drivers/net, ipv6: Select IPv6 fragment idents for virtio UFO packets
[pandora-kernel.git] / 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  *  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/ipv6.h>
68 #include <net/net_namespace.h>
69 #include <net/netns/generic.h>
70 #include <net/rtnetlink.h>
71 #include <net/sock.h>
72
73 #include <asm/system.h>
74 #include <asm/uaccess.h>
75
76 /* Uncomment to enable debugging */
77 /* #define TUN_DEBUG 1 */
78
79 #ifdef TUN_DEBUG
80 static int debug;
81
82 #define tun_debug(level, tun, fmt, args...)                     \
83 do {                                                            \
84         if (tun->debug)                                         \
85                 netdev_printk(level, tun->dev, fmt, ##args);    \
86 } while (0)
87 #define DBG1(level, fmt, args...)                               \
88 do {                                                            \
89         if (debug == 2)                                         \
90                 printk(level fmt, ##args);                      \
91 } while (0)
92 #else
93 #define tun_debug(level, tun, fmt, args...)                     \
94 do {                                                            \
95         if (0)                                                  \
96                 netdev_printk(level, tun->dev, fmt, ##args);    \
97 } while (0)
98 #define DBG1(level, fmt, args...)                               \
99 do {                                                            \
100         if (0)                                                  \
101                 printk(level fmt, ##args);                      \
102 } while (0)
103 #endif
104
105 #define FLT_EXACT_COUNT 8
106 struct tap_filter {
107         unsigned int    count;    /* Number of addrs. Zero means disabled */
108         u32             mask[2];  /* Mask of the hashed addrs */
109         unsigned char   addr[FLT_EXACT_COUNT][ETH_ALEN];
110 };
111
112 struct tun_file {
113         atomic_t count;
114         struct tun_struct *tun;
115         struct net *net;
116 };
117
118 struct tun_sock;
119
120 struct tun_struct {
121         struct tun_file         *tfile;
122         unsigned int            flags;
123         uid_t                   owner;
124         gid_t                   group;
125
126         struct net_device       *dev;
127         u32                     set_features;
128 #define TUN_USER_FEATURES (NETIF_F_HW_CSUM|NETIF_F_TSO_ECN|NETIF_F_TSO| \
129                           NETIF_F_TSO6|NETIF_F_UFO)
130         struct fasync_struct    *fasync;
131
132         struct tap_filter       txflt;
133         struct socket           socket;
134         struct socket_wq        wq;
135
136         int                     vnet_hdr_sz;
137
138 #ifdef TUN_DEBUG
139         int debug;
140 #endif
141 };
142
143 struct tun_sock {
144         struct sock             sk;
145         struct tun_struct       *tun;
146 };
147
148 static inline struct tun_sock *tun_sk(struct sock *sk)
149 {
150         return container_of(sk, struct tun_sock, sk);
151 }
152
153 static int tun_attach(struct tun_struct *tun, struct file *file)
154 {
155         struct tun_file *tfile = file->private_data;
156         int err;
157
158         ASSERT_RTNL();
159
160         netif_tx_lock_bh(tun->dev);
161
162         err = -EINVAL;
163         if (tfile->tun)
164                 goto out;
165
166         err = -EBUSY;
167         if (tun->tfile)
168                 goto out;
169
170         err = 0;
171         tfile->tun = tun;
172         tun->tfile = tfile;
173         tun->socket.file = file;
174         netif_carrier_on(tun->dev);
175         dev_hold(tun->dev);
176         sock_hold(tun->socket.sk);
177         atomic_inc(&tfile->count);
178
179 out:
180         netif_tx_unlock_bh(tun->dev);
181         return err;
182 }
183
184 static void __tun_detach(struct tun_struct *tun)
185 {
186         /* Detach from net device */
187         netif_tx_lock_bh(tun->dev);
188         netif_carrier_off(tun->dev);
189         tun->tfile = NULL;
190         tun->socket.file = NULL;
191         netif_tx_unlock_bh(tun->dev);
192
193         /* Drop read queue */
194         skb_queue_purge(&tun->socket.sk->sk_receive_queue);
195
196         /* Drop the extra count on the net device */
197         dev_put(tun->dev);
198 }
199
200 static void tun_detach(struct tun_struct *tun)
201 {
202         rtnl_lock();
203         __tun_detach(tun);
204         rtnl_unlock();
205 }
206
207 static struct tun_struct *__tun_get(struct tun_file *tfile)
208 {
209         struct tun_struct *tun = NULL;
210
211         if (atomic_inc_not_zero(&tfile->count))
212                 tun = tfile->tun;
213
214         return tun;
215 }
216
217 static struct tun_struct *tun_get(struct file *file)
218 {
219         return __tun_get(file->private_data);
220 }
221
222 static void tun_put(struct tun_struct *tun)
223 {
224         struct tun_file *tfile = tun->tfile;
225
226         if (atomic_dec_and_test(&tfile->count))
227                 tun_detach(tfile->tun);
228 }
229
230 /* TAP filtering */
231 static void addr_hash_set(u32 *mask, const u8 *addr)
232 {
233         int n = ether_crc(ETH_ALEN, addr) >> 26;
234         mask[n >> 5] |= (1 << (n & 31));
235 }
236
237 static unsigned int addr_hash_test(const u32 *mask, const u8 *addr)
238 {
239         int n = ether_crc(ETH_ALEN, addr) >> 26;
240         return mask[n >> 5] & (1 << (n & 31));
241 }
242
243 static int update_filter(struct tap_filter *filter, void __user *arg)
244 {
245         struct { u8 u[ETH_ALEN]; } *addr;
246         struct tun_filter uf;
247         int err, alen, n, nexact;
248
249         if (copy_from_user(&uf, arg, sizeof(uf)))
250                 return -EFAULT;
251
252         if (!uf.count) {
253                 /* Disabled */
254                 filter->count = 0;
255                 return 0;
256         }
257
258         alen = ETH_ALEN * uf.count;
259         addr = kmalloc(alen, GFP_KERNEL);
260         if (!addr)
261                 return -ENOMEM;
262
263         if (copy_from_user(addr, arg + sizeof(uf), alen)) {
264                 err = -EFAULT;
265                 goto done;
266         }
267
268         /* The filter is updated without holding any locks. Which is
269          * perfectly safe. We disable it first and in the worst
270          * case we'll accept a few undesired packets. */
271         filter->count = 0;
272         wmb();
273
274         /* Use first set of addresses as an exact filter */
275         for (n = 0; n < uf.count && n < FLT_EXACT_COUNT; n++)
276                 memcpy(filter->addr[n], addr[n].u, ETH_ALEN);
277
278         nexact = n;
279
280         /* Remaining multicast addresses are hashed,
281          * unicast will leave the filter disabled. */
282         memset(filter->mask, 0, sizeof(filter->mask));
283         for (; n < uf.count; n++) {
284                 if (!is_multicast_ether_addr(addr[n].u)) {
285                         err = 0; /* no filter */
286                         goto done;
287                 }
288                 addr_hash_set(filter->mask, addr[n].u);
289         }
290
291         /* For ALLMULTI just set the mask to all ones.
292          * This overrides the mask populated above. */
293         if ((uf.flags & TUN_FLT_ALLMULTI))
294                 memset(filter->mask, ~0, sizeof(filter->mask));
295
296         /* Now enable the filter */
297         wmb();
298         filter->count = nexact;
299
300         /* Return the number of exact filters */
301         err = nexact;
302
303 done:
304         kfree(addr);
305         return err;
306 }
307
308 /* Returns: 0 - drop, !=0 - accept */
309 static int run_filter(struct tap_filter *filter, const struct sk_buff *skb)
310 {
311         /* Cannot use eth_hdr(skb) here because skb_mac_hdr() is incorrect
312          * at this point. */
313         struct ethhdr *eh = (struct ethhdr *) skb->data;
314         int i;
315
316         /* Exact match */
317         for (i = 0; i < filter->count; i++)
318                 if (!compare_ether_addr(eh->h_dest, filter->addr[i]))
319                         return 1;
320
321         /* Inexact match (multicast only) */
322         if (is_multicast_ether_addr(eh->h_dest))
323                 return addr_hash_test(filter->mask, eh->h_dest);
324
325         return 0;
326 }
327
328 /*
329  * Checks whether the packet is accepted or not.
330  * Returns: 0 - drop, !=0 - accept
331  */
332 static int check_filter(struct tap_filter *filter, const struct sk_buff *skb)
333 {
334         if (!filter->count)
335                 return 1;
336
337         return run_filter(filter, skb);
338 }
339
340 /* Network device part of the driver */
341
342 static const struct ethtool_ops tun_ethtool_ops;
343
344 /* Net device detach from fd. */
345 static void tun_net_uninit(struct net_device *dev)
346 {
347         struct tun_struct *tun = netdev_priv(dev);
348         struct tun_file *tfile = tun->tfile;
349
350         /* Inform the methods they need to stop using the dev.
351          */
352         if (tfile) {
353                 wake_up_all(&tun->wq.wait);
354                 if (atomic_dec_and_test(&tfile->count))
355                         __tun_detach(tun);
356         }
357 }
358
359 static void tun_free_netdev(struct net_device *dev)
360 {
361         struct tun_struct *tun = netdev_priv(dev);
362
363         sock_put(tun->socket.sk);
364 }
365
366 /* Net device open. */
367 static int tun_net_open(struct net_device *dev)
368 {
369         netif_start_queue(dev);
370         return 0;
371 }
372
373 /* Net device close. */
374 static int tun_net_close(struct net_device *dev)
375 {
376         netif_stop_queue(dev);
377         return 0;
378 }
379
380 /* Net device start xmit */
381 static netdev_tx_t tun_net_xmit(struct sk_buff *skb, struct net_device *dev)
382 {
383         struct tun_struct *tun = netdev_priv(dev);
384
385         tun_debug(KERN_INFO, tun, "tun_net_xmit %d\n", skb->len);
386
387         /* Drop packet if interface is not attached */
388         if (!tun->tfile)
389                 goto drop;
390
391         /* Drop if the filter does not like it.
392          * This is a noop if the filter is disabled.
393          * Filter can be enabled only for the TAP devices. */
394         if (!check_filter(&tun->txflt, skb))
395                 goto drop;
396
397         if (tun->socket.sk->sk_filter &&
398             sk_filter(tun->socket.sk, skb))
399                 goto drop;
400
401         if (skb_queue_len(&tun->socket.sk->sk_receive_queue) >= dev->tx_queue_len) {
402                 if (!(tun->flags & TUN_ONE_QUEUE)) {
403                         /* Normal queueing mode. */
404                         /* Packet scheduler handles dropping of further packets. */
405                         netif_stop_queue(dev);
406
407                         /* We won't see all dropped packets individually, so overrun
408                          * error is more appropriate. */
409                         dev->stats.tx_fifo_errors++;
410                 } else {
411                         /* Single queue mode.
412                          * Driver handles dropping of all packets itself. */
413                         goto drop;
414                 }
415         }
416
417         /* Orphan the skb - required as we might hang on to it
418          * for indefinite time. */
419         skb_orphan(skb);
420
421         nf_reset(skb);
422
423         /* Enqueue packet */
424         skb_queue_tail(&tun->socket.sk->sk_receive_queue, skb);
425
426         /* Notify and wake up reader process */
427         if (tun->flags & TUN_FASYNC)
428                 kill_fasync(&tun->fasync, SIGIO, POLL_IN);
429         wake_up_interruptible_poll(&tun->wq.wait, POLLIN |
430                                    POLLRDNORM | POLLRDBAND);
431         return NETDEV_TX_OK;
432
433 drop:
434         dev->stats.tx_dropped++;
435         kfree_skb(skb);
436         return NETDEV_TX_OK;
437 }
438
439 static void tun_net_mclist(struct net_device *dev)
440 {
441         /*
442          * This callback is supposed to deal with mc filter in
443          * _rx_ path and has nothing to do with the _tx_ path.
444          * In rx path we always accept everything userspace gives us.
445          */
446 }
447
448 #define MIN_MTU 68
449 #define MAX_MTU 65535
450
451 static int
452 tun_net_change_mtu(struct net_device *dev, int new_mtu)
453 {
454         if (new_mtu < MIN_MTU || new_mtu + dev->hard_header_len > MAX_MTU)
455                 return -EINVAL;
456         dev->mtu = new_mtu;
457         return 0;
458 }
459
460 static u32 tun_net_fix_features(struct net_device *dev, u32 features)
461 {
462         struct tun_struct *tun = netdev_priv(dev);
463
464         return (features & tun->set_features) | (features & ~TUN_USER_FEATURES);
465 }
466 #ifdef CONFIG_NET_POLL_CONTROLLER
467 static void tun_poll_controller(struct net_device *dev)
468 {
469         /*
470          * Tun only receives frames when:
471          * 1) the char device endpoint gets data from user space
472          * 2) the tun socket gets a sendmsg call from user space
473          * Since both of those are syncronous operations, we are guaranteed
474          * never to have pending data when we poll for it
475          * so theres nothing to do here but return.
476          * We need this though so netpoll recognizes us as an interface that
477          * supports polling, which enables bridge devices in virt setups to
478          * still use netconsole
479          */
480         return;
481 }
482 #endif
483 static const struct net_device_ops tun_netdev_ops = {
484         .ndo_uninit             = tun_net_uninit,
485         .ndo_open               = tun_net_open,
486         .ndo_stop               = tun_net_close,
487         .ndo_start_xmit         = tun_net_xmit,
488         .ndo_change_mtu         = tun_net_change_mtu,
489         .ndo_fix_features       = tun_net_fix_features,
490 #ifdef CONFIG_NET_POLL_CONTROLLER
491         .ndo_poll_controller    = tun_poll_controller,
492 #endif
493 };
494
495 static const struct net_device_ops tap_netdev_ops = {
496         .ndo_uninit             = tun_net_uninit,
497         .ndo_open               = tun_net_open,
498         .ndo_stop               = tun_net_close,
499         .ndo_start_xmit         = tun_net_xmit,
500         .ndo_change_mtu         = tun_net_change_mtu,
501         .ndo_fix_features       = tun_net_fix_features,
502         .ndo_set_rx_mode        = tun_net_mclist,
503         .ndo_set_mac_address    = eth_mac_addr,
504         .ndo_validate_addr      = eth_validate_addr,
505 #ifdef CONFIG_NET_POLL_CONTROLLER
506         .ndo_poll_controller    = tun_poll_controller,
507 #endif
508 };
509
510 /* Initialize net device. */
511 static void tun_net_init(struct net_device *dev)
512 {
513         struct tun_struct *tun = netdev_priv(dev);
514
515         switch (tun->flags & TUN_TYPE_MASK) {
516         case TUN_TUN_DEV:
517                 dev->netdev_ops = &tun_netdev_ops;
518
519                 /* Point-to-Point TUN Device */
520                 dev->hard_header_len = 0;
521                 dev->addr_len = 0;
522                 dev->mtu = 1500;
523
524                 /* Zero header length */
525                 dev->type = ARPHRD_NONE;
526                 dev->flags = IFF_POINTOPOINT | IFF_NOARP | IFF_MULTICAST;
527                 dev->tx_queue_len = TUN_READQ_SIZE;  /* We prefer our own queue length */
528                 break;
529
530         case TUN_TAP_DEV:
531                 dev->netdev_ops = &tap_netdev_ops;
532                 /* Ethernet TAP Device */
533                 ether_setup(dev);
534                 dev->priv_flags &= ~IFF_TX_SKB_SHARING;
535
536                 random_ether_addr(dev->dev_addr);
537
538                 dev->tx_queue_len = TUN_READQ_SIZE;  /* We prefer our own queue length */
539                 break;
540         }
541 }
542
543 /* Character device part */
544
545 /* Poll */
546 static unsigned int tun_chr_poll(struct file *file, poll_table * wait)
547 {
548         struct tun_file *tfile = file->private_data;
549         struct tun_struct *tun = __tun_get(tfile);
550         struct sock *sk;
551         unsigned int mask = 0;
552
553         if (!tun)
554                 return POLLERR;
555
556         sk = tun->socket.sk;
557
558         tun_debug(KERN_INFO, tun, "tun_chr_poll\n");
559
560         poll_wait(file, &tun->wq.wait, wait);
561
562         if (!skb_queue_empty(&sk->sk_receive_queue))
563                 mask |= POLLIN | POLLRDNORM;
564
565         if (sock_writeable(sk) ||
566             (!test_and_set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags) &&
567              sock_writeable(sk)))
568                 mask |= POLLOUT | POLLWRNORM;
569
570         if (tun->dev->reg_state != NETREG_REGISTERED)
571                 mask = POLLERR;
572
573         tun_put(tun);
574         return mask;
575 }
576
577 /* prepad is the amount to reserve at front.  len is length after that.
578  * linear is a hint as to how much to copy (usually headers). */
579 static struct sk_buff *tun_alloc_skb(struct tun_struct *tun,
580                                      size_t prepad, size_t len,
581                                      size_t linear, int noblock)
582 {
583         struct sock *sk = tun->socket.sk;
584         struct sk_buff *skb;
585         int err;
586
587         sock_update_classid(sk);
588
589         /* Under a page?  Don't bother with paged skb. */
590         if (prepad + len < PAGE_SIZE || !linear)
591                 linear = len;
592
593         skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
594                                    &err);
595         if (!skb)
596                 return ERR_PTR(err);
597
598         skb_reserve(skb, prepad);
599         skb_put(skb, linear);
600         skb->data_len = len - linear;
601         skb->len += len - linear;
602
603         return skb;
604 }
605
606 /* Get packet from user space buffer */
607 static ssize_t tun_get_user(struct tun_struct *tun,
608                             const struct iovec *iv, size_t count,
609                             int noblock)
610 {
611         struct tun_pi pi = { 0, cpu_to_be16(ETH_P_IP) };
612         struct sk_buff *skb;
613         size_t len = count, align = NET_SKB_PAD;
614         struct virtio_net_hdr gso = { 0 };
615         int offset = 0;
616
617         if (!(tun->flags & TUN_NO_PI)) {
618                 if (len < sizeof(pi))
619                         return -EINVAL;
620                 len -= sizeof(pi);
621
622                 if (memcpy_fromiovecend((void *)&pi, iv, 0, sizeof(pi)))
623                         return -EFAULT;
624                 offset += sizeof(pi);
625         }
626
627         if (tun->flags & TUN_VNET_HDR) {
628                 if (len < tun->vnet_hdr_sz)
629                         return -EINVAL;
630                 len -= tun->vnet_hdr_sz;
631
632                 if (memcpy_fromiovecend((void *)&gso, iv, offset, sizeof(gso)))
633                         return -EFAULT;
634
635                 if ((gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
636                     gso.csum_start + gso.csum_offset + 2 > gso.hdr_len)
637                         gso.hdr_len = gso.csum_start + gso.csum_offset + 2;
638
639                 if (gso.hdr_len > len)
640                         return -EINVAL;
641                 offset += tun->vnet_hdr_sz;
642         }
643
644         if ((tun->flags & TUN_TYPE_MASK) == TUN_TAP_DEV) {
645                 align += NET_IP_ALIGN;
646                 if (unlikely(len < ETH_HLEN ||
647                              (gso.hdr_len && gso.hdr_len < ETH_HLEN)))
648                         return -EINVAL;
649         }
650
651         skb = tun_alloc_skb(tun, align, len, gso.hdr_len, noblock);
652         if (IS_ERR(skb)) {
653                 if (PTR_ERR(skb) != -EAGAIN)
654                         tun->dev->stats.rx_dropped++;
655                 return PTR_ERR(skb);
656         }
657
658         if (skb_copy_datagram_from_iovec(skb, 0, iv, offset, len)) {
659                 tun->dev->stats.rx_dropped++;
660                 kfree_skb(skb);
661                 return -EFAULT;
662         }
663
664         if (gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
665                 if (!skb_partial_csum_set(skb, gso.csum_start,
666                                           gso.csum_offset)) {
667                         tun->dev->stats.rx_frame_errors++;
668                         kfree_skb(skb);
669                         return -EINVAL;
670                 }
671         }
672
673         switch (tun->flags & TUN_TYPE_MASK) {
674         case TUN_TUN_DEV:
675                 if (tun->flags & TUN_NO_PI) {
676                         switch (skb->data[0] & 0xf0) {
677                         case 0x40:
678                                 pi.proto = htons(ETH_P_IP);
679                                 break;
680                         case 0x60:
681                                 pi.proto = htons(ETH_P_IPV6);
682                                 break;
683                         default:
684                                 tun->dev->stats.rx_dropped++;
685                                 kfree_skb(skb);
686                                 return -EINVAL;
687                         }
688                 }
689
690                 skb_reset_mac_header(skb);
691                 skb->protocol = pi.proto;
692                 skb->dev = tun->dev;
693                 break;
694         case TUN_TAP_DEV:
695                 skb->protocol = eth_type_trans(skb, tun->dev);
696                 break;
697         }
698
699         skb_reset_network_header(skb);
700
701         if (gso.gso_type != VIRTIO_NET_HDR_GSO_NONE) {
702                 pr_debug("GSO!\n");
703                 switch (gso.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
704                 case VIRTIO_NET_HDR_GSO_TCPV4:
705                         skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
706                         break;
707                 case VIRTIO_NET_HDR_GSO_TCPV6:
708                         skb_shinfo(skb)->gso_type = SKB_GSO_TCPV6;
709                         break;
710                 case VIRTIO_NET_HDR_GSO_UDP:
711                         skb_shinfo(skb)->gso_type = SKB_GSO_UDP;
712                         if (skb->protocol == htons(ETH_P_IPV6))
713                                 ipv6_proxy_select_ident(skb);
714                         break;
715                 default:
716                         tun->dev->stats.rx_frame_errors++;
717                         kfree_skb(skb);
718                         return -EINVAL;
719                 }
720
721                 if (gso.gso_type & VIRTIO_NET_HDR_GSO_ECN)
722                         skb_shinfo(skb)->gso_type |= SKB_GSO_TCP_ECN;
723
724                 skb_shinfo(skb)->gso_size = gso.gso_size;
725                 if (skb_shinfo(skb)->gso_size == 0) {
726                         tun->dev->stats.rx_frame_errors++;
727                         kfree_skb(skb);
728                         return -EINVAL;
729                 }
730
731                 /* Header must be checked, and gso_segs computed. */
732                 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
733                 skb_shinfo(skb)->gso_segs = 0;
734         }
735
736         netif_rx_ni(skb);
737
738         tun->dev->stats.rx_packets++;
739         tun->dev->stats.rx_bytes += len;
740
741         return count;
742 }
743
744 static ssize_t tun_chr_aio_write(struct kiocb *iocb, const struct iovec *iv,
745                               unsigned long count, loff_t pos)
746 {
747         struct file *file = iocb->ki_filp;
748         struct tun_struct *tun = tun_get(file);
749         ssize_t result;
750
751         if (!tun)
752                 return -EBADFD;
753
754         tun_debug(KERN_INFO, tun, "tun_chr_write %ld\n", count);
755
756         result = tun_get_user(tun, iv, iov_length(iv, count),
757                               file->f_flags & O_NONBLOCK);
758
759         tun_put(tun);
760         return result;
761 }
762
763 /* Put packet to the user space buffer */
764 static ssize_t tun_put_user(struct tun_struct *tun,
765                             struct sk_buff *skb,
766                             const struct iovec *iv, int len)
767 {
768         struct tun_pi pi = { 0, skb->protocol };
769         ssize_t total = 0;
770
771         if (!(tun->flags & TUN_NO_PI)) {
772                 if ((len -= sizeof(pi)) < 0)
773                         return -EINVAL;
774
775                 if (len < skb->len) {
776                         /* Packet will be striped */
777                         pi.flags |= TUN_PKT_STRIP;
778                 }
779
780                 if (memcpy_toiovecend(iv, (void *) &pi, 0, sizeof(pi)))
781                         return -EFAULT;
782                 total += sizeof(pi);
783         }
784
785         if (tun->flags & TUN_VNET_HDR) {
786                 struct virtio_net_hdr gso = { 0 }; /* no info leak */
787                 if ((len -= tun->vnet_hdr_sz) < 0)
788                         return -EINVAL;
789
790                 if (skb_is_gso(skb)) {
791                         struct skb_shared_info *sinfo = skb_shinfo(skb);
792
793                         /* This is a hint as to how much should be linear. */
794                         gso.hdr_len = skb_headlen(skb);
795                         gso.gso_size = sinfo->gso_size;
796                         if (sinfo->gso_type & SKB_GSO_TCPV4)
797                                 gso.gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
798                         else if (sinfo->gso_type & SKB_GSO_TCPV6)
799                                 gso.gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
800                         else if (sinfo->gso_type & SKB_GSO_UDP)
801                                 gso.gso_type = VIRTIO_NET_HDR_GSO_UDP;
802                         else {
803                                 pr_err("unexpected GSO type: "
804                                        "0x%x, gso_size %d, hdr_len %d\n",
805                                        sinfo->gso_type, gso.gso_size,
806                                        gso.hdr_len);
807                                 print_hex_dump(KERN_ERR, "tun: ",
808                                                DUMP_PREFIX_NONE,
809                                                16, 1, skb->head,
810                                                min((int)gso.hdr_len, 64), true);
811                                 WARN_ON_ONCE(1);
812                                 return -EINVAL;
813                         }
814                         if (sinfo->gso_type & SKB_GSO_TCP_ECN)
815                                 gso.gso_type |= VIRTIO_NET_HDR_GSO_ECN;
816                 } else
817                         gso.gso_type = VIRTIO_NET_HDR_GSO_NONE;
818
819                 if (skb->ip_summed == CHECKSUM_PARTIAL) {
820                         gso.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
821                         gso.csum_start = skb_checksum_start_offset(skb);
822                         gso.csum_offset = skb->csum_offset;
823                 } else if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
824                         gso.flags = VIRTIO_NET_HDR_F_DATA_VALID;
825                 } /* else everything is zero */
826
827                 if (unlikely(memcpy_toiovecend(iv, (void *)&gso, total,
828                                                sizeof(gso))))
829                         return -EFAULT;
830                 total += tun->vnet_hdr_sz;
831         }
832
833         len = min_t(int, skb->len, len);
834
835         skb_copy_datagram_const_iovec(skb, 0, iv, total, len);
836         total += skb->len;
837
838         tun->dev->stats.tx_packets++;
839         tun->dev->stats.tx_bytes += len;
840
841         return total;
842 }
843
844 static ssize_t tun_do_read(struct tun_struct *tun,
845                            struct kiocb *iocb, const struct iovec *iv,
846                            ssize_t len, int noblock)
847 {
848         DECLARE_WAITQUEUE(wait, current);
849         struct sk_buff *skb;
850         ssize_t ret = 0;
851
852         tun_debug(KERN_INFO, tun, "tun_chr_read\n");
853
854         if (unlikely(!noblock))
855                 add_wait_queue(&tun->wq.wait, &wait);
856         while (len) {
857                 current->state = TASK_INTERRUPTIBLE;
858
859                 /* Read frames from the queue */
860                 if (!(skb=skb_dequeue(&tun->socket.sk->sk_receive_queue))) {
861                         if (noblock) {
862                                 ret = -EAGAIN;
863                                 break;
864                         }
865                         if (signal_pending(current)) {
866                                 ret = -ERESTARTSYS;
867                                 break;
868                         }
869                         if (tun->dev->reg_state != NETREG_REGISTERED) {
870                                 ret = -EIO;
871                                 break;
872                         }
873
874                         /* Nothing to read, let's sleep */
875                         schedule();
876                         continue;
877                 }
878                 netif_wake_queue(tun->dev);
879
880                 ret = tun_put_user(tun, skb, iv, len);
881                 kfree_skb(skb);
882                 break;
883         }
884
885         current->state = TASK_RUNNING;
886         if (unlikely(!noblock))
887                 remove_wait_queue(&tun->wq.wait, &wait);
888
889         return ret;
890 }
891
892 static ssize_t tun_chr_aio_read(struct kiocb *iocb, const struct iovec *iv,
893                             unsigned long count, loff_t pos)
894 {
895         struct file *file = iocb->ki_filp;
896         struct tun_file *tfile = file->private_data;
897         struct tun_struct *tun = __tun_get(tfile);
898         ssize_t len, ret;
899
900         if (!tun)
901                 return -EBADFD;
902         len = iov_length(iv, count);
903         if (len < 0) {
904                 ret = -EINVAL;
905                 goto out;
906         }
907
908         ret = tun_do_read(tun, iocb, iv, len, file->f_flags & O_NONBLOCK);
909         ret = min_t(ssize_t, ret, len);
910 out:
911         tun_put(tun);
912         return ret;
913 }
914
915 static void tun_setup(struct net_device *dev)
916 {
917         struct tun_struct *tun = netdev_priv(dev);
918
919         tun->owner = -1;
920         tun->group = -1;
921
922         dev->ethtool_ops = &tun_ethtool_ops;
923         dev->destructor = tun_free_netdev;
924 }
925
926 /* Trivial set of netlink ops to allow deleting tun or tap
927  * device with netlink.
928  */
929 static int tun_validate(struct nlattr *tb[], struct nlattr *data[])
930 {
931         return -EINVAL;
932 }
933
934 static struct rtnl_link_ops tun_link_ops __read_mostly = {
935         .kind           = DRV_NAME,
936         .priv_size      = sizeof(struct tun_struct),
937         .setup          = tun_setup,
938         .validate       = tun_validate,
939 };
940
941 static void tun_sock_write_space(struct sock *sk)
942 {
943         struct tun_struct *tun;
944         wait_queue_head_t *wqueue;
945
946         if (!sock_writeable(sk))
947                 return;
948
949         if (!test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags))
950                 return;
951
952         wqueue = sk_sleep(sk);
953         if (wqueue && waitqueue_active(wqueue))
954                 wake_up_interruptible_sync_poll(wqueue, POLLOUT |
955                                                 POLLWRNORM | POLLWRBAND);
956
957         tun = tun_sk(sk)->tun;
958         kill_fasync(&tun->fasync, SIGIO, POLL_OUT);
959 }
960
961 static void tun_sock_destruct(struct sock *sk)
962 {
963         free_netdev(tun_sk(sk)->tun->dev);
964 }
965
966 static int tun_sendmsg(struct kiocb *iocb, struct socket *sock,
967                        struct msghdr *m, size_t total_len)
968 {
969         struct tun_struct *tun = container_of(sock, struct tun_struct, socket);
970         return tun_get_user(tun, m->msg_iov, total_len,
971                             m->msg_flags & MSG_DONTWAIT);
972 }
973
974 static int tun_recvmsg(struct kiocb *iocb, struct socket *sock,
975                        struct msghdr *m, size_t total_len,
976                        int flags)
977 {
978         struct tun_struct *tun = container_of(sock, struct tun_struct, socket);
979         int ret;
980         if (flags & ~(MSG_DONTWAIT|MSG_TRUNC))
981                 return -EINVAL;
982         ret = tun_do_read(tun, iocb, m->msg_iov, total_len,
983                           flags & MSG_DONTWAIT);
984         if (ret > total_len) {
985                 m->msg_flags |= MSG_TRUNC;
986                 ret = flags & MSG_TRUNC ? ret : total_len;
987         }
988         return ret;
989 }
990
991 /* Ops structure to mimic raw sockets with tun */
992 static const struct proto_ops tun_socket_ops = {
993         .sendmsg = tun_sendmsg,
994         .recvmsg = tun_recvmsg,
995 };
996
997 static struct proto tun_proto = {
998         .name           = "tun",
999         .owner          = THIS_MODULE,
1000         .obj_size       = sizeof(struct tun_sock),
1001 };
1002
1003 static int tun_flags(struct tun_struct *tun)
1004 {
1005         int flags = 0;
1006
1007         if (tun->flags & TUN_TUN_DEV)
1008                 flags |= IFF_TUN;
1009         else
1010                 flags |= IFF_TAP;
1011
1012         if (tun->flags & TUN_NO_PI)
1013                 flags |= IFF_NO_PI;
1014
1015         if (tun->flags & TUN_ONE_QUEUE)
1016                 flags |= IFF_ONE_QUEUE;
1017
1018         if (tun->flags & TUN_VNET_HDR)
1019                 flags |= IFF_VNET_HDR;
1020
1021         return flags;
1022 }
1023
1024 static ssize_t tun_show_flags(struct device *dev, struct device_attribute *attr,
1025                               char *buf)
1026 {
1027         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1028         return sprintf(buf, "0x%x\n", tun_flags(tun));
1029 }
1030
1031 static ssize_t tun_show_owner(struct device *dev, struct device_attribute *attr,
1032                               char *buf)
1033 {
1034         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1035         return sprintf(buf, "%d\n", tun->owner);
1036 }
1037
1038 static ssize_t tun_show_group(struct device *dev, struct device_attribute *attr,
1039                               char *buf)
1040 {
1041         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1042         return sprintf(buf, "%d\n", tun->group);
1043 }
1044
1045 static DEVICE_ATTR(tun_flags, 0444, tun_show_flags, NULL);
1046 static DEVICE_ATTR(owner, 0444, tun_show_owner, NULL);
1047 static DEVICE_ATTR(group, 0444, tun_show_group, NULL);
1048
1049 static int tun_set_iff(struct net *net, struct file *file, struct ifreq *ifr)
1050 {
1051         struct sock *sk;
1052         struct tun_struct *tun;
1053         struct net_device *dev;
1054         int err;
1055
1056         dev = __dev_get_by_name(net, ifr->ifr_name);
1057         if (dev) {
1058                 const struct cred *cred = current_cred();
1059
1060                 if (ifr->ifr_flags & IFF_TUN_EXCL)
1061                         return -EBUSY;
1062                 if ((ifr->ifr_flags & IFF_TUN) && dev->netdev_ops == &tun_netdev_ops)
1063                         tun = netdev_priv(dev);
1064                 else if ((ifr->ifr_flags & IFF_TAP) && dev->netdev_ops == &tap_netdev_ops)
1065                         tun = netdev_priv(dev);
1066                 else
1067                         return -EINVAL;
1068
1069                 if (((tun->owner != -1 && cred->euid != tun->owner) ||
1070                      (tun->group != -1 && !in_egroup_p(tun->group))) &&
1071                     !capable(CAP_NET_ADMIN))
1072                         return -EPERM;
1073                 err = security_tun_dev_attach(tun->socket.sk);
1074                 if (err < 0)
1075                         return err;
1076
1077                 err = tun_attach(tun, file);
1078                 if (err < 0)
1079                         return err;
1080         }
1081         else {
1082                 char *name;
1083                 unsigned long flags = 0;
1084
1085                 if (!capable(CAP_NET_ADMIN))
1086                         return -EPERM;
1087                 err = security_tun_dev_create();
1088                 if (err < 0)
1089                         return err;
1090
1091                 /* Set dev type */
1092                 if (ifr->ifr_flags & IFF_TUN) {
1093                         /* TUN device */
1094                         flags |= TUN_TUN_DEV;
1095                         name = "tun%d";
1096                 } else if (ifr->ifr_flags & IFF_TAP) {
1097                         /* TAP device */
1098                         flags |= TUN_TAP_DEV;
1099                         name = "tap%d";
1100                 } else
1101                         return -EINVAL;
1102
1103                 if (*ifr->ifr_name)
1104                         name = ifr->ifr_name;
1105
1106                 dev = alloc_netdev(sizeof(struct tun_struct), name,
1107                                    tun_setup);
1108                 if (!dev)
1109                         return -ENOMEM;
1110
1111                 dev_net_set(dev, net);
1112                 dev->rtnl_link_ops = &tun_link_ops;
1113
1114                 tun = netdev_priv(dev);
1115                 tun->dev = dev;
1116                 tun->flags = flags;
1117                 tun->txflt.count = 0;
1118                 tun->vnet_hdr_sz = sizeof(struct virtio_net_hdr);
1119
1120                 err = -ENOMEM;
1121                 sk = sk_alloc(net, AF_UNSPEC, GFP_KERNEL, &tun_proto);
1122                 if (!sk)
1123                         goto err_free_dev;
1124
1125                 tun->socket.wq = &tun->wq;
1126                 init_waitqueue_head(&tun->wq.wait);
1127                 tun->socket.ops = &tun_socket_ops;
1128                 sock_init_data(&tun->socket, sk);
1129                 sk->sk_write_space = tun_sock_write_space;
1130                 sk->sk_sndbuf = INT_MAX;
1131
1132                 tun_sk(sk)->tun = tun;
1133
1134                 security_tun_dev_post_create(sk);
1135
1136                 tun_net_init(dev);
1137
1138                 dev->hw_features = NETIF_F_SG | NETIF_F_FRAGLIST |
1139                         TUN_USER_FEATURES;
1140                 dev->features = dev->hw_features;
1141
1142                 err = register_netdevice(tun->dev);
1143                 if (err < 0)
1144                         goto err_free_sk;
1145
1146                 if (device_create_file(&tun->dev->dev, &dev_attr_tun_flags) ||
1147                     device_create_file(&tun->dev->dev, &dev_attr_owner) ||
1148                     device_create_file(&tun->dev->dev, &dev_attr_group))
1149                         pr_err("Failed to create tun sysfs files\n");
1150
1151                 sk->sk_destruct = tun_sock_destruct;
1152
1153                 err = tun_attach(tun, file);
1154                 if (err < 0)
1155                         goto failed;
1156         }
1157
1158         tun_debug(KERN_INFO, tun, "tun_set_iff\n");
1159
1160         if (ifr->ifr_flags & IFF_NO_PI)
1161                 tun->flags |= TUN_NO_PI;
1162         else
1163                 tun->flags &= ~TUN_NO_PI;
1164
1165         if (ifr->ifr_flags & IFF_ONE_QUEUE)
1166                 tun->flags |= TUN_ONE_QUEUE;
1167         else
1168                 tun->flags &= ~TUN_ONE_QUEUE;
1169
1170         if (ifr->ifr_flags & IFF_VNET_HDR)
1171                 tun->flags |= TUN_VNET_HDR;
1172         else
1173                 tun->flags &= ~TUN_VNET_HDR;
1174
1175         /* Make sure persistent devices do not get stuck in
1176          * xoff state.
1177          */
1178         if (netif_running(tun->dev))
1179                 netif_wake_queue(tun->dev);
1180
1181         strcpy(ifr->ifr_name, tun->dev->name);
1182         return 0;
1183
1184  err_free_sk:
1185         sock_put(sk);
1186  err_free_dev:
1187         free_netdev(dev);
1188  failed:
1189         return err;
1190 }
1191
1192 static int tun_get_iff(struct net *net, struct tun_struct *tun,
1193                        struct ifreq *ifr)
1194 {
1195         tun_debug(KERN_INFO, tun, "tun_get_iff\n");
1196
1197         strcpy(ifr->ifr_name, tun->dev->name);
1198
1199         ifr->ifr_flags = tun_flags(tun);
1200
1201         return 0;
1202 }
1203
1204 /* This is like a cut-down ethtool ops, except done via tun fd so no
1205  * privs required. */
1206 static int set_offload(struct tun_struct *tun, unsigned long arg)
1207 {
1208         u32 features = 0;
1209
1210         if (arg & TUN_F_CSUM) {
1211                 features |= NETIF_F_HW_CSUM;
1212                 arg &= ~TUN_F_CSUM;
1213
1214                 if (arg & (TUN_F_TSO4|TUN_F_TSO6)) {
1215                         if (arg & TUN_F_TSO_ECN) {
1216                                 features |= NETIF_F_TSO_ECN;
1217                                 arg &= ~TUN_F_TSO_ECN;
1218                         }
1219                         if (arg & TUN_F_TSO4)
1220                                 features |= NETIF_F_TSO;
1221                         if (arg & TUN_F_TSO6)
1222                                 features |= NETIF_F_TSO6;
1223                         arg &= ~(TUN_F_TSO4|TUN_F_TSO6);
1224                 }
1225
1226                 if (arg & TUN_F_UFO) {
1227                         features |= NETIF_F_UFO;
1228                         arg &= ~TUN_F_UFO;
1229                 }
1230         }
1231
1232         /* This gives the user a way to test for new features in future by
1233          * trying to set them. */
1234         if (arg)
1235                 return -EINVAL;
1236
1237         tun->set_features = features;
1238         netdev_update_features(tun->dev);
1239
1240         return 0;
1241 }
1242
1243 static long __tun_chr_ioctl(struct file *file, unsigned int cmd,
1244                             unsigned long arg, int ifreq_len)
1245 {
1246         struct tun_file *tfile = file->private_data;
1247         struct tun_struct *tun;
1248         void __user* argp = (void __user*)arg;
1249         struct sock_fprog fprog;
1250         struct ifreq ifr;
1251         int sndbuf;
1252         int vnet_hdr_sz;
1253         int ret;
1254
1255         if (cmd == TUNSETIFF || _IOC_TYPE(cmd) == 0x89) {
1256                 if (copy_from_user(&ifr, argp, ifreq_len))
1257                         return -EFAULT;
1258         } else {
1259                 memset(&ifr, 0, sizeof(ifr));
1260         }
1261         if (cmd == TUNGETFEATURES) {
1262                 /* Currently this just means: "what IFF flags are valid?".
1263                  * This is needed because we never checked for invalid flags on
1264                  * TUNSETIFF. */
1265                 return put_user(IFF_TUN | IFF_TAP | IFF_NO_PI | IFF_ONE_QUEUE |
1266                                 IFF_VNET_HDR,
1267                                 (unsigned int __user*)argp);
1268         }
1269
1270         rtnl_lock();
1271
1272         tun = __tun_get(tfile);
1273         if (cmd == TUNSETIFF && !tun) {
1274                 ifr.ifr_name[IFNAMSIZ-1] = '\0';
1275
1276                 ret = tun_set_iff(tfile->net, file, &ifr);
1277
1278                 if (ret)
1279                         goto unlock;
1280
1281                 if (copy_to_user(argp, &ifr, ifreq_len))
1282                         ret = -EFAULT;
1283                 goto unlock;
1284         }
1285
1286         ret = -EBADFD;
1287         if (!tun)
1288                 goto unlock;
1289
1290         tun_debug(KERN_INFO, tun, "tun_chr_ioctl cmd %d\n", cmd);
1291
1292         ret = 0;
1293         switch (cmd) {
1294         case TUNGETIFF:
1295                 ret = tun_get_iff(current->nsproxy->net_ns, tun, &ifr);
1296                 if (ret)
1297                         break;
1298
1299                 if (copy_to_user(argp, &ifr, ifreq_len))
1300                         ret = -EFAULT;
1301                 break;
1302
1303         case TUNSETNOCSUM:
1304                 /* Disable/Enable checksum */
1305
1306                 /* [unimplemented] */
1307                 tun_debug(KERN_INFO, tun, "ignored: set checksum %s\n",
1308                           arg ? "disabled" : "enabled");
1309                 break;
1310
1311         case TUNSETPERSIST:
1312                 /* Disable/Enable persist mode */
1313                 if (arg)
1314                         tun->flags |= TUN_PERSIST;
1315                 else
1316                         tun->flags &= ~TUN_PERSIST;
1317
1318                 tun_debug(KERN_INFO, tun, "persist %s\n",
1319                           arg ? "enabled" : "disabled");
1320                 break;
1321
1322         case TUNSETOWNER:
1323                 /* Set owner of the device */
1324                 tun->owner = (uid_t) arg;
1325
1326                 tun_debug(KERN_INFO, tun, "owner set to %d\n", tun->owner);
1327                 break;
1328
1329         case TUNSETGROUP:
1330                 /* Set group of the device */
1331                 tun->group= (gid_t) arg;
1332
1333                 tun_debug(KERN_INFO, tun, "group set to %d\n", tun->group);
1334                 break;
1335
1336         case TUNSETLINK:
1337                 /* Only allow setting the type when the interface is down */
1338                 if (tun->dev->flags & IFF_UP) {
1339                         tun_debug(KERN_INFO, tun,
1340                                   "Linktype set failed because interface is up\n");
1341                         ret = -EBUSY;
1342                 } else {
1343                         tun->dev->type = (int) arg;
1344                         tun_debug(KERN_INFO, tun, "linktype set to %d\n",
1345                                   tun->dev->type);
1346                         ret = 0;
1347                 }
1348                 break;
1349
1350 #ifdef TUN_DEBUG
1351         case TUNSETDEBUG:
1352                 tun->debug = arg;
1353                 break;
1354 #endif
1355         case TUNSETOFFLOAD:
1356                 ret = set_offload(tun, arg);
1357                 break;
1358
1359         case TUNSETTXFILTER:
1360                 /* Can be set only for TAPs */
1361                 ret = -EINVAL;
1362                 if ((tun->flags & TUN_TYPE_MASK) != TUN_TAP_DEV)
1363                         break;
1364                 ret = update_filter(&tun->txflt, (void __user *)arg);
1365                 break;
1366
1367         case SIOCGIFHWADDR:
1368                 /* Get hw address */
1369                 memcpy(ifr.ifr_hwaddr.sa_data, tun->dev->dev_addr, ETH_ALEN);
1370                 ifr.ifr_hwaddr.sa_family = tun->dev->type;
1371                 if (copy_to_user(argp, &ifr, ifreq_len))
1372                         ret = -EFAULT;
1373                 break;
1374
1375         case SIOCSIFHWADDR:
1376                 /* Set hw address */
1377                 tun_debug(KERN_DEBUG, tun, "set hw address: %pM\n",
1378                           ifr.ifr_hwaddr.sa_data);
1379
1380                 ret = dev_set_mac_address(tun->dev, &ifr.ifr_hwaddr);
1381                 break;
1382
1383         case TUNGETSNDBUF:
1384                 sndbuf = tun->socket.sk->sk_sndbuf;
1385                 if (copy_to_user(argp, &sndbuf, sizeof(sndbuf)))
1386                         ret = -EFAULT;
1387                 break;
1388
1389         case TUNSETSNDBUF:
1390                 if (copy_from_user(&sndbuf, argp, sizeof(sndbuf))) {
1391                         ret = -EFAULT;
1392                         break;
1393                 }
1394
1395                 tun->socket.sk->sk_sndbuf = sndbuf;
1396                 break;
1397
1398         case TUNGETVNETHDRSZ:
1399                 vnet_hdr_sz = tun->vnet_hdr_sz;
1400                 if (copy_to_user(argp, &vnet_hdr_sz, sizeof(vnet_hdr_sz)))
1401                         ret = -EFAULT;
1402                 break;
1403
1404         case TUNSETVNETHDRSZ:
1405                 if (copy_from_user(&vnet_hdr_sz, argp, sizeof(vnet_hdr_sz))) {
1406                         ret = -EFAULT;
1407                         break;
1408                 }
1409                 if (vnet_hdr_sz < (int)sizeof(struct virtio_net_hdr)) {
1410                         ret = -EINVAL;
1411                         break;
1412                 }
1413
1414                 tun->vnet_hdr_sz = vnet_hdr_sz;
1415                 break;
1416
1417         case TUNATTACHFILTER:
1418                 /* Can be set only for TAPs */
1419                 ret = -EINVAL;
1420                 if ((tun->flags & TUN_TYPE_MASK) != TUN_TAP_DEV)
1421                         break;
1422                 ret = -EFAULT;
1423                 if (copy_from_user(&fprog, argp, sizeof(fprog)))
1424                         break;
1425
1426                 ret = sk_attach_filter(&fprog, tun->socket.sk);
1427                 break;
1428
1429         case TUNDETACHFILTER:
1430                 /* Can be set only for TAPs */
1431                 ret = -EINVAL;
1432                 if ((tun->flags & TUN_TYPE_MASK) != TUN_TAP_DEV)
1433                         break;
1434                 ret = sk_detach_filter(tun->socket.sk);
1435                 break;
1436
1437         default:
1438                 ret = -EINVAL;
1439                 break;
1440         }
1441
1442 unlock:
1443         rtnl_unlock();
1444         if (tun)
1445                 tun_put(tun);
1446         return ret;
1447 }
1448
1449 static long tun_chr_ioctl(struct file *file,
1450                           unsigned int cmd, unsigned long arg)
1451 {
1452         return __tun_chr_ioctl(file, cmd, arg, sizeof (struct ifreq));
1453 }
1454
1455 #ifdef CONFIG_COMPAT
1456 static long tun_chr_compat_ioctl(struct file *file,
1457                          unsigned int cmd, unsigned long arg)
1458 {
1459         switch (cmd) {
1460         case TUNSETIFF:
1461         case TUNGETIFF:
1462         case TUNSETTXFILTER:
1463         case TUNGETSNDBUF:
1464         case TUNSETSNDBUF:
1465         case SIOCGIFHWADDR:
1466         case SIOCSIFHWADDR:
1467                 arg = (unsigned long)compat_ptr(arg);
1468                 break;
1469         default:
1470                 arg = (compat_ulong_t)arg;
1471                 break;
1472         }
1473
1474         /*
1475          * compat_ifreq is shorter than ifreq, so we must not access beyond
1476          * the end of that structure. All fields that are used in this
1477          * driver are compatible though, we don't need to convert the
1478          * contents.
1479          */
1480         return __tun_chr_ioctl(file, cmd, arg, sizeof(struct compat_ifreq));
1481 }
1482 #endif /* CONFIG_COMPAT */
1483
1484 static int tun_chr_fasync(int fd, struct file *file, int on)
1485 {
1486         struct tun_struct *tun = tun_get(file);
1487         int ret;
1488
1489         if (!tun)
1490                 return -EBADFD;
1491
1492         tun_debug(KERN_INFO, tun, "tun_chr_fasync %d\n", on);
1493
1494         if ((ret = fasync_helper(fd, file, on, &tun->fasync)) < 0)
1495                 goto out;
1496
1497         if (on) {
1498                 ret = __f_setown(file, task_pid(current), PIDTYPE_PID, 0);
1499                 if (ret)
1500                         goto out;
1501                 tun->flags |= TUN_FASYNC;
1502         } else
1503                 tun->flags &= ~TUN_FASYNC;
1504         ret = 0;
1505 out:
1506         tun_put(tun);
1507         return ret;
1508 }
1509
1510 static int tun_chr_open(struct inode *inode, struct file * file)
1511 {
1512         struct tun_file *tfile;
1513
1514         DBG1(KERN_INFO, "tunX: tun_chr_open\n");
1515
1516         tfile = kmalloc(sizeof(*tfile), GFP_KERNEL);
1517         if (!tfile)
1518                 return -ENOMEM;
1519         atomic_set(&tfile->count, 0);
1520         tfile->tun = NULL;
1521         tfile->net = get_net(current->nsproxy->net_ns);
1522         file->private_data = tfile;
1523         return 0;
1524 }
1525
1526 static int tun_chr_close(struct inode *inode, struct file *file)
1527 {
1528         struct tun_file *tfile = file->private_data;
1529         struct tun_struct *tun;
1530
1531         tun = __tun_get(tfile);
1532         if (tun) {
1533                 struct net_device *dev = tun->dev;
1534
1535                 tun_debug(KERN_INFO, tun, "tun_chr_close\n");
1536
1537                 __tun_detach(tun);
1538
1539                 /* If desirable, unregister the netdevice. */
1540                 if (!(tun->flags & TUN_PERSIST)) {
1541                         rtnl_lock();
1542                         if (dev->reg_state == NETREG_REGISTERED)
1543                                 unregister_netdevice(dev);
1544                         rtnl_unlock();
1545                 }
1546         }
1547
1548         tun = tfile->tun;
1549         if (tun)
1550                 sock_put(tun->socket.sk);
1551
1552         put_net(tfile->net);
1553         kfree(tfile);
1554
1555         return 0;
1556 }
1557
1558 static const struct file_operations tun_fops = {
1559         .owner  = THIS_MODULE,
1560         .llseek = no_llseek,
1561         .read  = do_sync_read,
1562         .aio_read  = tun_chr_aio_read,
1563         .write = do_sync_write,
1564         .aio_write = tun_chr_aio_write,
1565         .poll   = tun_chr_poll,
1566         .unlocked_ioctl = tun_chr_ioctl,
1567 #ifdef CONFIG_COMPAT
1568         .compat_ioctl = tun_chr_compat_ioctl,
1569 #endif
1570         .open   = tun_chr_open,
1571         .release = tun_chr_close,
1572         .fasync = tun_chr_fasync
1573 };
1574
1575 static struct miscdevice tun_miscdev = {
1576         .minor = TUN_MINOR,
1577         .name = "tun",
1578         .nodename = "net/tun",
1579         .fops = &tun_fops,
1580 };
1581
1582 /* ethtool interface */
1583
1584 static int tun_get_settings(struct net_device *dev, struct ethtool_cmd *cmd)
1585 {
1586         cmd->supported          = 0;
1587         cmd->advertising        = 0;
1588         ethtool_cmd_speed_set(cmd, SPEED_10);
1589         cmd->duplex             = DUPLEX_FULL;
1590         cmd->port               = PORT_TP;
1591         cmd->phy_address        = 0;
1592         cmd->transceiver        = XCVR_INTERNAL;
1593         cmd->autoneg            = AUTONEG_DISABLE;
1594         cmd->maxtxpkt           = 0;
1595         cmd->maxrxpkt           = 0;
1596         return 0;
1597 }
1598
1599 static void tun_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
1600 {
1601         struct tun_struct *tun = netdev_priv(dev);
1602
1603         strcpy(info->driver, DRV_NAME);
1604         strcpy(info->version, DRV_VERSION);
1605         strcpy(info->fw_version, "N/A");
1606
1607         switch (tun->flags & TUN_TYPE_MASK) {
1608         case TUN_TUN_DEV:
1609                 strcpy(info->bus_info, "tun");
1610                 break;
1611         case TUN_TAP_DEV:
1612                 strcpy(info->bus_info, "tap");
1613                 break;
1614         }
1615 }
1616
1617 static u32 tun_get_msglevel(struct net_device *dev)
1618 {
1619 #ifdef TUN_DEBUG
1620         struct tun_struct *tun = netdev_priv(dev);
1621         return tun->debug;
1622 #else
1623         return -EOPNOTSUPP;
1624 #endif
1625 }
1626
1627 static void tun_set_msglevel(struct net_device *dev, u32 value)
1628 {
1629 #ifdef TUN_DEBUG
1630         struct tun_struct *tun = netdev_priv(dev);
1631         tun->debug = value;
1632 #endif
1633 }
1634
1635 static const struct ethtool_ops tun_ethtool_ops = {
1636         .get_settings   = tun_get_settings,
1637         .get_drvinfo    = tun_get_drvinfo,
1638         .get_msglevel   = tun_get_msglevel,
1639         .set_msglevel   = tun_set_msglevel,
1640         .get_link       = ethtool_op_get_link,
1641 };
1642
1643
1644 static int __init tun_init(void)
1645 {
1646         int ret = 0;
1647
1648         pr_info("%s, %s\n", DRV_DESCRIPTION, DRV_VERSION);
1649         pr_info("%s\n", DRV_COPYRIGHT);
1650
1651         ret = rtnl_link_register(&tun_link_ops);
1652         if (ret) {
1653                 pr_err("Can't register link_ops\n");
1654                 goto err_linkops;
1655         }
1656
1657         ret = misc_register(&tun_miscdev);
1658         if (ret) {
1659                 pr_err("Can't register misc device %d\n", TUN_MINOR);
1660                 goto err_misc;
1661         }
1662         return  0;
1663 err_misc:
1664         rtnl_link_unregister(&tun_link_ops);
1665 err_linkops:
1666         return ret;
1667 }
1668
1669 static void tun_cleanup(void)
1670 {
1671         misc_deregister(&tun_miscdev);
1672         rtnl_link_unregister(&tun_link_ops);
1673 }
1674
1675 /* Get an underlying socket object from tun file.  Returns error unless file is
1676  * attached to a device.  The returned object works like a packet socket, it
1677  * can be used for sock_sendmsg/sock_recvmsg.  The caller is responsible for
1678  * holding a reference to the file for as long as the socket is in use. */
1679 struct socket *tun_get_socket(struct file *file)
1680 {
1681         struct tun_struct *tun;
1682         if (file->f_op != &tun_fops)
1683                 return ERR_PTR(-EINVAL);
1684         tun = tun_get(file);
1685         if (!tun)
1686                 return ERR_PTR(-EBADFD);
1687         tun_put(tun);
1688         return &tun->socket;
1689 }
1690 EXPORT_SYMBOL_GPL(tun_get_socket);
1691
1692 module_init(tun_init);
1693 module_exit(tun_cleanup);
1694 MODULE_DESCRIPTION(DRV_DESCRIPTION);
1695 MODULE_AUTHOR(DRV_COPYRIGHT);
1696 MODULE_LICENSE("GPL");
1697 MODULE_ALIAS_MISCDEV(TUN_MINOR);
1698 MODULE_ALIAS("devname:net/tun");