Version:  2.0.40 2.2.26 2.4.37 3.12 3.13 3.14 3.15 3.16 3.17 3.18 3.19 4.0 4.1 4.2 4.3 4.4 4.5 4.6 4.7 4.8 4.9

Linux/drivers/net/tun.c

  1 /*
  2  *  TUN - Universal TUN/TAP device driver.
  3  *  Copyright (C) 1999-2002 Maxim Krasnyansky <maxk@qualcomm.com>
  4  *
  5  *  This program is free software; you can redistribute it and/or modify
  6  *  it under the terms of the GNU General Public License as published by
  7  *  the Free Software Foundation; either version 2 of the License, or
  8  *  (at your option) any later version.
  9  *
 10  *  This program is distributed in the hope that it will be useful,
 11  *  but WITHOUT ANY WARRANTY; without even the implied warranty of
 12  *  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
 13  *  GNU General Public License for more details.
 14  *
 15  *  $Id: tun.c,v 1.15 2002/03/01 02:44:24 maxk Exp $
 16  */
 17 
 18 /*
 19  *  Changes:
 20  *
 21  *  Mike Kershaw <dragorn@kismetwireless.net> 2005/08/14
 22  *    Add TUNSETLINK ioctl to set the link encapsulation
 23  *
 24  *  Mark Smith <markzzzsmith@yahoo.com.au>
 25  *    Use eth_random_addr() for tap MAC address.
 26  *
 27  *  Harald Roelle <harald.roelle@ifi.lmu.de>  2004/04/20
 28  *    Fixes in packet dropping, queue length setting and queue wakeup.
 29  *    Increased default tx queue length.
 30  *    Added ethtool API.
 31  *    Minor cleanups
 32  *
 33  *  Daniel Podlejski <underley@underley.eu.org>
 34  *    Modifications for 2.3.99-pre5 kernel.
 35  */
 36 
 37 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
 38 
 39 #define DRV_NAME        "tun"
 40 #define DRV_VERSION     "1.6"
 41 #define DRV_DESCRIPTION "Universal TUN/TAP device driver"
 42 #define DRV_COPYRIGHT   "(C) 1999-2004 Max Krasnyansky <maxk@qualcomm.com>"
 43 
 44 #include <linux/module.h>
 45 #include <linux/errno.h>
 46 #include <linux/kernel.h>
 47 #include <linux/major.h>
 48 #include <linux/slab.h>
 49 #include <linux/poll.h>
 50 #include <linux/fcntl.h>
 51 #include <linux/init.h>
 52 #include <linux/skbuff.h>
 53 #include <linux/netdevice.h>
 54 #include <linux/etherdevice.h>
 55 #include <linux/miscdevice.h>
 56 #include <linux/ethtool.h>
 57 #include <linux/rtnetlink.h>
 58 #include <linux/compat.h>
 59 #include <linux/if.h>
 60 #include <linux/if_arp.h>
 61 #include <linux/if_ether.h>
 62 #include <linux/if_tun.h>
 63 #include <linux/if_vlan.h>
 64 #include <linux/crc32.h>
 65 #include <linux/nsproxy.h>
 66 #include <linux/virtio_net.h>
 67 #include <linux/rcupdate.h>
 68 #include <net/net_namespace.h>
 69 #include <net/netns/generic.h>
 70 #include <net/rtnetlink.h>
 71 #include <net/sock.h>
 72 #include <linux/seq_file.h>
 73 #include <linux/uio.h>
 74 #include <linux/skb_array.h>
 75 
 76 #include <asm/uaccess.h>
 77 
 78 /* Uncomment to enable debugging */
 79 /* #define TUN_DEBUG 1 */
 80 
 81 #ifdef TUN_DEBUG
 82 static int debug;
 83 
 84 #define tun_debug(level, tun, fmt, args...)                     \
 85 do {                                                            \
 86         if (tun->debug)                                         \
 87                 netdev_printk(level, tun->dev, fmt, ##args);    \
 88 } while (0)
 89 #define DBG1(level, fmt, args...)                               \
 90 do {                                                            \
 91         if (debug == 2)                                         \
 92                 printk(level fmt, ##args);                      \
 93 } while (0)
 94 #else
 95 #define tun_debug(level, tun, fmt, args...)                     \
 96 do {                                                            \
 97         if (0)                                                  \
 98                 netdev_printk(level, tun->dev, fmt, ##args);    \
 99 } while (0)
100 #define DBG1(level, fmt, args...)                               \
101 do {                                                            \
102         if (0)                                                  \
103                 printk(level fmt, ##args);                      \
104 } while (0)
105 #endif
106 
107 /* TUN device flags */
108 
109 /* IFF_ATTACH_QUEUE is never stored in device flags,
110  * overload it to mean fasync when stored there.
111  */
112 #define TUN_FASYNC      IFF_ATTACH_QUEUE
113 /* High bits in flags field are unused. */
114 #define TUN_VNET_LE     0x80000000
115 #define TUN_VNET_BE     0x40000000
116 
117 #define TUN_FEATURES (IFF_NO_PI | IFF_ONE_QUEUE | IFF_VNET_HDR | \
118                       IFF_MULTI_QUEUE)
119 #define GOODCOPY_LEN 128
120 
121 #define FLT_EXACT_COUNT 8
122 struct tap_filter {
123         unsigned int    count;    /* Number of addrs. Zero means disabled */
124         u32             mask[2];  /* Mask of the hashed addrs */
125         unsigned char   addr[FLT_EXACT_COUNT][ETH_ALEN];
126 };
127 
128 /* MAX_TAP_QUEUES 256 is chosen to allow rx/tx queues to be equal
129  * to max number of VCPUs in guest. */
130 #define MAX_TAP_QUEUES 256
131 #define MAX_TAP_FLOWS  4096
132 
133 #define TUN_FLOW_EXPIRE (3 * HZ)
134 
135 struct tun_pcpu_stats {
136         u64 rx_packets;
137         u64 rx_bytes;
138         u64 tx_packets;
139         u64 tx_bytes;
140         struct u64_stats_sync syncp;
141         u32 rx_dropped;
142         u32 tx_dropped;
143         u32 rx_frame_errors;
144 };
145 
146 /* A tun_file connects an open character device to a tuntap netdevice. It
147  * also contains all socket related structures (except sock_fprog and tap_filter)
148  * to serve as one transmit queue for tuntap device. The sock_fprog and
149  * tap_filter were kept in tun_struct since they were used for filtering for the
150  * netdevice not for a specific queue (at least I didn't see the requirement for
151  * this).
152  *
153  * RCU usage:
154  * The tun_file and tun_struct are loosely coupled, the pointer from one to the
155  * other can only be read while rcu_read_lock or rtnl_lock is held.
156  */
157 struct tun_file {
158         struct sock sk;
159         struct socket socket;
160         struct socket_wq wq;
161         struct tun_struct __rcu *tun;
162         struct fasync_struct *fasync;
163         /* only used for fasnyc */
164         unsigned int flags;
165         union {
166                 u16 queue_index;
167                 unsigned int ifindex;
168         };
169         struct list_head next;
170         struct tun_struct *detached;
171         struct skb_array tx_array;
172 };
173 
174 struct tun_flow_entry {
175         struct hlist_node hash_link;
176         struct rcu_head rcu;
177         struct tun_struct *tun;
178 
179         u32 rxhash;
180         u32 rps_rxhash;
181         int queue_index;
182         unsigned long updated;
183 };
184 
185 #define TUN_NUM_FLOW_ENTRIES 1024
186 
187 /* Since the socket were moved to tun_file, to preserve the behavior of persist
188  * device, socket filter, sndbuf and vnet header size were restore when the
189  * file were attached to a persist device.
190  */
191 struct tun_struct {
192         struct tun_file __rcu   *tfiles[MAX_TAP_QUEUES];
193         unsigned int            numqueues;
194         unsigned int            flags;
195         kuid_t                  owner;
196         kgid_t                  group;
197 
198         struct net_device       *dev;
199         netdev_features_t       set_features;
200 #define TUN_USER_FEATURES (NETIF_F_HW_CSUM|NETIF_F_TSO_ECN|NETIF_F_TSO| \
201                           NETIF_F_TSO6|NETIF_F_UFO)
202 
203         int                     align;
204         int                     vnet_hdr_sz;
205         int                     sndbuf;
206         struct tap_filter       txflt;
207         struct sock_fprog       fprog;
208         /* protected by rtnl lock */
209         bool                    filter_attached;
210 #ifdef TUN_DEBUG
211         int debug;
212 #endif
213         spinlock_t lock;
214         struct hlist_head flows[TUN_NUM_FLOW_ENTRIES];
215         struct timer_list flow_gc_timer;
216         unsigned long ageing_time;
217         unsigned int numdisabled;
218         struct list_head disabled;
219         void *security;
220         u32 flow_count;
221         struct tun_pcpu_stats __percpu *pcpu_stats;
222 };
223 
224 #ifdef CONFIG_TUN_VNET_CROSS_LE
225 static inline bool tun_legacy_is_little_endian(struct tun_struct *tun)
226 {
227         return tun->flags & TUN_VNET_BE ? false :
228                 virtio_legacy_is_little_endian();
229 }
230 
231 static long tun_get_vnet_be(struct tun_struct *tun, int __user *argp)
232 {
233         int be = !!(tun->flags & TUN_VNET_BE);
234 
235         if (put_user(be, argp))
236                 return -EFAULT;
237 
238         return 0;
239 }
240 
241 static long tun_set_vnet_be(struct tun_struct *tun, int __user *argp)
242 {
243         int be;
244 
245         if (get_user(be, argp))
246                 return -EFAULT;
247 
248         if (be)
249                 tun->flags |= TUN_VNET_BE;
250         else
251                 tun->flags &= ~TUN_VNET_BE;
252 
253         return 0;
254 }
255 #else
256 static inline bool tun_legacy_is_little_endian(struct tun_struct *tun)
257 {
258         return virtio_legacy_is_little_endian();
259 }
260 
261 static long tun_get_vnet_be(struct tun_struct *tun, int __user *argp)
262 {
263         return -EINVAL;
264 }
265 
266 static long tun_set_vnet_be(struct tun_struct *tun, int __user *argp)
267 {
268         return -EINVAL;
269 }
270 #endif /* CONFIG_TUN_VNET_CROSS_LE */
271 
272 static inline bool tun_is_little_endian(struct tun_struct *tun)
273 {
274         return tun->flags & TUN_VNET_LE ||
275                 tun_legacy_is_little_endian(tun);
276 }
277 
278 static inline u16 tun16_to_cpu(struct tun_struct *tun, __virtio16 val)
279 {
280         return __virtio16_to_cpu(tun_is_little_endian(tun), val);
281 }
282 
283 static inline __virtio16 cpu_to_tun16(struct tun_struct *tun, u16 val)
284 {
285         return __cpu_to_virtio16(tun_is_little_endian(tun), val);
286 }
287 
288 static inline u32 tun_hashfn(u32 rxhash)
289 {
290         return rxhash & 0x3ff;
291 }
292 
293 static struct tun_flow_entry *tun_flow_find(struct hlist_head *head, u32 rxhash)
294 {
295         struct tun_flow_entry *e;
296 
297         hlist_for_each_entry_rcu(e, head, hash_link) {
298                 if (e->rxhash == rxhash)
299                         return e;
300         }
301         return NULL;
302 }
303 
304 static struct tun_flow_entry *tun_flow_create(struct tun_struct *tun,
305                                               struct hlist_head *head,
306                                               u32 rxhash, u16 queue_index)
307 {
308         struct tun_flow_entry *e = kmalloc(sizeof(*e), GFP_ATOMIC);
309 
310         if (e) {
311                 tun_debug(KERN_INFO, tun, "create flow: hash %u index %u\n",
312                           rxhash, queue_index);
313                 e->updated = jiffies;
314                 e->rxhash = rxhash;
315                 e->rps_rxhash = 0;
316                 e->queue_index = queue_index;
317                 e->tun = tun;
318                 hlist_add_head_rcu(&e->hash_link, head);
319                 ++tun->flow_count;
320         }
321         return e;
322 }
323 
324 static void tun_flow_delete(struct tun_struct *tun, struct tun_flow_entry *e)
325 {
326         tun_debug(KERN_INFO, tun, "delete flow: hash %u index %u\n",
327                   e->rxhash, e->queue_index);
328         hlist_del_rcu(&e->hash_link);
329         kfree_rcu(e, rcu);
330         --tun->flow_count;
331 }
332 
333 static void tun_flow_flush(struct tun_struct *tun)
334 {
335         int i;
336 
337         spin_lock_bh(&tun->lock);
338         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
339                 struct tun_flow_entry *e;
340                 struct hlist_node *n;
341 
342                 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link)
343                         tun_flow_delete(tun, e);
344         }
345         spin_unlock_bh(&tun->lock);
346 }
347 
348 static void tun_flow_delete_by_queue(struct tun_struct *tun, u16 queue_index)
349 {
350         int i;
351 
352         spin_lock_bh(&tun->lock);
353         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
354                 struct tun_flow_entry *e;
355                 struct hlist_node *n;
356 
357                 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
358                         if (e->queue_index == queue_index)
359                                 tun_flow_delete(tun, e);
360                 }
361         }
362         spin_unlock_bh(&tun->lock);
363 }
364 
365 static void tun_flow_cleanup(unsigned long data)
366 {
367         struct tun_struct *tun = (struct tun_struct *)data;
368         unsigned long delay = tun->ageing_time;
369         unsigned long next_timer = jiffies + delay;
370         unsigned long count = 0;
371         int i;
372 
373         tun_debug(KERN_INFO, tun, "tun_flow_cleanup\n");
374 
375         spin_lock_bh(&tun->lock);
376         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
377                 struct tun_flow_entry *e;
378                 struct hlist_node *n;
379 
380                 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
381                         unsigned long this_timer;
382                         count++;
383                         this_timer = e->updated + delay;
384                         if (time_before_eq(this_timer, jiffies))
385                                 tun_flow_delete(tun, e);
386                         else if (time_before(this_timer, next_timer))
387                                 next_timer = this_timer;
388                 }
389         }
390 
391         if (count)
392                 mod_timer(&tun->flow_gc_timer, round_jiffies_up(next_timer));
393         spin_unlock_bh(&tun->lock);
394 }
395 
396 static void tun_flow_update(struct tun_struct *tun, u32 rxhash,
397                             struct tun_file *tfile)
398 {
399         struct hlist_head *head;
400         struct tun_flow_entry *e;
401         unsigned long delay = tun->ageing_time;
402         u16 queue_index = tfile->queue_index;
403 
404         if (!rxhash)
405                 return;
406         else
407                 head = &tun->flows[tun_hashfn(rxhash)];
408 
409         rcu_read_lock();
410 
411         /* We may get a very small possibility of OOO during switching, not
412          * worth to optimize.*/
413         if (tun->numqueues == 1 || tfile->detached)
414                 goto unlock;
415 
416         e = tun_flow_find(head, rxhash);
417         if (likely(e)) {
418                 /* TODO: keep queueing to old queue until it's empty? */
419                 e->queue_index = queue_index;
420                 e->updated = jiffies;
421                 sock_rps_record_flow_hash(e->rps_rxhash);
422         } else {
423                 spin_lock_bh(&tun->lock);
424                 if (!tun_flow_find(head, rxhash) &&
425                     tun->flow_count < MAX_TAP_FLOWS)
426                         tun_flow_create(tun, head, rxhash, queue_index);
427 
428                 if (!timer_pending(&tun->flow_gc_timer))
429                         mod_timer(&tun->flow_gc_timer,
430                                   round_jiffies_up(jiffies + delay));
431                 spin_unlock_bh(&tun->lock);
432         }
433 
434 unlock:
435         rcu_read_unlock();
436 }
437 
438 /**
439  * Save the hash received in the stack receive path and update the
440  * flow_hash table accordingly.
441  */
442 static inline void tun_flow_save_rps_rxhash(struct tun_flow_entry *e, u32 hash)
443 {
444         if (unlikely(e->rps_rxhash != hash))
445                 e->rps_rxhash = hash;
446 }
447 
448 /* We try to identify a flow through its rxhash first. The reason that
449  * we do not check rxq no. is because some cards(e.g 82599), chooses
450  * the rxq based on the txq where the last packet of the flow comes. As
451  * the userspace application move between processors, we may get a
452  * different rxq no. here. If we could not get rxhash, then we would
453  * hope the rxq no. may help here.
454  */
455 static u16 tun_select_queue(struct net_device *dev, struct sk_buff *skb,
456                             void *accel_priv, select_queue_fallback_t fallback)
457 {
458         struct tun_struct *tun = netdev_priv(dev);
459         struct tun_flow_entry *e;
460         u32 txq = 0;
461         u32 numqueues = 0;
462 
463         rcu_read_lock();
464         numqueues = ACCESS_ONCE(tun->numqueues);
465 
466         txq = skb_get_hash(skb);
467         if (txq) {
468                 e = tun_flow_find(&tun->flows[tun_hashfn(txq)], txq);
469                 if (e) {
470                         tun_flow_save_rps_rxhash(e, txq);
471                         txq = e->queue_index;
472                 } else
473                         /* use multiply and shift instead of expensive divide */
474                         txq = ((u64)txq * numqueues) >> 32;
475         } else if (likely(skb_rx_queue_recorded(skb))) {
476                 txq = skb_get_rx_queue(skb);
477                 while (unlikely(txq >= numqueues))
478                         txq -= numqueues;
479         }
480 
481         rcu_read_unlock();
482         return txq;
483 }
484 
485 static inline bool tun_not_capable(struct tun_struct *tun)
486 {
487         const struct cred *cred = current_cred();
488         struct net *net = dev_net(tun->dev);
489 
490         return ((uid_valid(tun->owner) && !uid_eq(cred->euid, tun->owner)) ||
491                   (gid_valid(tun->group) && !in_egroup_p(tun->group))) &&
492                 !ns_capable(net->user_ns, CAP_NET_ADMIN);
493 }
494 
495 static void tun_set_real_num_queues(struct tun_struct *tun)
496 {
497         netif_set_real_num_tx_queues(tun->dev, tun->numqueues);
498         netif_set_real_num_rx_queues(tun->dev, tun->numqueues);
499 }
500 
501 static void tun_disable_queue(struct tun_struct *tun, struct tun_file *tfile)
502 {
503         tfile->detached = tun;
504         list_add_tail(&tfile->next, &tun->disabled);
505         ++tun->numdisabled;
506 }
507 
508 static struct tun_struct *tun_enable_queue(struct tun_file *tfile)
509 {
510         struct tun_struct *tun = tfile->detached;
511 
512         tfile->detached = NULL;
513         list_del_init(&tfile->next);
514         --tun->numdisabled;
515         return tun;
516 }
517 
518 static void tun_queue_purge(struct tun_file *tfile)
519 {
520         struct sk_buff *skb;
521 
522         while ((skb = skb_array_consume(&tfile->tx_array)) != NULL)
523                 kfree_skb(skb);
524 
525         skb_queue_purge(&tfile->sk.sk_error_queue);
526 }
527 
528 static void __tun_detach(struct tun_file *tfile, bool clean)
529 {
530         struct tun_file *ntfile;
531         struct tun_struct *tun;
532 
533         tun = rtnl_dereference(tfile->tun);
534 
535         if (tun && !tfile->detached) {
536                 u16 index = tfile->queue_index;
537                 BUG_ON(index >= tun->numqueues);
538 
539                 rcu_assign_pointer(tun->tfiles[index],
540                                    tun->tfiles[tun->numqueues - 1]);
541                 ntfile = rtnl_dereference(tun->tfiles[index]);
542                 ntfile->queue_index = index;
543 
544                 --tun->numqueues;
545                 if (clean) {
546                         RCU_INIT_POINTER(tfile->tun, NULL);
547                         sock_put(&tfile->sk);
548                 } else
549                         tun_disable_queue(tun, tfile);
550 
551                 synchronize_net();
552                 tun_flow_delete_by_queue(tun, tun->numqueues + 1);
553                 /* Drop read queue */
554                 tun_queue_purge(tfile);
555                 tun_set_real_num_queues(tun);
556         } else if (tfile->detached && clean) {
557                 tun = tun_enable_queue(tfile);
558                 sock_put(&tfile->sk);
559         }
560 
561         if (clean) {
562                 if (tun && tun->numqueues == 0 && tun->numdisabled == 0) {
563                         netif_carrier_off(tun->dev);
564 
565                         if (!(tun->flags & IFF_PERSIST) &&
566                             tun->dev->reg_state == NETREG_REGISTERED)
567                                 unregister_netdevice(tun->dev);
568                 }
569                 if (tun)
570                         skb_array_cleanup(&tfile->tx_array);
571                 sock_put(&tfile->sk);
572         }
573 }
574 
575 static void tun_detach(struct tun_file *tfile, bool clean)
576 {
577         rtnl_lock();
578         __tun_detach(tfile, clean);
579         rtnl_unlock();
580 }
581 
582 static void tun_detach_all(struct net_device *dev)
583 {
584         struct tun_struct *tun = netdev_priv(dev);
585         struct tun_file *tfile, *tmp;
586         int i, n = tun->numqueues;
587 
588         for (i = 0; i < n; i++) {
589                 tfile = rtnl_dereference(tun->tfiles[i]);
590                 BUG_ON(!tfile);
591                 tfile->socket.sk->sk_shutdown = RCV_SHUTDOWN;
592                 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
593                 RCU_INIT_POINTER(tfile->tun, NULL);
594                 --tun->numqueues;
595         }
596         list_for_each_entry(tfile, &tun->disabled, next) {
597                 tfile->socket.sk->sk_shutdown = RCV_SHUTDOWN;
598                 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
599                 RCU_INIT_POINTER(tfile->tun, NULL);
600         }
601         BUG_ON(tun->numqueues != 0);
602 
603         synchronize_net();
604         for (i = 0; i < n; i++) {
605                 tfile = rtnl_dereference(tun->tfiles[i]);
606                 /* Drop read queue */
607                 tun_queue_purge(tfile);
608                 sock_put(&tfile->sk);
609         }
610         list_for_each_entry_safe(tfile, tmp, &tun->disabled, next) {
611                 tun_enable_queue(tfile);
612                 tun_queue_purge(tfile);
613                 sock_put(&tfile->sk);
614         }
615         BUG_ON(tun->numdisabled != 0);
616 
617         if (tun->flags & IFF_PERSIST)
618                 module_put(THIS_MODULE);
619 }
620 
621 static int tun_attach(struct tun_struct *tun, struct file *file, bool skip_filter)
622 {
623         struct tun_file *tfile = file->private_data;
624         struct net_device *dev = tun->dev;
625         int err;
626 
627         err = security_tun_dev_attach(tfile->socket.sk, tun->security);
628         if (err < 0)
629                 goto out;
630 
631         err = -EINVAL;
632         if (rtnl_dereference(tfile->tun) && !tfile->detached)
633                 goto out;
634 
635         err = -EBUSY;
636         if (!(tun->flags & IFF_MULTI_QUEUE) && tun->numqueues == 1)
637                 goto out;
638 
639         err = -E2BIG;
640         if (!tfile->detached &&
641             tun->numqueues + tun->numdisabled == MAX_TAP_QUEUES)
642                 goto out;
643 
644         err = 0;
645 
646         /* Re-attach the filter to persist device */
647         if (!skip_filter && (tun->filter_attached == true)) {
648                 lock_sock(tfile->socket.sk);
649                 err = sk_attach_filter(&tun->fprog, tfile->socket.sk);
650                 release_sock(tfile->socket.sk);
651                 if (!err)
652                         goto out;
653         }
654 
655         if (!tfile->detached &&
656             skb_array_init(&tfile->tx_array, dev->tx_queue_len, GFP_KERNEL)) {
657                 err = -ENOMEM;
658                 goto out;
659         }
660 
661         tfile->queue_index = tun->numqueues;
662         tfile->socket.sk->sk_shutdown &= ~RCV_SHUTDOWN;
663         rcu_assign_pointer(tfile->tun, tun);
664         rcu_assign_pointer(tun->tfiles[tun->numqueues], tfile);
665         tun->numqueues++;
666 
667         if (tfile->detached)
668                 tun_enable_queue(tfile);
669         else
670                 sock_hold(&tfile->sk);
671 
672         tun_set_real_num_queues(tun);
673 
674         /* device is allowed to go away first, so no need to hold extra
675          * refcnt.
676          */
677 
678 out:
679         return err;
680 }
681 
682 static struct tun_struct *__tun_get(struct tun_file *tfile)
683 {
684         struct tun_struct *tun;
685 
686         rcu_read_lock();
687         tun = rcu_dereference(tfile->tun);
688         if (tun)
689                 dev_hold(tun->dev);
690         rcu_read_unlock();
691 
692         return tun;
693 }
694 
695 static struct tun_struct *tun_get(struct file *file)
696 {
697         return __tun_get(file->private_data);
698 }
699 
700 static void tun_put(struct tun_struct *tun)
701 {
702         dev_put(tun->dev);
703 }
704 
705 /* TAP filtering */
706 static void addr_hash_set(u32 *mask, const u8 *addr)
707 {
708         int n = ether_crc(ETH_ALEN, addr) >> 26;
709         mask[n >> 5] |= (1 << (n & 31));
710 }
711 
712 static unsigned int addr_hash_test(const u32 *mask, const u8 *addr)
713 {
714         int n = ether_crc(ETH_ALEN, addr) >> 26;
715         return mask[n >> 5] & (1 << (n & 31));
716 }
717 
718 static int update_filter(struct tap_filter *filter, void __user *arg)
719 {
720         struct { u8 u[ETH_ALEN]; } *addr;
721         struct tun_filter uf;
722         int err, alen, n, nexact;
723 
724         if (copy_from_user(&uf, arg, sizeof(uf)))
725                 return -EFAULT;
726 
727         if (!uf.count) {
728                 /* Disabled */
729                 filter->count = 0;
730                 return 0;
731         }
732 
733         alen = ETH_ALEN * uf.count;
734         addr = memdup_user(arg + sizeof(uf), alen);
735         if (IS_ERR(addr))
736                 return PTR_ERR(addr);
737 
738         /* The filter is updated without holding any locks. Which is
739          * perfectly safe. We disable it first and in the worst
740          * case we'll accept a few undesired packets. */
741         filter->count = 0;
742         wmb();
743 
744         /* Use first set of addresses as an exact filter */
745         for (n = 0; n < uf.count && n < FLT_EXACT_COUNT; n++)
746                 memcpy(filter->addr[n], addr[n].u, ETH_ALEN);
747 
748         nexact = n;
749 
750         /* Remaining multicast addresses are hashed,
751          * unicast will leave the filter disabled. */
752         memset(filter->mask, 0, sizeof(filter->mask));
753         for (; n < uf.count; n++) {
754                 if (!is_multicast_ether_addr(addr[n].u)) {
755                         err = 0; /* no filter */
756                         goto free_addr;
757                 }
758                 addr_hash_set(filter->mask, addr[n].u);
759         }
760 
761         /* For ALLMULTI just set the mask to all ones.
762          * This overrides the mask populated above. */
763         if ((uf.flags & TUN_FLT_ALLMULTI))
764                 memset(filter->mask, ~0, sizeof(filter->mask));
765 
766         /* Now enable the filter */
767         wmb();
768         filter->count = nexact;
769 
770         /* Return the number of exact filters */
771         err = nexact;
772 free_addr:
773         kfree(addr);
774         return err;
775 }
776 
777 /* Returns: 0 - drop, !=0 - accept */
778 static int run_filter(struct tap_filter *filter, const struct sk_buff *skb)
779 {
780         /* Cannot use eth_hdr(skb) here because skb_mac_hdr() is incorrect
781          * at this point. */
782         struct ethhdr *eh = (struct ethhdr *) skb->data;
783         int i;
784 
785         /* Exact match */
786         for (i = 0; i < filter->count; i++)
787                 if (ether_addr_equal(eh->h_dest, filter->addr[i]))
788                         return 1;
789 
790         /* Inexact match (multicast only) */
791         if (is_multicast_ether_addr(eh->h_dest))
792                 return addr_hash_test(filter->mask, eh->h_dest);
793 
794         return 0;
795 }
796 
797 /*
798  * Checks whether the packet is accepted or not.
799  * Returns: 0 - drop, !=0 - accept
800  */
801 static int check_filter(struct tap_filter *filter, const struct sk_buff *skb)
802 {
803         if (!filter->count)
804                 return 1;
805 
806         return run_filter(filter, skb);
807 }
808 
809 /* Network device part of the driver */
810 
811 static const struct ethtool_ops tun_ethtool_ops;
812 
813 /* Net device detach from fd. */
814 static void tun_net_uninit(struct net_device *dev)
815 {
816         tun_detach_all(dev);
817 }
818 
819 /* Net device open. */
820 static int tun_net_open(struct net_device *dev)
821 {
822         netif_tx_start_all_queues(dev);
823         return 0;
824 }
825 
826 /* Net device close. */
827 static int tun_net_close(struct net_device *dev)
828 {
829         netif_tx_stop_all_queues(dev);
830         return 0;
831 }
832 
833 /* Net device start xmit */
834 static netdev_tx_t tun_net_xmit(struct sk_buff *skb, struct net_device *dev)
835 {
836         struct tun_struct *tun = netdev_priv(dev);
837         int txq = skb->queue_mapping;
838         struct tun_file *tfile;
839         u32 numqueues = 0;
840 
841         rcu_read_lock();
842         tfile = rcu_dereference(tun->tfiles[txq]);
843         numqueues = ACCESS_ONCE(tun->numqueues);
844 
845         /* Drop packet if interface is not attached */
846         if (txq >= numqueues)
847                 goto drop;
848 
849 #ifdef CONFIG_RPS
850         if (numqueues == 1 && static_key_false(&rps_needed)) {
851                 /* Select queue was not called for the skbuff, so we extract the
852                  * RPS hash and save it into the flow_table here.
853                  */
854                 __u32 rxhash;
855 
856                 rxhash = skb_get_hash(skb);
857                 if (rxhash) {
858                         struct tun_flow_entry *e;
859                         e = tun_flow_find(&tun->flows[tun_hashfn(rxhash)],
860                                         rxhash);
861                         if (e)
862                                 tun_flow_save_rps_rxhash(e, rxhash);
863                 }
864         }
865 #endif
866 
867         tun_debug(KERN_INFO, tun, "tun_net_xmit %d\n", skb->len);
868 
869         BUG_ON(!tfile);
870 
871         /* Drop if the filter does not like it.
872          * This is a noop if the filter is disabled.
873          * Filter can be enabled only for the TAP devices. */
874         if (!check_filter(&tun->txflt, skb))
875                 goto drop;
876 
877         if (tfile->socket.sk->sk_filter &&
878             sk_filter(tfile->socket.sk, skb))
879                 goto drop;
880 
881         /* Limit the number of packets queued by dividing txq length with the
882          * number of queues.
883          */
884         if (skb_queue_len(&tfile->socket.sk->sk_receive_queue) * numqueues
885                           >= dev->tx_queue_len)
886                 goto drop;
887 
888         if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC)))
889                 goto drop;
890 
891         skb_tx_timestamp(skb);
892 
893         /* Orphan the skb - required as we might hang on to it
894          * for indefinite time.
895          */
896         skb_orphan(skb);
897 
898         nf_reset(skb);
899 
900         if (skb_array_produce(&tfile->tx_array, skb))
901                 goto drop;
902 
903         /* Notify and wake up reader process */
904         if (tfile->flags & TUN_FASYNC)
905                 kill_fasync(&tfile->fasync, SIGIO, POLL_IN);
906         tfile->socket.sk->sk_data_ready(tfile->socket.sk);
907 
908         rcu_read_unlock();
909         return NETDEV_TX_OK;
910 
911 drop:
912         this_cpu_inc(tun->pcpu_stats->tx_dropped);
913         skb_tx_error(skb);
914         kfree_skb(skb);
915         rcu_read_unlock();
916         return NET_XMIT_DROP;
917 }
918 
919 static void tun_net_mclist(struct net_device *dev)
920 {
921         /*
922          * This callback is supposed to deal with mc filter in
923          * _rx_ path and has nothing to do with the _tx_ path.
924          * In rx path we always accept everything userspace gives us.
925          */
926 }
927 
928 #define MIN_MTU 68
929 #define MAX_MTU 65535
930 
931 static int
932 tun_net_change_mtu(struct net_device *dev, int new_mtu)
933 {
934         if (new_mtu < MIN_MTU || new_mtu + dev->hard_header_len > MAX_MTU)
935                 return -EINVAL;
936         dev->mtu = new_mtu;
937         return 0;
938 }
939 
940 static netdev_features_t tun_net_fix_features(struct net_device *dev,
941         netdev_features_t features)
942 {
943         struct tun_struct *tun = netdev_priv(dev);
944 
945         return (features & tun->set_features) | (features & ~TUN_USER_FEATURES);
946 }
947 #ifdef CONFIG_NET_POLL_CONTROLLER
948 static void tun_poll_controller(struct net_device *dev)
949 {
950         /*
951          * Tun only receives frames when:
952          * 1) the char device endpoint gets data from user space
953          * 2) the tun socket gets a sendmsg call from user space
954          * Since both of those are synchronous operations, we are guaranteed
955          * never to have pending data when we poll for it
956          * so there is nothing to do here but return.
957          * We need this though so netpoll recognizes us as an interface that
958          * supports polling, which enables bridge devices in virt setups to
959          * still use netconsole
960          */
961         return;
962 }
963 #endif
964 
965 static void tun_set_headroom(struct net_device *dev, int new_hr)
966 {
967         struct tun_struct *tun = netdev_priv(dev);
968 
969         if (new_hr < NET_SKB_PAD)
970                 new_hr = NET_SKB_PAD;
971 
972         tun->align = new_hr;
973 }
974 
975 static struct rtnl_link_stats64 *
976 tun_net_get_stats64(struct net_device *dev, struct rtnl_link_stats64 *stats)
977 {
978         u32 rx_dropped = 0, tx_dropped = 0, rx_frame_errors = 0;
979         struct tun_struct *tun = netdev_priv(dev);
980         struct tun_pcpu_stats *p;
981         int i;
982 
983         for_each_possible_cpu(i) {
984                 u64 rxpackets, rxbytes, txpackets, txbytes;
985                 unsigned int start;
986 
987                 p = per_cpu_ptr(tun->pcpu_stats, i);
988                 do {
989                         start = u64_stats_fetch_begin(&p->syncp);
990                         rxpackets       = p->rx_packets;
991                         rxbytes         = p->rx_bytes;
992                         txpackets       = p->tx_packets;
993                         txbytes         = p->tx_bytes;
994                 } while (u64_stats_fetch_retry(&p->syncp, start));
995 
996                 stats->rx_packets       += rxpackets;
997                 stats->rx_bytes         += rxbytes;
998                 stats->tx_packets       += txpackets;
999                 stats->tx_bytes         += txbytes;
1000 
1001                 /* u32 counters */
1002                 rx_dropped      += p->rx_dropped;
1003                 rx_frame_errors += p->rx_frame_errors;
1004                 tx_dropped      += p->tx_dropped;
1005         }
1006         stats->rx_dropped  = rx_dropped;
1007         stats->rx_frame_errors = rx_frame_errors;
1008         stats->tx_dropped = tx_dropped;
1009         return stats;
1010 }
1011 
1012 static const struct net_device_ops tun_netdev_ops = {
1013         .ndo_uninit             = tun_net_uninit,
1014         .ndo_open               = tun_net_open,
1015         .ndo_stop               = tun_net_close,
1016         .ndo_start_xmit         = tun_net_xmit,
1017         .ndo_change_mtu         = tun_net_change_mtu,
1018         .ndo_fix_features       = tun_net_fix_features,
1019         .ndo_select_queue       = tun_select_queue,
1020 #ifdef CONFIG_NET_POLL_CONTROLLER
1021         .ndo_poll_controller    = tun_poll_controller,
1022 #endif
1023         .ndo_set_rx_headroom    = tun_set_headroom,
1024         .ndo_get_stats64        = tun_net_get_stats64,
1025 };
1026 
1027 static const struct net_device_ops tap_netdev_ops = {
1028         .ndo_uninit             = tun_net_uninit,
1029         .ndo_open               = tun_net_open,
1030         .ndo_stop               = tun_net_close,
1031         .ndo_start_xmit         = tun_net_xmit,
1032         .ndo_change_mtu         = tun_net_change_mtu,
1033         .ndo_fix_features       = tun_net_fix_features,
1034         .ndo_set_rx_mode        = tun_net_mclist,
1035         .ndo_set_mac_address    = eth_mac_addr,
1036         .ndo_validate_addr      = eth_validate_addr,
1037         .ndo_select_queue       = tun_select_queue,
1038 #ifdef CONFIG_NET_POLL_CONTROLLER
1039         .ndo_poll_controller    = tun_poll_controller,
1040 #endif
1041         .ndo_features_check     = passthru_features_check,
1042         .ndo_set_rx_headroom    = tun_set_headroom,
1043         .ndo_get_stats64        = tun_net_get_stats64,
1044 };
1045 
1046 static void tun_flow_init(struct tun_struct *tun)
1047 {
1048         int i;
1049 
1050         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++)
1051                 INIT_HLIST_HEAD(&tun->flows[i]);
1052 
1053         tun->ageing_time = TUN_FLOW_EXPIRE;
1054         setup_timer(&tun->flow_gc_timer, tun_flow_cleanup, (unsigned long)tun);
1055         mod_timer(&tun->flow_gc_timer,
1056                   round_jiffies_up(jiffies + tun->ageing_time));
1057 }
1058 
1059 static void tun_flow_uninit(struct tun_struct *tun)
1060 {
1061         del_timer_sync(&tun->flow_gc_timer);
1062         tun_flow_flush(tun);
1063 }
1064 
1065 /* Initialize net device. */
1066 static void tun_net_init(struct net_device *dev)
1067 {
1068         struct tun_struct *tun = netdev_priv(dev);
1069 
1070         switch (tun->flags & TUN_TYPE_MASK) {
1071         case IFF_TUN:
1072                 dev->netdev_ops = &tun_netdev_ops;
1073 
1074                 /* Point-to-Point TUN Device */
1075                 dev->hard_header_len = 0;
1076                 dev->addr_len = 0;
1077                 dev->mtu = 1500;
1078 
1079                 /* Zero header length */
1080                 dev->type = ARPHRD_NONE;
1081                 dev->flags = IFF_POINTOPOINT | IFF_NOARP | IFF_MULTICAST;
1082                 break;
1083 
1084         case IFF_TAP:
1085                 dev->netdev_ops = &tap_netdev_ops;
1086                 /* Ethernet TAP Device */
1087                 ether_setup(dev);
1088                 dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1089                 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
1090 
1091                 eth_hw_addr_random(dev);
1092 
1093                 break;
1094         }
1095 }
1096 
1097 /* Character device part */
1098 
1099 /* Poll */
1100 static unsigned int tun_chr_poll(struct file *file, poll_table *wait)
1101 {
1102         struct tun_file *tfile = file->private_data;
1103         struct tun_struct *tun = __tun_get(tfile);
1104         struct sock *sk;
1105         unsigned int mask = 0;
1106 
1107         if (!tun)
1108                 return POLLERR;
1109 
1110         sk = tfile->socket.sk;
1111 
1112         tun_debug(KERN_INFO, tun, "tun_chr_poll\n");
1113 
1114         poll_wait(file, sk_sleep(sk), wait);
1115 
1116         if (!skb_array_empty(&tfile->tx_array))
1117                 mask |= POLLIN | POLLRDNORM;
1118 
1119         if (sock_writeable(sk) ||
1120             (!test_and_set_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags) &&
1121              sock_writeable(sk)))
1122                 mask |= POLLOUT | POLLWRNORM;
1123 
1124         if (tun->dev->reg_state != NETREG_REGISTERED)
1125                 mask = POLLERR;
1126 
1127         tun_put(tun);
1128         return mask;
1129 }
1130 
1131 /* prepad is the amount to reserve at front.  len is length after that.
1132  * linear is a hint as to how much to copy (usually headers). */
1133 static struct sk_buff *tun_alloc_skb(struct tun_file *tfile,
1134                                      size_t prepad, size_t len,
1135                                      size_t linear, int noblock)
1136 {
1137         struct sock *sk = tfile->socket.sk;
1138         struct sk_buff *skb;
1139         int err;
1140 
1141         /* Under a page?  Don't bother with paged skb. */
1142         if (prepad + len < PAGE_SIZE || !linear)
1143                 linear = len;
1144 
1145         skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
1146                                    &err, 0);
1147         if (!skb)
1148                 return ERR_PTR(err);
1149 
1150         skb_reserve(skb, prepad);
1151         skb_put(skb, linear);
1152         skb->data_len = len - linear;
1153         skb->len += len - linear;
1154 
1155         return skb;
1156 }
1157 
1158 /* Get packet from user space buffer */
1159 static ssize_t tun_get_user(struct tun_struct *tun, struct tun_file *tfile,
1160                             void *msg_control, struct iov_iter *from,
1161                             int noblock)
1162 {
1163         struct tun_pi pi = { 0, cpu_to_be16(ETH_P_IP) };
1164         struct sk_buff *skb;
1165         size_t total_len = iov_iter_count(from);
1166         size_t len = total_len, align = tun->align, linear;
1167         struct virtio_net_hdr gso = { 0 };
1168         struct tun_pcpu_stats *stats;
1169         int good_linear;
1170         int copylen;
1171         bool zerocopy = false;
1172         int err;
1173         u32 rxhash;
1174         ssize_t n;
1175 
1176         if (!(tun->dev->flags & IFF_UP))
1177                 return -EIO;
1178 
1179         if (!(tun->flags & IFF_NO_PI)) {
1180                 if (len < sizeof(pi))
1181                         return -EINVAL;
1182                 len -= sizeof(pi);
1183 
1184                 n = copy_from_iter(&pi, sizeof(pi), from);
1185                 if (n != sizeof(pi))
1186                         return -EFAULT;
1187         }
1188 
1189         if (tun->flags & IFF_VNET_HDR) {
1190                 if (len < tun->vnet_hdr_sz)
1191                         return -EINVAL;
1192                 len -= tun->vnet_hdr_sz;
1193 
1194                 n = copy_from_iter(&gso, sizeof(gso), from);
1195                 if (n != sizeof(gso))
1196                         return -EFAULT;
1197 
1198                 if ((gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
1199                     tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2 > tun16_to_cpu(tun, gso.hdr_len))
1200                         gso.hdr_len = cpu_to_tun16(tun, tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2);
1201 
1202                 if (tun16_to_cpu(tun, gso.hdr_len) > len)
1203                         return -EINVAL;
1204                 iov_iter_advance(from, tun->vnet_hdr_sz - sizeof(gso));
1205         }
1206 
1207         if ((tun->flags & TUN_TYPE_MASK) == IFF_TAP) {
1208                 align += NET_IP_ALIGN;
1209                 if (unlikely(len < ETH_HLEN ||
1210                              (gso.hdr_len && tun16_to_cpu(tun, gso.hdr_len) < ETH_HLEN)))
1211                         return -EINVAL;
1212         }
1213 
1214         good_linear = SKB_MAX_HEAD(align);
1215 
1216         if (msg_control) {
1217                 struct iov_iter i = *from;
1218 
1219                 /* There are 256 bytes to be copied in skb, so there is
1220                  * enough room for skb expand head in case it is used.
1221                  * The rest of the buffer is mapped from userspace.
1222                  */
1223                 copylen = gso.hdr_len ? tun16_to_cpu(tun, gso.hdr_len) : GOODCOPY_LEN;
1224                 if (copylen > good_linear)
1225                         copylen = good_linear;
1226                 linear = copylen;
1227                 iov_iter_advance(&i, copylen);
1228                 if (iov_iter_npages(&i, INT_MAX) <= MAX_SKB_FRAGS)
1229                         zerocopy = true;
1230         }
1231 
1232         if (!zerocopy) {
1233                 copylen = len;
1234                 if (tun16_to_cpu(tun, gso.hdr_len) > good_linear)
1235                         linear = good_linear;
1236                 else
1237                         linear = tun16_to_cpu(tun, gso.hdr_len);
1238         }
1239 
1240         skb = tun_alloc_skb(tfile, align, copylen, linear, noblock);
1241         if (IS_ERR(skb)) {
1242                 if (PTR_ERR(skb) != -EAGAIN)
1243                         this_cpu_inc(tun->pcpu_stats->rx_dropped);
1244                 return PTR_ERR(skb);
1245         }
1246 
1247         if (zerocopy)
1248                 err = zerocopy_sg_from_iter(skb, from);
1249         else
1250                 err = skb_copy_datagram_from_iter(skb, 0, from, len);
1251 
1252         if (err) {
1253                 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1254                 kfree_skb(skb);
1255                 return -EFAULT;
1256         }
1257 
1258         err = virtio_net_hdr_to_skb(skb, &gso, tun_is_little_endian(tun));
1259         if (err) {
1260                 this_cpu_inc(tun->pcpu_stats->rx_frame_errors);
1261                 kfree_skb(skb);
1262                 return -EINVAL;
1263         }
1264 
1265         switch (tun->flags & TUN_TYPE_MASK) {
1266         case IFF_TUN:
1267                 if (tun->flags & IFF_NO_PI) {
1268                         switch (skb->data[0] & 0xf0) {
1269                         case 0x40:
1270                                 pi.proto = htons(ETH_P_IP);
1271                                 break;
1272                         case 0x60:
1273                                 pi.proto = htons(ETH_P_IPV6);
1274                                 break;
1275                         default:
1276                                 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1277                                 kfree_skb(skb);
1278                                 return -EINVAL;
1279                         }
1280                 }
1281 
1282                 skb_reset_mac_header(skb);
1283                 skb->protocol = pi.proto;
1284                 skb->dev = tun->dev;
1285                 break;
1286         case IFF_TAP:
1287                 skb->protocol = eth_type_trans(skb, tun->dev);
1288                 break;
1289         }
1290 
1291         /* copy skb_ubuf_info for callback when skb has no error */
1292         if (zerocopy) {
1293                 skb_shinfo(skb)->destructor_arg = msg_control;
1294                 skb_shinfo(skb)->tx_flags |= SKBTX_DEV_ZEROCOPY;
1295                 skb_shinfo(skb)->tx_flags |= SKBTX_SHARED_FRAG;
1296         } else if (msg_control) {
1297                 struct ubuf_info *uarg = msg_control;
1298                 uarg->callback(uarg, false);
1299         }
1300 
1301         skb_reset_network_header(skb);
1302         skb_probe_transport_header(skb, 0);
1303 
1304         rxhash = skb_get_hash(skb);
1305         netif_rx_ni(skb);
1306 
1307         stats = get_cpu_ptr(tun->pcpu_stats);
1308         u64_stats_update_begin(&stats->syncp);
1309         stats->rx_packets++;
1310         stats->rx_bytes += len;
1311         u64_stats_update_end(&stats->syncp);
1312         put_cpu_ptr(stats);
1313 
1314         tun_flow_update(tun, rxhash, tfile);
1315         return total_len;
1316 }
1317 
1318 static ssize_t tun_chr_write_iter(struct kiocb *iocb, struct iov_iter *from)
1319 {
1320         struct file *file = iocb->ki_filp;
1321         struct tun_struct *tun = tun_get(file);
1322         struct tun_file *tfile = file->private_data;
1323         ssize_t result;
1324 
1325         if (!tun)
1326                 return -EBADFD;
1327 
1328         result = tun_get_user(tun, tfile, NULL, from, file->f_flags & O_NONBLOCK);
1329 
1330         tun_put(tun);
1331         return result;
1332 }
1333 
1334 /* Put packet to the user space buffer */
1335 static ssize_t tun_put_user(struct tun_struct *tun,
1336                             struct tun_file *tfile,
1337                             struct sk_buff *skb,
1338                             struct iov_iter *iter)
1339 {
1340         struct tun_pi pi = { 0, skb->protocol };
1341         struct tun_pcpu_stats *stats;
1342         ssize_t total;
1343         int vlan_offset = 0;
1344         int vlan_hlen = 0;
1345         int vnet_hdr_sz = 0;
1346 
1347         if (skb_vlan_tag_present(skb))
1348                 vlan_hlen = VLAN_HLEN;
1349 
1350         if (tun->flags & IFF_VNET_HDR)
1351                 vnet_hdr_sz = tun->vnet_hdr_sz;
1352 
1353         total = skb->len + vlan_hlen + vnet_hdr_sz;
1354 
1355         if (!(tun->flags & IFF_NO_PI)) {
1356                 if (iov_iter_count(iter) < sizeof(pi))
1357                         return -EINVAL;
1358 
1359                 total += sizeof(pi);
1360                 if (iov_iter_count(iter) < total) {
1361                         /* Packet will be striped */
1362                         pi.flags |= TUN_PKT_STRIP;
1363                 }
1364 
1365                 if (copy_to_iter(&pi, sizeof(pi), iter) != sizeof(pi))
1366                         return -EFAULT;
1367         }
1368 
1369         if (vnet_hdr_sz) {
1370                 struct virtio_net_hdr gso = { 0 }; /* no info leak */
1371                 int ret;
1372 
1373                 if (iov_iter_count(iter) < vnet_hdr_sz)
1374                         return -EINVAL;
1375 
1376                 ret = virtio_net_hdr_from_skb(skb, &gso,
1377                                               tun_is_little_endian(tun));
1378                 if (ret) {
1379                         struct skb_shared_info *sinfo = skb_shinfo(skb);
1380                         pr_err("unexpected GSO type: "
1381                                "0x%x, gso_size %d, hdr_len %d\n",
1382                                sinfo->gso_type, tun16_to_cpu(tun, gso.gso_size),
1383                                tun16_to_cpu(tun, gso.hdr_len));
1384                         print_hex_dump(KERN_ERR, "tun: ",
1385                                        DUMP_PREFIX_NONE,
1386                                        16, 1, skb->head,
1387                                        min((int)tun16_to_cpu(tun, gso.hdr_len), 64), true);
1388                         WARN_ON_ONCE(1);
1389                         return -EINVAL;
1390                 }
1391 
1392                 if (copy_to_iter(&gso, sizeof(gso), iter) != sizeof(gso))
1393                         return -EFAULT;
1394 
1395                 iov_iter_advance(iter, vnet_hdr_sz - sizeof(gso));
1396         }
1397 
1398         if (vlan_hlen) {
1399                 int ret;
1400                 struct {
1401                         __be16 h_vlan_proto;
1402                         __be16 h_vlan_TCI;
1403                 } veth;
1404 
1405                 veth.h_vlan_proto = skb->vlan_proto;
1406                 veth.h_vlan_TCI = htons(skb_vlan_tag_get(skb));
1407 
1408                 vlan_offset = offsetof(struct vlan_ethhdr, h_vlan_proto);
1409 
1410                 ret = skb_copy_datagram_iter(skb, 0, iter, vlan_offset);
1411                 if (ret || !iov_iter_count(iter))
1412                         goto done;
1413 
1414                 ret = copy_to_iter(&veth, sizeof(veth), iter);
1415                 if (ret != sizeof(veth) || !iov_iter_count(iter))
1416                         goto done;
1417         }
1418 
1419         skb_copy_datagram_iter(skb, vlan_offset, iter, skb->len - vlan_offset);
1420 
1421 done:
1422         /* caller is in process context, */
1423         stats = get_cpu_ptr(tun->pcpu_stats);
1424         u64_stats_update_begin(&stats->syncp);
1425         stats->tx_packets++;
1426         stats->tx_bytes += skb->len + vlan_hlen;
1427         u64_stats_update_end(&stats->syncp);
1428         put_cpu_ptr(tun->pcpu_stats);
1429 
1430         return total;
1431 }
1432 
1433 static struct sk_buff *tun_ring_recv(struct tun_file *tfile, int noblock,
1434                                      int *err)
1435 {
1436         DECLARE_WAITQUEUE(wait, current);
1437         struct sk_buff *skb = NULL;
1438         int error = 0;
1439 
1440         skb = skb_array_consume(&tfile->tx_array);
1441         if (skb)
1442                 goto out;
1443         if (noblock) {
1444                 error = -EAGAIN;
1445                 goto out;
1446         }
1447 
1448         add_wait_queue(&tfile->wq.wait, &wait);
1449         current->state = TASK_INTERRUPTIBLE;
1450 
1451         while (1) {
1452                 skb = skb_array_consume(&tfile->tx_array);
1453                 if (skb)
1454                         break;
1455                 if (signal_pending(current)) {
1456                         error = -ERESTARTSYS;
1457                         break;
1458                 }
1459                 if (tfile->socket.sk->sk_shutdown & RCV_SHUTDOWN) {
1460                         error = -EFAULT;
1461                         break;
1462                 }
1463 
1464                 schedule();
1465         }
1466 
1467         current->state = TASK_RUNNING;
1468         remove_wait_queue(&tfile->wq.wait, &wait);
1469 
1470 out:
1471         *err = error;
1472         return skb;
1473 }
1474 
1475 static ssize_t tun_do_read(struct tun_struct *tun, struct tun_file *tfile,
1476                            struct iov_iter *to,
1477                            int noblock)
1478 {
1479         struct sk_buff *skb;
1480         ssize_t ret;
1481         int err;
1482 
1483         tun_debug(KERN_INFO, tun, "tun_do_read\n");
1484 
1485         if (!iov_iter_count(to))
1486                 return 0;
1487 
1488         /* Read frames from ring */
1489         skb = tun_ring_recv(tfile, noblock, &err);
1490         if (!skb)
1491                 return err;
1492 
1493         ret = tun_put_user(tun, tfile, skb, to);
1494         if (unlikely(ret < 0))
1495                 kfree_skb(skb);
1496         else
1497                 consume_skb(skb);
1498 
1499         return ret;
1500 }
1501 
1502 static ssize_t tun_chr_read_iter(struct kiocb *iocb, struct iov_iter *to)
1503 {
1504         struct file *file = iocb->ki_filp;
1505         struct tun_file *tfile = file->private_data;
1506         struct tun_struct *tun = __tun_get(tfile);
1507         ssize_t len = iov_iter_count(to), ret;
1508 
1509         if (!tun)
1510                 return -EBADFD;
1511         ret = tun_do_read(tun, tfile, to, file->f_flags & O_NONBLOCK);
1512         ret = min_t(ssize_t, ret, len);
1513         if (ret > 0)
1514                 iocb->ki_pos = ret;
1515         tun_put(tun);
1516         return ret;
1517 }
1518 
1519 static void tun_free_netdev(struct net_device *dev)
1520 {
1521         struct tun_struct *tun = netdev_priv(dev);
1522 
1523         BUG_ON(!(list_empty(&tun->disabled)));
1524         free_percpu(tun->pcpu_stats);
1525         tun_flow_uninit(tun);
1526         security_tun_dev_free_security(tun->security);
1527         free_netdev(dev);
1528 }
1529 
1530 static void tun_setup(struct net_device *dev)
1531 {
1532         struct tun_struct *tun = netdev_priv(dev);
1533 
1534         tun->owner = INVALID_UID;
1535         tun->group = INVALID_GID;
1536 
1537         dev->ethtool_ops = &tun_ethtool_ops;
1538         dev->destructor = tun_free_netdev;
1539         /* We prefer our own queue length */
1540         dev->tx_queue_len = TUN_READQ_SIZE;
1541 }
1542 
1543 /* Trivial set of netlink ops to allow deleting tun or tap
1544  * device with netlink.
1545  */
1546 static int tun_validate(struct nlattr *tb[], struct nlattr *data[])
1547 {
1548         return -EINVAL;
1549 }
1550 
1551 static struct rtnl_link_ops tun_link_ops __read_mostly = {
1552         .kind           = DRV_NAME,
1553         .priv_size      = sizeof(struct tun_struct),
1554         .setup          = tun_setup,
1555         .validate       = tun_validate,
1556 };
1557 
1558 static void tun_sock_write_space(struct sock *sk)
1559 {
1560         struct tun_file *tfile;
1561         wait_queue_head_t *wqueue;
1562 
1563         if (!sock_writeable(sk))
1564                 return;
1565 
1566         if (!test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags))
1567                 return;
1568 
1569         wqueue = sk_sleep(sk);
1570         if (wqueue && waitqueue_active(wqueue))
1571                 wake_up_interruptible_sync_poll(wqueue, POLLOUT |
1572                                                 POLLWRNORM | POLLWRBAND);
1573 
1574         tfile = container_of(sk, struct tun_file, sk);
1575         kill_fasync(&tfile->fasync, SIGIO, POLL_OUT);
1576 }
1577 
1578 static int tun_sendmsg(struct socket *sock, struct msghdr *m, size_t total_len)
1579 {
1580         int ret;
1581         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
1582         struct tun_struct *tun = __tun_get(tfile);
1583 
1584         if (!tun)
1585                 return -EBADFD;
1586 
1587         ret = tun_get_user(tun, tfile, m->msg_control, &m->msg_iter,
1588                            m->msg_flags & MSG_DONTWAIT);
1589         tun_put(tun);
1590         return ret;
1591 }
1592 
1593 static int tun_recvmsg(struct socket *sock, struct msghdr *m, size_t total_len,
1594                        int flags)
1595 {
1596         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
1597         struct tun_struct *tun = __tun_get(tfile);
1598         int ret;
1599 
1600         if (!tun)
1601                 return -EBADFD;
1602 
1603         if (flags & ~(MSG_DONTWAIT|MSG_TRUNC|MSG_ERRQUEUE)) {
1604                 ret = -EINVAL;
1605                 goto out;
1606         }
1607         if (flags & MSG_ERRQUEUE) {
1608                 ret = sock_recv_errqueue(sock->sk, m, total_len,
1609                                          SOL_PACKET, TUN_TX_TIMESTAMP);
1610                 goto out;
1611         }
1612         ret = tun_do_read(tun, tfile, &m->msg_iter, flags & MSG_DONTWAIT);
1613         if (ret > (ssize_t)total_len) {
1614                 m->msg_flags |= MSG_TRUNC;
1615                 ret = flags & MSG_TRUNC ? ret : total_len;
1616         }
1617 out:
1618         tun_put(tun);
1619         return ret;
1620 }
1621 
1622 static int tun_peek_len(struct socket *sock)
1623 {
1624         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
1625         struct tun_struct *tun;
1626         int ret = 0;
1627 
1628         tun = __tun_get(tfile);
1629         if (!tun)
1630                 return 0;
1631 
1632         ret = skb_array_peek_len(&tfile->tx_array);
1633         tun_put(tun);
1634 
1635         return ret;
1636 }
1637 
1638 /* Ops structure to mimic raw sockets with tun */
1639 static const struct proto_ops tun_socket_ops = {
1640         .peek_len = tun_peek_len,
1641         .sendmsg = tun_sendmsg,
1642         .recvmsg = tun_recvmsg,
1643 };
1644 
1645 static struct proto tun_proto = {
1646         .name           = "tun",
1647         .owner          = THIS_MODULE,
1648         .obj_size       = sizeof(struct tun_file),
1649 };
1650 
1651 static int tun_flags(struct tun_struct *tun)
1652 {
1653         return tun->flags & (TUN_FEATURES | IFF_PERSIST | IFF_TUN | IFF_TAP);
1654 }
1655 
1656 static ssize_t tun_show_flags(struct device *dev, struct device_attribute *attr,
1657                               char *buf)
1658 {
1659         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1660         return sprintf(buf, "0x%x\n", tun_flags(tun));
1661 }
1662 
1663 static ssize_t tun_show_owner(struct device *dev, struct device_attribute *attr,
1664                               char *buf)
1665 {
1666         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1667         return uid_valid(tun->owner)?
1668                 sprintf(buf, "%u\n",
1669                         from_kuid_munged(current_user_ns(), tun->owner)):
1670                 sprintf(buf, "-1\n");
1671 }
1672 
1673 static ssize_t tun_show_group(struct device *dev, struct device_attribute *attr,
1674                               char *buf)
1675 {
1676         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1677         return gid_valid(tun->group) ?
1678                 sprintf(buf, "%u\n",
1679                         from_kgid_munged(current_user_ns(), tun->group)):
1680                 sprintf(buf, "-1\n");
1681 }
1682 
1683 static DEVICE_ATTR(tun_flags, 0444, tun_show_flags, NULL);
1684 static DEVICE_ATTR(owner, 0444, tun_show_owner, NULL);
1685 static DEVICE_ATTR(group, 0444, tun_show_group, NULL);
1686 
1687 static struct attribute *tun_dev_attrs[] = {
1688         &dev_attr_tun_flags.attr,
1689         &dev_attr_owner.attr,
1690         &dev_attr_group.attr,
1691         NULL
1692 };
1693 
1694 static const struct attribute_group tun_attr_group = {
1695         .attrs = tun_dev_attrs
1696 };
1697 
1698 static int tun_set_iff(struct net *net, struct file *file, struct ifreq *ifr)
1699 {
1700         struct tun_struct *tun;
1701         struct tun_file *tfile = file->private_data;
1702         struct net_device *dev;
1703         int err;
1704 
1705         if (tfile->detached)
1706                 return -EINVAL;
1707 
1708         dev = __dev_get_by_name(net, ifr->ifr_name);
1709         if (dev) {
1710                 if (ifr->ifr_flags & IFF_TUN_EXCL)
1711                         return -EBUSY;
1712                 if ((ifr->ifr_flags & IFF_TUN) && dev->netdev_ops == &tun_netdev_ops)
1713                         tun = netdev_priv(dev);
1714                 else if ((ifr->ifr_flags & IFF_TAP) && dev->netdev_ops == &tap_netdev_ops)
1715                         tun = netdev_priv(dev);
1716                 else
1717                         return -EINVAL;
1718 
1719                 if (!!(ifr->ifr_flags & IFF_MULTI_QUEUE) !=
1720                     !!(tun->flags & IFF_MULTI_QUEUE))
1721                         return -EINVAL;
1722 
1723                 if (tun_not_capable(tun))
1724                         return -EPERM;
1725                 err = security_tun_dev_open(tun->security);
1726                 if (err < 0)
1727                         return err;
1728 
1729                 err = tun_attach(tun, file, ifr->ifr_flags & IFF_NOFILTER);
1730                 if (err < 0)
1731                         return err;
1732 
1733                 if (tun->flags & IFF_MULTI_QUEUE &&
1734                     (tun->numqueues + tun->numdisabled > 1)) {
1735                         /* One or more queue has already been attached, no need
1736                          * to initialize the device again.
1737                          */
1738                         return 0;
1739                 }
1740         }
1741         else {
1742                 char *name;
1743                 unsigned long flags = 0;
1744                 int queues = ifr->ifr_flags & IFF_MULTI_QUEUE ?
1745                              MAX_TAP_QUEUES : 1;
1746 
1747                 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
1748                         return -EPERM;
1749                 err = security_tun_dev_create();
1750                 if (err < 0)
1751                         return err;
1752 
1753                 /* Set dev type */
1754                 if (ifr->ifr_flags & IFF_TUN) {
1755                         /* TUN device */
1756                         flags |= IFF_TUN;
1757                         name = "tun%d";
1758                 } else if (ifr->ifr_flags & IFF_TAP) {
1759                         /* TAP device */
1760                         flags |= IFF_TAP;
1761                         name = "tap%d";
1762                 } else
1763                         return -EINVAL;
1764 
1765                 if (*ifr->ifr_name)
1766                         name = ifr->ifr_name;
1767 
1768                 dev = alloc_netdev_mqs(sizeof(struct tun_struct), name,
1769                                        NET_NAME_UNKNOWN, tun_setup, queues,
1770                                        queues);
1771 
1772                 if (!dev)
1773                         return -ENOMEM;
1774 
1775                 dev_net_set(dev, net);
1776                 dev->rtnl_link_ops = &tun_link_ops;
1777                 dev->ifindex = tfile->ifindex;
1778                 dev->sysfs_groups[0] = &tun_attr_group;
1779 
1780                 tun = netdev_priv(dev);
1781                 tun->dev = dev;
1782                 tun->flags = flags;
1783                 tun->txflt.count = 0;
1784                 tun->vnet_hdr_sz = sizeof(struct virtio_net_hdr);
1785 
1786                 tun->align = NET_SKB_PAD;
1787                 tun->filter_attached = false;
1788                 tun->sndbuf = tfile->socket.sk->sk_sndbuf;
1789 
1790                 tun->pcpu_stats = netdev_alloc_pcpu_stats(struct tun_pcpu_stats);
1791                 if (!tun->pcpu_stats) {
1792                         err = -ENOMEM;
1793                         goto err_free_dev;
1794                 }
1795 
1796                 spin_lock_init(&tun->lock);
1797 
1798                 err = security_tun_dev_alloc_security(&tun->security);
1799                 if (err < 0)
1800                         goto err_free_stat;
1801 
1802                 tun_net_init(dev);
1803                 tun_flow_init(tun);
1804 
1805                 dev->hw_features = NETIF_F_SG | NETIF_F_FRAGLIST |
1806                                    TUN_USER_FEATURES | NETIF_F_HW_VLAN_CTAG_TX |
1807                                    NETIF_F_HW_VLAN_STAG_TX;
1808                 dev->features = dev->hw_features | NETIF_F_LLTX;
1809                 dev->vlan_features = dev->features &
1810                                      ~(NETIF_F_HW_VLAN_CTAG_TX |
1811                                        NETIF_F_HW_VLAN_STAG_TX);
1812 
1813                 INIT_LIST_HEAD(&tun->disabled);
1814                 err = tun_attach(tun, file, false);
1815                 if (err < 0)
1816                         goto err_free_flow;
1817 
1818                 err = register_netdevice(tun->dev);
1819                 if (err < 0)
1820                         goto err_detach;
1821         }
1822 
1823         netif_carrier_on(tun->dev);
1824 
1825         tun_debug(KERN_INFO, tun, "tun_set_iff\n");
1826 
1827         tun->flags = (tun->flags & ~TUN_FEATURES) |
1828                 (ifr->ifr_flags & TUN_FEATURES);
1829 
1830         /* Make sure persistent devices do not get stuck in
1831          * xoff state.
1832          */
1833         if (netif_running(tun->dev))
1834                 netif_tx_wake_all_queues(tun->dev);
1835 
1836         strcpy(ifr->ifr_name, tun->dev->name);
1837         return 0;
1838 
1839 err_detach:
1840         tun_detach_all(dev);
1841 err_free_flow:
1842         tun_flow_uninit(tun);
1843         security_tun_dev_free_security(tun->security);
1844 err_free_stat:
1845         free_percpu(tun->pcpu_stats);
1846 err_free_dev:
1847         free_netdev(dev);
1848         return err;
1849 }
1850 
1851 static void tun_get_iff(struct net *net, struct tun_struct *tun,
1852                        struct ifreq *ifr)
1853 {
1854         tun_debug(KERN_INFO, tun, "tun_get_iff\n");
1855 
1856         strcpy(ifr->ifr_name, tun->dev->name);
1857 
1858         ifr->ifr_flags = tun_flags(tun);
1859 
1860 }
1861 
1862 /* This is like a cut-down ethtool ops, except done via tun fd so no
1863  * privs required. */
1864 static int set_offload(struct tun_struct *tun, unsigned long arg)
1865 {
1866         netdev_features_t features = 0;
1867 
1868         if (arg & TUN_F_CSUM) {
1869                 features |= NETIF_F_HW_CSUM;
1870                 arg &= ~TUN_F_CSUM;
1871 
1872                 if (arg & (TUN_F_TSO4|TUN_F_TSO6)) {
1873                         if (arg & TUN_F_TSO_ECN) {
1874                                 features |= NETIF_F_TSO_ECN;
1875                                 arg &= ~TUN_F_TSO_ECN;
1876                         }
1877                         if (arg & TUN_F_TSO4)
1878                                 features |= NETIF_F_TSO;
1879                         if (arg & TUN_F_TSO6)
1880                                 features |= NETIF_F_TSO6;
1881                         arg &= ~(TUN_F_TSO4|TUN_F_TSO6);
1882                 }
1883 
1884                 if (arg & TUN_F_UFO) {
1885                         features |= NETIF_F_UFO;
1886                         arg &= ~TUN_F_UFO;
1887                 }
1888         }
1889 
1890         /* This gives the user a way to test for new features in future by
1891          * trying to set them. */
1892         if (arg)
1893                 return -EINVAL;
1894 
1895         tun->set_features = features;
1896         netdev_update_features(tun->dev);
1897 
1898         return 0;
1899 }
1900 
1901 static void tun_detach_filter(struct tun_struct *tun, int n)
1902 {
1903         int i;
1904         struct tun_file *tfile;
1905 
1906         for (i = 0; i < n; i++) {
1907                 tfile = rtnl_dereference(tun->tfiles[i]);
1908                 lock_sock(tfile->socket.sk);
1909                 sk_detach_filter(tfile->socket.sk);
1910                 release_sock(tfile->socket.sk);
1911         }
1912 
1913         tun->filter_attached = false;
1914 }
1915 
1916 static int tun_attach_filter(struct tun_struct *tun)
1917 {
1918         int i, ret = 0;
1919         struct tun_file *tfile;
1920 
1921         for (i = 0; i < tun->numqueues; i++) {
1922                 tfile = rtnl_dereference(tun->tfiles[i]);
1923                 lock_sock(tfile->socket.sk);
1924                 ret = sk_attach_filter(&tun->fprog, tfile->socket.sk);
1925                 release_sock(tfile->socket.sk);
1926                 if (ret) {
1927                         tun_detach_filter(tun, i);
1928                         return ret;
1929                 }
1930         }
1931 
1932         tun->filter_attached = true;
1933         return ret;
1934 }
1935 
1936 static void tun_set_sndbuf(struct tun_struct *tun)
1937 {
1938         struct tun_file *tfile;
1939         int i;
1940 
1941         for (i = 0; i < tun->numqueues; i++) {
1942                 tfile = rtnl_dereference(tun->tfiles[i]);
1943                 tfile->socket.sk->sk_sndbuf = tun->sndbuf;
1944         }
1945 }
1946 
1947 static int tun_set_queue(struct file *file, struct ifreq *ifr)
1948 {
1949         struct tun_file *tfile = file->private_data;
1950         struct tun_struct *tun;
1951         int ret = 0;
1952 
1953         rtnl_lock();
1954 
1955         if (ifr->ifr_flags & IFF_ATTACH_QUEUE) {
1956                 tun = tfile->detached;
1957                 if (!tun) {
1958                         ret = -EINVAL;
1959                         goto unlock;
1960                 }
1961                 ret = security_tun_dev_attach_queue(tun->security);
1962                 if (ret < 0)
1963                         goto unlock;
1964                 ret = tun_attach(tun, file, false);
1965         } else if (ifr->ifr_flags & IFF_DETACH_QUEUE) {
1966                 tun = rtnl_dereference(tfile->tun);
1967                 if (!tun || !(tun->flags & IFF_MULTI_QUEUE) || tfile->detached)
1968                         ret = -EINVAL;
1969                 else
1970                         __tun_detach(tfile, false);
1971         } else
1972                 ret = -EINVAL;
1973 
1974 unlock:
1975         rtnl_unlock();
1976         return ret;
1977 }
1978 
1979 static long __tun_chr_ioctl(struct file *file, unsigned int cmd,
1980                             unsigned long arg, int ifreq_len)
1981 {
1982         struct tun_file *tfile = file->private_data;
1983         struct tun_struct *tun;
1984         void __user* argp = (void __user*)arg;
1985         struct ifreq ifr;
1986         kuid_t owner;
1987         kgid_t group;
1988         int sndbuf;
1989         int vnet_hdr_sz;
1990         unsigned int ifindex;
1991         int le;
1992         int ret;
1993 
1994         if (cmd == TUNSETIFF || cmd == TUNSETQUEUE || _IOC_TYPE(cmd) == 0x89) {
1995                 if (copy_from_user(&ifr, argp, ifreq_len))
1996                         return -EFAULT;
1997         } else {
1998                 memset(&ifr, 0, sizeof(ifr));
1999         }
2000         if (cmd == TUNGETFEATURES) {
2001                 /* Currently this just means: "what IFF flags are valid?".
2002                  * This is needed because we never checked for invalid flags on
2003                  * TUNSETIFF.
2004                  */
2005                 return put_user(IFF_TUN | IFF_TAP | TUN_FEATURES,
2006                                 (unsigned int __user*)argp);
2007         } else if (cmd == TUNSETQUEUE)
2008                 return tun_set_queue(file, &ifr);
2009 
2010         ret = 0;
2011         rtnl_lock();
2012 
2013         tun = __tun_get(tfile);
2014         if (cmd == TUNSETIFF && !tun) {
2015                 ifr.ifr_name[IFNAMSIZ-1] = '\0';
2016 
2017                 ret = tun_set_iff(sock_net(&tfile->sk), file, &ifr);
2018 
2019                 if (ret)
2020                         goto unlock;
2021 
2022                 if (copy_to_user(argp, &ifr, ifreq_len))
2023                         ret = -EFAULT;
2024                 goto unlock;
2025         }
2026         if (cmd == TUNSETIFINDEX) {
2027                 ret = -EPERM;
2028                 if (tun)
2029                         goto unlock;
2030 
2031                 ret = -EFAULT;
2032                 if (copy_from_user(&ifindex, argp, sizeof(ifindex)))
2033                         goto unlock;
2034 
2035                 ret = 0;
2036                 tfile->ifindex = ifindex;
2037                 goto unlock;
2038         }
2039 
2040         ret = -EBADFD;
2041         if (!tun)
2042                 goto unlock;
2043 
2044         tun_debug(KERN_INFO, tun, "tun_chr_ioctl cmd %u\n", cmd);
2045 
2046         ret = 0;
2047         switch (cmd) {
2048         case TUNGETIFF:
2049                 tun_get_iff(current->nsproxy->net_ns, tun, &ifr);
2050 
2051                 if (tfile->detached)
2052                         ifr.ifr_flags |= IFF_DETACH_QUEUE;
2053                 if (!tfile->socket.sk->sk_filter)
2054                         ifr.ifr_flags |= IFF_NOFILTER;
2055 
2056                 if (copy_to_user(argp, &ifr, ifreq_len))
2057                         ret = -EFAULT;
2058                 break;
2059 
2060         case TUNSETNOCSUM:
2061                 /* Disable/Enable checksum */
2062 
2063                 /* [unimplemented] */
2064                 tun_debug(KERN_INFO, tun, "ignored: set checksum %s\n",
2065                           arg ? "disabled" : "enabled");
2066                 break;
2067 
2068         case TUNSETPERSIST:
2069                 /* Disable/Enable persist mode. Keep an extra reference to the
2070                  * module to prevent the module being unprobed.
2071                  */
2072                 if (arg && !(tun->flags & IFF_PERSIST)) {
2073                         tun->flags |= IFF_PERSIST;
2074                         __module_get(THIS_MODULE);
2075                 }
2076                 if (!arg && (tun->flags & IFF_PERSIST)) {
2077                         tun->flags &= ~IFF_PERSIST;
2078                         module_put(THIS_MODULE);
2079                 }
2080 
2081                 tun_debug(KERN_INFO, tun, "persist %s\n",
2082                           arg ? "enabled" : "disabled");
2083                 break;
2084 
2085         case TUNSETOWNER:
2086                 /* Set owner of the device */
2087                 owner = make_kuid(current_user_ns(), arg);
2088                 if (!uid_valid(owner)) {
2089                         ret = -EINVAL;
2090                         break;
2091                 }
2092                 tun->owner = owner;
2093                 tun_debug(KERN_INFO, tun, "owner set to %u\n",
2094                           from_kuid(&init_user_ns, tun->owner));
2095                 break;
2096 
2097         case TUNSETGROUP:
2098                 /* Set group of the device */
2099                 group = make_kgid(current_user_ns(), arg);
2100                 if (!gid_valid(group)) {
2101                         ret = -EINVAL;
2102                         break;
2103                 }
2104                 tun->group = group;
2105                 tun_debug(KERN_INFO, tun, "group set to %u\n",
2106                           from_kgid(&init_user_ns, tun->group));
2107                 break;
2108 
2109         case TUNSETLINK:
2110                 /* Only allow setting the type when the interface is down */
2111                 if (tun->dev->flags & IFF_UP) {
2112                         tun_debug(KERN_INFO, tun,
2113                                   "Linktype set failed because interface is up\n");
2114                         ret = -EBUSY;
2115                 } else {
2116                         tun->dev->type = (int) arg;
2117                         tun_debug(KERN_INFO, tun, "linktype set to %d\n",
2118                                   tun->dev->type);
2119                         ret = 0;
2120                 }
2121                 break;
2122 
2123 #ifdef TUN_DEBUG
2124         case TUNSETDEBUG:
2125                 tun->debug = arg;
2126                 break;
2127 #endif
2128         case TUNSETOFFLOAD:
2129                 ret = set_offload(tun, arg);
2130                 break;
2131 
2132         case TUNSETTXFILTER:
2133                 /* Can be set only for TAPs */
2134                 ret = -EINVAL;
2135                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
2136                         break;
2137                 ret = update_filter(&tun->txflt, (void __user *)arg);
2138                 break;
2139 
2140         case SIOCGIFHWADDR:
2141                 /* Get hw address */
2142                 memcpy(ifr.ifr_hwaddr.sa_data, tun->dev->dev_addr, ETH_ALEN);
2143                 ifr.ifr_hwaddr.sa_family = tun->dev->type;
2144                 if (copy_to_user(argp, &ifr, ifreq_len))
2145                         ret = -EFAULT;
2146                 break;
2147 
2148         case SIOCSIFHWADDR:
2149                 /* Set hw address */
2150                 tun_debug(KERN_DEBUG, tun, "set hw address: %pM\n",
2151                           ifr.ifr_hwaddr.sa_data);
2152 
2153                 ret = dev_set_mac_address(tun->dev, &ifr.ifr_hwaddr);
2154                 break;
2155 
2156         case TUNGETSNDBUF:
2157                 sndbuf = tfile->socket.sk->sk_sndbuf;
2158                 if (copy_to_user(argp, &sndbuf, sizeof(sndbuf)))
2159                         ret = -EFAULT;
2160                 break;
2161 
2162         case TUNSETSNDBUF:
2163                 if (copy_from_user(&sndbuf, argp, sizeof(sndbuf))) {
2164                         ret = -EFAULT;
2165                         break;
2166                 }
2167 
2168                 tun->sndbuf = sndbuf;
2169                 tun_set_sndbuf(tun);
2170                 break;
2171 
2172         case TUNGETVNETHDRSZ:
2173                 vnet_hdr_sz = tun->vnet_hdr_sz;
2174                 if (copy_to_user(argp, &vnet_hdr_sz, sizeof(vnet_hdr_sz)))
2175                         ret = -EFAULT;
2176                 break;
2177 
2178         case TUNSETVNETHDRSZ:
2179                 if (copy_from_user(&vnet_hdr_sz, argp, sizeof(vnet_hdr_sz))) {
2180                         ret = -EFAULT;
2181                         break;
2182                 }
2183                 if (vnet_hdr_sz < (int)sizeof(struct virtio_net_hdr)) {
2184                         ret = -EINVAL;
2185                         break;
2186                 }
2187 
2188                 tun->vnet_hdr_sz = vnet_hdr_sz;
2189                 break;
2190 
2191         case TUNGETVNETLE:
2192                 le = !!(tun->flags & TUN_VNET_LE);
2193                 if (put_user(le, (int __user *)argp))
2194                         ret = -EFAULT;
2195                 break;
2196 
2197         case TUNSETVNETLE:
2198                 if (get_user(le, (int __user *)argp)) {
2199                         ret = -EFAULT;
2200                         break;
2201                 }
2202                 if (le)
2203                         tun->flags |= TUN_VNET_LE;
2204                 else
2205                         tun->flags &= ~TUN_VNET_LE;
2206                 break;
2207 
2208         case TUNGETVNETBE:
2209                 ret = tun_get_vnet_be(tun, argp);
2210                 break;
2211 
2212         case TUNSETVNETBE:
2213                 ret = tun_set_vnet_be(tun, argp);
2214                 break;
2215 
2216         case TUNATTACHFILTER:
2217                 /* Can be set only for TAPs */
2218                 ret = -EINVAL;
2219                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
2220                         break;
2221                 ret = -EFAULT;
2222                 if (copy_from_user(&tun->fprog, argp, sizeof(tun->fprog)))
2223                         break;
2224 
2225                 ret = tun_attach_filter(tun);
2226                 break;
2227 
2228         case TUNDETACHFILTER:
2229                 /* Can be set only for TAPs */
2230                 ret = -EINVAL;
2231                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
2232                         break;
2233                 ret = 0;
2234                 tun_detach_filter(tun, tun->numqueues);
2235                 break;
2236 
2237         case TUNGETFILTER:
2238                 ret = -EINVAL;
2239                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
2240                         break;
2241                 ret = -EFAULT;
2242                 if (copy_to_user(argp, &tun->fprog, sizeof(tun->fprog)))
2243                         break;
2244                 ret = 0;
2245                 break;
2246 
2247         default:
2248                 ret = -EINVAL;
2249                 break;
2250         }
2251 
2252 unlock:
2253         rtnl_unlock();
2254         if (tun)
2255                 tun_put(tun);
2256         return ret;
2257 }
2258 
2259 static long tun_chr_ioctl(struct file *file,
2260                           unsigned int cmd, unsigned long arg)
2261 {
2262         return __tun_chr_ioctl(file, cmd, arg, sizeof (struct ifreq));
2263 }
2264 
2265 #ifdef CONFIG_COMPAT
2266 static long tun_chr_compat_ioctl(struct file *file,
2267                          unsigned int cmd, unsigned long arg)
2268 {
2269         switch (cmd) {
2270         case TUNSETIFF:
2271         case TUNGETIFF:
2272         case TUNSETTXFILTER:
2273         case TUNGETSNDBUF:
2274         case TUNSETSNDBUF:
2275         case SIOCGIFHWADDR:
2276         case SIOCSIFHWADDR:
2277                 arg = (unsigned long)compat_ptr(arg);
2278                 break;
2279         default:
2280                 arg = (compat_ulong_t)arg;
2281                 break;
2282         }
2283 
2284         /*
2285          * compat_ifreq is shorter than ifreq, so we must not access beyond
2286          * the end of that structure. All fields that are used in this
2287          * driver are compatible though, we don't need to convert the
2288          * contents.
2289          */
2290         return __tun_chr_ioctl(file, cmd, arg, sizeof(struct compat_ifreq));
2291 }
2292 #endif /* CONFIG_COMPAT */
2293 
2294 static int tun_chr_fasync(int fd, struct file *file, int on)
2295 {
2296         struct tun_file *tfile = file->private_data;
2297         int ret;
2298 
2299         if ((ret = fasync_helper(fd, file, on, &tfile->fasync)) < 0)
2300                 goto out;
2301 
2302         if (on) {
2303                 __f_setown(file, task_pid(current), PIDTYPE_PID, 0);
2304                 tfile->flags |= TUN_FASYNC;
2305         } else
2306                 tfile->flags &= ~TUN_FASYNC;
2307         ret = 0;
2308 out:
2309         return ret;
2310 }
2311 
2312 static int tun_chr_open(struct inode *inode, struct file * file)
2313 {
2314         struct net *net = current->nsproxy->net_ns;
2315         struct tun_file *tfile;
2316 
2317         DBG1(KERN_INFO, "tunX: tun_chr_open\n");
2318 
2319         tfile = (struct tun_file *)sk_alloc(net, AF_UNSPEC, GFP_KERNEL,
2320                                             &tun_proto, 0);
2321         if (!tfile)
2322                 return -ENOMEM;
2323         RCU_INIT_POINTER(tfile->tun, NULL);
2324         tfile->flags = 0;
2325         tfile->ifindex = 0;
2326 
2327         init_waitqueue_head(&tfile->wq.wait);
2328         RCU_INIT_POINTER(tfile->socket.wq, &tfile->wq);
2329 
2330         tfile->socket.file = file;
2331         tfile->socket.ops = &tun_socket_ops;
2332 
2333         sock_init_data(&tfile->socket, &tfile->sk);
2334 
2335         tfile->sk.sk_write_space = tun_sock_write_space;
2336         tfile->sk.sk_sndbuf = INT_MAX;
2337 
2338         file->private_data = tfile;
2339         INIT_LIST_HEAD(&tfile->next);
2340 
2341         sock_set_flag(&tfile->sk, SOCK_ZEROCOPY);
2342 
2343         return 0;
2344 }
2345 
2346 static int tun_chr_close(struct inode *inode, struct file *file)
2347 {
2348         struct tun_file *tfile = file->private_data;
2349 
2350         tun_detach(tfile, true);
2351 
2352         return 0;
2353 }
2354 
2355 #ifdef CONFIG_PROC_FS
2356 static void tun_chr_show_fdinfo(struct seq_file *m, struct file *f)
2357 {
2358         struct tun_struct *tun;
2359         struct ifreq ifr;
2360 
2361         memset(&ifr, 0, sizeof(ifr));
2362 
2363         rtnl_lock();
2364         tun = tun_get(f);
2365         if (tun)
2366                 tun_get_iff(current->nsproxy->net_ns, tun, &ifr);
2367         rtnl_unlock();
2368 
2369         if (tun)
2370                 tun_put(tun);
2371 
2372         seq_printf(m, "iff:\t%s\n", ifr.ifr_name);
2373 }
2374 #endif
2375 
2376 static const struct file_operations tun_fops = {
2377         .owner  = THIS_MODULE,
2378         .llseek = no_llseek,
2379         .read_iter  = tun_chr_read_iter,
2380         .write_iter = tun_chr_write_iter,
2381         .poll   = tun_chr_poll,
2382         .unlocked_ioctl = tun_chr_ioctl,
2383 #ifdef CONFIG_COMPAT
2384         .compat_ioctl = tun_chr_compat_ioctl,
2385 #endif
2386         .open   = tun_chr_open,
2387         .release = tun_chr_close,
2388         .fasync = tun_chr_fasync,
2389 #ifdef CONFIG_PROC_FS
2390         .show_fdinfo = tun_chr_show_fdinfo,
2391 #endif
2392 };
2393 
2394 static struct miscdevice tun_miscdev = {
2395         .minor = TUN_MINOR,
2396         .name = "tun",
2397         .nodename = "net/tun",
2398         .fops = &tun_fops,
2399 };
2400 
2401 /* ethtool interface */
2402 
2403 static int tun_get_settings(struct net_device *dev, struct ethtool_cmd *cmd)
2404 {
2405         cmd->supported          = 0;
2406         cmd->advertising        = 0;
2407         ethtool_cmd_speed_set(cmd, SPEED_10);
2408         cmd->duplex             = DUPLEX_FULL;
2409         cmd->port               = PORT_TP;
2410         cmd->phy_address        = 0;
2411         cmd->transceiver        = XCVR_INTERNAL;
2412         cmd->autoneg            = AUTONEG_DISABLE;
2413         cmd->maxtxpkt           = 0;
2414         cmd->maxrxpkt           = 0;
2415         return 0;
2416 }
2417 
2418 static void tun_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
2419 {
2420         struct tun_struct *tun = netdev_priv(dev);
2421 
2422         strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
2423         strlcpy(info->version, DRV_VERSION, sizeof(info->version));
2424 
2425         switch (tun->flags & TUN_TYPE_MASK) {
2426         case IFF_TUN:
2427                 strlcpy(info->bus_info, "tun", sizeof(info->bus_info));
2428                 break;
2429         case IFF_TAP:
2430                 strlcpy(info->bus_info, "tap", sizeof(info->bus_info));
2431                 break;
2432         }
2433 }
2434 
2435 static u32 tun_get_msglevel(struct net_device *dev)
2436 {
2437 #ifdef TUN_DEBUG
2438         struct tun_struct *tun = netdev_priv(dev);
2439         return tun->debug;
2440 #else
2441         return -EOPNOTSUPP;
2442 #endif
2443 }
2444 
2445 static void tun_set_msglevel(struct net_device *dev, u32 value)
2446 {
2447 #ifdef TUN_DEBUG
2448         struct tun_struct *tun = netdev_priv(dev);
2449         tun->debug = value;
2450 #endif
2451 }
2452 
2453 static const struct ethtool_ops tun_ethtool_ops = {
2454         .get_settings   = tun_get_settings,
2455         .get_drvinfo    = tun_get_drvinfo,
2456         .get_msglevel   = tun_get_msglevel,
2457         .set_msglevel   = tun_set_msglevel,
2458         .get_link       = ethtool_op_get_link,
2459         .get_ts_info    = ethtool_op_get_ts_info,
2460 };
2461 
2462 static int tun_queue_resize(struct tun_struct *tun)
2463 {
2464         struct net_device *dev = tun->dev;
2465         struct tun_file *tfile;
2466         struct skb_array **arrays;
2467         int n = tun->numqueues + tun->numdisabled;
2468         int ret, i;
2469 
2470         arrays = kmalloc(sizeof *arrays * n, GFP_KERNEL);
2471         if (!arrays)
2472                 return -ENOMEM;
2473 
2474         for (i = 0; i < tun->numqueues; i++) {
2475                 tfile = rtnl_dereference(tun->tfiles[i]);
2476                 arrays[i] = &tfile->tx_array;
2477         }
2478         list_for_each_entry(tfile, &tun->disabled, next)
2479                 arrays[i++] = &tfile->tx_array;
2480 
2481         ret = skb_array_resize_multiple(arrays, n,
2482                                         dev->tx_queue_len, GFP_KERNEL);
2483 
2484         kfree(arrays);
2485         return ret;
2486 }
2487 
2488 static int tun_device_event(struct notifier_block *unused,
2489                             unsigned long event, void *ptr)
2490 {
2491         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
2492         struct tun_struct *tun = netdev_priv(dev);
2493 
2494         if (dev->rtnl_link_ops != &tun_link_ops)
2495                 return NOTIFY_DONE;
2496 
2497         switch (event) {
2498         case NETDEV_CHANGE_TX_QUEUE_LEN:
2499                 if (tun_queue_resize(tun))
2500                         return NOTIFY_BAD;
2501                 break;
2502         default:
2503                 break;
2504         }
2505 
2506         return NOTIFY_DONE;
2507 }
2508 
2509 static struct notifier_block tun_notifier_block __read_mostly = {
2510         .notifier_call  = tun_device_event,
2511 };
2512 
2513 static int __init tun_init(void)
2514 {
2515         int ret = 0;
2516 
2517         pr_info("%s, %s\n", DRV_DESCRIPTION, DRV_VERSION);
2518         pr_info("%s\n", DRV_COPYRIGHT);
2519 
2520         ret = rtnl_link_register(&tun_link_ops);
2521         if (ret) {
2522                 pr_err("Can't register link_ops\n");
2523                 goto err_linkops;
2524         }
2525 
2526         ret = misc_register(&tun_miscdev);
2527         if (ret) {
2528                 pr_err("Can't register misc device %d\n", TUN_MINOR);
2529                 goto err_misc;
2530         }
2531 
2532         register_netdevice_notifier(&tun_notifier_block);
2533         return  0;
2534 err_misc:
2535         rtnl_link_unregister(&tun_link_ops);
2536 err_linkops:
2537         return ret;
2538 }
2539 
2540 static void tun_cleanup(void)
2541 {
2542         misc_deregister(&tun_miscdev);
2543         rtnl_link_unregister(&tun_link_ops);
2544         unregister_netdevice_notifier(&tun_notifier_block);
2545 }
2546 
2547 /* Get an underlying socket object from tun file.  Returns error unless file is
2548  * attached to a device.  The returned object works like a packet socket, it
2549  * can be used for sock_sendmsg/sock_recvmsg.  The caller is responsible for
2550  * holding a reference to the file for as long as the socket is in use. */
2551 struct socket *tun_get_socket(struct file *file)
2552 {
2553         struct tun_file *tfile;
2554         if (file->f_op != &tun_fops)
2555                 return ERR_PTR(-EINVAL);
2556         tfile = file->private_data;
2557         if (!tfile)
2558                 return ERR_PTR(-EBADFD);
2559         return &tfile->socket;
2560 }
2561 EXPORT_SYMBOL_GPL(tun_get_socket);
2562 
2563 module_init(tun_init);
2564 module_exit(tun_cleanup);
2565 MODULE_DESCRIPTION(DRV_DESCRIPTION);
2566 MODULE_AUTHOR(DRV_COPYRIGHT);
2567 MODULE_LICENSE("GPL");
2568 MODULE_ALIAS_MISCDEV(TUN_MINOR);
2569 MODULE_ALIAS("devname:net/tun");
2570 

This page was automatically generated by LXR 0.3.1 (source).  •  Linux is a registered trademark of Linus Torvalds  •  Contact us