3d4c24572ecdcda8ffdd47070ce75dcb625441da
[muen/linux.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/sched/signal.h>
48 #include <linux/major.h>
49 #include <linux/slab.h>
50 #include <linux/poll.h>
51 #include <linux/fcntl.h>
52 #include <linux/init.h>
53 #include <linux/skbuff.h>
54 #include <linux/netdevice.h>
55 #include <linux/etherdevice.h>
56 #include <linux/miscdevice.h>
57 #include <linux/ethtool.h>
58 #include <linux/rtnetlink.h>
59 #include <linux/compat.h>
60 #include <linux/if.h>
61 #include <linux/if_arp.h>
62 #include <linux/if_ether.h>
63 #include <linux/if_tun.h>
64 #include <linux/if_vlan.h>
65 #include <linux/crc32.h>
66 #include <linux/nsproxy.h>
67 #include <linux/virtio_net.h>
68 #include <linux/rcupdate.h>
69 #include <net/net_namespace.h>
70 #include <net/netns/generic.h>
71 #include <net/rtnetlink.h>
72 #include <net/sock.h>
73 #include <linux/seq_file.h>
74 #include <linux/uio.h>
75 #include <linux/skb_array.h>
76
77 #include <linux/uaccess.h>
78
79 /* Uncomment to enable debugging */
80 /* #define TUN_DEBUG 1 */
81
82 #ifdef TUN_DEBUG
83 static int debug;
84
85 #define tun_debug(level, tun, fmt, args...)                     \
86 do {                                                            \
87         if (tun->debug)                                         \
88                 netdev_printk(level, tun->dev, fmt, ##args);    \
89 } while (0)
90 #define DBG1(level, fmt, args...)                               \
91 do {                                                            \
92         if (debug == 2)                                         \
93                 printk(level fmt, ##args);                      \
94 } while (0)
95 #else
96 #define tun_debug(level, tun, fmt, args...)                     \
97 do {                                                            \
98         if (0)                                                  \
99                 netdev_printk(level, tun->dev, fmt, ##args);    \
100 } while (0)
101 #define DBG1(level, fmt, args...)                               \
102 do {                                                            \
103         if (0)                                                  \
104                 printk(level fmt, ##args);                      \
105 } while (0)
106 #endif
107
108 /* TUN device flags */
109
110 /* IFF_ATTACH_QUEUE is never stored in device flags,
111  * overload it to mean fasync when stored there.
112  */
113 #define TUN_FASYNC      IFF_ATTACH_QUEUE
114 /* High bits in flags field are unused. */
115 #define TUN_VNET_LE     0x80000000
116 #define TUN_VNET_BE     0x40000000
117
118 #define TUN_FEATURES (IFF_NO_PI | IFF_ONE_QUEUE | IFF_VNET_HDR | \
119                       IFF_MULTI_QUEUE)
120 #define GOODCOPY_LEN 128
121
122 #define FLT_EXACT_COUNT 8
123 struct tap_filter {
124         unsigned int    count;    /* Number of addrs. Zero means disabled */
125         u32             mask[2];  /* Mask of the hashed addrs */
126         unsigned char   addr[FLT_EXACT_COUNT][ETH_ALEN];
127 };
128
129 /* MAX_TAP_QUEUES 256 is chosen to allow rx/tx queues to be equal
130  * to max number of VCPUs in guest. */
131 #define MAX_TAP_QUEUES 256
132 #define MAX_TAP_FLOWS  4096
133
134 #define TUN_FLOW_EXPIRE (3 * HZ)
135
136 struct tun_pcpu_stats {
137         u64 rx_packets;
138         u64 rx_bytes;
139         u64 tx_packets;
140         u64 tx_bytes;
141         struct u64_stats_sync syncp;
142         u32 rx_dropped;
143         u32 tx_dropped;
144         u32 rx_frame_errors;
145 };
146
147 /* A tun_file connects an open character device to a tuntap netdevice. It
148  * also contains all socket related structures (except sock_fprog and tap_filter)
149  * to serve as one transmit queue for tuntap device. The sock_fprog and
150  * tap_filter were kept in tun_struct since they were used for filtering for the
151  * netdevice not for a specific queue (at least I didn't see the requirement for
152  * this).
153  *
154  * RCU usage:
155  * The tun_file and tun_struct are loosely coupled, the pointer from one to the
156  * other can only be read while rcu_read_lock or rtnl_lock is held.
157  */
158 struct tun_file {
159         struct sock sk;
160         struct socket socket;
161         struct socket_wq wq;
162         struct tun_struct __rcu *tun;
163         struct fasync_struct *fasync;
164         /* only used for fasnyc */
165         unsigned int flags;
166         union {
167                 u16 queue_index;
168                 unsigned int ifindex;
169         };
170         struct list_head next;
171         struct tun_struct *detached;
172         struct skb_array tx_array;
173 };
174
175 struct tun_flow_entry {
176         struct hlist_node hash_link;
177         struct rcu_head rcu;
178         struct tun_struct *tun;
179
180         u32 rxhash;
181         u32 rps_rxhash;
182         int queue_index;
183         unsigned long updated;
184 };
185
186 #define TUN_NUM_FLOW_ENTRIES 1024
187
188 /* Since the socket were moved to tun_file, to preserve the behavior of persist
189  * device, socket filter, sndbuf and vnet header size were restore when the
190  * file were attached to a persist device.
191  */
192 struct tun_struct {
193         struct tun_file __rcu   *tfiles[MAX_TAP_QUEUES];
194         unsigned int            numqueues;
195         unsigned int            flags;
196         kuid_t                  owner;
197         kgid_t                  group;
198
199         struct net_device       *dev;
200         netdev_features_t       set_features;
201 #define TUN_USER_FEATURES (NETIF_F_HW_CSUM|NETIF_F_TSO_ECN|NETIF_F_TSO| \
202                           NETIF_F_TSO6|NETIF_F_UFO)
203
204         int                     align;
205         int                     vnet_hdr_sz;
206         int                     sndbuf;
207         struct tap_filter       txflt;
208         struct sock_fprog       fprog;
209         /* protected by rtnl lock */
210         bool                    filter_attached;
211 #ifdef TUN_DEBUG
212         int debug;
213 #endif
214         spinlock_t lock;
215         struct hlist_head flows[TUN_NUM_FLOW_ENTRIES];
216         struct timer_list flow_gc_timer;
217         unsigned long ageing_time;
218         unsigned int numdisabled;
219         struct list_head disabled;
220         void *security;
221         u32 flow_count;
222         u32 rx_batched;
223         struct tun_pcpu_stats __percpu *pcpu_stats;
224 };
225
226 #ifdef CONFIG_TUN_VNET_CROSS_LE
227 static inline bool tun_legacy_is_little_endian(struct tun_struct *tun)
228 {
229         return tun->flags & TUN_VNET_BE ? false :
230                 virtio_legacy_is_little_endian();
231 }
232
233 static long tun_get_vnet_be(struct tun_struct *tun, int __user *argp)
234 {
235         int be = !!(tun->flags & TUN_VNET_BE);
236
237         if (put_user(be, argp))
238                 return -EFAULT;
239
240         return 0;
241 }
242
243 static long tun_set_vnet_be(struct tun_struct *tun, int __user *argp)
244 {
245         int be;
246
247         if (get_user(be, argp))
248                 return -EFAULT;
249
250         if (be)
251                 tun->flags |= TUN_VNET_BE;
252         else
253                 tun->flags &= ~TUN_VNET_BE;
254
255         return 0;
256 }
257 #else
258 static inline bool tun_legacy_is_little_endian(struct tun_struct *tun)
259 {
260         return virtio_legacy_is_little_endian();
261 }
262
263 static long tun_get_vnet_be(struct tun_struct *tun, int __user *argp)
264 {
265         return -EINVAL;
266 }
267
268 static long tun_set_vnet_be(struct tun_struct *tun, int __user *argp)
269 {
270         return -EINVAL;
271 }
272 #endif /* CONFIG_TUN_VNET_CROSS_LE */
273
274 static inline bool tun_is_little_endian(struct tun_struct *tun)
275 {
276         return tun->flags & TUN_VNET_LE ||
277                 tun_legacy_is_little_endian(tun);
278 }
279
280 static inline u16 tun16_to_cpu(struct tun_struct *tun, __virtio16 val)
281 {
282         return __virtio16_to_cpu(tun_is_little_endian(tun), val);
283 }
284
285 static inline __virtio16 cpu_to_tun16(struct tun_struct *tun, u16 val)
286 {
287         return __cpu_to_virtio16(tun_is_little_endian(tun), val);
288 }
289
290 static inline u32 tun_hashfn(u32 rxhash)
291 {
292         return rxhash & 0x3ff;
293 }
294
295 static struct tun_flow_entry *tun_flow_find(struct hlist_head *head, u32 rxhash)
296 {
297         struct tun_flow_entry *e;
298
299         hlist_for_each_entry_rcu(e, head, hash_link) {
300                 if (e->rxhash == rxhash)
301                         return e;
302         }
303         return NULL;
304 }
305
306 static struct tun_flow_entry *tun_flow_create(struct tun_struct *tun,
307                                               struct hlist_head *head,
308                                               u32 rxhash, u16 queue_index)
309 {
310         struct tun_flow_entry *e = kmalloc(sizeof(*e), GFP_ATOMIC);
311
312         if (e) {
313                 tun_debug(KERN_INFO, tun, "create flow: hash %u index %u\n",
314                           rxhash, queue_index);
315                 e->updated = jiffies;
316                 e->rxhash = rxhash;
317                 e->rps_rxhash = 0;
318                 e->queue_index = queue_index;
319                 e->tun = tun;
320                 hlist_add_head_rcu(&e->hash_link, head);
321                 ++tun->flow_count;
322         }
323         return e;
324 }
325
326 static void tun_flow_delete(struct tun_struct *tun, struct tun_flow_entry *e)
327 {
328         tun_debug(KERN_INFO, tun, "delete flow: hash %u index %u\n",
329                   e->rxhash, e->queue_index);
330         hlist_del_rcu(&e->hash_link);
331         kfree_rcu(e, rcu);
332         --tun->flow_count;
333 }
334
335 static void tun_flow_flush(struct tun_struct *tun)
336 {
337         int i;
338
339         spin_lock_bh(&tun->lock);
340         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
341                 struct tun_flow_entry *e;
342                 struct hlist_node *n;
343
344                 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link)
345                         tun_flow_delete(tun, e);
346         }
347         spin_unlock_bh(&tun->lock);
348 }
349
350 static void tun_flow_delete_by_queue(struct tun_struct *tun, u16 queue_index)
351 {
352         int i;
353
354         spin_lock_bh(&tun->lock);
355         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
356                 struct tun_flow_entry *e;
357                 struct hlist_node *n;
358
359                 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
360                         if (e->queue_index == queue_index)
361                                 tun_flow_delete(tun, e);
362                 }
363         }
364         spin_unlock_bh(&tun->lock);
365 }
366
367 static void tun_flow_cleanup(unsigned long data)
368 {
369         struct tun_struct *tun = (struct tun_struct *)data;
370         unsigned long delay = tun->ageing_time;
371         unsigned long next_timer = jiffies + delay;
372         unsigned long count = 0;
373         int i;
374
375         tun_debug(KERN_INFO, tun, "tun_flow_cleanup\n");
376
377         spin_lock_bh(&tun->lock);
378         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
379                 struct tun_flow_entry *e;
380                 struct hlist_node *n;
381
382                 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
383                         unsigned long this_timer;
384                         count++;
385                         this_timer = e->updated + delay;
386                         if (time_before_eq(this_timer, jiffies))
387                                 tun_flow_delete(tun, e);
388                         else if (time_before(this_timer, next_timer))
389                                 next_timer = this_timer;
390                 }
391         }
392
393         if (count)
394                 mod_timer(&tun->flow_gc_timer, round_jiffies_up(next_timer));
395         spin_unlock_bh(&tun->lock);
396 }
397
398 static void tun_flow_update(struct tun_struct *tun, u32 rxhash,
399                             struct tun_file *tfile)
400 {
401         struct hlist_head *head;
402         struct tun_flow_entry *e;
403         unsigned long delay = tun->ageing_time;
404         u16 queue_index = tfile->queue_index;
405
406         if (!rxhash)
407                 return;
408         else
409                 head = &tun->flows[tun_hashfn(rxhash)];
410
411         rcu_read_lock();
412
413         /* We may get a very small possibility of OOO during switching, not
414          * worth to optimize.*/
415         if (tun->numqueues == 1 || tfile->detached)
416                 goto unlock;
417
418         e = tun_flow_find(head, rxhash);
419         if (likely(e)) {
420                 /* TODO: keep queueing to old queue until it's empty? */
421                 e->queue_index = queue_index;
422                 e->updated = jiffies;
423                 sock_rps_record_flow_hash(e->rps_rxhash);
424         } else {
425                 spin_lock_bh(&tun->lock);
426                 if (!tun_flow_find(head, rxhash) &&
427                     tun->flow_count < MAX_TAP_FLOWS)
428                         tun_flow_create(tun, head, rxhash, queue_index);
429
430                 if (!timer_pending(&tun->flow_gc_timer))
431                         mod_timer(&tun->flow_gc_timer,
432                                   round_jiffies_up(jiffies + delay));
433                 spin_unlock_bh(&tun->lock);
434         }
435
436 unlock:
437         rcu_read_unlock();
438 }
439
440 /**
441  * Save the hash received in the stack receive path and update the
442  * flow_hash table accordingly.
443  */
444 static inline void tun_flow_save_rps_rxhash(struct tun_flow_entry *e, u32 hash)
445 {
446         if (unlikely(e->rps_rxhash != hash))
447                 e->rps_rxhash = hash;
448 }
449
450 /* We try to identify a flow through its rxhash first. The reason that
451  * we do not check rxq no. is because some cards(e.g 82599), chooses
452  * the rxq based on the txq where the last packet of the flow comes. As
453  * the userspace application move between processors, we may get a
454  * different rxq no. here. If we could not get rxhash, then we would
455  * hope the rxq no. may help here.
456  */
457 static u16 tun_select_queue(struct net_device *dev, struct sk_buff *skb,
458                             void *accel_priv, select_queue_fallback_t fallback)
459 {
460         struct tun_struct *tun = netdev_priv(dev);
461         struct tun_flow_entry *e;
462         u32 txq = 0;
463         u32 numqueues = 0;
464
465         rcu_read_lock();
466         numqueues = ACCESS_ONCE(tun->numqueues);
467
468         txq = __skb_get_hash_symmetric(skb);
469         if (txq) {
470                 e = tun_flow_find(&tun->flows[tun_hashfn(txq)], txq);
471                 if (e) {
472                         tun_flow_save_rps_rxhash(e, txq);
473                         txq = e->queue_index;
474                 } else
475                         /* use multiply and shift instead of expensive divide */
476                         txq = ((u64)txq * numqueues) >> 32;
477         } else if (likely(skb_rx_queue_recorded(skb))) {
478                 txq = skb_get_rx_queue(skb);
479                 while (unlikely(txq >= numqueues))
480                         txq -= numqueues;
481         }
482
483         rcu_read_unlock();
484         return txq;
485 }
486
487 static inline bool tun_not_capable(struct tun_struct *tun)
488 {
489         const struct cred *cred = current_cred();
490         struct net *net = dev_net(tun->dev);
491
492         return ((uid_valid(tun->owner) && !uid_eq(cred->euid, tun->owner)) ||
493                   (gid_valid(tun->group) && !in_egroup_p(tun->group))) &&
494                 !ns_capable(net->user_ns, CAP_NET_ADMIN);
495 }
496
497 static void tun_set_real_num_queues(struct tun_struct *tun)
498 {
499         netif_set_real_num_tx_queues(tun->dev, tun->numqueues);
500         netif_set_real_num_rx_queues(tun->dev, tun->numqueues);
501 }
502
503 static void tun_disable_queue(struct tun_struct *tun, struct tun_file *tfile)
504 {
505         tfile->detached = tun;
506         list_add_tail(&tfile->next, &tun->disabled);
507         ++tun->numdisabled;
508 }
509
510 static struct tun_struct *tun_enable_queue(struct tun_file *tfile)
511 {
512         struct tun_struct *tun = tfile->detached;
513
514         tfile->detached = NULL;
515         list_del_init(&tfile->next);
516         --tun->numdisabled;
517         return tun;
518 }
519
520 static void tun_queue_purge(struct tun_file *tfile)
521 {
522         struct sk_buff *skb;
523
524         while ((skb = skb_array_consume(&tfile->tx_array)) != NULL)
525                 kfree_skb(skb);
526
527         skb_queue_purge(&tfile->sk.sk_write_queue);
528         skb_queue_purge(&tfile->sk.sk_error_queue);
529 }
530
531 static void __tun_detach(struct tun_file *tfile, bool clean)
532 {
533         struct tun_file *ntfile;
534         struct tun_struct *tun;
535
536         tun = rtnl_dereference(tfile->tun);
537
538         if (tun && !tfile->detached) {
539                 u16 index = tfile->queue_index;
540                 BUG_ON(index >= tun->numqueues);
541
542                 rcu_assign_pointer(tun->tfiles[index],
543                                    tun->tfiles[tun->numqueues - 1]);
544                 ntfile = rtnl_dereference(tun->tfiles[index]);
545                 ntfile->queue_index = index;
546
547                 --tun->numqueues;
548                 if (clean) {
549                         RCU_INIT_POINTER(tfile->tun, NULL);
550                         sock_put(&tfile->sk);
551                 } else
552                         tun_disable_queue(tun, tfile);
553
554                 synchronize_net();
555                 tun_flow_delete_by_queue(tun, tun->numqueues + 1);
556                 /* Drop read queue */
557                 tun_queue_purge(tfile);
558                 tun_set_real_num_queues(tun);
559         } else if (tfile->detached && clean) {
560                 tun = tun_enable_queue(tfile);
561                 sock_put(&tfile->sk);
562         }
563
564         if (clean) {
565                 if (tun && tun->numqueues == 0 && tun->numdisabled == 0) {
566                         netif_carrier_off(tun->dev);
567
568                         if (!(tun->flags & IFF_PERSIST) &&
569                             tun->dev->reg_state == NETREG_REGISTERED)
570                                 unregister_netdevice(tun->dev);
571                 }
572                 if (tun)
573                         skb_array_cleanup(&tfile->tx_array);
574                 sock_put(&tfile->sk);
575         }
576 }
577
578 static void tun_detach(struct tun_file *tfile, bool clean)
579 {
580         rtnl_lock();
581         __tun_detach(tfile, clean);
582         rtnl_unlock();
583 }
584
585 static void tun_detach_all(struct net_device *dev)
586 {
587         struct tun_struct *tun = netdev_priv(dev);
588         struct tun_file *tfile, *tmp;
589         int i, n = tun->numqueues;
590
591         for (i = 0; i < n; i++) {
592                 tfile = rtnl_dereference(tun->tfiles[i]);
593                 BUG_ON(!tfile);
594                 tfile->socket.sk->sk_shutdown = RCV_SHUTDOWN;
595                 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
596                 RCU_INIT_POINTER(tfile->tun, NULL);
597                 --tun->numqueues;
598         }
599         list_for_each_entry(tfile, &tun->disabled, next) {
600                 tfile->socket.sk->sk_shutdown = RCV_SHUTDOWN;
601                 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
602                 RCU_INIT_POINTER(tfile->tun, NULL);
603         }
604         BUG_ON(tun->numqueues != 0);
605
606         synchronize_net();
607         for (i = 0; i < n; i++) {
608                 tfile = rtnl_dereference(tun->tfiles[i]);
609                 /* Drop read queue */
610                 tun_queue_purge(tfile);
611                 sock_put(&tfile->sk);
612         }
613         list_for_each_entry_safe(tfile, tmp, &tun->disabled, next) {
614                 tun_enable_queue(tfile);
615                 tun_queue_purge(tfile);
616                 sock_put(&tfile->sk);
617         }
618         BUG_ON(tun->numdisabled != 0);
619
620         if (tun->flags & IFF_PERSIST)
621                 module_put(THIS_MODULE);
622 }
623
624 static int tun_attach(struct tun_struct *tun, struct file *file, bool skip_filter)
625 {
626         struct tun_file *tfile = file->private_data;
627         struct net_device *dev = tun->dev;
628         int err;
629
630         err = security_tun_dev_attach(tfile->socket.sk, tun->security);
631         if (err < 0)
632                 goto out;
633
634         err = -EINVAL;
635         if (rtnl_dereference(tfile->tun) && !tfile->detached)
636                 goto out;
637
638         err = -EBUSY;
639         if (!(tun->flags & IFF_MULTI_QUEUE) && tun->numqueues == 1)
640                 goto out;
641
642         err = -E2BIG;
643         if (!tfile->detached &&
644             tun->numqueues + tun->numdisabled == MAX_TAP_QUEUES)
645                 goto out;
646
647         err = 0;
648
649         /* Re-attach the filter to persist device */
650         if (!skip_filter && (tun->filter_attached == true)) {
651                 lock_sock(tfile->socket.sk);
652                 err = sk_attach_filter(&tun->fprog, tfile->socket.sk);
653                 release_sock(tfile->socket.sk);
654                 if (!err)
655                         goto out;
656         }
657
658         if (!tfile->detached &&
659             skb_array_init(&tfile->tx_array, dev->tx_queue_len, GFP_KERNEL)) {
660                 err = -ENOMEM;
661                 goto out;
662         }
663
664         tfile->queue_index = tun->numqueues;
665         tfile->socket.sk->sk_shutdown &= ~RCV_SHUTDOWN;
666         rcu_assign_pointer(tfile->tun, tun);
667         rcu_assign_pointer(tun->tfiles[tun->numqueues], tfile);
668         tun->numqueues++;
669
670         if (tfile->detached)
671                 tun_enable_queue(tfile);
672         else
673                 sock_hold(&tfile->sk);
674
675         tun_set_real_num_queues(tun);
676
677         /* device is allowed to go away first, so no need to hold extra
678          * refcnt.
679          */
680
681 out:
682         return err;
683 }
684
685 static struct tun_struct *__tun_get(struct tun_file *tfile)
686 {
687         struct tun_struct *tun;
688
689         rcu_read_lock();
690         tun = rcu_dereference(tfile->tun);
691         if (tun)
692                 dev_hold(tun->dev);
693         rcu_read_unlock();
694
695         return tun;
696 }
697
698 static struct tun_struct *tun_get(struct file *file)
699 {
700         return __tun_get(file->private_data);
701 }
702
703 static void tun_put(struct tun_struct *tun)
704 {
705         dev_put(tun->dev);
706 }
707
708 /* TAP filtering */
709 static void addr_hash_set(u32 *mask, const u8 *addr)
710 {
711         int n = ether_crc(ETH_ALEN, addr) >> 26;
712         mask[n >> 5] |= (1 << (n & 31));
713 }
714
715 static unsigned int addr_hash_test(const u32 *mask, const u8 *addr)
716 {
717         int n = ether_crc(ETH_ALEN, addr) >> 26;
718         return mask[n >> 5] & (1 << (n & 31));
719 }
720
721 static int update_filter(struct tap_filter *filter, void __user *arg)
722 {
723         struct { u8 u[ETH_ALEN]; } *addr;
724         struct tun_filter uf;
725         int err, alen, n, nexact;
726
727         if (copy_from_user(&uf, arg, sizeof(uf)))
728                 return -EFAULT;
729
730         if (!uf.count) {
731                 /* Disabled */
732                 filter->count = 0;
733                 return 0;
734         }
735
736         alen = ETH_ALEN * uf.count;
737         addr = memdup_user(arg + sizeof(uf), alen);
738         if (IS_ERR(addr))
739                 return PTR_ERR(addr);
740
741         /* The filter is updated without holding any locks. Which is
742          * perfectly safe. We disable it first and in the worst
743          * case we'll accept a few undesired packets. */
744         filter->count = 0;
745         wmb();
746
747         /* Use first set of addresses as an exact filter */
748         for (n = 0; n < uf.count && n < FLT_EXACT_COUNT; n++)
749                 memcpy(filter->addr[n], addr[n].u, ETH_ALEN);
750
751         nexact = n;
752
753         /* Remaining multicast addresses are hashed,
754          * unicast will leave the filter disabled. */
755         memset(filter->mask, 0, sizeof(filter->mask));
756         for (; n < uf.count; n++) {
757                 if (!is_multicast_ether_addr(addr[n].u)) {
758                         err = 0; /* no filter */
759                         goto free_addr;
760                 }
761                 addr_hash_set(filter->mask, addr[n].u);
762         }
763
764         /* For ALLMULTI just set the mask to all ones.
765          * This overrides the mask populated above. */
766         if ((uf.flags & TUN_FLT_ALLMULTI))
767                 memset(filter->mask, ~0, sizeof(filter->mask));
768
769         /* Now enable the filter */
770         wmb();
771         filter->count = nexact;
772
773         /* Return the number of exact filters */
774         err = nexact;
775 free_addr:
776         kfree(addr);
777         return err;
778 }
779
780 /* Returns: 0 - drop, !=0 - accept */
781 static int run_filter(struct tap_filter *filter, const struct sk_buff *skb)
782 {
783         /* Cannot use eth_hdr(skb) here because skb_mac_hdr() is incorrect
784          * at this point. */
785         struct ethhdr *eh = (struct ethhdr *) skb->data;
786         int i;
787
788         /* Exact match */
789         for (i = 0; i < filter->count; i++)
790                 if (ether_addr_equal(eh->h_dest, filter->addr[i]))
791                         return 1;
792
793         /* Inexact match (multicast only) */
794         if (is_multicast_ether_addr(eh->h_dest))
795                 return addr_hash_test(filter->mask, eh->h_dest);
796
797         return 0;
798 }
799
800 /*
801  * Checks whether the packet is accepted or not.
802  * Returns: 0 - drop, !=0 - accept
803  */
804 static int check_filter(struct tap_filter *filter, const struct sk_buff *skb)
805 {
806         if (!filter->count)
807                 return 1;
808
809         return run_filter(filter, skb);
810 }
811
812 /* Network device part of the driver */
813
814 static const struct ethtool_ops tun_ethtool_ops;
815
816 /* Net device detach from fd. */
817 static void tun_net_uninit(struct net_device *dev)
818 {
819         tun_detach_all(dev);
820 }
821
822 /* Net device open. */
823 static int tun_net_open(struct net_device *dev)
824 {
825         struct tun_struct *tun = netdev_priv(dev);
826         int i;
827
828         netif_tx_start_all_queues(dev);
829
830         for (i = 0; i < tun->numqueues; i++) {
831                 struct tun_file *tfile;
832
833                 tfile = rtnl_dereference(tun->tfiles[i]);
834                 tfile->socket.sk->sk_write_space(tfile->socket.sk);
835         }
836
837         return 0;
838 }
839
840 /* Net device close. */
841 static int tun_net_close(struct net_device *dev)
842 {
843         netif_tx_stop_all_queues(dev);
844         return 0;
845 }
846
847 /* Net device start xmit */
848 static netdev_tx_t tun_net_xmit(struct sk_buff *skb, struct net_device *dev)
849 {
850         struct tun_struct *tun = netdev_priv(dev);
851         int txq = skb->queue_mapping;
852         struct tun_file *tfile;
853         u32 numqueues = 0;
854
855         rcu_read_lock();
856         tfile = rcu_dereference(tun->tfiles[txq]);
857         numqueues = ACCESS_ONCE(tun->numqueues);
858
859         /* Drop packet if interface is not attached */
860         if (txq >= numqueues)
861                 goto drop;
862
863 #ifdef CONFIG_RPS
864         if (numqueues == 1 && static_key_false(&rps_needed)) {
865                 /* Select queue was not called for the skbuff, so we extract the
866                  * RPS hash and save it into the flow_table here.
867                  */
868                 __u32 rxhash;
869
870                 rxhash = __skb_get_hash_symmetric(skb);
871                 if (rxhash) {
872                         struct tun_flow_entry *e;
873                         e = tun_flow_find(&tun->flows[tun_hashfn(rxhash)],
874                                         rxhash);
875                         if (e)
876                                 tun_flow_save_rps_rxhash(e, rxhash);
877                 }
878         }
879 #endif
880
881         tun_debug(KERN_INFO, tun, "tun_net_xmit %d\n", skb->len);
882
883         BUG_ON(!tfile);
884
885         /* Drop if the filter does not like it.
886          * This is a noop if the filter is disabled.
887          * Filter can be enabled only for the TAP devices. */
888         if (!check_filter(&tun->txflt, skb))
889                 goto drop;
890
891         if (tfile->socket.sk->sk_filter &&
892             sk_filter(tfile->socket.sk, skb))
893                 goto drop;
894
895         if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC)))
896                 goto drop;
897
898         skb_tx_timestamp(skb);
899
900         /* Orphan the skb - required as we might hang on to it
901          * for indefinite time.
902          */
903         skb_orphan(skb);
904
905         nf_reset(skb);
906
907         if (skb_array_produce(&tfile->tx_array, skb))
908                 goto drop;
909
910         /* Notify and wake up reader process */
911         if (tfile->flags & TUN_FASYNC)
912                 kill_fasync(&tfile->fasync, SIGIO, POLL_IN);
913         tfile->socket.sk->sk_data_ready(tfile->socket.sk);
914
915         rcu_read_unlock();
916         return NETDEV_TX_OK;
917
918 drop:
919         this_cpu_inc(tun->pcpu_stats->tx_dropped);
920         skb_tx_error(skb);
921         kfree_skb(skb);
922         rcu_read_unlock();
923         return NET_XMIT_DROP;
924 }
925
926 static void tun_net_mclist(struct net_device *dev)
927 {
928         /*
929          * This callback is supposed to deal with mc filter in
930          * _rx_ path and has nothing to do with the _tx_ path.
931          * In rx path we always accept everything userspace gives us.
932          */
933 }
934
935 static netdev_features_t tun_net_fix_features(struct net_device *dev,
936         netdev_features_t features)
937 {
938         struct tun_struct *tun = netdev_priv(dev);
939
940         return (features & tun->set_features) | (features & ~TUN_USER_FEATURES);
941 }
942 #ifdef CONFIG_NET_POLL_CONTROLLER
943 static void tun_poll_controller(struct net_device *dev)
944 {
945         /*
946          * Tun only receives frames when:
947          * 1) the char device endpoint gets data from user space
948          * 2) the tun socket gets a sendmsg call from user space
949          * Since both of those are synchronous operations, we are guaranteed
950          * never to have pending data when we poll for it
951          * so there is nothing to do here but return.
952          * We need this though so netpoll recognizes us as an interface that
953          * supports polling, which enables bridge devices in virt setups to
954          * still use netconsole
955          */
956         return;
957 }
958 #endif
959
960 static void tun_set_headroom(struct net_device *dev, int new_hr)
961 {
962         struct tun_struct *tun = netdev_priv(dev);
963
964         if (new_hr < NET_SKB_PAD)
965                 new_hr = NET_SKB_PAD;
966
967         tun->align = new_hr;
968 }
969
970 static void
971 tun_net_get_stats64(struct net_device *dev, struct rtnl_link_stats64 *stats)
972 {
973         u32 rx_dropped = 0, tx_dropped = 0, rx_frame_errors = 0;
974         struct tun_struct *tun = netdev_priv(dev);
975         struct tun_pcpu_stats *p;
976         int i;
977
978         for_each_possible_cpu(i) {
979                 u64 rxpackets, rxbytes, txpackets, txbytes;
980                 unsigned int start;
981
982                 p = per_cpu_ptr(tun->pcpu_stats, i);
983                 do {
984                         start = u64_stats_fetch_begin(&p->syncp);
985                         rxpackets       = p->rx_packets;
986                         rxbytes         = p->rx_bytes;
987                         txpackets       = p->tx_packets;
988                         txbytes         = p->tx_bytes;
989                 } while (u64_stats_fetch_retry(&p->syncp, start));
990
991                 stats->rx_packets       += rxpackets;
992                 stats->rx_bytes         += rxbytes;
993                 stats->tx_packets       += txpackets;
994                 stats->tx_bytes         += txbytes;
995
996                 /* u32 counters */
997                 rx_dropped      += p->rx_dropped;
998                 rx_frame_errors += p->rx_frame_errors;
999                 tx_dropped      += p->tx_dropped;
1000         }
1001         stats->rx_dropped  = rx_dropped;
1002         stats->rx_frame_errors = rx_frame_errors;
1003         stats->tx_dropped = tx_dropped;
1004 }
1005
1006 static const struct net_device_ops tun_netdev_ops = {
1007         .ndo_uninit             = tun_net_uninit,
1008         .ndo_open               = tun_net_open,
1009         .ndo_stop               = tun_net_close,
1010         .ndo_start_xmit         = tun_net_xmit,
1011         .ndo_fix_features       = tun_net_fix_features,
1012         .ndo_select_queue       = tun_select_queue,
1013 #ifdef CONFIG_NET_POLL_CONTROLLER
1014         .ndo_poll_controller    = tun_poll_controller,
1015 #endif
1016         .ndo_set_rx_headroom    = tun_set_headroom,
1017         .ndo_get_stats64        = tun_net_get_stats64,
1018 };
1019
1020 static const struct net_device_ops tap_netdev_ops = {
1021         .ndo_uninit             = tun_net_uninit,
1022         .ndo_open               = tun_net_open,
1023         .ndo_stop               = tun_net_close,
1024         .ndo_start_xmit         = tun_net_xmit,
1025         .ndo_fix_features       = tun_net_fix_features,
1026         .ndo_set_rx_mode        = tun_net_mclist,
1027         .ndo_set_mac_address    = eth_mac_addr,
1028         .ndo_validate_addr      = eth_validate_addr,
1029         .ndo_select_queue       = tun_select_queue,
1030 #ifdef CONFIG_NET_POLL_CONTROLLER
1031         .ndo_poll_controller    = tun_poll_controller,
1032 #endif
1033         .ndo_features_check     = passthru_features_check,
1034         .ndo_set_rx_headroom    = tun_set_headroom,
1035         .ndo_get_stats64        = tun_net_get_stats64,
1036 };
1037
1038 static void tun_flow_init(struct tun_struct *tun)
1039 {
1040         int i;
1041
1042         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++)
1043                 INIT_HLIST_HEAD(&tun->flows[i]);
1044
1045         tun->ageing_time = TUN_FLOW_EXPIRE;
1046         setup_timer(&tun->flow_gc_timer, tun_flow_cleanup, (unsigned long)tun);
1047         mod_timer(&tun->flow_gc_timer,
1048                   round_jiffies_up(jiffies + tun->ageing_time));
1049 }
1050
1051 static void tun_flow_uninit(struct tun_struct *tun)
1052 {
1053         del_timer_sync(&tun->flow_gc_timer);
1054         tun_flow_flush(tun);
1055 }
1056
1057 #define MIN_MTU 68
1058 #define MAX_MTU 65535
1059
1060 /* Initialize net device. */
1061 static void tun_net_init(struct net_device *dev)
1062 {
1063         struct tun_struct *tun = netdev_priv(dev);
1064
1065         switch (tun->flags & TUN_TYPE_MASK) {
1066         case IFF_TUN:
1067                 dev->netdev_ops = &tun_netdev_ops;
1068
1069                 /* Point-to-Point TUN Device */
1070                 dev->hard_header_len = 0;
1071                 dev->addr_len = 0;
1072                 dev->mtu = 1500;
1073
1074                 /* Zero header length */
1075                 dev->type = ARPHRD_NONE;
1076                 dev->flags = IFF_POINTOPOINT | IFF_NOARP | IFF_MULTICAST;
1077                 break;
1078
1079         case IFF_TAP:
1080                 dev->netdev_ops = &tap_netdev_ops;
1081                 /* Ethernet TAP Device */
1082                 ether_setup(dev);
1083                 dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1084                 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
1085
1086                 eth_hw_addr_random(dev);
1087
1088                 break;
1089         }
1090
1091         dev->min_mtu = MIN_MTU;
1092         dev->max_mtu = MAX_MTU - dev->hard_header_len;
1093 }
1094
1095 /* Character device part */
1096
1097 /* Poll */
1098 static unsigned int tun_chr_poll(struct file *file, poll_table *wait)
1099 {
1100         struct tun_file *tfile = file->private_data;
1101         struct tun_struct *tun = __tun_get(tfile);
1102         struct sock *sk;
1103         unsigned int mask = 0;
1104
1105         if (!tun)
1106                 return POLLERR;
1107
1108         sk = tfile->socket.sk;
1109
1110         tun_debug(KERN_INFO, tun, "tun_chr_poll\n");
1111
1112         poll_wait(file, sk_sleep(sk), wait);
1113
1114         if (!skb_array_empty(&tfile->tx_array))
1115                 mask |= POLLIN | POLLRDNORM;
1116
1117         if (tun->dev->flags & IFF_UP &&
1118             (sock_writeable(sk) ||
1119              (!test_and_set_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags) &&
1120               sock_writeable(sk))))
1121                 mask |= POLLOUT | POLLWRNORM;
1122
1123         if (tun->dev->reg_state != NETREG_REGISTERED)
1124                 mask = POLLERR;
1125
1126         tun_put(tun);
1127         return mask;
1128 }
1129
1130 /* prepad is the amount to reserve at front.  len is length after that.
1131  * linear is a hint as to how much to copy (usually headers). */
1132 static struct sk_buff *tun_alloc_skb(struct tun_file *tfile,
1133                                      size_t prepad, size_t len,
1134                                      size_t linear, int noblock)
1135 {
1136         struct sock *sk = tfile->socket.sk;
1137         struct sk_buff *skb;
1138         int err;
1139
1140         /* Under a page?  Don't bother with paged skb. */
1141         if (prepad + len < PAGE_SIZE || !linear)
1142                 linear = len;
1143
1144         skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
1145                                    &err, 0);
1146         if (!skb)
1147                 return ERR_PTR(err);
1148
1149         skb_reserve(skb, prepad);
1150         skb_put(skb, linear);
1151         skb->data_len = len - linear;
1152         skb->len += len - linear;
1153
1154         return skb;
1155 }
1156
1157 static void tun_rx_batched(struct tun_struct *tun, struct tun_file *tfile,
1158                            struct sk_buff *skb, int more)
1159 {
1160         struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
1161         struct sk_buff_head process_queue;
1162         u32 rx_batched = tun->rx_batched;
1163         bool rcv = false;
1164
1165         if (!rx_batched || (!more && skb_queue_empty(queue))) {
1166                 local_bh_disable();
1167                 netif_receive_skb(skb);
1168                 local_bh_enable();
1169                 return;
1170         }
1171
1172         spin_lock(&queue->lock);
1173         if (!more || skb_queue_len(queue) == rx_batched) {
1174                 __skb_queue_head_init(&process_queue);
1175                 skb_queue_splice_tail_init(queue, &process_queue);
1176                 rcv = true;
1177         } else {
1178                 __skb_queue_tail(queue, skb);
1179         }
1180         spin_unlock(&queue->lock);
1181
1182         if (rcv) {
1183                 struct sk_buff *nskb;
1184
1185                 local_bh_disable();
1186                 while ((nskb = __skb_dequeue(&process_queue)))
1187                         netif_receive_skb(nskb);
1188                 netif_receive_skb(skb);
1189                 local_bh_enable();
1190         }
1191 }
1192
1193 /* Get packet from user space buffer */
1194 static ssize_t tun_get_user(struct tun_struct *tun, struct tun_file *tfile,
1195                             void *msg_control, struct iov_iter *from,
1196                             int noblock, bool more)
1197 {
1198         struct tun_pi pi = { 0, cpu_to_be16(ETH_P_IP) };
1199         struct sk_buff *skb;
1200         size_t total_len = iov_iter_count(from);
1201         size_t len = total_len, align = tun->align, linear;
1202         struct virtio_net_hdr gso = { 0 };
1203         struct tun_pcpu_stats *stats;
1204         int good_linear;
1205         int copylen;
1206         bool zerocopy = false;
1207         int err;
1208         u32 rxhash;
1209
1210         if (!(tun->dev->flags & IFF_UP))
1211                 return -EIO;
1212
1213         if (!(tun->flags & IFF_NO_PI)) {
1214                 if (len < sizeof(pi))
1215                         return -EINVAL;
1216                 len -= sizeof(pi);
1217
1218                 if (!copy_from_iter_full(&pi, sizeof(pi), from))
1219                         return -EFAULT;
1220         }
1221
1222         if (tun->flags & IFF_VNET_HDR) {
1223                 int vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
1224
1225                 if (len < vnet_hdr_sz)
1226                         return -EINVAL;
1227                 len -= vnet_hdr_sz;
1228
1229                 if (!copy_from_iter_full(&gso, sizeof(gso), from))
1230                         return -EFAULT;
1231
1232                 if ((gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
1233                     tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2 > tun16_to_cpu(tun, gso.hdr_len))
1234                         gso.hdr_len = cpu_to_tun16(tun, tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2);
1235
1236                 if (tun16_to_cpu(tun, gso.hdr_len) > len)
1237                         return -EINVAL;
1238                 iov_iter_advance(from, vnet_hdr_sz - sizeof(gso));
1239         }
1240
1241         if ((tun->flags & TUN_TYPE_MASK) == IFF_TAP) {
1242                 align += NET_IP_ALIGN;
1243                 if (unlikely(len < ETH_HLEN ||
1244                              (gso.hdr_len && tun16_to_cpu(tun, gso.hdr_len) < ETH_HLEN)))
1245                         return -EINVAL;
1246         }
1247
1248         good_linear = SKB_MAX_HEAD(align);
1249
1250         if (msg_control) {
1251                 struct iov_iter i = *from;
1252
1253                 /* There are 256 bytes to be copied in skb, so there is
1254                  * enough room for skb expand head in case it is used.
1255                  * The rest of the buffer is mapped from userspace.
1256                  */
1257                 copylen = gso.hdr_len ? tun16_to_cpu(tun, gso.hdr_len) : GOODCOPY_LEN;
1258                 if (copylen > good_linear)
1259                         copylen = good_linear;
1260                 linear = copylen;
1261                 iov_iter_advance(&i, copylen);
1262                 if (iov_iter_npages(&i, INT_MAX) <= MAX_SKB_FRAGS)
1263                         zerocopy = true;
1264         }
1265
1266         if (!zerocopy) {
1267                 copylen = len;
1268                 if (tun16_to_cpu(tun, gso.hdr_len) > good_linear)
1269                         linear = good_linear;
1270                 else
1271                         linear = tun16_to_cpu(tun, gso.hdr_len);
1272         }
1273
1274         skb = tun_alloc_skb(tfile, align, copylen, linear, noblock);
1275         if (IS_ERR(skb)) {
1276                 if (PTR_ERR(skb) != -EAGAIN)
1277                         this_cpu_inc(tun->pcpu_stats->rx_dropped);
1278                 return PTR_ERR(skb);
1279         }
1280
1281         if (zerocopy)
1282                 err = zerocopy_sg_from_iter(skb, from);
1283         else
1284                 err = skb_copy_datagram_from_iter(skb, 0, from, len);
1285
1286         if (err) {
1287                 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1288                 kfree_skb(skb);
1289                 return -EFAULT;
1290         }
1291
1292         if (virtio_net_hdr_to_skb(skb, &gso, tun_is_little_endian(tun))) {
1293                 this_cpu_inc(tun->pcpu_stats->rx_frame_errors);
1294                 kfree_skb(skb);
1295                 return -EINVAL;
1296         }
1297
1298         switch (tun->flags & TUN_TYPE_MASK) {
1299         case IFF_TUN:
1300                 if (tun->flags & IFF_NO_PI) {
1301                         switch (skb->data[0] & 0xf0) {
1302                         case 0x40:
1303                                 pi.proto = htons(ETH_P_IP);
1304                                 break;
1305                         case 0x60:
1306                                 pi.proto = htons(ETH_P_IPV6);
1307                                 break;
1308                         default:
1309                                 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1310                                 kfree_skb(skb);
1311                                 return -EINVAL;
1312                         }
1313                 }
1314
1315                 skb_reset_mac_header(skb);
1316                 skb->protocol = pi.proto;
1317                 skb->dev = tun->dev;
1318                 break;
1319         case IFF_TAP:
1320                 skb->protocol = eth_type_trans(skb, tun->dev);
1321                 break;
1322         }
1323
1324         /* copy skb_ubuf_info for callback when skb has no error */
1325         if (zerocopy) {
1326                 skb_shinfo(skb)->destructor_arg = msg_control;
1327                 skb_shinfo(skb)->tx_flags |= SKBTX_DEV_ZEROCOPY;
1328                 skb_shinfo(skb)->tx_flags |= SKBTX_SHARED_FRAG;
1329         } else if (msg_control) {
1330                 struct ubuf_info *uarg = msg_control;
1331                 uarg->callback(uarg, false);
1332         }
1333
1334         skb_reset_network_header(skb);
1335         skb_probe_transport_header(skb, 0);
1336
1337         rxhash = __skb_get_hash_symmetric(skb);
1338 #ifndef CONFIG_4KSTACKS
1339         tun_rx_batched(tun, tfile, skb, more);
1340 #else
1341         netif_rx_ni(skb);
1342 #endif
1343
1344         stats = get_cpu_ptr(tun->pcpu_stats);
1345         u64_stats_update_begin(&stats->syncp);
1346         stats->rx_packets++;
1347         stats->rx_bytes += len;
1348         u64_stats_update_end(&stats->syncp);
1349         put_cpu_ptr(stats);
1350
1351         tun_flow_update(tun, rxhash, tfile);
1352         return total_len;
1353 }
1354
1355 static ssize_t tun_chr_write_iter(struct kiocb *iocb, struct iov_iter *from)
1356 {
1357         struct file *file = iocb->ki_filp;
1358         struct tun_struct *tun = tun_get(file);
1359         struct tun_file *tfile = file->private_data;
1360         ssize_t result;
1361
1362         if (!tun)
1363                 return -EBADFD;
1364
1365         result = tun_get_user(tun, tfile, NULL, from,
1366                               file->f_flags & O_NONBLOCK, false);
1367
1368         tun_put(tun);
1369         return result;
1370 }
1371
1372 /* Put packet to the user space buffer */
1373 static ssize_t tun_put_user(struct tun_struct *tun,
1374                             struct tun_file *tfile,
1375                             struct sk_buff *skb,
1376                             struct iov_iter *iter)
1377 {
1378         struct tun_pi pi = { 0, skb->protocol };
1379         struct tun_pcpu_stats *stats;
1380         ssize_t total;
1381         int vlan_offset = 0;
1382         int vlan_hlen = 0;
1383         int vnet_hdr_sz = 0;
1384
1385         if (skb_vlan_tag_present(skb))
1386                 vlan_hlen = VLAN_HLEN;
1387
1388         if (tun->flags & IFF_VNET_HDR)
1389                 vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
1390
1391         total = skb->len + vlan_hlen + vnet_hdr_sz;
1392
1393         if (!(tun->flags & IFF_NO_PI)) {
1394                 if (iov_iter_count(iter) < sizeof(pi))
1395                         return -EINVAL;
1396
1397                 total += sizeof(pi);
1398                 if (iov_iter_count(iter) < total) {
1399                         /* Packet will be striped */
1400                         pi.flags |= TUN_PKT_STRIP;
1401                 }
1402
1403                 if (copy_to_iter(&pi, sizeof(pi), iter) != sizeof(pi))
1404                         return -EFAULT;
1405         }
1406
1407         if (vnet_hdr_sz) {
1408                 struct virtio_net_hdr gso;
1409
1410                 if (iov_iter_count(iter) < vnet_hdr_sz)
1411                         return -EINVAL;
1412
1413                 if (virtio_net_hdr_from_skb(skb, &gso,
1414                                             tun_is_little_endian(tun), true)) {
1415                         struct skb_shared_info *sinfo = skb_shinfo(skb);
1416                         pr_err("unexpected GSO type: "
1417                                "0x%x, gso_size %d, hdr_len %d\n",
1418                                sinfo->gso_type, tun16_to_cpu(tun, gso.gso_size),
1419                                tun16_to_cpu(tun, gso.hdr_len));
1420                         print_hex_dump(KERN_ERR, "tun: ",
1421                                        DUMP_PREFIX_NONE,
1422                                        16, 1, skb->head,
1423                                        min((int)tun16_to_cpu(tun, gso.hdr_len), 64), true);
1424                         WARN_ON_ONCE(1);
1425                         return -EINVAL;
1426                 }
1427
1428                 if (copy_to_iter(&gso, sizeof(gso), iter) != sizeof(gso))
1429                         return -EFAULT;
1430
1431                 iov_iter_advance(iter, vnet_hdr_sz - sizeof(gso));
1432         }
1433
1434         if (vlan_hlen) {
1435                 int ret;
1436                 struct {
1437                         __be16 h_vlan_proto;
1438                         __be16 h_vlan_TCI;
1439                 } veth;
1440
1441                 veth.h_vlan_proto = skb->vlan_proto;
1442                 veth.h_vlan_TCI = htons(skb_vlan_tag_get(skb));
1443
1444                 vlan_offset = offsetof(struct vlan_ethhdr, h_vlan_proto);
1445
1446                 ret = skb_copy_datagram_iter(skb, 0, iter, vlan_offset);
1447                 if (ret || !iov_iter_count(iter))
1448                         goto done;
1449
1450                 ret = copy_to_iter(&veth, sizeof(veth), iter);
1451                 if (ret != sizeof(veth) || !iov_iter_count(iter))
1452                         goto done;
1453         }
1454
1455         skb_copy_datagram_iter(skb, vlan_offset, iter, skb->len - vlan_offset);
1456
1457 done:
1458         /* caller is in process context, */
1459         stats = get_cpu_ptr(tun->pcpu_stats);
1460         u64_stats_update_begin(&stats->syncp);
1461         stats->tx_packets++;
1462         stats->tx_bytes += skb->len + vlan_hlen;
1463         u64_stats_update_end(&stats->syncp);
1464         put_cpu_ptr(tun->pcpu_stats);
1465
1466         return total;
1467 }
1468
1469 static struct sk_buff *tun_ring_recv(struct tun_file *tfile, int noblock,
1470                                      int *err)
1471 {
1472         DECLARE_WAITQUEUE(wait, current);
1473         struct sk_buff *skb = NULL;
1474         int error = 0;
1475
1476         skb = skb_array_consume(&tfile->tx_array);
1477         if (skb)
1478                 goto out;
1479         if (noblock) {
1480                 error = -EAGAIN;
1481                 goto out;
1482         }
1483
1484         add_wait_queue(&tfile->wq.wait, &wait);
1485         current->state = TASK_INTERRUPTIBLE;
1486
1487         while (1) {
1488                 skb = skb_array_consume(&tfile->tx_array);
1489                 if (skb)
1490                         break;
1491                 if (signal_pending(current)) {
1492                         error = -ERESTARTSYS;
1493                         break;
1494                 }
1495                 if (tfile->socket.sk->sk_shutdown & RCV_SHUTDOWN) {
1496                         error = -EFAULT;
1497                         break;
1498                 }
1499
1500                 schedule();
1501         }
1502
1503         current->state = TASK_RUNNING;
1504         remove_wait_queue(&tfile->wq.wait, &wait);
1505
1506 out:
1507         *err = error;
1508         return skb;
1509 }
1510
1511 static ssize_t tun_do_read(struct tun_struct *tun, struct tun_file *tfile,
1512                            struct iov_iter *to,
1513                            int noblock, struct sk_buff *skb)
1514 {
1515         ssize_t ret;
1516         int err;
1517
1518         tun_debug(KERN_INFO, tun, "tun_do_read\n");
1519
1520         if (!iov_iter_count(to))
1521                 return 0;
1522
1523         if (!skb) {
1524                 /* Read frames from ring */
1525                 skb = tun_ring_recv(tfile, noblock, &err);
1526                 if (!skb)
1527                         return err;
1528         }
1529
1530         ret = tun_put_user(tun, tfile, skb, to);
1531         if (unlikely(ret < 0))
1532                 kfree_skb(skb);
1533         else
1534                 consume_skb(skb);
1535
1536         return ret;
1537 }
1538
1539 static ssize_t tun_chr_read_iter(struct kiocb *iocb, struct iov_iter *to)
1540 {
1541         struct file *file = iocb->ki_filp;
1542         struct tun_file *tfile = file->private_data;
1543         struct tun_struct *tun = __tun_get(tfile);
1544         ssize_t len = iov_iter_count(to), ret;
1545
1546         if (!tun)
1547                 return -EBADFD;
1548         ret = tun_do_read(tun, tfile, to, file->f_flags & O_NONBLOCK, NULL);
1549         ret = min_t(ssize_t, ret, len);
1550         if (ret > 0)
1551                 iocb->ki_pos = ret;
1552         tun_put(tun);
1553         return ret;
1554 }
1555
1556 static void tun_free_netdev(struct net_device *dev)
1557 {
1558         struct tun_struct *tun = netdev_priv(dev);
1559
1560         BUG_ON(!(list_empty(&tun->disabled)));
1561         free_percpu(tun->pcpu_stats);
1562         tun_flow_uninit(tun);
1563         security_tun_dev_free_security(tun->security);
1564 }
1565
1566 static void tun_setup(struct net_device *dev)
1567 {
1568         struct tun_struct *tun = netdev_priv(dev);
1569
1570         tun->owner = INVALID_UID;
1571         tun->group = INVALID_GID;
1572
1573         dev->ethtool_ops = &tun_ethtool_ops;
1574         dev->needs_free_netdev = true;
1575         dev->priv_destructor = tun_free_netdev;
1576         /* We prefer our own queue length */
1577         dev->tx_queue_len = TUN_READQ_SIZE;
1578 }
1579
1580 /* Trivial set of netlink ops to allow deleting tun or tap
1581  * device with netlink.
1582  */
1583 static int tun_validate(struct nlattr *tb[], struct nlattr *data[],
1584                         struct netlink_ext_ack *extack)
1585 {
1586         return -EINVAL;
1587 }
1588
1589 static struct rtnl_link_ops tun_link_ops __read_mostly = {
1590         .kind           = DRV_NAME,
1591         .priv_size      = sizeof(struct tun_struct),
1592         .setup          = tun_setup,
1593         .validate       = tun_validate,
1594 };
1595
1596 static void tun_sock_write_space(struct sock *sk)
1597 {
1598         struct tun_file *tfile;
1599         wait_queue_head_t *wqueue;
1600
1601         if (!sock_writeable(sk))
1602                 return;
1603
1604         if (!test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags))
1605                 return;
1606
1607         wqueue = sk_sleep(sk);
1608         if (wqueue && waitqueue_active(wqueue))
1609                 wake_up_interruptible_sync_poll(wqueue, POLLOUT |
1610                                                 POLLWRNORM | POLLWRBAND);
1611
1612         tfile = container_of(sk, struct tun_file, sk);
1613         kill_fasync(&tfile->fasync, SIGIO, POLL_OUT);
1614 }
1615
1616 static int tun_sendmsg(struct socket *sock, struct msghdr *m, size_t total_len)
1617 {
1618         int ret;
1619         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
1620         struct tun_struct *tun = __tun_get(tfile);
1621
1622         if (!tun)
1623                 return -EBADFD;
1624
1625         ret = tun_get_user(tun, tfile, m->msg_control, &m->msg_iter,
1626                            m->msg_flags & MSG_DONTWAIT,
1627                            m->msg_flags & MSG_MORE);
1628         tun_put(tun);
1629         return ret;
1630 }
1631
1632 static int tun_recvmsg(struct socket *sock, struct msghdr *m, size_t total_len,
1633                        int flags)
1634 {
1635         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
1636         struct tun_struct *tun = __tun_get(tfile);
1637         int ret;
1638
1639         if (!tun)
1640                 return -EBADFD;
1641
1642         if (flags & ~(MSG_DONTWAIT|MSG_TRUNC|MSG_ERRQUEUE)) {
1643                 ret = -EINVAL;
1644                 goto out;
1645         }
1646         if (flags & MSG_ERRQUEUE) {
1647                 ret = sock_recv_errqueue(sock->sk, m, total_len,
1648                                          SOL_PACKET, TUN_TX_TIMESTAMP);
1649                 goto out;
1650         }
1651         ret = tun_do_read(tun, tfile, &m->msg_iter, flags & MSG_DONTWAIT,
1652                           m->msg_control);
1653         if (ret > (ssize_t)total_len) {
1654                 m->msg_flags |= MSG_TRUNC;
1655                 ret = flags & MSG_TRUNC ? ret : total_len;
1656         }
1657 out:
1658         tun_put(tun);
1659         return ret;
1660 }
1661
1662 static int tun_peek_len(struct socket *sock)
1663 {
1664         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
1665         struct tun_struct *tun;
1666         int ret = 0;
1667
1668         tun = __tun_get(tfile);
1669         if (!tun)
1670                 return 0;
1671
1672         ret = skb_array_peek_len(&tfile->tx_array);
1673         tun_put(tun);
1674
1675         return ret;
1676 }
1677
1678 /* Ops structure to mimic raw sockets with tun */
1679 static const struct proto_ops tun_socket_ops = {
1680         .peek_len = tun_peek_len,
1681         .sendmsg = tun_sendmsg,
1682         .recvmsg = tun_recvmsg,
1683 };
1684
1685 static struct proto tun_proto = {
1686         .name           = "tun",
1687         .owner          = THIS_MODULE,
1688         .obj_size       = sizeof(struct tun_file),
1689 };
1690
1691 static int tun_flags(struct tun_struct *tun)
1692 {
1693         return tun->flags & (TUN_FEATURES | IFF_PERSIST | IFF_TUN | IFF_TAP);
1694 }
1695
1696 static ssize_t tun_show_flags(struct device *dev, struct device_attribute *attr,
1697                               char *buf)
1698 {
1699         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1700         return sprintf(buf, "0x%x\n", tun_flags(tun));
1701 }
1702
1703 static ssize_t tun_show_owner(struct device *dev, struct device_attribute *attr,
1704                               char *buf)
1705 {
1706         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1707         return uid_valid(tun->owner)?
1708                 sprintf(buf, "%u\n",
1709                         from_kuid_munged(current_user_ns(), tun->owner)):
1710                 sprintf(buf, "-1\n");
1711 }
1712
1713 static ssize_t tun_show_group(struct device *dev, struct device_attribute *attr,
1714                               char *buf)
1715 {
1716         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1717         return gid_valid(tun->group) ?
1718                 sprintf(buf, "%u\n",
1719                         from_kgid_munged(current_user_ns(), tun->group)):
1720                 sprintf(buf, "-1\n");
1721 }
1722
1723 static DEVICE_ATTR(tun_flags, 0444, tun_show_flags, NULL);
1724 static DEVICE_ATTR(owner, 0444, tun_show_owner, NULL);
1725 static DEVICE_ATTR(group, 0444, tun_show_group, NULL);
1726
1727 static struct attribute *tun_dev_attrs[] = {
1728         &dev_attr_tun_flags.attr,
1729         &dev_attr_owner.attr,
1730         &dev_attr_group.attr,
1731         NULL
1732 };
1733
1734 static const struct attribute_group tun_attr_group = {
1735         .attrs = tun_dev_attrs
1736 };
1737
1738 static int tun_set_iff(struct net *net, struct file *file, struct ifreq *ifr)
1739 {
1740         struct tun_struct *tun;
1741         struct tun_file *tfile = file->private_data;
1742         struct net_device *dev;
1743         int err;
1744
1745         if (tfile->detached)
1746                 return -EINVAL;
1747
1748         dev = __dev_get_by_name(net, ifr->ifr_name);
1749         if (dev) {
1750                 if (ifr->ifr_flags & IFF_TUN_EXCL)
1751                         return -EBUSY;
1752                 if ((ifr->ifr_flags & IFF_TUN) && dev->netdev_ops == &tun_netdev_ops)
1753                         tun = netdev_priv(dev);
1754                 else if ((ifr->ifr_flags & IFF_TAP) && dev->netdev_ops == &tap_netdev_ops)
1755                         tun = netdev_priv(dev);
1756                 else
1757                         return -EINVAL;
1758
1759                 if (!!(ifr->ifr_flags & IFF_MULTI_QUEUE) !=
1760                     !!(tun->flags & IFF_MULTI_QUEUE))
1761                         return -EINVAL;
1762
1763                 if (tun_not_capable(tun))
1764                         return -EPERM;
1765                 err = security_tun_dev_open(tun->security);
1766                 if (err < 0)
1767                         return err;
1768
1769                 err = tun_attach(tun, file, ifr->ifr_flags & IFF_NOFILTER);
1770                 if (err < 0)
1771                         return err;
1772
1773                 if (tun->flags & IFF_MULTI_QUEUE &&
1774                     (tun->numqueues + tun->numdisabled > 1)) {
1775                         /* One or more queue has already been attached, no need
1776                          * to initialize the device again.
1777                          */
1778                         return 0;
1779                 }
1780         }
1781         else {
1782                 char *name;
1783                 unsigned long flags = 0;
1784                 int queues = ifr->ifr_flags & IFF_MULTI_QUEUE ?
1785                              MAX_TAP_QUEUES : 1;
1786
1787                 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
1788                         return -EPERM;
1789                 err = security_tun_dev_create();
1790                 if (err < 0)
1791                         return err;
1792
1793                 /* Set dev type */
1794                 if (ifr->ifr_flags & IFF_TUN) {
1795                         /* TUN device */
1796                         flags |= IFF_TUN;
1797                         name = "tun%d";
1798                 } else if (ifr->ifr_flags & IFF_TAP) {
1799                         /* TAP device */
1800                         flags |= IFF_TAP;
1801                         name = "tap%d";
1802                 } else
1803                         return -EINVAL;
1804
1805                 if (*ifr->ifr_name)
1806                         name = ifr->ifr_name;
1807
1808                 dev = alloc_netdev_mqs(sizeof(struct tun_struct), name,
1809                                        NET_NAME_UNKNOWN, tun_setup, queues,
1810                                        queues);
1811
1812                 if (!dev)
1813                         return -ENOMEM;
1814
1815                 dev_net_set(dev, net);
1816                 dev->rtnl_link_ops = &tun_link_ops;
1817                 dev->ifindex = tfile->ifindex;
1818                 dev->sysfs_groups[0] = &tun_attr_group;
1819
1820                 tun = netdev_priv(dev);
1821                 tun->dev = dev;
1822                 tun->flags = flags;
1823                 tun->txflt.count = 0;
1824                 tun->vnet_hdr_sz = sizeof(struct virtio_net_hdr);
1825
1826                 tun->align = NET_SKB_PAD;
1827                 tun->filter_attached = false;
1828                 tun->sndbuf = tfile->socket.sk->sk_sndbuf;
1829                 tun->rx_batched = 0;
1830
1831                 tun->pcpu_stats = netdev_alloc_pcpu_stats(struct tun_pcpu_stats);
1832                 if (!tun->pcpu_stats) {
1833                         err = -ENOMEM;
1834                         goto err_free_dev;
1835                 }
1836
1837                 spin_lock_init(&tun->lock);
1838
1839                 err = security_tun_dev_alloc_security(&tun->security);
1840                 if (err < 0)
1841                         goto err_free_stat;
1842
1843                 tun_net_init(dev);
1844                 tun_flow_init(tun);
1845
1846                 dev->hw_features = NETIF_F_SG | NETIF_F_FRAGLIST |
1847                                    TUN_USER_FEATURES | NETIF_F_HW_VLAN_CTAG_TX |
1848                                    NETIF_F_HW_VLAN_STAG_TX;
1849                 dev->features = dev->hw_features | NETIF_F_LLTX;
1850                 dev->vlan_features = dev->features &
1851                                      ~(NETIF_F_HW_VLAN_CTAG_TX |
1852                                        NETIF_F_HW_VLAN_STAG_TX);
1853
1854                 INIT_LIST_HEAD(&tun->disabled);
1855                 err = tun_attach(tun, file, false);
1856                 if (err < 0)
1857                         goto err_free_flow;
1858
1859                 err = register_netdevice(tun->dev);
1860                 if (err < 0)
1861                         goto err_detach;
1862         }
1863
1864         netif_carrier_on(tun->dev);
1865
1866         tun_debug(KERN_INFO, tun, "tun_set_iff\n");
1867
1868         tun->flags = (tun->flags & ~TUN_FEATURES) |
1869                 (ifr->ifr_flags & TUN_FEATURES);
1870
1871         /* Make sure persistent devices do not get stuck in
1872          * xoff state.
1873          */
1874         if (netif_running(tun->dev))
1875                 netif_tx_wake_all_queues(tun->dev);
1876
1877         strcpy(ifr->ifr_name, tun->dev->name);
1878         return 0;
1879
1880 err_detach:
1881         tun_detach_all(dev);
1882 err_free_flow:
1883         tun_flow_uninit(tun);
1884         security_tun_dev_free_security(tun->security);
1885 err_free_stat:
1886         free_percpu(tun->pcpu_stats);
1887 err_free_dev:
1888         free_netdev(dev);
1889         return err;
1890 }
1891
1892 static void tun_get_iff(struct net *net, struct tun_struct *tun,
1893                        struct ifreq *ifr)
1894 {
1895         tun_debug(KERN_INFO, tun, "tun_get_iff\n");
1896
1897         strcpy(ifr->ifr_name, tun->dev->name);
1898
1899         ifr->ifr_flags = tun_flags(tun);
1900
1901 }
1902
1903 /* This is like a cut-down ethtool ops, except done via tun fd so no
1904  * privs required. */
1905 static int set_offload(struct tun_struct *tun, unsigned long arg)
1906 {
1907         netdev_features_t features = 0;
1908
1909         if (arg & TUN_F_CSUM) {
1910                 features |= NETIF_F_HW_CSUM;
1911                 arg &= ~TUN_F_CSUM;
1912
1913                 if (arg & (TUN_F_TSO4|TUN_F_TSO6)) {
1914                         if (arg & TUN_F_TSO_ECN) {
1915                                 features |= NETIF_F_TSO_ECN;
1916                                 arg &= ~TUN_F_TSO_ECN;
1917                         }
1918                         if (arg & TUN_F_TSO4)
1919                                 features |= NETIF_F_TSO;
1920                         if (arg & TUN_F_TSO6)
1921                                 features |= NETIF_F_TSO6;
1922                         arg &= ~(TUN_F_TSO4|TUN_F_TSO6);
1923                 }
1924
1925                 if (arg & TUN_F_UFO) {
1926                         features |= NETIF_F_UFO;
1927                         arg &= ~TUN_F_UFO;
1928                 }
1929         }
1930
1931         /* This gives the user a way to test for new features in future by
1932          * trying to set them. */
1933         if (arg)
1934                 return -EINVAL;
1935
1936         tun->set_features = features;
1937         tun->dev->wanted_features &= ~TUN_USER_FEATURES;
1938         tun->dev->wanted_features |= features;
1939         netdev_update_features(tun->dev);
1940
1941         return 0;
1942 }
1943
1944 static void tun_detach_filter(struct tun_struct *tun, int n)
1945 {
1946         int i;
1947         struct tun_file *tfile;
1948
1949         for (i = 0; i < n; i++) {
1950                 tfile = rtnl_dereference(tun->tfiles[i]);
1951                 lock_sock(tfile->socket.sk);
1952                 sk_detach_filter(tfile->socket.sk);
1953                 release_sock(tfile->socket.sk);
1954         }
1955
1956         tun->filter_attached = false;
1957 }
1958
1959 static int tun_attach_filter(struct tun_struct *tun)
1960 {
1961         int i, ret = 0;
1962         struct tun_file *tfile;
1963
1964         for (i = 0; i < tun->numqueues; i++) {
1965                 tfile = rtnl_dereference(tun->tfiles[i]);
1966                 lock_sock(tfile->socket.sk);
1967                 ret = sk_attach_filter(&tun->fprog, tfile->socket.sk);
1968                 release_sock(tfile->socket.sk);
1969                 if (ret) {
1970                         tun_detach_filter(tun, i);
1971                         return ret;
1972                 }
1973         }
1974
1975         tun->filter_attached = true;
1976         return ret;
1977 }
1978
1979 static void tun_set_sndbuf(struct tun_struct *tun)
1980 {
1981         struct tun_file *tfile;
1982         int i;
1983
1984         for (i = 0; i < tun->numqueues; i++) {
1985                 tfile = rtnl_dereference(tun->tfiles[i]);
1986                 tfile->socket.sk->sk_sndbuf = tun->sndbuf;
1987         }
1988 }
1989
1990 static int tun_set_queue(struct file *file, struct ifreq *ifr)
1991 {
1992         struct tun_file *tfile = file->private_data;
1993         struct tun_struct *tun;
1994         int ret = 0;
1995
1996         rtnl_lock();
1997
1998         if (ifr->ifr_flags & IFF_ATTACH_QUEUE) {
1999                 tun = tfile->detached;
2000                 if (!tun) {
2001                         ret = -EINVAL;
2002                         goto unlock;
2003                 }
2004                 ret = security_tun_dev_attach_queue(tun->security);
2005                 if (ret < 0)
2006                         goto unlock;
2007                 ret = tun_attach(tun, file, false);
2008         } else if (ifr->ifr_flags & IFF_DETACH_QUEUE) {
2009                 tun = rtnl_dereference(tfile->tun);
2010                 if (!tun || !(tun->flags & IFF_MULTI_QUEUE) || tfile->detached)
2011                         ret = -EINVAL;
2012                 else
2013                         __tun_detach(tfile, false);
2014         } else
2015                 ret = -EINVAL;
2016
2017 unlock:
2018         rtnl_unlock();
2019         return ret;
2020 }
2021
2022 static long __tun_chr_ioctl(struct file *file, unsigned int cmd,
2023                             unsigned long arg, int ifreq_len)
2024 {
2025         struct tun_file *tfile = file->private_data;
2026         struct tun_struct *tun;
2027         void __user* argp = (void __user*)arg;
2028         struct ifreq ifr;
2029         kuid_t owner;
2030         kgid_t group;
2031         int sndbuf;
2032         int vnet_hdr_sz;
2033         unsigned int ifindex;
2034         int le;
2035         int ret;
2036
2037         if (cmd == TUNSETIFF || cmd == TUNSETQUEUE || _IOC_TYPE(cmd) == SOCK_IOC_TYPE) {
2038                 if (copy_from_user(&ifr, argp, ifreq_len))
2039                         return -EFAULT;
2040         } else {
2041                 memset(&ifr, 0, sizeof(ifr));
2042         }
2043         if (cmd == TUNGETFEATURES) {
2044                 /* Currently this just means: "what IFF flags are valid?".
2045                  * This is needed because we never checked for invalid flags on
2046                  * TUNSETIFF.
2047                  */
2048                 return put_user(IFF_TUN | IFF_TAP | TUN_FEATURES,
2049                                 (unsigned int __user*)argp);
2050         } else if (cmd == TUNSETQUEUE)
2051                 return tun_set_queue(file, &ifr);
2052
2053         ret = 0;
2054         rtnl_lock();
2055
2056         tun = __tun_get(tfile);
2057         if (cmd == TUNSETIFF) {
2058                 ret = -EEXIST;
2059                 if (tun)
2060                         goto unlock;
2061
2062                 ifr.ifr_name[IFNAMSIZ-1] = '\0';
2063
2064                 ret = tun_set_iff(sock_net(&tfile->sk), file, &ifr);
2065
2066                 if (ret)
2067                         goto unlock;
2068
2069                 if (copy_to_user(argp, &ifr, ifreq_len))
2070                         ret = -EFAULT;
2071                 goto unlock;
2072         }
2073         if (cmd == TUNSETIFINDEX) {
2074                 ret = -EPERM;
2075                 if (tun)
2076                         goto unlock;
2077
2078                 ret = -EFAULT;
2079                 if (copy_from_user(&ifindex, argp, sizeof(ifindex)))
2080                         goto unlock;
2081
2082                 ret = 0;
2083                 tfile->ifindex = ifindex;
2084                 goto unlock;
2085         }
2086
2087         ret = -EBADFD;
2088         if (!tun)
2089                 goto unlock;
2090
2091         tun_debug(KERN_INFO, tun, "tun_chr_ioctl cmd %u\n", cmd);
2092
2093         ret = 0;
2094         switch (cmd) {
2095         case TUNGETIFF:
2096                 tun_get_iff(current->nsproxy->net_ns, tun, &ifr);
2097
2098                 if (tfile->detached)
2099                         ifr.ifr_flags |= IFF_DETACH_QUEUE;
2100                 if (!tfile->socket.sk->sk_filter)
2101                         ifr.ifr_flags |= IFF_NOFILTER;
2102
2103                 if (copy_to_user(argp, &ifr, ifreq_len))
2104                         ret = -EFAULT;
2105                 break;
2106
2107         case TUNSETNOCSUM:
2108                 /* Disable/Enable checksum */
2109
2110                 /* [unimplemented] */
2111                 tun_debug(KERN_INFO, tun, "ignored: set checksum %s\n",
2112                           arg ? "disabled" : "enabled");
2113                 break;
2114
2115         case TUNSETPERSIST:
2116                 /* Disable/Enable persist mode. Keep an extra reference to the
2117                  * module to prevent the module being unprobed.
2118                  */
2119                 if (arg && !(tun->flags & IFF_PERSIST)) {
2120                         tun->flags |= IFF_PERSIST;
2121                         __module_get(THIS_MODULE);
2122                 }
2123                 if (!arg && (tun->flags & IFF_PERSIST)) {
2124                         tun->flags &= ~IFF_PERSIST;
2125                         module_put(THIS_MODULE);
2126                 }
2127
2128                 tun_debug(KERN_INFO, tun, "persist %s\n",
2129                           arg ? "enabled" : "disabled");
2130                 break;
2131
2132         case TUNSETOWNER:
2133                 /* Set owner of the device */
2134                 owner = make_kuid(current_user_ns(), arg);
2135                 if (!uid_valid(owner)) {
2136                         ret = -EINVAL;
2137                         break;
2138                 }
2139                 tun->owner = owner;
2140                 tun_debug(KERN_INFO, tun, "owner set to %u\n",
2141                           from_kuid(&init_user_ns, tun->owner));
2142                 break;
2143
2144         case TUNSETGROUP:
2145                 /* Set group of the device */
2146                 group = make_kgid(current_user_ns(), arg);
2147                 if (!gid_valid(group)) {
2148                         ret = -EINVAL;
2149                         break;
2150                 }
2151                 tun->group = group;
2152                 tun_debug(KERN_INFO, tun, "group set to %u\n",
2153                           from_kgid(&init_user_ns, tun->group));
2154                 break;
2155
2156         case TUNSETLINK:
2157                 /* Only allow setting the type when the interface is down */
2158                 if (tun->dev->flags & IFF_UP) {
2159                         tun_debug(KERN_INFO, tun,
2160                                   "Linktype set failed because interface is up\n");
2161                         ret = -EBUSY;
2162                 } else {
2163                         tun->dev->type = (int) arg;
2164                         tun_debug(KERN_INFO, tun, "linktype set to %d\n",
2165                                   tun->dev->type);
2166                         ret = 0;
2167                 }
2168                 break;
2169
2170 #ifdef TUN_DEBUG
2171         case TUNSETDEBUG:
2172                 tun->debug = arg;
2173                 break;
2174 #endif
2175         case TUNSETOFFLOAD:
2176                 ret = set_offload(tun, arg);
2177                 break;
2178
2179         case TUNSETTXFILTER:
2180                 /* Can be set only for TAPs */
2181                 ret = -EINVAL;
2182                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
2183                         break;
2184                 ret = update_filter(&tun->txflt, (void __user *)arg);
2185                 break;
2186
2187         case SIOCGIFHWADDR:
2188                 /* Get hw address */
2189                 memcpy(ifr.ifr_hwaddr.sa_data, tun->dev->dev_addr, ETH_ALEN);
2190                 ifr.ifr_hwaddr.sa_family = tun->dev->type;
2191                 if (copy_to_user(argp, &ifr, ifreq_len))
2192                         ret = -EFAULT;
2193                 break;
2194
2195         case SIOCSIFHWADDR:
2196                 /* Set hw address */
2197                 tun_debug(KERN_DEBUG, tun, "set hw address: %pM\n",
2198                           ifr.ifr_hwaddr.sa_data);
2199
2200                 ret = dev_set_mac_address(tun->dev, &ifr.ifr_hwaddr);
2201                 break;
2202
2203         case TUNGETSNDBUF:
2204                 sndbuf = tfile->socket.sk->sk_sndbuf;
2205                 if (copy_to_user(argp, &sndbuf, sizeof(sndbuf)))
2206                         ret = -EFAULT;
2207                 break;
2208
2209         case TUNSETSNDBUF:
2210                 if (copy_from_user(&sndbuf, argp, sizeof(sndbuf))) {
2211                         ret = -EFAULT;
2212                         break;
2213                 }
2214
2215                 tun->sndbuf = sndbuf;
2216                 tun_set_sndbuf(tun);
2217                 break;
2218
2219         case TUNGETVNETHDRSZ:
2220                 vnet_hdr_sz = tun->vnet_hdr_sz;
2221                 if (copy_to_user(argp, &vnet_hdr_sz, sizeof(vnet_hdr_sz)))
2222                         ret = -EFAULT;
2223                 break;
2224
2225         case TUNSETVNETHDRSZ:
2226                 if (copy_from_user(&vnet_hdr_sz, argp, sizeof(vnet_hdr_sz))) {
2227                         ret = -EFAULT;
2228                         break;
2229                 }
2230                 if (vnet_hdr_sz < (int)sizeof(struct virtio_net_hdr)) {
2231                         ret = -EINVAL;
2232                         break;
2233                 }
2234
2235                 tun->vnet_hdr_sz = vnet_hdr_sz;
2236                 break;
2237
2238         case TUNGETVNETLE:
2239                 le = !!(tun->flags & TUN_VNET_LE);
2240                 if (put_user(le, (int __user *)argp))
2241                         ret = -EFAULT;
2242                 break;
2243
2244         case TUNSETVNETLE:
2245                 if (get_user(le, (int __user *)argp)) {
2246                         ret = -EFAULT;
2247                         break;
2248                 }
2249                 if (le)
2250                         tun->flags |= TUN_VNET_LE;
2251                 else
2252                         tun->flags &= ~TUN_VNET_LE;
2253                 break;
2254
2255         case TUNGETVNETBE:
2256                 ret = tun_get_vnet_be(tun, argp);
2257                 break;
2258
2259         case TUNSETVNETBE:
2260                 ret = tun_set_vnet_be(tun, argp);
2261                 break;
2262
2263         case TUNATTACHFILTER:
2264                 /* Can be set only for TAPs */
2265                 ret = -EINVAL;
2266                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
2267                         break;
2268                 ret = -EFAULT;
2269                 if (copy_from_user(&tun->fprog, argp, sizeof(tun->fprog)))
2270                         break;
2271
2272                 ret = tun_attach_filter(tun);
2273                 break;
2274
2275         case TUNDETACHFILTER:
2276                 /* Can be set only for TAPs */
2277                 ret = -EINVAL;
2278                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
2279                         break;
2280                 ret = 0;
2281                 tun_detach_filter(tun, tun->numqueues);
2282                 break;
2283
2284         case TUNGETFILTER:
2285                 ret = -EINVAL;
2286                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
2287                         break;
2288                 ret = -EFAULT;
2289                 if (copy_to_user(argp, &tun->fprog, sizeof(tun->fprog)))
2290                         break;
2291                 ret = 0;
2292                 break;
2293
2294         default:
2295                 ret = -EINVAL;
2296                 break;
2297         }
2298
2299 unlock:
2300         rtnl_unlock();
2301         if (tun)
2302                 tun_put(tun);
2303         return ret;
2304 }
2305
2306 static long tun_chr_ioctl(struct file *file,
2307                           unsigned int cmd, unsigned long arg)
2308 {
2309         return __tun_chr_ioctl(file, cmd, arg, sizeof (struct ifreq));
2310 }
2311
2312 #ifdef CONFIG_COMPAT
2313 static long tun_chr_compat_ioctl(struct file *file,
2314                          unsigned int cmd, unsigned long arg)
2315 {
2316         switch (cmd) {
2317         case TUNSETIFF:
2318         case TUNGETIFF:
2319         case TUNSETTXFILTER:
2320         case TUNGETSNDBUF:
2321         case TUNSETSNDBUF:
2322         case SIOCGIFHWADDR:
2323         case SIOCSIFHWADDR:
2324                 arg = (unsigned long)compat_ptr(arg);
2325                 break;
2326         default:
2327                 arg = (compat_ulong_t)arg;
2328                 break;
2329         }
2330
2331         /*
2332          * compat_ifreq is shorter than ifreq, so we must not access beyond
2333          * the end of that structure. All fields that are used in this
2334          * driver are compatible though, we don't need to convert the
2335          * contents.
2336          */
2337         return __tun_chr_ioctl(file, cmd, arg, sizeof(struct compat_ifreq));
2338 }
2339 #endif /* CONFIG_COMPAT */
2340
2341 static int tun_chr_fasync(int fd, struct file *file, int on)
2342 {
2343         struct tun_file *tfile = file->private_data;
2344         int ret;
2345
2346         if ((ret = fasync_helper(fd, file, on, &tfile->fasync)) < 0)
2347                 goto out;
2348
2349         if (on) {
2350                 __f_setown(file, task_pid(current), PIDTYPE_PID, 0);
2351                 tfile->flags |= TUN_FASYNC;
2352         } else
2353                 tfile->flags &= ~TUN_FASYNC;
2354         ret = 0;
2355 out:
2356         return ret;
2357 }
2358
2359 static int tun_chr_open(struct inode *inode, struct file * file)
2360 {
2361         struct net *net = current->nsproxy->net_ns;
2362         struct tun_file *tfile;
2363
2364         DBG1(KERN_INFO, "tunX: tun_chr_open\n");
2365
2366         tfile = (struct tun_file *)sk_alloc(net, AF_UNSPEC, GFP_KERNEL,
2367                                             &tun_proto, 0);
2368         if (!tfile)
2369                 return -ENOMEM;
2370         RCU_INIT_POINTER(tfile->tun, NULL);
2371         tfile->flags = 0;
2372         tfile->ifindex = 0;
2373
2374         init_waitqueue_head(&tfile->wq.wait);
2375         RCU_INIT_POINTER(tfile->socket.wq, &tfile->wq);
2376
2377         tfile->socket.file = file;
2378         tfile->socket.ops = &tun_socket_ops;
2379
2380         sock_init_data(&tfile->socket, &tfile->sk);
2381
2382         tfile->sk.sk_write_space = tun_sock_write_space;
2383         tfile->sk.sk_sndbuf = INT_MAX;
2384
2385         file->private_data = tfile;
2386         INIT_LIST_HEAD(&tfile->next);
2387
2388         sock_set_flag(&tfile->sk, SOCK_ZEROCOPY);
2389
2390         return 0;
2391 }
2392
2393 static int tun_chr_close(struct inode *inode, struct file *file)
2394 {
2395         struct tun_file *tfile = file->private_data;
2396
2397         tun_detach(tfile, true);
2398
2399         return 0;
2400 }
2401
2402 #ifdef CONFIG_PROC_FS
2403 static void tun_chr_show_fdinfo(struct seq_file *m, struct file *f)
2404 {
2405         struct tun_struct *tun;
2406         struct ifreq ifr;
2407
2408         memset(&ifr, 0, sizeof(ifr));
2409
2410         rtnl_lock();
2411         tun = tun_get(f);
2412         if (tun)
2413                 tun_get_iff(current->nsproxy->net_ns, tun, &ifr);
2414         rtnl_unlock();
2415
2416         if (tun)
2417                 tun_put(tun);
2418
2419         seq_printf(m, "iff:\t%s\n", ifr.ifr_name);
2420 }
2421 #endif
2422
2423 static const struct file_operations tun_fops = {
2424         .owner  = THIS_MODULE,
2425         .llseek = no_llseek,
2426         .read_iter  = tun_chr_read_iter,
2427         .write_iter = tun_chr_write_iter,
2428         .poll   = tun_chr_poll,
2429         .unlocked_ioctl = tun_chr_ioctl,
2430 #ifdef CONFIG_COMPAT
2431         .compat_ioctl = tun_chr_compat_ioctl,
2432 #endif
2433         .open   = tun_chr_open,
2434         .release = tun_chr_close,
2435         .fasync = tun_chr_fasync,
2436 #ifdef CONFIG_PROC_FS
2437         .show_fdinfo = tun_chr_show_fdinfo,
2438 #endif
2439 };
2440
2441 static struct miscdevice tun_miscdev = {
2442         .minor = TUN_MINOR,
2443         .name = "tun",
2444         .nodename = "net/tun",
2445         .fops = &tun_fops,
2446 };
2447
2448 /* ethtool interface */
2449
2450 static int tun_get_link_ksettings(struct net_device *dev,
2451                                   struct ethtool_link_ksettings *cmd)
2452 {
2453         ethtool_link_ksettings_zero_link_mode(cmd, supported);
2454         ethtool_link_ksettings_zero_link_mode(cmd, advertising);
2455         cmd->base.speed         = SPEED_10;
2456         cmd->base.duplex        = DUPLEX_FULL;
2457         cmd->base.port          = PORT_TP;
2458         cmd->base.phy_address   = 0;
2459         cmd->base.autoneg       = AUTONEG_DISABLE;
2460         return 0;
2461 }
2462
2463 static void tun_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
2464 {
2465         struct tun_struct *tun = netdev_priv(dev);
2466
2467         strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
2468         strlcpy(info->version, DRV_VERSION, sizeof(info->version));
2469
2470         switch (tun->flags & TUN_TYPE_MASK) {
2471         case IFF_TUN:
2472                 strlcpy(info->bus_info, "tun", sizeof(info->bus_info));
2473                 break;
2474         case IFF_TAP:
2475                 strlcpy(info->bus_info, "tap", sizeof(info->bus_info));
2476                 break;
2477         }
2478 }
2479
2480 static u32 tun_get_msglevel(struct net_device *dev)
2481 {
2482 #ifdef TUN_DEBUG
2483         struct tun_struct *tun = netdev_priv(dev);
2484         return tun->debug;
2485 #else
2486         return -EOPNOTSUPP;
2487 #endif
2488 }
2489
2490 static void tun_set_msglevel(struct net_device *dev, u32 value)
2491 {
2492 #ifdef TUN_DEBUG
2493         struct tun_struct *tun = netdev_priv(dev);
2494         tun->debug = value;
2495 #endif
2496 }
2497
2498 static int tun_get_coalesce(struct net_device *dev,
2499                             struct ethtool_coalesce *ec)
2500 {
2501         struct tun_struct *tun = netdev_priv(dev);
2502
2503         ec->rx_max_coalesced_frames = tun->rx_batched;
2504
2505         return 0;
2506 }
2507
2508 static int tun_set_coalesce(struct net_device *dev,
2509                             struct ethtool_coalesce *ec)
2510 {
2511         struct tun_struct *tun = netdev_priv(dev);
2512
2513         if (ec->rx_max_coalesced_frames > NAPI_POLL_WEIGHT)
2514                 tun->rx_batched = NAPI_POLL_WEIGHT;
2515         else
2516                 tun->rx_batched = ec->rx_max_coalesced_frames;
2517
2518         return 0;
2519 }
2520
2521 static const struct ethtool_ops tun_ethtool_ops = {
2522         .get_drvinfo    = tun_get_drvinfo,
2523         .get_msglevel   = tun_get_msglevel,
2524         .set_msglevel   = tun_set_msglevel,
2525         .get_link       = ethtool_op_get_link,
2526         .get_ts_info    = ethtool_op_get_ts_info,
2527         .get_coalesce   = tun_get_coalesce,
2528         .set_coalesce   = tun_set_coalesce,
2529         .get_link_ksettings = tun_get_link_ksettings,
2530 };
2531
2532 static int tun_queue_resize(struct tun_struct *tun)
2533 {
2534         struct net_device *dev = tun->dev;
2535         struct tun_file *tfile;
2536         struct skb_array **arrays;
2537         int n = tun->numqueues + tun->numdisabled;
2538         int ret, i;
2539
2540         arrays = kmalloc(sizeof *arrays * n, GFP_KERNEL);
2541         if (!arrays)
2542                 return -ENOMEM;
2543
2544         for (i = 0; i < tun->numqueues; i++) {
2545                 tfile = rtnl_dereference(tun->tfiles[i]);
2546                 arrays[i] = &tfile->tx_array;
2547         }
2548         list_for_each_entry(tfile, &tun->disabled, next)
2549                 arrays[i++] = &tfile->tx_array;
2550
2551         ret = skb_array_resize_multiple(arrays, n,
2552                                         dev->tx_queue_len, GFP_KERNEL);
2553
2554         kfree(arrays);
2555         return ret;
2556 }
2557
2558 static int tun_device_event(struct notifier_block *unused,
2559                             unsigned long event, void *ptr)
2560 {
2561         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
2562         struct tun_struct *tun = netdev_priv(dev);
2563
2564         if (dev->rtnl_link_ops != &tun_link_ops)
2565                 return NOTIFY_DONE;
2566
2567         switch (event) {
2568         case NETDEV_CHANGE_TX_QUEUE_LEN:
2569                 if (tun_queue_resize(tun))
2570                         return NOTIFY_BAD;
2571                 break;
2572         default:
2573                 break;
2574         }
2575
2576         return NOTIFY_DONE;
2577 }
2578
2579 static struct notifier_block tun_notifier_block __read_mostly = {
2580         .notifier_call  = tun_device_event,
2581 };
2582
2583 static int __init tun_init(void)
2584 {
2585         int ret = 0;
2586
2587         pr_info("%s, %s\n", DRV_DESCRIPTION, DRV_VERSION);
2588
2589         ret = rtnl_link_register(&tun_link_ops);
2590         if (ret) {
2591                 pr_err("Can't register link_ops\n");
2592                 goto err_linkops;
2593         }
2594
2595         ret = misc_register(&tun_miscdev);
2596         if (ret) {
2597                 pr_err("Can't register misc device %d\n", TUN_MINOR);
2598                 goto err_misc;
2599         }
2600
2601         register_netdevice_notifier(&tun_notifier_block);
2602         return  0;
2603 err_misc:
2604         rtnl_link_unregister(&tun_link_ops);
2605 err_linkops:
2606         return ret;
2607 }
2608
2609 static void tun_cleanup(void)
2610 {
2611         misc_deregister(&tun_miscdev);
2612         rtnl_link_unregister(&tun_link_ops);
2613         unregister_netdevice_notifier(&tun_notifier_block);
2614 }
2615
2616 /* Get an underlying socket object from tun file.  Returns error unless file is
2617  * attached to a device.  The returned object works like a packet socket, it
2618  * can be used for sock_sendmsg/sock_recvmsg.  The caller is responsible for
2619  * holding a reference to the file for as long as the socket is in use. */
2620 struct socket *tun_get_socket(struct file *file)
2621 {
2622         struct tun_file *tfile;
2623         if (file->f_op != &tun_fops)
2624                 return ERR_PTR(-EINVAL);
2625         tfile = file->private_data;
2626         if (!tfile)
2627                 return ERR_PTR(-EBADFD);
2628         return &tfile->socket;
2629 }
2630 EXPORT_SYMBOL_GPL(tun_get_socket);
2631
2632 struct skb_array *tun_get_skb_array(struct file *file)
2633 {
2634         struct tun_file *tfile;
2635
2636         if (file->f_op != &tun_fops)
2637                 return ERR_PTR(-EINVAL);
2638         tfile = file->private_data;
2639         if (!tfile)
2640                 return ERR_PTR(-EBADFD);
2641         return &tfile->tx_array;
2642 }
2643 EXPORT_SYMBOL_GPL(tun_get_skb_array);
2644
2645 module_init(tun_init);
2646 module_exit(tun_cleanup);
2647 MODULE_DESCRIPTION(DRV_DESCRIPTION);
2648 MODULE_AUTHOR(DRV_COPYRIGHT);
2649 MODULE_LICENSE("GPL");
2650 MODULE_ALIAS_MISCDEV(TUN_MINOR);
2651 MODULE_ALIAS("devname:net/tun");