Merge branch 'master' of git://eden-feed.erg.abdn.ac.uk/net-next-2.6
[linux-2.6] / drivers / net / tun.c
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  *  Brian Braunstein <linuxkernel@bristyle.com> 2007/03/23
22  *    Fixed hw address handling.  Now net_device.dev_addr is kept consistent
23  *    with tun.dev_addr when the address is set by this module.
24  *
25  *  Mike Kershaw <dragorn@kismetwireless.net> 2005/08/14
26  *    Add TUNSETLINK ioctl to set the link encapsulation
27  *
28  *  Mark Smith <markzzzsmith@yahoo.com.au>
29  *   Use random_ether_addr() for tap MAC address.
30  *
31  *  Harald Roelle <harald.roelle@ifi.lmu.de>  2004/04/20
32  *    Fixes in packet dropping, queue length setting and queue wakeup.
33  *    Increased default tx queue length.
34  *    Added ethtool API.
35  *    Minor cleanups
36  *
37  *  Daniel Podlejski <underley@underley.eu.org>
38  *    Modifications for 2.3.99-pre5 kernel.
39  */
40
41 #define DRV_NAME        "tun"
42 #define DRV_VERSION     "1.6"
43 #define DRV_DESCRIPTION "Universal TUN/TAP device driver"
44 #define DRV_COPYRIGHT   "(C) 1999-2004 Max Krasnyansky <maxk@qualcomm.com>"
45
46 #include <linux/module.h>
47 #include <linux/errno.h>
48 #include <linux/kernel.h>
49 #include <linux/major.h>
50 #include <linux/slab.h>
51 #include <linux/poll.h>
52 #include <linux/fcntl.h>
53 #include <linux/init.h>
54 #include <linux/skbuff.h>
55 #include <linux/netdevice.h>
56 #include <linux/etherdevice.h>
57 #include <linux/miscdevice.h>
58 #include <linux/ethtool.h>
59 #include <linux/rtnetlink.h>
60 #include <linux/if.h>
61 #include <linux/if_arp.h>
62 #include <linux/if_ether.h>
63 #include <linux/if_tun.h>
64 #include <linux/crc32.h>
65 #include <linux/nsproxy.h>
66 #include <linux/virtio_net.h>
67 #include <net/net_namespace.h>
68 #include <net/netns/generic.h>
69
70 #include <asm/system.h>
71 #include <asm/uaccess.h>
72
73 /* Uncomment to enable debugging */
74 /* #define TUN_DEBUG 1 */
75
76 #ifdef TUN_DEBUG
77 static int debug;
78
79 #define DBG  if(tun->debug)printk
80 #define DBG1 if(debug==2)printk
81 #else
82 #define DBG( a... )
83 #define DBG1( a... )
84 #endif
85
86 struct tun_struct {
87         struct list_head        list;
88         unsigned long           flags;
89         int                     attached;
90         uid_t                   owner;
91         gid_t                   group;
92
93         wait_queue_head_t       read_wait;
94         struct sk_buff_head     readq;
95
96         struct net_device       *dev;
97
98         struct fasync_struct    *fasync;
99
100         unsigned long if_flags;
101         u8 dev_addr[ETH_ALEN];
102         u32 chr_filter[2];
103         u32 net_filter[2];
104
105 #ifdef TUN_DEBUG
106         int debug;
107 #endif
108 };
109
110 /* Network device part of the driver */
111
112 static unsigned int tun_net_id;
113 struct tun_net {
114         struct list_head dev_list;
115 };
116
117 static const struct ethtool_ops tun_ethtool_ops;
118
119 /* Net device open. */
120 static int tun_net_open(struct net_device *dev)
121 {
122         netif_start_queue(dev);
123         return 0;
124 }
125
126 /* Net device close. */
127 static int tun_net_close(struct net_device *dev)
128 {
129         netif_stop_queue(dev);
130         return 0;
131 }
132
133 /* Net device start xmit */
134 static int tun_net_xmit(struct sk_buff *skb, struct net_device *dev)
135 {
136         struct tun_struct *tun = netdev_priv(dev);
137
138         DBG(KERN_INFO "%s: tun_net_xmit %d\n", tun->dev->name, skb->len);
139
140         /* Drop packet if interface is not attached */
141         if (!tun->attached)
142                 goto drop;
143
144         /* Packet dropping */
145         if (skb_queue_len(&tun->readq) >= dev->tx_queue_len) {
146                 if (!(tun->flags & TUN_ONE_QUEUE)) {
147                         /* Normal queueing mode. */
148                         /* Packet scheduler handles dropping of further packets. */
149                         netif_stop_queue(dev);
150
151                         /* We won't see all dropped packets individually, so overrun
152                          * error is more appropriate. */
153                         dev->stats.tx_fifo_errors++;
154                 } else {
155                         /* Single queue mode.
156                          * Driver handles dropping of all packets itself. */
157                         goto drop;
158                 }
159         }
160
161         /* Queue packet */
162         skb_queue_tail(&tun->readq, skb);
163         dev->trans_start = jiffies;
164
165         /* Notify and wake up reader process */
166         if (tun->flags & TUN_FASYNC)
167                 kill_fasync(&tun->fasync, SIGIO, POLL_IN);
168         wake_up_interruptible(&tun->read_wait);
169         return 0;
170
171 drop:
172         dev->stats.tx_dropped++;
173         kfree_skb(skb);
174         return 0;
175 }
176
177 /** Add the specified Ethernet address to this multicast filter. */
178 static void
179 add_multi(u32* filter, const u8* addr)
180 {
181         int bit_nr = ether_crc(ETH_ALEN, addr) >> 26;
182         filter[bit_nr >> 5] |= 1 << (bit_nr & 31);
183 }
184
185 /** Remove the specified Ethernet addres from this multicast filter. */
186 static void
187 del_multi(u32* filter, const u8* addr)
188 {
189         int bit_nr = ether_crc(ETH_ALEN, addr) >> 26;
190         filter[bit_nr >> 5] &= ~(1 << (bit_nr & 31));
191 }
192
193 /** Update the list of multicast groups to which the network device belongs.
194  * This list is used to filter packets being sent from the character device to
195  * the network device. */
196 static void
197 tun_net_mclist(struct net_device *dev)
198 {
199         struct tun_struct *tun = netdev_priv(dev);
200         const struct dev_mc_list *mclist;
201         int i;
202         DECLARE_MAC_BUF(mac);
203         DBG(KERN_DEBUG "%s: tun_net_mclist: mc_count %d\n",
204                         dev->name, dev->mc_count);
205         memset(tun->chr_filter, 0, sizeof tun->chr_filter);
206         for (i = 0, mclist = dev->mc_list; i < dev->mc_count && mclist != NULL;
207                         i++, mclist = mclist->next) {
208                 add_multi(tun->net_filter, mclist->dmi_addr);
209                 DBG(KERN_DEBUG "%s: tun_net_mclist: %s\n",
210                     dev->name, print_mac(mac, mclist->dmi_addr));
211         }
212 }
213
214 #define MIN_MTU 68
215 #define MAX_MTU 65535
216
217 static int
218 tun_net_change_mtu(struct net_device *dev, int new_mtu)
219 {
220         if (new_mtu < MIN_MTU || new_mtu + dev->hard_header_len > MAX_MTU)
221                 return -EINVAL;
222         dev->mtu = new_mtu;
223         return 0;
224 }
225
226 /* Initialize net device. */
227 static void tun_net_init(struct net_device *dev)
228 {
229         struct tun_struct *tun = netdev_priv(dev);
230
231         switch (tun->flags & TUN_TYPE_MASK) {
232         case TUN_TUN_DEV:
233                 /* Point-to-Point TUN Device */
234                 dev->hard_header_len = 0;
235                 dev->addr_len = 0;
236                 dev->mtu = 1500;
237                 dev->change_mtu = tun_net_change_mtu;
238
239                 /* Zero header length */
240                 dev->type = ARPHRD_NONE;
241                 dev->flags = IFF_POINTOPOINT | IFF_NOARP | IFF_MULTICAST;
242                 dev->tx_queue_len = TUN_READQ_SIZE;  /* We prefer our own queue length */
243                 break;
244
245         case TUN_TAP_DEV:
246                 /* Ethernet TAP Device */
247                 dev->set_multicast_list = tun_net_mclist;
248
249                 ether_setup(dev);
250                 dev->change_mtu = tun_net_change_mtu;
251
252                 /* random address already created for us by tun_set_iff, use it */
253                 memcpy(dev->dev_addr, tun->dev_addr, min(sizeof(tun->dev_addr), sizeof(dev->dev_addr)) );
254
255                 dev->tx_queue_len = TUN_READQ_SIZE;  /* We prefer our own queue length */
256                 break;
257         }
258 }
259
260 /* Character device part */
261
262 /* Poll */
263 static unsigned int tun_chr_poll(struct file *file, poll_table * wait)
264 {
265         struct tun_struct *tun = file->private_data;
266         unsigned int mask = POLLOUT | POLLWRNORM;
267
268         if (!tun)
269                 return -EBADFD;
270
271         DBG(KERN_INFO "%s: tun_chr_poll\n", tun->dev->name);
272
273         poll_wait(file, &tun->read_wait, wait);
274
275         if (!skb_queue_empty(&tun->readq))
276                 mask |= POLLIN | POLLRDNORM;
277
278         return mask;
279 }
280
281 /* Get packet from user space buffer */
282 static __inline__ ssize_t tun_get_user(struct tun_struct *tun, struct iovec *iv, size_t count)
283 {
284         struct tun_pi pi = { 0, __constant_htons(ETH_P_IP) };
285         struct sk_buff *skb;
286         size_t len = count, align = 0;
287         struct virtio_net_hdr gso = { 0 };
288
289         if (!(tun->flags & TUN_NO_PI)) {
290                 if ((len -= sizeof(pi)) > count)
291                         return -EINVAL;
292
293                 if(memcpy_fromiovec((void *)&pi, iv, sizeof(pi)))
294                         return -EFAULT;
295         }
296
297         if (tun->flags & TUN_VNET_HDR) {
298                 if ((len -= sizeof(gso)) > count)
299                         return -EINVAL;
300
301                 if (memcpy_fromiovec((void *)&gso, iv, sizeof(gso)))
302                         return -EFAULT;
303
304                 if (gso.hdr_len > len)
305                         return -EINVAL;
306         }
307
308         if ((tun->flags & TUN_TYPE_MASK) == TUN_TAP_DEV) {
309                 align = NET_IP_ALIGN;
310                 if (unlikely(len < ETH_HLEN))
311                         return -EINVAL;
312         }
313
314         if (!(skb = alloc_skb(len + align, GFP_KERNEL))) {
315                 tun->dev->stats.rx_dropped++;
316                 return -ENOMEM;
317         }
318
319         if (align)
320                 skb_reserve(skb, align);
321         if (memcpy_fromiovec(skb_put(skb, len), iv, len)) {
322                 tun->dev->stats.rx_dropped++;
323                 kfree_skb(skb);
324                 return -EFAULT;
325         }
326
327         if (gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
328                 if (!skb_partial_csum_set(skb, gso.csum_start,
329                                           gso.csum_offset)) {
330                         tun->dev->stats.rx_frame_errors++;
331                         kfree_skb(skb);
332                         return -EINVAL;
333                 }
334         } else if (tun->flags & TUN_NOCHECKSUM)
335                 skb->ip_summed = CHECKSUM_UNNECESSARY;
336
337         switch (tun->flags & TUN_TYPE_MASK) {
338         case TUN_TUN_DEV:
339                 if (tun->flags & TUN_NO_PI) {
340                         switch (skb->data[0] & 0xf0) {
341                         case 0x40:
342                                 pi.proto = htons(ETH_P_IP);
343                                 break;
344                         case 0x60:
345                                 pi.proto = htons(ETH_P_IPV6);
346                                 break;
347                         default:
348                                 tun->dev->stats.rx_dropped++;
349                                 kfree_skb(skb);
350                                 return -EINVAL;
351                         }
352                 }
353
354                 skb_reset_mac_header(skb);
355                 skb->protocol = pi.proto;
356                 skb->dev = tun->dev;
357                 break;
358         case TUN_TAP_DEV:
359                 skb->protocol = eth_type_trans(skb, tun->dev);
360                 break;
361         };
362
363         if (gso.gso_type != VIRTIO_NET_HDR_GSO_NONE) {
364                 pr_debug("GSO!\n");
365                 switch (gso.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
366                 case VIRTIO_NET_HDR_GSO_TCPV4:
367                         skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
368                         break;
369                 case VIRTIO_NET_HDR_GSO_TCPV6:
370                         skb_shinfo(skb)->gso_type = SKB_GSO_TCPV6;
371                         break;
372                 default:
373                         tun->dev->stats.rx_frame_errors++;
374                         kfree_skb(skb);
375                         return -EINVAL;
376                 }
377
378                 if (gso.gso_type & VIRTIO_NET_HDR_GSO_ECN)
379                         skb_shinfo(skb)->gso_type |= SKB_GSO_TCP_ECN;
380
381                 skb_shinfo(skb)->gso_size = gso.gso_size;
382                 if (skb_shinfo(skb)->gso_size == 0) {
383                         tun->dev->stats.rx_frame_errors++;
384                         kfree_skb(skb);
385                         return -EINVAL;
386                 }
387
388                 /* Header must be checked, and gso_segs computed. */
389                 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
390                 skb_shinfo(skb)->gso_segs = 0;
391         }
392
393         netif_rx_ni(skb);
394         tun->dev->last_rx = jiffies;
395
396         tun->dev->stats.rx_packets++;
397         tun->dev->stats.rx_bytes += len;
398
399         return count;
400 }
401
402 static ssize_t tun_chr_aio_write(struct kiocb *iocb, const struct iovec *iv,
403                               unsigned long count, loff_t pos)
404 {
405         struct tun_struct *tun = iocb->ki_filp->private_data;
406
407         if (!tun)
408                 return -EBADFD;
409
410         DBG(KERN_INFO "%s: tun_chr_write %ld\n", tun->dev->name, count);
411
412         return tun_get_user(tun, (struct iovec *) iv, iov_length(iv, count));
413 }
414
415 /* Put packet to the user space buffer */
416 static __inline__ ssize_t tun_put_user(struct tun_struct *tun,
417                                        struct sk_buff *skb,
418                                        struct iovec *iv, int len)
419 {
420         struct tun_pi pi = { 0, skb->protocol };
421         ssize_t total = 0;
422
423         if (!(tun->flags & TUN_NO_PI)) {
424                 if ((len -= sizeof(pi)) < 0)
425                         return -EINVAL;
426
427                 if (len < skb->len) {
428                         /* Packet will be striped */
429                         pi.flags |= TUN_PKT_STRIP;
430                 }
431
432                 if (memcpy_toiovec(iv, (void *) &pi, sizeof(pi)))
433                         return -EFAULT;
434                 total += sizeof(pi);
435         }
436
437         if (tun->flags & TUN_VNET_HDR) {
438                 struct virtio_net_hdr gso = { 0 }; /* no info leak */
439                 if ((len -= sizeof(gso)) < 0)
440                         return -EINVAL;
441
442                 if (skb_is_gso(skb)) {
443                         struct skb_shared_info *sinfo = skb_shinfo(skb);
444
445                         /* This is a hint as to how much should be linear. */
446                         gso.hdr_len = skb_headlen(skb);
447                         gso.gso_size = sinfo->gso_size;
448                         if (sinfo->gso_type & SKB_GSO_TCPV4)
449                                 gso.gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
450                         else if (sinfo->gso_type & SKB_GSO_TCPV6)
451                                 gso.gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
452                         else
453                                 BUG();
454                         if (sinfo->gso_type & SKB_GSO_TCP_ECN)
455                                 gso.gso_type |= VIRTIO_NET_HDR_GSO_ECN;
456                 } else
457                         gso.gso_type = VIRTIO_NET_HDR_GSO_NONE;
458
459                 if (skb->ip_summed == CHECKSUM_PARTIAL) {
460                         gso.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
461                         gso.csum_start = skb->csum_start - skb_headroom(skb);
462                         gso.csum_offset = skb->csum_offset;
463                 } /* else everything is zero */
464
465                 if (unlikely(memcpy_toiovec(iv, (void *)&gso, sizeof(gso))))
466                         return -EFAULT;
467                 total += sizeof(gso);
468         }
469
470         len = min_t(int, skb->len, len);
471
472         skb_copy_datagram_iovec(skb, 0, iv, len);
473         total += len;
474
475         tun->dev->stats.tx_packets++;
476         tun->dev->stats.tx_bytes += len;
477
478         return total;
479 }
480
481 static ssize_t tun_chr_aio_read(struct kiocb *iocb, const struct iovec *iv,
482                             unsigned long count, loff_t pos)
483 {
484         struct file *file = iocb->ki_filp;
485         struct tun_struct *tun = file->private_data;
486         DECLARE_WAITQUEUE(wait, current);
487         struct sk_buff *skb;
488         ssize_t len, ret = 0;
489         DECLARE_MAC_BUF(mac);
490
491         if (!tun)
492                 return -EBADFD;
493
494         DBG(KERN_INFO "%s: tun_chr_read\n", tun->dev->name);
495
496         len = iov_length(iv, count);
497         if (len < 0)
498                 return -EINVAL;
499
500         add_wait_queue(&tun->read_wait, &wait);
501         while (len) {
502                 const u8 ones[ ETH_ALEN] = { 0xff, 0xff, 0xff, 0xff, 0xff, 0xff };
503                 u8 addr[ ETH_ALEN];
504                 int bit_nr;
505
506                 current->state = TASK_INTERRUPTIBLE;
507
508                 /* Read frames from the queue */
509                 if (!(skb=skb_dequeue(&tun->readq))) {
510                         if (file->f_flags & O_NONBLOCK) {
511                                 ret = -EAGAIN;
512                                 break;
513                         }
514                         if (signal_pending(current)) {
515                                 ret = -ERESTARTSYS;
516                                 break;
517                         }
518
519                         /* Nothing to read, let's sleep */
520                         schedule();
521                         continue;
522                 }
523                 netif_wake_queue(tun->dev);
524
525                 /** Decide whether to accept this packet. This code is designed to
526                  * behave identically to an Ethernet interface. Accept the packet if
527                  * - we are promiscuous.
528                  * - the packet is addressed to us.
529                  * - the packet is broadcast.
530                  * - the packet is multicast and
531                  *   - we are multicast promiscous.
532                  *   - we belong to the multicast group.
533                  */
534                 skb_copy_from_linear_data(skb, addr, min_t(size_t, sizeof addr,
535                                                                    skb->len));
536                 bit_nr = ether_crc(sizeof addr, addr) >> 26;
537                 if ((tun->if_flags & IFF_PROMISC) ||
538                                 memcmp(addr, tun->dev_addr, sizeof addr) == 0 ||
539                                 memcmp(addr, ones, sizeof addr) == 0 ||
540                                 (((addr[0] == 1 && addr[1] == 0 && addr[2] == 0x5e) ||
541                                   (addr[0] == 0x33 && addr[1] == 0x33)) &&
542                                  ((tun->if_flags & IFF_ALLMULTI) ||
543                                   (tun->chr_filter[bit_nr >> 5] & (1 << (bit_nr & 31)))))) {
544                         DBG(KERN_DEBUG "%s: tun_chr_readv: accepted: %s\n",
545                                         tun->dev->name, print_mac(mac, addr));
546                         ret = tun_put_user(tun, skb, (struct iovec *) iv, len);
547                         kfree_skb(skb);
548                         break;
549                 } else {
550                         DBG(KERN_DEBUG "%s: tun_chr_readv: rejected: %s\n",
551                                         tun->dev->name, print_mac(mac, addr));
552                         kfree_skb(skb);
553                         continue;
554                 }
555         }
556
557         current->state = TASK_RUNNING;
558         remove_wait_queue(&tun->read_wait, &wait);
559
560         return ret;
561 }
562
563 static void tun_setup(struct net_device *dev)
564 {
565         struct tun_struct *tun = netdev_priv(dev);
566
567         skb_queue_head_init(&tun->readq);
568         init_waitqueue_head(&tun->read_wait);
569
570         tun->owner = -1;
571         tun->group = -1;
572
573         dev->open = tun_net_open;
574         dev->hard_start_xmit = tun_net_xmit;
575         dev->stop = tun_net_close;
576         dev->ethtool_ops = &tun_ethtool_ops;
577         dev->destructor = free_netdev;
578         dev->features |= NETIF_F_NETNS_LOCAL;
579 }
580
581 static struct tun_struct *tun_get_by_name(struct tun_net *tn, const char *name)
582 {
583         struct tun_struct *tun;
584
585         ASSERT_RTNL();
586         list_for_each_entry(tun, &tn->dev_list, list) {
587                 if (!strncmp(tun->dev->name, name, IFNAMSIZ))
588                     return tun;
589         }
590
591         return NULL;
592 }
593
594 static int tun_set_iff(struct net *net, struct file *file, struct ifreq *ifr)
595 {
596         struct tun_net *tn;
597         struct tun_struct *tun;
598         struct net_device *dev;
599         int err;
600
601         tn = net_generic(net, tun_net_id);
602         tun = tun_get_by_name(tn, ifr->ifr_name);
603         if (tun) {
604                 if (tun->attached)
605                         return -EBUSY;
606
607                 /* Check permissions */
608                 if (((tun->owner != -1 &&
609                       current->euid != tun->owner) ||
610                      (tun->group != -1 &&
611                       current->egid != tun->group)) &&
612                      !capable(CAP_NET_ADMIN))
613                         return -EPERM;
614         }
615         else if (__dev_get_by_name(net, ifr->ifr_name))
616                 return -EINVAL;
617         else {
618                 char *name;
619                 unsigned long flags = 0;
620
621                 err = -EINVAL;
622
623                 if (!capable(CAP_NET_ADMIN))
624                         return -EPERM;
625
626                 /* Set dev type */
627                 if (ifr->ifr_flags & IFF_TUN) {
628                         /* TUN device */
629                         flags |= TUN_TUN_DEV;
630                         name = "tun%d";
631                 } else if (ifr->ifr_flags & IFF_TAP) {
632                         /* TAP device */
633                         flags |= TUN_TAP_DEV;
634                         name = "tap%d";
635                 } else
636                         goto failed;
637
638                 if (*ifr->ifr_name)
639                         name = ifr->ifr_name;
640
641                 dev = alloc_netdev(sizeof(struct tun_struct), name,
642                                    tun_setup);
643                 if (!dev)
644                         return -ENOMEM;
645
646                 dev_net_set(dev, net);
647                 tun = netdev_priv(dev);
648                 tun->dev = dev;
649                 tun->flags = flags;
650                 /* Be promiscuous by default to maintain previous behaviour. */
651                 tun->if_flags = IFF_PROMISC;
652                 /* Generate random Ethernet address. */
653                 *(__be16 *)tun->dev_addr = htons(0x00FF);
654                 get_random_bytes(tun->dev_addr + sizeof(u16), 4);
655                 memset(tun->chr_filter, 0, sizeof tun->chr_filter);
656
657                 tun_net_init(dev);
658
659                 if (strchr(dev->name, '%')) {
660                         err = dev_alloc_name(dev, dev->name);
661                         if (err < 0)
662                                 goto err_free_dev;
663                 }
664
665                 err = register_netdevice(tun->dev);
666                 if (err < 0)
667                         goto err_free_dev;
668
669                 list_add(&tun->list, &tn->dev_list);
670         }
671
672         DBG(KERN_INFO "%s: tun_set_iff\n", tun->dev->name);
673
674         if (ifr->ifr_flags & IFF_NO_PI)
675                 tun->flags |= TUN_NO_PI;
676         else
677                 tun->flags &= ~TUN_NO_PI;
678
679         if (ifr->ifr_flags & IFF_ONE_QUEUE)
680                 tun->flags |= TUN_ONE_QUEUE;
681         else
682                 tun->flags &= ~TUN_ONE_QUEUE;
683
684         if (ifr->ifr_flags & IFF_VNET_HDR)
685                 tun->flags |= TUN_VNET_HDR;
686         else
687                 tun->flags &= ~TUN_VNET_HDR;
688
689         file->private_data = tun;
690         tun->attached = 1;
691         get_net(dev_net(tun->dev));
692
693         /* Make sure persistent devices do not get stuck in
694          * xoff state.
695          */
696         if (netif_running(tun->dev))
697                 netif_wake_queue(tun->dev);
698
699         strcpy(ifr->ifr_name, tun->dev->name);
700         return 0;
701
702  err_free_dev:
703         free_netdev(dev);
704  failed:
705         return err;
706 }
707
708 /* This is like a cut-down ethtool ops, except done via tun fd so no
709  * privs required. */
710 static int set_offload(struct net_device *dev, unsigned long arg)
711 {
712         unsigned int old_features, features;
713
714         old_features = dev->features;
715         /* Unset features, set them as we chew on the arg. */
716         features = (old_features & ~(NETIF_F_HW_CSUM|NETIF_F_SG|NETIF_F_FRAGLIST
717                                     |NETIF_F_TSO_ECN|NETIF_F_TSO|NETIF_F_TSO6));
718
719         if (arg & TUN_F_CSUM) {
720                 features |= NETIF_F_HW_CSUM|NETIF_F_SG|NETIF_F_FRAGLIST;
721                 arg &= ~TUN_F_CSUM;
722
723                 if (arg & (TUN_F_TSO4|TUN_F_TSO6)) {
724                         if (arg & TUN_F_TSO_ECN) {
725                                 features |= NETIF_F_TSO_ECN;
726                                 arg &= ~TUN_F_TSO_ECN;
727                         }
728                         if (arg & TUN_F_TSO4)
729                                 features |= NETIF_F_TSO;
730                         if (arg & TUN_F_TSO6)
731                                 features |= NETIF_F_TSO6;
732                         arg &= ~(TUN_F_TSO4|TUN_F_TSO6);
733                 }
734         }
735
736         /* This gives the user a way to test for new features in future by
737          * trying to set them. */
738         if (arg)
739                 return -EINVAL;
740
741         dev->features = features;
742         if (old_features != dev->features)
743                 netdev_features_change(dev);
744
745         return 0;
746 }
747
748 static int tun_chr_ioctl(struct inode *inode, struct file *file,
749                          unsigned int cmd, unsigned long arg)
750 {
751         struct tun_struct *tun = file->private_data;
752         void __user* argp = (void __user*)arg;
753         struct ifreq ifr;
754         DECLARE_MAC_BUF(mac);
755
756         if (cmd == TUNSETIFF || _IOC_TYPE(cmd) == 0x89)
757                 if (copy_from_user(&ifr, argp, sizeof ifr))
758                         return -EFAULT;
759
760         if (cmd == TUNSETIFF && !tun) {
761                 int err;
762
763                 ifr.ifr_name[IFNAMSIZ-1] = '\0';
764
765                 rtnl_lock();
766                 err = tun_set_iff(current->nsproxy->net_ns, file, &ifr);
767                 rtnl_unlock();
768
769                 if (err)
770                         return err;
771
772                 if (copy_to_user(argp, &ifr, sizeof(ifr)))
773                         return -EFAULT;
774                 return 0;
775         }
776
777         if (cmd == TUNGETFEATURES) {
778                 /* Currently this just means: "what IFF flags are valid?".
779                  * This is needed because we never checked for invalid flags on
780                  * TUNSETIFF. */
781                 return put_user(IFF_TUN | IFF_TAP | IFF_NO_PI | IFF_ONE_QUEUE |
782                                 IFF_VNET_HDR,
783                                 (unsigned int __user*)argp);
784         }
785
786         if (!tun)
787                 return -EBADFD;
788
789         DBG(KERN_INFO "%s: tun_chr_ioctl cmd %d\n", tun->dev->name, cmd);
790
791         switch (cmd) {
792         case TUNSETNOCSUM:
793                 /* Disable/Enable checksum */
794                 if (arg)
795                         tun->flags |= TUN_NOCHECKSUM;
796                 else
797                         tun->flags &= ~TUN_NOCHECKSUM;
798
799                 DBG(KERN_INFO "%s: checksum %s\n",
800                     tun->dev->name, arg ? "disabled" : "enabled");
801                 break;
802
803         case TUNSETPERSIST:
804                 /* Disable/Enable persist mode */
805                 if (arg)
806                         tun->flags |= TUN_PERSIST;
807                 else
808                         tun->flags &= ~TUN_PERSIST;
809
810                 DBG(KERN_INFO "%s: persist %s\n",
811                     tun->dev->name, arg ? "enabled" : "disabled");
812                 break;
813
814         case TUNSETOWNER:
815                 /* Set owner of the device */
816                 tun->owner = (uid_t) arg;
817
818                 DBG(KERN_INFO "%s: owner set to %d\n", tun->dev->name, tun->owner);
819                 break;
820
821         case TUNSETGROUP:
822                 /* Set group of the device */
823                 tun->group= (gid_t) arg;
824
825                 DBG(KERN_INFO "%s: group set to %d\n", tun->dev->name, tun->group);
826                 break;
827
828         case TUNSETLINK:
829         {
830                 int ret;
831
832                 /* Only allow setting the type when the interface is down */
833                 rtnl_lock();
834                 if (tun->dev->flags & IFF_UP) {
835                         DBG(KERN_INFO "%s: Linktype set failed because interface is up\n",
836                                 tun->dev->name);
837                         ret = -EBUSY;
838                 } else {
839                         tun->dev->type = (int) arg;
840                         DBG(KERN_INFO "%s: linktype set to %d\n", tun->dev->name, tun->dev->type);
841                         ret = 0;
842                 }
843                 rtnl_unlock();
844                 return ret;
845         }
846
847 #ifdef TUN_DEBUG
848         case TUNSETDEBUG:
849                 tun->debug = arg;
850                 break;
851 #endif
852
853         case TUNSETOFFLOAD:
854         {
855                 int ret;
856                 rtnl_lock();
857                 ret = set_offload(tun->dev, arg);
858                 rtnl_unlock();
859                 return ret;
860         }
861
862         case SIOCGIFFLAGS:
863                 ifr.ifr_flags = tun->if_flags;
864                 if (copy_to_user( argp, &ifr, sizeof ifr))
865                         return -EFAULT;
866                 return 0;
867
868         case SIOCSIFFLAGS:
869                 /** Set the character device's interface flags. Currently only
870                  * IFF_PROMISC and IFF_ALLMULTI are used. */
871                 tun->if_flags = ifr.ifr_flags;
872                 DBG(KERN_INFO "%s: interface flags 0x%lx\n",
873                                 tun->dev->name, tun->if_flags);
874                 return 0;
875
876         case SIOCGIFHWADDR:
877                 /* Note: the actual net device's address may be different */
878                 memcpy(ifr.ifr_hwaddr.sa_data, tun->dev_addr,
879                                 min(sizeof ifr.ifr_hwaddr.sa_data, sizeof tun->dev_addr));
880                 if (copy_to_user( argp, &ifr, sizeof ifr))
881                         return -EFAULT;
882                 return 0;
883
884         case SIOCSIFHWADDR:
885         {
886                 /* try to set the actual net device's hw address */
887                 int ret;
888
889                 rtnl_lock();
890                 ret = dev_set_mac_address(tun->dev, &ifr.ifr_hwaddr);
891                 rtnl_unlock();
892
893                 if (ret == 0) {
894                         /** Set the character device's hardware address. This is used when
895                          * filtering packets being sent from the network device to the character
896                          * device. */
897                         memcpy(tun->dev_addr, ifr.ifr_hwaddr.sa_data,
898                                         min(sizeof ifr.ifr_hwaddr.sa_data, sizeof tun->dev_addr));
899                         DBG(KERN_DEBUG "%s: set hardware address: %x:%x:%x:%x:%x:%x\n",
900                                         tun->dev->name,
901                                         tun->dev_addr[0], tun->dev_addr[1], tun->dev_addr[2],
902                                         tun->dev_addr[3], tun->dev_addr[4], tun->dev_addr[5]);
903                 }
904
905                 return  ret;
906         }
907
908         case SIOCADDMULTI:
909                 /** Add the specified group to the character device's multicast filter
910                  * list. */
911                 rtnl_lock();
912                 netif_tx_lock_bh(tun->dev);
913                 add_multi(tun->chr_filter, ifr.ifr_hwaddr.sa_data);
914                 netif_tx_unlock_bh(tun->dev);
915                 rtnl_unlock();
916
917                 DBG(KERN_DEBUG "%s: add multi: %s\n",
918                     tun->dev->name, print_mac(mac, ifr.ifr_hwaddr.sa_data));
919                 return 0;
920
921         case SIOCDELMULTI:
922                 /** Remove the specified group from the character device's multicast
923                  * filter list. */
924                 rtnl_lock();
925                 netif_tx_lock_bh(tun->dev);
926                 del_multi(tun->chr_filter, ifr.ifr_hwaddr.sa_data);
927                 netif_tx_unlock_bh(tun->dev);
928                 rtnl_unlock();
929
930                 DBG(KERN_DEBUG "%s: del multi: %s\n",
931                     tun->dev->name, print_mac(mac, ifr.ifr_hwaddr.sa_data));
932                 return 0;
933
934         default:
935                 return -EINVAL;
936         };
937
938         return 0;
939 }
940
941 static int tun_chr_fasync(int fd, struct file *file, int on)
942 {
943         struct tun_struct *tun = file->private_data;
944         int ret;
945
946         if (!tun)
947                 return -EBADFD;
948
949         DBG(KERN_INFO "%s: tun_chr_fasync %d\n", tun->dev->name, on);
950
951         if ((ret = fasync_helper(fd, file, on, &tun->fasync)) < 0)
952                 return ret;
953
954         if (on) {
955                 ret = __f_setown(file, task_pid(current), PIDTYPE_PID, 0);
956                 if (ret)
957                         return ret;
958                 tun->flags |= TUN_FASYNC;
959         } else
960                 tun->flags &= ~TUN_FASYNC;
961
962         return 0;
963 }
964
965 static int tun_chr_open(struct inode *inode, struct file * file)
966 {
967         DBG1(KERN_INFO "tunX: tun_chr_open\n");
968         file->private_data = NULL;
969         return 0;
970 }
971
972 static int tun_chr_close(struct inode *inode, struct file *file)
973 {
974         struct tun_struct *tun = file->private_data;
975
976         if (!tun)
977                 return 0;
978
979         DBG(KERN_INFO "%s: tun_chr_close\n", tun->dev->name);
980
981         tun_chr_fasync(-1, file, 0);
982
983         rtnl_lock();
984
985         /* Detach from net device */
986         file->private_data = NULL;
987         tun->attached = 0;
988         put_net(dev_net(tun->dev));
989
990         /* Drop read queue */
991         skb_queue_purge(&tun->readq);
992
993         if (!(tun->flags & TUN_PERSIST)) {
994                 list_del(&tun->list);
995                 unregister_netdevice(tun->dev);
996         }
997
998         rtnl_unlock();
999
1000         return 0;
1001 }
1002
1003 static const struct file_operations tun_fops = {
1004         .owner  = THIS_MODULE,
1005         .llseek = no_llseek,
1006         .read  = do_sync_read,
1007         .aio_read  = tun_chr_aio_read,
1008         .write = do_sync_write,
1009         .aio_write = tun_chr_aio_write,
1010         .poll   = tun_chr_poll,
1011         .ioctl  = tun_chr_ioctl,
1012         .open   = tun_chr_open,
1013         .release = tun_chr_close,
1014         .fasync = tun_chr_fasync
1015 };
1016
1017 static struct miscdevice tun_miscdev = {
1018         .minor = TUN_MINOR,
1019         .name = "tun",
1020         .fops = &tun_fops,
1021 };
1022
1023 /* ethtool interface */
1024
1025 static int tun_get_settings(struct net_device *dev, struct ethtool_cmd *cmd)
1026 {
1027         cmd->supported          = 0;
1028         cmd->advertising        = 0;
1029         cmd->speed              = SPEED_10;
1030         cmd->duplex             = DUPLEX_FULL;
1031         cmd->port               = PORT_TP;
1032         cmd->phy_address        = 0;
1033         cmd->transceiver        = XCVR_INTERNAL;
1034         cmd->autoneg            = AUTONEG_DISABLE;
1035         cmd->maxtxpkt           = 0;
1036         cmd->maxrxpkt           = 0;
1037         return 0;
1038 }
1039
1040 static void tun_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
1041 {
1042         struct tun_struct *tun = netdev_priv(dev);
1043
1044         strcpy(info->driver, DRV_NAME);
1045         strcpy(info->version, DRV_VERSION);
1046         strcpy(info->fw_version, "N/A");
1047
1048         switch (tun->flags & TUN_TYPE_MASK) {
1049         case TUN_TUN_DEV:
1050                 strcpy(info->bus_info, "tun");
1051                 break;
1052         case TUN_TAP_DEV:
1053                 strcpy(info->bus_info, "tap");
1054                 break;
1055         }
1056 }
1057
1058 static u32 tun_get_msglevel(struct net_device *dev)
1059 {
1060 #ifdef TUN_DEBUG
1061         struct tun_struct *tun = netdev_priv(dev);
1062         return tun->debug;
1063 #else
1064         return -EOPNOTSUPP;
1065 #endif
1066 }
1067
1068 static void tun_set_msglevel(struct net_device *dev, u32 value)
1069 {
1070 #ifdef TUN_DEBUG
1071         struct tun_struct *tun = netdev_priv(dev);
1072         tun->debug = value;
1073 #endif
1074 }
1075
1076 static u32 tun_get_link(struct net_device *dev)
1077 {
1078         struct tun_struct *tun = netdev_priv(dev);
1079         return tun->attached;
1080 }
1081
1082 static u32 tun_get_rx_csum(struct net_device *dev)
1083 {
1084         struct tun_struct *tun = netdev_priv(dev);
1085         return (tun->flags & TUN_NOCHECKSUM) == 0;
1086 }
1087
1088 static int tun_set_rx_csum(struct net_device *dev, u32 data)
1089 {
1090         struct tun_struct *tun = netdev_priv(dev);
1091         if (data)
1092                 tun->flags &= ~TUN_NOCHECKSUM;
1093         else
1094                 tun->flags |= TUN_NOCHECKSUM;
1095         return 0;
1096 }
1097
1098 static const struct ethtool_ops tun_ethtool_ops = {
1099         .get_settings   = tun_get_settings,
1100         .get_drvinfo    = tun_get_drvinfo,
1101         .get_msglevel   = tun_get_msglevel,
1102         .set_msglevel   = tun_set_msglevel,
1103         .get_link       = tun_get_link,
1104         .get_rx_csum    = tun_get_rx_csum,
1105         .set_rx_csum    = tun_set_rx_csum
1106 };
1107
1108 static int tun_init_net(struct net *net)
1109 {
1110         struct tun_net *tn;
1111
1112         tn = kmalloc(sizeof(*tn), GFP_KERNEL);
1113         if (tn == NULL)
1114                 return -ENOMEM;
1115
1116         INIT_LIST_HEAD(&tn->dev_list);
1117
1118         if (net_assign_generic(net, tun_net_id, tn)) {
1119                 kfree(tn);
1120                 return -ENOMEM;
1121         }
1122
1123         return 0;
1124 }
1125
1126 static void tun_exit_net(struct net *net)
1127 {
1128         struct tun_net *tn;
1129         struct tun_struct *tun, *nxt;
1130
1131         tn = net_generic(net, tun_net_id);
1132
1133         rtnl_lock();
1134         list_for_each_entry_safe(tun, nxt, &tn->dev_list, list) {
1135                 DBG(KERN_INFO "%s cleaned up\n", tun->dev->name);
1136                 unregister_netdevice(tun->dev);
1137         }
1138         rtnl_unlock();
1139
1140         kfree(tn);
1141 }
1142
1143 static struct pernet_operations tun_net_ops = {
1144         .init = tun_init_net,
1145         .exit = tun_exit_net,
1146 };
1147
1148 static int __init tun_init(void)
1149 {
1150         int ret = 0;
1151
1152         printk(KERN_INFO "tun: %s, %s\n", DRV_DESCRIPTION, DRV_VERSION);
1153         printk(KERN_INFO "tun: %s\n", DRV_COPYRIGHT);
1154
1155         ret = register_pernet_gen_device(&tun_net_id, &tun_net_ops);
1156         if (ret) {
1157                 printk(KERN_ERR "tun: Can't register pernet ops\n");
1158                 goto err_pernet;
1159         }
1160
1161         ret = misc_register(&tun_miscdev);
1162         if (ret) {
1163                 printk(KERN_ERR "tun: Can't register misc device %d\n", TUN_MINOR);
1164                 goto err_misc;
1165         }
1166         return 0;
1167
1168 err_misc:
1169         unregister_pernet_gen_device(tun_net_id, &tun_net_ops);
1170 err_pernet:
1171         return ret;
1172 }
1173
1174 static void tun_cleanup(void)
1175 {
1176         misc_deregister(&tun_miscdev);
1177         unregister_pernet_gen_device(tun_net_id, &tun_net_ops);
1178 }
1179
1180 module_init(tun_init);
1181 module_exit(tun_cleanup);
1182 MODULE_DESCRIPTION(DRV_DESCRIPTION);
1183 MODULE_AUTHOR(DRV_COPYRIGHT);
1184 MODULE_LICENSE("GPL");
1185 MODULE_ALIAS_MISCDEV(TUN_MINOR);