Version:  2.0.40 2.2.26 2.4.37 3.11 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

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 = kmalloc(alen, GFP_KERNEL);
735         if (!addr)
736                 return -ENOMEM;
737 
738         if (copy_from_user(addr, arg + sizeof(uf), alen)) {
739                 err = -EFAULT;
740                 goto done;
741         }
742 
743         /* The filter is updated without holding any locks. Which is
744          * perfectly safe. We disable it first and in the worst
745          * case we'll accept a few undesired packets. */
746         filter->count = 0;
747         wmb();
748 
749         /* Use first set of addresses as an exact filter */
750         for (n = 0; n < uf.count && n < FLT_EXACT_COUNT; n++)
751                 memcpy(filter->addr[n], addr[n].u, ETH_ALEN);
752 
753         nexact = n;
754 
755         /* Remaining multicast addresses are hashed,
756          * unicast will leave the filter disabled. */
757         memset(filter->mask, 0, sizeof(filter->mask));
758         for (; n < uf.count; n++) {
759                 if (!is_multicast_ether_addr(addr[n].u)) {
760                         err = 0; /* no filter */
761                         goto done;
762                 }
763                 addr_hash_set(filter->mask, addr[n].u);
764         }
765 
766         /* For ALLMULTI just set the mask to all ones.
767          * This overrides the mask populated above. */
768         if ((uf.flags & TUN_FLT_ALLMULTI))
769                 memset(filter->mask, ~0, sizeof(filter->mask));
770 
771         /* Now enable the filter */
772         wmb();
773         filter->count = nexact;
774 
775         /* Return the number of exact filters */
776         err = nexact;
777 
778 done:
779         kfree(addr);
780         return err;
781 }
782 
783 /* Returns: 0 - drop, !=0 - accept */
784 static int run_filter(struct tap_filter *filter, const struct sk_buff *skb)
785 {
786         /* Cannot use eth_hdr(skb) here because skb_mac_hdr() is incorrect
787          * at this point. */
788         struct ethhdr *eh = (struct ethhdr *) skb->data;
789         int i;
790 
791         /* Exact match */
792         for (i = 0; i < filter->count; i++)
793                 if (ether_addr_equal(eh->h_dest, filter->addr[i]))
794                         return 1;
795 
796         /* Inexact match (multicast only) */
797         if (is_multicast_ether_addr(eh->h_dest))
798                 return addr_hash_test(filter->mask, eh->h_dest);
799 
800         return 0;
801 }
802 
803 /*
804  * Checks whether the packet is accepted or not.
805  * Returns: 0 - drop, !=0 - accept
806  */
807 static int check_filter(struct tap_filter *filter, const struct sk_buff *skb)
808 {
809         if (!filter->count)
810                 return 1;
811 
812         return run_filter(filter, skb);
813 }
814 
815 /* Network device part of the driver */
816 
817 static const struct ethtool_ops tun_ethtool_ops;
818 
819 /* Net device detach from fd. */
820 static void tun_net_uninit(struct net_device *dev)
821 {
822         tun_detach_all(dev);
823 }
824 
825 /* Net device open. */
826 static int tun_net_open(struct net_device *dev)
827 {
828         netif_tx_start_all_queues(dev);
829         return 0;
830 }
831 
832 /* Net device close. */
833 static int tun_net_close(struct net_device *dev)
834 {
835         netif_tx_stop_all_queues(dev);
836         return 0;
837 }
838 
839 /* Net device start xmit */
840 static netdev_tx_t tun_net_xmit(struct sk_buff *skb, struct net_device *dev)
841 {
842         struct tun_struct *tun = netdev_priv(dev);
843         int txq = skb->queue_mapping;
844         struct tun_file *tfile;
845         u32 numqueues = 0;
846 
847         rcu_read_lock();
848         tfile = rcu_dereference(tun->tfiles[txq]);
849         numqueues = ACCESS_ONCE(tun->numqueues);
850 
851         /* Drop packet if interface is not attached */
852         if (txq >= numqueues)
853                 goto drop;
854 
855 #ifdef CONFIG_RPS
856         if (numqueues == 1 && static_key_false(&rps_needed)) {
857                 /* Select queue was not called for the skbuff, so we extract the
858                  * RPS hash and save it into the flow_table here.
859                  */
860                 __u32 rxhash;
861 
862                 rxhash = skb_get_hash(skb);
863                 if (rxhash) {
864                         struct tun_flow_entry *e;
865                         e = tun_flow_find(&tun->flows[tun_hashfn(rxhash)],
866                                         rxhash);
867                         if (e)
868                                 tun_flow_save_rps_rxhash(e, rxhash);
869                 }
870         }
871 #endif
872 
873         tun_debug(KERN_INFO, tun, "tun_net_xmit %d\n", skb->len);
874 
875         BUG_ON(!tfile);
876 
877         /* Drop if the filter does not like it.
878          * This is a noop if the filter is disabled.
879          * Filter can be enabled only for the TAP devices. */
880         if (!check_filter(&tun->txflt, skb))
881                 goto drop;
882 
883         if (tfile->socket.sk->sk_filter &&
884             sk_filter(tfile->socket.sk, skb))
885                 goto drop;
886 
887         /* Limit the number of packets queued by dividing txq length with the
888          * number of queues.
889          */
890         if (skb_queue_len(&tfile->socket.sk->sk_receive_queue) * numqueues
891                           >= dev->tx_queue_len)
892                 goto drop;
893 
894         if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC)))
895                 goto drop;
896 
897         skb_tx_timestamp(skb);
898 
899         /* Orphan the skb - required as we might hang on to it
900          * for indefinite time.
901          */
902         skb_orphan(skb);
903 
904         nf_reset(skb);
905 
906         if (skb_array_produce(&tfile->tx_array, skb))
907                 goto drop;
908 
909         /* Notify and wake up reader process */
910         if (tfile->flags & TUN_FASYNC)
911                 kill_fasync(&tfile->fasync, SIGIO, POLL_IN);
912         tfile->socket.sk->sk_data_ready(tfile->socket.sk);
913 
914         rcu_read_unlock();
915         return NETDEV_TX_OK;
916 
917 drop:
918         this_cpu_inc(tun->pcpu_stats->tx_dropped);
919         skb_tx_error(skb);
920         kfree_skb(skb);
921         rcu_read_unlock();
922         return NET_XMIT_DROP;
923 }
924 
925 static void tun_net_mclist(struct net_device *dev)
926 {
927         /*
928          * This callback is supposed to deal with mc filter in
929          * _rx_ path and has nothing to do with the _tx_ path.
930          * In rx path we always accept everything userspace gives us.
931          */
932 }
933 
934 #define MIN_MTU 68
935 #define MAX_MTU 65535
936 
937 static int
938 tun_net_change_mtu(struct net_device *dev, int new_mtu)
939 {
940         if (new_mtu < MIN_MTU || new_mtu + dev->hard_header_len > MAX_MTU)
941                 return -EINVAL;
942         dev->mtu = new_mtu;
943         return 0;
944 }
945 
946 static netdev_features_t tun_net_fix_features(struct net_device *dev,
947         netdev_features_t features)
948 {
949         struct tun_struct *tun = netdev_priv(dev);
950 
951         return (features & tun->set_features) | (features & ~TUN_USER_FEATURES);
952 }
953 #ifdef CONFIG_NET_POLL_CONTROLLER
954 static void tun_poll_controller(struct net_device *dev)
955 {
956         /*
957          * Tun only receives frames when:
958          * 1) the char device endpoint gets data from user space
959          * 2) the tun socket gets a sendmsg call from user space
960          * Since both of those are synchronous operations, we are guaranteed
961          * never to have pending data when we poll for it
962          * so there is nothing to do here but return.
963          * We need this though so netpoll recognizes us as an interface that
964          * supports polling, which enables bridge devices in virt setups to
965          * still use netconsole
966          */
967         return;
968 }
969 #endif
970 
971 static void tun_set_headroom(struct net_device *dev, int new_hr)
972 {
973         struct tun_struct *tun = netdev_priv(dev);
974 
975         if (new_hr < NET_SKB_PAD)
976                 new_hr = NET_SKB_PAD;
977 
978         tun->align = new_hr;
979 }
980 
981 static struct rtnl_link_stats64 *
982 tun_net_get_stats64(struct net_device *dev, struct rtnl_link_stats64 *stats)
983 {
984         u32 rx_dropped = 0, tx_dropped = 0, rx_frame_errors = 0;
985         struct tun_struct *tun = netdev_priv(dev);
986         struct tun_pcpu_stats *p;
987         int i;
988 
989         for_each_possible_cpu(i) {
990                 u64 rxpackets, rxbytes, txpackets, txbytes;
991                 unsigned int start;
992 
993                 p = per_cpu_ptr(tun->pcpu_stats, i);
994                 do {
995                         start = u64_stats_fetch_begin(&p->syncp);
996                         rxpackets       = p->rx_packets;
997                         rxbytes         = p->rx_bytes;
998                         txpackets       = p->tx_packets;
999                         txbytes         = p->tx_bytes;
1000                 } while (u64_stats_fetch_retry(&p->syncp, start));
1001 
1002                 stats->rx_packets       += rxpackets;
1003                 stats->rx_bytes         += rxbytes;
1004                 stats->tx_packets       += txpackets;
1005                 stats->tx_bytes         += txbytes;
1006 
1007                 /* u32 counters */
1008                 rx_dropped      += p->rx_dropped;
1009                 rx_frame_errors += p->rx_frame_errors;
1010                 tx_dropped      += p->tx_dropped;
1011         }
1012         stats->rx_dropped  = rx_dropped;
1013         stats->rx_frame_errors = rx_frame_errors;
1014         stats->tx_dropped = tx_dropped;
1015         return stats;
1016 }
1017 
1018 static const struct net_device_ops tun_netdev_ops = {
1019         .ndo_uninit             = tun_net_uninit,
1020         .ndo_open               = tun_net_open,
1021         .ndo_stop               = tun_net_close,
1022         .ndo_start_xmit         = tun_net_xmit,
1023         .ndo_change_mtu         = tun_net_change_mtu,
1024         .ndo_fix_features       = tun_net_fix_features,
1025         .ndo_select_queue       = tun_select_queue,
1026 #ifdef CONFIG_NET_POLL_CONTROLLER
1027         .ndo_poll_controller    = tun_poll_controller,
1028 #endif
1029         .ndo_set_rx_headroom    = tun_set_headroom,
1030         .ndo_get_stats64        = tun_net_get_stats64,
1031 };
1032 
1033 static const struct net_device_ops tap_netdev_ops = {
1034         .ndo_uninit             = tun_net_uninit,
1035         .ndo_open               = tun_net_open,
1036         .ndo_stop               = tun_net_close,
1037         .ndo_start_xmit         = tun_net_xmit,
1038         .ndo_change_mtu         = tun_net_change_mtu,
1039         .ndo_fix_features       = tun_net_fix_features,
1040         .ndo_set_rx_mode        = tun_net_mclist,
1041         .ndo_set_mac_address    = eth_mac_addr,
1042         .ndo_validate_addr      = eth_validate_addr,
1043         .ndo_select_queue       = tun_select_queue,
1044 #ifdef CONFIG_NET_POLL_CONTROLLER
1045         .ndo_poll_controller    = tun_poll_controller,
1046 #endif
1047         .ndo_features_check     = passthru_features_check,
1048         .ndo_set_rx_headroom    = tun_set_headroom,
1049         .ndo_get_stats64        = tun_net_get_stats64,
1050 };
1051 
1052 static void tun_flow_init(struct tun_struct *tun)
1053 {
1054         int i;
1055 
1056         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++)
1057                 INIT_HLIST_HEAD(&tun->flows[i]);
1058 
1059         tun->ageing_time = TUN_FLOW_EXPIRE;
1060         setup_timer(&tun->flow_gc_timer, tun_flow_cleanup, (unsigned long)tun);
1061         mod_timer(&tun->flow_gc_timer,
1062                   round_jiffies_up(jiffies + tun->ageing_time));
1063 }
1064 
1065 static void tun_flow_uninit(struct tun_struct *tun)
1066 {
1067         del_timer_sync(&tun->flow_gc_timer);
1068         tun_flow_flush(tun);
1069 }
1070 
1071 /* Initialize net device. */
1072 static void tun_net_init(struct net_device *dev)
1073 {
1074         struct tun_struct *tun = netdev_priv(dev);
1075 
1076         switch (tun->flags & TUN_TYPE_MASK) {
1077         case IFF_TUN:
1078                 dev->netdev_ops = &tun_netdev_ops;
1079 
1080                 /* Point-to-Point TUN Device */
1081                 dev->hard_header_len = 0;
1082                 dev->addr_len = 0;
1083                 dev->mtu = 1500;
1084 
1085                 /* Zero header length */
1086                 dev->type = ARPHRD_NONE;
1087                 dev->flags = IFF_POINTOPOINT | IFF_NOARP | IFF_MULTICAST;
1088                 break;
1089 
1090         case IFF_TAP:
1091                 dev->netdev_ops = &tap_netdev_ops;
1092                 /* Ethernet TAP Device */
1093                 ether_setup(dev);
1094                 dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1095                 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
1096 
1097                 eth_hw_addr_random(dev);
1098 
1099                 break;
1100         }
1101 }
1102 
1103 /* Character device part */
1104 
1105 /* Poll */
1106 static unsigned int tun_chr_poll(struct file *file, poll_table *wait)
1107 {
1108         struct tun_file *tfile = file->private_data;
1109         struct tun_struct *tun = __tun_get(tfile);
1110         struct sock *sk;
1111         unsigned int mask = 0;
1112 
1113         if (!tun)
1114                 return POLLERR;
1115 
1116         sk = tfile->socket.sk;
1117 
1118         tun_debug(KERN_INFO, tun, "tun_chr_poll\n");
1119 
1120         poll_wait(file, sk_sleep(sk), wait);
1121 
1122         if (!skb_array_empty(&tfile->tx_array))
1123                 mask |= POLLIN | POLLRDNORM;
1124 
1125         if (sock_writeable(sk) ||
1126             (!test_and_set_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags) &&
1127              sock_writeable(sk)))
1128                 mask |= POLLOUT | POLLWRNORM;
1129 
1130         if (tun->dev->reg_state != NETREG_REGISTERED)
1131                 mask = POLLERR;
1132 
1133         tun_put(tun);
1134         return mask;
1135 }
1136 
1137 /* prepad is the amount to reserve at front.  len is length after that.
1138  * linear is a hint as to how much to copy (usually headers). */
1139 static struct sk_buff *tun_alloc_skb(struct tun_file *tfile,
1140                                      size_t prepad, size_t len,
1141                                      size_t linear, int noblock)
1142 {
1143         struct sock *sk = tfile->socket.sk;
1144         struct sk_buff *skb;
1145         int err;
1146 
1147         /* Under a page?  Don't bother with paged skb. */
1148         if (prepad + len < PAGE_SIZE || !linear)
1149                 linear = len;
1150 
1151         skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
1152                                    &err, 0);
1153         if (!skb)
1154                 return ERR_PTR(err);
1155 
1156         skb_reserve(skb, prepad);
1157         skb_put(skb, linear);
1158         skb->data_len = len - linear;
1159         skb->len += len - linear;
1160 
1161         return skb;
1162 }
1163 
1164 /* Get packet from user space buffer */
1165 static ssize_t tun_get_user(struct tun_struct *tun, struct tun_file *tfile,
1166                             void *msg_control, struct iov_iter *from,
1167                             int noblock)
1168 {
1169         struct tun_pi pi = { 0, cpu_to_be16(ETH_P_IP) };
1170         struct sk_buff *skb;
1171         size_t total_len = iov_iter_count(from);
1172         size_t len = total_len, align = tun->align, linear;
1173         struct virtio_net_hdr gso = { 0 };
1174         struct tun_pcpu_stats *stats;
1175         int good_linear;
1176         int copylen;
1177         bool zerocopy = false;
1178         int err;
1179         u32 rxhash;
1180         ssize_t n;
1181 
1182         if (!(tun->dev->flags & IFF_UP))
1183                 return -EIO;
1184 
1185         if (!(tun->flags & IFF_NO_PI)) {
1186                 if (len < sizeof(pi))
1187                         return -EINVAL;
1188                 len -= sizeof(pi);
1189 
1190                 n = copy_from_iter(&pi, sizeof(pi), from);
1191                 if (n != sizeof(pi))
1192                         return -EFAULT;
1193         }
1194 
1195         if (tun->flags & IFF_VNET_HDR) {
1196                 if (len < tun->vnet_hdr_sz)
1197                         return -EINVAL;
1198                 len -= tun->vnet_hdr_sz;
1199 
1200                 n = copy_from_iter(&gso, sizeof(gso), from);
1201                 if (n != sizeof(gso))
1202                         return -EFAULT;
1203 
1204                 if ((gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
1205                     tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2 > tun16_to_cpu(tun, gso.hdr_len))
1206                         gso.hdr_len = cpu_to_tun16(tun, tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2);
1207 
1208                 if (tun16_to_cpu(tun, gso.hdr_len) > len)
1209                         return -EINVAL;
1210                 iov_iter_advance(from, tun->vnet_hdr_sz - sizeof(gso));
1211         }
1212 
1213         if ((tun->flags & TUN_TYPE_MASK) == IFF_TAP) {
1214                 align += NET_IP_ALIGN;
1215                 if (unlikely(len < ETH_HLEN ||
1216                              (gso.hdr_len && tun16_to_cpu(tun, gso.hdr_len) < ETH_HLEN)))
1217                         return -EINVAL;
1218         }
1219 
1220         good_linear = SKB_MAX_HEAD(align);
1221 
1222         if (msg_control) {
1223                 struct iov_iter i = *from;
1224 
1225                 /* There are 256 bytes to be copied in skb, so there is
1226                  * enough room for skb expand head in case it is used.
1227                  * The rest of the buffer is mapped from userspace.
1228                  */
1229                 copylen = gso.hdr_len ? tun16_to_cpu(tun, gso.hdr_len) : GOODCOPY_LEN;
1230                 if (copylen > good_linear)
1231                         copylen = good_linear;
1232                 linear = copylen;
1233                 iov_iter_advance(&i, copylen);
1234                 if (iov_iter_npages(&i, INT_MAX) <= MAX_SKB_FRAGS)
1235                         zerocopy = true;
1236         }
1237 
1238         if (!zerocopy) {
1239                 copylen = len;
1240                 if (tun16_to_cpu(tun, gso.hdr_len) > good_linear)
1241                         linear = good_linear;
1242                 else
1243                         linear = tun16_to_cpu(tun, gso.hdr_len);
1244         }
1245 
1246         skb = tun_alloc_skb(tfile, align, copylen, linear, noblock);
1247         if (IS_ERR(skb)) {
1248                 if (PTR_ERR(skb) != -EAGAIN)
1249                         this_cpu_inc(tun->pcpu_stats->rx_dropped);
1250                 return PTR_ERR(skb);
1251         }
1252 
1253         if (zerocopy)
1254                 err = zerocopy_sg_from_iter(skb, from);
1255         else {
1256                 err = skb_copy_datagram_from_iter(skb, 0, from, len);
1257                 if (!err && msg_control) {
1258                         struct ubuf_info *uarg = msg_control;
1259                         uarg->callback(uarg, false);
1260                 }
1261         }
1262 
1263         if (err) {
1264                 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1265                 kfree_skb(skb);
1266                 return -EFAULT;
1267         }
1268 
1269         err = virtio_net_hdr_to_skb(skb, &gso, tun_is_little_endian(tun));
1270         if (err) {
1271                 this_cpu_inc(tun->pcpu_stats->rx_frame_errors);
1272                 kfree_skb(skb);
1273                 return -EINVAL;
1274         }
1275 
1276         switch (tun->flags & TUN_TYPE_MASK) {
1277         case IFF_TUN:
1278                 if (tun->flags & IFF_NO_PI) {
1279                         switch (skb->data[0] & 0xf0) {
1280                         case 0x40:
1281                                 pi.proto = htons(ETH_P_IP);
1282                                 break;
1283                         case 0x60:
1284                                 pi.proto = htons(ETH_P_IPV6);
1285                                 break;
1286                         default:
1287                                 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1288                                 kfree_skb(skb);
1289                                 return -EINVAL;
1290                         }
1291                 }
1292 
1293                 skb_reset_mac_header(skb);
1294                 skb->protocol = pi.proto;
1295                 skb->dev = tun->dev;
1296                 break;
1297         case IFF_TAP:
1298                 skb->protocol = eth_type_trans(skb, tun->dev);
1299                 break;
1300         }
1301 
1302         /* copy skb_ubuf_info for callback when skb has no error */
1303         if (zerocopy) {
1304                 skb_shinfo(skb)->destructor_arg = msg_control;
1305                 skb_shinfo(skb)->tx_flags |= SKBTX_DEV_ZEROCOPY;
1306                 skb_shinfo(skb)->tx_flags |= SKBTX_SHARED_FRAG;
1307         }
1308 
1309         skb_reset_network_header(skb);
1310         skb_probe_transport_header(skb, 0);
1311 
1312         rxhash = skb_get_hash(skb);
1313         netif_rx_ni(skb);
1314 
1315         stats = get_cpu_ptr(tun->pcpu_stats);
1316         u64_stats_update_begin(&stats->syncp);
1317         stats->rx_packets++;
1318         stats->rx_bytes += len;
1319         u64_stats_update_end(&stats->syncp);
1320         put_cpu_ptr(stats);
1321 
1322         tun_flow_update(tun, rxhash, tfile);
1323         return total_len;
1324 }
1325 
1326 static ssize_t tun_chr_write_iter(struct kiocb *iocb, struct iov_iter *from)
1327 {
1328         struct file *file = iocb->ki_filp;
1329         struct tun_struct *tun = tun_get(file);
1330         struct tun_file *tfile = file->private_data;
1331         ssize_t result;
1332 
1333         if (!tun)
1334                 return -EBADFD;
1335 
1336         result = tun_get_user(tun, tfile, NULL, from, file->f_flags & O_NONBLOCK);
1337 
1338         tun_put(tun);
1339         return result;
1340 }
1341 
1342 /* Put packet to the user space buffer */
1343 static ssize_t tun_put_user(struct tun_struct *tun,
1344                             struct tun_file *tfile,
1345                             struct sk_buff *skb,
1346                             struct iov_iter *iter)
1347 {
1348         struct tun_pi pi = { 0, skb->protocol };
1349         struct tun_pcpu_stats *stats;
1350         ssize_t total;
1351         int vlan_offset = 0;
1352         int vlan_hlen = 0;
1353         int vnet_hdr_sz = 0;
1354 
1355         if (skb_vlan_tag_present(skb))
1356                 vlan_hlen = VLAN_HLEN;
1357 
1358         if (tun->flags & IFF_VNET_HDR)
1359                 vnet_hdr_sz = tun->vnet_hdr_sz;
1360 
1361         total = skb->len + vlan_hlen + vnet_hdr_sz;
1362 
1363         if (!(tun->flags & IFF_NO_PI)) {
1364                 if (iov_iter_count(iter) < sizeof(pi))
1365                         return -EINVAL;
1366 
1367                 total += sizeof(pi);
1368                 if (iov_iter_count(iter) < total) {
1369                         /* Packet will be striped */
1370                         pi.flags |= TUN_PKT_STRIP;
1371                 }
1372 
1373                 if (copy_to_iter(&pi, sizeof(pi), iter) != sizeof(pi))
1374                         return -EFAULT;
1375         }
1376 
1377         if (vnet_hdr_sz) {
1378                 struct virtio_net_hdr gso = { 0 }; /* no info leak */
1379                 int ret;
1380 
1381                 if (iov_iter_count(iter) < vnet_hdr_sz)
1382                         return -EINVAL;
1383 
1384                 ret = virtio_net_hdr_from_skb(skb, &gso,
1385                                               tun_is_little_endian(tun));
1386                 if (ret) {
1387                         struct skb_shared_info *sinfo = skb_shinfo(skb);
1388                         pr_err("unexpected GSO type: "
1389                                "0x%x, gso_size %d, hdr_len %d\n",
1390                                sinfo->gso_type, tun16_to_cpu(tun, gso.gso_size),
1391                                tun16_to_cpu(tun, gso.hdr_len));
1392                         print_hex_dump(KERN_ERR, "tun: ",
1393                                        DUMP_PREFIX_NONE,
1394                                        16, 1, skb->head,
1395                                        min((int)tun16_to_cpu(tun, gso.hdr_len), 64), true);
1396                         WARN_ON_ONCE(1);
1397                         return -EINVAL;
1398                 }
1399 
1400                 if (copy_to_iter(&gso, sizeof(gso), iter) != sizeof(gso))
1401                         return -EFAULT;
1402 
1403                 iov_iter_advance(iter, vnet_hdr_sz - sizeof(gso));
1404         }
1405 
1406         if (vlan_hlen) {
1407                 int ret;
1408                 struct {
1409                         __be16 h_vlan_proto;
1410                         __be16 h_vlan_TCI;
1411                 } veth;
1412 
1413                 veth.h_vlan_proto = skb->vlan_proto;
1414                 veth.h_vlan_TCI = htons(skb_vlan_tag_get(skb));
1415 
1416                 vlan_offset = offsetof(struct vlan_ethhdr, h_vlan_proto);
1417 
1418                 ret = skb_copy_datagram_iter(skb, 0, iter, vlan_offset);
1419                 if (ret || !iov_iter_count(iter))
1420                         goto done;
1421 
1422                 ret = copy_to_iter(&veth, sizeof(veth), iter);
1423                 if (ret != sizeof(veth) || !iov_iter_count(iter))
1424                         goto done;
1425         }
1426 
1427         skb_copy_datagram_iter(skb, vlan_offset, iter, skb->len - vlan_offset);
1428 
1429 done:
1430         /* caller is in process context, */
1431         stats = get_cpu_ptr(tun->pcpu_stats);
1432         u64_stats_update_begin(&stats->syncp);
1433         stats->tx_packets++;
1434         stats->tx_bytes += skb->len + vlan_hlen;
1435         u64_stats_update_end(&stats->syncp);
1436         put_cpu_ptr(tun->pcpu_stats);
1437 
1438         return total;
1439 }
1440 
1441 static struct sk_buff *tun_ring_recv(struct tun_file *tfile, int noblock,
1442                                      int *err)
1443 {
1444         DECLARE_WAITQUEUE(wait, current);
1445         struct sk_buff *skb = NULL;
1446         int error = 0;
1447 
1448         skb = skb_array_consume(&tfile->tx_array);
1449         if (skb)
1450                 goto out;
1451         if (noblock) {
1452                 error = -EAGAIN;
1453                 goto out;
1454         }
1455 
1456         add_wait_queue(&tfile->wq.wait, &wait);
1457         current->state = TASK_INTERRUPTIBLE;
1458 
1459         while (1) {
1460                 skb = skb_array_consume(&tfile->tx_array);
1461                 if (skb)
1462                         break;
1463                 if (signal_pending(current)) {
1464                         error = -ERESTARTSYS;
1465                         break;
1466                 }
1467                 if (tfile->socket.sk->sk_shutdown & RCV_SHUTDOWN) {
1468                         error = -EFAULT;
1469                         break;
1470                 }
1471 
1472                 schedule();
1473         }
1474 
1475         current->state = TASK_RUNNING;
1476         remove_wait_queue(&tfile->wq.wait, &wait);
1477 
1478 out:
1479         *err = error;
1480         return skb;
1481 }
1482 
1483 static ssize_t tun_do_read(struct tun_struct *tun, struct tun_file *tfile,
1484                            struct iov_iter *to,
1485                            int noblock)
1486 {
1487         struct sk_buff *skb;
1488         ssize_t ret;
1489         int err;
1490 
1491         tun_debug(KERN_INFO, tun, "tun_do_read\n");
1492 
1493         if (!iov_iter_count(to))
1494                 return 0;
1495 
1496         /* Read frames from ring */
1497         skb = tun_ring_recv(tfile, noblock, &err);
1498         if (!skb)
1499                 return err;
1500 
1501         ret = tun_put_user(tun, tfile, skb, to);
1502         if (unlikely(ret < 0))
1503                 kfree_skb(skb);
1504         else
1505                 consume_skb(skb);
1506 
1507         return ret;
1508 }
1509 
1510 static ssize_t tun_chr_read_iter(struct kiocb *iocb, struct iov_iter *to)
1511 {
1512         struct file *file = iocb->ki_filp;
1513         struct tun_file *tfile = file->private_data;
1514         struct tun_struct *tun = __tun_get(tfile);
1515         ssize_t len = iov_iter_count(to), ret;
1516 
1517         if (!tun)
1518                 return -EBADFD;
1519         ret = tun_do_read(tun, tfile, to, file->f_flags & O_NONBLOCK);
1520         ret = min_t(ssize_t, ret, len);
1521         if (ret > 0)
1522                 iocb->ki_pos = ret;
1523         tun_put(tun);
1524         return ret;
1525 }
1526 
1527 static void tun_free_netdev(struct net_device *dev)
1528 {
1529         struct tun_struct *tun = netdev_priv(dev);
1530 
1531         BUG_ON(!(list_empty(&tun->disabled)));
1532         free_percpu(tun->pcpu_stats);
1533         tun_flow_uninit(tun);
1534         security_tun_dev_free_security(tun->security);
1535         free_netdev(dev);
1536 }
1537 
1538 static void tun_setup(struct net_device *dev)
1539 {
1540         struct tun_struct *tun = netdev_priv(dev);
1541 
1542         tun->owner = INVALID_UID;
1543         tun->group = INVALID_GID;
1544 
1545         dev->ethtool_ops = &tun_ethtool_ops;
1546         dev->destructor = tun_free_netdev;
1547         /* We prefer our own queue length */
1548         dev->tx_queue_len = TUN_READQ_SIZE;
1549 }
1550 
1551 /* Trivial set of netlink ops to allow deleting tun or tap
1552  * device with netlink.
1553  */
1554 static int tun_validate(struct nlattr *tb[], struct nlattr *data[])
1555 {
1556         return -EINVAL;
1557 }
1558 
1559 static struct rtnl_link_ops tun_link_ops __read_mostly = {
1560         .kind           = DRV_NAME,
1561         .priv_size      = sizeof(struct tun_struct),
1562         .setup          = tun_setup,
1563         .validate       = tun_validate,
1564 };
1565 
1566 static void tun_sock_write_space(struct sock *sk)
1567 {
1568         struct tun_file *tfile;
1569         wait_queue_head_t *wqueue;
1570 
1571         if (!sock_writeable(sk))
1572                 return;
1573 
1574         if (!test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags))
1575                 return;
1576 
1577         wqueue = sk_sleep(sk);
1578         if (wqueue && waitqueue_active(wqueue))
1579                 wake_up_interruptible_sync_poll(wqueue, POLLOUT |
1580                                                 POLLWRNORM | POLLWRBAND);
1581 
1582         tfile = container_of(sk, struct tun_file, sk);
1583         kill_fasync(&tfile->fasync, SIGIO, POLL_OUT);
1584 }
1585 
1586 static int tun_sendmsg(struct socket *sock, struct msghdr *m, size_t total_len)
1587 {
1588         int ret;
1589         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
1590         struct tun_struct *tun = __tun_get(tfile);
1591 
1592         if (!tun)
1593                 return -EBADFD;
1594 
1595         ret = tun_get_user(tun, tfile, m->msg_control, &m->msg_iter,
1596                            m->msg_flags & MSG_DONTWAIT);
1597         tun_put(tun);
1598         return ret;
1599 }
1600 
1601 static int tun_recvmsg(struct socket *sock, struct msghdr *m, size_t total_len,
1602                        int flags)
1603 {
1604         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
1605         struct tun_struct *tun = __tun_get(tfile);
1606         int ret;
1607 
1608         if (!tun)
1609                 return -EBADFD;
1610 
1611         if (flags & ~(MSG_DONTWAIT|MSG_TRUNC|MSG_ERRQUEUE)) {
1612                 ret = -EINVAL;
1613                 goto out;
1614         }
1615         if (flags & MSG_ERRQUEUE) {
1616                 ret = sock_recv_errqueue(sock->sk, m, total_len,
1617                                          SOL_PACKET, TUN_TX_TIMESTAMP);
1618                 goto out;
1619         }
1620         ret = tun_do_read(tun, tfile, &m->msg_iter, flags & MSG_DONTWAIT);
1621         if (ret > (ssize_t)total_len) {
1622                 m->msg_flags |= MSG_TRUNC;
1623                 ret = flags & MSG_TRUNC ? ret : total_len;
1624         }
1625 out:
1626         tun_put(tun);
1627         return ret;
1628 }
1629 
1630 static int tun_peek_len(struct socket *sock)
1631 {
1632         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
1633         struct tun_struct *tun;
1634         int ret = 0;
1635 
1636         tun = __tun_get(tfile);
1637         if (!tun)
1638                 return 0;
1639 
1640         ret = skb_array_peek_len(&tfile->tx_array);
1641         tun_put(tun);
1642 
1643         return ret;
1644 }
1645 
1646 /* Ops structure to mimic raw sockets with tun */
1647 static const struct proto_ops tun_socket_ops = {
1648         .peek_len = tun_peek_len,
1649         .sendmsg = tun_sendmsg,
1650         .recvmsg = tun_recvmsg,
1651 };
1652 
1653 static struct proto tun_proto = {
1654         .name           = "tun",
1655         .owner          = THIS_MODULE,
1656         .obj_size       = sizeof(struct tun_file),
1657 };
1658 
1659 static int tun_flags(struct tun_struct *tun)
1660 {
1661         return tun->flags & (TUN_FEATURES | IFF_PERSIST | IFF_TUN | IFF_TAP);
1662 }
1663 
1664 static ssize_t tun_show_flags(struct device *dev, struct device_attribute *attr,
1665                               char *buf)
1666 {
1667         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1668         return sprintf(buf, "0x%x\n", tun_flags(tun));
1669 }
1670 
1671 static ssize_t tun_show_owner(struct device *dev, struct device_attribute *attr,
1672                               char *buf)
1673 {
1674         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1675         return uid_valid(tun->owner)?
1676                 sprintf(buf, "%u\n",
1677                         from_kuid_munged(current_user_ns(), tun->owner)):
1678                 sprintf(buf, "-1\n");
1679 }
1680 
1681 static ssize_t tun_show_group(struct device *dev, struct device_attribute *attr,
1682                               char *buf)
1683 {
1684         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1685         return gid_valid(tun->group) ?
1686                 sprintf(buf, "%u\n",
1687                         from_kgid_munged(current_user_ns(), tun->group)):
1688                 sprintf(buf, "-1\n");
1689 }
1690 
1691 static DEVICE_ATTR(tun_flags, 0444, tun_show_flags, NULL);
1692 static DEVICE_ATTR(owner, 0444, tun_show_owner, NULL);
1693 static DEVICE_ATTR(group, 0444, tun_show_group, NULL);
1694 
1695 static struct attribute *tun_dev_attrs[] = {
1696         &dev_attr_tun_flags.attr,
1697         &dev_attr_owner.attr,
1698         &dev_attr_group.attr,
1699         NULL
1700 };
1701 
1702 static const struct attribute_group tun_attr_group = {
1703         .attrs = tun_dev_attrs
1704 };
1705 
1706 static int tun_set_iff(struct net *net, struct file *file, struct ifreq *ifr)
1707 {
1708         struct tun_struct *tun;
1709         struct tun_file *tfile = file->private_data;
1710         struct net_device *dev;
1711         int err;
1712 
1713         if (tfile->detached)
1714                 return -EINVAL;
1715 
1716         dev = __dev_get_by_name(net, ifr->ifr_name);
1717         if (dev) {
1718                 if (ifr->ifr_flags & IFF_TUN_EXCL)
1719                         return -EBUSY;
1720                 if ((ifr->ifr_flags & IFF_TUN) && dev->netdev_ops == &tun_netdev_ops)
1721                         tun = netdev_priv(dev);
1722                 else if ((ifr->ifr_flags & IFF_TAP) && dev->netdev_ops == &tap_netdev_ops)
1723                         tun = netdev_priv(dev);
1724                 else
1725                         return -EINVAL;
1726 
1727                 if (!!(ifr->ifr_flags & IFF_MULTI_QUEUE) !=
1728                     !!(tun->flags & IFF_MULTI_QUEUE))
1729                         return -EINVAL;
1730 
1731                 if (tun_not_capable(tun))
1732                         return -EPERM;
1733                 err = security_tun_dev_open(tun->security);
1734                 if (err < 0)
1735                         return err;
1736 
1737                 err = tun_attach(tun, file, ifr->ifr_flags & IFF_NOFILTER);
1738                 if (err < 0)
1739                         return err;
1740 
1741                 if (tun->flags & IFF_MULTI_QUEUE &&
1742                     (tun->numqueues + tun->numdisabled > 1)) {
1743                         /* One or more queue has already been attached, no need
1744                          * to initialize the device again.
1745                          */
1746                         return 0;
1747                 }
1748         }
1749         else {
1750                 char *name;
1751                 unsigned long flags = 0;
1752                 int queues = ifr->ifr_flags & IFF_MULTI_QUEUE ?
1753                              MAX_TAP_QUEUES : 1;
1754 
1755                 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
1756                         return -EPERM;
1757                 err = security_tun_dev_create();
1758                 if (err < 0)
1759                         return err;
1760 
1761                 /* Set dev type */
1762                 if (ifr->ifr_flags & IFF_TUN) {
1763                         /* TUN device */
1764                         flags |= IFF_TUN;
1765                         name = "tun%d";
1766                 } else if (ifr->ifr_flags & IFF_TAP) {
1767                         /* TAP device */
1768                         flags |= IFF_TAP;
1769                         name = "tap%d";
1770                 } else
1771                         return -EINVAL;
1772 
1773                 if (*ifr->ifr_name)
1774                         name = ifr->ifr_name;
1775 
1776                 dev = alloc_netdev_mqs(sizeof(struct tun_struct), name,
1777                                        NET_NAME_UNKNOWN, tun_setup, queues,
1778                                        queues);
1779 
1780                 if (!dev)
1781                         return -ENOMEM;
1782 
1783                 dev_net_set(dev, net);
1784                 dev->rtnl_link_ops = &tun_link_ops;
1785                 dev->ifindex = tfile->ifindex;
1786                 dev->sysfs_groups[0] = &tun_attr_group;
1787 
1788                 tun = netdev_priv(dev);
1789                 tun->dev = dev;
1790                 tun->flags = flags;
1791                 tun->txflt.count = 0;
1792                 tun->vnet_hdr_sz = sizeof(struct virtio_net_hdr);
1793 
1794                 tun->align = NET_SKB_PAD;
1795                 tun->filter_attached = false;
1796                 tun->sndbuf = tfile->socket.sk->sk_sndbuf;
1797 
1798                 tun->pcpu_stats = netdev_alloc_pcpu_stats(struct tun_pcpu_stats);
1799                 if (!tun->pcpu_stats) {
1800                         err = -ENOMEM;
1801                         goto err_free_dev;
1802                 }
1803 
1804                 spin_lock_init(&tun->lock);
1805 
1806                 err = security_tun_dev_alloc_security(&tun->security);
1807                 if (err < 0)
1808                         goto err_free_stat;
1809 
1810                 tun_net_init(dev);
1811                 tun_flow_init(tun);
1812 
1813                 dev->hw_features = NETIF_F_SG | NETIF_F_FRAGLIST |
1814                                    TUN_USER_FEATURES | NETIF_F_HW_VLAN_CTAG_TX |
1815                                    NETIF_F_HW_VLAN_STAG_TX;
1816                 dev->features = dev->hw_features | NETIF_F_LLTX;
1817                 dev->vlan_features = dev->features &
1818                                      ~(NETIF_F_HW_VLAN_CTAG_TX |
1819                                        NETIF_F_HW_VLAN_STAG_TX);
1820 
1821                 INIT_LIST_HEAD(&tun->disabled);
1822                 err = tun_attach(tun, file, false);
1823                 if (err < 0)
1824                         goto err_free_flow;
1825 
1826                 err = register_netdevice(tun->dev);
1827                 if (err < 0)
1828                         goto err_detach;
1829         }
1830 
1831         netif_carrier_on(tun->dev);
1832 
1833         tun_debug(KERN_INFO, tun, "tun_set_iff\n");
1834 
1835         tun->flags = (tun->flags & ~TUN_FEATURES) |
1836                 (ifr->ifr_flags & TUN_FEATURES);
1837 
1838         /* Make sure persistent devices do not get stuck in
1839          * xoff state.
1840          */
1841         if (netif_running(tun->dev))
1842                 netif_tx_wake_all_queues(tun->dev);
1843 
1844         strcpy(ifr->ifr_name, tun->dev->name);
1845         return 0;
1846 
1847 err_detach:
1848         tun_detach_all(dev);
1849 err_free_flow:
1850         tun_flow_uninit(tun);
1851         security_tun_dev_free_security(tun->security);
1852 err_free_stat:
1853         free_percpu(tun->pcpu_stats);
1854 err_free_dev:
1855         free_netdev(dev);
1856         return err;
1857 }
1858 
1859 static void tun_get_iff(struct net *net, struct tun_struct *tun,
1860                        struct ifreq *ifr)
1861 {
1862         tun_debug(KERN_INFO, tun, "tun_get_iff\n");
1863 
1864         strcpy(ifr->ifr_name, tun->dev->name);
1865 
1866         ifr->ifr_flags = tun_flags(tun);
1867 
1868 }
1869 
1870 /* This is like a cut-down ethtool ops, except done via tun fd so no
1871  * privs required. */
1872 static int set_offload(struct tun_struct *tun, unsigned long arg)
1873 {
1874         netdev_features_t features = 0;
1875 
1876         if (arg & TUN_F_CSUM) {
1877                 features |= NETIF_F_HW_CSUM;
1878                 arg &= ~TUN_F_CSUM;
1879 
1880                 if (arg & (TUN_F_TSO4|TUN_F_TSO6)) {
1881                         if (arg & TUN_F_TSO_ECN) {
1882                                 features |= NETIF_F_TSO_ECN;
1883                                 arg &= ~TUN_F_TSO_ECN;
1884                         }
1885                         if (arg & TUN_F_TSO4)
1886                                 features |= NETIF_F_TSO;
1887                         if (arg & TUN_F_TSO6)
1888                                 features |= NETIF_F_TSO6;
1889                         arg &= ~(TUN_F_TSO4|TUN_F_TSO6);
1890                 }
1891 
1892                 if (arg & TUN_F_UFO) {
1893                         features |= NETIF_F_UFO;
1894                         arg &= ~TUN_F_UFO;
1895                 }
1896         }
1897 
1898         /* This gives the user a way to test for new features in future by
1899          * trying to set them. */
1900         if (arg)
1901                 return -EINVAL;
1902 
1903         tun->set_features = features;
1904         netdev_update_features(tun->dev);
1905 
1906         return 0;
1907 }
1908 
1909 static void tun_detach_filter(struct tun_struct *tun, int n)
1910 {
1911         int i;
1912         struct tun_file *tfile;
1913 
1914         for (i = 0; i < n; i++) {
1915                 tfile = rtnl_dereference(tun->tfiles[i]);
1916                 lock_sock(tfile->socket.sk);
1917                 sk_detach_filter(tfile->socket.sk);
1918                 release_sock(tfile->socket.sk);
1919         }
1920 
1921         tun->filter_attached = false;
1922 }
1923 
1924 static int tun_attach_filter(struct tun_struct *tun)
1925 {
1926         int i, ret = 0;
1927         struct tun_file *tfile;
1928 
1929         for (i = 0; i < tun->numqueues; i++) {
1930                 tfile = rtnl_dereference(tun->tfiles[i]);
1931                 lock_sock(tfile->socket.sk);
1932                 ret = sk_attach_filter(&tun->fprog, tfile->socket.sk);
1933                 release_sock(tfile->socket.sk);
1934                 if (ret) {
1935                         tun_detach_filter(tun, i);
1936                         return ret;
1937                 }
1938         }
1939 
1940         tun->filter_attached = true;
1941         return ret;
1942 }
1943 
1944 static void tun_set_sndbuf(struct tun_struct *tun)
1945 {
1946         struct tun_file *tfile;
1947         int i;
1948 
1949         for (i = 0; i < tun->numqueues; i++) {
1950                 tfile = rtnl_dereference(tun->tfiles[i]);
1951                 tfile->socket.sk->sk_sndbuf = tun->sndbuf;
1952         }
1953 }
1954 
1955 static int tun_set_queue(struct file *file, struct ifreq *ifr)
1956 {
1957         struct tun_file *tfile = file->private_data;
1958         struct tun_struct *tun;
1959         int ret = 0;
1960 
1961         rtnl_lock();
1962 
1963         if (ifr->ifr_flags & IFF_ATTACH_QUEUE) {
1964                 tun = tfile->detached;
1965                 if (!tun) {
1966                         ret = -EINVAL;
1967                         goto unlock;
1968                 }
1969                 ret = security_tun_dev_attach_queue(tun->security);
1970                 if (ret < 0)
1971                         goto unlock;
1972                 ret = tun_attach(tun, file, false);
1973         } else if (ifr->ifr_flags & IFF_DETACH_QUEUE) {
1974                 tun = rtnl_dereference(tfile->tun);
1975                 if (!tun || !(tun->flags & IFF_MULTI_QUEUE) || tfile->detached)
1976                         ret = -EINVAL;
1977                 else
1978                         __tun_detach(tfile, false);
1979         } else
1980                 ret = -EINVAL;
1981 
1982 unlock:
1983         rtnl_unlock();
1984         return ret;
1985 }
1986 
1987 static long __tun_chr_ioctl(struct file *file, unsigned int cmd,
1988                             unsigned long arg, int ifreq_len)
1989 {
1990         struct tun_file *tfile = file->private_data;
1991         struct tun_struct *tun;
1992         void __user* argp = (void __user*)arg;
1993         struct ifreq ifr;
1994         kuid_t owner;
1995         kgid_t group;
1996         int sndbuf;
1997         int vnet_hdr_sz;
1998         unsigned int ifindex;
1999         int le;
2000         int ret;
2001 
2002         if (cmd == TUNSETIFF || cmd == TUNSETQUEUE || _IOC_TYPE(cmd) == 0x89) {
2003                 if (copy_from_user(&ifr, argp, ifreq_len))
2004                         return -EFAULT;
2005         } else {
2006                 memset(&ifr, 0, sizeof(ifr));
2007         }
2008         if (cmd == TUNGETFEATURES) {
2009                 /* Currently this just means: "what IFF flags are valid?".
2010                  * This is needed because we never checked for invalid flags on
2011                  * TUNSETIFF.
2012                  */
2013                 return put_user(IFF_TUN | IFF_TAP | TUN_FEATURES,
2014                                 (unsigned int __user*)argp);
2015         } else if (cmd == TUNSETQUEUE)
2016                 return tun_set_queue(file, &ifr);
2017 
2018         ret = 0;
2019         rtnl_lock();
2020 
2021         tun = __tun_get(tfile);
2022         if (cmd == TUNSETIFF && !tun) {
2023                 ifr.ifr_name[IFNAMSIZ-1] = '\0';
2024 
2025                 ret = tun_set_iff(sock_net(&tfile->sk), file, &ifr);
2026 
2027                 if (ret)
2028                         goto unlock;
2029 
2030                 if (copy_to_user(argp, &ifr, ifreq_len))
2031                         ret = -EFAULT;
2032                 goto unlock;
2033         }
2034         if (cmd == TUNSETIFINDEX) {
2035                 ret = -EPERM;
2036                 if (tun)
2037                         goto unlock;
2038 
2039                 ret = -EFAULT;
2040                 if (copy_from_user(&ifindex, argp, sizeof(ifindex)))
2041                         goto unlock;
2042 
2043                 ret = 0;
2044                 tfile->ifindex = ifindex;
2045                 goto unlock;
2046         }
2047 
2048         ret = -EBADFD;
2049         if (!tun)
2050                 goto unlock;
2051 
2052         tun_debug(KERN_INFO, tun, "tun_chr_ioctl cmd %u\n", cmd);
2053 
2054         ret = 0;
2055         switch (cmd) {
2056         case TUNGETIFF:
2057                 tun_get_iff(current->nsproxy->net_ns, tun, &ifr);
2058 
2059                 if (tfile->detached)
2060                         ifr.ifr_flags |= IFF_DETACH_QUEUE;
2061                 if (!tfile->socket.sk->sk_filter)
2062                         ifr.ifr_flags |= IFF_NOFILTER;
2063 
2064                 if (copy_to_user(argp, &ifr, ifreq_len))
2065                         ret = -EFAULT;
2066                 break;
2067 
2068         case TUNSETNOCSUM:
2069                 /* Disable/Enable checksum */
2070 
2071                 /* [unimplemented] */
2072                 tun_debug(KERN_INFO, tun, "ignored: set checksum %s\n",
2073                           arg ? "disabled" : "enabled");
2074                 break;
2075 
2076         case TUNSETPERSIST:
2077                 /* Disable/Enable persist mode. Keep an extra reference to the
2078                  * module to prevent the module being unprobed.
2079                  */
2080                 if (arg && !(tun->flags & IFF_PERSIST)) {
2081                         tun->flags |= IFF_PERSIST;
2082                         __module_get(THIS_MODULE);
2083                 }
2084                 if (!arg && (tun->flags & IFF_PERSIST)) {
2085                         tun->flags &= ~IFF_PERSIST;
2086                         module_put(THIS_MODULE);
2087                 }
2088 
2089                 tun_debug(KERN_INFO, tun, "persist %s\n",
2090                           arg ? "enabled" : "disabled");
2091                 break;
2092 
2093         case TUNSETOWNER:
2094                 /* Set owner of the device */
2095                 owner = make_kuid(current_user_ns(), arg);
2096                 if (!uid_valid(owner)) {
2097                         ret = -EINVAL;
2098                         break;
2099                 }
2100                 tun->owner = owner;
2101                 tun_debug(KERN_INFO, tun, "owner set to %u\n",
2102                           from_kuid(&init_user_ns, tun->owner));
2103                 break;
2104 
2105         case TUNSETGROUP:
2106                 /* Set group of the device */
2107                 group = make_kgid(current_user_ns(), arg);
2108                 if (!gid_valid(group)) {
2109                         ret = -EINVAL;
2110                         break;
2111                 }
2112                 tun->group = group;
2113                 tun_debug(KERN_INFO, tun, "group set to %u\n",
2114                           from_kgid(&init_user_ns, tun->group));
2115                 break;
2116 
2117         case TUNSETLINK:
2118                 /* Only allow setting the type when the interface is down */
2119                 if (tun->dev->flags & IFF_UP) {
2120                         tun_debug(KERN_INFO, tun,
2121                                   "Linktype set failed because interface is up\n");
2122                         ret = -EBUSY;
2123                 } else {
2124                         tun->dev->type = (int) arg;
2125                         tun_debug(KERN_INFO, tun, "linktype set to %d\n",
2126                                   tun->dev->type);
2127                         ret = 0;
2128                 }
2129                 break;
2130 
2131 #ifdef TUN_DEBUG
2132         case TUNSETDEBUG:
2133                 tun->debug = arg;
2134                 break;
2135 #endif
2136         case TUNSETOFFLOAD:
2137                 ret = set_offload(tun, arg);
2138                 break;
2139 
2140         case TUNSETTXFILTER:
2141                 /* Can be set only for TAPs */
2142                 ret = -EINVAL;
2143                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
2144                         break;
2145                 ret = update_filter(&tun->txflt, (void __user *)arg);
2146                 break;
2147 
2148         case SIOCGIFHWADDR:
2149                 /* Get hw address */
2150                 memcpy(ifr.ifr_hwaddr.sa_data, tun->dev->dev_addr, ETH_ALEN);
2151                 ifr.ifr_hwaddr.sa_family = tun->dev->type;
2152                 if (copy_to_user(argp, &ifr, ifreq_len))
2153                         ret = -EFAULT;
2154                 break;
2155 
2156         case SIOCSIFHWADDR:
2157                 /* Set hw address */
2158                 tun_debug(KERN_DEBUG, tun, "set hw address: %pM\n",
2159                           ifr.ifr_hwaddr.sa_data);
2160 
2161                 ret = dev_set_mac_address(tun->dev, &ifr.ifr_hwaddr);
2162                 break;
2163 
2164         case TUNGETSNDBUF:
2165                 sndbuf = tfile->socket.sk->sk_sndbuf;
2166                 if (copy_to_user(argp, &sndbuf, sizeof(sndbuf)))
2167                         ret = -EFAULT;
2168                 break;
2169 
2170         case TUNSETSNDBUF:
2171                 if (copy_from_user(&sndbuf, argp, sizeof(sndbuf))) {
2172                         ret = -EFAULT;
2173                         break;
2174                 }
2175 
2176                 tun->sndbuf = sndbuf;
2177                 tun_set_sndbuf(tun);
2178                 break;
2179 
2180         case TUNGETVNETHDRSZ:
2181                 vnet_hdr_sz = tun->vnet_hdr_sz;
2182                 if (copy_to_user(argp, &vnet_hdr_sz, sizeof(vnet_hdr_sz)))
2183                         ret = -EFAULT;
2184                 break;
2185 
2186         case TUNSETVNETHDRSZ:
2187                 if (copy_from_user(&vnet_hdr_sz, argp, sizeof(vnet_hdr_sz))) {
2188                         ret = -EFAULT;
2189                         break;
2190                 }
2191                 if (vnet_hdr_sz < (int)sizeof(struct virtio_net_hdr)) {
2192                         ret = -EINVAL;
2193                         break;
2194                 }
2195 
2196                 tun->vnet_hdr_sz = vnet_hdr_sz;
2197                 break;
2198 
2199         case TUNGETVNETLE:
2200                 le = !!(tun->flags & TUN_VNET_LE);
2201                 if (put_user(le, (int __user *)argp))
2202                         ret = -EFAULT;
2203                 break;
2204 
2205         case TUNSETVNETLE:
2206                 if (get_user(le, (int __user *)argp)) {
2207                         ret = -EFAULT;
2208                         break;
2209                 }
2210                 if (le)
2211                         tun->flags |= TUN_VNET_LE;
2212                 else
2213                         tun->flags &= ~TUN_VNET_LE;
2214                 break;
2215 
2216         case TUNGETVNETBE:
2217                 ret = tun_get_vnet_be(tun, argp);
2218                 break;
2219 
2220         case TUNSETVNETBE:
2221                 ret = tun_set_vnet_be(tun, argp);
2222                 break;
2223 
2224         case TUNATTACHFILTER:
2225                 /* Can be set only for TAPs */
2226                 ret = -EINVAL;
2227                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
2228                         break;
2229                 ret = -EFAULT;
2230                 if (copy_from_user(&tun->fprog, argp, sizeof(tun->fprog)))
2231                         break;
2232 
2233                 ret = tun_attach_filter(tun);
2234                 break;
2235 
2236         case TUNDETACHFILTER:
2237                 /* Can be set only for TAPs */
2238                 ret = -EINVAL;
2239                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
2240                         break;
2241                 ret = 0;
2242                 tun_detach_filter(tun, tun->numqueues);
2243                 break;
2244 
2245         case TUNGETFILTER:
2246                 ret = -EINVAL;
2247                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
2248                         break;
2249                 ret = -EFAULT;
2250                 if (copy_to_user(argp, &tun->fprog, sizeof(tun->fprog)))
2251                         break;
2252                 ret = 0;
2253                 break;
2254 
2255         default:
2256                 ret = -EINVAL;
2257                 break;
2258         }
2259 
2260 unlock:
2261         rtnl_unlock();
2262         if (tun)
2263                 tun_put(tun);
2264         return ret;
2265 }
2266 
2267 static long tun_chr_ioctl(struct file *file,
2268                           unsigned int cmd, unsigned long arg)
2269 {
2270         return __tun_chr_ioctl(file, cmd, arg, sizeof (struct ifreq));
2271 }
2272 
2273 #ifdef CONFIG_COMPAT
2274 static long tun_chr_compat_ioctl(struct file *file,
2275                          unsigned int cmd, unsigned long arg)
2276 {
2277         switch (cmd) {
2278         case TUNSETIFF:
2279         case TUNGETIFF:
2280         case TUNSETTXFILTER:
2281         case TUNGETSNDBUF:
2282         case TUNSETSNDBUF:
2283         case SIOCGIFHWADDR:
2284         case SIOCSIFHWADDR:
2285                 arg = (unsigned long)compat_ptr(arg);
2286                 break;
2287         default:
2288                 arg = (compat_ulong_t)arg;
2289                 break;
2290         }
2291 
2292         /*
2293          * compat_ifreq is shorter than ifreq, so we must not access beyond
2294          * the end of that structure. All fields that are used in this
2295          * driver are compatible though, we don't need to convert the
2296          * contents.
2297          */
2298         return __tun_chr_ioctl(file, cmd, arg, sizeof(struct compat_ifreq));
2299 }
2300 #endif /* CONFIG_COMPAT */
2301 
2302 static int tun_chr_fasync(int fd, struct file *file, int on)
2303 {
2304         struct tun_file *tfile = file->private_data;
2305         int ret;
2306 
2307         if ((ret = fasync_helper(fd, file, on, &tfile->fasync)) < 0)
2308                 goto out;
2309 
2310         if (on) {
2311                 __f_setown(file, task_pid(current), PIDTYPE_PID, 0);
2312                 tfile->flags |= TUN_FASYNC;
2313         } else
2314                 tfile->flags &= ~TUN_FASYNC;
2315         ret = 0;
2316 out:
2317         return ret;
2318 }
2319 
2320 static int tun_chr_open(struct inode *inode, struct file * file)
2321 {
2322         struct net *net = current->nsproxy->net_ns;
2323         struct tun_file *tfile;
2324 
2325         DBG1(KERN_INFO, "tunX: tun_chr_open\n");
2326 
2327         tfile = (struct tun_file *)sk_alloc(net, AF_UNSPEC, GFP_KERNEL,
2328                                             &tun_proto, 0);
2329         if (!tfile)
2330                 return -ENOMEM;
2331         RCU_INIT_POINTER(tfile->tun, NULL);
2332         tfile->flags = 0;
2333         tfile->ifindex = 0;
2334 
2335         init_waitqueue_head(&tfile->wq.wait);
2336         RCU_INIT_POINTER(tfile->socket.wq, &tfile->wq);
2337 
2338         tfile->socket.file = file;
2339         tfile->socket.ops = &tun_socket_ops;
2340 
2341         sock_init_data(&tfile->socket, &tfile->sk);
2342 
2343         tfile->sk.sk_write_space = tun_sock_write_space;
2344         tfile->sk.sk_sndbuf = INT_MAX;
2345 
2346         file->private_data = tfile;
2347         INIT_LIST_HEAD(&tfile->next);
2348 
2349         sock_set_flag(&tfile->sk, SOCK_ZEROCOPY);
2350 
2351         return 0;
2352 }
2353 
2354 static int tun_chr_close(struct inode *inode, struct file *file)
2355 {
2356         struct tun_file *tfile = file->private_data;
2357 
2358         tun_detach(tfile, true);
2359 
2360         return 0;
2361 }
2362 
2363 #ifdef CONFIG_PROC_FS
2364 static void tun_chr_show_fdinfo(struct seq_file *m, struct file *f)
2365 {
2366         struct tun_struct *tun;
2367         struct ifreq ifr;
2368 
2369         memset(&ifr, 0, sizeof(ifr));
2370 
2371         rtnl_lock();
2372         tun = tun_get(f);
2373         if (tun)
2374                 tun_get_iff(current->nsproxy->net_ns, tun, &ifr);
2375         rtnl_unlock();
2376 
2377         if (tun)
2378                 tun_put(tun);
2379 
2380         seq_printf(m, "iff:\t%s\n", ifr.ifr_name);
2381 }
2382 #endif
2383 
2384 static const struct file_operations tun_fops = {
2385         .owner  = THIS_MODULE,
2386         .llseek = no_llseek,
2387         .read_iter  = tun_chr_read_iter,
2388         .write_iter = tun_chr_write_iter,
2389         .poll   = tun_chr_poll,
2390         .unlocked_ioctl = tun_chr_ioctl,
2391 #ifdef CONFIG_COMPAT
2392         .compat_ioctl = tun_chr_compat_ioctl,
2393 #endif
2394         .open   = tun_chr_open,
2395         .release = tun_chr_close,
2396         .fasync = tun_chr_fasync,
2397 #ifdef CONFIG_PROC_FS
2398         .show_fdinfo = tun_chr_show_fdinfo,
2399 #endif
2400 };
2401 
2402 static struct miscdevice tun_miscdev = {
2403         .minor = TUN_MINOR,
2404         .name = "tun",
2405         .nodename = "net/tun",
2406         .fops = &tun_fops,
2407 };
2408 
2409 /* ethtool interface */
2410 
2411 static int tun_get_settings(struct net_device *dev, struct ethtool_cmd *cmd)
2412 {
2413         cmd->supported          = 0;
2414         cmd->advertising        = 0;
2415         ethtool_cmd_speed_set(cmd, SPEED_10);
2416         cmd->duplex             = DUPLEX_FULL;
2417         cmd->port               = PORT_TP;
2418         cmd->phy_address        = 0;
2419         cmd->transceiver        = XCVR_INTERNAL;
2420         cmd->autoneg            = AUTONEG_DISABLE;
2421         cmd->maxtxpkt           = 0;
2422         cmd->maxrxpkt           = 0;
2423         return 0;
2424 }
2425 
2426 static void tun_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
2427 {
2428         struct tun_struct *tun = netdev_priv(dev);
2429 
2430         strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
2431         strlcpy(info->version, DRV_VERSION, sizeof(info->version));
2432 
2433         switch (tun->flags & TUN_TYPE_MASK) {
2434         case IFF_TUN:
2435                 strlcpy(info->bus_info, "tun", sizeof(info->bus_info));
2436                 break;
2437         case IFF_TAP:
2438                 strlcpy(info->bus_info, "tap", sizeof(info->bus_info));
2439                 break;
2440         }
2441 }
2442 
2443 static u32 tun_get_msglevel(struct net_device *dev)
2444 {
2445 #ifdef TUN_DEBUG
2446         struct tun_struct *tun = netdev_priv(dev);
2447         return tun->debug;
2448 #else
2449         return -EOPNOTSUPP;
2450 #endif
2451 }
2452 
2453 static void tun_set_msglevel(struct net_device *dev, u32 value)
2454 {
2455 #ifdef TUN_DEBUG
2456         struct tun_struct *tun = netdev_priv(dev);
2457         tun->debug = value;
2458 #endif
2459 }
2460 
2461 static const struct ethtool_ops tun_ethtool_ops = {
2462         .get_settings   = tun_get_settings,
2463         .get_drvinfo    = tun_get_drvinfo,
2464         .get_msglevel   = tun_get_msglevel,
2465         .set_msglevel   = tun_set_msglevel,
2466         .get_link       = ethtool_op_get_link,
2467         .get_ts_info    = ethtool_op_get_ts_info,
2468 };
2469 
2470 static int tun_queue_resize(struct tun_struct *tun)
2471 {
2472         struct net_device *dev = tun->dev;
2473         struct tun_file *tfile;
2474         struct skb_array **arrays;
2475         int n = tun->numqueues + tun->numdisabled;
2476         int ret, i;
2477 
2478         arrays = kmalloc(sizeof *arrays * n, GFP_KERNEL);
2479         if (!arrays)
2480                 return -ENOMEM;
2481 
2482         for (i = 0; i < tun->numqueues; i++) {
2483                 tfile = rtnl_dereference(tun->tfiles[i]);
2484                 arrays[i] = &tfile->tx_array;
2485         }
2486         list_for_each_entry(tfile, &tun->disabled, next)
2487                 arrays[i++] = &tfile->tx_array;
2488 
2489         ret = skb_array_resize_multiple(arrays, n,
2490                                         dev->tx_queue_len, GFP_KERNEL);
2491 
2492         kfree(arrays);
2493         return ret;
2494 }
2495 
2496 static int tun_device_event(struct notifier_block *unused,
2497                             unsigned long event, void *ptr)
2498 {
2499         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
2500         struct tun_struct *tun = netdev_priv(dev);
2501 
2502         if (dev->rtnl_link_ops != &tun_link_ops)
2503                 return NOTIFY_DONE;
2504 
2505         switch (event) {
2506         case NETDEV_CHANGE_TX_QUEUE_LEN:
2507                 if (tun_queue_resize(tun))
2508                         return NOTIFY_BAD;
2509                 break;
2510         default:
2511                 break;
2512         }
2513 
2514         return NOTIFY_DONE;
2515 }
2516 
2517 static struct notifier_block tun_notifier_block __read_mostly = {
2518         .notifier_call  = tun_device_event,
2519 };
2520 
2521 static int __init tun_init(void)
2522 {
2523         int ret = 0;
2524 
2525         pr_info("%s, %s\n", DRV_DESCRIPTION, DRV_VERSION);
2526         pr_info("%s\n", DRV_COPYRIGHT);
2527 
2528         ret = rtnl_link_register(&tun_link_ops);
2529         if (ret) {
2530                 pr_err("Can't register link_ops\n");
2531                 goto err_linkops;
2532         }
2533 
2534         ret = misc_register(&tun_miscdev);
2535         if (ret) {
2536                 pr_err("Can't register misc device %d\n", TUN_MINOR);
2537                 goto err_misc;
2538         }
2539 
2540         register_netdevice_notifier(&tun_notifier_block);
2541         return  0;
2542 err_misc:
2543         rtnl_link_unregister(&tun_link_ops);
2544 err_linkops:
2545         return ret;
2546 }
2547 
2548 static void tun_cleanup(void)
2549 {
2550         misc_deregister(&tun_miscdev);
2551         rtnl_link_unregister(&tun_link_ops);
2552         unregister_netdevice_notifier(&tun_notifier_block);
2553 }
2554 
2555 /* Get an underlying socket object from tun file.  Returns error unless file is
2556  * attached to a device.  The returned object works like a packet socket, it
2557  * can be used for sock_sendmsg/sock_recvmsg.  The caller is responsible for
2558  * holding a reference to the file for as long as the socket is in use. */
2559 struct socket *tun_get_socket(struct file *file)
2560 {
2561         struct tun_file *tfile;
2562         if (file->f_op != &tun_fops)
2563                 return ERR_PTR(-EINVAL);
2564         tfile = file->private_data;
2565         if (!tfile)
2566                 return ERR_PTR(-EBADFD);
2567         return &tfile->socket;
2568 }
2569 EXPORT_SYMBOL_GPL(tun_get_socket);
2570 
2571 module_init(tun_init);
2572 module_exit(tun_cleanup);
2573 MODULE_DESCRIPTION(DRV_DESCRIPTION);
2574 MODULE_AUTHOR(DRV_COPYRIGHT);
2575 MODULE_LICENSE("GPL");
2576 MODULE_ALIAS_MISCDEV(TUN_MINOR);
2577 MODULE_ALIAS("devname:net/tun");
2578 

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