?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
mfd: wm8350-i2c: Make sure the i2c regmap functions are compiled
[linux/fpc-iii.git] / drivers / net / tun.c
blob813750d096802c82f162338df1bc93a5a0525df2
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/ipv6.h>
69 #include <net/net_namespace.h>
70 #include <net/netns/generic.h>
71 #include <net/rtnetlink.h>
72 #include <net/sock.h>
73
74 #include <asm/uaccess.h>
75
76 /* Uncomment to enable debugging */
77 /* #define TUN_DEBUG 1 */
78
79 #ifdef TUN_DEBUG
80 static int debug;
81
82 #define tun_debug(level, tun, fmt, args...) \
83 do { \
84 if (tun->debug) \
85 netdev_printk(level, tun->dev, fmt, ##args); \
86 } while (0)
87 #define DBG1(level, fmt, args...) \
88 do { \
89 if (debug == 2) \
90 printk(level fmt, ##args); \
91 } while (0)
92 #else
93 #define tun_debug(level, tun, fmt, args...) \
94 do { \
95 if (0) \
96 netdev_printk(level, tun->dev, fmt, ##args); \
97 } while (0)
98 #define DBG1(level, fmt, args...) \
99 do { \
100 if (0) \
101 printk(level fmt, ##args); \
102 } while (0)
103 #endif
104
105 #define GOODCOPY_LEN 128
106
107 #define FLT_EXACT_COUNT 8
108 struct tap_filter {
109 unsigned int count; /* Number of addrs. Zero means disabled */
110 u32 mask[2]; /* Mask of the hashed addrs */
111 unsigned char addr[FLT_EXACT_COUNT][ETH_ALEN];
112 };
113
114 /* DEFAULT_MAX_NUM_RSS_QUEUES were choosed to let the rx/tx queues allocated for
115 * the netdevice to be fit in one page. So we can make sure the success of
116 * memory allocation. TODO: increase the limit. */
117 #define MAX_TAP_QUEUES DEFAULT_MAX_NUM_RSS_QUEUES
118 #define MAX_TAP_FLOWS 4096
119
120 #define TUN_FLOW_EXPIRE (3 * HZ)
121
122 /* A tun_file connects an open character device to a tuntap netdevice. It
123 * also contains all socket related strctures (except sock_fprog and tap_filter)
124 * to serve as one transmit queue for tuntap device. The sock_fprog and
125 * tap_filter were kept in tun_struct since they were used for filtering for the
126 * netdevice not for a specific queue (at least I didn't see the requirement for
127 * this).
128 *
129 * RCU usage:
130 * The tun_file and tun_struct are loosely coupled, the pointer from one to the
131 * other can only be read while rcu_read_lock or rtnl_lock is held.
132 */
133 struct tun_file {
134 struct sock sk;
135 struct socket socket;
136 struct socket_wq wq;
137 struct tun_struct __rcu *tun;
138 struct net *net;
139 struct fasync_struct *fasync;
140 /* only used for fasnyc */
141 unsigned int flags;
142 union {
143 u16 queue_index;
144 unsigned int ifindex;
145 };
146 struct list_head next;
147 struct tun_struct *detached;
148 };
149
150 struct tun_flow_entry {
151 struct hlist_node hash_link;
152 struct rcu_head rcu;
153 struct tun_struct *tun;
154
155 u32 rxhash;
156 int queue_index;
157 unsigned long updated;
158 };
159
160 #define TUN_NUM_FLOW_ENTRIES 1024
161
162 /* Since the socket were moved to tun_file, to preserve the behavior of persist
163 * device, socket filter, sndbuf and vnet header size were restore when the
164 * file were attached to a persist device.
165 */
166 struct tun_struct {
167 struct tun_file __rcu *tfiles[MAX_TAP_QUEUES];
168 unsigned int numqueues;
169 unsigned int flags;
170 kuid_t owner;
171 kgid_t group;
172
173 struct net_device *dev;
174 netdev_features_t set_features;
175 #define TUN_USER_FEATURES (NETIF_F_HW_CSUM|NETIF_F_TSO_ECN|NETIF_F_TSO| \
176 NETIF_F_TSO6|NETIF_F_UFO)
177
178 int vnet_hdr_sz;
179 int sndbuf;
180 struct tap_filter txflt;
181 struct sock_fprog fprog;
182 /* protected by rtnl lock */
183 bool filter_attached;
184 #ifdef TUN_DEBUG
185 int debug;
186 #endif
187 spinlock_t lock;
188 struct hlist_head flows[TUN_NUM_FLOW_ENTRIES];
189 struct timer_list flow_gc_timer;
190 unsigned long ageing_time;
191 unsigned int numdisabled;
192 struct list_head disabled;
193 void *security;
194 u32 flow_count;
195 };
196
197 static inline u32 tun_hashfn(u32 rxhash)
198 {
199 return rxhash & 0x3ff;
200 }
201
202 static struct tun_flow_entry *tun_flow_find(struct hlist_head *head, u32 rxhash)
203 {
204 struct tun_flow_entry *e;
205
206 hlist_for_each_entry_rcu(e, head, hash_link) {
207 if (e->rxhash == rxhash)
208 return e;
209 }
210 return NULL;
211 }
212
213 static struct tun_flow_entry *tun_flow_create(struct tun_struct *tun,
214 struct hlist_head *head,
215 u32 rxhash, u16 queue_index)
216 {
217 struct tun_flow_entry *e = kmalloc(sizeof(*e), GFP_ATOMIC);
218
219 if (e) {
220 tun_debug(KERN_INFO, tun, "create flow: hash %u index %u\n",
221 rxhash, queue_index);
222 e->updated = jiffies;
223 e->rxhash = rxhash;
224 e->queue_index = queue_index;
225 e->tun = tun;
226 hlist_add_head_rcu(&e->hash_link, head);
227 ++tun->flow_count;
228 }
229 return e;
230 }
231
232 static void tun_flow_delete(struct tun_struct *tun, struct tun_flow_entry *e)
233 {
234 tun_debug(KERN_INFO, tun, "delete flow: hash %u index %u\n",
235 e->rxhash, e->queue_index);
236 hlist_del_rcu(&e->hash_link);
237 kfree_rcu(e, rcu);
238 --tun->flow_count;
239 }
240
241 static void tun_flow_flush(struct tun_struct *tun)
242 {
243 int i;
244
245 spin_lock_bh(&tun->lock);
246 for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
247 struct tun_flow_entry *e;
248 struct hlist_node *n;
249
250 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link)
251 tun_flow_delete(tun, e);
252 }
253 spin_unlock_bh(&tun->lock);
254 }
255
256 static void tun_flow_delete_by_queue(struct tun_struct *tun, u16 queue_index)
257 {
258 int i;
259
260 spin_lock_bh(&tun->lock);
261 for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
262 struct tun_flow_entry *e;
263 struct hlist_node *n;
264
265 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
266 if (e->queue_index == queue_index)
267 tun_flow_delete(tun, e);
268 }
269 }
270 spin_unlock_bh(&tun->lock);
271 }
272
273 static void tun_flow_cleanup(unsigned long data)
274 {
275 struct tun_struct *tun = (struct tun_struct *)data;
276 unsigned long delay = tun->ageing_time;
277 unsigned long next_timer = jiffies + delay;
278 unsigned long count = 0;
279 int i;
280
281 tun_debug(KERN_INFO, tun, "tun_flow_cleanup\n");
282
283 spin_lock_bh(&tun->lock);
284 for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
285 struct tun_flow_entry *e;
286 struct hlist_node *n;
287
288 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
289 unsigned long this_timer;
290 count++;
291 this_timer = e->updated + delay;
292 if (time_before_eq(this_timer, jiffies))
293 tun_flow_delete(tun, e);
294 else if (time_before(this_timer, next_timer))
295 next_timer = this_timer;
296 }
297 }
298
299 if (count)
300 mod_timer(&tun->flow_gc_timer, round_jiffies_up(next_timer));
301 spin_unlock_bh(&tun->lock);
302 }
303
304 static void tun_flow_update(struct tun_struct *tun, u32 rxhash,
305 struct tun_file *tfile)
306 {
307 struct hlist_head *head;
308 struct tun_flow_entry *e;
309 unsigned long delay = tun->ageing_time;
310 u16 queue_index = tfile->queue_index;
311
312 if (!rxhash)
313 return;
314 else
315 head = &tun->flows[tun_hashfn(rxhash)];
316
317 rcu_read_lock();
318
319 /* We may get a very small possibility of OOO during switching, not
320 * worth to optimize.*/
321 if (tun->numqueues == 1 || tfile->detached)
322 goto unlock;
323
324 e = tun_flow_find(head, rxhash);
325 if (likely(e)) {
326 /* TODO: keep queueing to old queue until it's empty? */
327 e->queue_index = queue_index;
328 e->updated = jiffies;
329 } else {
330 spin_lock_bh(&tun->lock);
331 if (!tun_flow_find(head, rxhash) &&
332 tun->flow_count < MAX_TAP_FLOWS)
333 tun_flow_create(tun, head, rxhash, queue_index);
334
335 if (!timer_pending(&tun->flow_gc_timer))
336 mod_timer(&tun->flow_gc_timer,
337 round_jiffies_up(jiffies + delay));
338 spin_unlock_bh(&tun->lock);
339 }
340
341 unlock:
342 rcu_read_unlock();
343 }
344
345 /* We try to identify a flow through its rxhash first. The reason that
346 * we do not check rxq no. is becuase some cards(e.g 82599), chooses
347 * the rxq based on the txq where the last packet of the flow comes. As
348 * the userspace application move between processors, we may get a
349 * different rxq no. here. If we could not get rxhash, then we would
350 * hope the rxq no. may help here.
351 */
352 static u16 tun_select_queue(struct net_device *dev, struct sk_buff *skb)
353 {
354 struct tun_struct *tun = netdev_priv(dev);
355 struct tun_flow_entry *e;
356 u32 txq = 0;
357 u32 numqueues = 0;
358
359 rcu_read_lock();
360 numqueues = ACCESS_ONCE(tun->numqueues);
361
362 txq = skb_get_rxhash(skb);
363 if (txq) {
364 e = tun_flow_find(&tun->flows[tun_hashfn(txq)], txq);
365 if (e)
366 txq = e->queue_index;
367 else
368 /* use multiply and shift instead of expensive divide */
369 txq = ((u64)txq * numqueues) >> 32;
370 } else if (likely(skb_rx_queue_recorded(skb))) {
371 txq = skb_get_rx_queue(skb);
372 while (unlikely(txq >= numqueues))
373 txq -= numqueues;
374 }
375
376 rcu_read_unlock();
377 return txq;
378 }
379
380 static inline bool tun_not_capable(struct tun_struct *tun)
381 {
382 const struct cred *cred = current_cred();
383 struct net *net = dev_net(tun->dev);
384
385 return ((uid_valid(tun->owner) && !uid_eq(cred->euid, tun->owner)) ||
386 (gid_valid(tun->group) && !in_egroup_p(tun->group))) &&
387 !ns_capable(net->user_ns, CAP_NET_ADMIN);
388 }
389
390 static void tun_set_real_num_queues(struct tun_struct *tun)
391 {
392 netif_set_real_num_tx_queues(tun->dev, tun->numqueues);
393 netif_set_real_num_rx_queues(tun->dev, tun->numqueues);
394 }
395
396 static void tun_disable_queue(struct tun_struct *tun, struct tun_file *tfile)
397 {
398 tfile->detached = tun;
399 list_add_tail(&tfile->next, &tun->disabled);
400 ++tun->numdisabled;
401 }
402
403 static struct tun_struct *tun_enable_queue(struct tun_file *tfile)
404 {
405 struct tun_struct *tun = tfile->detached;
406
407 tfile->detached = NULL;
408 list_del_init(&tfile->next);
409 --tun->numdisabled;
410 return tun;
411 }
412
413 static void tun_queue_purge(struct tun_file *tfile)
414 {
415 skb_queue_purge(&tfile->sk.sk_receive_queue);
416 skb_queue_purge(&tfile->sk.sk_error_queue);
417 }
418
419 static void __tun_detach(struct tun_file *tfile, bool clean)
420 {
421 struct tun_file *ntfile;
422 struct tun_struct *tun;
423
424 tun = rtnl_dereference(tfile->tun);
425
426 if (tun && !tfile->detached) {
427 u16 index = tfile->queue_index;
428 BUG_ON(index >= tun->numqueues);
429
430 rcu_assign_pointer(tun->tfiles[index],
431 tun->tfiles[tun->numqueues - 1]);
432 ntfile = rtnl_dereference(tun->tfiles[index]);
433 ntfile->queue_index = index;
434
435 --tun->numqueues;
436 if (clean) {
437 rcu_assign_pointer(tfile->tun, NULL);
438 sock_put(&tfile->sk);
439 } else
440 tun_disable_queue(tun, tfile);
441
442 synchronize_net();
443 tun_flow_delete_by_queue(tun, tun->numqueues + 1);
444 /* Drop read queue */
445 tun_queue_purge(tfile);
446 tun_set_real_num_queues(tun);
447 } else if (tfile->detached && clean) {
448 tun = tun_enable_queue(tfile);
449 sock_put(&tfile->sk);
450 }
451
452 if (clean) {
453 if (tun && tun->numqueues == 0 && tun->numdisabled == 0) {
454 netif_carrier_off(tun->dev);
455
456 if (!(tun->flags & TUN_PERSIST) &&
457 tun->dev->reg_state == NETREG_REGISTERED)
458 unregister_netdevice(tun->dev);
459 }
460
461 BUG_ON(!test_bit(SOCK_EXTERNALLY_ALLOCATED,
462 &tfile->socket.flags));
463 sk_release_kernel(&tfile->sk);
464 }
465 }
466
467 static void tun_detach(struct tun_file *tfile, bool clean)
468 {
469 rtnl_lock();
470 __tun_detach(tfile, clean);
471 rtnl_unlock();
472 }
473
474 static void tun_detach_all(struct net_device *dev)
475 {
476 struct tun_struct *tun = netdev_priv(dev);
477 struct tun_file *tfile, *tmp;
478 int i, n = tun->numqueues;
479
480 for (i = 0; i < n; i++) {
481 tfile = rtnl_dereference(tun->tfiles[i]);
482 BUG_ON(!tfile);
483 wake_up_all(&tfile->wq.wait);
484 rcu_assign_pointer(tfile->tun, NULL);
485 --tun->numqueues;
486 }
487 list_for_each_entry(tfile, &tun->disabled, next) {
488 wake_up_all(&tfile->wq.wait);
489 rcu_assign_pointer(tfile->tun, NULL);
490 }
491 BUG_ON(tun->numqueues != 0);
492
493 synchronize_net();
494 for (i = 0; i < n; i++) {
495 tfile = rtnl_dereference(tun->tfiles[i]);
496 /* Drop read queue */
497 tun_queue_purge(tfile);
498 sock_put(&tfile->sk);
499 }
500 list_for_each_entry_safe(tfile, tmp, &tun->disabled, next) {
501 tun_enable_queue(tfile);
502 tun_queue_purge(tfile);
503 sock_put(&tfile->sk);
504 }
505 BUG_ON(tun->numdisabled != 0);
506
507 if (tun->flags & TUN_PERSIST)
508 module_put(THIS_MODULE);
509 }
510
511 static int tun_attach(struct tun_struct *tun, struct file *file, bool skip_filter)
512 {
513 struct tun_file *tfile = file->private_data;
514 int err;
515
516 err = security_tun_dev_attach(tfile->socket.sk, tun->security);
517 if (err < 0)
518 goto out;
519
520 err = -EINVAL;
521 if (rtnl_dereference(tfile->tun) && !tfile->detached)
522 goto out;
523
524 err = -EBUSY;
525 if (!(tun->flags & TUN_TAP_MQ) && tun->numqueues == 1)
526 goto out;
527
528 err = -E2BIG;
529 if (!tfile->detached &&
530 tun->numqueues + tun->numdisabled == MAX_TAP_QUEUES)
531 goto out;
532
533 err = 0;
534
535 /* Re-attach the filter to presist device */
536 if (!skip_filter && (tun->filter_attached == true)) {
537 err = sk_attach_filter(&tun->fprog, tfile->socket.sk);
538 if (!err)
539 goto out;
540 }
541 tfile->queue_index = tun->numqueues;
542 rcu_assign_pointer(tfile->tun, tun);
543 rcu_assign_pointer(tun->tfiles[tun->numqueues], tfile);
544 tun->numqueues++;
545
546 if (tfile->detached)
547 tun_enable_queue(tfile);
548 else
549 sock_hold(&tfile->sk);
550
551 tun_set_real_num_queues(tun);
552
553 /* device is allowed to go away first, so no need to hold extra
554 * refcnt.
555 */
556
557 out:
558 return err;
559 }
560
561 static struct tun_struct *__tun_get(struct tun_file *tfile)
562 {
563 struct tun_struct *tun;
564
565 rcu_read_lock();
566 tun = rcu_dereference(tfile->tun);
567 if (tun)
568 dev_hold(tun->dev);
569 rcu_read_unlock();
570
571 return tun;
572 }
573
574 static struct tun_struct *tun_get(struct file *file)
575 {
576 return __tun_get(file->private_data);
577 }
578
579 static void tun_put(struct tun_struct *tun)
580 {
581 dev_put(tun->dev);
582 }
583
584 /* TAP filtering */
585 static void addr_hash_set(u32 *mask, const u8 *addr)
586 {
587 int n = ether_crc(ETH_ALEN, addr) >> 26;
588 mask[n >> 5] |= (1 << (n & 31));
589 }
590
591 static unsigned int addr_hash_test(const u32 *mask, const u8 *addr)
592 {
593 int n = ether_crc(ETH_ALEN, addr) >> 26;
594 return mask[n >> 5] & (1 << (n & 31));
595 }
596
597 static int update_filter(struct tap_filter *filter, void __user *arg)
598 {
599 struct { u8 u[ETH_ALEN]; } *addr;
600 struct tun_filter uf;
601 int err, alen, n, nexact;
602
603 if (copy_from_user(&uf, arg, sizeof(uf)))
604 return -EFAULT;
605
606 if (!uf.count) {
607 /* Disabled */
608 filter->count = 0;
609 return 0;
610 }
611
612 alen = ETH_ALEN * uf.count;
613 addr = kmalloc(alen, GFP_KERNEL);
614 if (!addr)
615 return -ENOMEM;
616
617 if (copy_from_user(addr, arg + sizeof(uf), alen)) {
618 err = -EFAULT;
619 goto done;
620 }
621
622 /* The filter is updated without holding any locks. Which is
623 * perfectly safe. We disable it first and in the worst
624 * case we'll accept a few undesired packets. */
625 filter->count = 0;
626 wmb();
627
628 /* Use first set of addresses as an exact filter */
629 for (n = 0; n < uf.count && n < FLT_EXACT_COUNT; n++)
630 memcpy(filter->addr[n], addr[n].u, ETH_ALEN);
631
632 nexact = n;
633
634 /* Remaining multicast addresses are hashed,
635 * unicast will leave the filter disabled. */
636 memset(filter->mask, 0, sizeof(filter->mask));
637 for (; n < uf.count; n++) {
638 if (!is_multicast_ether_addr(addr[n].u)) {
639 err = 0; /* no filter */
640 goto done;
641 }
642 addr_hash_set(filter->mask, addr[n].u);
643 }
644
645 /* For ALLMULTI just set the mask to all ones.
646 * This overrides the mask populated above. */
647 if ((uf.flags & TUN_FLT_ALLMULTI))
648 memset(filter->mask, ~0, sizeof(filter->mask));
649
650 /* Now enable the filter */
651 wmb();
652 filter->count = nexact;
653
654 /* Return the number of exact filters */
655 err = nexact;
656
657 done:
658 kfree(addr);
659 return err;
660 }
661
662 /* Returns: 0 - drop, !=0 - accept */
663 static int run_filter(struct tap_filter *filter, const struct sk_buff *skb)
664 {
665 /* Cannot use eth_hdr(skb) here because skb_mac_hdr() is incorrect
666 * at this point. */
667 struct ethhdr *eh = (struct ethhdr *) skb->data;
668 int i;
669
670 /* Exact match */
671 for (i = 0; i < filter->count; i++)
672 if (ether_addr_equal(eh->h_dest, filter->addr[i]))
673 return 1;
674
675 /* Inexact match (multicast only) */
676 if (is_multicast_ether_addr(eh->h_dest))
677 return addr_hash_test(filter->mask, eh->h_dest);
678
679 return 0;
680 }
681
682 /*
683 * Checks whether the packet is accepted or not.
684 * Returns: 0 - drop, !=0 - accept
685 */
686 static int check_filter(struct tap_filter *filter, const struct sk_buff *skb)
687 {
688 if (!filter->count)
689 return 1;
690
691 return run_filter(filter, skb);
692 }
693
694 /* Network device part of the driver */
695
696 static const struct ethtool_ops tun_ethtool_ops;
697
698 /* Net device detach from fd. */
699 static void tun_net_uninit(struct net_device *dev)
700 {
701 tun_detach_all(dev);
702 }
703
704 /* Net device open. */
705 static int tun_net_open(struct net_device *dev)
706 {
707 netif_tx_start_all_queues(dev);
708 return 0;
709 }
710
711 /* Net device close. */
712 static int tun_net_close(struct net_device *dev)
713 {
714 netif_tx_stop_all_queues(dev);
715 return 0;
716 }
717
718 /* Net device start xmit */
719 static netdev_tx_t tun_net_xmit(struct sk_buff *skb, struct net_device *dev)
720 {
721 struct tun_struct *tun = netdev_priv(dev);
722 int txq = skb->queue_mapping;
723 struct tun_file *tfile;
724
725 rcu_read_lock();
726 tfile = rcu_dereference(tun->tfiles[txq]);
727
728 /* Drop packet if interface is not attached */
729 if (txq >= tun->numqueues)
730 goto drop;
731
732 tun_debug(KERN_INFO, tun, "tun_net_xmit %d\n", skb->len);
733
734 BUG_ON(!tfile);
735
736 /* Drop if the filter does not like it.
737 * This is a noop if the filter is disabled.
738 * Filter can be enabled only for the TAP devices. */
739 if (!check_filter(&tun->txflt, skb))
740 goto drop;
741
742 if (tfile->socket.sk->sk_filter &&
743 sk_filter(tfile->socket.sk, skb))
744 goto drop;
745
746 /* Limit the number of packets queued by dividing txq length with the
747 * number of queues.
748 */
749 if (skb_queue_len(&tfile->socket.sk->sk_receive_queue)
750 >= dev->tx_queue_len / tun->numqueues)
751 goto drop;
752
753 if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC)))
754 goto drop;
755
756 skb_tx_timestamp(skb);
757
758 /* Orphan the skb - required as we might hang on to it
759 * for indefinite time.
760 */
761 skb_orphan(skb);
762
763 nf_reset(skb);
764
765 /* Enqueue packet */
766 skb_queue_tail(&tfile->socket.sk->sk_receive_queue, skb);
767
768 /* Notify and wake up reader process */
769 if (tfile->flags & TUN_FASYNC)
770 kill_fasync(&tfile->fasync, SIGIO, POLL_IN);
771 wake_up_interruptible_poll(&tfile->wq.wait, POLLIN |
772 POLLRDNORM | POLLRDBAND);
773
774 rcu_read_unlock();
775 return NETDEV_TX_OK;
776
777 drop:
778 dev->stats.tx_dropped++;
779 skb_tx_error(skb);
780 kfree_skb(skb);
781 rcu_read_unlock();
782 return NETDEV_TX_OK;
783 }
784
785 static void tun_net_mclist(struct net_device *dev)
786 {
787 /*
788 * This callback is supposed to deal with mc filter in
789 * _rx_ path and has nothing to do with the _tx_ path.
790 * In rx path we always accept everything userspace gives us.
791 */
792 }
793
794 #define MIN_MTU 68
795 #define MAX_MTU 65535
796
797 static int
798 tun_net_change_mtu(struct net_device *dev, int new_mtu)
799 {
800 if (new_mtu < MIN_MTU || new_mtu + dev->hard_header_len > MAX_MTU)
801 return -EINVAL;
802 dev->mtu = new_mtu;
803 return 0;
804 }
805
806 static netdev_features_t tun_net_fix_features(struct net_device *dev,
807 netdev_features_t features)
808 {
809 struct tun_struct *tun = netdev_priv(dev);
810
811 return (features & tun->set_features) | (features & ~TUN_USER_FEATURES);
812 }
813 #ifdef CONFIG_NET_POLL_CONTROLLER
814 static void tun_poll_controller(struct net_device *dev)
815 {
816 /*
817 * Tun only receives frames when:
818 * 1) the char device endpoint gets data from user space
819 * 2) the tun socket gets a sendmsg call from user space
820 * Since both of those are syncronous operations, we are guaranteed
821 * never to have pending data when we poll for it
822 * so theres nothing to do here but return.
823 * We need this though so netpoll recognizes us as an interface that
824 * supports polling, which enables bridge devices in virt setups to
825 * still use netconsole
826 */
827 return;
828 }
829 #endif
830 static const struct net_device_ops tun_netdev_ops = {
831 .ndo_uninit = tun_net_uninit,
832 .ndo_open = tun_net_open,
833 .ndo_stop = tun_net_close,
834 .ndo_start_xmit = tun_net_xmit,
835 .ndo_change_mtu = tun_net_change_mtu,
836 .ndo_fix_features = tun_net_fix_features,
837 .ndo_select_queue = tun_select_queue,
838 #ifdef CONFIG_NET_POLL_CONTROLLER
839 .ndo_poll_controller = tun_poll_controller,
840 #endif
841 };
842
843 static const struct net_device_ops tap_netdev_ops = {
844 .ndo_uninit = tun_net_uninit,
845 .ndo_open = tun_net_open,
846 .ndo_stop = tun_net_close,
847 .ndo_start_xmit = tun_net_xmit,
848 .ndo_change_mtu = tun_net_change_mtu,
849 .ndo_fix_features = tun_net_fix_features,
850 .ndo_set_rx_mode = tun_net_mclist,
851 .ndo_set_mac_address = eth_mac_addr,
852 .ndo_validate_addr = eth_validate_addr,
853 .ndo_select_queue = tun_select_queue,
854 #ifdef CONFIG_NET_POLL_CONTROLLER
855 .ndo_poll_controller = tun_poll_controller,
856 #endif
857 };
858
859 static void tun_flow_init(struct tun_struct *tun)
860 {
861 int i;
862
863 for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++)
864 INIT_HLIST_HEAD(&tun->flows[i]);
865
866 tun->ageing_time = TUN_FLOW_EXPIRE;
867 setup_timer(&tun->flow_gc_timer, tun_flow_cleanup, (unsigned long)tun);
868 mod_timer(&tun->flow_gc_timer,
869 round_jiffies_up(jiffies + tun->ageing_time));
870 }
871
872 static void tun_flow_uninit(struct tun_struct *tun)
873 {
874 del_timer_sync(&tun->flow_gc_timer);
875 tun_flow_flush(tun);
876 }
877
878 /* Initialize net device. */
879 static void tun_net_init(struct net_device *dev)
880 {
881 struct tun_struct *tun = netdev_priv(dev);
882
883 switch (tun->flags & TUN_TYPE_MASK) {
884 case TUN_TUN_DEV:
885 dev->netdev_ops = &tun_netdev_ops;
886
887 /* Point-to-Point TUN Device */
888 dev->hard_header_len = 0;
889 dev->addr_len = 0;
890 dev->mtu = 1500;
891
892 /* Zero header length */
893 dev->type = ARPHRD_NONE;
894 dev->flags = IFF_POINTOPOINT | IFF_NOARP | IFF_MULTICAST;
895 dev->tx_queue_len = TUN_READQ_SIZE; /* We prefer our own queue length */
896 break;
897
898 case TUN_TAP_DEV:
899 dev->netdev_ops = &tap_netdev_ops;
900 /* Ethernet TAP Device */
901 ether_setup(dev);
902 dev->priv_flags &= ~IFF_TX_SKB_SHARING;
903 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
904
905 eth_hw_addr_random(dev);
906
907 dev->tx_queue_len = TUN_READQ_SIZE; /* We prefer our own queue length */
908 break;
909 }
910 }
911
912 /* Character device part */
913
914 /* Poll */
915 static unsigned int tun_chr_poll(struct file *file, poll_table *wait)
916 {
917 struct tun_file *tfile = file->private_data;
918 struct tun_struct *tun = __tun_get(tfile);
919 struct sock *sk;
920 unsigned int mask = 0;
921
922 if (!tun)
923 return POLLERR;
924
925 sk = tfile->socket.sk;
926
927 tun_debug(KERN_INFO, tun, "tun_chr_poll\n");
928
929 poll_wait(file, &tfile->wq.wait, wait);
930
931 if (!skb_queue_empty(&sk->sk_receive_queue))
932 mask |= POLLIN | POLLRDNORM;
933
934 if (sock_writeable(sk) ||
935 (!test_and_set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags) &&
936 sock_writeable(sk)))
937 mask |= POLLOUT | POLLWRNORM;
938
939 if (tun->dev->reg_state != NETREG_REGISTERED)
940 mask = POLLERR;
941
942 tun_put(tun);
943 return mask;
944 }
945
946 /* prepad is the amount to reserve at front. len is length after that.
947 * linear is a hint as to how much to copy (usually headers). */
948 static struct sk_buff *tun_alloc_skb(struct tun_file *tfile,
949 size_t prepad, size_t len,
950 size_t linear, int noblock)
951 {
952 struct sock *sk = tfile->socket.sk;
953 struct sk_buff *skb;
954 int err;
955
956 /* Under a page? Don't bother with paged skb. */
957 if (prepad + len < PAGE_SIZE || !linear)
958 linear = len;
959
960 skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
961 &err, 0);
962 if (!skb)
963 return ERR_PTR(err);
964
965 skb_reserve(skb, prepad);
966 skb_put(skb, linear);
967 skb->data_len = len - linear;
968 skb->len += len - linear;
969
970 return skb;
971 }
972
973 /* Get packet from user space buffer */
974 static ssize_t tun_get_user(struct tun_struct *tun, struct tun_file *tfile,
975 void *msg_control, const struct iovec *iv,
976 size_t total_len, size_t count, int noblock)
977 {
978 struct tun_pi pi = { 0, cpu_to_be16(ETH_P_IP) };
979 struct sk_buff *skb;
980 size_t len = total_len, align = NET_SKB_PAD, linear;
981 struct virtio_net_hdr gso = { 0 };
982 int good_linear;
983 int offset = 0;
984 int copylen;
985 bool zerocopy = false;
986 int err;
987 u32 rxhash;
988
989 if (!(tun->flags & TUN_NO_PI)) {
990 if (len < sizeof(pi))
991 return -EINVAL;
992 len -= sizeof(pi);
993
994 if (memcpy_fromiovecend((void *)&pi, iv, 0, sizeof(pi)))
995 return -EFAULT;
996 offset += sizeof(pi);
997 }
998
999 if (tun->flags & TUN_VNET_HDR) {
1000 if (len < tun->vnet_hdr_sz)
1001 return -EINVAL;
1002 len -= tun->vnet_hdr_sz;
1003
1004 if (memcpy_fromiovecend((void *)&gso, iv, offset, sizeof(gso)))
1005 return -EFAULT;
1006
1007 if ((gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
1008 gso.csum_start + gso.csum_offset + 2 > gso.hdr_len)
1009 gso.hdr_len = gso.csum_start + gso.csum_offset + 2;
1010
1011 if (gso.hdr_len > len)
1012 return -EINVAL;
1013 offset += tun->vnet_hdr_sz;
1014 }
1015
1016 if ((tun->flags & TUN_TYPE_MASK) == TUN_TAP_DEV) {
1017 align += NET_IP_ALIGN;
1018 if (unlikely(len < ETH_HLEN ||
1019 (gso.hdr_len && gso.hdr_len < ETH_HLEN)))
1020 return -EINVAL;
1021 }
1022
1023 good_linear = SKB_MAX_HEAD(align);
1024
1025 if (msg_control) {
1026 /* There are 256 bytes to be copied in skb, so there is
1027 * enough room for skb expand head in case it is used.
1028 * The rest of the buffer is mapped from userspace.
1029 */
1030 copylen = gso.hdr_len ? gso.hdr_len : GOODCOPY_LEN;
1031 if (copylen > good_linear)
1032 copylen = good_linear;
1033 linear = copylen;
1034 if (iov_pages(iv, offset + copylen, count) <= MAX_SKB_FRAGS)
1035 zerocopy = true;
1036 }
1037
1038 if (!zerocopy) {
1039 copylen = len;
1040 if (gso.hdr_len > good_linear)
1041 linear = good_linear;
1042 else
1043 linear = gso.hdr_len;
1044 }
1045
1046 skb = tun_alloc_skb(tfile, align, copylen, linear, noblock);
1047 if (IS_ERR(skb)) {
1048 if (PTR_ERR(skb) != -EAGAIN)
1049 tun->dev->stats.rx_dropped++;
1050 return PTR_ERR(skb);
1051 }
1052
1053 if (zerocopy)
1054 err = zerocopy_sg_from_iovec(skb, iv, offset, count);
1055 else {
1056 err = skb_copy_datagram_from_iovec(skb, 0, iv, offset, len);
1057 if (!err && msg_control) {
1058 struct ubuf_info *uarg = msg_control;
1059 uarg->callback(uarg, false);
1060 }
1061 }
1062
1063 if (err) {
1064 tun->dev->stats.rx_dropped++;
1065 kfree_skb(skb);
1066 return -EFAULT;
1067 }
1068
1069 if (gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
1070 if (!skb_partial_csum_set(skb, gso.csum_start,
1071 gso.csum_offset)) {
1072 tun->dev->stats.rx_frame_errors++;
1073 kfree_skb(skb);
1074 return -EINVAL;
1075 }
1076 }
1077
1078 switch (tun->flags & TUN_TYPE_MASK) {
1079 case TUN_TUN_DEV:
1080 if (tun->flags & TUN_NO_PI) {
1081 switch (skb->data[0] & 0xf0) {
1082 case 0x40:
1083 pi.proto = htons(ETH_P_IP);
1084 break;
1085 case 0x60:
1086 pi.proto = htons(ETH_P_IPV6);
1087 break;
1088 default:
1089 tun->dev->stats.rx_dropped++;
1090 kfree_skb(skb);
1091 return -EINVAL;
1092 }
1093 }
1094
1095 skb_reset_mac_header(skb);
1096 skb->protocol = pi.proto;
1097 skb->dev = tun->dev;
1098 break;
1099 case TUN_TAP_DEV:
1100 skb->protocol = eth_type_trans(skb, tun->dev);
1101 break;
1102 }
1103
1104 skb_reset_network_header(skb);
1105
1106 if (gso.gso_type != VIRTIO_NET_HDR_GSO_NONE) {
1107 pr_debug("GSO!\n");
1108 switch (gso.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
1109 case VIRTIO_NET_HDR_GSO_TCPV4:
1110 skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
1111 break;
1112 case VIRTIO_NET_HDR_GSO_TCPV6:
1113 skb_shinfo(skb)->gso_type = SKB_GSO_TCPV6;
1114 break;
1115 case VIRTIO_NET_HDR_GSO_UDP:
1116 skb_shinfo(skb)->gso_type = SKB_GSO_UDP;
1117 if (skb->protocol == htons(ETH_P_IPV6))
1118 ipv6_proxy_select_ident(skb);
1119 break;
1120 default:
1121 tun->dev->stats.rx_frame_errors++;
1122 kfree_skb(skb);
1123 return -EINVAL;
1124 }
1125
1126 if (gso.gso_type & VIRTIO_NET_HDR_GSO_ECN)
1127 skb_shinfo(skb)->gso_type |= SKB_GSO_TCP_ECN;
1128
1129 skb_shinfo(skb)->gso_size = gso.gso_size;
1130 if (skb_shinfo(skb)->gso_size == 0) {
1131 tun->dev->stats.rx_frame_errors++;
1132 kfree_skb(skb);
1133 return -EINVAL;
1134 }
1135
1136 /* Header must be checked, and gso_segs computed. */
1137 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
1138 skb_shinfo(skb)->gso_segs = 0;
1139 }
1140
1141 /* copy skb_ubuf_info for callback when skb has no error */
1142 if (zerocopy) {
1143 skb_shinfo(skb)->destructor_arg = msg_control;
1144 skb_shinfo(skb)->tx_flags |= SKBTX_DEV_ZEROCOPY;
1145 skb_shinfo(skb)->tx_flags |= SKBTX_SHARED_FRAG;
1146 }
1147
1148 skb_probe_transport_header(skb, 0);
1149
1150 rxhash = skb_get_rxhash(skb);
1151 netif_rx_ni(skb);
1152
1153 tun->dev->stats.rx_packets++;
1154 tun->dev->stats.rx_bytes += len;
1155
1156 tun_flow_update(tun, rxhash, tfile);
1157 return total_len;
1158 }
1159
1160 static ssize_t tun_chr_aio_write(struct kiocb *iocb, const struct iovec *iv,
1161 unsigned long count, loff_t pos)
1162 {
1163 struct file *file = iocb->ki_filp;
1164 struct tun_struct *tun = tun_get(file);
1165 struct tun_file *tfile = file->private_data;
1166 ssize_t result;
1167
1168 if (!tun)
1169 return -EBADFD;
1170
1171 tun_debug(KERN_INFO, tun, "tun_chr_write %ld\n", count);
1172
1173 result = tun_get_user(tun, tfile, NULL, iv, iov_length(iv, count),
1174 count, file->f_flags & O_NONBLOCK);
1175
1176 tun_put(tun);
1177 return result;
1178 }
1179
1180 /* Put packet to the user space buffer */
1181 static ssize_t tun_put_user(struct tun_struct *tun,
1182 struct tun_file *tfile,
1183 struct sk_buff *skb,
1184 const struct iovec *iv, int len)
1185 {
1186 struct tun_pi pi = { 0, skb->protocol };
1187 ssize_t total = 0;
1188 int vlan_offset = 0, copied;
1189 int vlan_hlen = 0;
1190
1191 if (vlan_tx_tag_present(skb))
1192 vlan_hlen = VLAN_HLEN;
1193
1194 if (!(tun->flags & TUN_NO_PI)) {
1195 if ((len -= sizeof(pi)) < 0)
1196 return -EINVAL;
1197
1198 if (len < skb->len) {
1199 /* Packet will be striped */
1200 pi.flags |= TUN_PKT_STRIP;
1201 }
1202
1203 if (memcpy_toiovecend(iv, (void *) &pi, 0, sizeof(pi)))
1204 return -EFAULT;
1205 total += sizeof(pi);
1206 }
1207
1208 if (tun->flags & TUN_VNET_HDR) {
1209 struct virtio_net_hdr gso = { 0 }; /* no info leak */
1210 if ((len -= tun->vnet_hdr_sz) < 0)
1211 return -EINVAL;
1212
1213 if (skb_is_gso(skb)) {
1214 struct skb_shared_info *sinfo = skb_shinfo(skb);
1215
1216 /* This is a hint as to how much should be linear. */
1217 gso.hdr_len = skb_headlen(skb);
1218 gso.gso_size = sinfo->gso_size;
1219 if (sinfo->gso_type & SKB_GSO_TCPV4)
1220 gso.gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
1221 else if (sinfo->gso_type & SKB_GSO_TCPV6)
1222 gso.gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
1223 else if (sinfo->gso_type & SKB_GSO_UDP)
1224 gso.gso_type = VIRTIO_NET_HDR_GSO_UDP;
1225 else {
1226 pr_err("unexpected GSO type: "
1227 "0x%x, gso_size %d, hdr_len %d\n",
1228 sinfo->gso_type, gso.gso_size,
1229 gso.hdr_len);
1230 print_hex_dump(KERN_ERR, "tun: ",
1231 DUMP_PREFIX_NONE,
1232 16, 1, skb->head,
1233 min((int)gso.hdr_len, 64), true);
1234 WARN_ON_ONCE(1);
1235 return -EINVAL;
1236 }
1237 if (sinfo->gso_type & SKB_GSO_TCP_ECN)
1238 gso.gso_type |= VIRTIO_NET_HDR_GSO_ECN;
1239 } else
1240 gso.gso_type = VIRTIO_NET_HDR_GSO_NONE;
1241
1242 if (skb->ip_summed == CHECKSUM_PARTIAL) {
1243 gso.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
1244 gso.csum_start = skb_checksum_start_offset(skb) +
1245 vlan_hlen;
1246 gso.csum_offset = skb->csum_offset;
1247 } else if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
1248 gso.flags = VIRTIO_NET_HDR_F_DATA_VALID;
1249 } /* else everything is zero */
1250
1251 if (unlikely(memcpy_toiovecend(iv, (void *)&gso, total,
1252 sizeof(gso))))
1253 return -EFAULT;
1254 total += tun->vnet_hdr_sz;
1255 }
1256
1257 copied = total;
1258 len = min_t(int, skb->len + vlan_hlen, len);
1259 total += skb->len + vlan_hlen;
1260 if (vlan_hlen) {
1261 int copy, ret;
1262 struct {
1263 __be16 h_vlan_proto;
1264 __be16 h_vlan_TCI;
1265 } veth;
1266
1267 veth.h_vlan_proto = skb->vlan_proto;
1268 veth.h_vlan_TCI = htons(vlan_tx_tag_get(skb));
1269
1270 vlan_offset = offsetof(struct vlan_ethhdr, h_vlan_proto);
1271
1272 copy = min_t(int, vlan_offset, len);
1273 ret = skb_copy_datagram_const_iovec(skb, 0, iv, copied, copy);
1274 len -= copy;
1275 copied += copy;
1276 if (ret || !len)
1277 goto done;
1278
1279 copy = min_t(int, sizeof(veth), len);
1280 ret = memcpy_toiovecend(iv, (void *)&veth, copied, copy);
1281 len -= copy;
1282 copied += copy;
1283 if (ret || !len)
1284 goto done;
1285 }
1286
1287 skb_copy_datagram_const_iovec(skb, vlan_offset, iv, copied, len);
1288
1289 done:
1290 tun->dev->stats.tx_packets++;
1291 tun->dev->stats.tx_bytes += len;
1292
1293 return total;
1294 }
1295
1296 static ssize_t tun_do_read(struct tun_struct *tun, struct tun_file *tfile,
1297 struct kiocb *iocb, const struct iovec *iv,
1298 ssize_t len, int noblock)
1299 {
1300 DECLARE_WAITQUEUE(wait, current);
1301 struct sk_buff *skb;
1302 ssize_t ret = 0;
1303
1304 tun_debug(KERN_INFO, tun, "tun_do_read\n");
1305
1306 if (unlikely(!noblock))
1307 add_wait_queue(&tfile->wq.wait, &wait);
1308 while (len) {
1309 if (unlikely(!noblock))
1310 current->state = TASK_INTERRUPTIBLE;
1311
1312 /* Read frames from the queue */
1313 if (!(skb = skb_dequeue(&tfile->socket.sk->sk_receive_queue))) {
1314 if (noblock) {
1315 ret = -EAGAIN;
1316 break;
1317 }
1318 if (signal_pending(current)) {
1319 ret = -ERESTARTSYS;
1320 break;
1321 }
1322 if (tun->dev->reg_state != NETREG_REGISTERED) {
1323 ret = -EIO;
1324 break;
1325 }
1326
1327 /* Nothing to read, let's sleep */
1328 schedule();
1329 continue;
1330 }
1331
1332 ret = tun_put_user(tun, tfile, skb, iv, len);
1333 kfree_skb(skb);
1334 break;
1335 }
1336
1337 if (unlikely(!noblock)) {
1338 current->state = TASK_RUNNING;
1339 remove_wait_queue(&tfile->wq.wait, &wait);
1340 }
1341
1342 return ret;
1343 }
1344
1345 static ssize_t tun_chr_aio_read(struct kiocb *iocb, const struct iovec *iv,
1346 unsigned long count, loff_t pos)
1347 {
1348 struct file *file = iocb->ki_filp;
1349 struct tun_file *tfile = file->private_data;
1350 struct tun_struct *tun = __tun_get(tfile);
1351 ssize_t len, ret;
1352
1353 if (!tun)
1354 return -EBADFD;
1355 len = iov_length(iv, count);
1356 if (len < 0) {
1357 ret = -EINVAL;
1358 goto out;
1359 }
1360
1361 ret = tun_do_read(tun, tfile, iocb, iv, len,
1362 file->f_flags & O_NONBLOCK);
1363 ret = min_t(ssize_t, ret, len);
1364 if (ret > 0)
1365 iocb->ki_pos = ret;
1366 out:
1367 tun_put(tun);
1368 return ret;
1369 }
1370
1371 static void tun_free_netdev(struct net_device *dev)
1372 {
1373 struct tun_struct *tun = netdev_priv(dev);
1374
1375 BUG_ON(!(list_empty(&tun->disabled)));
1376 tun_flow_uninit(tun);
1377 security_tun_dev_free_security(tun->security);
1378 free_netdev(dev);
1379 }
1380
1381 static void tun_setup(struct net_device *dev)
1382 {
1383 struct tun_struct *tun = netdev_priv(dev);
1384
1385 tun->owner = INVALID_UID;
1386 tun->group = INVALID_GID;
1387
1388 dev->ethtool_ops = &tun_ethtool_ops;
1389 dev->destructor = tun_free_netdev;
1390 }
1391
1392 /* Trivial set of netlink ops to allow deleting tun or tap
1393 * device with netlink.
1394 */
1395 static int tun_validate(struct nlattr *tb[], struct nlattr *data[])
1396 {
1397 return -EINVAL;
1398 }
1399
1400 static struct rtnl_link_ops tun_link_ops __read_mostly = {
1401 .kind = DRV_NAME,
1402 .priv_size = sizeof(struct tun_struct),
1403 .setup = tun_setup,
1404 .validate = tun_validate,
1405 };
1406
1407 static void tun_sock_write_space(struct sock *sk)
1408 {
1409 struct tun_file *tfile;
1410 wait_queue_head_t *wqueue;
1411
1412 if (!sock_writeable(sk))
1413 return;
1414
1415 if (!test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags))
1416 return;
1417
1418 wqueue = sk_sleep(sk);
1419 if (wqueue && waitqueue_active(wqueue))
1420 wake_up_interruptible_sync_poll(wqueue, POLLOUT |
1421 POLLWRNORM | POLLWRBAND);
1422
1423 tfile = container_of(sk, struct tun_file, sk);
1424 kill_fasync(&tfile->fasync, SIGIO, POLL_OUT);
1425 }
1426
1427 static int tun_sendmsg(struct kiocb *iocb, struct socket *sock,
1428 struct msghdr *m, size_t total_len)
1429 {
1430 int ret;
1431 struct tun_file *tfile = container_of(sock, struct tun_file, socket);
1432 struct tun_struct *tun = __tun_get(tfile);
1433
1434 if (!tun)
1435 return -EBADFD;
1436 ret = tun_get_user(tun, tfile, m->msg_control, m->msg_iov, total_len,
1437 m->msg_iovlen, m->msg_flags & MSG_DONTWAIT);
1438 tun_put(tun);
1439 return ret;
1440 }
1441
1442 static int tun_recvmsg(struct kiocb *iocb, struct socket *sock,
1443 struct msghdr *m, size_t total_len,
1444 int flags)
1445 {
1446 struct tun_file *tfile = container_of(sock, struct tun_file, socket);
1447 struct tun_struct *tun = __tun_get(tfile);
1448 int ret;
1449
1450 if (!tun)
1451 return -EBADFD;
1452
1453 if (flags & ~(MSG_DONTWAIT|MSG_TRUNC|MSG_ERRQUEUE)) {
1454 ret = -EINVAL;
1455 goto out;
1456 }
1457 if (flags & MSG_ERRQUEUE) {
1458 ret = sock_recv_errqueue(sock->sk, m, total_len,
1459 SOL_PACKET, TUN_TX_TIMESTAMP);
1460 goto out;
1461 }
1462 ret = tun_do_read(tun, tfile, iocb, m->msg_iov, total_len,
1463 flags & MSG_DONTWAIT);
1464 if (ret > total_len) {
1465 m->msg_flags |= MSG_TRUNC;
1466 ret = flags & MSG_TRUNC ? ret : total_len;
1467 }
1468 out:
1469 tun_put(tun);
1470 return ret;
1471 }
1472
1473 static int tun_release(struct socket *sock)
1474 {
1475 if (sock->sk)
1476 sock_put(sock->sk);
1477 return 0;
1478 }
1479
1480 /* Ops structure to mimic raw sockets with tun */
1481 static const struct proto_ops tun_socket_ops = {
1482 .sendmsg = tun_sendmsg,
1483 .recvmsg = tun_recvmsg,
1484 .release = tun_release,
1485 };
1486
1487 static struct proto tun_proto = {
1488 .name = "tun",
1489 .owner = THIS_MODULE,
1490 .obj_size = sizeof(struct tun_file),
1491 };
1492
1493 static int tun_flags(struct tun_struct *tun)
1494 {
1495 int flags = 0;
1496
1497 if (tun->flags & TUN_TUN_DEV)
1498 flags |= IFF_TUN;
1499 else
1500 flags |= IFF_TAP;
1501
1502 if (tun->flags & TUN_NO_PI)
1503 flags |= IFF_NO_PI;
1504
1505 /* This flag has no real effect. We track the value for backwards
1506 * compatibility.
1507 */
1508 if (tun->flags & TUN_ONE_QUEUE)
1509 flags |= IFF_ONE_QUEUE;
1510
1511 if (tun->flags & TUN_VNET_HDR)
1512 flags |= IFF_VNET_HDR;
1513
1514 if (tun->flags & TUN_TAP_MQ)
1515 flags |= IFF_MULTI_QUEUE;
1516
1517 if (tun->flags & TUN_PERSIST)
1518 flags |= IFF_PERSIST;
1519
1520 return flags;
1521 }
1522
1523 static ssize_t tun_show_flags(struct device *dev, struct device_attribute *attr,
1524 char *buf)
1525 {
1526 struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1527 return sprintf(buf, "0x%x\n", tun_flags(tun));
1528 }
1529
1530 static ssize_t tun_show_owner(struct device *dev, struct device_attribute *attr,
1531 char *buf)
1532 {
1533 struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1534 return uid_valid(tun->owner)?
1535 sprintf(buf, "%u\n",
1536 from_kuid_munged(current_user_ns(), tun->owner)):
1537 sprintf(buf, "-1\n");
1538 }
1539
1540 static ssize_t tun_show_group(struct device *dev, struct device_attribute *attr,
1541 char *buf)
1542 {
1543 struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1544 return gid_valid(tun->group) ?
1545 sprintf(buf, "%u\n",
1546 from_kgid_munged(current_user_ns(), tun->group)):
1547 sprintf(buf, "-1\n");
1548 }
1549
1550 static DEVICE_ATTR(tun_flags, 0444, tun_show_flags, NULL);
1551 static DEVICE_ATTR(owner, 0444, tun_show_owner, NULL);
1552 static DEVICE_ATTR(group, 0444, tun_show_group, NULL);
1553
1554 static int tun_set_iff(struct net *net, struct file *file, struct ifreq *ifr)
1555 {
1556 struct tun_struct *tun;
1557 struct tun_file *tfile = file->private_data;
1558 struct net_device *dev;
1559 int err;
1560
1561 if (tfile->detached)
1562 return -EINVAL;
1563
1564 dev = __dev_get_by_name(net, ifr->ifr_name);
1565 if (dev) {
1566 if (ifr->ifr_flags & IFF_TUN_EXCL)
1567 return -EBUSY;
1568 if ((ifr->ifr_flags & IFF_TUN) && dev->netdev_ops == &tun_netdev_ops)
1569 tun = netdev_priv(dev);
1570 else if ((ifr->ifr_flags & IFF_TAP) && dev->netdev_ops == &tap_netdev_ops)
1571 tun = netdev_priv(dev);
1572 else
1573 return -EINVAL;
1574
1575 if (!!(ifr->ifr_flags & IFF_MULTI_QUEUE) !=
1576 !!(tun->flags & TUN_TAP_MQ))
1577 return -EINVAL;
1578
1579 if (tun_not_capable(tun))
1580 return -EPERM;
1581 err = security_tun_dev_open(tun->security);
1582 if (err < 0)
1583 return err;
1584
1585 err = tun_attach(tun, file, ifr->ifr_flags & IFF_NOFILTER);
1586 if (err < 0)
1587 return err;
1588
1589 if (tun->flags & TUN_TAP_MQ &&
1590 (tun->numqueues + tun->numdisabled > 1)) {
1591 /* One or more queue has already been attached, no need
1592 * to initialize the device again.
1593 */
1594 return 0;
1595 }
1596 }
1597 else {
1598 char *name;
1599 unsigned long flags = 0;
1600 int queues = ifr->ifr_flags & IFF_MULTI_QUEUE ?
1601 MAX_TAP_QUEUES : 1;
1602
1603 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
1604 return -EPERM;
1605 err = security_tun_dev_create();
1606 if (err < 0)
1607 return err;
1608
1609 /* Set dev type */
1610 if (ifr->ifr_flags & IFF_TUN) {
1611 /* TUN device */
1612 flags |= TUN_TUN_DEV;
1613 name = "tun%d";
1614 } else if (ifr->ifr_flags & IFF_TAP) {
1615 /* TAP device */
1616 flags |= TUN_TAP_DEV;
1617 name = "tap%d";
1618 } else
1619 return -EINVAL;
1620
1621 if (*ifr->ifr_name)
1622 name = ifr->ifr_name;
1623
1624 dev = alloc_netdev_mqs(sizeof(struct tun_struct), name,
1625 tun_setup, queues, queues);
1626
1627 if (!dev)
1628 return -ENOMEM;
1629
1630 dev_net_set(dev, net);
1631 dev->rtnl_link_ops = &tun_link_ops;
1632 dev->ifindex = tfile->ifindex;
1633
1634 tun = netdev_priv(dev);
1635 tun->dev = dev;
1636 tun->flags = flags;
1637 tun->txflt.count = 0;
1638 tun->vnet_hdr_sz = sizeof(struct virtio_net_hdr);
1639
1640 tun->filter_attached = false;
1641 tun->sndbuf = tfile->socket.sk->sk_sndbuf;
1642
1643 spin_lock_init(&tun->lock);
1644
1645 err = security_tun_dev_alloc_security(&tun->security);
1646 if (err < 0)
1647 goto err_free_dev;
1648
1649 tun_net_init(dev);
1650 tun_flow_init(tun);
1651
1652 dev->hw_features = NETIF_F_SG | NETIF_F_FRAGLIST |
1653 TUN_USER_FEATURES | NETIF_F_HW_VLAN_CTAG_TX |
1654 NETIF_F_HW_VLAN_STAG_TX;
1655 dev->features = dev->hw_features;
1656 dev->vlan_features = dev->features &
1657 ~(NETIF_F_HW_VLAN_CTAG_TX |
1658 NETIF_F_HW_VLAN_STAG_TX);
1659
1660 INIT_LIST_HEAD(&tun->disabled);
1661 err = tun_attach(tun, file, false);
1662 if (err < 0)
1663 goto err_free_flow;
1664
1665 err = register_netdevice(tun->dev);
1666 if (err < 0)
1667 goto err_detach;
1668
1669 if (device_create_file(&tun->dev->dev, &dev_attr_tun_flags) ||
1670 device_create_file(&tun->dev->dev, &dev_attr_owner) ||
1671 device_create_file(&tun->dev->dev, &dev_attr_group))
1672 pr_err("Failed to create tun sysfs files\n");
1673 }
1674
1675 netif_carrier_on(tun->dev);
1676
1677 tun_debug(KERN_INFO, tun, "tun_set_iff\n");
1678
1679 if (ifr->ifr_flags & IFF_NO_PI)
1680 tun->flags |= TUN_NO_PI;
1681 else
1682 tun->flags &= ~TUN_NO_PI;
1683
1684 /* This flag has no real effect. We track the value for backwards
1685 * compatibility.
1686 */
1687 if (ifr->ifr_flags & IFF_ONE_QUEUE)
1688 tun->flags |= TUN_ONE_QUEUE;
1689 else
1690 tun->flags &= ~TUN_ONE_QUEUE;
1691
1692 if (ifr->ifr_flags & IFF_VNET_HDR)
1693 tun->flags |= TUN_VNET_HDR;
1694 else
1695 tun->flags &= ~TUN_VNET_HDR;
1696
1697 if (ifr->ifr_flags & IFF_MULTI_QUEUE)
1698 tun->flags |= TUN_TAP_MQ;
1699 else
1700 tun->flags &= ~TUN_TAP_MQ;
1701
1702 /* Make sure persistent devices do not get stuck in
1703 * xoff state.
1704 */
1705 if (netif_running(tun->dev))
1706 netif_tx_wake_all_queues(tun->dev);
1707
1708 strcpy(ifr->ifr_name, tun->dev->name);
1709 return 0;
1710
1711 err_detach:
1712 tun_detach_all(dev);
1713 err_free_flow:
1714 tun_flow_uninit(tun);
1715 security_tun_dev_free_security(tun->security);
1716 err_free_dev:
1717 free_netdev(dev);
1718 return err;
1719 }
1720
1721 static void tun_get_iff(struct net *net, struct tun_struct *tun,
1722 struct ifreq *ifr)
1723 {
1724 tun_debug(KERN_INFO, tun, "tun_get_iff\n");
1725
1726 strcpy(ifr->ifr_name, tun->dev->name);
1727
1728 ifr->ifr_flags = tun_flags(tun);
1729
1730 }
1731
1732 /* This is like a cut-down ethtool ops, except done via tun fd so no
1733 * privs required. */
1734 static int set_offload(struct tun_struct *tun, unsigned long arg)
1735 {
1736 netdev_features_t features = 0;
1737
1738 if (arg & TUN_F_CSUM) {
1739 features |= NETIF_F_HW_CSUM;
1740 arg &= ~TUN_F_CSUM;
1741
1742 if (arg & (TUN_F_TSO4|TUN_F_TSO6)) {
1743 if (arg & TUN_F_TSO_ECN) {
1744 features |= NETIF_F_TSO_ECN;
1745 arg &= ~TUN_F_TSO_ECN;
1746 }
1747 if (arg & TUN_F_TSO4)
1748 features |= NETIF_F_TSO;
1749 if (arg & TUN_F_TSO6)
1750 features |= NETIF_F_TSO6;
1751 arg &= ~(TUN_F_TSO4|TUN_F_TSO6);
1752 }
1753
1754 if (arg & TUN_F_UFO) {
1755 features |= NETIF_F_UFO;
1756 arg &= ~TUN_F_UFO;
1757 }
1758 }
1759
1760 /* This gives the user a way to test for new features in future by
1761 * trying to set them. */
1762 if (arg)
1763 return -EINVAL;
1764
1765 tun->set_features = features;
1766 netdev_update_features(tun->dev);
1767
1768 return 0;
1769 }
1770
1771 static void tun_detach_filter(struct tun_struct *tun, int n)
1772 {
1773 int i;
1774 struct tun_file *tfile;
1775
1776 for (i = 0; i < n; i++) {
1777 tfile = rtnl_dereference(tun->tfiles[i]);
1778 sk_detach_filter(tfile->socket.sk);
1779 }
1780
1781 tun->filter_attached = false;
1782 }
1783
1784 static int tun_attach_filter(struct tun_struct *tun)
1785 {
1786 int i, ret = 0;
1787 struct tun_file *tfile;
1788
1789 for (i = 0; i < tun->numqueues; i++) {
1790 tfile = rtnl_dereference(tun->tfiles[i]);
1791 ret = sk_attach_filter(&tun->fprog, tfile->socket.sk);
1792 if (ret) {
1793 tun_detach_filter(tun, i);
1794 return ret;
1795 }
1796 }
1797
1798 tun->filter_attached = true;
1799 return ret;
1800 }
1801
1802 static void tun_set_sndbuf(struct tun_struct *tun)
1803 {
1804 struct tun_file *tfile;
1805 int i;
1806
1807 for (i = 0; i < tun->numqueues; i++) {
1808 tfile = rtnl_dereference(tun->tfiles[i]);
1809 tfile->socket.sk->sk_sndbuf = tun->sndbuf;
1810 }
1811 }
1812
1813 static int tun_set_queue(struct file *file, struct ifreq *ifr)
1814 {
1815 struct tun_file *tfile = file->private_data;
1816 struct tun_struct *tun;
1817 int ret = 0;
1818
1819 rtnl_lock();
1820
1821 if (ifr->ifr_flags & IFF_ATTACH_QUEUE) {
1822 tun = tfile->detached;
1823 if (!tun) {
1824 ret = -EINVAL;
1825 goto unlock;
1826 }
1827 ret = security_tun_dev_attach_queue(tun->security);
1828 if (ret < 0)
1829 goto unlock;
1830 ret = tun_attach(tun, file, false);
1831 } else if (ifr->ifr_flags & IFF_DETACH_QUEUE) {
1832 tun = rtnl_dereference(tfile->tun);
1833 if (!tun || !(tun->flags & TUN_TAP_MQ) || tfile->detached)
1834 ret = -EINVAL;
1835 else
1836 __tun_detach(tfile, false);
1837 } else
1838 ret = -EINVAL;
1839
1840 unlock:
1841 rtnl_unlock();
1842 return ret;
1843 }
1844
1845 static long __tun_chr_ioctl(struct file *file, unsigned int cmd,
1846 unsigned long arg, int ifreq_len)
1847 {
1848 struct tun_file *tfile = file->private_data;
1849 struct tun_struct *tun;
1850 void __user* argp = (void __user*)arg;
1851 struct ifreq ifr;
1852 kuid_t owner;
1853 kgid_t group;
1854 int sndbuf;
1855 int vnet_hdr_sz;
1856 unsigned int ifindex;
1857 int ret;
1858
1859 if (cmd == TUNSETIFF || cmd == TUNSETQUEUE || _IOC_TYPE(cmd) == 0x89) {
1860 if (copy_from_user(&ifr, argp, ifreq_len))
1861 return -EFAULT;
1862 } else {
1863 memset(&ifr, 0, sizeof(ifr));
1864 }
1865 if (cmd == TUNGETFEATURES) {
1866 /* Currently this just means: "what IFF flags are valid?".
1867 * This is needed because we never checked for invalid flags on
1868 * TUNSETIFF. */
1869 return put_user(IFF_TUN | IFF_TAP | IFF_NO_PI | IFF_ONE_QUEUE |
1870 IFF_VNET_HDR | IFF_MULTI_QUEUE,
1871 (unsigned int __user*)argp);
1872 } else if (cmd == TUNSETQUEUE)
1873 return tun_set_queue(file, &ifr);
1874
1875 ret = 0;
1876 rtnl_lock();
1877
1878 tun = __tun_get(tfile);
1879 if (cmd == TUNSETIFF && !tun) {
1880 ifr.ifr_name[IFNAMSIZ-1] = '\0';
1881
1882 ret = tun_set_iff(tfile->net, file, &ifr);
1883
1884 if (ret)
1885 goto unlock;
1886
1887 if (copy_to_user(argp, &ifr, ifreq_len))
1888 ret = -EFAULT;
1889 goto unlock;
1890 }
1891 if (cmd == TUNSETIFINDEX) {
1892 ret = -EPERM;
1893 if (tun)
1894 goto unlock;
1895
1896 ret = -EFAULT;
1897 if (copy_from_user(&ifindex, argp, sizeof(ifindex)))
1898 goto unlock;
1899
1900 ret = 0;
1901 tfile->ifindex = ifindex;
1902 goto unlock;
1903 }
1904
1905 ret = -EBADFD;
1906 if (!tun)
1907 goto unlock;
1908
1909 tun_debug(KERN_INFO, tun, "tun_chr_ioctl cmd %u\n", cmd);
1910
1911 ret = 0;
1912 switch (cmd) {
1913 case TUNGETIFF:
1914 tun_get_iff(current->nsproxy->net_ns, tun, &ifr);
1915
1916 if (tfile->detached)
1917 ifr.ifr_flags |= IFF_DETACH_QUEUE;
1918 if (!tfile->socket.sk->sk_filter)
1919 ifr.ifr_flags |= IFF_NOFILTER;
1920
1921 if (copy_to_user(argp, &ifr, ifreq_len))
1922 ret = -EFAULT;
1923 break;
1924
1925 case TUNSETNOCSUM:
1926 /* Disable/Enable checksum */
1927
1928 /* [unimplemented] */
1929 tun_debug(KERN_INFO, tun, "ignored: set checksum %s\n",
1930 arg ? "disabled" : "enabled");
1931 break;
1932
1933 case TUNSETPERSIST:
1934 /* Disable/Enable persist mode. Keep an extra reference to the
1935 * module to prevent the module being unprobed.
1936 */
1937 if (arg && !(tun->flags & TUN_PERSIST)) {
1938 tun->flags |= TUN_PERSIST;
1939 __module_get(THIS_MODULE);
1940 }
1941 if (!arg && (tun->flags & TUN_PERSIST)) {
1942 tun->flags &= ~TUN_PERSIST;
1943 module_put(THIS_MODULE);
1944 }
1945
1946 tun_debug(KERN_INFO, tun, "persist %s\n",
1947 arg ? "enabled" : "disabled");
1948 break;
1949
1950 case TUNSETOWNER:
1951 /* Set owner of the device */
1952 owner = make_kuid(current_user_ns(), arg);
1953 if (!uid_valid(owner)) {
1954 ret = -EINVAL;
1955 break;
1956 }
1957 tun->owner = owner;
1958 tun_debug(KERN_INFO, tun, "owner set to %u\n",
1959 from_kuid(&init_user_ns, tun->owner));
1960 break;
1961
1962 case TUNSETGROUP:
1963 /* Set group of the device */
1964 group = make_kgid(current_user_ns(), arg);
1965 if (!gid_valid(group)) {
1966 ret = -EINVAL;
1967 break;
1968 }
1969 tun->group = group;
1970 tun_debug(KERN_INFO, tun, "group set to %u\n",
1971 from_kgid(&init_user_ns, tun->group));
1972 break;
1973
1974 case TUNSETLINK:
1975 /* Only allow setting the type when the interface is down */
1976 if (tun->dev->flags & IFF_UP) {
1977 tun_debug(KERN_INFO, tun,
1978 "Linktype set failed because interface is up\n");
1979 ret = -EBUSY;
1980 } else {
1981 tun->dev->type = (int) arg;
1982 tun_debug(KERN_INFO, tun, "linktype set to %d\n",
1983 tun->dev->type);
1984 ret = 0;
1985 }
1986 break;
1987
1988 #ifdef TUN_DEBUG
1989 case TUNSETDEBUG:
1990 tun->debug = arg;
1991 break;
1992 #endif
1993 case TUNSETOFFLOAD:
1994 ret = set_offload(tun, arg);
1995 break;
1996
1997 case TUNSETTXFILTER:
1998 /* Can be set only for TAPs */
1999 ret = -EINVAL;
2000 if ((tun->flags & TUN_TYPE_MASK) != TUN_TAP_DEV)
2001 break;
2002 ret = update_filter(&tun->txflt, (void __user *)arg);
2003 break;
2004
2005 case SIOCGIFHWADDR:
2006 /* Get hw address */
2007 memcpy(ifr.ifr_hwaddr.sa_data, tun->dev->dev_addr, ETH_ALEN);
2008 ifr.ifr_hwaddr.sa_family = tun->dev->type;
2009 if (copy_to_user(argp, &ifr, ifreq_len))
2010 ret = -EFAULT;
2011 break;
2012
2013 case SIOCSIFHWADDR:
2014 /* Set hw address */
2015 tun_debug(KERN_DEBUG, tun, "set hw address: %pM\n",
2016 ifr.ifr_hwaddr.sa_data);
2017
2018 ret = dev_set_mac_address(tun->dev, &ifr.ifr_hwaddr);
2019 break;
2020
2021 case TUNGETSNDBUF:
2022 sndbuf = tfile->socket.sk->sk_sndbuf;
2023 if (copy_to_user(argp, &sndbuf, sizeof(sndbuf)))
2024 ret = -EFAULT;
2025 break;
2026
2027 case TUNSETSNDBUF:
2028 if (copy_from_user(&sndbuf, argp, sizeof(sndbuf))) {
2029 ret = -EFAULT;
2030 break;
2031 }
2032
2033 tun->sndbuf = sndbuf;
2034 tun_set_sndbuf(tun);
2035 break;
2036
2037 case TUNGETVNETHDRSZ:
2038 vnet_hdr_sz = tun->vnet_hdr_sz;
2039 if (copy_to_user(argp, &vnet_hdr_sz, sizeof(vnet_hdr_sz)))
2040 ret = -EFAULT;
2041 break;
2042
2043 case TUNSETVNETHDRSZ:
2044 if (copy_from_user(&vnet_hdr_sz, argp, sizeof(vnet_hdr_sz))) {
2045 ret = -EFAULT;
2046 break;
2047 }
2048 if (vnet_hdr_sz < (int)sizeof(struct virtio_net_hdr)) {
2049 ret = -EINVAL;
2050 break;
2051 }
2052
2053 tun->vnet_hdr_sz = vnet_hdr_sz;
2054 break;
2055
2056 case TUNATTACHFILTER:
2057 /* Can be set only for TAPs */
2058 ret = -EINVAL;
2059 if ((tun->flags & TUN_TYPE_MASK) != TUN_TAP_DEV)
2060 break;
2061 ret = -EFAULT;
2062 if (copy_from_user(&tun->fprog, argp, sizeof(tun->fprog)))
2063 break;
2064
2065 ret = tun_attach_filter(tun);
2066 break;
2067
2068 case TUNDETACHFILTER:
2069 /* Can be set only for TAPs */
2070 ret = -EINVAL;
2071 if ((tun->flags & TUN_TYPE_MASK) != TUN_TAP_DEV)
2072 break;
2073 ret = 0;
2074 tun_detach_filter(tun, tun->numqueues);
2075 break;
2076
2077 case TUNGETFILTER:
2078 ret = -EINVAL;
2079 if ((tun->flags & TUN_TYPE_MASK) != TUN_TAP_DEV)
2080 break;
2081 ret = -EFAULT;
2082 if (copy_to_user(argp, &tun->fprog, sizeof(tun->fprog)))
2083 break;
2084 ret = 0;
2085 break;
2086
2087 default:
2088 ret = -EINVAL;
2089 break;
2090 }
2091
2092 unlock:
2093 rtnl_unlock();
2094 if (tun)
2095 tun_put(tun);
2096 return ret;
2097 }
2098
2099 static long tun_chr_ioctl(struct file *file,
2100 unsigned int cmd, unsigned long arg)
2101 {
2102 return __tun_chr_ioctl(file, cmd, arg, sizeof (struct ifreq));
2103 }
2104
2105 #ifdef CONFIG_COMPAT
2106 static long tun_chr_compat_ioctl(struct file *file,
2107 unsigned int cmd, unsigned long arg)
2108 {
2109 switch (cmd) {
2110 case TUNSETIFF:
2111 case TUNGETIFF:
2112 case TUNSETTXFILTER:
2113 case TUNGETSNDBUF:
2114 case TUNSETSNDBUF:
2115 case SIOCGIFHWADDR:
2116 case SIOCSIFHWADDR:
2117 arg = (unsigned long)compat_ptr(arg);
2118 break;
2119 default:
2120 arg = (compat_ulong_t)arg;
2121 break;
2122 }
2123
2124 /*
2125 * compat_ifreq is shorter than ifreq, so we must not access beyond
2126 * the end of that structure. All fields that are used in this
2127 * driver are compatible though, we don't need to convert the
2128 * contents.
2129 */
2130 return __tun_chr_ioctl(file, cmd, arg, sizeof(struct compat_ifreq));
2131 }
2132 #endif /* CONFIG_COMPAT */
2133
2134 static int tun_chr_fasync(int fd, struct file *file, int on)
2135 {
2136 struct tun_file *tfile = file->private_data;
2137 int ret;
2138
2139 if ((ret = fasync_helper(fd, file, on, &tfile->fasync)) < 0)
2140 goto out;
2141
2142 if (on) {
2143 ret = __f_setown(file, task_pid(current), PIDTYPE_PID, 0);
2144 if (ret)
2145 goto out;
2146 tfile->flags |= TUN_FASYNC;
2147 } else
2148 tfile->flags &= ~TUN_FASYNC;
2149 ret = 0;
2150 out:
2151 return ret;
2152 }
2153
2154 static int tun_chr_open(struct inode *inode, struct file * file)
2155 {
2156 struct tun_file *tfile;
2157
2158 DBG1(KERN_INFO, "tunX: tun_chr_open\n");
2159
2160 tfile = (struct tun_file *)sk_alloc(&init_net, AF_UNSPEC, GFP_KERNEL,
2161 &tun_proto);
2162 if (!tfile)
2163 return -ENOMEM;
2164 rcu_assign_pointer(tfile->tun, NULL);
2165 tfile->net = get_net(current->nsproxy->net_ns);
2166 tfile->flags = 0;
2167 tfile->ifindex = 0;
2168
2169 rcu_assign_pointer(tfile->socket.wq, &tfile->wq);
2170 init_waitqueue_head(&tfile->wq.wait);
2171
2172 tfile->socket.file = file;
2173 tfile->socket.ops = &tun_socket_ops;
2174
2175 sock_init_data(&tfile->socket, &tfile->sk);
2176 sk_change_net(&tfile->sk, tfile->net);
2177
2178 tfile->sk.sk_write_space = tun_sock_write_space;
2179 tfile->sk.sk_sndbuf = INT_MAX;
2180
2181 file->private_data = tfile;
2182 set_bit(SOCK_EXTERNALLY_ALLOCATED, &tfile->socket.flags);
2183 INIT_LIST_HEAD(&tfile->next);
2184
2185 sock_set_flag(&tfile->sk, SOCK_ZEROCOPY);
2186
2187 return 0;
2188 }
2189
2190 static int tun_chr_close(struct inode *inode, struct file *file)
2191 {
2192 struct tun_file *tfile = file->private_data;
2193 struct net *net = tfile->net;
2194
2195 tun_detach(tfile, true);
2196 put_net(net);
2197
2198 return 0;
2199 }
2200
2201 static const struct file_operations tun_fops = {
2202 .owner = THIS_MODULE,
2203 .llseek = no_llseek,
2204 .read = do_sync_read,
2205 .aio_read = tun_chr_aio_read,
2206 .write = do_sync_write,
2207 .aio_write = tun_chr_aio_write,
2208 .poll = tun_chr_poll,
2209 .unlocked_ioctl = tun_chr_ioctl,
2210 #ifdef CONFIG_COMPAT
2211 .compat_ioctl = tun_chr_compat_ioctl,
2212 #endif
2213 .open = tun_chr_open,
2214 .release = tun_chr_close,
2215 .fasync = tun_chr_fasync
2216 };
2217
2218 static struct miscdevice tun_miscdev = {
2219 .minor = TUN_MINOR,
2220 .name = "tun",
2221 .nodename = "net/tun",
2222 .fops = &tun_fops,
2223 };
2224
2225 /* ethtool interface */
2226
2227 static int tun_get_settings(struct net_device *dev, struct ethtool_cmd *cmd)
2228 {
2229 cmd->supported = 0;
2230 cmd->advertising = 0;
2231 ethtool_cmd_speed_set(cmd, SPEED_10);
2232 cmd->duplex = DUPLEX_FULL;
2233 cmd->port = PORT_TP;
2234 cmd->phy_address = 0;
2235 cmd->transceiver = XCVR_INTERNAL;
2236 cmd->autoneg = AUTONEG_DISABLE;
2237 cmd->maxtxpkt = 0;
2238 cmd->maxrxpkt = 0;
2239 return 0;
2240 }
2241
2242 static void tun_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
2243 {
2244 struct tun_struct *tun = netdev_priv(dev);
2245
2246 strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
2247 strlcpy(info->version, DRV_VERSION, sizeof(info->version));
2248
2249 switch (tun->flags & TUN_TYPE_MASK) {
2250 case TUN_TUN_DEV:
2251 strlcpy(info->bus_info, "tun", sizeof(info->bus_info));
2252 break;
2253 case TUN_TAP_DEV:
2254 strlcpy(info->bus_info, "tap", sizeof(info->bus_info));
2255 break;
2256 }
2257 }
2258
2259 static u32 tun_get_msglevel(struct net_device *dev)
2260 {
2261 #ifdef TUN_DEBUG
2262 struct tun_struct *tun = netdev_priv(dev);
2263 return tun->debug;
2264 #else
2265 return -EOPNOTSUPP;
2266 #endif
2267 }
2268
2269 static void tun_set_msglevel(struct net_device *dev, u32 value)
2270 {
2271 #ifdef TUN_DEBUG
2272 struct tun_struct *tun = netdev_priv(dev);
2273 tun->debug = value;
2274 #endif
2275 }
2276
2277 static const struct ethtool_ops tun_ethtool_ops = {
2278 .get_settings = tun_get_settings,
2279 .get_drvinfo = tun_get_drvinfo,
2280 .get_msglevel = tun_get_msglevel,
2281 .set_msglevel = tun_set_msglevel,
2282 .get_link = ethtool_op_get_link,
2283 .get_ts_info = ethtool_op_get_ts_info,
2284 };
2285
2286
2287 static int __init tun_init(void)
2288 {
2289 int ret = 0;
2290
2291 pr_info("%s, %s\n", DRV_DESCRIPTION, DRV_VERSION);
2292 pr_info("%s\n", DRV_COPYRIGHT);
2293
2294 ret = rtnl_link_register(&tun_link_ops);
2295 if (ret) {
2296 pr_err("Can't register link_ops\n");
2297 goto err_linkops;
2298 }
2299
2300 ret = misc_register(&tun_miscdev);
2301 if (ret) {
2302 pr_err("Can't register misc device %d\n", TUN_MINOR);
2303 goto err_misc;
2304 }
2305 return 0;
2306 err_misc:
2307 rtnl_link_unregister(&tun_link_ops);
2308 err_linkops:
2309 return ret;
2310 }
2311
2312 static void tun_cleanup(void)
2313 {
2314 misc_deregister(&tun_miscdev);
2315 rtnl_link_unregister(&tun_link_ops);
2316 }
2317
2318 /* Get an underlying socket object from tun file. Returns error unless file is
2319 * attached to a device. The returned object works like a packet socket, it
2320 * can be used for sock_sendmsg/sock_recvmsg. The caller is responsible for
2321 * holding a reference to the file for as long as the socket is in use. */
2322 struct socket *tun_get_socket(struct file *file)
2323 {
2324 struct tun_file *tfile;
2325 if (file->f_op != &tun_fops)
2326 return ERR_PTR(-EINVAL);
2327 tfile = file->private_data;
2328 if (!tfile)
2329 return ERR_PTR(-EBADFD);
2330 return &tfile->socket;
2331 }
2332 EXPORT_SYMBOL_GPL(tun_get_socket);
2333
2334 module_init(tun_init);
2335 module_exit(tun_cleanup);
2336 MODULE_DESCRIPTION(DRV_DESCRIPTION);
2337 MODULE_AUTHOR(DRV_COPYRIGHT);
2338 MODULE_LICENSE("GPL");
2339 MODULE_ALIAS_MISCDEV(TUN_MINOR);
2340 MODULE_ALIAS("devname:net/tun");
Linux with added FPC-III support