?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
tuntap: initialize vlan_features
[linux/fpc-iii.git] / drivers / net / tun.c
blob316c759bd988e6d602e7ab7dae37022f46be9ffa
1 /*
2 * TUN - Universal TUN/TAP device driver.
3 * Copyright (C) 1999-2002 Maxim Krasnyansky <maxk@qualcomm.com>
4 *
5 * This program is free software; you can redistribute it and/or modify
6 * it under the terms of the GNU General Public License as published by
7 * the Free Software Foundation; either version 2 of the License, or
8 * (at your option) any later version.
9 *
10 * This program is distributed in the hope that it will be useful,
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
13 * GNU General Public License for more details.
14 *
15 * $Id: tun.c,v 1.15 2002/03/01 02:44:24 maxk Exp $
16 */
17
18 /*
19 * Changes:
20 *
21 * Mike Kershaw <dragorn@kismetwireless.net> 2005/08/14
22 * Add TUNSETLINK ioctl to set the link encapsulation
23 *
24 * Mark Smith <markzzzsmith@yahoo.com.au>
25 * Use eth_random_addr() for tap MAC address.
26 *
27 * Harald Roelle <harald.roelle@ifi.lmu.de> 2004/04/20
28 * Fixes in packet dropping, queue length setting and queue wakeup.
29 * Increased default tx queue length.
30 * Added ethtool API.
31 * Minor cleanups
32 *
33 * Daniel Podlejski <underley@underley.eu.org>
34 * Modifications for 2.3.99-pre5 kernel.
35 */
36
37 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
38
39 #define DRV_NAME "tun"
40 #define DRV_VERSION "1.6"
41 #define DRV_DESCRIPTION "Universal TUN/TAP device driver"
42 #define DRV_COPYRIGHT "(C) 1999-2004 Max Krasnyansky <maxk@qualcomm.com>"
43
44 #include <linux/module.h>
45 #include <linux/errno.h>
46 #include <linux/kernel.h>
47 #include <linux/major.h>
48 #include <linux/slab.h>
49 #include <linux/poll.h>
50 #include <linux/fcntl.h>
51 #include <linux/init.h>
52 #include <linux/skbuff.h>
53 #include <linux/netdevice.h>
54 #include <linux/etherdevice.h>
55 #include <linux/miscdevice.h>
56 #include <linux/ethtool.h>
57 #include <linux/rtnetlink.h>
58 #include <linux/compat.h>
59 #include <linux/if.h>
60 #include <linux/if_arp.h>
61 #include <linux/if_ether.h>
62 #include <linux/if_tun.h>
63 #include <linux/crc32.h>
64 #include <linux/nsproxy.h>
65 #include <linux/virtio_net.h>
66 #include <linux/rcupdate.h>
67 #include <net/net_namespace.h>
68 #include <net/netns/generic.h>
69 #include <net/rtnetlink.h>
70 #include <net/sock.h>
71
72 #include <asm/uaccess.h>
73
74 /* Uncomment to enable debugging */
75 /* #define TUN_DEBUG 1 */
76
77 #ifdef TUN_DEBUG
78 static int debug;
79
80 #define tun_debug(level, tun, fmt, args...) \
81 do { \
82 if (tun->debug) \
83 netdev_printk(level, tun->dev, fmt, ##args); \
84 } while (0)
85 #define DBG1(level, fmt, args...) \
86 do { \
87 if (debug == 2) \
88 printk(level fmt, ##args); \
89 } while (0)
90 #else
91 #define tun_debug(level, tun, fmt, args...) \
92 do { \
93 if (0) \
94 netdev_printk(level, tun->dev, fmt, ##args); \
95 } while (0)
96 #define DBG1(level, fmt, args...) \
97 do { \
98 if (0) \
99 printk(level fmt, ##args); \
100 } while (0)
101 #endif
102
103 #define GOODCOPY_LEN 128
104
105 #define FLT_EXACT_COUNT 8
106 struct tap_filter {
107 unsigned int count; /* Number of addrs. Zero means disabled */
108 u32 mask[2]; /* Mask of the hashed addrs */
109 unsigned char addr[FLT_EXACT_COUNT][ETH_ALEN];
110 };
111
112 /* DEFAULT_MAX_NUM_RSS_QUEUES were choosed to let the rx/tx queues allocated for
113 * the netdevice to be fit in one page. So we can make sure the success of
114 * memory allocation. TODO: increase the limit. */
115 #define MAX_TAP_QUEUES DEFAULT_MAX_NUM_RSS_QUEUES
116 #define MAX_TAP_FLOWS 4096
117
118 #define TUN_FLOW_EXPIRE (3 * HZ)
119
120 /* A tun_file connects an open character device to a tuntap netdevice. It
121 * also contains all socket related strctures (except sock_fprog and tap_filter)
122 * to serve as one transmit queue for tuntap device. The sock_fprog and
123 * tap_filter were kept in tun_struct since they were used for filtering for the
124 * netdevice not for a specific queue (at least I didn't see the requirement for
125 * this).
126 *
127 * RCU usage:
128 * The tun_file and tun_struct are loosely coupled, the pointer from one to the
129 * other can only be read while rcu_read_lock or rtnl_lock is held.
130 */
131 struct tun_file {
132 struct sock sk;
133 struct socket socket;
134 struct socket_wq wq;
135 struct tun_struct __rcu *tun;
136 struct net *net;
137 struct fasync_struct *fasync;
138 /* only used for fasnyc */
139 unsigned int flags;
140 u16 queue_index;
141 struct list_head next;
142 struct tun_struct *detached;
143 };
144
145 struct tun_flow_entry {
146 struct hlist_node hash_link;
147 struct rcu_head rcu;
148 struct tun_struct *tun;
149
150 u32 rxhash;
151 int queue_index;
152 unsigned long updated;
153 };
154
155 #define TUN_NUM_FLOW_ENTRIES 1024
156
157 /* Since the socket were moved to tun_file, to preserve the behavior of persist
158 * device, socket filter, sndbuf and vnet header size were restore when the
159 * file were attached to a persist device.
160 */
161 struct tun_struct {
162 struct tun_file __rcu *tfiles[MAX_TAP_QUEUES];
163 unsigned int numqueues;
164 unsigned int flags;
165 kuid_t owner;
166 kgid_t group;
167
168 struct net_device *dev;
169 netdev_features_t set_features;
170 #define TUN_USER_FEATURES (NETIF_F_HW_CSUM|NETIF_F_TSO_ECN|NETIF_F_TSO| \
171 NETIF_F_TSO6|NETIF_F_UFO)
172
173 int vnet_hdr_sz;
174 int sndbuf;
175 struct tap_filter txflt;
176 struct sock_fprog fprog;
177 /* protected by rtnl lock */
178 bool filter_attached;
179 #ifdef TUN_DEBUG
180 int debug;
181 #endif
182 spinlock_t lock;
183 struct hlist_head flows[TUN_NUM_FLOW_ENTRIES];
184 struct timer_list flow_gc_timer;
185 unsigned long ageing_time;
186 unsigned int numdisabled;
187 struct list_head disabled;
188 void *security;
189 u32 flow_count;
190 };
191
192 static inline u32 tun_hashfn(u32 rxhash)
193 {
194 return rxhash & 0x3ff;
195 }
196
197 static struct tun_flow_entry *tun_flow_find(struct hlist_head *head, u32 rxhash)
198 {
199 struct tun_flow_entry *e;
200
201 hlist_for_each_entry_rcu(e, head, hash_link) {
202 if (e->rxhash == rxhash)
203 return e;
204 }
205 return NULL;
206 }
207
208 static struct tun_flow_entry *tun_flow_create(struct tun_struct *tun,
209 struct hlist_head *head,
210 u32 rxhash, u16 queue_index)
211 {
212 struct tun_flow_entry *e = kmalloc(sizeof(*e), GFP_ATOMIC);
213
214 if (e) {
215 tun_debug(KERN_INFO, tun, "create flow: hash %u index %u\n",
216 rxhash, queue_index);
217 e->updated = jiffies;
218 e->rxhash = rxhash;
219 e->queue_index = queue_index;
220 e->tun = tun;
221 hlist_add_head_rcu(&e->hash_link, head);
222 ++tun->flow_count;
223 }
224 return e;
225 }
226
227 static void tun_flow_delete(struct tun_struct *tun, struct tun_flow_entry *e)
228 {
229 tun_debug(KERN_INFO, tun, "delete flow: hash %u index %u\n",
230 e->rxhash, e->queue_index);
231 hlist_del_rcu(&e->hash_link);
232 kfree_rcu(e, rcu);
233 --tun->flow_count;
234 }
235
236 static void tun_flow_flush(struct tun_struct *tun)
237 {
238 int i;
239
240 spin_lock_bh(&tun->lock);
241 for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
242 struct tun_flow_entry *e;
243 struct hlist_node *n;
244
245 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link)
246 tun_flow_delete(tun, e);
247 }
248 spin_unlock_bh(&tun->lock);
249 }
250
251 static void tun_flow_delete_by_queue(struct tun_struct *tun, u16 queue_index)
252 {
253 int i;
254
255 spin_lock_bh(&tun->lock);
256 for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
257 struct tun_flow_entry *e;
258 struct hlist_node *n;
259
260 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
261 if (e->queue_index == queue_index)
262 tun_flow_delete(tun, e);
263 }
264 }
265 spin_unlock_bh(&tun->lock);
266 }
267
268 static void tun_flow_cleanup(unsigned long data)
269 {
270 struct tun_struct *tun = (struct tun_struct *)data;
271 unsigned long delay = tun->ageing_time;
272 unsigned long next_timer = jiffies + delay;
273 unsigned long count = 0;
274 int i;
275
276 tun_debug(KERN_INFO, tun, "tun_flow_cleanup\n");
277
278 spin_lock_bh(&tun->lock);
279 for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
280 struct tun_flow_entry *e;
281 struct hlist_node *n;
282
283 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
284 unsigned long this_timer;
285 count++;
286 this_timer = e->updated + delay;
287 if (time_before_eq(this_timer, jiffies))
288 tun_flow_delete(tun, e);
289 else if (time_before(this_timer, next_timer))
290 next_timer = this_timer;
291 }
292 }
293
294 if (count)
295 mod_timer(&tun->flow_gc_timer, round_jiffies_up(next_timer));
296 spin_unlock_bh(&tun->lock);
297 }
298
299 static void tun_flow_update(struct tun_struct *tun, u32 rxhash,
300 struct tun_file *tfile)
301 {
302 struct hlist_head *head;
303 struct tun_flow_entry *e;
304 unsigned long delay = tun->ageing_time;
305 u16 queue_index = tfile->queue_index;
306
307 if (!rxhash)
308 return;
309 else
310 head = &tun->flows[tun_hashfn(rxhash)];
311
312 rcu_read_lock();
313
314 /* We may get a very small possibility of OOO during switching, not
315 * worth to optimize.*/
316 if (tun->numqueues == 1 || tfile->detached)
317 goto unlock;
318
319 e = tun_flow_find(head, rxhash);
320 if (likely(e)) {
321 /* TODO: keep queueing to old queue until it's empty? */
322 e->queue_index = queue_index;
323 e->updated = jiffies;
324 } else {
325 spin_lock_bh(&tun->lock);
326 if (!tun_flow_find(head, rxhash) &&
327 tun->flow_count < MAX_TAP_FLOWS)
328 tun_flow_create(tun, head, rxhash, queue_index);
329
330 if (!timer_pending(&tun->flow_gc_timer))
331 mod_timer(&tun->flow_gc_timer,
332 round_jiffies_up(jiffies + delay));
333 spin_unlock_bh(&tun->lock);
334 }
335
336 unlock:
337 rcu_read_unlock();
338 }
339
340 /* We try to identify a flow through its rxhash first. The reason that
341 * we do not check rxq no. is becuase some cards(e.g 82599), chooses
342 * the rxq based on the txq where the last packet of the flow comes. As
343 * the userspace application move between processors, we may get a
344 * different rxq no. here. If we could not get rxhash, then we would
345 * hope the rxq no. may help here.
346 */
347 static u16 tun_select_queue(struct net_device *dev, struct sk_buff *skb)
348 {
349 struct tun_struct *tun = netdev_priv(dev);
350 struct tun_flow_entry *e;
351 u32 txq = 0;
352 u32 numqueues = 0;
353
354 rcu_read_lock();
355 numqueues = tun->numqueues;
356
357 txq = skb_get_rxhash(skb);
358 if (txq) {
359 e = tun_flow_find(&tun->flows[tun_hashfn(txq)], txq);
360 if (e)
361 txq = e->queue_index;
362 else
363 /* use multiply and shift instead of expensive divide */
364 txq = ((u64)txq * numqueues) >> 32;
365 } else if (likely(skb_rx_queue_recorded(skb))) {
366 txq = skb_get_rx_queue(skb);
367 while (unlikely(txq >= numqueues))
368 txq -= numqueues;
369 }
370
371 rcu_read_unlock();
372 return txq;
373 }
374
375 static inline bool tun_not_capable(struct tun_struct *tun)
376 {
377 const struct cred *cred = current_cred();
378 struct net *net = dev_net(tun->dev);
379
380 return ((uid_valid(tun->owner) && !uid_eq(cred->euid, tun->owner)) ||
381 (gid_valid(tun->group) && !in_egroup_p(tun->group))) &&
382 !ns_capable(net->user_ns, CAP_NET_ADMIN);
383 }
384
385 static void tun_set_real_num_queues(struct tun_struct *tun)
386 {
387 netif_set_real_num_tx_queues(tun->dev, tun->numqueues);
388 netif_set_real_num_rx_queues(tun->dev, tun->numqueues);
389 }
390
391 static void tun_disable_queue(struct tun_struct *tun, struct tun_file *tfile)
392 {
393 tfile->detached = tun;
394 list_add_tail(&tfile->next, &tun->disabled);
395 ++tun->numdisabled;
396 }
397
398 static struct tun_struct *tun_enable_queue(struct tun_file *tfile)
399 {
400 struct tun_struct *tun = tfile->detached;
401
402 tfile->detached = NULL;
403 list_del_init(&tfile->next);
404 --tun->numdisabled;
405 return tun;
406 }
407
408 static void __tun_detach(struct tun_file *tfile, bool clean)
409 {
410 struct tun_file *ntfile;
411 struct tun_struct *tun;
412
413 tun = rtnl_dereference(tfile->tun);
414
415 if (tun && !tfile->detached) {
416 u16 index = tfile->queue_index;
417 BUG_ON(index >= tun->numqueues);
418
419 rcu_assign_pointer(tun->tfiles[index],
420 tun->tfiles[tun->numqueues - 1]);
421 ntfile = rtnl_dereference(tun->tfiles[index]);
422 ntfile->queue_index = index;
423
424 --tun->numqueues;
425 if (clean) {
426 rcu_assign_pointer(tfile->tun, NULL);
427 sock_put(&tfile->sk);
428 } else
429 tun_disable_queue(tun, tfile);
430
431 synchronize_net();
432 tun_flow_delete_by_queue(tun, tun->numqueues + 1);
433 /* Drop read queue */
434 skb_queue_purge(&tfile->sk.sk_receive_queue);
435 tun_set_real_num_queues(tun);
436 } else if (tfile->detached && clean) {
437 tun = tun_enable_queue(tfile);
438 sock_put(&tfile->sk);
439 }
440
441 if (clean) {
442 if (tun && tun->numqueues == 0 && tun->numdisabled == 0) {
443 netif_carrier_off(tun->dev);
444
445 if (!(tun->flags & TUN_PERSIST) &&
446 tun->dev->reg_state == NETREG_REGISTERED)
447 unregister_netdevice(tun->dev);
448 }
449
450 BUG_ON(!test_bit(SOCK_EXTERNALLY_ALLOCATED,
451 &tfile->socket.flags));
452 sk_release_kernel(&tfile->sk);
453 }
454 }
455
456 static void tun_detach(struct tun_file *tfile, bool clean)
457 {
458 rtnl_lock();
459 __tun_detach(tfile, clean);
460 rtnl_unlock();
461 }
462
463 static void tun_detach_all(struct net_device *dev)
464 {
465 struct tun_struct *tun = netdev_priv(dev);
466 struct tun_file *tfile, *tmp;
467 int i, n = tun->numqueues;
468
469 for (i = 0; i < n; i++) {
470 tfile = rtnl_dereference(tun->tfiles[i]);
471 BUG_ON(!tfile);
472 wake_up_all(&tfile->wq.wait);
473 rcu_assign_pointer(tfile->tun, NULL);
474 --tun->numqueues;
475 }
476 list_for_each_entry(tfile, &tun->disabled, next) {
477 wake_up_all(&tfile->wq.wait);
478 rcu_assign_pointer(tfile->tun, NULL);
479 }
480 BUG_ON(tun->numqueues != 0);
481
482 synchronize_net();
483 for (i = 0; i < n; i++) {
484 tfile = rtnl_dereference(tun->tfiles[i]);
485 /* Drop read queue */
486 skb_queue_purge(&tfile->sk.sk_receive_queue);
487 sock_put(&tfile->sk);
488 }
489 list_for_each_entry_safe(tfile, tmp, &tun->disabled, next) {
490 tun_enable_queue(tfile);
491 skb_queue_purge(&tfile->sk.sk_receive_queue);
492 sock_put(&tfile->sk);
493 }
494 BUG_ON(tun->numdisabled != 0);
495
496 if (tun->flags & TUN_PERSIST)
497 module_put(THIS_MODULE);
498 }
499
500 static int tun_attach(struct tun_struct *tun, struct file *file)
501 {
502 struct tun_file *tfile = file->private_data;
503 int err;
504
505 err = security_tun_dev_attach(tfile->socket.sk, tun->security);
506 if (err < 0)
507 goto out;
508
509 err = -EINVAL;
510 if (rtnl_dereference(tfile->tun) && !tfile->detached)
511 goto out;
512
513 err = -EBUSY;
514 if (!(tun->flags & TUN_TAP_MQ) && tun->numqueues == 1)
515 goto out;
516
517 err = -E2BIG;
518 if (!tfile->detached &&
519 tun->numqueues + tun->numdisabled == MAX_TAP_QUEUES)
520 goto out;
521
522 err = 0;
523
524 /* Re-attach the filter to presist device */
525 if (tun->filter_attached == true) {
526 err = sk_attach_filter(&tun->fprog, tfile->socket.sk);
527 if (!err)
528 goto out;
529 }
530 tfile->queue_index = tun->numqueues;
531 rcu_assign_pointer(tfile->tun, tun);
532 rcu_assign_pointer(tun->tfiles[tun->numqueues], tfile);
533 tun->numqueues++;
534
535 if (tfile->detached)
536 tun_enable_queue(tfile);
537 else
538 sock_hold(&tfile->sk);
539
540 tun_set_real_num_queues(tun);
541
542 /* device is allowed to go away first, so no need to hold extra
543 * refcnt.
544 */
545
546 out:
547 return err;
548 }
549
550 static struct tun_struct *__tun_get(struct tun_file *tfile)
551 {
552 struct tun_struct *tun;
553
554 rcu_read_lock();
555 tun = rcu_dereference(tfile->tun);
556 if (tun)
557 dev_hold(tun->dev);
558 rcu_read_unlock();
559
560 return tun;
561 }
562
563 static struct tun_struct *tun_get(struct file *file)
564 {
565 return __tun_get(file->private_data);
566 }
567
568 static void tun_put(struct tun_struct *tun)
569 {
570 dev_put(tun->dev);
571 }
572
573 /* TAP filtering */
574 static void addr_hash_set(u32 *mask, const u8 *addr)
575 {
576 int n = ether_crc(ETH_ALEN, addr) >> 26;
577 mask[n >> 5] |= (1 << (n & 31));
578 }
579
580 static unsigned int addr_hash_test(const u32 *mask, const u8 *addr)
581 {
582 int n = ether_crc(ETH_ALEN, addr) >> 26;
583 return mask[n >> 5] & (1 << (n & 31));
584 }
585
586 static int update_filter(struct tap_filter *filter, void __user *arg)
587 {
588 struct { u8 u[ETH_ALEN]; } *addr;
589 struct tun_filter uf;
590 int err, alen, n, nexact;
591
592 if (copy_from_user(&uf, arg, sizeof(uf)))
593 return -EFAULT;
594
595 if (!uf.count) {
596 /* Disabled */
597 filter->count = 0;
598 return 0;
599 }
600
601 alen = ETH_ALEN * uf.count;
602 addr = kmalloc(alen, GFP_KERNEL);
603 if (!addr)
604 return -ENOMEM;
605
606 if (copy_from_user(addr, arg + sizeof(uf), alen)) {
607 err = -EFAULT;
608 goto done;
609 }
610
611 /* The filter is updated without holding any locks. Which is
612 * perfectly safe. We disable it first and in the worst
613 * case we'll accept a few undesired packets. */
614 filter->count = 0;
615 wmb();
616
617 /* Use first set of addresses as an exact filter */
618 for (n = 0; n < uf.count && n < FLT_EXACT_COUNT; n++)
619 memcpy(filter->addr[n], addr[n].u, ETH_ALEN);
620
621 nexact = n;
622
623 /* Remaining multicast addresses are hashed,
624 * unicast will leave the filter disabled. */
625 memset(filter->mask, 0, sizeof(filter->mask));
626 for (; n < uf.count; n++) {
627 if (!is_multicast_ether_addr(addr[n].u)) {
628 err = 0; /* no filter */
629 goto done;
630 }
631 addr_hash_set(filter->mask, addr[n].u);
632 }
633
634 /* For ALLMULTI just set the mask to all ones.
635 * This overrides the mask populated above. */
636 if ((uf.flags & TUN_FLT_ALLMULTI))
637 memset(filter->mask, ~0, sizeof(filter->mask));
638
639 /* Now enable the filter */
640 wmb();
641 filter->count = nexact;
642
643 /* Return the number of exact filters */
644 err = nexact;
645
646 done:
647 kfree(addr);
648 return err;
649 }
650
651 /* Returns: 0 - drop, !=0 - accept */
652 static int run_filter(struct tap_filter *filter, const struct sk_buff *skb)
653 {
654 /* Cannot use eth_hdr(skb) here because skb_mac_hdr() is incorrect
655 * at this point. */
656 struct ethhdr *eh = (struct ethhdr *) skb->data;
657 int i;
658
659 /* Exact match */
660 for (i = 0; i < filter->count; i++)
661 if (ether_addr_equal(eh->h_dest, filter->addr[i]))
662 return 1;
663
664 /* Inexact match (multicast only) */
665 if (is_multicast_ether_addr(eh->h_dest))
666 return addr_hash_test(filter->mask, eh->h_dest);
667
668 return 0;
669 }
670
671 /*
672 * Checks whether the packet is accepted or not.
673 * Returns: 0 - drop, !=0 - accept
674 */
675 static int check_filter(struct tap_filter *filter, const struct sk_buff *skb)
676 {
677 if (!filter->count)
678 return 1;
679
680 return run_filter(filter, skb);
681 }
682
683 /* Network device part of the driver */
684
685 static const struct ethtool_ops tun_ethtool_ops;
686
687 /* Net device detach from fd. */
688 static void tun_net_uninit(struct net_device *dev)
689 {
690 tun_detach_all(dev);
691 }
692
693 /* Net device open. */
694 static int tun_net_open(struct net_device *dev)
695 {
696 netif_tx_start_all_queues(dev);
697 return 0;
698 }
699
700 /* Net device close. */
701 static int tun_net_close(struct net_device *dev)
702 {
703 netif_tx_stop_all_queues(dev);
704 return 0;
705 }
706
707 /* Net device start xmit */
708 static netdev_tx_t tun_net_xmit(struct sk_buff *skb, struct net_device *dev)
709 {
710 struct tun_struct *tun = netdev_priv(dev);
711 int txq = skb->queue_mapping;
712 struct tun_file *tfile;
713
714 rcu_read_lock();
715 tfile = rcu_dereference(tun->tfiles[txq]);
716
717 /* Drop packet if interface is not attached */
718 if (txq >= tun->numqueues)
719 goto drop;
720
721 tun_debug(KERN_INFO, tun, "tun_net_xmit %d\n", skb->len);
722
723 BUG_ON(!tfile);
724
725 /* Drop if the filter does not like it.
726 * This is a noop if the filter is disabled.
727 * Filter can be enabled only for the TAP devices. */
728 if (!check_filter(&tun->txflt, skb))
729 goto drop;
730
731 if (tfile->socket.sk->sk_filter &&
732 sk_filter(tfile->socket.sk, skb))
733 goto drop;
734
735 /* Limit the number of packets queued by dividing txq length with the
736 * number of queues.
737 */
738 if (skb_queue_len(&tfile->socket.sk->sk_receive_queue)
739 >= dev->tx_queue_len / tun->numqueues)
740 goto drop;
741
742 /* Orphan the skb - required as we might hang on to it
743 * for indefinite time. */
744 if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC)))
745 goto drop;
746 skb_orphan(skb);
747
748 nf_reset(skb);
749
750 /* Enqueue packet */
751 skb_queue_tail(&tfile->socket.sk->sk_receive_queue, skb);
752
753 /* Notify and wake up reader process */
754 if (tfile->flags & TUN_FASYNC)
755 kill_fasync(&tfile->fasync, SIGIO, POLL_IN);
756 wake_up_interruptible_poll(&tfile->wq.wait, POLLIN |
757 POLLRDNORM | POLLRDBAND);
758
759 rcu_read_unlock();
760 return NETDEV_TX_OK;
761
762 drop:
763 dev->stats.tx_dropped++;
764 skb_tx_error(skb);
765 kfree_skb(skb);
766 rcu_read_unlock();
767 return NETDEV_TX_OK;
768 }
769
770 static void tun_net_mclist(struct net_device *dev)
771 {
772 /*
773 * This callback is supposed to deal with mc filter in
774 * _rx_ path and has nothing to do with the _tx_ path.
775 * In rx path we always accept everything userspace gives us.
776 */
777 }
778
779 #define MIN_MTU 68
780 #define MAX_MTU 65535
781
782 static int
783 tun_net_change_mtu(struct net_device *dev, int new_mtu)
784 {
785 if (new_mtu < MIN_MTU || new_mtu + dev->hard_header_len > MAX_MTU)
786 return -EINVAL;
787 dev->mtu = new_mtu;
788 return 0;
789 }
790
791 static netdev_features_t tun_net_fix_features(struct net_device *dev,
792 netdev_features_t features)
793 {
794 struct tun_struct *tun = netdev_priv(dev);
795
796 return (features & tun->set_features) | (features & ~TUN_USER_FEATURES);
797 }
798 #ifdef CONFIG_NET_POLL_CONTROLLER
799 static void tun_poll_controller(struct net_device *dev)
800 {
801 /*
802 * Tun only receives frames when:
803 * 1) the char device endpoint gets data from user space
804 * 2) the tun socket gets a sendmsg call from user space
805 * Since both of those are syncronous operations, we are guaranteed
806 * never to have pending data when we poll for it
807 * so theres nothing to do here but return.
808 * We need this though so netpoll recognizes us as an interface that
809 * supports polling, which enables bridge devices in virt setups to
810 * still use netconsole
811 */
812 return;
813 }
814 #endif
815 static const struct net_device_ops tun_netdev_ops = {
816 .ndo_uninit = tun_net_uninit,
817 .ndo_open = tun_net_open,
818 .ndo_stop = tun_net_close,
819 .ndo_start_xmit = tun_net_xmit,
820 .ndo_change_mtu = tun_net_change_mtu,
821 .ndo_fix_features = tun_net_fix_features,
822 .ndo_select_queue = tun_select_queue,
823 #ifdef CONFIG_NET_POLL_CONTROLLER
824 .ndo_poll_controller = tun_poll_controller,
825 #endif
826 };
827
828 static const struct net_device_ops tap_netdev_ops = {
829 .ndo_uninit = tun_net_uninit,
830 .ndo_open = tun_net_open,
831 .ndo_stop = tun_net_close,
832 .ndo_start_xmit = tun_net_xmit,
833 .ndo_change_mtu = tun_net_change_mtu,
834 .ndo_fix_features = tun_net_fix_features,
835 .ndo_set_rx_mode = tun_net_mclist,
836 .ndo_set_mac_address = eth_mac_addr,
837 .ndo_validate_addr = eth_validate_addr,
838 .ndo_select_queue = tun_select_queue,
839 #ifdef CONFIG_NET_POLL_CONTROLLER
840 .ndo_poll_controller = tun_poll_controller,
841 #endif
842 };
843
844 static int tun_flow_init(struct tun_struct *tun)
845 {
846 int i;
847
848 for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++)
849 INIT_HLIST_HEAD(&tun->flows[i]);
850
851 tun->ageing_time = TUN_FLOW_EXPIRE;
852 setup_timer(&tun->flow_gc_timer, tun_flow_cleanup, (unsigned long)tun);
853 mod_timer(&tun->flow_gc_timer,
854 round_jiffies_up(jiffies + tun->ageing_time));
855
856 return 0;
857 }
858
859 static void tun_flow_uninit(struct tun_struct *tun)
860 {
861 del_timer_sync(&tun->flow_gc_timer);
862 tun_flow_flush(tun);
863 }
864
865 /* Initialize net device. */
866 static void tun_net_init(struct net_device *dev)
867 {
868 struct tun_struct *tun = netdev_priv(dev);
869
870 switch (tun->flags & TUN_TYPE_MASK) {
871 case TUN_TUN_DEV:
872 dev->netdev_ops = &tun_netdev_ops;
873
874 /* Point-to-Point TUN Device */
875 dev->hard_header_len = 0;
876 dev->addr_len = 0;
877 dev->mtu = 1500;
878
879 /* Zero header length */
880 dev->type = ARPHRD_NONE;
881 dev->flags = IFF_POINTOPOINT | IFF_NOARP | IFF_MULTICAST;
882 dev->tx_queue_len = TUN_READQ_SIZE; /* We prefer our own queue length */
883 break;
884
885 case TUN_TAP_DEV:
886 dev->netdev_ops = &tap_netdev_ops;
887 /* Ethernet TAP Device */
888 ether_setup(dev);
889 dev->priv_flags &= ~IFF_TX_SKB_SHARING;
890 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
891
892 eth_hw_addr_random(dev);
893
894 dev->tx_queue_len = TUN_READQ_SIZE; /* We prefer our own queue length */
895 break;
896 }
897 }
898
899 /* Character device part */
900
901 /* Poll */
902 static unsigned int tun_chr_poll(struct file *file, poll_table *wait)
903 {
904 struct tun_file *tfile = file->private_data;
905 struct tun_struct *tun = __tun_get(tfile);
906 struct sock *sk;
907 unsigned int mask = 0;
908
909 if (!tun)
910 return POLLERR;
911
912 sk = tfile->socket.sk;
913
914 tun_debug(KERN_INFO, tun, "tun_chr_poll\n");
915
916 poll_wait(file, &tfile->wq.wait, wait);
917
918 if (!skb_queue_empty(&sk->sk_receive_queue))
919 mask |= POLLIN | POLLRDNORM;
920
921 if (sock_writeable(sk) ||
922 (!test_and_set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags) &&
923 sock_writeable(sk)))
924 mask |= POLLOUT | POLLWRNORM;
925
926 if (tun->dev->reg_state != NETREG_REGISTERED)
927 mask = POLLERR;
928
929 tun_put(tun);
930 return mask;
931 }
932
933 /* prepad is the amount to reserve at front. len is length after that.
934 * linear is a hint as to how much to copy (usually headers). */
935 static struct sk_buff *tun_alloc_skb(struct tun_file *tfile,
936 size_t prepad, size_t len,
937 size_t linear, int noblock)
938 {
939 struct sock *sk = tfile->socket.sk;
940 struct sk_buff *skb;
941 int err;
942
943 /* Under a page? Don't bother with paged skb. */
944 if (prepad + len < PAGE_SIZE || !linear)
945 linear = len;
946
947 skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
948 &err);
949 if (!skb)
950 return ERR_PTR(err);
951
952 skb_reserve(skb, prepad);
953 skb_put(skb, linear);
954 skb->data_len = len - linear;
955 skb->len += len - linear;
956
957 return skb;
958 }
959
960 /* set skb frags from iovec, this can move to core network code for reuse */
961 static int zerocopy_sg_from_iovec(struct sk_buff *skb, const struct iovec *from,
962 int offset, size_t count)
963 {
964 int len = iov_length(from, count) - offset;
965 int copy = skb_headlen(skb);
966 int size, offset1 = 0;
967 int i = 0;
968
969 /* Skip over from offset */
970 while (count && (offset >= from->iov_len)) {
971 offset -= from->iov_len;
972 ++from;
973 --count;
974 }
975
976 /* copy up to skb headlen */
977 while (count && (copy > 0)) {
978 size = min_t(unsigned int, copy, from->iov_len - offset);
979 if (copy_from_user(skb->data + offset1, from->iov_base + offset,
980 size))
981 return -EFAULT;
982 if (copy > size) {
983 ++from;
984 --count;
985 offset = 0;
986 } else
987 offset += size;
988 copy -= size;
989 offset1 += size;
990 }
991
992 if (len == offset1)
993 return 0;
994
995 while (count--) {
996 struct page *page[MAX_SKB_FRAGS];
997 int num_pages;
998 unsigned long base;
999 unsigned long truesize;
1000
1001 len = from->iov_len - offset;
1002 if (!len) {
1003 offset = 0;
1004 ++from;
1005 continue;
1006 }
1007 base = (unsigned long)from->iov_base + offset;
1008 size = ((base & ~PAGE_MASK) + len + ~PAGE_MASK) >> PAGE_SHIFT;
1009 if (i + size > MAX_SKB_FRAGS)
1010 return -EMSGSIZE;
1011 num_pages = get_user_pages_fast(base, size, 0, &page[i]);
1012 if (num_pages != size) {
1013 for (i = 0; i < num_pages; i++)
1014 put_page(page[i]);
1015 return -EFAULT;
1016 }
1017 truesize = size * PAGE_SIZE;
1018 skb->data_len += len;
1019 skb->len += len;
1020 skb->truesize += truesize;
1021 atomic_add(truesize, &skb->sk->sk_wmem_alloc);
1022 while (len) {
1023 int off = base & ~PAGE_MASK;
1024 int size = min_t(int, len, PAGE_SIZE - off);
1025 __skb_fill_page_desc(skb, i, page[i], off, size);
1026 skb_shinfo(skb)->nr_frags++;
1027 /* increase sk_wmem_alloc */
1028 base += size;
1029 len -= size;
1030 i++;
1031 }
1032 offset = 0;
1033 ++from;
1034 }
1035 return 0;
1036 }
1037
1038 /* Get packet from user space buffer */
1039 static ssize_t tun_get_user(struct tun_struct *tun, struct tun_file *tfile,
1040 void *msg_control, const struct iovec *iv,
1041 size_t total_len, size_t count, int noblock)
1042 {
1043 struct tun_pi pi = { 0, cpu_to_be16(ETH_P_IP) };
1044 struct sk_buff *skb;
1045 size_t len = total_len, align = NET_SKB_PAD;
1046 struct virtio_net_hdr gso = { 0 };
1047 int offset = 0;
1048 int copylen;
1049 bool zerocopy = false;
1050 int err;
1051 u32 rxhash;
1052
1053 if (!(tun->flags & TUN_NO_PI)) {
1054 if ((len -= sizeof(pi)) > total_len)
1055 return -EINVAL;
1056
1057 if (memcpy_fromiovecend((void *)&pi, iv, 0, sizeof(pi)))
1058 return -EFAULT;
1059 offset += sizeof(pi);
1060 }
1061
1062 if (tun->flags & TUN_VNET_HDR) {
1063 if ((len -= tun->vnet_hdr_sz) > total_len)
1064 return -EINVAL;
1065
1066 if (memcpy_fromiovecend((void *)&gso, iv, offset, sizeof(gso)))
1067 return -EFAULT;
1068
1069 if ((gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
1070 gso.csum_start + gso.csum_offset + 2 > gso.hdr_len)
1071 gso.hdr_len = gso.csum_start + gso.csum_offset + 2;
1072
1073 if (gso.hdr_len > len)
1074 return -EINVAL;
1075 offset += tun->vnet_hdr_sz;
1076 }
1077
1078 if ((tun->flags & TUN_TYPE_MASK) == TUN_TAP_DEV) {
1079 align += NET_IP_ALIGN;
1080 if (unlikely(len < ETH_HLEN ||
1081 (gso.hdr_len && gso.hdr_len < ETH_HLEN)))
1082 return -EINVAL;
1083 }
1084
1085 if (msg_control)
1086 zerocopy = true;
1087
1088 if (zerocopy) {
1089 /* Userspace may produce vectors with count greater than
1090 * MAX_SKB_FRAGS, so we need to linearize parts of the skb
1091 * to let the rest of data to be fit in the frags.
1092 */
1093 if (count > MAX_SKB_FRAGS) {
1094 copylen = iov_length(iv, count - MAX_SKB_FRAGS);
1095 if (copylen < offset)
1096 copylen = 0;
1097 else
1098 copylen -= offset;
1099 } else
1100 copylen = 0;
1101 /* There are 256 bytes to be copied in skb, so there is enough
1102 * room for skb expand head in case it is used.
1103 * The rest of the buffer is mapped from userspace.
1104 */
1105 if (copylen < gso.hdr_len)
1106 copylen = gso.hdr_len;
1107 if (!copylen)
1108 copylen = GOODCOPY_LEN;
1109 } else
1110 copylen = len;
1111
1112 skb = tun_alloc_skb(tfile, align, copylen, gso.hdr_len, noblock);
1113 if (IS_ERR(skb)) {
1114 if (PTR_ERR(skb) != -EAGAIN)
1115 tun->dev->stats.rx_dropped++;
1116 return PTR_ERR(skb);
1117 }
1118
1119 if (zerocopy)
1120 err = zerocopy_sg_from_iovec(skb, iv, offset, count);
1121 else
1122 err = skb_copy_datagram_from_iovec(skb, 0, iv, offset, len);
1123
1124 if (err) {
1125 tun->dev->stats.rx_dropped++;
1126 kfree_skb(skb);
1127 return -EFAULT;
1128 }
1129
1130 if (gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
1131 if (!skb_partial_csum_set(skb, gso.csum_start,
1132 gso.csum_offset)) {
1133 tun->dev->stats.rx_frame_errors++;
1134 kfree_skb(skb);
1135 return -EINVAL;
1136 }
1137 }
1138
1139 switch (tun->flags & TUN_TYPE_MASK) {
1140 case TUN_TUN_DEV:
1141 if (tun->flags & TUN_NO_PI) {
1142 switch (skb->data[0] & 0xf0) {
1143 case 0x40:
1144 pi.proto = htons(ETH_P_IP);
1145 break;
1146 case 0x60:
1147 pi.proto = htons(ETH_P_IPV6);
1148 break;
1149 default:
1150 tun->dev->stats.rx_dropped++;
1151 kfree_skb(skb);
1152 return -EINVAL;
1153 }
1154 }
1155
1156 skb_reset_mac_header(skb);
1157 skb->protocol = pi.proto;
1158 skb->dev = tun->dev;
1159 break;
1160 case TUN_TAP_DEV:
1161 skb->protocol = eth_type_trans(skb, tun->dev);
1162 break;
1163 }
1164
1165 if (gso.gso_type != VIRTIO_NET_HDR_GSO_NONE) {
1166 pr_debug("GSO!\n");
1167 switch (gso.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
1168 case VIRTIO_NET_HDR_GSO_TCPV4:
1169 skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
1170 break;
1171 case VIRTIO_NET_HDR_GSO_TCPV6:
1172 skb_shinfo(skb)->gso_type = SKB_GSO_TCPV6;
1173 break;
1174 case VIRTIO_NET_HDR_GSO_UDP:
1175 skb_shinfo(skb)->gso_type = SKB_GSO_UDP;
1176 break;
1177 default:
1178 tun->dev->stats.rx_frame_errors++;
1179 kfree_skb(skb);
1180 return -EINVAL;
1181 }
1182
1183 if (gso.gso_type & VIRTIO_NET_HDR_GSO_ECN)
1184 skb_shinfo(skb)->gso_type |= SKB_GSO_TCP_ECN;
1185
1186 skb_shinfo(skb)->gso_size = gso.gso_size;
1187 if (skb_shinfo(skb)->gso_size == 0) {
1188 tun->dev->stats.rx_frame_errors++;
1189 kfree_skb(skb);
1190 return -EINVAL;
1191 }
1192
1193 /* Header must be checked, and gso_segs computed. */
1194 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
1195 skb_shinfo(skb)->gso_segs = 0;
1196 }
1197
1198 /* copy skb_ubuf_info for callback when skb has no error */
1199 if (zerocopy) {
1200 skb_shinfo(skb)->destructor_arg = msg_control;
1201 skb_shinfo(skb)->tx_flags |= SKBTX_DEV_ZEROCOPY;
1202 skb_shinfo(skb)->tx_flags |= SKBTX_SHARED_FRAG;
1203 }
1204
1205 skb_reset_network_header(skb);
1206 skb_probe_transport_header(skb, 0);
1207
1208 rxhash = skb_get_rxhash(skb);
1209 netif_rx_ni(skb);
1210
1211 tun->dev->stats.rx_packets++;
1212 tun->dev->stats.rx_bytes += len;
1213
1214 tun_flow_update(tun, rxhash, tfile);
1215 return total_len;
1216 }
1217
1218 static ssize_t tun_chr_aio_write(struct kiocb *iocb, const struct iovec *iv,
1219 unsigned long count, loff_t pos)
1220 {
1221 struct file *file = iocb->ki_filp;
1222 struct tun_struct *tun = tun_get(file);
1223 struct tun_file *tfile = file->private_data;
1224 ssize_t result;
1225
1226 if (!tun)
1227 return -EBADFD;
1228
1229 tun_debug(KERN_INFO, tun, "tun_chr_write %ld\n", count);
1230
1231 result = tun_get_user(tun, tfile, NULL, iv, iov_length(iv, count),
1232 count, file->f_flags & O_NONBLOCK);
1233
1234 tun_put(tun);
1235 return result;
1236 }
1237
1238 /* Put packet to the user space buffer */
1239 static ssize_t tun_put_user(struct tun_struct *tun,
1240 struct tun_file *tfile,
1241 struct sk_buff *skb,
1242 const struct iovec *iv, int len)
1243 {
1244 struct tun_pi pi = { 0, skb->protocol };
1245 ssize_t total = 0;
1246
1247 if (!(tun->flags & TUN_NO_PI)) {
1248 if ((len -= sizeof(pi)) < 0)
1249 return -EINVAL;
1250
1251 if (len < skb->len) {
1252 /* Packet will be striped */
1253 pi.flags |= TUN_PKT_STRIP;
1254 }
1255
1256 if (memcpy_toiovecend(iv, (void *) &pi, 0, sizeof(pi)))
1257 return -EFAULT;
1258 total += sizeof(pi);
1259 }
1260
1261 if (tun->flags & TUN_VNET_HDR) {
1262 struct virtio_net_hdr gso = { 0 }; /* no info leak */
1263 if ((len -= tun->vnet_hdr_sz) < 0)
1264 return -EINVAL;
1265
1266 if (skb_is_gso(skb)) {
1267 struct skb_shared_info *sinfo = skb_shinfo(skb);
1268
1269 /* This is a hint as to how much should be linear. */
1270 gso.hdr_len = skb_headlen(skb);
1271 gso.gso_size = sinfo->gso_size;
1272 if (sinfo->gso_type & SKB_GSO_TCPV4)
1273 gso.gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
1274 else if (sinfo->gso_type & SKB_GSO_TCPV6)
1275 gso.gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
1276 else if (sinfo->gso_type & SKB_GSO_UDP)
1277 gso.gso_type = VIRTIO_NET_HDR_GSO_UDP;
1278 else {
1279 pr_err("unexpected GSO type: "
1280 "0x%x, gso_size %d, hdr_len %d\n",
1281 sinfo->gso_type, gso.gso_size,
1282 gso.hdr_len);
1283 print_hex_dump(KERN_ERR, "tun: ",
1284 DUMP_PREFIX_NONE,
1285 16, 1, skb->head,
1286 min((int)gso.hdr_len, 64), true);
1287 WARN_ON_ONCE(1);
1288 return -EINVAL;
1289 }
1290 if (sinfo->gso_type & SKB_GSO_TCP_ECN)
1291 gso.gso_type |= VIRTIO_NET_HDR_GSO_ECN;
1292 } else
1293 gso.gso_type = VIRTIO_NET_HDR_GSO_NONE;
1294
1295 if (skb->ip_summed == CHECKSUM_PARTIAL) {
1296 gso.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
1297 gso.csum_start = skb_checksum_start_offset(skb);
1298 gso.csum_offset = skb->csum_offset;
1299 } else if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
1300 gso.flags = VIRTIO_NET_HDR_F_DATA_VALID;
1301 } /* else everything is zero */
1302
1303 if (unlikely(memcpy_toiovecend(iv, (void *)&gso, total,
1304 sizeof(gso))))
1305 return -EFAULT;
1306 total += tun->vnet_hdr_sz;
1307 }
1308
1309 len = min_t(int, skb->len, len);
1310
1311 skb_copy_datagram_const_iovec(skb, 0, iv, total, len);
1312 total += skb->len;
1313
1314 tun->dev->stats.tx_packets++;
1315 tun->dev->stats.tx_bytes += len;
1316
1317 return total;
1318 }
1319
1320 static ssize_t tun_do_read(struct tun_struct *tun, struct tun_file *tfile,
1321 struct kiocb *iocb, const struct iovec *iv,
1322 ssize_t len, int noblock)
1323 {
1324 DECLARE_WAITQUEUE(wait, current);
1325 struct sk_buff *skb;
1326 ssize_t ret = 0;
1327
1328 tun_debug(KERN_INFO, tun, "tun_do_read\n");
1329
1330 if (unlikely(!noblock))
1331 add_wait_queue(&tfile->wq.wait, &wait);
1332 while (len) {
1333 current->state = TASK_INTERRUPTIBLE;
1334
1335 /* Read frames from the queue */
1336 if (!(skb = skb_dequeue(&tfile->socket.sk->sk_receive_queue))) {
1337 if (noblock) {
1338 ret = -EAGAIN;
1339 break;
1340 }
1341 if (signal_pending(current)) {
1342 ret = -ERESTARTSYS;
1343 break;
1344 }
1345 if (tun->dev->reg_state != NETREG_REGISTERED) {
1346 ret = -EIO;
1347 break;
1348 }
1349
1350 /* Nothing to read, let's sleep */
1351 schedule();
1352 continue;
1353 }
1354
1355 ret = tun_put_user(tun, tfile, skb, iv, len);
1356 kfree_skb(skb);
1357 break;
1358 }
1359
1360 current->state = TASK_RUNNING;
1361 if (unlikely(!noblock))
1362 remove_wait_queue(&tfile->wq.wait, &wait);
1363
1364 return ret;
1365 }
1366
1367 static ssize_t tun_chr_aio_read(struct kiocb *iocb, const struct iovec *iv,
1368 unsigned long count, loff_t pos)
1369 {
1370 struct file *file = iocb->ki_filp;
1371 struct tun_file *tfile = file->private_data;
1372 struct tun_struct *tun = __tun_get(tfile);
1373 ssize_t len, ret;
1374
1375 if (!tun)
1376 return -EBADFD;
1377 len = iov_length(iv, count);
1378 if (len < 0) {
1379 ret = -EINVAL;
1380 goto out;
1381 }
1382
1383 ret = tun_do_read(tun, tfile, iocb, iv, len,
1384 file->f_flags & O_NONBLOCK);
1385 ret = min_t(ssize_t, ret, len);
1386 out:
1387 tun_put(tun);
1388 return ret;
1389 }
1390
1391 static void tun_free_netdev(struct net_device *dev)
1392 {
1393 struct tun_struct *tun = netdev_priv(dev);
1394
1395 BUG_ON(!(list_empty(&tun->disabled)));
1396 tun_flow_uninit(tun);
1397 security_tun_dev_free_security(tun->security);
1398 free_netdev(dev);
1399 }
1400
1401 static void tun_setup(struct net_device *dev)
1402 {
1403 struct tun_struct *tun = netdev_priv(dev);
1404
1405 tun->owner = INVALID_UID;
1406 tun->group = INVALID_GID;
1407
1408 dev->ethtool_ops = &tun_ethtool_ops;
1409 dev->destructor = tun_free_netdev;
1410 }
1411
1412 /* Trivial set of netlink ops to allow deleting tun or tap
1413 * device with netlink.
1414 */
1415 static int tun_validate(struct nlattr *tb[], struct nlattr *data[])
1416 {
1417 return -EINVAL;
1418 }
1419
1420 static struct rtnl_link_ops tun_link_ops __read_mostly = {
1421 .kind = DRV_NAME,
1422 .priv_size = sizeof(struct tun_struct),
1423 .setup = tun_setup,
1424 .validate = tun_validate,
1425 };
1426
1427 static void tun_sock_write_space(struct sock *sk)
1428 {
1429 struct tun_file *tfile;
1430 wait_queue_head_t *wqueue;
1431
1432 if (!sock_writeable(sk))
1433 return;
1434
1435 if (!test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags))
1436 return;
1437
1438 wqueue = sk_sleep(sk);
1439 if (wqueue && waitqueue_active(wqueue))
1440 wake_up_interruptible_sync_poll(wqueue, POLLOUT |
1441 POLLWRNORM | POLLWRBAND);
1442
1443 tfile = container_of(sk, struct tun_file, sk);
1444 kill_fasync(&tfile->fasync, SIGIO, POLL_OUT);
1445 }
1446
1447 static int tun_sendmsg(struct kiocb *iocb, struct socket *sock,
1448 struct msghdr *m, size_t total_len)
1449 {
1450 int ret;
1451 struct tun_file *tfile = container_of(sock, struct tun_file, socket);
1452 struct tun_struct *tun = __tun_get(tfile);
1453
1454 if (!tun)
1455 return -EBADFD;
1456 ret = tun_get_user(tun, tfile, m->msg_control, m->msg_iov, total_len,
1457 m->msg_iovlen, m->msg_flags & MSG_DONTWAIT);
1458 tun_put(tun);
1459 return ret;
1460 }
1461
1462
1463 static int tun_recvmsg(struct kiocb *iocb, struct socket *sock,
1464 struct msghdr *m, size_t total_len,
1465 int flags)
1466 {
1467 struct tun_file *tfile = container_of(sock, struct tun_file, socket);
1468 struct tun_struct *tun = __tun_get(tfile);
1469 int ret;
1470
1471 if (!tun)
1472 return -EBADFD;
1473
1474 if (flags & ~(MSG_DONTWAIT|MSG_TRUNC))
1475 return -EINVAL;
1476 ret = tun_do_read(tun, tfile, iocb, m->msg_iov, total_len,
1477 flags & MSG_DONTWAIT);
1478 if (ret > total_len) {
1479 m->msg_flags |= MSG_TRUNC;
1480 ret = flags & MSG_TRUNC ? ret : total_len;
1481 }
1482 tun_put(tun);
1483 return ret;
1484 }
1485
1486 static int tun_release(struct socket *sock)
1487 {
1488 if (sock->sk)
1489 sock_put(sock->sk);
1490 return 0;
1491 }
1492
1493 /* Ops structure to mimic raw sockets with tun */
1494 static const struct proto_ops tun_socket_ops = {
1495 .sendmsg = tun_sendmsg,
1496 .recvmsg = tun_recvmsg,
1497 .release = tun_release,
1498 };
1499
1500 static struct proto tun_proto = {
1501 .name = "tun",
1502 .owner = THIS_MODULE,
1503 .obj_size = sizeof(struct tun_file),
1504 };
1505
1506 static int tun_flags(struct tun_struct *tun)
1507 {
1508 int flags = 0;
1509
1510 if (tun->flags & TUN_TUN_DEV)
1511 flags |= IFF_TUN;
1512 else
1513 flags |= IFF_TAP;
1514
1515 if (tun->flags & TUN_NO_PI)
1516 flags |= IFF_NO_PI;
1517
1518 /* This flag has no real effect. We track the value for backwards
1519 * compatibility.
1520 */
1521 if (tun->flags & TUN_ONE_QUEUE)
1522 flags |= IFF_ONE_QUEUE;
1523
1524 if (tun->flags & TUN_VNET_HDR)
1525 flags |= IFF_VNET_HDR;
1526
1527 if (tun->flags & TUN_TAP_MQ)
1528 flags |= IFF_MULTI_QUEUE;
1529
1530 return flags;
1531 }
1532
1533 static ssize_t tun_show_flags(struct device *dev, struct device_attribute *attr,
1534 char *buf)
1535 {
1536 struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1537 return sprintf(buf, "0x%x\n", tun_flags(tun));
1538 }
1539
1540 static ssize_t tun_show_owner(struct device *dev, struct device_attribute *attr,
1541 char *buf)
1542 {
1543 struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1544 return uid_valid(tun->owner)?
1545 sprintf(buf, "%u\n",
1546 from_kuid_munged(current_user_ns(), tun->owner)):
1547 sprintf(buf, "-1\n");
1548 }
1549
1550 static ssize_t tun_show_group(struct device *dev, struct device_attribute *attr,
1551 char *buf)
1552 {
1553 struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1554 return gid_valid(tun->group) ?
1555 sprintf(buf, "%u\n",
1556 from_kgid_munged(current_user_ns(), tun->group)):
1557 sprintf(buf, "-1\n");
1558 }
1559
1560 static DEVICE_ATTR(tun_flags, 0444, tun_show_flags, NULL);
1561 static DEVICE_ATTR(owner, 0444, tun_show_owner, NULL);
1562 static DEVICE_ATTR(group, 0444, tun_show_group, NULL);
1563
1564 static int tun_set_iff(struct net *net, struct file *file, struct ifreq *ifr)
1565 {
1566 struct tun_struct *tun;
1567 struct tun_file *tfile = file->private_data;
1568 struct net_device *dev;
1569 int err;
1570
1571 if (tfile->detached)
1572 return -EINVAL;
1573
1574 dev = __dev_get_by_name(net, ifr->ifr_name);
1575 if (dev) {
1576 if (ifr->ifr_flags & IFF_TUN_EXCL)
1577 return -EBUSY;
1578 if ((ifr->ifr_flags & IFF_TUN) && dev->netdev_ops == &tun_netdev_ops)
1579 tun = netdev_priv(dev);
1580 else if ((ifr->ifr_flags & IFF_TAP) && dev->netdev_ops == &tap_netdev_ops)
1581 tun = netdev_priv(dev);
1582 else
1583 return -EINVAL;
1584
1585 if (tun_not_capable(tun))
1586 return -EPERM;
1587 err = security_tun_dev_open(tun->security);
1588 if (err < 0)
1589 return err;
1590
1591 err = tun_attach(tun, file);
1592 if (err < 0)
1593 return err;
1594
1595 if (tun->flags & TUN_TAP_MQ &&
1596 (tun->numqueues + tun->numdisabled > 1))
1597 return err;
1598 }
1599 else {
1600 char *name;
1601 unsigned long flags = 0;
1602 int queues = ifr->ifr_flags & IFF_MULTI_QUEUE ?
1603 MAX_TAP_QUEUES : 1;
1604
1605 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
1606 return -EPERM;
1607 err = security_tun_dev_create();
1608 if (err < 0)
1609 return err;
1610
1611 /* Set dev type */
1612 if (ifr->ifr_flags & IFF_TUN) {
1613 /* TUN device */
1614 flags |= TUN_TUN_DEV;
1615 name = "tun%d";
1616 } else if (ifr->ifr_flags & IFF_TAP) {
1617 /* TAP device */
1618 flags |= TUN_TAP_DEV;
1619 name = "tap%d";
1620 } else
1621 return -EINVAL;
1622
1623 if (*ifr->ifr_name)
1624 name = ifr->ifr_name;
1625
1626 dev = alloc_netdev_mqs(sizeof(struct tun_struct), name,
1627 tun_setup, queues, queues);
1628
1629 if (!dev)
1630 return -ENOMEM;
1631
1632 dev_net_set(dev, net);
1633 dev->rtnl_link_ops = &tun_link_ops;
1634
1635 tun = netdev_priv(dev);
1636 tun->dev = dev;
1637 tun->flags = flags;
1638 tun->txflt.count = 0;
1639 tun->vnet_hdr_sz = sizeof(struct virtio_net_hdr);
1640
1641 tun->filter_attached = false;
1642 tun->sndbuf = tfile->socket.sk->sk_sndbuf;
1643
1644 spin_lock_init(&tun->lock);
1645
1646 err = security_tun_dev_alloc_security(&tun->security);
1647 if (err < 0)
1648 goto err_free_dev;
1649
1650 tun_net_init(dev);
1651
1652 err = tun_flow_init(tun);
1653 if (err < 0)
1654 goto err_free_dev;
1655
1656 dev->hw_features = NETIF_F_SG | NETIF_F_FRAGLIST |
1657 TUN_USER_FEATURES;
1658 dev->features = dev->hw_features;
1659 dev->vlan_features = dev->features;
1660
1661 INIT_LIST_HEAD(&tun->disabled);
1662 err = tun_attach(tun, file);
1663 if (err < 0)
1664 goto err_free_dev;
1665
1666 err = register_netdevice(tun->dev);
1667 if (err < 0)
1668 goto err_free_dev;
1669
1670 if (device_create_file(&tun->dev->dev, &dev_attr_tun_flags) ||
1671 device_create_file(&tun->dev->dev, &dev_attr_owner) ||
1672 device_create_file(&tun->dev->dev, &dev_attr_group))
1673 pr_err("Failed to create tun sysfs files\n");
1674 }
1675
1676 netif_carrier_on(tun->dev);
1677
1678 tun_debug(KERN_INFO, tun, "tun_set_iff\n");
1679
1680 if (ifr->ifr_flags & IFF_NO_PI)
1681 tun->flags |= TUN_NO_PI;
1682 else
1683 tun->flags &= ~TUN_NO_PI;
1684
1685 /* This flag has no real effect. We track the value for backwards
1686 * compatibility.
1687 */
1688 if (ifr->ifr_flags & IFF_ONE_QUEUE)
1689 tun->flags |= TUN_ONE_QUEUE;
1690 else
1691 tun->flags &= ~TUN_ONE_QUEUE;
1692
1693 if (ifr->ifr_flags & IFF_VNET_HDR)
1694 tun->flags |= TUN_VNET_HDR;
1695 else
1696 tun->flags &= ~TUN_VNET_HDR;
1697
1698 if (ifr->ifr_flags & IFF_MULTI_QUEUE)
1699 tun->flags |= TUN_TAP_MQ;
1700 else
1701 tun->flags &= ~TUN_TAP_MQ;
1702
1703 /* Make sure persistent devices do not get stuck in
1704 * xoff state.
1705 */
1706 if (netif_running(tun->dev))
1707 netif_tx_wake_all_queues(tun->dev);
1708
1709 strcpy(ifr->ifr_name, tun->dev->name);
1710 return 0;
1711
1712 err_free_dev:
1713 free_netdev(dev);
1714 return err;
1715 }
1716
1717 static void tun_get_iff(struct net *net, struct tun_struct *tun,
1718 struct ifreq *ifr)
1719 {
1720 tun_debug(KERN_INFO, tun, "tun_get_iff\n");
1721
1722 strcpy(ifr->ifr_name, tun->dev->name);
1723
1724 ifr->ifr_flags = tun_flags(tun);
1725
1726 }
1727
1728 /* This is like a cut-down ethtool ops, except done via tun fd so no
1729 * privs required. */
1730 static int set_offload(struct tun_struct *tun, unsigned long arg)
1731 {
1732 netdev_features_t features = 0;
1733
1734 if (arg & TUN_F_CSUM) {
1735 features |= NETIF_F_HW_CSUM;
1736 arg &= ~TUN_F_CSUM;
1737
1738 if (arg & (TUN_F_TSO4|TUN_F_TSO6)) {
1739 if (arg & TUN_F_TSO_ECN) {
1740 features |= NETIF_F_TSO_ECN;
1741 arg &= ~TUN_F_TSO_ECN;
1742 }
1743 if (arg & TUN_F_TSO4)
1744 features |= NETIF_F_TSO;
1745 if (arg & TUN_F_TSO6)
1746 features |= NETIF_F_TSO6;
1747 arg &= ~(TUN_F_TSO4|TUN_F_TSO6);
1748 }
1749
1750 if (arg & TUN_F_UFO) {
1751 features |= NETIF_F_UFO;
1752 arg &= ~TUN_F_UFO;
1753 }
1754 }
1755
1756 /* This gives the user a way to test for new features in future by
1757 * trying to set them. */
1758 if (arg)
1759 return -EINVAL;
1760
1761 tun->set_features = features;
1762 netdev_update_features(tun->dev);
1763
1764 return 0;
1765 }
1766
1767 static void tun_detach_filter(struct tun_struct *tun, int n)
1768 {
1769 int i;
1770 struct tun_file *tfile;
1771
1772 for (i = 0; i < n; i++) {
1773 tfile = rtnl_dereference(tun->tfiles[i]);
1774 sk_detach_filter(tfile->socket.sk);
1775 }
1776
1777 tun->filter_attached = false;
1778 }
1779
1780 static int tun_attach_filter(struct tun_struct *tun)
1781 {
1782 int i, ret = 0;
1783 struct tun_file *tfile;
1784
1785 for (i = 0; i < tun->numqueues; i++) {
1786 tfile = rtnl_dereference(tun->tfiles[i]);
1787 ret = sk_attach_filter(&tun->fprog, tfile->socket.sk);
1788 if (ret) {
1789 tun_detach_filter(tun, i);
1790 return ret;
1791 }
1792 }
1793
1794 tun->filter_attached = true;
1795 return ret;
1796 }
1797
1798 static void tun_set_sndbuf(struct tun_struct *tun)
1799 {
1800 struct tun_file *tfile;
1801 int i;
1802
1803 for (i = 0; i < tun->numqueues; i++) {
1804 tfile = rtnl_dereference(tun->tfiles[i]);
1805 tfile->socket.sk->sk_sndbuf = tun->sndbuf;
1806 }
1807 }
1808
1809 static int tun_set_queue(struct file *file, struct ifreq *ifr)
1810 {
1811 struct tun_file *tfile = file->private_data;
1812 struct tun_struct *tun;
1813 int ret = 0;
1814
1815 rtnl_lock();
1816
1817 if (ifr->ifr_flags & IFF_ATTACH_QUEUE) {
1818 tun = tfile->detached;
1819 if (!tun) {
1820 ret = -EINVAL;
1821 goto unlock;
1822 }
1823 ret = security_tun_dev_attach_queue(tun->security);
1824 if (ret < 0)
1825 goto unlock;
1826 ret = tun_attach(tun, file);
1827 } else if (ifr->ifr_flags & IFF_DETACH_QUEUE) {
1828 tun = rtnl_dereference(tfile->tun);
1829 if (!tun || !(tun->flags & TUN_TAP_MQ) || tfile->detached)
1830 ret = -EINVAL;
1831 else
1832 __tun_detach(tfile, false);
1833 } else
1834 ret = -EINVAL;
1835
1836 unlock:
1837 rtnl_unlock();
1838 return ret;
1839 }
1840
1841 static long __tun_chr_ioctl(struct file *file, unsigned int cmd,
1842 unsigned long arg, int ifreq_len)
1843 {
1844 struct tun_file *tfile = file->private_data;
1845 struct tun_struct *tun;
1846 void __user* argp = (void __user*)arg;
1847 struct ifreq ifr;
1848 kuid_t owner;
1849 kgid_t group;
1850 int sndbuf;
1851 int vnet_hdr_sz;
1852 int ret;
1853
1854 if (cmd == TUNSETIFF || cmd == TUNSETQUEUE || _IOC_TYPE(cmd) == 0x89) {
1855 if (copy_from_user(&ifr, argp, ifreq_len))
1856 return -EFAULT;
1857 } else {
1858 memset(&ifr, 0, sizeof(ifr));
1859 }
1860 if (cmd == TUNGETFEATURES) {
1861 /* Currently this just means: "what IFF flags are valid?".
1862 * This is needed because we never checked for invalid flags on
1863 * TUNSETIFF. */
1864 return put_user(IFF_TUN | IFF_TAP | IFF_NO_PI | IFF_ONE_QUEUE |
1865 IFF_VNET_HDR | IFF_MULTI_QUEUE,
1866 (unsigned int __user*)argp);
1867 } else if (cmd == TUNSETQUEUE)
1868 return tun_set_queue(file, &ifr);
1869
1870 ret = 0;
1871 rtnl_lock();
1872
1873 tun = __tun_get(tfile);
1874 if (cmd == TUNSETIFF && !tun) {
1875 ifr.ifr_name[IFNAMSIZ-1] = '\0';
1876
1877 ret = tun_set_iff(tfile->net, file, &ifr);
1878
1879 if (ret)
1880 goto unlock;
1881
1882 if (copy_to_user(argp, &ifr, ifreq_len))
1883 ret = -EFAULT;
1884 goto unlock;
1885 }
1886
1887 ret = -EBADFD;
1888 if (!tun)
1889 goto unlock;
1890
1891 tun_debug(KERN_INFO, tun, "tun_chr_ioctl cmd %u\n", cmd);
1892
1893 ret = 0;
1894 switch (cmd) {
1895 case TUNGETIFF:
1896 tun_get_iff(current->nsproxy->net_ns, tun, &ifr);
1897
1898 if (copy_to_user(argp, &ifr, ifreq_len))
1899 ret = -EFAULT;
1900 break;
1901
1902 case TUNSETNOCSUM:
1903 /* Disable/Enable checksum */
1904
1905 /* [unimplemented] */
1906 tun_debug(KERN_INFO, tun, "ignored: set checksum %s\n",
1907 arg ? "disabled" : "enabled");
1908 break;
1909
1910 case TUNSETPERSIST:
1911 /* Disable/Enable persist mode. Keep an extra reference to the
1912 * module to prevent the module being unprobed.
1913 */
1914 if (arg && !(tun->flags & TUN_PERSIST)) {
1915 tun->flags |= TUN_PERSIST;
1916 __module_get(THIS_MODULE);
1917 }
1918 if (!arg && (tun->flags & TUN_PERSIST)) {
1919 tun->flags &= ~TUN_PERSIST;
1920 module_put(THIS_MODULE);
1921 }
1922
1923 tun_debug(KERN_INFO, tun, "persist %s\n",
1924 arg ? "enabled" : "disabled");
1925 break;
1926
1927 case TUNSETOWNER:
1928 /* Set owner of the device */
1929 owner = make_kuid(current_user_ns(), arg);
1930 if (!uid_valid(owner)) {
1931 ret = -EINVAL;
1932 break;
1933 }
1934 tun->owner = owner;
1935 tun_debug(KERN_INFO, tun, "owner set to %u\n",
1936 from_kuid(&init_user_ns, tun->owner));
1937 break;
1938
1939 case TUNSETGROUP:
1940 /* Set group of the device */
1941 group = make_kgid(current_user_ns(), arg);
1942 if (!gid_valid(group)) {
1943 ret = -EINVAL;
1944 break;
1945 }
1946 tun->group = group;
1947 tun_debug(KERN_INFO, tun, "group set to %u\n",
1948 from_kgid(&init_user_ns, tun->group));
1949 break;
1950
1951 case TUNSETLINK:
1952 /* Only allow setting the type when the interface is down */
1953 if (tun->dev->flags & IFF_UP) {
1954 tun_debug(KERN_INFO, tun,
1955 "Linktype set failed because interface is up\n");
1956 ret = -EBUSY;
1957 } else {
1958 tun->dev->type = (int) arg;
1959 tun_debug(KERN_INFO, tun, "linktype set to %d\n",
1960 tun->dev->type);
1961 ret = 0;
1962 }
1963 break;
1964
1965 #ifdef TUN_DEBUG
1966 case TUNSETDEBUG:
1967 tun->debug = arg;
1968 break;
1969 #endif
1970 case TUNSETOFFLOAD:
1971 ret = set_offload(tun, arg);
1972 break;
1973
1974 case TUNSETTXFILTER:
1975 /* Can be set only for TAPs */
1976 ret = -EINVAL;
1977 if ((tun->flags & TUN_TYPE_MASK) != TUN_TAP_DEV)
1978 break;
1979 ret = update_filter(&tun->txflt, (void __user *)arg);
1980 break;
1981
1982 case SIOCGIFHWADDR:
1983 /* Get hw address */
1984 memcpy(ifr.ifr_hwaddr.sa_data, tun->dev->dev_addr, ETH_ALEN);
1985 ifr.ifr_hwaddr.sa_family = tun->dev->type;
1986 if (copy_to_user(argp, &ifr, ifreq_len))
1987 ret = -EFAULT;
1988 break;
1989
1990 case SIOCSIFHWADDR:
1991 /* Set hw address */
1992 tun_debug(KERN_DEBUG, tun, "set hw address: %pM\n",
1993 ifr.ifr_hwaddr.sa_data);
1994
1995 ret = dev_set_mac_address(tun->dev, &ifr.ifr_hwaddr);
1996 break;
1997
1998 case TUNGETSNDBUF:
1999 sndbuf = tfile->socket.sk->sk_sndbuf;
2000 if (copy_to_user(argp, &sndbuf, sizeof(sndbuf)))
2001 ret = -EFAULT;
2002 break;
2003
2004 case TUNSETSNDBUF:
2005 if (copy_from_user(&sndbuf, argp, sizeof(sndbuf))) {
2006 ret = -EFAULT;
2007 break;
2008 }
2009
2010 tun->sndbuf = sndbuf;
2011 tun_set_sndbuf(tun);
2012 break;
2013
2014 case TUNGETVNETHDRSZ:
2015 vnet_hdr_sz = tun->vnet_hdr_sz;
2016 if (copy_to_user(argp, &vnet_hdr_sz, sizeof(vnet_hdr_sz)))
2017 ret = -EFAULT;
2018 break;
2019
2020 case TUNSETVNETHDRSZ:
2021 if (copy_from_user(&vnet_hdr_sz, argp, sizeof(vnet_hdr_sz))) {
2022 ret = -EFAULT;
2023 break;
2024 }
2025 if (vnet_hdr_sz < (int)sizeof(struct virtio_net_hdr)) {
2026 ret = -EINVAL;
2027 break;
2028 }
2029
2030 tun->vnet_hdr_sz = vnet_hdr_sz;
2031 break;
2032
2033 case TUNATTACHFILTER:
2034 /* Can be set only for TAPs */
2035 ret = -EINVAL;
2036 if ((tun->flags & TUN_TYPE_MASK) != TUN_TAP_DEV)
2037 break;
2038 ret = -EFAULT;
2039 if (copy_from_user(&tun->fprog, argp, sizeof(tun->fprog)))
2040 break;
2041
2042 ret = tun_attach_filter(tun);
2043 break;
2044
2045 case TUNDETACHFILTER:
2046 /* Can be set only for TAPs */
2047 ret = -EINVAL;
2048 if ((tun->flags & TUN_TYPE_MASK) != TUN_TAP_DEV)
2049 break;
2050 ret = 0;
2051 tun_detach_filter(tun, tun->numqueues);
2052 break;
2053
2054 default:
2055 ret = -EINVAL;
2056 break;
2057 }
2058
2059 unlock:
2060 rtnl_unlock();
2061 if (tun)
2062 tun_put(tun);
2063 return ret;
2064 }
2065
2066 static long tun_chr_ioctl(struct file *file,
2067 unsigned int cmd, unsigned long arg)
2068 {
2069 return __tun_chr_ioctl(file, cmd, arg, sizeof (struct ifreq));
2070 }
2071
2072 #ifdef CONFIG_COMPAT
2073 static long tun_chr_compat_ioctl(struct file *file,
2074 unsigned int cmd, unsigned long arg)
2075 {
2076 switch (cmd) {
2077 case TUNSETIFF:
2078 case TUNGETIFF:
2079 case TUNSETTXFILTER:
2080 case TUNGETSNDBUF:
2081 case TUNSETSNDBUF:
2082 case SIOCGIFHWADDR:
2083 case SIOCSIFHWADDR:
2084 arg = (unsigned long)compat_ptr(arg);
2085 break;
2086 default:
2087 arg = (compat_ulong_t)arg;
2088 break;
2089 }
2090
2091 /*
2092 * compat_ifreq is shorter than ifreq, so we must not access beyond
2093 * the end of that structure. All fields that are used in this
2094 * driver are compatible though, we don't need to convert the
2095 * contents.
2096 */
2097 return __tun_chr_ioctl(file, cmd, arg, sizeof(struct compat_ifreq));
2098 }
2099 #endif /* CONFIG_COMPAT */
2100
2101 static int tun_chr_fasync(int fd, struct file *file, int on)
2102 {
2103 struct tun_file *tfile = file->private_data;
2104 int ret;
2105
2106 if ((ret = fasync_helper(fd, file, on, &tfile->fasync)) < 0)
2107 goto out;
2108
2109 if (on) {
2110 ret = __f_setown(file, task_pid(current), PIDTYPE_PID, 0);
2111 if (ret)
2112 goto out;
2113 tfile->flags |= TUN_FASYNC;
2114 } else
2115 tfile->flags &= ~TUN_FASYNC;
2116 ret = 0;
2117 out:
2118 return ret;
2119 }
2120
2121 static int tun_chr_open(struct inode *inode, struct file * file)
2122 {
2123 struct tun_file *tfile;
2124
2125 DBG1(KERN_INFO, "tunX: tun_chr_open\n");
2126
2127 tfile = (struct tun_file *)sk_alloc(&init_net, AF_UNSPEC, GFP_KERNEL,
2128 &tun_proto);
2129 if (!tfile)
2130 return -ENOMEM;
2131 rcu_assign_pointer(tfile->tun, NULL);
2132 tfile->net = get_net(current->nsproxy->net_ns);
2133 tfile->flags = 0;
2134
2135 rcu_assign_pointer(tfile->socket.wq, &tfile->wq);
2136 init_waitqueue_head(&tfile->wq.wait);
2137
2138 tfile->socket.file = file;
2139 tfile->socket.ops = &tun_socket_ops;
2140
2141 sock_init_data(&tfile->socket, &tfile->sk);
2142 sk_change_net(&tfile->sk, tfile->net);
2143
2144 tfile->sk.sk_write_space = tun_sock_write_space;
2145 tfile->sk.sk_sndbuf = INT_MAX;
2146
2147 file->private_data = tfile;
2148 set_bit(SOCK_EXTERNALLY_ALLOCATED, &tfile->socket.flags);
2149 INIT_LIST_HEAD(&tfile->next);
2150
2151 return 0;
2152 }
2153
2154 static int tun_chr_close(struct inode *inode, struct file *file)
2155 {
2156 struct tun_file *tfile = file->private_data;
2157 struct net *net = tfile->net;
2158
2159 tun_detach(tfile, true);
2160 put_net(net);
2161
2162 return 0;
2163 }
2164
2165 static const struct file_operations tun_fops = {
2166 .owner = THIS_MODULE,
2167 .llseek = no_llseek,
2168 .read = do_sync_read,
2169 .aio_read = tun_chr_aio_read,
2170 .write = do_sync_write,
2171 .aio_write = tun_chr_aio_write,
2172 .poll = tun_chr_poll,
2173 .unlocked_ioctl = tun_chr_ioctl,
2174 #ifdef CONFIG_COMPAT
2175 .compat_ioctl = tun_chr_compat_ioctl,
2176 #endif
2177 .open = tun_chr_open,
2178 .release = tun_chr_close,
2179 .fasync = tun_chr_fasync
2180 };
2181
2182 static struct miscdevice tun_miscdev = {
2183 .minor = TUN_MINOR,
2184 .name = "tun",
2185 .nodename = "net/tun",
2186 .fops = &tun_fops,
2187 };
2188
2189 /* ethtool interface */
2190
2191 static int tun_get_settings(struct net_device *dev, struct ethtool_cmd *cmd)
2192 {
2193 cmd->supported = 0;
2194 cmd->advertising = 0;
2195 ethtool_cmd_speed_set(cmd, SPEED_10);
2196 cmd->duplex = DUPLEX_FULL;
2197 cmd->port = PORT_TP;
2198 cmd->phy_address = 0;
2199 cmd->transceiver = XCVR_INTERNAL;
2200 cmd->autoneg = AUTONEG_DISABLE;
2201 cmd->maxtxpkt = 0;
2202 cmd->maxrxpkt = 0;
2203 return 0;
2204 }
2205
2206 static void tun_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
2207 {
2208 struct tun_struct *tun = netdev_priv(dev);
2209
2210 strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
2211 strlcpy(info->version, DRV_VERSION, sizeof(info->version));
2212
2213 switch (tun->flags & TUN_TYPE_MASK) {
2214 case TUN_TUN_DEV:
2215 strlcpy(info->bus_info, "tun", sizeof(info->bus_info));
2216 break;
2217 case TUN_TAP_DEV:
2218 strlcpy(info->bus_info, "tap", sizeof(info->bus_info));
2219 break;
2220 }
2221 }
2222
2223 static u32 tun_get_msglevel(struct net_device *dev)
2224 {
2225 #ifdef TUN_DEBUG
2226 struct tun_struct *tun = netdev_priv(dev);
2227 return tun->debug;
2228 #else
2229 return -EOPNOTSUPP;
2230 #endif
2231 }
2232
2233 static void tun_set_msglevel(struct net_device *dev, u32 value)
2234 {
2235 #ifdef TUN_DEBUG
2236 struct tun_struct *tun = netdev_priv(dev);
2237 tun->debug = value;
2238 #endif
2239 }
2240
2241 static const struct ethtool_ops tun_ethtool_ops = {
2242 .get_settings = tun_get_settings,
2243 .get_drvinfo = tun_get_drvinfo,
2244 .get_msglevel = tun_get_msglevel,
2245 .set_msglevel = tun_set_msglevel,
2246 .get_link = ethtool_op_get_link,
2247 };
2248
2249
2250 static int __init tun_init(void)
2251 {
2252 int ret = 0;
2253
2254 pr_info("%s, %s\n", DRV_DESCRIPTION, DRV_VERSION);
2255 pr_info("%s\n", DRV_COPYRIGHT);
2256
2257 ret = rtnl_link_register(&tun_link_ops);
2258 if (ret) {
2259 pr_err("Can't register link_ops\n");
2260 goto err_linkops;
2261 }
2262
2263 ret = misc_register(&tun_miscdev);
2264 if (ret) {
2265 pr_err("Can't register misc device %d\n", TUN_MINOR);
2266 goto err_misc;
2267 }
2268 return 0;
2269 err_misc:
2270 rtnl_link_unregister(&tun_link_ops);
2271 err_linkops:
2272 return ret;
2273 }
2274
2275 static void tun_cleanup(void)
2276 {
2277 misc_deregister(&tun_miscdev);
2278 rtnl_link_unregister(&tun_link_ops);
2279 }
2280
2281 /* Get an underlying socket object from tun file. Returns error unless file is
2282 * attached to a device. The returned object works like a packet socket, it
2283 * can be used for sock_sendmsg/sock_recvmsg. The caller is responsible for
2284 * holding a reference to the file for as long as the socket is in use. */
2285 struct socket *tun_get_socket(struct file *file)
2286 {
2287 struct tun_file *tfile;
2288 if (file->f_op != &tun_fops)
2289 return ERR_PTR(-EINVAL);
2290 tfile = file->private_data;
2291 if (!tfile)
2292 return ERR_PTR(-EBADFD);
2293 return &tfile->socket;
2294 }
2295 EXPORT_SYMBOL_GPL(tun_get_socket);
2296
2297 module_init(tun_init);
2298 module_exit(tun_cleanup);
2299 MODULE_DESCRIPTION(DRV_DESCRIPTION);
2300 MODULE_AUTHOR(DRV_COPYRIGHT);
2301 MODULE_LICENSE("GPL");
2302 MODULE_ALIAS_MISCDEV(TUN_MINOR);
2303 MODULE_ALIAS("devname:net/tun");
Linux with added FPC-III support