xref: /openbmc/linux/drivers/net/tun.c (revision 545e4006)
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 DRV_NAME	"tun"
38 #define DRV_VERSION	"1.6"
39 #define DRV_DESCRIPTION	"Universal TUN/TAP device driver"
40 #define DRV_COPYRIGHT	"(C) 1999-2004 Max Krasnyansky <maxk@qualcomm.com>"
41 
42 #include <linux/module.h>
43 #include <linux/errno.h>
44 #include <linux/kernel.h>
45 #include <linux/major.h>
46 #include <linux/slab.h>
47 #include <linux/smp_lock.h>
48 #include <linux/poll.h>
49 #include <linux/fcntl.h>
50 #include <linux/init.h>
51 #include <linux/skbuff.h>
52 #include <linux/netdevice.h>
53 #include <linux/etherdevice.h>
54 #include <linux/miscdevice.h>
55 #include <linux/ethtool.h>
56 #include <linux/rtnetlink.h>
57 #include <linux/if.h>
58 #include <linux/if_arp.h>
59 #include <linux/if_ether.h>
60 #include <linux/if_tun.h>
61 #include <linux/crc32.h>
62 #include <linux/nsproxy.h>
63 #include <linux/virtio_net.h>
64 #include <net/net_namespace.h>
65 #include <net/netns/generic.h>
66 
67 #include <asm/system.h>
68 #include <asm/uaccess.h>
69 
70 /* Uncomment to enable debugging */
71 /* #define TUN_DEBUG 1 */
72 
73 #ifdef TUN_DEBUG
74 static int debug;
75 
76 #define DBG  if(tun->debug)printk
77 #define DBG1 if(debug==2)printk
78 #else
79 #define DBG( a... )
80 #define DBG1( a... )
81 #endif
82 
83 #define FLT_EXACT_COUNT 8
84 struct tap_filter {
85 	unsigned int    count;    /* Number of addrs. Zero means disabled */
86 	u32             mask[2];  /* Mask of the hashed addrs */
87 	unsigned char	addr[FLT_EXACT_COUNT][ETH_ALEN];
88 };
89 
90 struct tun_struct {
91 	struct list_head        list;
92 	unsigned int 		flags;
93 	int			attached;
94 	uid_t			owner;
95 	gid_t			group;
96 
97 	wait_queue_head_t	read_wait;
98 	struct sk_buff_head	readq;
99 
100 	struct net_device	*dev;
101 	struct fasync_struct	*fasync;
102 
103 	struct tap_filter       txflt;
104 
105 #ifdef TUN_DEBUG
106 	int debug;
107 #endif
108 };
109 
110 /* TAP filterting */
111 static void addr_hash_set(u32 *mask, const u8 *addr)
112 {
113 	int n = ether_crc(ETH_ALEN, addr) >> 26;
114 	mask[n >> 5] |= (1 << (n & 31));
115 }
116 
117 static unsigned int addr_hash_test(const u32 *mask, const u8 *addr)
118 {
119 	int n = ether_crc(ETH_ALEN, addr) >> 26;
120 	return mask[n >> 5] & (1 << (n & 31));
121 }
122 
123 static int update_filter(struct tap_filter *filter, void __user *arg)
124 {
125 	struct { u8 u[ETH_ALEN]; } *addr;
126 	struct tun_filter uf;
127 	int err, alen, n, nexact;
128 
129 	if (copy_from_user(&uf, arg, sizeof(uf)))
130 		return -EFAULT;
131 
132 	if (!uf.count) {
133 		/* Disabled */
134 		filter->count = 0;
135 		return 0;
136 	}
137 
138 	alen = ETH_ALEN * uf.count;
139 	addr = kmalloc(alen, GFP_KERNEL);
140 	if (!addr)
141 		return -ENOMEM;
142 
143 	if (copy_from_user(addr, arg + sizeof(uf), alen)) {
144 		err = -EFAULT;
145 		goto done;
146 	}
147 
148 	/* The filter is updated without holding any locks. Which is
149 	 * perfectly safe. We disable it first and in the worst
150 	 * case we'll accept a few undesired packets. */
151 	filter->count = 0;
152 	wmb();
153 
154 	/* Use first set of addresses as an exact filter */
155 	for (n = 0; n < uf.count && n < FLT_EXACT_COUNT; n++)
156 		memcpy(filter->addr[n], addr[n].u, ETH_ALEN);
157 
158 	nexact = n;
159 
160 	/* The rest is hashed */
161 	memset(filter->mask, 0, sizeof(filter->mask));
162 	for (; n < uf.count; n++)
163 		addr_hash_set(filter->mask, addr[n].u);
164 
165 	/* For ALLMULTI just set the mask to all ones.
166 	 * This overrides the mask populated above. */
167 	if ((uf.flags & TUN_FLT_ALLMULTI))
168 		memset(filter->mask, ~0, sizeof(filter->mask));
169 
170 	/* Now enable the filter */
171 	wmb();
172 	filter->count = nexact;
173 
174 	/* Return the number of exact filters */
175 	err = nexact;
176 
177 done:
178 	kfree(addr);
179 	return err;
180 }
181 
182 /* Returns: 0 - drop, !=0 - accept */
183 static int run_filter(struct tap_filter *filter, const struct sk_buff *skb)
184 {
185 	/* Cannot use eth_hdr(skb) here because skb_mac_hdr() is incorrect
186 	 * at this point. */
187 	struct ethhdr *eh = (struct ethhdr *) skb->data;
188 	int i;
189 
190 	/* Exact match */
191 	for (i = 0; i < filter->count; i++)
192 		if (!compare_ether_addr(eh->h_dest, filter->addr[i]))
193 			return 1;
194 
195 	/* Inexact match (multicast only) */
196 	if (is_multicast_ether_addr(eh->h_dest))
197 		return addr_hash_test(filter->mask, eh->h_dest);
198 
199 	return 0;
200 }
201 
202 /*
203  * Checks whether the packet is accepted or not.
204  * Returns: 0 - drop, !=0 - accept
205  */
206 static int check_filter(struct tap_filter *filter, const struct sk_buff *skb)
207 {
208 	if (!filter->count)
209 		return 1;
210 
211 	return run_filter(filter, skb);
212 }
213 
214 /* Network device part of the driver */
215 
216 static unsigned int tun_net_id;
217 struct tun_net {
218 	struct list_head dev_list;
219 };
220 
221 static const struct ethtool_ops tun_ethtool_ops;
222 
223 /* Net device open. */
224 static int tun_net_open(struct net_device *dev)
225 {
226 	netif_start_queue(dev);
227 	return 0;
228 }
229 
230 /* Net device close. */
231 static int tun_net_close(struct net_device *dev)
232 {
233 	netif_stop_queue(dev);
234 	return 0;
235 }
236 
237 /* Net device start xmit */
238 static int tun_net_xmit(struct sk_buff *skb, struct net_device *dev)
239 {
240 	struct tun_struct *tun = netdev_priv(dev);
241 
242 	DBG(KERN_INFO "%s: tun_net_xmit %d\n", tun->dev->name, skb->len);
243 
244 	/* Drop packet if interface is not attached */
245 	if (!tun->attached)
246 		goto drop;
247 
248 	/* Drop if the filter does not like it.
249 	 * This is a noop if the filter is disabled.
250 	 * Filter can be enabled only for the TAP devices. */
251 	if (!check_filter(&tun->txflt, skb))
252 		goto drop;
253 
254 	if (skb_queue_len(&tun->readq) >= dev->tx_queue_len) {
255 		if (!(tun->flags & TUN_ONE_QUEUE)) {
256 			/* Normal queueing mode. */
257 			/* Packet scheduler handles dropping of further packets. */
258 			netif_stop_queue(dev);
259 
260 			/* We won't see all dropped packets individually, so overrun
261 			 * error is more appropriate. */
262 			dev->stats.tx_fifo_errors++;
263 		} else {
264 			/* Single queue mode.
265 			 * Driver handles dropping of all packets itself. */
266 			goto drop;
267 		}
268 	}
269 
270 	/* Enqueue packet */
271 	skb_queue_tail(&tun->readq, skb);
272 	dev->trans_start = jiffies;
273 
274 	/* Notify and wake up reader process */
275 	if (tun->flags & TUN_FASYNC)
276 		kill_fasync(&tun->fasync, SIGIO, POLL_IN);
277 	wake_up_interruptible(&tun->read_wait);
278 	return 0;
279 
280 drop:
281 	dev->stats.tx_dropped++;
282 	kfree_skb(skb);
283 	return 0;
284 }
285 
286 static void tun_net_mclist(struct net_device *dev)
287 {
288 	/*
289 	 * This callback is supposed to deal with mc filter in
290 	 * _rx_ path and has nothing to do with the _tx_ path.
291 	 * In rx path we always accept everything userspace gives us.
292 	 */
293 	return;
294 }
295 
296 #define MIN_MTU 68
297 #define MAX_MTU 65535
298 
299 static int
300 tun_net_change_mtu(struct net_device *dev, int new_mtu)
301 {
302 	if (new_mtu < MIN_MTU || new_mtu + dev->hard_header_len > MAX_MTU)
303 		return -EINVAL;
304 	dev->mtu = new_mtu;
305 	return 0;
306 }
307 
308 /* Initialize net device. */
309 static void tun_net_init(struct net_device *dev)
310 {
311 	struct tun_struct *tun = netdev_priv(dev);
312 
313 	switch (tun->flags & TUN_TYPE_MASK) {
314 	case TUN_TUN_DEV:
315 		/* Point-to-Point TUN Device */
316 		dev->hard_header_len = 0;
317 		dev->addr_len = 0;
318 		dev->mtu = 1500;
319 		dev->change_mtu = tun_net_change_mtu;
320 
321 		/* Zero header length */
322 		dev->type = ARPHRD_NONE;
323 		dev->flags = IFF_POINTOPOINT | IFF_NOARP | IFF_MULTICAST;
324 		dev->tx_queue_len = TUN_READQ_SIZE;  /* We prefer our own queue length */
325 		break;
326 
327 	case TUN_TAP_DEV:
328 		/* Ethernet TAP Device */
329 		ether_setup(dev);
330 		dev->change_mtu         = tun_net_change_mtu;
331 		dev->set_multicast_list = tun_net_mclist;
332 
333 		random_ether_addr(dev->dev_addr);
334 
335 		dev->tx_queue_len = TUN_READQ_SIZE;  /* We prefer our own queue length */
336 		break;
337 	}
338 }
339 
340 /* Character device part */
341 
342 /* Poll */
343 static unsigned int tun_chr_poll(struct file *file, poll_table * wait)
344 {
345 	struct tun_struct *tun = file->private_data;
346 	unsigned int mask = POLLOUT | POLLWRNORM;
347 
348 	if (!tun)
349 		return -EBADFD;
350 
351 	DBG(KERN_INFO "%s: tun_chr_poll\n", tun->dev->name);
352 
353 	poll_wait(file, &tun->read_wait, wait);
354 
355 	if (!skb_queue_empty(&tun->readq))
356 		mask |= POLLIN | POLLRDNORM;
357 
358 	return mask;
359 }
360 
361 /* Get packet from user space buffer */
362 static __inline__ ssize_t tun_get_user(struct tun_struct *tun, struct iovec *iv, size_t count)
363 {
364 	struct tun_pi pi = { 0, __constant_htons(ETH_P_IP) };
365 	struct sk_buff *skb;
366 	size_t len = count, align = 0;
367 	struct virtio_net_hdr gso = { 0 };
368 
369 	if (!(tun->flags & TUN_NO_PI)) {
370 		if ((len -= sizeof(pi)) > count)
371 			return -EINVAL;
372 
373 		if(memcpy_fromiovec((void *)&pi, iv, sizeof(pi)))
374 			return -EFAULT;
375 	}
376 
377 	if (tun->flags & TUN_VNET_HDR) {
378 		if ((len -= sizeof(gso)) > count)
379 			return -EINVAL;
380 
381 		if (memcpy_fromiovec((void *)&gso, iv, sizeof(gso)))
382 			return -EFAULT;
383 
384 		if (gso.hdr_len > len)
385 			return -EINVAL;
386 	}
387 
388 	if ((tun->flags & TUN_TYPE_MASK) == TUN_TAP_DEV) {
389 		align = NET_IP_ALIGN;
390 		if (unlikely(len < ETH_HLEN))
391 			return -EINVAL;
392 	}
393 
394 	if (!(skb = alloc_skb(len + align, GFP_KERNEL))) {
395 		tun->dev->stats.rx_dropped++;
396 		return -ENOMEM;
397 	}
398 
399 	if (align)
400 		skb_reserve(skb, align);
401 	if (memcpy_fromiovec(skb_put(skb, len), iv, len)) {
402 		tun->dev->stats.rx_dropped++;
403 		kfree_skb(skb);
404 		return -EFAULT;
405 	}
406 
407 	if (gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
408 		if (!skb_partial_csum_set(skb, gso.csum_start,
409 					  gso.csum_offset)) {
410 			tun->dev->stats.rx_frame_errors++;
411 			kfree_skb(skb);
412 			return -EINVAL;
413 		}
414 	} else if (tun->flags & TUN_NOCHECKSUM)
415 		skb->ip_summed = CHECKSUM_UNNECESSARY;
416 
417 	switch (tun->flags & TUN_TYPE_MASK) {
418 	case TUN_TUN_DEV:
419 		if (tun->flags & TUN_NO_PI) {
420 			switch (skb->data[0] & 0xf0) {
421 			case 0x40:
422 				pi.proto = htons(ETH_P_IP);
423 				break;
424 			case 0x60:
425 				pi.proto = htons(ETH_P_IPV6);
426 				break;
427 			default:
428 				tun->dev->stats.rx_dropped++;
429 				kfree_skb(skb);
430 				return -EINVAL;
431 			}
432 		}
433 
434 		skb_reset_mac_header(skb);
435 		skb->protocol = pi.proto;
436 		skb->dev = tun->dev;
437 		break;
438 	case TUN_TAP_DEV:
439 		skb->protocol = eth_type_trans(skb, tun->dev);
440 		break;
441 	};
442 
443 	if (gso.gso_type != VIRTIO_NET_HDR_GSO_NONE) {
444 		pr_debug("GSO!\n");
445 		switch (gso.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
446 		case VIRTIO_NET_HDR_GSO_TCPV4:
447 			skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
448 			break;
449 		case VIRTIO_NET_HDR_GSO_TCPV6:
450 			skb_shinfo(skb)->gso_type = SKB_GSO_TCPV6;
451 			break;
452 		default:
453 			tun->dev->stats.rx_frame_errors++;
454 			kfree_skb(skb);
455 			return -EINVAL;
456 		}
457 
458 		if (gso.gso_type & VIRTIO_NET_HDR_GSO_ECN)
459 			skb_shinfo(skb)->gso_type |= SKB_GSO_TCP_ECN;
460 
461 		skb_shinfo(skb)->gso_size = gso.gso_size;
462 		if (skb_shinfo(skb)->gso_size == 0) {
463 			tun->dev->stats.rx_frame_errors++;
464 			kfree_skb(skb);
465 			return -EINVAL;
466 		}
467 
468 		/* Header must be checked, and gso_segs computed. */
469 		skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
470 		skb_shinfo(skb)->gso_segs = 0;
471 	}
472 
473 	netif_rx_ni(skb);
474 	tun->dev->last_rx = jiffies;
475 
476 	tun->dev->stats.rx_packets++;
477 	tun->dev->stats.rx_bytes += len;
478 
479 	return count;
480 }
481 
482 static ssize_t tun_chr_aio_write(struct kiocb *iocb, const struct iovec *iv,
483 			      unsigned long count, loff_t pos)
484 {
485 	struct tun_struct *tun = iocb->ki_filp->private_data;
486 
487 	if (!tun)
488 		return -EBADFD;
489 
490 	DBG(KERN_INFO "%s: tun_chr_write %ld\n", tun->dev->name, count);
491 
492 	return tun_get_user(tun, (struct iovec *) iv, iov_length(iv, count));
493 }
494 
495 /* Put packet to the user space buffer */
496 static __inline__ ssize_t tun_put_user(struct tun_struct *tun,
497 				       struct sk_buff *skb,
498 				       struct iovec *iv, int len)
499 {
500 	struct tun_pi pi = { 0, skb->protocol };
501 	ssize_t total = 0;
502 
503 	if (!(tun->flags & TUN_NO_PI)) {
504 		if ((len -= sizeof(pi)) < 0)
505 			return -EINVAL;
506 
507 		if (len < skb->len) {
508 			/* Packet will be striped */
509 			pi.flags |= TUN_PKT_STRIP;
510 		}
511 
512 		if (memcpy_toiovec(iv, (void *) &pi, sizeof(pi)))
513 			return -EFAULT;
514 		total += sizeof(pi);
515 	}
516 
517 	if (tun->flags & TUN_VNET_HDR) {
518 		struct virtio_net_hdr gso = { 0 }; /* no info leak */
519 		if ((len -= sizeof(gso)) < 0)
520 			return -EINVAL;
521 
522 		if (skb_is_gso(skb)) {
523 			struct skb_shared_info *sinfo = skb_shinfo(skb);
524 
525 			/* This is a hint as to how much should be linear. */
526 			gso.hdr_len = skb_headlen(skb);
527 			gso.gso_size = sinfo->gso_size;
528 			if (sinfo->gso_type & SKB_GSO_TCPV4)
529 				gso.gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
530 			else if (sinfo->gso_type & SKB_GSO_TCPV6)
531 				gso.gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
532 			else
533 				BUG();
534 			if (sinfo->gso_type & SKB_GSO_TCP_ECN)
535 				gso.gso_type |= VIRTIO_NET_HDR_GSO_ECN;
536 		} else
537 			gso.gso_type = VIRTIO_NET_HDR_GSO_NONE;
538 
539 		if (skb->ip_summed == CHECKSUM_PARTIAL) {
540 			gso.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
541 			gso.csum_start = skb->csum_start - skb_headroom(skb);
542 			gso.csum_offset = skb->csum_offset;
543 		} /* else everything is zero */
544 
545 		if (unlikely(memcpy_toiovec(iv, (void *)&gso, sizeof(gso))))
546 			return -EFAULT;
547 		total += sizeof(gso);
548 	}
549 
550 	len = min_t(int, skb->len, len);
551 
552 	skb_copy_datagram_iovec(skb, 0, iv, len);
553 	total += len;
554 
555 	tun->dev->stats.tx_packets++;
556 	tun->dev->stats.tx_bytes += len;
557 
558 	return total;
559 }
560 
561 static ssize_t tun_chr_aio_read(struct kiocb *iocb, const struct iovec *iv,
562 			    unsigned long count, loff_t pos)
563 {
564 	struct file *file = iocb->ki_filp;
565 	struct tun_struct *tun = file->private_data;
566 	DECLARE_WAITQUEUE(wait, current);
567 	struct sk_buff *skb;
568 	ssize_t len, ret = 0;
569 
570 	if (!tun)
571 		return -EBADFD;
572 
573 	DBG(KERN_INFO "%s: tun_chr_read\n", tun->dev->name);
574 
575 	len = iov_length(iv, count);
576 	if (len < 0)
577 		return -EINVAL;
578 
579 	add_wait_queue(&tun->read_wait, &wait);
580 	while (len) {
581 		current->state = TASK_INTERRUPTIBLE;
582 
583 		/* Read frames from the queue */
584 		if (!(skb=skb_dequeue(&tun->readq))) {
585 			if (file->f_flags & O_NONBLOCK) {
586 				ret = -EAGAIN;
587 				break;
588 			}
589 			if (signal_pending(current)) {
590 				ret = -ERESTARTSYS;
591 				break;
592 			}
593 
594 			/* Nothing to read, let's sleep */
595 			schedule();
596 			continue;
597 		}
598 		netif_wake_queue(tun->dev);
599 
600 		ret = tun_put_user(tun, skb, (struct iovec *) iv, len);
601 		kfree_skb(skb);
602 		break;
603 	}
604 
605 	current->state = TASK_RUNNING;
606 	remove_wait_queue(&tun->read_wait, &wait);
607 
608 	return ret;
609 }
610 
611 static void tun_setup(struct net_device *dev)
612 {
613 	struct tun_struct *tun = netdev_priv(dev);
614 
615 	skb_queue_head_init(&tun->readq);
616 	init_waitqueue_head(&tun->read_wait);
617 
618 	tun->owner = -1;
619 	tun->group = -1;
620 
621 	dev->open = tun_net_open;
622 	dev->hard_start_xmit = tun_net_xmit;
623 	dev->stop = tun_net_close;
624 	dev->ethtool_ops = &tun_ethtool_ops;
625 	dev->destructor = free_netdev;
626 	dev->features |= NETIF_F_NETNS_LOCAL;
627 }
628 
629 static struct tun_struct *tun_get_by_name(struct tun_net *tn, const char *name)
630 {
631 	struct tun_struct *tun;
632 
633 	ASSERT_RTNL();
634 	list_for_each_entry(tun, &tn->dev_list, list) {
635 		if (!strncmp(tun->dev->name, name, IFNAMSIZ))
636 		    return tun;
637 	}
638 
639 	return NULL;
640 }
641 
642 static int tun_set_iff(struct net *net, struct file *file, struct ifreq *ifr)
643 {
644 	struct tun_net *tn;
645 	struct tun_struct *tun;
646 	struct net_device *dev;
647 	int err;
648 
649 	tn = net_generic(net, tun_net_id);
650 	tun = tun_get_by_name(tn, ifr->ifr_name);
651 	if (tun) {
652 		if (tun->attached)
653 			return -EBUSY;
654 
655 		/* Check permissions */
656 		if (((tun->owner != -1 &&
657 		      current->euid != tun->owner) ||
658 		     (tun->group != -1 &&
659 		      current->egid != tun->group)) &&
660 		     !capable(CAP_NET_ADMIN))
661 			return -EPERM;
662 	}
663 	else if (__dev_get_by_name(net, ifr->ifr_name))
664 		return -EINVAL;
665 	else {
666 		char *name;
667 		unsigned long flags = 0;
668 
669 		err = -EINVAL;
670 
671 		if (!capable(CAP_NET_ADMIN))
672 			return -EPERM;
673 
674 		/* Set dev type */
675 		if (ifr->ifr_flags & IFF_TUN) {
676 			/* TUN device */
677 			flags |= TUN_TUN_DEV;
678 			name = "tun%d";
679 		} else if (ifr->ifr_flags & IFF_TAP) {
680 			/* TAP device */
681 			flags |= TUN_TAP_DEV;
682 			name = "tap%d";
683 		} else
684 			goto failed;
685 
686 		if (*ifr->ifr_name)
687 			name = ifr->ifr_name;
688 
689 		dev = alloc_netdev(sizeof(struct tun_struct), name,
690 				   tun_setup);
691 		if (!dev)
692 			return -ENOMEM;
693 
694 		dev_net_set(dev, net);
695 		tun = netdev_priv(dev);
696 		tun->dev = dev;
697 		tun->flags = flags;
698 		tun->txflt.count = 0;
699 
700 		tun_net_init(dev);
701 
702 		if (strchr(dev->name, '%')) {
703 			err = dev_alloc_name(dev, dev->name);
704 			if (err < 0)
705 				goto err_free_dev;
706 		}
707 
708 		err = register_netdevice(tun->dev);
709 		if (err < 0)
710 			goto err_free_dev;
711 
712 		list_add(&tun->list, &tn->dev_list);
713 	}
714 
715 	DBG(KERN_INFO "%s: tun_set_iff\n", tun->dev->name);
716 
717 	if (ifr->ifr_flags & IFF_NO_PI)
718 		tun->flags |= TUN_NO_PI;
719 	else
720 		tun->flags &= ~TUN_NO_PI;
721 
722 	if (ifr->ifr_flags & IFF_ONE_QUEUE)
723 		tun->flags |= TUN_ONE_QUEUE;
724 	else
725 		tun->flags &= ~TUN_ONE_QUEUE;
726 
727 	if (ifr->ifr_flags & IFF_VNET_HDR)
728 		tun->flags |= TUN_VNET_HDR;
729 	else
730 		tun->flags &= ~TUN_VNET_HDR;
731 
732 	file->private_data = tun;
733 	tun->attached = 1;
734 	get_net(dev_net(tun->dev));
735 
736 	/* Make sure persistent devices do not get stuck in
737 	 * xoff state.
738 	 */
739 	if (netif_running(tun->dev))
740 		netif_wake_queue(tun->dev);
741 
742 	strcpy(ifr->ifr_name, tun->dev->name);
743 	return 0;
744 
745  err_free_dev:
746 	free_netdev(dev);
747  failed:
748 	return err;
749 }
750 
751 /* This is like a cut-down ethtool ops, except done via tun fd so no
752  * privs required. */
753 static int set_offload(struct net_device *dev, unsigned long arg)
754 {
755 	unsigned int old_features, features;
756 
757 	old_features = dev->features;
758 	/* Unset features, set them as we chew on the arg. */
759 	features = (old_features & ~(NETIF_F_HW_CSUM|NETIF_F_SG|NETIF_F_FRAGLIST
760 				    |NETIF_F_TSO_ECN|NETIF_F_TSO|NETIF_F_TSO6));
761 
762 	if (arg & TUN_F_CSUM) {
763 		features |= NETIF_F_HW_CSUM|NETIF_F_SG|NETIF_F_FRAGLIST;
764 		arg &= ~TUN_F_CSUM;
765 
766 		if (arg & (TUN_F_TSO4|TUN_F_TSO6)) {
767 			if (arg & TUN_F_TSO_ECN) {
768 				features |= NETIF_F_TSO_ECN;
769 				arg &= ~TUN_F_TSO_ECN;
770 			}
771 			if (arg & TUN_F_TSO4)
772 				features |= NETIF_F_TSO;
773 			if (arg & TUN_F_TSO6)
774 				features |= NETIF_F_TSO6;
775 			arg &= ~(TUN_F_TSO4|TUN_F_TSO6);
776 		}
777 	}
778 
779 	/* This gives the user a way to test for new features in future by
780 	 * trying to set them. */
781 	if (arg)
782 		return -EINVAL;
783 
784 	dev->features = features;
785 	if (old_features != dev->features)
786 		netdev_features_change(dev);
787 
788 	return 0;
789 }
790 
791 static int tun_chr_ioctl(struct inode *inode, struct file *file,
792 			 unsigned int cmd, unsigned long arg)
793 {
794 	struct tun_struct *tun = file->private_data;
795 	void __user* argp = (void __user*)arg;
796 	struct ifreq ifr;
797 	int ret;
798 	DECLARE_MAC_BUF(mac);
799 
800 	if (cmd == TUNSETIFF || _IOC_TYPE(cmd) == 0x89)
801 		if (copy_from_user(&ifr, argp, sizeof ifr))
802 			return -EFAULT;
803 
804 	if (cmd == TUNSETIFF && !tun) {
805 		int err;
806 
807 		ifr.ifr_name[IFNAMSIZ-1] = '\0';
808 
809 		rtnl_lock();
810 		err = tun_set_iff(current->nsproxy->net_ns, file, &ifr);
811 		rtnl_unlock();
812 
813 		if (err)
814 			return err;
815 
816 		if (copy_to_user(argp, &ifr, sizeof(ifr)))
817 			return -EFAULT;
818 		return 0;
819 	}
820 
821 	if (cmd == TUNGETFEATURES) {
822 		/* Currently this just means: "what IFF flags are valid?".
823 		 * This is needed because we never checked for invalid flags on
824 		 * TUNSETIFF. */
825 		return put_user(IFF_TUN | IFF_TAP | IFF_NO_PI | IFF_ONE_QUEUE |
826 				IFF_VNET_HDR,
827 				(unsigned int __user*)argp);
828 	}
829 
830 	if (!tun)
831 		return -EBADFD;
832 
833 	DBG(KERN_INFO "%s: tun_chr_ioctl cmd %d\n", tun->dev->name, cmd);
834 
835 	switch (cmd) {
836 	case TUNSETNOCSUM:
837 		/* Disable/Enable checksum */
838 		if (arg)
839 			tun->flags |= TUN_NOCHECKSUM;
840 		else
841 			tun->flags &= ~TUN_NOCHECKSUM;
842 
843 		DBG(KERN_INFO "%s: checksum %s\n",
844 		    tun->dev->name, arg ? "disabled" : "enabled");
845 		break;
846 
847 	case TUNSETPERSIST:
848 		/* Disable/Enable persist mode */
849 		if (arg)
850 			tun->flags |= TUN_PERSIST;
851 		else
852 			tun->flags &= ~TUN_PERSIST;
853 
854 		DBG(KERN_INFO "%s: persist %s\n",
855 		    tun->dev->name, arg ? "enabled" : "disabled");
856 		break;
857 
858 	case TUNSETOWNER:
859 		/* Set owner of the device */
860 		tun->owner = (uid_t) arg;
861 
862 		DBG(KERN_INFO "%s: owner set to %d\n", tun->dev->name, tun->owner);
863 		break;
864 
865 	case TUNSETGROUP:
866 		/* Set group of the device */
867 		tun->group= (gid_t) arg;
868 
869 		DBG(KERN_INFO "%s: group set to %d\n", tun->dev->name, tun->group);
870 		break;
871 
872 	case TUNSETLINK:
873 		/* Only allow setting the type when the interface is down */
874 		rtnl_lock();
875 		if (tun->dev->flags & IFF_UP) {
876 			DBG(KERN_INFO "%s: Linktype set failed because interface is up\n",
877 				tun->dev->name);
878 			ret = -EBUSY;
879 		} else {
880 			tun->dev->type = (int) arg;
881 			DBG(KERN_INFO "%s: linktype set to %d\n", tun->dev->name, tun->dev->type);
882 			ret = 0;
883 		}
884 		rtnl_unlock();
885 		return ret;
886 
887 #ifdef TUN_DEBUG
888 	case TUNSETDEBUG:
889 		tun->debug = arg;
890 		break;
891 #endif
892 	case TUNSETOFFLOAD:
893 		rtnl_lock();
894 		ret = set_offload(tun->dev, arg);
895 		rtnl_unlock();
896 		return ret;
897 
898 	case TUNSETTXFILTER:
899 		/* Can be set only for TAPs */
900 		if ((tun->flags & TUN_TYPE_MASK) != TUN_TAP_DEV)
901 			return -EINVAL;
902 		rtnl_lock();
903 		ret = update_filter(&tun->txflt, (void __user *)arg);
904 		rtnl_unlock();
905 		return ret;
906 
907 	case SIOCGIFHWADDR:
908 		/* Get hw addres */
909 		memcpy(ifr.ifr_hwaddr.sa_data, tun->dev->dev_addr, ETH_ALEN);
910 		ifr.ifr_hwaddr.sa_family = tun->dev->type;
911 		if (copy_to_user(argp, &ifr, sizeof ifr))
912 			return -EFAULT;
913 		return 0;
914 
915 	case SIOCSIFHWADDR:
916 		/* Set hw address */
917 		DBG(KERN_DEBUG "%s: set hw address: %s\n",
918 			tun->dev->name, print_mac(mac, ifr.ifr_hwaddr.sa_data));
919 
920 		rtnl_lock();
921 		ret = dev_set_mac_address(tun->dev, &ifr.ifr_hwaddr);
922 		rtnl_unlock();
923 		return ret;
924 
925 	default:
926 		return -EINVAL;
927 	};
928 
929 	return 0;
930 }
931 
932 static int tun_chr_fasync(int fd, struct file *file, int on)
933 {
934 	struct tun_struct *tun = file->private_data;
935 	int ret;
936 
937 	if (!tun)
938 		return -EBADFD;
939 
940 	DBG(KERN_INFO "%s: tun_chr_fasync %d\n", tun->dev->name, on);
941 
942 	lock_kernel();
943 	if ((ret = fasync_helper(fd, file, on, &tun->fasync)) < 0)
944 		goto out;
945 
946 	if (on) {
947 		ret = __f_setown(file, task_pid(current), PIDTYPE_PID, 0);
948 		if (ret)
949 			goto out;
950 		tun->flags |= TUN_FASYNC;
951 	} else
952 		tun->flags &= ~TUN_FASYNC;
953 	ret = 0;
954 out:
955 	unlock_kernel();
956 	return ret;
957 }
958 
959 static int tun_chr_open(struct inode *inode, struct file * file)
960 {
961 	cycle_kernel_lock();
962 	DBG1(KERN_INFO "tunX: tun_chr_open\n");
963 	file->private_data = NULL;
964 	return 0;
965 }
966 
967 static int tun_chr_close(struct inode *inode, struct file *file)
968 {
969 	struct tun_struct *tun = file->private_data;
970 
971 	if (!tun)
972 		return 0;
973 
974 	DBG(KERN_INFO "%s: tun_chr_close\n", tun->dev->name);
975 
976 	tun_chr_fasync(-1, file, 0);
977 
978 	rtnl_lock();
979 
980 	/* Detach from net device */
981 	file->private_data = NULL;
982 	tun->attached = 0;
983 	put_net(dev_net(tun->dev));
984 
985 	/* Drop read queue */
986 	skb_queue_purge(&tun->readq);
987 
988 	if (!(tun->flags & TUN_PERSIST)) {
989 		list_del(&tun->list);
990 		unregister_netdevice(tun->dev);
991 	}
992 
993 	rtnl_unlock();
994 
995 	return 0;
996 }
997 
998 static const struct file_operations tun_fops = {
999 	.owner	= THIS_MODULE,
1000 	.llseek = no_llseek,
1001 	.read  = do_sync_read,
1002 	.aio_read  = tun_chr_aio_read,
1003 	.write = do_sync_write,
1004 	.aio_write = tun_chr_aio_write,
1005 	.poll	= tun_chr_poll,
1006 	.ioctl	= tun_chr_ioctl,
1007 	.open	= tun_chr_open,
1008 	.release = tun_chr_close,
1009 	.fasync = tun_chr_fasync
1010 };
1011 
1012 static struct miscdevice tun_miscdev = {
1013 	.minor = TUN_MINOR,
1014 	.name = "tun",
1015 	.fops = &tun_fops,
1016 };
1017 
1018 /* ethtool interface */
1019 
1020 static int tun_get_settings(struct net_device *dev, struct ethtool_cmd *cmd)
1021 {
1022 	cmd->supported		= 0;
1023 	cmd->advertising	= 0;
1024 	cmd->speed		= SPEED_10;
1025 	cmd->duplex		= DUPLEX_FULL;
1026 	cmd->port		= PORT_TP;
1027 	cmd->phy_address	= 0;
1028 	cmd->transceiver	= XCVR_INTERNAL;
1029 	cmd->autoneg		= AUTONEG_DISABLE;
1030 	cmd->maxtxpkt		= 0;
1031 	cmd->maxrxpkt		= 0;
1032 	return 0;
1033 }
1034 
1035 static void tun_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
1036 {
1037 	struct tun_struct *tun = netdev_priv(dev);
1038 
1039 	strcpy(info->driver, DRV_NAME);
1040 	strcpy(info->version, DRV_VERSION);
1041 	strcpy(info->fw_version, "N/A");
1042 
1043 	switch (tun->flags & TUN_TYPE_MASK) {
1044 	case TUN_TUN_DEV:
1045 		strcpy(info->bus_info, "tun");
1046 		break;
1047 	case TUN_TAP_DEV:
1048 		strcpy(info->bus_info, "tap");
1049 		break;
1050 	}
1051 }
1052 
1053 static u32 tun_get_msglevel(struct net_device *dev)
1054 {
1055 #ifdef TUN_DEBUG
1056 	struct tun_struct *tun = netdev_priv(dev);
1057 	return tun->debug;
1058 #else
1059 	return -EOPNOTSUPP;
1060 #endif
1061 }
1062 
1063 static void tun_set_msglevel(struct net_device *dev, u32 value)
1064 {
1065 #ifdef TUN_DEBUG
1066 	struct tun_struct *tun = netdev_priv(dev);
1067 	tun->debug = value;
1068 #endif
1069 }
1070 
1071 static u32 tun_get_link(struct net_device *dev)
1072 {
1073 	struct tun_struct *tun = netdev_priv(dev);
1074 	return tun->attached;
1075 }
1076 
1077 static u32 tun_get_rx_csum(struct net_device *dev)
1078 {
1079 	struct tun_struct *tun = netdev_priv(dev);
1080 	return (tun->flags & TUN_NOCHECKSUM) == 0;
1081 }
1082 
1083 static int tun_set_rx_csum(struct net_device *dev, u32 data)
1084 {
1085 	struct tun_struct *tun = netdev_priv(dev);
1086 	if (data)
1087 		tun->flags &= ~TUN_NOCHECKSUM;
1088 	else
1089 		tun->flags |= TUN_NOCHECKSUM;
1090 	return 0;
1091 }
1092 
1093 static const struct ethtool_ops tun_ethtool_ops = {
1094 	.get_settings	= tun_get_settings,
1095 	.get_drvinfo	= tun_get_drvinfo,
1096 	.get_msglevel	= tun_get_msglevel,
1097 	.set_msglevel	= tun_set_msglevel,
1098 	.get_link	= tun_get_link,
1099 	.get_rx_csum	= tun_get_rx_csum,
1100 	.set_rx_csum	= tun_set_rx_csum
1101 };
1102 
1103 static int tun_init_net(struct net *net)
1104 {
1105 	struct tun_net *tn;
1106 
1107 	tn = kmalloc(sizeof(*tn), GFP_KERNEL);
1108 	if (tn == NULL)
1109 		return -ENOMEM;
1110 
1111 	INIT_LIST_HEAD(&tn->dev_list);
1112 
1113 	if (net_assign_generic(net, tun_net_id, tn)) {
1114 		kfree(tn);
1115 		return -ENOMEM;
1116 	}
1117 
1118 	return 0;
1119 }
1120 
1121 static void tun_exit_net(struct net *net)
1122 {
1123 	struct tun_net *tn;
1124 	struct tun_struct *tun, *nxt;
1125 
1126 	tn = net_generic(net, tun_net_id);
1127 
1128 	rtnl_lock();
1129 	list_for_each_entry_safe(tun, nxt, &tn->dev_list, list) {
1130 		DBG(KERN_INFO "%s cleaned up\n", tun->dev->name);
1131 		unregister_netdevice(tun->dev);
1132 	}
1133 	rtnl_unlock();
1134 
1135 	kfree(tn);
1136 }
1137 
1138 static struct pernet_operations tun_net_ops = {
1139 	.init = tun_init_net,
1140 	.exit = tun_exit_net,
1141 };
1142 
1143 static int __init tun_init(void)
1144 {
1145 	int ret = 0;
1146 
1147 	printk(KERN_INFO "tun: %s, %s\n", DRV_DESCRIPTION, DRV_VERSION);
1148 	printk(KERN_INFO "tun: %s\n", DRV_COPYRIGHT);
1149 
1150 	ret = register_pernet_gen_device(&tun_net_id, &tun_net_ops);
1151 	if (ret) {
1152 		printk(KERN_ERR "tun: Can't register pernet ops\n");
1153 		goto err_pernet;
1154 	}
1155 
1156 	ret = misc_register(&tun_miscdev);
1157 	if (ret) {
1158 		printk(KERN_ERR "tun: Can't register misc device %d\n", TUN_MINOR);
1159 		goto err_misc;
1160 	}
1161 	return 0;
1162 
1163 err_misc:
1164 	unregister_pernet_gen_device(tun_net_id, &tun_net_ops);
1165 err_pernet:
1166 	return ret;
1167 }
1168 
1169 static void tun_cleanup(void)
1170 {
1171 	misc_deregister(&tun_miscdev);
1172 	unregister_pernet_gen_device(tun_net_id, &tun_net_ops);
1173 }
1174 
1175 module_init(tun_init);
1176 module_exit(tun_cleanup);
1177 MODULE_DESCRIPTION(DRV_DESCRIPTION);
1178 MODULE_AUTHOR(DRV_COPYRIGHT);
1179 MODULE_LICENSE("GPL");
1180 MODULE_ALIAS_MISCDEV(TUN_MINOR);
1181