->dev->stats.tx_bytes += len;
831 
832 	return total;
833 }
834 
835 static ssize_t tun_do_read(struct tun_struct *tun,
836 			   struct kiocb *iocb, const struct iovec *iv,
837 			   ssize_t len, int noblock)
838 {
839 	DECLARE_WAITQUEUE(wait, current);
840 	struct sk_buff *skb;
841 	ssize_t ret = 0;
842 
843 	tun_debug(KERN_INFO, tun, "tun_chr_read\n");
844 
845 	if (unlikely(!noblock))
846 		add_wait_queue(&tun->wq.wait, &wait);
847 	while (len) {
848 		current->state = TASK_INTERRUPTIBLE;
849 
850 		/* Read frames from the queue */
851 		if (!(skb=skb_dequeue(&tun->socket.sk->sk_receive_queue))) {
852 			if (noblock) {
853 				ret = -EAGAIN;
854 				break;
855 			}
856 			if (signal_pending(current)) {
857 				ret = -ERESTARTSYS;
858 				break;
859 			}
860 			if (tun->dev->reg_state != NETREG_REGISTERED) {
861 				ret = -EIO;
862 				break;
863 			}
864 
865 			/* Nothing to read, let's sleep */
866 			schedule();
867 			continue;
868 		}
869 		netif_wake_queue(tun->dev);
870 
871 		ret = tun_put_user(tun, skb, iv, len);
872 		kfree_skb(skb);
873 		break;
874 	}
875 
876 	current->state = TASK_RUNNING;
877 	if (unlikely(!noblock))
878 		remove_wait_queue(&tun->wq.wait, &wait);
879 
880 	return ret;
881 }
882 
883 static ssize_t tun_chr_aio_read(struct kiocb *iocb, const struct iovec *iv,
884 			    unsigned long count, loff_t pos)
885 {
886 	struct file *file = iocb->ki_filp;
887 	struct tun_file *tfile = file->private_data;
888 	struct tun_struct *tun = __tun_get(tfile);
889 	ssize_t len, ret;
890 
891 	if (!tun)
892 		return -EBADFD;
893 	len = iov_length(iv, count);
894 	if (len < 0) {
895 		ret = -EINVAL;
896 		goto out;
897 	}
898 
899 	ret = tun_do_read(tun, iocb, iv, len, file->f_flags & O_NONBLOCK);
900 	ret = min_t(ssize_t, ret, len);
901 out:
902 	tun_put(tun);
903 	return ret;
904 }
905 
906 static void tun_setup(struct net_device *dev)
907 {
908 	struct tun_struct *tun = netdev_priv(dev);
909 
910 	tun->owner = -1;
911 	tun->group = -1;
912 
913 	dev->ethtool_ops = &tun_ethtool_ops;
914 	dev->destructor = tun_free_netdev;
915 }
916 
917 /* Trivial set of netlink ops to allow deleting tun or tap
918  * device with netlink.
919  */
920 static int tun_validate(struct nlattr *tb[], struct nlattr *data[])
921 {
922 	return -EINVAL;
923 }
924 
925 static struct rtnl_link_ops tun_link_ops __read_mostly = {
926 	.kind		= DRV_NAME,
927 	.priv_size	= sizeof(struct tun_struct),
928 	.setup		= tun_setup,
929 	.validate	= tun_validate,
930 };
931 
932 static void tun_sock_write_space(struct sock *sk)
933 {
934 	struct tun_struct *tun;
935 	wait_queue_head_t *wqueue;
936 
937 	if (!sock_writeable(sk))
938 		return;
939 
940 	if (!test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags))
941 		return;
942 
943 	wqueue = sk_sleep(sk);
944 	if (wqueue && waitqueue_active(wqueue))
945 		wake_up_interruptible_sync_poll(wqueue, POLLOUT |
946 						POLLWRNORM | POLLWRBAND);
947 
948 	tun = tun_sk(sk)->tun;
949 	kill_fasync(&tun->fasync, SIGIO, POLL_OUT);
950 }
951 
952 static void tun_sock_destruct(struct sock *sk)
953 {
954 	free_netdev(tun_sk(sk)->tun->dev);
955 }
956 
957 static int tun_sendmsg(struct kiocb *iocb, struct socket *sock,
958 		       struct msghdr *m, size_t total_len)
959 {
960 	struct tun_struct *tun = container_of(sock, struct tun_struct, socket);
961 	return tun_get_user(tun, m->msg_iov, total_len,
962 			    m->msg_flags & MSG_DONTWAIT);
963 }
964 
965 static int tun_recvmsg(struct kiocb *iocb, struct socket *sock,
966 		       struct msghdr *m, size_t total_len,
967 		       int flags)
968 {
969 	struct tun_struct *tun = container_of(sock, struct tun_struct, socket);
970 	int ret;
971 	if (flags & ~(MSG_DONTWAIT|MSG_TRUNC))
972 		return -EINVAL;
973 	ret = tun_do_read(tun, iocb, m->msg_iov, total_len,
974 			  flags & MSG_DONTWAIT);
975 	if (ret > total_len) {
976 		m->msg_flags |= MSG_TRUNC;
977 		ret = flags & MSG_TRUNC ? ret : total_len;
978 	}
979 	return ret;
980 }
981 
982 /* Ops structure to mimic raw sockets with tun */
983 static const struct proto_ops tun_socket_ops = {
984 	.sendmsg = tun_sendmsg,
985 	.recvmsg = tun_recvmsg,
986 };
987 
988 static struct proto tun_proto = {
989 	.name		= "tun",
990 	.owner		= THIS_MODULE,
991 	.obj_size	= sizeof(struct tun_sock),
992 };
993 
994 static int tun_flags(struct tun_struct *tun)
995 {
996 	int flags = 0;
997 
998 	if (tun->flags & TUN_TUN_DEV)
999 		flags |= IFF_TUN;
1000 	else
1001 		flags |= IFF_TAP;
1002 
1003 	if (tun->flags & TUN_NO_PI)
1004 		flags |= IFF_NO_PI;
1005 
1006 	if (tun->flags & TUN_ONE_QUEUE)
1007 		flags |= IFF_ONE_QUEUE;
1008 
1009 	if (tun->flags & TUN_VNET_HDR)
1010 		flags |= IFF_VNET_HDR;
1011 
1012 	return flags;
1013 }
1014 
1015 static ssize_t tun_show_flags(struct device *dev, struct device_attribute *attr,
1016 			      char *buf)
1017 {
1018 	struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1019 	return sprintf(buf, "0x%x\n", tun_flags(tun));
1020 }
1021 
1022 static ssize_t tun_show_owner(struct device *dev, struct device_attribute *attr,
1023 			      char *buf)
1024 {
1025 	struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1026 	return sprintf(buf, "%d\n", tun->owner);
1027 }
1028 
1029 static ssize_t tun_show_group(struct device *dev, struct device_attribute *attr,
1030 			      char *buf)
1031 {
1032 	struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1033 	return sprintf(buf, "%d\n", tun->group);
1034 }
1035 
1036 static DEVICE_ATTR(tun_flags, 0444, tun_show_flags, NULL);
1037 static DEVICE_ATTR(owner, 0444, tun_show_owner, NULL);
1038 static DEVICE_ATTR(group, 0444, tun_show_group, NULL);
1039 
1040 static int tun_set_iff(struct net *net, struct file *file, struct ifreq *ifr)
1041 {
1042 	struct sock *sk;
1043 	struct tun_struct *tun;
1044 	struct net_device *dev;
1045 	int err;
1046 
1047 	dev = __dev_get_by_name(net, ifr->ifr_name);
1048 	if (dev) {
1049 		const struct cred *cred = current_cred();
1050 
1051 		if (ifr->ifr_flags & IFF_TUN_EXCL)
1052 			return -EBUSY;
1053 		if ((ifr->ifr_flags & IFF_TUN) && dev->netdev_ops == &tun_netdev_ops)
1054 			tun = netdev_priv(dev);
1055 		else if ((ifr->ifr_flags & IFF_TAP) && dev->netdev_ops == &tap_netdev_ops)
1056 			tun = netdev_priv(dev);
1057 		else
1058 			return -EINVAL;
1059 
1060 		if (((tun->owner != -1 && cred->euid != tun->owner) ||
1061 		     (tun->group != -1 && !in_egroup_p(tun->group))) &&
1062 		    !capable(CAP_NET_ADMIN))
1063 			return -EPERM;
1064 		err = security_tun_dev_attach(tun->socket.sk);
1065 		if (err < 0)
1066 			return err;
1067 
1068 		err = tun_attach(tun, file);
1069 		if (err < 0)
1070 			return err;
1071 	}
1072 	else {
1073 		char *name;
1074 		unsigned long flags = 0;
1075 
1076 		if (!capable(CAP_NET_ADMIN))
1077 			return -EPERM;
1078 		err = security_tun_dev_create();
1079 		if (err < 0)
1080 			return err;
1081 
1082 		/* Set dev type */
1083 		if (ifr->ifr_flags & IFF_TUN) {
1084 			/* TUN device */
1085 			flags |= TUN_TUN_DEV;
1086 			name = "tun%d";
1087 		} else if (ifr->ifr_flags & IFF_TAP) {
1088 			/* TAP device */
1089 			flags |= TUN_TAP_DEV;
1090 			name = "tap%d";
1091 		} else
1092 			return -EINVAL;
1093 
1094 		if (*ifr->ifr_name)
1095 			name = ifr->ifr_name;
1096 
1097 		dev = alloc_netdev(sizeof(struct tun_struct), name,
1098 				   tun_setup);
1099 		if (!dev)
1100 			return -ENOMEM;
1101 
1102 		dev_net_set(dev, net);
1103 		dev->rtnl_link_ops = &tun_link_ops;
1104 
1105 		tun = netdev_priv(dev);
1106 		tun->dev = dev;
1107 		tun->flags = flags;
1108 		tun->txflt.count = 0;
1109 		tun->vnet_hdr_sz = sizeof(struct virtio_net_hdr);
1110 
1111 		err = -ENOMEM;
1112 		sk = sk_alloc(net, AF_UNSPEC, GFP_KERNEL, &tun_proto);
1113 		if (!sk)
1114 			goto err_free_dev;
1115 
1116 		tun->socket.wq = &tun->wq;
1117 		init_waitqueue_head(&tun->wq.wait);
1118 		tun->socket.ops = &tun_socket_ops;
1119 		sock_init_data(&tun->socket, sk);
1120 		sk->sk_write_space = tun_sock_write_space;
1121 		sk->sk_sndbuf = INT_MAX;
1122 
1123 		tun_sk(sk)->tun = tun;
1124 
1125 		security_tun_dev_post_create(sk);
1126 
1127 		tun_net_init(dev);
1128 
1129 		dev->hw_features = NETIF_F_SG | NETIF_F_FRAGLIST |
1130 			TUN_USER_FEATURES;
1131 		dev->features = dev->hw_features;
1132 
1133 		err = register_netdevice(tun->dev);
1134 		if (err < 0)
1135 			goto err_free_sk;
1136 
1137 		if (device_create_file(&tun->dev->dev, &dev_attr_tun_flags) ||
1138 		    device_create_file(&tun->dev->dev, &dev_attr_owner) ||
1139 		    device_create_file(&tun->dev->dev, &dev_attr_group))
1140 			pr_err("Failed to create tun sysfs files\n");
1141 
1142 		sk->sk_destruct = tun_sock_destruct;
1143 
1144 		err = tun_attach(tun, file);
1145 		if (err < 0)
1146 			goto failed;
1147 	}
1148 
1149 	tun_debug(KERN_INFO, tun, "tun_set_iff\n");
1150 
1151 	if (ifr->ifr_flags & IFF_NO_PI)
1152 		tun->flags |= TUN_NO_PI;
1153 	else
1154 		tun->flags &= ~TUN_NO_PI;
1155 
1156 	if (ifr->ifr_flags & IFF_ONE_QUEUE)
1157 		tun->flags |= TUN_ONE_QUEUE;
1158 	else
1159 		tun->flags &= ~TUN_ONE_QUEUE;
1160 
1161 	if (ifr->ifr_flags & IFF_VNET_HDR)
1162 		tun->flags |= TUN_VNET_HDR;
1163 	else
1164 		tun->flags &= ~TUN_VNET_HDR;
1165 
1166 	/* Make sure persistent devices do not get stuck in
1167 	 * xoff state.
1168 	 */
1169 	if (netif_running(tun->dev))
1170 		netif_wake_queue(tun->dev);
1171 
1172 	strcpy(ifr->ifr_name, tun->dev->name);
1173 	return 0;
1174 
1175  err_free_sk:
1176 	sock_put(sk);
1177  err_free_dev:
1178 	free_netdev(dev);
1179  failed:
1180 	return err;
1181 }
1182 
1183 static int tun_get_iff(struct net *net, struct tun_struct *tun,
1184 		       struct ifreq *ifr)
1185 {
1186 	tun_debug(KERN_INFO, tun, "tun_get_iff\n");
1187 
1188 	strcpy(ifr->ifr_name, tun->dev->name);
1189 
1190 	ifr->ifr_flags = tun_flags(tun);
1191 
1192 	return 0;
1193 }
1194 
1195 /* This is like a cut-down ethtool ops, except done via tun fd so no
1196  * privs required. */
1197 static int set_offload(struct tun_struct *tun, unsigned long arg)
1198 {
1199 	u32 features = 0;
1200 
1201 	if (arg & TUN_F_CSUM) {
1202 		features |= NETIF_F_HW_CSUM;
1203 		arg &= ~TUN_F_CSUM;
1204 
1205 		if (arg & (TUN_F_TSO4|TUN_F_TSO6)) {
1206 			if (arg & TUN_F_TSO_ECN) {
1207 				features |= NETIF_F_TSO_ECN;
1208 				arg &= ~TUN_F_TSO_ECN;
1209 			}
1210 			if (arg & TUN_F_TSO4)
1211 				features |= NETIF_F_TSO;
1212 			if (arg & TUN_F_TSO6)
1213 				features |= NETIF_F_TSO6;
1214 			arg &= ~(TUN_F_TSO4|TUN_F_TSO6);
1215 		}
1216 
1217 		if (arg & TUN_F_UFO) {
1218 			features |= NETIF_F_UFO;
1219 			arg &= ~TUN_F_UFO;
1220 		}
1221 	}
1222 
1223 	/* This gives the user a way to test for new features in future by
1224 	 * trying to set them. */
1225 	if (arg)
1226 		return -EINVAL;
1227 
1228 	tun->set_features = features;
1229 	netdev_update_features(tun->dev);
1230 
1231 	return 0;
1232 }
1233 
1234 static long __tun_chr_ioctl(struct file *file, unsigned int cmd,
1235 			    unsigned long arg, int ifreq_len)
1236 {
1237 	struct tun_file *tfile = file->private_data;
1238 	struct tun_struct *tun;
1239 	void __user* argp = (void __user*)arg;
1240 	struct sock_fprog fprog;
1241 	struct ifreq ifr;
1242 	int sndbuf;
1243 	int vnet_hdr_sz;
1244 	int ret;
1245 
1246 	if (cmd == TUNSETIFF || _IOC_TYPE(cmd) == 0x89)
1247 		if (copy_from_user(&ifr, argp, ifreq_len))
1248 			return -EFAULT;
1249 
1250 	if (cmd == TUNGETFEATURES) {
1251 		/* Currently this just means: "what IFF flags are valid?".
1252 		 * This is needed because we never checked for invalid flags on
1253 		 * TUNSETIFF. */
1254 		return put_user(IFF_TUN | IFF_TAP | IFF_NO_PI | IFF_ONE_QUEUE |
1255 				IFF_VNET_HDR,
1256 				(unsigned int __user*)argp);
1257 	}
1258 
1259 	rtnl_lock();
1260 
1261 	tun = __tun_get(tfile);
1262 	if (cmd == TUNSETIFF && !tun) {
1263 		ifr.ifr_name[IFNAMSIZ-1] = '\0';
1264 
1265 		ret = tun_set_iff(tfile->net, file, &ifr);
1266 
1267 		if (ret)
1268 			goto unlock;
1269 
1270 		if (copy_to_user(argp, &ifr, ifreq_len))
1271 			ret = -EFAULT;
1272 		goto unlock;
1273 	}
1274 
1275 	ret = -EBADFD;
1276 	if (!tun)
1277 		goto unlock;
1278 
1279 	tun_debug(KERN_INFO, tun, "tun_chr_ioctl cmd %d\n", cmd);
1280 
1281 	ret = 0;
1282 	switch (cmd) {
1283 	case TUNGETIFF:
1284 		ret = tun_get_iff(current->nsproxy->net_ns, tun, &ifr);
1285 		if (ret)
1286 			break;
1287 
1288 		if (copy_to_user(argp, &ifr, ifreq_len))
1289 			ret = -EFAULT;
1290 		break;
1291 
1292 	case TUNSETNOCSUM:
1293 		/* Disable/Enable checksum */
1294 
1295 		/* [unimplemented] */
1296 		tun_debug(KERN_INFO, tun, "ignored: set checksum %s\n",
1297 			  arg ? "disabled" : "enabled");
1298 		break;
1299 
1300 	case TUNSETPERSIST:
1301 		/* Disable/Enable persist mode */
1302 		if (arg)
1303 			tun->flags |= TUN_PERSIST;
1304 		else
1305 			tun->flags &= ~TUN_PERSIST;
1306 
1307 		tun_debug(KERN_INFO, tun, "persist %s\n",
1308 			  arg ? "enabled" : "disabled");
1309 		break;
1310 
1311 	case TUNSETOWNER:
1312 		/* Set owner of the device */
1313 		tun->owner = (uid_t) arg;
1314 
1315 		tun_debug(KERN_INFO, tun, "owner set to %d\n", tun->owner);
1316 		break;
1317 
1318 	case TUNSETGROUP:
1319 		/* Set group of the device */
1320 		tun->group= (gid_t) arg;
1321 
1322 		tun_debug(KERN_INFO, tun, "group set to %d\n", tun->group);
1323 		break;
1324 
1325 	case TUNSETLINK:
1326 		/* Only allow setting the type when the interface is down */
1327 		if (tun->dev->flags & IFF_UP) {
1328 			tun_debug(KERN_INFO, tun,
1329 				  "Linktype set failed because interface is up\n");
1330 			ret = -EBUSY;
1331 		} else {
1332 			tun->dev->type = (int) arg;
1333 			tun_debug(KERN_INFO, tun, "linktype set to %d\n",
1334 				  tun->dev->type);
1335 			ret = 0;
1336 		}
1337 		break;
1338 
1339 #ifdef TUN_DEBUG
1340 	case TUNSETDEBUG:
1341 		tun->debug = arg;
1342 		break;
1343 #endif
1344 	case TUNSETOFFLOAD:
1345 		ret = set_offload(tun, arg);
1346 		break;
1347 
1348 	case TUNSETTXFILTER:
1349 		/* Can be set only for TAPs */
1350 		ret = -EINVAL;
1351 		if ((tun->flags & TUN_TYPE_MASK) != TUN_TAP_DEV)
1352 			break;
1353 		ret = update_filter(&tun->txflt, (void __user *)arg);
1354 		break;
1355 
1356 	case SIOCGIFHWADDR:
1357 		/* Get hw address */
1358 		memcpy(ifr.ifr_hwaddr.sa_data, tun->dev->dev_addr, ETH_ALEN);
1359 		ifr.ifr_hwaddr.sa_family = tun->dev->type;
1360 		if (copy_to_user(argp, &ifr, ifreq_len))
1361 			ret = -EFAULT;
1362 		break;
1363 
1364 	case SIOCSIFHWADDR:
1365 		/* Set hw address */
1366 		tun_debug(KERN_DEBUG, tun, "set hw address: %pM\n",
1367 			  ifr.ifr_hwaddr.sa_data);
1368 
1369 		ret = dev_set_mac_address(tun->dev, &ifr.ifr_hwaddr);
1370 		break;
1371 
1372 	case TUNGETSNDBUF:
1373 		sndbuf = tun->socket.sk->sk_sndbuf;
1374 		if (copy_to_user(argp, &sndbuf, sizeof(sndbuf)))
1375 			ret = -EFAULT;
1376 		break;
1377 
1378 	case TUNSETSNDBUF:
1379 		if (copy_from_user(&sndbuf, argp, sizeof(sndbuf))) {
1380 			ret = -EFAULT;
1381 			break;
1382 		}
1383 
1384 		tun->socket.sk->sk_sndbuf = sndbuf;
1385 		break;
1386 
1387 	case TUNGETVNETHDRSZ:
1388 		vnet_hdr_sz = tun->vnet_hdr_sz;
1389 		if (copy_to_user(argp, &vnet_hdr_sz, sizeof(vnet_hdr_sz)))
1390 			ret = -EFAULT;
1391 		break;
1392 
1393 	case TUNSETVNETHDRSZ:
1394 		if (copy_from_user(&vnet_hdr_sz, argp, sizeof(vnet_hdr_sz))) {
1395 			ret = -EFAULT;
1396 			break;
1397 		}
1398 		if (vnet_hdr_sz < (int)sizeof(struct virtio_net_hdr)) {
1399 			ret = -EINVAL;
1400 			break;
1401 		}
1402 
1403 		tun->vnet_hdr_sz = vnet_hdr_sz;
1404 		break;
1405 
1406 	case TUNATTACHFILTER:
1407 		/* Can be set only for TAPs */
1408 		ret = -EINVAL;
1409 		if ((tun->flags & TUN_TYPE_MASK) != TUN_TAP_DEV)
1410 			break;
1411 		ret = -EFAULT;
1412 		if (copy_from_user(&fprog, argp, sizeof(fprog)))
1413 			break;
1414 
1415 		ret = sk_attach_filter(&fprog, tun->socket.sk);
1416 		break;
1417 
1418 	case TUNDETACHFILTER:
1419 		/* Can be set only for TAPs */
1420 		ret = -EINVAL;
1421 		if ((tun->flags & TUN_TYPE_MASK) != TUN_TAP_DEV)
1422 			break;
1423 		ret = sk_detach_filter(tun->socket.sk);
1424 		break;
1425 
1426 	default:
1427 		ret = -EINVAL;
1428 		break;
1429 	}
1430 
1431 unlock:
1432 	rtnl_unlock();
1433 	if (tun)
1434 		tun_put(tun);
1435 	return ret;
1436 }
1437 
1438 static long tun_chr_ioctl(struct file *file,
1439 			  unsigned int cmd, unsigned long arg)
1440 {
1441 	return __tun_chr_ioctl(file, cmd, arg, sizeof (struct ifreq));
1442 }
1443 
1444 #ifdef CONFIG_COMPAT
1445 static long tun_chr_compat_ioctl(struct file *file,
1446 			 unsigned int cmd, unsigned long arg)
1447 {
1448 	switch (cmd) {
1449 	case TUNSETIFF:
1450 	case TUNGETIFF:
1451 	case TUNSETTXFILTER:
1452 	case TUNGETSNDBUF:
1453 	case TUNSETSNDBUF:
1454 	case SIOCGIFHWADDR:
1455 	case SIOCSIFHWADDR:
1456 		arg = (unsigned long)compat_ptr(arg);
1457 		break;
1458 	default:
1459 		arg = (compat_ulong_t)arg;
1460 		break;
1461 	}
1462 
1463 	/*
1464 	 * compat_ifreq is shorter than ifreq, so we must not access beyond
1465 	 * the end of that structure. All fields that are used in this
1466 	 * driver are compatible though, we don't need to convert the
1467 	 * contents.
1468 	 */
1469 	return __tun_chr_ioctl(file, cmd, arg, sizeof(struct compat_ifreq));
1470 }
1471 #endif /* CONFIG_COMPAT */
1472 
1473 static int tun_chr_fasync(int fd, struct file *file, int on)
1474 {
1475 	struct tun_struct *tun = tun_get(file);
1476 	int ret;
1477 
1478 	if (!tun)
1479 		return -EBADFD;
1480 
1481 	tun_debug(KERN_INFO, tun, "tun_chr_fasync %d\n", on);
1482 
1483 	if ((ret = fasync_helper(fd, file, on, &tun->fasync)) < 0)
1484 		goto out;
1485 
1486 	if (on) {
1487 		ret = __f_setown(file, task_pid(current), PIDTYPE_PID, 0);
1488 		if (ret)
1489 			goto out;
1490 		tun->flags |= TUN_FASYNC;
1491 	} else
1492 		tun->flags &= ~TUN_FASYNC;
1493 	ret = 0;
1494 out:
1495 	tun_put(tun);
1496 	return ret;
1497 }
1498 
1499 static int tun_chr_open(struct inode *inode, struct file * file)
1500 {
1501 	struct tun_file *tfile;
1502 
1503 	DBG1(KERN_INFO, "tunX: tun_chr_open\n");
1504 
1505 	tfile = kmalloc(sizeof(*tfile), GFP_KERNEL);
1506 	if (!tfile)
1507 		return -ENOMEM;
1508 	atomic_set(&tfile->count, 0);
1509 	tfile->tun = NULL;
1510 	tfile->net = get_net(current->nsproxy->net_ns);
1511 	file->private_data = tfile;
1512 	return 0;
1513 }
1514 
1515 static int tun_chr_close(struct inode *inode, struct file *file)
1516 {
1517 	struct tun_file *tfile = file->private_data;
1518 	struct tun_struct *tun;
1519 
1520 	tun = __tun_get(tfile);
1521 	if (tun) {
1522 		struct net_device *dev = tun->dev;
1523 
1524 		tun_debug(KERN_INFO, tun, "tun_chr_close\n");
1525 
1526 		__tun_detach(tun);
1527 
1528 		/* If desirable, unregister the netdevice. */
1529 		if (!(tun->flags & TUN_PERSIST)) {
1530 			rtnl_lock();
1531 			if (dev->reg_state == NETREG_REGISTERED)
1532 				unregister_netdevice(dev);
1533 			rtnl_unlock();
1534 		}
1535 	}
1536 
1537 	tun = tfile->tun;
1538 	if (tun)
1539 		sock_put(tun->socket.sk);
1540 
1541 	put_net(tfile->net);
1542 	kfree(tfile);
1543 
1544 	return 0;
1545 }
1546 
1547 static const struct file_operations tun_fops = {
1548 	.owner	= THIS_MODULE,
1549 	.llseek = no_llseek,
1550 	.read  = do_sync_read,
1551 	.aio_read  = tun_chr_aio_read,
1552 	.write = do_sync_write,
1553 	.aio_write = tun_chr_aio_write,
1554 	.poll	= tun_chr_poll,
1555 	.unlocked_ioctl	= tun_chr_ioctl,
1556 #ifdef CONFIG_COMPAT
1557 	.compat_ioctl = tun_chr_compat_ioctl,
1558 #endif
1559 	.open	= tun_chr_open,
1560 	.release = tun_chr_close,
1561 	.fasync = tun_chr_fasync
1562 };
1563 
1564 static struct miscdevice tun_miscdev = {
1565 	.minor = TUN_MINOR,
1566 	.name = "tun",
1567 	.nodename = "net/tun",
1568 	.fops = &tun_fops,
1569 };
1570 
1571 /* ethtool interface */
1572 
1573 static int tun_get_settings(struct net_device *dev, struct ethtool_cmd *cmd)
1574 {
1575 	cmd->supported		= 0;
1576 	cmd->advertising	= 0;
1577 	ethtool_cmd_speed_set(cmd, SPEED_10);
1578 	cmd->duplex		= DUPLEX_FULL;
1579 	cmd->port		= PORT_TP;
1580 	cmd->phy_address	= 0;
1581 	cmd->transceiver	= XCVR_INTERNAL;
1582 	cmd->autoneg		= AUTONEG_DISABLE;
1583 	cmd->maxtxpkt		= 0;
1584 	cmd->maxrxpkt		= 0;
1585 	return 0;
1586 }
1587 
1588 static void tun_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
1589 {
1590 	struct tun_struct *tun = netdev_priv(dev);
1591 
1592 	strcpy(info->driver, DRV_NAME);
1593 	strcpy(info->version, DRV_VERSION);
1594 	strcpy(info->fw_version, "N/A");
1595 
1596 	switch (tun->flags & TUN_TYPE_MASK) {
1597 	case TUN_TUN_DEV:
1598 		strcpy(info->bus_info, "tun");
1599 		break;
1600 	case TUN_TAP_DEV:
1601 		strcpy(info->bus_info, "tap");
1602 		break;
1603 	}
1604 }
1605 
1606 static u32 tun_get_msglevel(struct net_device *dev)
1607 {
1608 #ifdef TUN_DEBUG
1609 	struct tun_struct *tun = netdev_priv(dev);
1610 	return tun->debug;
1611 #else
1612 	return -EOPNOTSUPP;
1613 #endif
1614 }
1615 
1616 static void tun_set_msglevel(struct net_device *dev, u32 value)
1617 {
1618 #ifdef TUN_DEBUG
1619 	struct tun_struct *tun = netdev_priv(dev);
1620 	tun->debug = value;
1621 #endif
1622 }
1623 
1624 static const struct ethtool_ops tun_ethtool_ops = {
1625 	.get_settings	= tun_get_settings,
1626 	.get_drvinfo	= tun_get_drvinfo,
1627 	.get_msglevel	= tun_get_msglevel,
1628 	.set_msglevel	= tun_set_msglevel,
1629 	.get_link	= ethtool_op_get_link,
1630 };
1631 
1632 
1633 static int __init tun_init(void)
1634 {
1635 	int ret = 0;
1636 
1637 	pr_info("%s, %s\n", DRV_DESCRIPTION, DRV_VERSION);
1638 	pr_info("%s\n", DRV_COPYRIGHT);
1639 
1640 	ret = rtnl_link_register(&tun_link_ops);
1641 	if (ret) {
1642 		pr_err("Can't register link_ops\n");
1643 		goto err_linkops;
1644 	}
1645 
1646 	ret = misc_register(&tun_miscdev);
1647 	if (ret) {
1648 		pr_err("Can't register misc device %d\n", TUN_MINOR);
1649 		goto err_misc;
1650 	}
1651 	return  0;
1652 err_misc:
1653 	rtnl_link_unregister(&tun_link_ops);
1654 err_linkops:
1655 	return ret;
1656 }
1657 
1658 static void tun_cleanup(void)
1659 {
1660 	misc_deregister(&tun_miscdev);
1661 	rtnl_link_unregister(&tun_link_ops);
1662 }
1663 
1664 /* Get an underlying socket object from tun file.  Returns error unless file is
1665  * attached to a device.  The returned object works like a packet socket, it
1666  * can be used for sock_sendmsg/sock_recvmsg.  The caller is responsible for
1667  * holding a reference to the file for as long as the socket is in use. */
1668 struct socket *tun_get_socket(struct file *file)
1669 {
1670 	struct tun_struct *tun;
1671 	if (file->f_op != &tun_fops)
1672 		return ERR_PTR(-EINVAL);
1673 	tun = tun_get(file);
1674 	if (!tun)
1675 		return ERR_PTR(-EBADFD);
1676 	tun_put(tun);
1677 	return &tun->socket;
1678 }
1679 EXPORT_SYMBOL_GPL(tun_get_socket);
1680 
1681 module_init(tun_init);
1682 module_exit(tun_cleanup);
1683 MODULE_DESCRIPTION(DRV_DESCRIPTION);
1684 MODULE_AUTHOR(DRV_COPYRIGHT);
1685 MODULE_LICENSE("GPL");
1686 MODULE_ALIAS_MISCDEV(TUN_MINOR);
1687 MODULE_ALIAS("devname:net/tun");
1688