Merge commit 'a88f9e27498afaea615ad3e93af4f26df1f84987' into after-upstream-android
[linux-3.10.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 eth_random_addr() for tap MAC address.
26  *
27  *  Harald Roelle <harald.roelle@ifi.lmu.de>  2004/04/20
28  *    Fixes in packet dropping, queue length setting and queue wakeup.
29  *    Increased default tx queue length.
30  *    Added ethtool API.
31  *    Minor cleanups
32  *
33  *  Daniel Podlejski <underley@underley.eu.org>
34  *    Modifications for 2.3.99-pre5 kernel.
35  */
36
37 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
38
39 #define DRV_NAME        "tun"
40 #define DRV_VERSION     "1.6"
41 #define DRV_DESCRIPTION "Universal TUN/TAP device driver"
42 #define DRV_COPYRIGHT   "(C) 1999-2004 Max Krasnyansky <maxk@qualcomm.com>"
43
44 #include <linux/module.h>
45 #include <linux/errno.h>
46 #include <linux/kernel.h>
47 #include <linux/major.h>
48 #include <linux/slab.h>
49 #include <linux/poll.h>
50 #include <linux/fcntl.h>
51 #include <linux/init.h>
52 #include <linux/skbuff.h>
53 #include <linux/netdevice.h>
54 #include <linux/etherdevice.h>
55 #include <linux/miscdevice.h>
56 #include <linux/ethtool.h>
57 #include <linux/rtnetlink.h>
58 #include <linux/compat.h>
59 #include <linux/if.h>
60 #include <linux/if_arp.h>
61 #include <linux/if_ether.h>
62 #include <linux/if_tun.h>
63 #include <linux/crc32.h>
64 #include <linux/nsproxy.h>
65 #include <linux/virtio_net.h>
66 #include <linux/rcupdate.h>
67 #include <net/net_namespace.h>
68 #include <net/netns/generic.h>
69 #include <net/rtnetlink.h>
70 #include <net/sock.h>
71
72 #include <asm/uaccess.h>
73
74 /* Uncomment to enable debugging */
75 /* #define TUN_DEBUG 1 */
76
77 #ifdef TUN_DEBUG
78 static int debug;
79
80 #define tun_debug(level, tun, fmt, args...)                     \
81 do {                                                            \
82         if (tun->debug)                                         \
83                 netdev_printk(level, tun->dev, fmt, ##args);    \
84 } while (0)
85 #define DBG1(level, fmt, args...)                               \
86 do {                                                            \
87         if (debug == 2)                                         \
88                 printk(level fmt, ##args);                      \
89 } while (0)
90 #else
91 #define tun_debug(level, tun, fmt, args...)                     \
92 do {                                                            \
93         if (0)                                                  \
94                 netdev_printk(level, tun->dev, fmt, ##args);    \
95 } while (0)
96 #define DBG1(level, fmt, args...)                               \
97 do {                                                            \
98         if (0)                                                  \
99                 printk(level fmt, ##args);                      \
100 } while (0)
101 #endif
102
103 #define GOODCOPY_LEN 128
104
105 #define FLT_EXACT_COUNT 8
106 struct tap_filter {
107         unsigned int    count;    /* Number of addrs. Zero means disabled */
108         u32             mask[2];  /* Mask of the hashed addrs */
109         unsigned char   addr[FLT_EXACT_COUNT][ETH_ALEN];
110 };
111
112 /* DEFAULT_MAX_NUM_RSS_QUEUES were choosed to let the rx/tx queues allocated for
113  * the netdevice to be fit in one page. So we can make sure the success of
114  * memory allocation. TODO: increase the limit. */
115 #define MAX_TAP_QUEUES DEFAULT_MAX_NUM_RSS_QUEUES
116 #define MAX_TAP_FLOWS  4096
117
118 #define TUN_FLOW_EXPIRE (3 * HZ)
119
120 /* A tun_file connects an open character device to a tuntap netdevice. It
121  * also contains all socket related strctures (except sock_fprog and tap_filter)
122  * to serve as one transmit queue for tuntap device. The sock_fprog and
123  * tap_filter were kept in tun_struct since they were used for filtering for the
124  * netdevice not for a specific queue (at least I didn't see the requirement for
125  * this).
126  *
127  * RCU usage:
128  * The tun_file and tun_struct are loosely coupled, the pointer from one to the
129  * other can only be read while rcu_read_lock or rtnl_lock is held.
130  */
131 struct tun_file {
132         struct sock sk;
133         struct socket socket;
134         struct socket_wq wq;
135         struct tun_struct __rcu *tun;
136         struct net *net;
137         struct fasync_struct *fasync;
138         /* only used for fasnyc */
139         unsigned int flags;
140         u16 queue_index;
141         struct list_head next;
142         struct tun_struct *detached;
143 };
144
145 struct tun_flow_entry {
146         struct hlist_node hash_link;
147         struct rcu_head rcu;
148         struct tun_struct *tun;
149
150         u32 rxhash;
151         int queue_index;
152         unsigned long updated;
153 };
154
155 #define TUN_NUM_FLOW_ENTRIES 1024
156
157 /* Since the socket were moved to tun_file, to preserve the behavior of persist
158  * device, socket filter, sndbuf and vnet header size were restore when the
159  * file were attached to a persist device.
160  */
161 struct tun_struct {
162         struct tun_file __rcu   *tfiles[MAX_TAP_QUEUES];
163         unsigned int            numqueues;
164         unsigned int            flags;
165         kuid_t                  owner;
166         kgid_t                  group;
167
168         struct net_device       *dev;
169         netdev_features_t       set_features;
170 #define TUN_USER_FEATURES (NETIF_F_HW_CSUM|NETIF_F_TSO_ECN|NETIF_F_TSO| \
171                           NETIF_F_TSO6|NETIF_F_UFO)
172
173         int                     vnet_hdr_sz;
174         int                     sndbuf;
175         struct tap_filter       txflt;
176         struct sock_fprog       fprog;
177         /* protected by rtnl lock */
178         bool                    filter_attached;
179 #ifdef TUN_DEBUG
180         int debug;
181 #endif
182         spinlock_t lock;
183         struct hlist_head flows[TUN_NUM_FLOW_ENTRIES];
184         struct timer_list flow_gc_timer;
185         unsigned long ageing_time;
186         unsigned int numdisabled;
187         struct list_head disabled;
188         void *security;
189         u32 flow_count;
190 };
191
192 static inline u32 tun_hashfn(u32 rxhash)
193 {
194         return rxhash & 0x3ff;
195 }
196
197 static struct tun_flow_entry *tun_flow_find(struct hlist_head *head, u32 rxhash)
198 {
199         struct tun_flow_entry *e;
200
201         hlist_for_each_entry_rcu(e, head, hash_link) {
202                 if (e->rxhash == rxhash)
203                         return e;
204         }
205         return NULL;
206 }
207
208 static struct tun_flow_entry *tun_flow_create(struct tun_struct *tun,
209                                               struct hlist_head *head,
210                                               u32 rxhash, u16 queue_index)
211 {
212         struct tun_flow_entry *e = kmalloc(sizeof(*e), GFP_ATOMIC);
213
214         if (e) {
215                 tun_debug(KERN_INFO, tun, "create flow: hash %u index %u\n",
216                           rxhash, queue_index);
217                 e->updated = jiffies;
218                 e->rxhash = rxhash;
219                 e->queue_index = queue_index;
220                 e->tun = tun;
221                 hlist_add_head_rcu(&e->hash_link, head);
222                 ++tun->flow_count;
223         }
224         return e;
225 }
226
227 static void tun_flow_delete(struct tun_struct *tun, struct tun_flow_entry *e)
228 {
229         tun_debug(KERN_INFO, tun, "delete flow: hash %u index %u\n",
230                   e->rxhash, e->queue_index);
231         hlist_del_rcu(&e->hash_link);
232         kfree_rcu(e, rcu);
233         --tun->flow_count;
234 }
235
236 static void tun_flow_flush(struct tun_struct *tun)
237 {
238         int i;
239
240         spin_lock_bh(&tun->lock);
241         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
242                 struct tun_flow_entry *e;
243                 struct hlist_node *n;
244
245                 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link)
246                         tun_flow_delete(tun, e);
247         }
248         spin_unlock_bh(&tun->lock);
249 }
250
251 static void tun_flow_delete_by_queue(struct tun_struct *tun, u16 queue_index)
252 {
253         int i;
254
255         spin_lock_bh(&tun->lock);
256         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
257                 struct tun_flow_entry *e;
258                 struct hlist_node *n;
259
260                 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
261                         if (e->queue_index == queue_index)
262                                 tun_flow_delete(tun, e);
263                 }
264         }
265         spin_unlock_bh(&tun->lock);
266 }
267
268 static void tun_flow_cleanup(unsigned long data)
269 {
270         struct tun_struct *tun = (struct tun_struct *)data;
271         unsigned long delay = tun->ageing_time;
272         unsigned long next_timer = jiffies + delay;
273         unsigned long count = 0;
274         int i;
275
276         tun_debug(KERN_INFO, tun, "tun_flow_cleanup\n");
277
278         spin_lock_bh(&tun->lock);
279         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
280                 struct tun_flow_entry *e;
281                 struct hlist_node *n;
282
283                 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
284                         unsigned long this_timer;
285                         count++;
286                         this_timer = e->updated + delay;
287                         if (time_before_eq(this_timer, jiffies))
288                                 tun_flow_delete(tun, e);
289                         else if (time_before(this_timer, next_timer))
290                                 next_timer = this_timer;
291                 }
292         }
293
294         if (count)
295                 mod_timer(&tun->flow_gc_timer, round_jiffies_up(next_timer));
296         spin_unlock_bh(&tun->lock);
297 }
298
299 static void tun_flow_update(struct tun_struct *tun, u32 rxhash,
300                             struct tun_file *tfile)
301 {
302         struct hlist_head *head;
303         struct tun_flow_entry *e;
304         unsigned long delay = tun->ageing_time;
305         u16 queue_index = tfile->queue_index;
306
307         if (!rxhash)
308                 return;
309         else
310                 head = &tun->flows[tun_hashfn(rxhash)];
311
312         rcu_read_lock();
313
314         /* We may get a very small possibility of OOO during switching, not
315          * worth to optimize.*/
316         if (tun->numqueues == 1 || tfile->detached)
317                 goto unlock;
318
319         e = tun_flow_find(head, rxhash);
320         if (likely(e)) {
321                 /* TODO: keep queueing to old queue until it's empty? */
322                 e->queue_index = queue_index;
323                 e->updated = jiffies;
324         } else {
325                 spin_lock_bh(&tun->lock);
326                 if (!tun_flow_find(head, rxhash) &&
327                     tun->flow_count < MAX_TAP_FLOWS)
328                         tun_flow_create(tun, head, rxhash, queue_index);
329
330                 if (!timer_pending(&tun->flow_gc_timer))
331                         mod_timer(&tun->flow_gc_timer,
332                                   round_jiffies_up(jiffies + delay));
333                 spin_unlock_bh(&tun->lock);
334         }
335
336 unlock:
337         rcu_read_unlock();
338 }
339
340 /* We try to identify a flow through its rxhash first. The reason that
341  * we do not check rxq no. is becuase some cards(e.g 82599), chooses
342  * the rxq based on the txq where the last packet of the flow comes. As
343  * the userspace application move between processors, we may get a
344  * different rxq no. here. If we could not get rxhash, then we would
345  * hope the rxq no. may help here.
346  */
347 static u16 tun_select_queue(struct net_device *dev, struct sk_buff *skb)
348 {
349         struct tun_struct *tun = netdev_priv(dev);
350         struct tun_flow_entry *e;
351         u32 txq = 0;
352         u32 numqueues = 0;
353
354         rcu_read_lock();
355         numqueues = ACCESS_ONCE(tun->numqueues);
356
357         txq = skb_get_rxhash(skb);
358         if (txq) {
359                 e = tun_flow_find(&tun->flows[tun_hashfn(txq)], txq);
360                 if (e)
361                         txq = e->queue_index;
362                 else
363                         /* use multiply and shift instead of expensive divide */
364                         txq = ((u64)txq * numqueues) >> 32;
365         } else if (likely(skb_rx_queue_recorded(skb))) {
366                 txq = skb_get_rx_queue(skb);
367                 while (unlikely(txq >= numqueues))
368                         txq -= numqueues;
369         }
370
371         rcu_read_unlock();
372         return txq;
373 }
374
375 static inline bool tun_not_capable(struct tun_struct *tun)
376 {
377         const struct cred *cred = current_cred();
378         struct net *net = dev_net(tun->dev);
379
380         return ((uid_valid(tun->owner) && !uid_eq(cred->euid, tun->owner)) ||
381                   (gid_valid(tun->group) && !in_egroup_p(tun->group))) &&
382                 !ns_capable(net->user_ns, CAP_NET_ADMIN);
383 }
384
385 static void tun_set_real_num_queues(struct tun_struct *tun)
386 {
387         netif_set_real_num_tx_queues(tun->dev, tun->numqueues);
388         netif_set_real_num_rx_queues(tun->dev, tun->numqueues);
389 }
390
391 static void tun_disable_queue(struct tun_struct *tun, struct tun_file *tfile)
392 {
393         tfile->detached = tun;
394         list_add_tail(&tfile->next, &tun->disabled);
395         ++tun->numdisabled;
396 }
397
398 static struct tun_struct *tun_enable_queue(struct tun_file *tfile)
399 {
400         struct tun_struct *tun = tfile->detached;
401
402         tfile->detached = NULL;
403         list_del_init(&tfile->next);
404         --tun->numdisabled;
405         return tun;
406 }
407
408 static void __tun_detach(struct tun_file *tfile, bool clean)
409 {
410         struct tun_file *ntfile;
411         struct tun_struct *tun;
412
413         tun = rtnl_dereference(tfile->tun);
414
415         if (tun && !tfile->detached) {
416                 u16 index = tfile->queue_index;
417                 BUG_ON(index >= tun->numqueues);
418
419                 rcu_assign_pointer(tun->tfiles[index],
420                                    tun->tfiles[tun->numqueues - 1]);
421                 ntfile = rtnl_dereference(tun->tfiles[index]);
422                 ntfile->queue_index = index;
423
424                 --tun->numqueues;
425                 if (clean) {
426                         rcu_assign_pointer(tfile->tun, NULL);
427                         sock_put(&tfile->sk);
428                 } else
429                         tun_disable_queue(tun, tfile);
430
431                 synchronize_net();
432                 tun_flow_delete_by_queue(tun, tun->numqueues + 1);
433                 /* Drop read queue */
434                 skb_queue_purge(&tfile->sk.sk_receive_queue);
435                 tun_set_real_num_queues(tun);
436         } else if (tfile->detached && clean) {
437                 tun = tun_enable_queue(tfile);
438                 sock_put(&tfile->sk);
439         }
440
441         if (clean) {
442                 if (tun && tun->numqueues == 0 && tun->numdisabled == 0) {
443                         netif_carrier_off(tun->dev);
444
445                         if (!(tun->flags & TUN_PERSIST) &&
446                             tun->dev->reg_state == NETREG_REGISTERED)
447                                 unregister_netdevice(tun->dev);
448                 }
449
450                 BUG_ON(!test_bit(SOCK_EXTERNALLY_ALLOCATED,
451                                  &tfile->socket.flags));
452                 sk_release_kernel(&tfile->sk);
453         }
454 }
455
456 static void tun_detach(struct tun_file *tfile, bool clean)
457 {
458         rtnl_lock();
459         __tun_detach(tfile, clean);
460         rtnl_unlock();
461 }
462
463 static void tun_detach_all(struct net_device *dev)
464 {
465         struct tun_struct *tun = netdev_priv(dev);
466         struct tun_file *tfile, *tmp;
467         int i, n = tun->numqueues;
468
469         for (i = 0; i < n; i++) {
470                 tfile = rtnl_dereference(tun->tfiles[i]);
471                 BUG_ON(!tfile);
472                 wake_up_all(&tfile->wq.wait);
473                 rcu_assign_pointer(tfile->tun, NULL);
474                 --tun->numqueues;
475         }
476         list_for_each_entry(tfile, &tun->disabled, next) {
477                 wake_up_all(&tfile->wq.wait);
478                 rcu_assign_pointer(tfile->tun, NULL);
479         }
480         BUG_ON(tun->numqueues != 0);
481
482         synchronize_net();
483         for (i = 0; i < n; i++) {
484                 tfile = rtnl_dereference(tun->tfiles[i]);
485                 /* Drop read queue */
486                 skb_queue_purge(&tfile->sk.sk_receive_queue);
487                 sock_put(&tfile->sk);
488         }
489         list_for_each_entry_safe(tfile, tmp, &tun->disabled, next) {
490                 tun_enable_queue(tfile);
491                 skb_queue_purge(&tfile->sk.sk_receive_queue);
492                 sock_put(&tfile->sk);
493         }
494         BUG_ON(tun->numdisabled != 0);
495
496         if (tun->flags & TUN_PERSIST)
497                 module_put(THIS_MODULE);
498 }
499
500 static int tun_attach(struct tun_struct *tun, struct file *file)
501 {
502         struct tun_file *tfile = file->private_data;
503         int err;
504
505         err = security_tun_dev_attach(tfile->socket.sk, tun->security);
506         if (err < 0)
507                 goto out;
508
509         err = -EINVAL;
510         if (rtnl_dereference(tfile->tun) && !tfile->detached)
511                 goto out;
512
513         err = -EBUSY;
514         if (!(tun->flags & TUN_TAP_MQ) && tun->numqueues == 1)
515                 goto out;
516
517         err = -E2BIG;
518         if (!tfile->detached &&
519             tun->numqueues + tun->numdisabled == MAX_TAP_QUEUES)
520                 goto out;
521
522         err = 0;
523
524         /* Re-attach the filter to presist device */
525         if (tun->filter_attached == true) {
526                 err = sk_attach_filter(&tun->fprog, tfile->socket.sk);
527                 if (!err)
528                         goto out;
529         }
530         tfile->queue_index = tun->numqueues;
531         rcu_assign_pointer(tfile->tun, tun);
532         rcu_assign_pointer(tun->tfiles[tun->numqueues], tfile);
533         tun->numqueues++;
534
535         if (tfile->detached)
536                 tun_enable_queue(tfile);
537         else
538                 sock_hold(&tfile->sk);
539
540         tun_set_real_num_queues(tun);
541
542         /* device is allowed to go away first, so no need to hold extra
543          * refcnt.
544          */
545
546 out:
547         return err;
548 }
549
550 static struct tun_struct *__tun_get(struct tun_file *tfile)
551 {
552         struct tun_struct *tun;
553
554         rcu_read_lock();
555         tun = rcu_dereference(tfile->tun);
556         if (tun)
557                 dev_hold(tun->dev);
558         rcu_read_unlock();
559
560         return tun;
561 }
562
563 static struct tun_struct *tun_get(struct file *file)
564 {
565         return __tun_get(file->private_data);
566 }
567
568 static void tun_put(struct tun_struct *tun)
569 {
570         dev_put(tun->dev);
571 }
572
573 /* TAP filtering */
574 static void addr_hash_set(u32 *mask, const u8 *addr)
575 {
576         int n = ether_crc(ETH_ALEN, addr) >> 26;
577         mask[n >> 5] |= (1 << (n & 31));
578 }
579
580 static unsigned int addr_hash_test(const u32 *mask, const u8 *addr)
581 {
582         int n = ether_crc(ETH_ALEN, addr) >> 26;
583         return mask[n >> 5] & (1 << (n & 31));
584 }
585
586 static int update_filter(struct tap_filter *filter, void __user *arg)
587 {
588         struct { u8 u[ETH_ALEN]; } *addr;
589         struct tun_filter uf;
590         int err, alen, n, nexact;
591
592         if (copy_from_user(&uf, arg, sizeof(uf)))
593                 return -EFAULT;
594
595         if (!uf.count) {
596                 /* Disabled */
597                 filter->count = 0;
598                 return 0;
599         }
600
601         alen = ETH_ALEN * uf.count;
602         addr = kmalloc(alen, GFP_KERNEL);
603         if (!addr)
604                 return -ENOMEM;
605
606         if (copy_from_user(addr, arg + sizeof(uf), alen)) {
607                 err = -EFAULT;
608                 goto done;
609         }
610
611         /* The filter is updated without holding any locks. Which is
612          * perfectly safe. We disable it first and in the worst
613          * case we'll accept a few undesired packets. */
614         filter->count = 0;
615         wmb();
616
617         /* Use first set of addresses as an exact filter */
618         for (n = 0; n < uf.count && n < FLT_EXACT_COUNT; n++)
619                 memcpy(filter->addr[n], addr[n].u, ETH_ALEN);
620
621         nexact = n;
622
623         /* Remaining multicast addresses are hashed,
624          * unicast will leave the filter disabled. */
625         memset(filter->mask, 0, sizeof(filter->mask));
626         for (; n < uf.count; n++) {
627                 if (!is_multicast_ether_addr(addr[n].u)) {
628                         err = 0; /* no filter */
629                         goto done;
630                 }
631                 addr_hash_set(filter->mask, addr[n].u);
632         }
633
634         /* For ALLMULTI just set the mask to all ones.
635          * This overrides the mask populated above. */
636         if ((uf.flags & TUN_FLT_ALLMULTI))
637                 memset(filter->mask, ~0, sizeof(filter->mask));
638
639         /* Now enable the filter */
640         wmb();
641         filter->count = nexact;
642
643         /* Return the number of exact filters */
644         err = nexact;
645
646 done:
647         kfree(addr);
648         return err;
649 }
650
651 /* Returns: 0 - drop, !=0 - accept */
652 static int run_filter(struct tap_filter *filter, const struct sk_buff *skb)
653 {
654         /* Cannot use eth_hdr(skb) here because skb_mac_hdr() is incorrect
655          * at this point. */
656         struct ethhdr *eh = (struct ethhdr *) skb->data;
657         int i;
658
659         /* Exact match */
660         for (i = 0; i < filter->count; i++)
661                 if (ether_addr_equal(eh->h_dest, filter->addr[i]))
662                         return 1;
663
664         /* Inexact match (multicast only) */
665         if (is_multicast_ether_addr(eh->h_dest))
666                 return addr_hash_test(filter->mask, eh->h_dest);
667
668         return 0;
669 }
670
671 /*
672  * Checks whether the packet is accepted or not.
673  * Returns: 0 - drop, !=0 - accept
674  */
675 static int check_filter(struct tap_filter *filter, const struct sk_buff *skb)
676 {
677         if (!filter->count)
678                 return 1;
679
680         return run_filter(filter, skb);
681 }
682
683 /* Network device part of the driver */
684
685 static const struct ethtool_ops tun_ethtool_ops;
686
687 /* Net device detach from fd. */
688 static void tun_net_uninit(struct net_device *dev)
689 {
690         tun_detach_all(dev);
691 }
692
693 /* Net device open. */
694 static int tun_net_open(struct net_device *dev)
695 {
696         netif_tx_start_all_queues(dev);
697         return 0;
698 }
699
700 /* Net device close. */
701 static int tun_net_close(struct net_device *dev)
702 {
703         netif_tx_stop_all_queues(dev);
704         return 0;
705 }
706
707 /* Net device start xmit */
708 static netdev_tx_t tun_net_xmit(struct sk_buff *skb, struct net_device *dev)
709 {
710         struct tun_struct *tun = netdev_priv(dev);
711         int txq = skb->queue_mapping;
712         struct tun_file *tfile;
713
714         rcu_read_lock();
715         tfile = rcu_dereference(tun->tfiles[txq]);
716
717         /* Drop packet if interface is not attached */
718         if (txq >= tun->numqueues)
719                 goto drop;
720
721         tun_debug(KERN_INFO, tun, "tun_net_xmit %d\n", skb->len);
722
723         BUG_ON(!tfile);
724
725         /* Drop if the filter does not like it.
726          * This is a noop if the filter is disabled.
727          * Filter can be enabled only for the TAP devices. */
728         if (!check_filter(&tun->txflt, skb))
729                 goto drop;
730
731         if (tfile->socket.sk->sk_filter &&
732             sk_filter(tfile->socket.sk, skb))
733                 goto drop;
734
735         /* Limit the number of packets queued by dividing txq length with the
736          * number of queues.
737          */
738         if (skb_queue_len(&tfile->socket.sk->sk_receive_queue)
739                           >= dev->tx_queue_len / tun->numqueues)
740                 goto drop;
741
742         /* Orphan the skb - required as we might hang on to it
743          * for indefinite time. */
744         if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC)))
745                 goto drop;
746         skb_orphan(skb);
747
748         nf_reset(skb);
749
750         /* Enqueue packet */
751         skb_queue_tail(&tfile->socket.sk->sk_receive_queue, skb);
752
753         /* Notify and wake up reader process */
754         if (tfile->flags & TUN_FASYNC)
755                 kill_fasync(&tfile->fasync, SIGIO, POLL_IN);
756         wake_up_interruptible_poll(&tfile->wq.wait, POLLIN |
757                                    POLLRDNORM | POLLRDBAND);
758
759         rcu_read_unlock();
760         return NETDEV_TX_OK;
761
762 drop:
763         dev->stats.tx_dropped++;
764         skb_tx_error(skb);
765         kfree_skb(skb);
766         rcu_read_unlock();
767         return NETDEV_TX_OK;
768 }
769
770 static void tun_net_mclist(struct net_device *dev)
771 {
772         /*
773          * This callback is supposed to deal with mc filter in
774          * _rx_ path and has nothing to do with the _tx_ path.
775          * In rx path we always accept everything userspace gives us.
776          */
777 }
778
779 #define MIN_MTU 68
780 #define MAX_MTU 65535
781
782 static int
783 tun_net_change_mtu(struct net_device *dev, int new_mtu)
784 {
785         if (new_mtu < MIN_MTU || new_mtu + dev->hard_header_len > MAX_MTU)
786                 return -EINVAL;
787         dev->mtu = new_mtu;
788         return 0;
789 }
790
791 static netdev_features_t tun_net_fix_features(struct net_device *dev,
792         netdev_features_t features)
793 {
794         struct tun_struct *tun = netdev_priv(dev);
795
796         return (features & tun->set_features) | (features & ~TUN_USER_FEATURES);
797 }
798 #ifdef CONFIG_NET_POLL_CONTROLLER
799 static void tun_poll_controller(struct net_device *dev)
800 {
801         /*
802          * Tun only receives frames when:
803          * 1) the char device endpoint gets data from user space
804          * 2) the tun socket gets a sendmsg call from user space
805          * Since both of those are syncronous operations, we are guaranteed
806          * never to have pending data when we poll for it
807          * so theres nothing to do here but return.
808          * We need this though so netpoll recognizes us as an interface that
809          * supports polling, which enables bridge devices in virt setups to
810          * still use netconsole
811          */
812         return;
813 }
814 #endif
815 static const struct net_device_ops tun_netdev_ops = {
816         .ndo_uninit             = tun_net_uninit,
817         .ndo_open               = tun_net_open,
818         .ndo_stop               = tun_net_close,
819         .ndo_start_xmit         = tun_net_xmit,
820         .ndo_change_mtu         = tun_net_change_mtu,
821         .ndo_fix_features       = tun_net_fix_features,
822         .ndo_select_queue       = tun_select_queue,
823 #ifdef CONFIG_NET_POLL_CONTROLLER
824         .ndo_poll_controller    = tun_poll_controller,
825 #endif
826 };
827
828 static const struct net_device_ops tap_netdev_ops = {
829         .ndo_uninit             = tun_net_uninit,
830         .ndo_open               = tun_net_open,
831         .ndo_stop               = tun_net_close,
832         .ndo_start_xmit         = tun_net_xmit,
833         .ndo_change_mtu         = tun_net_change_mtu,
834         .ndo_fix_features       = tun_net_fix_features,
835         .ndo_set_rx_mode        = tun_net_mclist,
836         .ndo_set_mac_address    = eth_mac_addr,
837         .ndo_validate_addr      = eth_validate_addr,
838         .ndo_select_queue       = tun_select_queue,
839 #ifdef CONFIG_NET_POLL_CONTROLLER
840         .ndo_poll_controller    = tun_poll_controller,
841 #endif
842 };
843
844 static int tun_flow_init(struct tun_struct *tun)
845 {
846         int i;
847
848         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++)
849                 INIT_HLIST_HEAD(&tun->flows[i]);
850
851         tun->ageing_time = TUN_FLOW_EXPIRE;
852         setup_timer(&tun->flow_gc_timer, tun_flow_cleanup, (unsigned long)tun);
853         mod_timer(&tun->flow_gc_timer,
854                   round_jiffies_up(jiffies + tun->ageing_time));
855
856         return 0;
857 }
858
859 static void tun_flow_uninit(struct tun_struct *tun)
860 {
861         del_timer_sync(&tun->flow_gc_timer);
862         tun_flow_flush(tun);
863 }
864
865 /* Initialize net device. */
866 static void tun_net_init(struct net_device *dev)
867 {
868         struct tun_struct *tun = netdev_priv(dev);
869
870         switch (tun->flags & TUN_TYPE_MASK) {
871         case TUN_TUN_DEV:
872                 dev->netdev_ops = &tun_netdev_ops;
873
874                 /* Point-to-Point TUN Device */
875                 dev->hard_header_len = 0;
876                 dev->addr_len = 0;
877                 dev->mtu = 1500;
878
879                 /* Zero header length */
880                 dev->type = ARPHRD_NONE;
881                 dev->flags = IFF_POINTOPOINT | IFF_NOARP | IFF_MULTICAST;
882                 dev->tx_queue_len = TUN_READQ_SIZE;  /* We prefer our own queue length */
883                 break;
884
885         case TUN_TAP_DEV:
886                 dev->netdev_ops = &tap_netdev_ops;
887                 /* Ethernet TAP Device */
888                 ether_setup(dev);
889                 dev->priv_flags &= ~IFF_TX_SKB_SHARING;
890                 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
891
892                 eth_hw_addr_random(dev);
893
894                 dev->tx_queue_len = TUN_READQ_SIZE;  /* We prefer our own queue length */
895                 break;
896         }
897 }
898
899 /* Character device part */
900
901 /* Poll */
902 static unsigned int tun_chr_poll(struct file *file, poll_table *wait)
903 {
904         struct tun_file *tfile = file->private_data;
905         struct tun_struct *tun = __tun_get(tfile);
906         struct sock *sk;
907         unsigned int mask = 0;
908
909         if (!tun)
910                 return POLLERR;
911
912         sk = tfile->socket.sk;
913
914         tun_debug(KERN_INFO, tun, "tun_chr_poll\n");
915
916         poll_wait(file, &tfile->wq.wait, wait);
917
918         if (!skb_queue_empty(&sk->sk_receive_queue))
919                 mask |= POLLIN | POLLRDNORM;
920
921         if (sock_writeable(sk) ||
922             (!test_and_set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags) &&
923              sock_writeable(sk)))
924                 mask |= POLLOUT | POLLWRNORM;
925
926         if (tun->dev->reg_state != NETREG_REGISTERED)
927                 mask = POLLERR;
928
929         tun_put(tun);
930         return mask;
931 }
932
933 /* prepad is the amount to reserve at front.  len is length after that.
934  * linear is a hint as to how much to copy (usually headers). */
935 static struct sk_buff *tun_alloc_skb(struct tun_file *tfile,
936                                      size_t prepad, size_t len,
937                                      size_t linear, int noblock)
938 {
939         struct sock *sk = tfile->socket.sk;
940         struct sk_buff *skb;
941         int err;
942
943         /* Under a page?  Don't bother with paged skb. */
944         if (prepad + len < PAGE_SIZE || !linear)
945                 linear = len;
946
947         skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
948                                    &err);
949         if (!skb)
950                 return ERR_PTR(err);
951
952         skb_reserve(skb, prepad);
953         skb_put(skb, linear);
954         skb->data_len = len - linear;
955         skb->len += len - linear;
956
957         return skb;
958 }
959
960 /* set skb frags from iovec, this can move to core network code for reuse */
961 static int zerocopy_sg_from_iovec(struct sk_buff *skb, const struct iovec *from,
962                                   int offset, size_t count)
963 {
964         int len = iov_length(from, count) - offset;
965         int copy = skb_headlen(skb);
966         int size, offset1 = 0;
967         int i = 0;
968
969         /* Skip over from offset */
970         while (count && (offset >= from->iov_len)) {
971                 offset -= from->iov_len;
972                 ++from;
973                 --count;
974         }
975
976         /* copy up to skb headlen */
977         while (count && (copy > 0)) {
978                 size = min_t(unsigned int, copy, from->iov_len - offset);
979                 if (copy_from_user(skb->data + offset1, from->iov_base + offset,
980                                    size))
981                         return -EFAULT;
982                 if (copy > size) {
983                         ++from;
984                         --count;
985                         offset = 0;
986                 } else
987                         offset += size;
988                 copy -= size;
989                 offset1 += size;
990         }
991
992         if (len == offset1)
993                 return 0;
994
995         while (count--) {
996                 struct page *page[MAX_SKB_FRAGS];
997                 int num_pages;
998                 unsigned long base;
999                 unsigned long truesize;
1000
1001                 len = from->iov_len - offset;
1002                 if (!len) {
1003                         offset = 0;
1004                         ++from;
1005                         continue;
1006                 }
1007                 base = (unsigned long)from->iov_base + offset;
1008                 size = ((base & ~PAGE_MASK) + len + ~PAGE_MASK) >> PAGE_SHIFT;
1009                 if (i + size > MAX_SKB_FRAGS)
1010                         return -EMSGSIZE;
1011                 num_pages = get_user_pages_fast(base, size, 0, &page[i]);
1012                 if (num_pages != size) {
1013                         int j;
1014
1015                         for (j = 0; j < num_pages; j++)
1016                                 put_page(page[i + j]);
1017                         return -EFAULT;
1018                 }
1019                 truesize = size * PAGE_SIZE;
1020                 skb->data_len += len;
1021                 skb->len += len;
1022                 skb->truesize += truesize;
1023                 atomic_add(truesize, &skb->sk->sk_wmem_alloc);
1024                 while (len) {
1025                         int off = base & ~PAGE_MASK;
1026                         int size = min_t(int, len, PAGE_SIZE - off);
1027                         __skb_fill_page_desc(skb, i, page[i], off, size);
1028                         skb_shinfo(skb)->nr_frags++;
1029                         /* increase sk_wmem_alloc */
1030                         base += size;
1031                         len -= size;
1032                         i++;
1033                 }
1034                 offset = 0;
1035                 ++from;
1036         }
1037         return 0;
1038 }
1039
1040 static unsigned long iov_pages(const struct iovec *iv, int offset,
1041                                unsigned long nr_segs)
1042 {
1043         unsigned long seg, base;
1044         int pages = 0, len, size;
1045
1046         while (nr_segs && (offset >= iv->iov_len)) {
1047                 offset -= iv->iov_len;
1048                 ++iv;
1049                 --nr_segs;
1050         }
1051
1052         for (seg = 0; seg < nr_segs; seg++) {
1053                 base = (unsigned long)iv[seg].iov_base + offset;
1054                 len = iv[seg].iov_len - offset;
1055                 size = ((base & ~PAGE_MASK) + len + ~PAGE_MASK) >> PAGE_SHIFT;
1056                 pages += size;
1057                 offset = 0;
1058         }
1059
1060         return pages;
1061 }
1062
1063 /* Get packet from user space buffer */
1064 static ssize_t tun_get_user(struct tun_struct *tun, struct tun_file *tfile,
1065                             void *msg_control, const struct iovec *iv,
1066                             size_t total_len, size_t count, int noblock)
1067 {
1068         struct tun_pi pi = { 0, cpu_to_be16(ETH_P_IP) };
1069         struct sk_buff *skb;
1070         size_t len = total_len, align = NET_SKB_PAD, linear;
1071         struct virtio_net_hdr gso = { 0 };
1072         int offset = 0;
1073         int copylen;
1074         bool zerocopy = false;
1075         int err;
1076         u32 rxhash;
1077
1078         if (!(tun->flags & TUN_NO_PI)) {
1079                 if ((len -= sizeof(pi)) > total_len)
1080                         return -EINVAL;
1081
1082                 if (memcpy_fromiovecend((void *)&pi, iv, 0, sizeof(pi)))
1083                         return -EFAULT;
1084                 offset += sizeof(pi);
1085         }
1086
1087         if (tun->flags & TUN_VNET_HDR) {
1088                 if ((len -= tun->vnet_hdr_sz) > total_len)
1089                         return -EINVAL;
1090
1091                 if (memcpy_fromiovecend((void *)&gso, iv, offset, sizeof(gso)))
1092                         return -EFAULT;
1093
1094                 if ((gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
1095                     gso.csum_start + gso.csum_offset + 2 > gso.hdr_len)
1096                         gso.hdr_len = gso.csum_start + gso.csum_offset + 2;
1097
1098                 if (gso.hdr_len > len)
1099                         return -EINVAL;
1100                 offset += tun->vnet_hdr_sz;
1101         }
1102
1103         if ((tun->flags & TUN_TYPE_MASK) == TUN_TAP_DEV) {
1104                 align += NET_IP_ALIGN;
1105                 if (unlikely(len < ETH_HLEN ||
1106                              (gso.hdr_len && gso.hdr_len < ETH_HLEN)))
1107                         return -EINVAL;
1108         }
1109
1110         if (msg_control) {
1111                 /* There are 256 bytes to be copied in skb, so there is
1112                  * enough room for skb expand head in case it is used.
1113                  * The rest of the buffer is mapped from userspace.
1114                  */
1115                 copylen = gso.hdr_len ? gso.hdr_len : GOODCOPY_LEN;
1116                 linear = copylen;
1117                 if (iov_pages(iv, offset + copylen, count) <= MAX_SKB_FRAGS)
1118                         zerocopy = true;
1119         }
1120
1121         if (!zerocopy) {
1122                 copylen = len;
1123                 linear = gso.hdr_len;
1124         }
1125
1126         skb = tun_alloc_skb(tfile, align, copylen, linear, noblock);
1127         if (IS_ERR(skb)) {
1128                 if (PTR_ERR(skb) != -EAGAIN)
1129                         tun->dev->stats.rx_dropped++;
1130                 return PTR_ERR(skb);
1131         }
1132
1133         if (zerocopy)
1134                 err = zerocopy_sg_from_iovec(skb, iv, offset, count);
1135         else {
1136                 err = skb_copy_datagram_from_iovec(skb, 0, iv, offset, len);
1137                 if (!err && msg_control) {
1138                         struct ubuf_info *uarg = msg_control;
1139                         uarg->callback(uarg, false);
1140                 }
1141         }
1142
1143         if (err) {
1144                 tun->dev->stats.rx_dropped++;
1145                 kfree_skb(skb);
1146                 return -EFAULT;
1147         }
1148
1149         if (gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
1150                 if (!skb_partial_csum_set(skb, gso.csum_start,
1151                                           gso.csum_offset)) {
1152                         tun->dev->stats.rx_frame_errors++;
1153                         kfree_skb(skb);
1154                         return -EINVAL;
1155                 }
1156         }
1157
1158         switch (tun->flags & TUN_TYPE_MASK) {
1159         case TUN_TUN_DEV:
1160                 if (tun->flags & TUN_NO_PI) {
1161                         switch (skb->data[0] & 0xf0) {
1162                         case 0x40:
1163                                 pi.proto = htons(ETH_P_IP);
1164                                 break;
1165                         case 0x60:
1166                                 pi.proto = htons(ETH_P_IPV6);
1167                                 break;
1168                         default:
1169                                 tun->dev->stats.rx_dropped++;
1170                                 kfree_skb(skb);
1171                                 return -EINVAL;
1172                         }
1173                 }
1174
1175                 skb_reset_mac_header(skb);
1176                 skb->protocol = pi.proto;
1177                 skb->dev = tun->dev;
1178                 break;
1179         case TUN_TAP_DEV:
1180                 skb->protocol = eth_type_trans(skb, tun->dev);
1181                 break;
1182         }
1183
1184         if (gso.gso_type != VIRTIO_NET_HDR_GSO_NONE) {
1185                 pr_debug("GSO!\n");
1186                 switch (gso.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
1187                 case VIRTIO_NET_HDR_GSO_TCPV4:
1188                         skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
1189                         break;
1190                 case VIRTIO_NET_HDR_GSO_TCPV6:
1191                         skb_shinfo(skb)->gso_type = SKB_GSO_TCPV6;
1192                         break;
1193                 case VIRTIO_NET_HDR_GSO_UDP:
1194                         skb_shinfo(skb)->gso_type = SKB_GSO_UDP;
1195                         break;
1196                 default:
1197                         tun->dev->stats.rx_frame_errors++;
1198                         kfree_skb(skb);
1199                         return -EINVAL;
1200                 }
1201
1202                 if (gso.gso_type & VIRTIO_NET_HDR_GSO_ECN)
1203                         skb_shinfo(skb)->gso_type |= SKB_GSO_TCP_ECN;
1204
1205                 skb_shinfo(skb)->gso_size = gso.gso_size;
1206                 if (skb_shinfo(skb)->gso_size == 0) {
1207                         tun->dev->stats.rx_frame_errors++;
1208                         kfree_skb(skb);
1209                         return -EINVAL;
1210                 }
1211
1212                 /* Header must be checked, and gso_segs computed. */
1213                 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
1214                 skb_shinfo(skb)->gso_segs = 0;
1215         }
1216
1217         /* copy skb_ubuf_info for callback when skb has no error */
1218         if (zerocopy) {
1219                 skb_shinfo(skb)->destructor_arg = msg_control;
1220                 skb_shinfo(skb)->tx_flags |= SKBTX_DEV_ZEROCOPY;
1221                 skb_shinfo(skb)->tx_flags |= SKBTX_SHARED_FRAG;
1222         }
1223
1224         skb_reset_network_header(skb);
1225         skb_probe_transport_header(skb, 0);
1226
1227         rxhash = skb_get_rxhash(skb);
1228         netif_rx_ni(skb);
1229
1230         tun->dev->stats.rx_packets++;
1231         tun->dev->stats.rx_bytes += len;
1232
1233         tun_flow_update(tun, rxhash, tfile);
1234         return total_len;
1235 }
1236
1237 static ssize_t tun_chr_aio_write(struct kiocb *iocb, const struct iovec *iv,
1238                               unsigned long count, loff_t pos)
1239 {
1240         struct file *file = iocb->ki_filp;
1241         struct tun_struct *tun = tun_get(file);
1242         struct tun_file *tfile = file->private_data;
1243         ssize_t result;
1244
1245         if (!tun)
1246                 return -EBADFD;
1247
1248         tun_debug(KERN_INFO, tun, "tun_chr_write %ld\n", count);
1249
1250         result = tun_get_user(tun, tfile, NULL, iv, iov_length(iv, count),
1251                               count, file->f_flags & O_NONBLOCK);
1252
1253         tun_put(tun);
1254         return result;
1255 }
1256
1257 /* Put packet to the user space buffer */
1258 static ssize_t tun_put_user(struct tun_struct *tun,
1259                             struct tun_file *tfile,
1260                             struct sk_buff *skb,
1261                             const struct iovec *iv, int len)
1262 {
1263         struct tun_pi pi = { 0, skb->protocol };
1264         ssize_t total = 0;
1265
1266         if (!(tun->flags & TUN_NO_PI)) {
1267                 if ((len -= sizeof(pi)) < 0)
1268                         return -EINVAL;
1269
1270                 if (len < skb->len) {
1271                         /* Packet will be striped */
1272                         pi.flags |= TUN_PKT_STRIP;
1273                 }
1274
1275                 if (memcpy_toiovecend(iv, (void *) &pi, 0, sizeof(pi)))
1276                         return -EFAULT;
1277                 total += sizeof(pi);
1278         }
1279
1280         if (tun->flags & TUN_VNET_HDR) {
1281                 struct virtio_net_hdr gso = { 0 }; /* no info leak */
1282                 if ((len -= tun->vnet_hdr_sz) < 0)
1283                         return -EINVAL;
1284
1285                 if (skb_is_gso(skb)) {
1286                         struct skb_shared_info *sinfo = skb_shinfo(skb);
1287
1288                         /* This is a hint as to how much should be linear. */
1289                         gso.hdr_len = skb_headlen(skb);
1290                         gso.gso_size = sinfo->gso_size;
1291                         if (sinfo->gso_type & SKB_GSO_TCPV4)
1292                                 gso.gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
1293                         else if (sinfo->gso_type & SKB_GSO_TCPV6)
1294                                 gso.gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
1295                         else if (sinfo->gso_type & SKB_GSO_UDP)
1296                                 gso.gso_type = VIRTIO_NET_HDR_GSO_UDP;
1297                         else {
1298                                 pr_err("unexpected GSO type: "
1299                                        "0x%x, gso_size %d, hdr_len %d\n",
1300                                        sinfo->gso_type, gso.gso_size,
1301                                        gso.hdr_len);
1302                                 print_hex_dump(KERN_ERR, "tun: ",
1303                                                DUMP_PREFIX_NONE,
1304                                                16, 1, skb->head,
1305                                                min((int)gso.hdr_len, 64), true);
1306                                 WARN_ON_ONCE(1);
1307                                 return -EINVAL;
1308                         }
1309                         if (sinfo->gso_type & SKB_GSO_TCP_ECN)
1310                                 gso.gso_type |= VIRTIO_NET_HDR_GSO_ECN;
1311                 } else
1312                         gso.gso_type = VIRTIO_NET_HDR_GSO_NONE;
1313
1314                 if (skb->ip_summed == CHECKSUM_PARTIAL) {
1315                         gso.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
1316                         gso.csum_start = skb_checksum_start_offset(skb);
1317                         gso.csum_offset = skb->csum_offset;
1318                 } else if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
1319                         gso.flags = VIRTIO_NET_HDR_F_DATA_VALID;
1320                 } /* else everything is zero */
1321
1322                 if (unlikely(memcpy_toiovecend(iv, (void *)&gso, total,
1323                                                sizeof(gso))))
1324                         return -EFAULT;
1325                 total += tun->vnet_hdr_sz;
1326         }
1327
1328         len = min_t(int, skb->len, len);
1329
1330         skb_copy_datagram_const_iovec(skb, 0, iv, total, len);
1331         total += skb->len;
1332
1333         tun->dev->stats.tx_packets++;
1334         tun->dev->stats.tx_bytes += len;
1335
1336         return total;
1337 }
1338
1339 static ssize_t tun_do_read(struct tun_struct *tun, struct tun_file *tfile,
1340                            struct kiocb *iocb, const struct iovec *iv,
1341                            ssize_t len, int noblock)
1342 {
1343         DECLARE_WAITQUEUE(wait, current);
1344         struct sk_buff *skb;
1345         ssize_t ret = 0;
1346
1347         tun_debug(KERN_INFO, tun, "tun_do_read\n");
1348
1349         if (unlikely(!noblock))
1350                 add_wait_queue(&tfile->wq.wait, &wait);
1351         while (len) {
1352                 current->state = TASK_INTERRUPTIBLE;
1353
1354                 /* Read frames from the queue */
1355                 if (!(skb = skb_dequeue(&tfile->socket.sk->sk_receive_queue))) {
1356                         if (noblock) {
1357                                 ret = -EAGAIN;
1358                                 break;
1359                         }
1360                         if (signal_pending(current)) {
1361                                 ret = -ERESTARTSYS;
1362                                 break;
1363                         }
1364                         if (tun->dev->reg_state != NETREG_REGISTERED) {
1365                                 ret = -EIO;
1366                                 break;
1367                         }
1368
1369                         /* Nothing to read, let's sleep */
1370                         schedule();
1371                         continue;
1372                 }
1373
1374                 ret = tun_put_user(tun, tfile, skb, iv, len);
1375                 kfree_skb(skb);
1376                 break;
1377         }
1378
1379         current->state = TASK_RUNNING;
1380         if (unlikely(!noblock))
1381                 remove_wait_queue(&tfile->wq.wait, &wait);
1382
1383         return ret;
1384 }
1385
1386 static ssize_t tun_chr_aio_read(struct kiocb *iocb, const struct iovec *iv,
1387                             unsigned long count, loff_t pos)
1388 {
1389         struct file *file = iocb->ki_filp;
1390         struct tun_file *tfile = file->private_data;
1391         struct tun_struct *tun = __tun_get(tfile);
1392         ssize_t len, ret;
1393
1394         if (!tun)
1395                 return -EBADFD;
1396         len = iov_length(iv, count);
1397         if (len < 0) {
1398                 ret = -EINVAL;
1399                 goto out;
1400         }
1401
1402         ret = tun_do_read(tun, tfile, iocb, iv, len,
1403                           file->f_flags & O_NONBLOCK);
1404         ret = min_t(ssize_t, ret, len);
1405 out:
1406         tun_put(tun);
1407         return ret;
1408 }
1409
1410 static void tun_free_netdev(struct net_device *dev)
1411 {
1412         struct tun_struct *tun = netdev_priv(dev);
1413
1414         BUG_ON(!(list_empty(&tun->disabled)));
1415         tun_flow_uninit(tun);
1416         security_tun_dev_free_security(tun->security);
1417         free_netdev(dev);
1418 }
1419
1420 static void tun_setup(struct net_device *dev)
1421 {
1422         struct tun_struct *tun = netdev_priv(dev);
1423
1424         tun->owner = INVALID_UID;
1425         tun->group = INVALID_GID;
1426
1427         dev->ethtool_ops = &tun_ethtool_ops;
1428         dev->destructor = tun_free_netdev;
1429 }
1430
1431 /* Trivial set of netlink ops to allow deleting tun or tap
1432  * device with netlink.
1433  */
1434 static int tun_validate(struct nlattr *tb[], struct nlattr *data[])
1435 {
1436         return -EINVAL;
1437 }
1438
1439 static struct rtnl_link_ops tun_link_ops __read_mostly = {
1440         .kind           = DRV_NAME,
1441         .priv_size      = sizeof(struct tun_struct),
1442         .setup          = tun_setup,
1443         .validate       = tun_validate,
1444 };
1445
1446 static void tun_sock_write_space(struct sock *sk)
1447 {
1448         struct tun_file *tfile;
1449         wait_queue_head_t *wqueue;
1450
1451         if (!sock_writeable(sk))
1452                 return;
1453
1454         if (!test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags))
1455                 return;
1456
1457         wqueue = sk_sleep(sk);
1458         if (wqueue && waitqueue_active(wqueue))
1459                 wake_up_interruptible_sync_poll(wqueue, POLLOUT |
1460                                                 POLLWRNORM | POLLWRBAND);
1461
1462         tfile = container_of(sk, struct tun_file, sk);
1463         kill_fasync(&tfile->fasync, SIGIO, POLL_OUT);
1464 }
1465
1466 static int tun_sendmsg(struct kiocb *iocb, struct socket *sock,
1467                        struct msghdr *m, size_t total_len)
1468 {
1469         int ret;
1470         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
1471         struct tun_struct *tun = __tun_get(tfile);
1472
1473         if (!tun)
1474                 return -EBADFD;
1475         ret = tun_get_user(tun, tfile, m->msg_control, m->msg_iov, total_len,
1476                            m->msg_iovlen, m->msg_flags & MSG_DONTWAIT);
1477         tun_put(tun);
1478         return ret;
1479 }
1480
1481
1482 static int tun_recvmsg(struct kiocb *iocb, struct socket *sock,
1483                        struct msghdr *m, size_t total_len,
1484                        int flags)
1485 {
1486         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
1487         struct tun_struct *tun = __tun_get(tfile);
1488         int ret;
1489
1490         if (!tun)
1491                 return -EBADFD;
1492
1493         if (flags & ~(MSG_DONTWAIT|MSG_TRUNC)) {
1494                 ret = -EINVAL;
1495                 goto out;
1496         }
1497         ret = tun_do_read(tun, tfile, iocb, m->msg_iov, total_len,
1498                           flags & MSG_DONTWAIT);
1499         if (ret > total_len) {
1500                 m->msg_flags |= MSG_TRUNC;
1501                 ret = flags & MSG_TRUNC ? ret : total_len;
1502         }
1503 out:
1504         tun_put(tun);
1505         return ret;
1506 }
1507
1508 static int tun_release(struct socket *sock)
1509 {
1510         if (sock->sk)
1511                 sock_put(sock->sk);
1512         return 0;
1513 }
1514
1515 /* Ops structure to mimic raw sockets with tun */
1516 static const struct proto_ops tun_socket_ops = {
1517         .sendmsg = tun_sendmsg,
1518         .recvmsg = tun_recvmsg,
1519         .release = tun_release,
1520 };
1521
1522 static struct proto tun_proto = {
1523         .name           = "tun",
1524         .owner          = THIS_MODULE,
1525         .obj_size       = sizeof(struct tun_file),
1526 };
1527
1528 static int tun_flags(struct tun_struct *tun)
1529 {
1530         int flags = 0;
1531
1532         if (tun->flags & TUN_TUN_DEV)
1533                 flags |= IFF_TUN;
1534         else
1535                 flags |= IFF_TAP;
1536
1537         if (tun->flags & TUN_NO_PI)
1538                 flags |= IFF_NO_PI;
1539
1540         /* This flag has no real effect.  We track the value for backwards
1541          * compatibility.
1542          */
1543         if (tun->flags & TUN_ONE_QUEUE)
1544                 flags |= IFF_ONE_QUEUE;
1545
1546         if (tun->flags & TUN_VNET_HDR)
1547                 flags |= IFF_VNET_HDR;
1548
1549         if (tun->flags & TUN_TAP_MQ)
1550                 flags |= IFF_MULTI_QUEUE;
1551
1552         return flags;
1553 }
1554
1555 static ssize_t tun_show_flags(struct device *dev, struct device_attribute *attr,
1556                               char *buf)
1557 {
1558         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1559         return sprintf(buf, "0x%x\n", tun_flags(tun));
1560 }
1561
1562 static ssize_t tun_show_owner(struct device *dev, struct device_attribute *attr,
1563                               char *buf)
1564 {
1565         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1566         return uid_valid(tun->owner)?
1567                 sprintf(buf, "%u\n",
1568                         from_kuid_munged(current_user_ns(), tun->owner)):
1569                 sprintf(buf, "-1\n");
1570 }
1571
1572 static ssize_t tun_show_group(struct device *dev, struct device_attribute *attr,
1573                               char *buf)
1574 {
1575         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1576         return gid_valid(tun->group) ?
1577                 sprintf(buf, "%u\n",
1578                         from_kgid_munged(current_user_ns(), tun->group)):
1579                 sprintf(buf, "-1\n");
1580 }
1581
1582 static DEVICE_ATTR(tun_flags, 0444, tun_show_flags, NULL);
1583 static DEVICE_ATTR(owner, 0444, tun_show_owner, NULL);
1584 static DEVICE_ATTR(group, 0444, tun_show_group, NULL);
1585
1586 static int tun_set_iff(struct net *net, struct file *file, struct ifreq *ifr)
1587 {
1588         struct tun_struct *tun;
1589         struct tun_file *tfile = file->private_data;
1590         struct net_device *dev;
1591         int err;
1592
1593         if (tfile->detached)
1594                 return -EINVAL;
1595
1596         dev = __dev_get_by_name(net, ifr->ifr_name);
1597         if (dev) {
1598                 if (ifr->ifr_flags & IFF_TUN_EXCL)
1599                         return -EBUSY;
1600                 if ((ifr->ifr_flags & IFF_TUN) && dev->netdev_ops == &tun_netdev_ops)
1601                         tun = netdev_priv(dev);
1602                 else if ((ifr->ifr_flags & IFF_TAP) && dev->netdev_ops == &tap_netdev_ops)
1603                         tun = netdev_priv(dev);
1604                 else
1605                         return -EINVAL;
1606
1607                 if (!!(ifr->ifr_flags & IFF_MULTI_QUEUE) !=
1608                     !!(tun->flags & TUN_TAP_MQ))
1609                         return -EINVAL;
1610
1611                 if (tun_not_capable(tun))
1612                         return -EPERM;
1613                 err = security_tun_dev_open(tun->security);
1614                 if (err < 0)
1615                         return err;
1616
1617                 err = tun_attach(tun, file);
1618                 if (err < 0)
1619                         return err;
1620
1621                 if (tun->flags & TUN_TAP_MQ &&
1622                     (tun->numqueues + tun->numdisabled > 1)) {
1623                         /* One or more queue has already been attached, no need
1624                          * to initialize the device again.
1625                          */
1626                         return 0;
1627                 }
1628         }
1629         else {
1630                 char *name;
1631                 unsigned long flags = 0;
1632                 int queues = ifr->ifr_flags & IFF_MULTI_QUEUE ?
1633                              MAX_TAP_QUEUES : 1;
1634
1635                 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
1636                         return -EPERM;
1637                 err = security_tun_dev_create();
1638                 if (err < 0)
1639                         return err;
1640
1641                 /* Set dev type */
1642                 if (ifr->ifr_flags & IFF_TUN) {
1643                         /* TUN device */
1644                         flags |= TUN_TUN_DEV;
1645                         name = "tun%d";
1646                 } else if (ifr->ifr_flags & IFF_TAP) {
1647                         /* TAP device */
1648                         flags |= TUN_TAP_DEV;
1649                         name = "tap%d";
1650                 } else
1651                         return -EINVAL;
1652
1653                 if (*ifr->ifr_name)
1654                         name = ifr->ifr_name;
1655
1656                 dev = alloc_netdev_mqs(sizeof(struct tun_struct), name,
1657                                        tun_setup, queues, queues);
1658
1659                 if (!dev)
1660                         return -ENOMEM;
1661
1662                 dev_net_set(dev, net);
1663                 dev->rtnl_link_ops = &tun_link_ops;
1664
1665                 tun = netdev_priv(dev);
1666                 tun->dev = dev;
1667                 tun->flags = flags;
1668                 tun->txflt.count = 0;
1669                 tun->vnet_hdr_sz = sizeof(struct virtio_net_hdr);
1670
1671                 tun->filter_attached = false;
1672                 tun->sndbuf = tfile->socket.sk->sk_sndbuf;
1673
1674                 spin_lock_init(&tun->lock);
1675
1676                 err = security_tun_dev_alloc_security(&tun->security);
1677                 if (err < 0)
1678                         goto err_free_dev;
1679
1680                 tun_net_init(dev);
1681
1682                 err = tun_flow_init(tun);
1683                 if (err < 0)
1684                         goto err_free_dev;
1685
1686                 dev->hw_features = NETIF_F_SG | NETIF_F_FRAGLIST |
1687                         TUN_USER_FEATURES;
1688                 dev->features = dev->hw_features;
1689                 dev->vlan_features = dev->features;
1690
1691                 INIT_LIST_HEAD(&tun->disabled);
1692                 err = tun_attach(tun, file);
1693                 if (err < 0)
1694                         goto err_free_dev;
1695
1696                 err = register_netdevice(tun->dev);
1697                 if (err < 0)
1698                         goto err_free_dev;
1699
1700                 if (device_create_file(&tun->dev->dev, &dev_attr_tun_flags) ||
1701                     device_create_file(&tun->dev->dev, &dev_attr_owner) ||
1702                     device_create_file(&tun->dev->dev, &dev_attr_group))
1703                         pr_err("Failed to create tun sysfs files\n");
1704         }
1705
1706         netif_carrier_on(tun->dev);
1707
1708         tun_debug(KERN_INFO, tun, "tun_set_iff\n");
1709
1710         if (ifr->ifr_flags & IFF_NO_PI)
1711                 tun->flags |= TUN_NO_PI;
1712         else
1713                 tun->flags &= ~TUN_NO_PI;
1714
1715         /* This flag has no real effect.  We track the value for backwards
1716          * compatibility.
1717          */
1718         if (ifr->ifr_flags & IFF_ONE_QUEUE)
1719                 tun->flags |= TUN_ONE_QUEUE;
1720         else
1721                 tun->flags &= ~TUN_ONE_QUEUE;
1722
1723         if (ifr->ifr_flags & IFF_VNET_HDR)
1724                 tun->flags |= TUN_VNET_HDR;
1725         else
1726                 tun->flags &= ~TUN_VNET_HDR;
1727
1728         if (ifr->ifr_flags & IFF_MULTI_QUEUE)
1729                 tun->flags |= TUN_TAP_MQ;
1730         else
1731                 tun->flags &= ~TUN_TAP_MQ;
1732
1733         /* Make sure persistent devices do not get stuck in
1734          * xoff state.
1735          */
1736         if (netif_running(tun->dev))
1737                 netif_tx_wake_all_queues(tun->dev);
1738
1739         strcpy(ifr->ifr_name, tun->dev->name);
1740         return 0;
1741
1742  err_free_dev:
1743         free_netdev(dev);
1744         return err;
1745 }
1746
1747 static void tun_get_iff(struct net *net, struct tun_struct *tun,
1748                        struct ifreq *ifr)
1749 {
1750         tun_debug(KERN_INFO, tun, "tun_get_iff\n");
1751
1752         strcpy(ifr->ifr_name, tun->dev->name);
1753
1754         ifr->ifr_flags = tun_flags(tun);
1755
1756 }
1757
1758 /* This is like a cut-down ethtool ops, except done via tun fd so no
1759  * privs required. */
1760 static int set_offload(struct tun_struct *tun, unsigned long arg)
1761 {
1762         netdev_features_t features = 0;
1763
1764         if (arg & TUN_F_CSUM) {
1765                 features |= NETIF_F_HW_CSUM;
1766                 arg &= ~TUN_F_CSUM;
1767
1768                 if (arg & (TUN_F_TSO4|TUN_F_TSO6)) {
1769                         if (arg & TUN_F_TSO_ECN) {
1770                                 features |= NETIF_F_TSO_ECN;
1771                                 arg &= ~TUN_F_TSO_ECN;
1772                         }
1773                         if (arg & TUN_F_TSO4)
1774                                 features |= NETIF_F_TSO;
1775                         if (arg & TUN_F_TSO6)
1776                                 features |= NETIF_F_TSO6;
1777                         arg &= ~(TUN_F_TSO4|TUN_F_TSO6);
1778                 }
1779
1780                 if (arg & TUN_F_UFO) {
1781                         features |= NETIF_F_UFO;
1782                         arg &= ~TUN_F_UFO;
1783                 }
1784         }
1785
1786         /* This gives the user a way to test for new features in future by
1787          * trying to set them. */
1788         if (arg)
1789                 return -EINVAL;
1790
1791         tun->set_features = features;
1792         netdev_update_features(tun->dev);
1793
1794         return 0;
1795 }
1796
1797 static void tun_detach_filter(struct tun_struct *tun, int n)
1798 {
1799         int i;
1800         struct tun_file *tfile;
1801
1802         for (i = 0; i < n; i++) {
1803                 tfile = rtnl_dereference(tun->tfiles[i]);
1804                 sk_detach_filter(tfile->socket.sk);
1805         }
1806
1807         tun->filter_attached = false;
1808 }
1809
1810 static int tun_attach_filter(struct tun_struct *tun)
1811 {
1812         int i, ret = 0;
1813         struct tun_file *tfile;
1814
1815         for (i = 0; i < tun->numqueues; i++) {
1816                 tfile = rtnl_dereference(tun->tfiles[i]);
1817                 ret = sk_attach_filter(&tun->fprog, tfile->socket.sk);
1818                 if (ret) {
1819                         tun_detach_filter(tun, i);
1820                         return ret;
1821                 }
1822         }
1823
1824         tun->filter_attached = true;
1825         return ret;
1826 }
1827
1828 static void tun_set_sndbuf(struct tun_struct *tun)
1829 {
1830         struct tun_file *tfile;
1831         int i;
1832
1833         for (i = 0; i < tun->numqueues; i++) {
1834                 tfile = rtnl_dereference(tun->tfiles[i]);
1835                 tfile->socket.sk->sk_sndbuf = tun->sndbuf;
1836         }
1837 }
1838
1839 static int tun_set_queue(struct file *file, struct ifreq *ifr)
1840 {
1841         struct tun_file *tfile = file->private_data;
1842         struct tun_struct *tun;
1843         int ret = 0;
1844
1845         rtnl_lock();
1846
1847         if (ifr->ifr_flags & IFF_ATTACH_QUEUE) {
1848                 tun = tfile->detached;
1849                 if (!tun) {
1850                         ret = -EINVAL;
1851                         goto unlock;
1852                 }
1853                 ret = security_tun_dev_attach_queue(tun->security);
1854                 if (ret < 0)
1855                         goto unlock;
1856                 ret = tun_attach(tun, file);
1857         } else if (ifr->ifr_flags & IFF_DETACH_QUEUE) {
1858                 tun = rtnl_dereference(tfile->tun);
1859                 if (!tun || !(tun->flags & TUN_TAP_MQ) || tfile->detached)
1860                         ret = -EINVAL;
1861                 else
1862                         __tun_detach(tfile, false);
1863         } else
1864                 ret = -EINVAL;
1865
1866 unlock:
1867         rtnl_unlock();
1868         return ret;
1869 }
1870
1871 static long __tun_chr_ioctl(struct file *file, unsigned int cmd,
1872                             unsigned long arg, int ifreq_len)
1873 {
1874         struct tun_file *tfile = file->private_data;
1875         struct tun_struct *tun;
1876         void __user* argp = (void __user*)arg;
1877         struct ifreq ifr;
1878         kuid_t owner;
1879         kgid_t group;
1880         int sndbuf;
1881         int vnet_hdr_sz;
1882         int ret;
1883
1884 #ifdef CONFIG_ANDROID_PARANOID_NETWORK
1885         if (cmd != TUNGETIFF && !capable(CAP_NET_ADMIN)) {
1886                 return -EPERM;
1887         }
1888 #endif
1889
1890         if (cmd == TUNSETIFF || cmd == TUNSETQUEUE || _IOC_TYPE(cmd) == 0x89) {
1891                 if (copy_from_user(&ifr, argp, ifreq_len))
1892                         return -EFAULT;
1893         } else {
1894                 memset(&ifr, 0, sizeof(ifr));
1895         }
1896         if (cmd == TUNGETFEATURES) {
1897                 /* Currently this just means: "what IFF flags are valid?".
1898                  * This is needed because we never checked for invalid flags on
1899                  * TUNSETIFF. */
1900                 return put_user(IFF_TUN | IFF_TAP | IFF_NO_PI | IFF_ONE_QUEUE |
1901                                 IFF_VNET_HDR | IFF_MULTI_QUEUE,
1902                                 (unsigned int __user*)argp);
1903         } else if (cmd == TUNSETQUEUE)
1904                 return tun_set_queue(file, &ifr);
1905
1906         ret = 0;
1907         rtnl_lock();
1908
1909         tun = __tun_get(tfile);
1910         if (cmd == TUNSETIFF && !tun) {
1911                 ifr.ifr_name[IFNAMSIZ-1] = '\0';
1912
1913                 ret = tun_set_iff(tfile->net, file, &ifr);
1914
1915                 if (ret)
1916                         goto unlock;
1917
1918                 if (copy_to_user(argp, &ifr, ifreq_len))
1919                         ret = -EFAULT;
1920                 goto unlock;
1921         }
1922
1923         ret = -EBADFD;
1924         if (!tun)
1925                 goto unlock;
1926
1927         tun_debug(KERN_INFO, tun, "tun_chr_ioctl cmd %u\n", cmd);
1928
1929         ret = 0;
1930         switch (cmd) {
1931         case TUNGETIFF:
1932                 tun_get_iff(current->nsproxy->net_ns, tun, &ifr);
1933
1934                 if (copy_to_user(argp, &ifr, ifreq_len))
1935                         ret = -EFAULT;
1936                 break;
1937
1938         case TUNSETNOCSUM:
1939                 /* Disable/Enable checksum */
1940
1941                 /* [unimplemented] */
1942                 tun_debug(KERN_INFO, tun, "ignored: set checksum %s\n",
1943                           arg ? "disabled" : "enabled");
1944                 break;
1945
1946         case TUNSETPERSIST:
1947                 /* Disable/Enable persist mode. Keep an extra reference to the
1948                  * module to prevent the module being unprobed.
1949                  */
1950                 if (arg && !(tun->flags & TUN_PERSIST)) {
1951                         tun->flags |= TUN_PERSIST;
1952                         __module_get(THIS_MODULE);
1953                 }
1954                 if (!arg && (tun->flags & TUN_PERSIST)) {
1955                         tun->flags &= ~TUN_PERSIST;
1956                         module_put(THIS_MODULE);
1957                 }
1958
1959                 tun_debug(KERN_INFO, tun, "persist %s\n",
1960                           arg ? "enabled" : "disabled");
1961                 break;
1962
1963         case TUNSETOWNER:
1964                 /* Set owner of the device */
1965                 owner = make_kuid(current_user_ns(), arg);
1966                 if (!uid_valid(owner)) {
1967                         ret = -EINVAL;
1968                         break;
1969                 }
1970                 tun->owner = owner;
1971                 tun_debug(KERN_INFO, tun, "owner set to %u\n",
1972                           from_kuid(&init_user_ns, tun->owner));
1973                 break;
1974
1975         case TUNSETGROUP:
1976                 /* Set group of the device */
1977                 group = make_kgid(current_user_ns(), arg);
1978                 if (!gid_valid(group)) {
1979                         ret = -EINVAL;
1980                         break;
1981                 }
1982                 tun->group = group;
1983                 tun_debug(KERN_INFO, tun, "group set to %u\n",
1984                           from_kgid(&init_user_ns, tun->group));
1985                 break;
1986
1987         case TUNSETLINK:
1988                 /* Only allow setting the type when the interface is down */
1989                 if (tun->dev->flags & IFF_UP) {
1990                         tun_debug(KERN_INFO, tun,
1991                                   "Linktype set failed because interface is up\n");
1992                         ret = -EBUSY;
1993                 } else {
1994                         tun->dev->type = (int) arg;
1995                         tun_debug(KERN_INFO, tun, "linktype set to %d\n",
1996                                   tun->dev->type);
1997                         ret = 0;
1998                 }
1999                 break;
2000
2001 #ifdef TUN_DEBUG
2002         case TUNSETDEBUG:
2003                 tun->debug = arg;
2004                 break;
2005 #endif
2006         case TUNSETOFFLOAD:
2007                 ret = set_offload(tun, arg);
2008                 break;
2009
2010         case TUNSETTXFILTER:
2011                 /* Can be set only for TAPs */
2012                 ret = -EINVAL;
2013                 if ((tun->flags & TUN_TYPE_MASK) != TUN_TAP_DEV)
2014                         break;
2015                 ret = update_filter(&tun->txflt, (void __user *)arg);
2016                 break;
2017
2018         case SIOCGIFHWADDR:
2019                 /* Get hw address */
2020                 memcpy(ifr.ifr_hwaddr.sa_data, tun->dev->dev_addr, ETH_ALEN);
2021                 ifr.ifr_hwaddr.sa_family = tun->dev->type;
2022                 if (copy_to_user(argp, &ifr, ifreq_len))
2023                         ret = -EFAULT;
2024                 break;
2025
2026         case SIOCSIFHWADDR:
2027                 /* Set hw address */
2028                 tun_debug(KERN_DEBUG, tun, "set hw address: %pM\n",
2029                           ifr.ifr_hwaddr.sa_data);
2030
2031                 ret = dev_set_mac_address(tun->dev, &ifr.ifr_hwaddr);
2032                 break;
2033
2034         case TUNGETSNDBUF:
2035                 sndbuf = tfile->socket.sk->sk_sndbuf;
2036                 if (copy_to_user(argp, &sndbuf, sizeof(sndbuf)))
2037                         ret = -EFAULT;
2038                 break;
2039
2040         case TUNSETSNDBUF:
2041                 if (copy_from_user(&sndbuf, argp, sizeof(sndbuf))) {
2042                         ret = -EFAULT;
2043                         break;
2044                 }
2045
2046                 tun->sndbuf = sndbuf;
2047                 tun_set_sndbuf(tun);
2048                 break;
2049
2050         case TUNGETVNETHDRSZ:
2051                 vnet_hdr_sz = tun->vnet_hdr_sz;
2052                 if (copy_to_user(argp, &vnet_hdr_sz, sizeof(vnet_hdr_sz)))
2053                         ret = -EFAULT;
2054                 break;
2055
2056         case TUNSETVNETHDRSZ:
2057                 if (copy_from_user(&vnet_hdr_sz, argp, sizeof(vnet_hdr_sz))) {
2058                         ret = -EFAULT;
2059                         break;
2060                 }
2061                 if (vnet_hdr_sz < (int)sizeof(struct virtio_net_hdr)) {
2062                         ret = -EINVAL;
2063                         break;
2064                 }
2065
2066                 tun->vnet_hdr_sz = vnet_hdr_sz;
2067                 break;
2068
2069         case TUNATTACHFILTER:
2070                 /* Can be set only for TAPs */
2071                 ret = -EINVAL;
2072                 if ((tun->flags & TUN_TYPE_MASK) != TUN_TAP_DEV)
2073                         break;
2074                 ret = -EFAULT;
2075                 if (copy_from_user(&tun->fprog, argp, sizeof(tun->fprog)))
2076                         break;
2077
2078                 ret = tun_attach_filter(tun);
2079                 break;
2080
2081         case TUNDETACHFILTER:
2082                 /* Can be set only for TAPs */
2083                 ret = -EINVAL;
2084                 if ((tun->flags & TUN_TYPE_MASK) != TUN_TAP_DEV)
2085                         break;
2086                 ret = 0;
2087                 tun_detach_filter(tun, tun->numqueues);
2088                 break;
2089
2090         default:
2091                 ret = -EINVAL;
2092                 break;
2093         }
2094
2095 unlock:
2096         rtnl_unlock();
2097         if (tun)
2098                 tun_put(tun);
2099         return ret;
2100 }
2101
2102 static long tun_chr_ioctl(struct file *file,
2103                           unsigned int cmd, unsigned long arg)
2104 {
2105         return __tun_chr_ioctl(file, cmd, arg, sizeof (struct ifreq));
2106 }
2107
2108 #ifdef CONFIG_COMPAT
2109 static long tun_chr_compat_ioctl(struct file *file,
2110                          unsigned int cmd, unsigned long arg)
2111 {
2112         switch (cmd) {
2113         case TUNSETIFF:
2114         case TUNGETIFF:
2115         case TUNSETTXFILTER:
2116         case TUNGETSNDBUF:
2117         case TUNSETSNDBUF:
2118         case SIOCGIFHWADDR:
2119         case SIOCSIFHWADDR:
2120                 arg = (unsigned long)compat_ptr(arg);
2121                 break;
2122         default:
2123                 arg = (compat_ulong_t)arg;
2124                 break;
2125         }
2126
2127         /*
2128          * compat_ifreq is shorter than ifreq, so we must not access beyond
2129          * the end of that structure. All fields that are used in this
2130          * driver are compatible though, we don't need to convert the
2131          * contents.
2132          */
2133         return __tun_chr_ioctl(file, cmd, arg, sizeof(struct compat_ifreq));
2134 }
2135 #endif /* CONFIG_COMPAT */
2136
2137 static int tun_chr_fasync(int fd, struct file *file, int on)
2138 {
2139         struct tun_file *tfile = file->private_data;
2140         int ret;
2141
2142         if ((ret = fasync_helper(fd, file, on, &tfile->fasync)) < 0)
2143                 goto out;
2144
2145         if (on) {
2146                 ret = __f_setown(file, task_pid(current), PIDTYPE_PID, 0);
2147                 if (ret)
2148                         goto out;
2149                 tfile->flags |= TUN_FASYNC;
2150         } else
2151                 tfile->flags &= ~TUN_FASYNC;
2152         ret = 0;
2153 out:
2154         return ret;
2155 }
2156
2157 static int tun_chr_open(struct inode *inode, struct file * file)
2158 {
2159         struct tun_file *tfile;
2160
2161         DBG1(KERN_INFO, "tunX: tun_chr_open\n");
2162
2163         tfile = (struct tun_file *)sk_alloc(&init_net, AF_UNSPEC, GFP_KERNEL,
2164                                             &tun_proto);
2165         if (!tfile)
2166                 return -ENOMEM;
2167         rcu_assign_pointer(tfile->tun, NULL);
2168         tfile->net = get_net(current->nsproxy->net_ns);
2169         tfile->flags = 0;
2170
2171         rcu_assign_pointer(tfile->socket.wq, &tfile->wq);
2172         init_waitqueue_head(&tfile->wq.wait);
2173
2174         tfile->socket.file = file;
2175         tfile->socket.ops = &tun_socket_ops;
2176
2177         sock_init_data(&tfile->socket, &tfile->sk);
2178         sk_change_net(&tfile->sk, tfile->net);
2179
2180         tfile->sk.sk_write_space = tun_sock_write_space;
2181         tfile->sk.sk_sndbuf = INT_MAX;
2182
2183         file->private_data = tfile;
2184         set_bit(SOCK_EXTERNALLY_ALLOCATED, &tfile->socket.flags);
2185         INIT_LIST_HEAD(&tfile->next);
2186
2187         sock_set_flag(&tfile->sk, SOCK_ZEROCOPY);
2188
2189         return 0;
2190 }
2191
2192 static int tun_chr_close(struct inode *inode, struct file *file)
2193 {
2194         struct tun_file *tfile = file->private_data;
2195         struct net *net = tfile->net;
2196
2197         tun_detach(tfile, true);
2198         put_net(net);
2199
2200         return 0;
2201 }
2202
2203 static const struct file_operations tun_fops = {
2204         .owner  = THIS_MODULE,
2205         .llseek = no_llseek,
2206         .read  = do_sync_read,
2207         .aio_read  = tun_chr_aio_read,
2208         .write = do_sync_write,
2209         .aio_write = tun_chr_aio_write,
2210         .poll   = tun_chr_poll,
2211         .unlocked_ioctl = tun_chr_ioctl,
2212 #ifdef CONFIG_COMPAT
2213         .compat_ioctl = tun_chr_compat_ioctl,
2214 #endif
2215         .open   = tun_chr_open,
2216         .release = tun_chr_close,
2217         .fasync = tun_chr_fasync
2218 };
2219
2220 static struct miscdevice tun_miscdev = {
2221         .minor = TUN_MINOR,
2222         .name = "tun",
2223         .nodename = "net/tun",
2224         .fops = &tun_fops,
2225 };
2226
2227 /* ethtool interface */
2228
2229 static int tun_get_settings(struct net_device *dev, struct ethtool_cmd *cmd)
2230 {
2231         cmd->supported          = 0;
2232         cmd->advertising        = 0;
2233         ethtool_cmd_speed_set(cmd, SPEED_10);
2234         cmd->duplex             = DUPLEX_FULL;
2235         cmd->port               = PORT_TP;
2236         cmd->phy_address        = 0;
2237         cmd->transceiver        = XCVR_INTERNAL;
2238         cmd->autoneg            = AUTONEG_DISABLE;
2239         cmd->maxtxpkt           = 0;
2240         cmd->maxrxpkt           = 0;
2241         return 0;
2242 }
2243
2244 static void tun_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
2245 {
2246         struct tun_struct *tun = netdev_priv(dev);
2247
2248         strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
2249         strlcpy(info->version, DRV_VERSION, sizeof(info->version));
2250
2251         switch (tun->flags & TUN_TYPE_MASK) {
2252         case TUN_TUN_DEV:
2253                 strlcpy(info->bus_info, "tun", sizeof(info->bus_info));
2254                 break;
2255         case TUN_TAP_DEV:
2256                 strlcpy(info->bus_info, "tap", sizeof(info->bus_info));
2257                 break;
2258         }
2259 }
2260
2261 static u32 tun_get_msglevel(struct net_device *dev)
2262 {
2263 #ifdef TUN_DEBUG
2264         struct tun_struct *tun = netdev_priv(dev);
2265         return tun->debug;
2266 #else
2267         return -EOPNOTSUPP;
2268 #endif
2269 }
2270
2271 static void tun_set_msglevel(struct net_device *dev, u32 value)
2272 {
2273 #ifdef TUN_DEBUG
2274         struct tun_struct *tun = netdev_priv(dev);
2275         tun->debug = value;
2276 #endif
2277 }
2278
2279 static const struct ethtool_ops tun_ethtool_ops = {
2280         .get_settings   = tun_get_settings,
2281         .get_drvinfo    = tun_get_drvinfo,
2282         .get_msglevel   = tun_get_msglevel,
2283         .set_msglevel   = tun_set_msglevel,
2284         .get_link       = ethtool_op_get_link,
2285 };
2286
2287
2288 static int __init tun_init(void)
2289 {
2290         int ret = 0;
2291
2292         pr_info("%s, %s\n", DRV_DESCRIPTION, DRV_VERSION);
2293         pr_info("%s\n", DRV_COPYRIGHT);
2294
2295         ret = rtnl_link_register(&tun_link_ops);
2296         if (ret) {
2297                 pr_err("Can't register link_ops\n");
2298                 goto err_linkops;
2299         }
2300
2301         ret = misc_register(&tun_miscdev);
2302         if (ret) {
2303                 pr_err("Can't register misc device %d\n", TUN_MINOR);
2304                 goto err_misc;
2305         }
2306         return  0;
2307 err_misc:
2308         rtnl_link_unregister(&tun_link_ops);
2309 err_linkops:
2310         return ret;
2311 }
2312
2313 static void tun_cleanup(void)
2314 {
2315         misc_deregister(&tun_miscdev);
2316         rtnl_link_unregister(&tun_link_ops);
2317 }
2318
2319 /* Get an underlying socket object from tun file.  Returns error unless file is
2320  * attached to a device.  The returned object works like a packet socket, it
2321  * can be used for sock_sendmsg/sock_recvmsg.  The caller is responsible for
2322  * holding a reference to the file for as long as the socket is in use. */
2323 struct socket *tun_get_socket(struct file *file)
2324 {
2325         struct tun_file *tfile;
2326         if (file->f_op != &tun_fops)
2327                 return ERR_PTR(-EINVAL);
2328         tfile = file->private_data;
2329         if (!tfile)
2330                 return ERR_PTR(-EBADFD);
2331         return &tfile->socket;
2332 }
2333 EXPORT_SYMBOL_GPL(tun_get_socket);
2334
2335 module_init(tun_init);
2336 module_exit(tun_cleanup);
2337 MODULE_DESCRIPTION(DRV_DESCRIPTION);
2338 MODULE_AUTHOR(DRV_COPYRIGHT);
2339 MODULE_LICENSE("GPL");
2340 MODULE_ALIAS_MISCDEV(TUN_MINOR);
2341 MODULE_ALIAS("devname:net/tun");