?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
jme: Do not enable NIC WoL functions on S0
[linux/fpc-iii.git] / drivers / net / tun.c
blob999484feb37770a227285dd7186e56648da15b86
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 #include <linux/seq_file.h>
74
75 #include <asm/uaccess.h>
76
77 /* Uncomment to enable debugging */
78 /* #define TUN_DEBUG 1 */
79
80 #ifdef TUN_DEBUG
81 static int debug;
82
83 #define tun_debug(level, tun, fmt, args...) \
84 do { \
85 if (tun->debug) \
86 netdev_printk(level, tun->dev, fmt, ##args); \
87 } while (0)
88 #define DBG1(level, fmt, args...) \
89 do { \
90 if (debug == 2) \
91 printk(level fmt, ##args); \
92 } while (0)
93 #else
94 #define tun_debug(level, tun, fmt, args...) \
95 do { \
96 if (0) \
97 netdev_printk(level, tun->dev, fmt, ##args); \
98 } while (0)
99 #define DBG1(level, fmt, args...) \
100 do { \
101 if (0) \
102 printk(level fmt, ##args); \
103 } while (0)
104 #endif
105
106 #define GOODCOPY_LEN 128
107
108 #define FLT_EXACT_COUNT 8
109 struct tap_filter {
110 unsigned int count; /* Number of addrs. Zero means disabled */
111 u32 mask[2]; /* Mask of the hashed addrs */
112 unsigned char addr[FLT_EXACT_COUNT][ETH_ALEN];
113 };
114
115 /* DEFAULT_MAX_NUM_RSS_QUEUES were chosen to let the rx/tx queues allocated for
116 * the netdevice to be fit in one page. So we can make sure the success of
117 * memory allocation. TODO: increase the limit. */
118 #define MAX_TAP_QUEUES DEFAULT_MAX_NUM_RSS_QUEUES
119 #define MAX_TAP_FLOWS 4096
120
121 #define TUN_FLOW_EXPIRE (3 * HZ)
122
123 /* A tun_file connects an open character device to a tuntap netdevice. It
124 * also contains all socket related structures (except sock_fprog and tap_filter)
125 * to serve as one transmit queue for tuntap device. The sock_fprog and
126 * tap_filter were kept in tun_struct since they were used for filtering for the
127 * netdevice not for a specific queue (at least I didn't see the requirement for
128 * this).
129 *
130 * RCU usage:
131 * The tun_file and tun_struct are loosely coupled, the pointer from one to the
132 * other can only be read while rcu_read_lock or rtnl_lock is held.
133 */
134 struct tun_file {
135 struct sock sk;
136 struct socket socket;
137 struct socket_wq wq;
138 struct tun_struct __rcu *tun;
139 struct net *net;
140 struct fasync_struct *fasync;
141 /* only used for fasnyc */
142 unsigned int flags;
143 union {
144 u16 queue_index;
145 unsigned int ifindex;
146 };
147 struct list_head next;
148 struct tun_struct *detached;
149 };
150
151 struct tun_flow_entry {
152 struct hlist_node hash_link;
153 struct rcu_head rcu;
154 struct tun_struct *tun;
155
156 u32 rxhash;
157 u32 rps_rxhash;
158 int queue_index;
159 unsigned long updated;
160 };
161
162 #define TUN_NUM_FLOW_ENTRIES 1024
163
164 /* Since the socket were moved to tun_file, to preserve the behavior of persist
165 * device, socket filter, sndbuf and vnet header size were restore when the
166 * file were attached to a persist device.
167 */
168 struct tun_struct {
169 struct tun_file __rcu *tfiles[MAX_TAP_QUEUES];
170 unsigned int numqueues;
171 unsigned int flags;
172 kuid_t owner;
173 kgid_t group;
174
175 struct net_device *dev;
176 netdev_features_t set_features;
177 #define TUN_USER_FEATURES (NETIF_F_HW_CSUM|NETIF_F_TSO_ECN|NETIF_F_TSO| \
178 NETIF_F_TSO6|NETIF_F_UFO)
179
180 int vnet_hdr_sz;
181 int sndbuf;
182 struct tap_filter txflt;
183 struct sock_fprog fprog;
184 /* protected by rtnl lock */
185 bool filter_attached;
186 #ifdef TUN_DEBUG
187 int debug;
188 #endif
189 spinlock_t lock;
190 struct hlist_head flows[TUN_NUM_FLOW_ENTRIES];
191 struct timer_list flow_gc_timer;
192 unsigned long ageing_time;
193 unsigned int numdisabled;
194 struct list_head disabled;
195 void *security;
196 u32 flow_count;
197 };
198
199 static inline u32 tun_hashfn(u32 rxhash)
200 {
201 return rxhash & 0x3ff;
202 }
203
204 static struct tun_flow_entry *tun_flow_find(struct hlist_head *head, u32 rxhash)
205 {
206 struct tun_flow_entry *e;
207
208 hlist_for_each_entry_rcu(e, head, hash_link) {
209 if (e->rxhash == rxhash)
210 return e;
211 }
212 return NULL;
213 }
214
215 static struct tun_flow_entry *tun_flow_create(struct tun_struct *tun,
216 struct hlist_head *head,
217 u32 rxhash, u16 queue_index)
218 {
219 struct tun_flow_entry *e = kmalloc(sizeof(*e), GFP_ATOMIC);
220
221 if (e) {
222 tun_debug(KERN_INFO, tun, "create flow: hash %u index %u\n",
223 rxhash, queue_index);
224 e->updated = jiffies;
225 e->rxhash = rxhash;
226 e->rps_rxhash = 0;
227 e->queue_index = queue_index;
228 e->tun = tun;
229 hlist_add_head_rcu(&e->hash_link, head);
230 ++tun->flow_count;
231 }
232 return e;
233 }
234
235 static void tun_flow_delete(struct tun_struct *tun, struct tun_flow_entry *e)
236 {
237 tun_debug(KERN_INFO, tun, "delete flow: hash %u index %u\n",
238 e->rxhash, e->queue_index);
239 sock_rps_reset_flow_hash(e->rps_rxhash);
240 hlist_del_rcu(&e->hash_link);
241 kfree_rcu(e, rcu);
242 --tun->flow_count;
243 }
244
245 static void tun_flow_flush(struct tun_struct *tun)
246 {
247 int i;
248
249 spin_lock_bh(&tun->lock);
250 for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
251 struct tun_flow_entry *e;
252 struct hlist_node *n;
253
254 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link)
255 tun_flow_delete(tun, e);
256 }
257 spin_unlock_bh(&tun->lock);
258 }
259
260 static void tun_flow_delete_by_queue(struct tun_struct *tun, u16 queue_index)
261 {
262 int i;
263
264 spin_lock_bh(&tun->lock);
265 for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
266 struct tun_flow_entry *e;
267 struct hlist_node *n;
268
269 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
270 if (e->queue_index == queue_index)
271 tun_flow_delete(tun, e);
272 }
273 }
274 spin_unlock_bh(&tun->lock);
275 }
276
277 static void tun_flow_cleanup(unsigned long data)
278 {
279 struct tun_struct *tun = (struct tun_struct *)data;
280 unsigned long delay = tun->ageing_time;
281 unsigned long next_timer = jiffies + delay;
282 unsigned long count = 0;
283 int i;
284
285 tun_debug(KERN_INFO, tun, "tun_flow_cleanup\n");
286
287 spin_lock_bh(&tun->lock);
288 for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
289 struct tun_flow_entry *e;
290 struct hlist_node *n;
291
292 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
293 unsigned long this_timer;
294 count++;
295 this_timer = e->updated + delay;
296 if (time_before_eq(this_timer, jiffies))
297 tun_flow_delete(tun, e);
298 else if (time_before(this_timer, next_timer))
299 next_timer = this_timer;
300 }
301 }
302
303 if (count)
304 mod_timer(&tun->flow_gc_timer, round_jiffies_up(next_timer));
305 spin_unlock_bh(&tun->lock);
306 }
307
308 static void tun_flow_update(struct tun_struct *tun, u32 rxhash,
309 struct tun_file *tfile)
310 {
311 struct hlist_head *head;
312 struct tun_flow_entry *e;
313 unsigned long delay = tun->ageing_time;
314 u16 queue_index = tfile->queue_index;
315
316 if (!rxhash)
317 return;
318 else
319 head = &tun->flows[tun_hashfn(rxhash)];
320
321 rcu_read_lock();
322
323 /* We may get a very small possibility of OOO during switching, not
324 * worth to optimize.*/
325 if (tun->numqueues == 1 || tfile->detached)
326 goto unlock;
327
328 e = tun_flow_find(head, rxhash);
329 if (likely(e)) {
330 /* TODO: keep queueing to old queue until it's empty? */
331 e->queue_index = queue_index;
332 e->updated = jiffies;
333 sock_rps_record_flow_hash(e->rps_rxhash);
334 } else {
335 spin_lock_bh(&tun->lock);
336 if (!tun_flow_find(head, rxhash) &&
337 tun->flow_count < MAX_TAP_FLOWS)
338 tun_flow_create(tun, head, rxhash, queue_index);
339
340 if (!timer_pending(&tun->flow_gc_timer))
341 mod_timer(&tun->flow_gc_timer,
342 round_jiffies_up(jiffies + delay));
343 spin_unlock_bh(&tun->lock);
344 }
345
346 unlock:
347 rcu_read_unlock();
348 }
349
350 /**
351 * Save the hash received in the stack receive path and update the
352 * flow_hash table accordingly.
353 */
354 static inline void tun_flow_save_rps_rxhash(struct tun_flow_entry *e, u32 hash)
355 {
356 if (unlikely(e->rps_rxhash != hash)) {
357 sock_rps_reset_flow_hash(e->rps_rxhash);
358 e->rps_rxhash = hash;
359 }
360 }
361
362 /* We try to identify a flow through its rxhash first. The reason that
363 * we do not check rxq no. is because some cards(e.g 82599), chooses
364 * the rxq based on the txq where the last packet of the flow comes. As
365 * the userspace application move between processors, we may get a
366 * different rxq no. here. If we could not get rxhash, then we would
367 * hope the rxq no. may help here.
368 */
369 static u16 tun_select_queue(struct net_device *dev, struct sk_buff *skb,
370 void *accel_priv, select_queue_fallback_t fallback)
371 {
372 struct tun_struct *tun = netdev_priv(dev);
373 struct tun_flow_entry *e;
374 u32 txq = 0;
375 u32 numqueues = 0;
376
377 rcu_read_lock();
378 numqueues = ACCESS_ONCE(tun->numqueues);
379
380 txq = skb_get_hash(skb);
381 if (txq) {
382 e = tun_flow_find(&tun->flows[tun_hashfn(txq)], txq);
383 if (e) {
384 tun_flow_save_rps_rxhash(e, txq);
385 txq = e->queue_index;
386 } else
387 /* use multiply and shift instead of expensive divide */
388 txq = ((u64)txq * numqueues) >> 32;
389 } else if (likely(skb_rx_queue_recorded(skb))) {
390 txq = skb_get_rx_queue(skb);
391 while (unlikely(txq >= numqueues))
392 txq -= numqueues;
393 }
394
395 rcu_read_unlock();
396 return txq;
397 }
398
399 static inline bool tun_not_capable(struct tun_struct *tun)
400 {
401 const struct cred *cred = current_cred();
402 struct net *net = dev_net(tun->dev);
403
404 return ((uid_valid(tun->owner) && !uid_eq(cred->euid, tun->owner)) ||
405 (gid_valid(tun->group) && !in_egroup_p(tun->group))) &&
406 !ns_capable(net->user_ns, CAP_NET_ADMIN);
407 }
408
409 static void tun_set_real_num_queues(struct tun_struct *tun)
410 {
411 netif_set_real_num_tx_queues(tun->dev, tun->numqueues);
412 netif_set_real_num_rx_queues(tun->dev, tun->numqueues);
413 }
414
415 static void tun_disable_queue(struct tun_struct *tun, struct tun_file *tfile)
416 {
417 tfile->detached = tun;
418 list_add_tail(&tfile->next, &tun->disabled);
419 ++tun->numdisabled;
420 }
421
422 static struct tun_struct *tun_enable_queue(struct tun_file *tfile)
423 {
424 struct tun_struct *tun = tfile->detached;
425
426 tfile->detached = NULL;
427 list_del_init(&tfile->next);
428 --tun->numdisabled;
429 return tun;
430 }
431
432 static void tun_queue_purge(struct tun_file *tfile)
433 {
434 skb_queue_purge(&tfile->sk.sk_receive_queue);
435 skb_queue_purge(&tfile->sk.sk_error_queue);
436 }
437
438 static void __tun_detach(struct tun_file *tfile, bool clean)
439 {
440 struct tun_file *ntfile;
441 struct tun_struct *tun;
442
443 tun = rtnl_dereference(tfile->tun);
444
445 if (tun && !tfile->detached) {
446 u16 index = tfile->queue_index;
447 BUG_ON(index >= tun->numqueues);
448
449 rcu_assign_pointer(tun->tfiles[index],
450 tun->tfiles[tun->numqueues - 1]);
451 ntfile = rtnl_dereference(tun->tfiles[index]);
452 ntfile->queue_index = index;
453
454 --tun->numqueues;
455 if (clean) {
456 RCU_INIT_POINTER(tfile->tun, NULL);
457 sock_put(&tfile->sk);
458 } else
459 tun_disable_queue(tun, tfile);
460
461 synchronize_net();
462 tun_flow_delete_by_queue(tun, tun->numqueues + 1);
463 /* Drop read queue */
464 tun_queue_purge(tfile);
465 tun_set_real_num_queues(tun);
466 } else if (tfile->detached && clean) {
467 tun = tun_enable_queue(tfile);
468 sock_put(&tfile->sk);
469 }
470
471 if (clean) {
472 if (tun && tun->numqueues == 0 && tun->numdisabled == 0) {
473 netif_carrier_off(tun->dev);
474
475 if (!(tun->flags & TUN_PERSIST) &&
476 tun->dev->reg_state == NETREG_REGISTERED)
477 unregister_netdevice(tun->dev);
478 }
479
480 BUG_ON(!test_bit(SOCK_EXTERNALLY_ALLOCATED,
481 &tfile->socket.flags));
482 sk_release_kernel(&tfile->sk);
483 }
484 }
485
486 static void tun_detach(struct tun_file *tfile, bool clean)
487 {
488 rtnl_lock();
489 __tun_detach(tfile, clean);
490 rtnl_unlock();
491 }
492
493 static void tun_detach_all(struct net_device *dev)
494 {
495 struct tun_struct *tun = netdev_priv(dev);
496 struct tun_file *tfile, *tmp;
497 int i, n = tun->numqueues;
498
499 for (i = 0; i < n; i++) {
500 tfile = rtnl_dereference(tun->tfiles[i]);
501 BUG_ON(!tfile);
502 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
503 RCU_INIT_POINTER(tfile->tun, NULL);
504 --tun->numqueues;
505 }
506 list_for_each_entry(tfile, &tun->disabled, next) {
507 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
508 RCU_INIT_POINTER(tfile->tun, NULL);
509 }
510 BUG_ON(tun->numqueues != 0);
511
512 synchronize_net();
513 for (i = 0; i < n; i++) {
514 tfile = rtnl_dereference(tun->tfiles[i]);
515 /* Drop read queue */
516 tun_queue_purge(tfile);
517 sock_put(&tfile->sk);
518 }
519 list_for_each_entry_safe(tfile, tmp, &tun->disabled, next) {
520 tun_enable_queue(tfile);
521 tun_queue_purge(tfile);
522 sock_put(&tfile->sk);
523 }
524 BUG_ON(tun->numdisabled != 0);
525
526 if (tun->flags & TUN_PERSIST)
527 module_put(THIS_MODULE);
528 }
529
530 static int tun_attach(struct tun_struct *tun, struct file *file, bool skip_filter)
531 {
532 struct tun_file *tfile = file->private_data;
533 int err;
534
535 err = security_tun_dev_attach(tfile->socket.sk, tun->security);
536 if (err < 0)
537 goto out;
538
539 err = -EINVAL;
540 if (rtnl_dereference(tfile->tun) && !tfile->detached)
541 goto out;
542
543 err = -EBUSY;
544 if (!(tun->flags & TUN_TAP_MQ) && tun->numqueues == 1)
545 goto out;
546
547 err = -E2BIG;
548 if (!tfile->detached &&
549 tun->numqueues + tun->numdisabled == MAX_TAP_QUEUES)
550 goto out;
551
552 err = 0;
553
554 /* Re-attach the filter to persist device */
555 if (!skip_filter && (tun->filter_attached == true)) {
556 err = __sk_attach_filter(&tun->fprog, tfile->socket.sk,
557 lockdep_rtnl_is_held());
558 if (!err)
559 goto out;
560 }
561 tfile->queue_index = tun->numqueues;
562 rcu_assign_pointer(tfile->tun, tun);
563 rcu_assign_pointer(tun->tfiles[tun->numqueues], tfile);
564 tun->numqueues++;
565
566 if (tfile->detached)
567 tun_enable_queue(tfile);
568 else
569 sock_hold(&tfile->sk);
570
571 tun_set_real_num_queues(tun);
572
573 /* device is allowed to go away first, so no need to hold extra
574 * refcnt.
575 */
576
577 out:
578 return err;
579 }
580
581 static struct tun_struct *__tun_get(struct tun_file *tfile)
582 {
583 struct tun_struct *tun;
584
585 rcu_read_lock();
586 tun = rcu_dereference(tfile->tun);
587 if (tun)
588 dev_hold(tun->dev);
589 rcu_read_unlock();
590
591 return tun;
592 }
593
594 static struct tun_struct *tun_get(struct file *file)
595 {
596 return __tun_get(file->private_data);
597 }
598
599 static void tun_put(struct tun_struct *tun)
600 {
601 dev_put(tun->dev);
602 }
603
604 /* TAP filtering */
605 static void addr_hash_set(u32 *mask, const u8 *addr)
606 {
607 int n = ether_crc(ETH_ALEN, addr) >> 26;
608 mask[n >> 5] |= (1 << (n & 31));
609 }
610
611 static unsigned int addr_hash_test(const u32 *mask, const u8 *addr)
612 {
613 int n = ether_crc(ETH_ALEN, addr) >> 26;
614 return mask[n >> 5] & (1 << (n & 31));
615 }
616
617 static int update_filter(struct tap_filter *filter, void __user *arg)
618 {
619 struct { u8 u[ETH_ALEN]; } *addr;
620 struct tun_filter uf;
621 int err, alen, n, nexact;
622
623 if (copy_from_user(&uf, arg, sizeof(uf)))
624 return -EFAULT;
625
626 if (!uf.count) {
627 /* Disabled */
628 filter->count = 0;
629 return 0;
630 }
631
632 alen = ETH_ALEN * uf.count;
633 addr = kmalloc(alen, GFP_KERNEL);
634 if (!addr)
635 return -ENOMEM;
636
637 if (copy_from_user(addr, arg + sizeof(uf), alen)) {
638 err = -EFAULT;
639 goto done;
640 }
641
642 /* The filter is updated without holding any locks. Which is
643 * perfectly safe. We disable it first and in the worst
644 * case we'll accept a few undesired packets. */
645 filter->count = 0;
646 wmb();
647
648 /* Use first set of addresses as an exact filter */
649 for (n = 0; n < uf.count && n < FLT_EXACT_COUNT; n++)
650 memcpy(filter->addr[n], addr[n].u, ETH_ALEN);
651
652 nexact = n;
653
654 /* Remaining multicast addresses are hashed,
655 * unicast will leave the filter disabled. */
656 memset(filter->mask, 0, sizeof(filter->mask));
657 for (; n < uf.count; n++) {
658 if (!is_multicast_ether_addr(addr[n].u)) {
659 err = 0; /* no filter */
660 goto done;
661 }
662 addr_hash_set(filter->mask, addr[n].u);
663 }
664
665 /* For ALLMULTI just set the mask to all ones.
666 * This overrides the mask populated above. */
667 if ((uf.flags & TUN_FLT_ALLMULTI))
668 memset(filter->mask, ~0, sizeof(filter->mask));
669
670 /* Now enable the filter */
671 wmb();
672 filter->count = nexact;
673
674 /* Return the number of exact filters */
675 err = nexact;
676
677 done:
678 kfree(addr);
679 return err;
680 }
681
682 /* Returns: 0 - drop, !=0 - accept */
683 static int run_filter(struct tap_filter *filter, const struct sk_buff *skb)
684 {
685 /* Cannot use eth_hdr(skb) here because skb_mac_hdr() is incorrect
686 * at this point. */
687 struct ethhdr *eh = (struct ethhdr *) skb->data;
688 int i;
689
690 /* Exact match */
691 for (i = 0; i < filter->count; i++)
692 if (ether_addr_equal(eh->h_dest, filter->addr[i]))
693 return 1;
694
695 /* Inexact match (multicast only) */
696 if (is_multicast_ether_addr(eh->h_dest))
697 return addr_hash_test(filter->mask, eh->h_dest);
698
699 return 0;
700 }
701
702 /*
703 * Checks whether the packet is accepted or not.
704 * Returns: 0 - drop, !=0 - accept
705 */
706 static int check_filter(struct tap_filter *filter, const struct sk_buff *skb)
707 {
708 if (!filter->count)
709 return 1;
710
711 return run_filter(filter, skb);
712 }
713
714 /* Network device part of the driver */
715
716 static const struct ethtool_ops tun_ethtool_ops;
717
718 /* Net device detach from fd. */
719 static void tun_net_uninit(struct net_device *dev)
720 {
721 tun_detach_all(dev);
722 }
723
724 /* Net device open. */
725 static int tun_net_open(struct net_device *dev)
726 {
727 netif_tx_start_all_queues(dev);
728 return 0;
729 }
730
731 /* Net device close. */
732 static int tun_net_close(struct net_device *dev)
733 {
734 netif_tx_stop_all_queues(dev);
735 return 0;
736 }
737
738 /* Net device start xmit */
739 static netdev_tx_t tun_net_xmit(struct sk_buff *skb, struct net_device *dev)
740 {
741 struct tun_struct *tun = netdev_priv(dev);
742 int txq = skb->queue_mapping;
743 struct tun_file *tfile;
744 u32 numqueues = 0;
745
746 rcu_read_lock();
747 tfile = rcu_dereference(tun->tfiles[txq]);
748 numqueues = ACCESS_ONCE(tun->numqueues);
749
750 /* Drop packet if interface is not attached */
751 if (txq >= numqueues)
752 goto drop;
753
754 if (numqueues == 1) {
755 /* Select queue was not called for the skbuff, so we extract the
756 * RPS hash and save it into the flow_table here.
757 */
758 __u32 rxhash;
759
760 rxhash = skb_get_hash(skb);
761 if (rxhash) {
762 struct tun_flow_entry *e;
763 e = tun_flow_find(&tun->flows[tun_hashfn(rxhash)],
764 rxhash);
765 if (e)
766 tun_flow_save_rps_rxhash(e, rxhash);
767 }
768 }
769
770 tun_debug(KERN_INFO, tun, "tun_net_xmit %d\n", skb->len);
771
772 BUG_ON(!tfile);
773
774 /* Drop if the filter does not like it.
775 * This is a noop if the filter is disabled.
776 * Filter can be enabled only for the TAP devices. */
777 if (!check_filter(&tun->txflt, skb))
778 goto drop;
779
780 if (tfile->socket.sk->sk_filter &&
781 sk_filter(tfile->socket.sk, skb))
782 goto drop;
783
784 /* Limit the number of packets queued by dividing txq length with the
785 * number of queues.
786 */
787 if (skb_queue_len(&tfile->socket.sk->sk_receive_queue) * numqueues
788 >= dev->tx_queue_len)
789 goto drop;
790
791 if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC)))
792 goto drop;
793
794 if (skb->sk) {
795 sock_tx_timestamp(skb->sk, &skb_shinfo(skb)->tx_flags);
796 sw_tx_timestamp(skb);
797 }
798
799 /* Orphan the skb - required as we might hang on to it
800 * for indefinite time.
801 */
802 skb_orphan(skb);
803
804 nf_reset(skb);
805
806 /* Enqueue packet */
807 skb_queue_tail(&tfile->socket.sk->sk_receive_queue, skb);
808
809 /* Notify and wake up reader process */
810 if (tfile->flags & TUN_FASYNC)
811 kill_fasync(&tfile->fasync, SIGIO, POLL_IN);
812 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
813
814 rcu_read_unlock();
815 return NETDEV_TX_OK;
816
817 drop:
818 dev->stats.tx_dropped++;
819 skb_tx_error(skb);
820 kfree_skb(skb);
821 rcu_read_unlock();
822 return NETDEV_TX_OK;
823 }
824
825 static void tun_net_mclist(struct net_device *dev)
826 {
827 /*
828 * This callback is supposed to deal with mc filter in
829 * _rx_ path and has nothing to do with the _tx_ path.
830 * In rx path we always accept everything userspace gives us.
831 */
832 }
833
834 #define MIN_MTU 68
835 #define MAX_MTU 65535
836
837 static int
838 tun_net_change_mtu(struct net_device *dev, int new_mtu)
839 {
840 if (new_mtu < MIN_MTU || new_mtu + dev->hard_header_len > MAX_MTU)
841 return -EINVAL;
842 dev->mtu = new_mtu;
843 return 0;
844 }
845
846 static netdev_features_t tun_net_fix_features(struct net_device *dev,
847 netdev_features_t features)
848 {
849 struct tun_struct *tun = netdev_priv(dev);
850
851 return (features & tun->set_features) | (features & ~TUN_USER_FEATURES);
852 }
853 #ifdef CONFIG_NET_POLL_CONTROLLER
854 static void tun_poll_controller(struct net_device *dev)
855 {
856 /*
857 * Tun only receives frames when:
858 * 1) the char device endpoint gets data from user space
859 * 2) the tun socket gets a sendmsg call from user space
860 * Since both of those are synchronous operations, we are guaranteed
861 * never to have pending data when we poll for it
862 * so there is nothing to do here but return.
863 * We need this though so netpoll recognizes us as an interface that
864 * supports polling, which enables bridge devices in virt setups to
865 * still use netconsole
866 */
867 return;
868 }
869 #endif
870 static const struct net_device_ops tun_netdev_ops = {
871 .ndo_uninit = tun_net_uninit,
872 .ndo_open = tun_net_open,
873 .ndo_stop = tun_net_close,
874 .ndo_start_xmit = tun_net_xmit,
875 .ndo_change_mtu = tun_net_change_mtu,
876 .ndo_fix_features = tun_net_fix_features,
877 .ndo_select_queue = tun_select_queue,
878 #ifdef CONFIG_NET_POLL_CONTROLLER
879 .ndo_poll_controller = tun_poll_controller,
880 #endif
881 };
882
883 static const struct net_device_ops tap_netdev_ops = {
884 .ndo_uninit = tun_net_uninit,
885 .ndo_open = tun_net_open,
886 .ndo_stop = tun_net_close,
887 .ndo_start_xmit = tun_net_xmit,
888 .ndo_change_mtu = tun_net_change_mtu,
889 .ndo_fix_features = tun_net_fix_features,
890 .ndo_set_rx_mode = tun_net_mclist,
891 .ndo_set_mac_address = eth_mac_addr,
892 .ndo_validate_addr = eth_validate_addr,
893 .ndo_select_queue = tun_select_queue,
894 #ifdef CONFIG_NET_POLL_CONTROLLER
895 .ndo_poll_controller = tun_poll_controller,
896 #endif
897 };
898
899 static void tun_flow_init(struct tun_struct *tun)
900 {
901 int i;
902
903 for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++)
904 INIT_HLIST_HEAD(&tun->flows[i]);
905
906 tun->ageing_time = TUN_FLOW_EXPIRE;
907 setup_timer(&tun->flow_gc_timer, tun_flow_cleanup, (unsigned long)tun);
908 mod_timer(&tun->flow_gc_timer,
909 round_jiffies_up(jiffies + tun->ageing_time));
910 }
911
912 static void tun_flow_uninit(struct tun_struct *tun)
913 {
914 del_timer_sync(&tun->flow_gc_timer);
915 tun_flow_flush(tun);
916 }
917
918 /* Initialize net device. */
919 static void tun_net_init(struct net_device *dev)
920 {
921 struct tun_struct *tun = netdev_priv(dev);
922
923 switch (tun->flags & TUN_TYPE_MASK) {
924 case TUN_TUN_DEV:
925 dev->netdev_ops = &tun_netdev_ops;
926
927 /* Point-to-Point TUN Device */
928 dev->hard_header_len = 0;
929 dev->addr_len = 0;
930 dev->mtu = 1500;
931
932 /* Zero header length */
933 dev->type = ARPHRD_NONE;
934 dev->flags = IFF_POINTOPOINT | IFF_NOARP | IFF_MULTICAST;
935 dev->tx_queue_len = TUN_READQ_SIZE; /* We prefer our own queue length */
936 break;
937
938 case TUN_TAP_DEV:
939 dev->netdev_ops = &tap_netdev_ops;
940 /* Ethernet TAP Device */
941 ether_setup(dev);
942 dev->priv_flags &= ~IFF_TX_SKB_SHARING;
943 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
944
945 eth_hw_addr_random(dev);
946
947 dev->tx_queue_len = TUN_READQ_SIZE; /* We prefer our own queue length */
948 break;
949 }
950 }
951
952 /* Character device part */
953
954 /* Poll */
955 static unsigned int tun_chr_poll(struct file *file, poll_table *wait)
956 {
957 struct tun_file *tfile = file->private_data;
958 struct tun_struct *tun = __tun_get(tfile);
959 struct sock *sk;
960 unsigned int mask = 0;
961
962 if (!tun)
963 return POLLERR;
964
965 sk = tfile->socket.sk;
966
967 tun_debug(KERN_INFO, tun, "tun_chr_poll\n");
968
969 poll_wait(file, sk_sleep(sk), wait);
970
971 if (!skb_queue_empty(&sk->sk_receive_queue))
972 mask |= POLLIN | POLLRDNORM;
973
974 if (sock_writeable(sk) ||
975 (!test_and_set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags) &&
976 sock_writeable(sk)))
977 mask |= POLLOUT | POLLWRNORM;
978
979 if (tun->dev->reg_state != NETREG_REGISTERED)
980 mask = POLLERR;
981
982 tun_put(tun);
983 return mask;
984 }
985
986 /* prepad is the amount to reserve at front. len is length after that.
987 * linear is a hint as to how much to copy (usually headers). */
988 static struct sk_buff *tun_alloc_skb(struct tun_file *tfile,
989 size_t prepad, size_t len,
990 size_t linear, int noblock)
991 {
992 struct sock *sk = tfile->socket.sk;
993 struct sk_buff *skb;
994 int err;
995
996 /* Under a page? Don't bother with paged skb. */
997 if (prepad + len < PAGE_SIZE || !linear)
998 linear = len;
999
1000 skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
1001 &err, 0);
1002 if (!skb)
1003 return ERR_PTR(err);
1004
1005 skb_reserve(skb, prepad);
1006 skb_put(skb, linear);
1007 skb->data_len = len - linear;
1008 skb->len += len - linear;
1009
1010 return skb;
1011 }
1012
1013 /* Get packet from user space buffer */
1014 static ssize_t tun_get_user(struct tun_struct *tun, struct tun_file *tfile,
1015 void *msg_control, const struct iovec *iv,
1016 size_t total_len, size_t count, int noblock)
1017 {
1018 struct tun_pi pi = { 0, cpu_to_be16(ETH_P_IP) };
1019 struct sk_buff *skb;
1020 size_t len = total_len, align = NET_SKB_PAD, linear;
1021 struct virtio_net_hdr gso = { 0 };
1022 int good_linear;
1023 int offset = 0;
1024 int copylen;
1025 bool zerocopy = false;
1026 int err;
1027 u32 rxhash;
1028
1029 if (!(tun->flags & TUN_NO_PI)) {
1030 if (len < sizeof(pi))
1031 return -EINVAL;
1032 len -= sizeof(pi);
1033
1034 if (memcpy_fromiovecend((void *)&pi, iv, 0, sizeof(pi)))
1035 return -EFAULT;
1036 offset += sizeof(pi);
1037 }
1038
1039 if (tun->flags & TUN_VNET_HDR) {
1040 if (len < tun->vnet_hdr_sz)
1041 return -EINVAL;
1042 len -= tun->vnet_hdr_sz;
1043
1044 if (memcpy_fromiovecend((void *)&gso, iv, offset, sizeof(gso)))
1045 return -EFAULT;
1046
1047 if ((gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
1048 gso.csum_start + gso.csum_offset + 2 > gso.hdr_len)
1049 gso.hdr_len = gso.csum_start + gso.csum_offset + 2;
1050
1051 if (gso.hdr_len > len)
1052 return -EINVAL;
1053 offset += tun->vnet_hdr_sz;
1054 }
1055
1056 if ((tun->flags & TUN_TYPE_MASK) == TUN_TAP_DEV) {
1057 align += NET_IP_ALIGN;
1058 if (unlikely(len < ETH_HLEN ||
1059 (gso.hdr_len && gso.hdr_len < ETH_HLEN)))
1060 return -EINVAL;
1061 }
1062
1063 good_linear = SKB_MAX_HEAD(align);
1064
1065 if (msg_control) {
1066 /* There are 256 bytes to be copied in skb, so there is
1067 * enough room for skb expand head in case it is used.
1068 * The rest of the buffer is mapped from userspace.
1069 */
1070 copylen = gso.hdr_len ? gso.hdr_len : GOODCOPY_LEN;
1071 if (copylen > good_linear)
1072 copylen = good_linear;
1073 linear = copylen;
1074 if (iov_pages(iv, offset + copylen, count) <= MAX_SKB_FRAGS)
1075 zerocopy = true;
1076 }
1077
1078 if (!zerocopy) {
1079 copylen = len;
1080 if (gso.hdr_len > good_linear)
1081 linear = good_linear;
1082 else
1083 linear = gso.hdr_len;
1084 }
1085
1086 skb = tun_alloc_skb(tfile, align, copylen, linear, noblock);
1087 if (IS_ERR(skb)) {
1088 if (PTR_ERR(skb) != -EAGAIN)
1089 tun->dev->stats.rx_dropped++;
1090 return PTR_ERR(skb);
1091 }
1092
1093 if (zerocopy)
1094 err = zerocopy_sg_from_iovec(skb, iv, offset, count);
1095 else {
1096 err = skb_copy_datagram_from_iovec(skb, 0, iv, offset, len);
1097 if (!err && msg_control) {
1098 struct ubuf_info *uarg = msg_control;
1099 uarg->callback(uarg, false);
1100 }
1101 }
1102
1103 if (err) {
1104 tun->dev->stats.rx_dropped++;
1105 kfree_skb(skb);
1106 return -EFAULT;
1107 }
1108
1109 if (gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
1110 if (!skb_partial_csum_set(skb, gso.csum_start,
1111 gso.csum_offset)) {
1112 tun->dev->stats.rx_frame_errors++;
1113 kfree_skb(skb);
1114 return -EINVAL;
1115 }
1116 }
1117
1118 switch (tun->flags & TUN_TYPE_MASK) {
1119 case TUN_TUN_DEV:
1120 if (tun->flags & TUN_NO_PI) {
1121 switch (skb->data[0] & 0xf0) {
1122 case 0x40:
1123 pi.proto = htons(ETH_P_IP);
1124 break;
1125 case 0x60:
1126 pi.proto = htons(ETH_P_IPV6);
1127 break;
1128 default:
1129 tun->dev->stats.rx_dropped++;
1130 kfree_skb(skb);
1131 return -EINVAL;
1132 }
1133 }
1134
1135 skb_reset_mac_header(skb);
1136 skb->protocol = pi.proto;
1137 skb->dev = tun->dev;
1138 break;
1139 case TUN_TAP_DEV:
1140 skb->protocol = eth_type_trans(skb, tun->dev);
1141 break;
1142 }
1143
1144 skb_reset_network_header(skb);
1145
1146 if (gso.gso_type != VIRTIO_NET_HDR_GSO_NONE) {
1147 pr_debug("GSO!\n");
1148 switch (gso.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
1149 case VIRTIO_NET_HDR_GSO_TCPV4:
1150 skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
1151 break;
1152 case VIRTIO_NET_HDR_GSO_TCPV6:
1153 skb_shinfo(skb)->gso_type = SKB_GSO_TCPV6;
1154 break;
1155 case VIRTIO_NET_HDR_GSO_UDP:
1156 skb_shinfo(skb)->gso_type = SKB_GSO_UDP;
1157 if (skb->protocol == htons(ETH_P_IPV6))
1158 ipv6_proxy_select_ident(skb);
1159 break;
1160 default:
1161 tun->dev->stats.rx_frame_errors++;
1162 kfree_skb(skb);
1163 return -EINVAL;
1164 }
1165
1166 if (gso.gso_type & VIRTIO_NET_HDR_GSO_ECN)
1167 skb_shinfo(skb)->gso_type |= SKB_GSO_TCP_ECN;
1168
1169 skb_shinfo(skb)->gso_size = gso.gso_size;
1170 if (skb_shinfo(skb)->gso_size == 0) {
1171 tun->dev->stats.rx_frame_errors++;
1172 kfree_skb(skb);
1173 return -EINVAL;
1174 }
1175
1176 /* Header must be checked, and gso_segs computed. */
1177 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
1178 skb_shinfo(skb)->gso_segs = 0;
1179 }
1180
1181 /* copy skb_ubuf_info for callback when skb has no error */
1182 if (zerocopy) {
1183 skb_shinfo(skb)->destructor_arg = msg_control;
1184 skb_shinfo(skb)->tx_flags |= SKBTX_DEV_ZEROCOPY;
1185 skb_shinfo(skb)->tx_flags |= SKBTX_SHARED_FRAG;
1186 }
1187
1188 skb_probe_transport_header(skb, 0);
1189
1190 rxhash = skb_get_hash(skb);
1191 netif_rx_ni(skb);
1192
1193 tun->dev->stats.rx_packets++;
1194 tun->dev->stats.rx_bytes += len;
1195
1196 tun_flow_update(tun, rxhash, tfile);
1197 return total_len;
1198 }
1199
1200 static ssize_t tun_chr_aio_write(struct kiocb *iocb, const struct iovec *iv,
1201 unsigned long count, loff_t pos)
1202 {
1203 struct file *file = iocb->ki_filp;
1204 struct tun_struct *tun = tun_get(file);
1205 struct tun_file *tfile = file->private_data;
1206 ssize_t result;
1207
1208 if (!tun)
1209 return -EBADFD;
1210
1211 tun_debug(KERN_INFO, tun, "tun_chr_write %ld\n", count);
1212
1213 result = tun_get_user(tun, tfile, NULL, iv, iov_length(iv, count),
1214 count, file->f_flags & O_NONBLOCK);
1215
1216 tun_put(tun);
1217 return result;
1218 }
1219
1220 /* Put packet to the user space buffer */
1221 static ssize_t tun_put_user(struct tun_struct *tun,
1222 struct tun_file *tfile,
1223 struct sk_buff *skb,
1224 const struct iovec *iv, int len)
1225 {
1226 struct tun_pi pi = { 0, skb->protocol };
1227 ssize_t total = 0;
1228 int vlan_offset = 0, copied;
1229 int vlan_hlen = 0;
1230
1231 if (vlan_tx_tag_present(skb))
1232 vlan_hlen = VLAN_HLEN;
1233
1234 if (!(tun->flags & TUN_NO_PI)) {
1235 if ((len -= sizeof(pi)) < 0)
1236 return -EINVAL;
1237
1238 if (len < skb->len) {
1239 /* Packet will be striped */
1240 pi.flags |= TUN_PKT_STRIP;
1241 }
1242
1243 if (memcpy_toiovecend(iv, (void *) &pi, 0, sizeof(pi)))
1244 return -EFAULT;
1245 total += sizeof(pi);
1246 }
1247
1248 if (tun->flags & TUN_VNET_HDR) {
1249 struct virtio_net_hdr gso = { 0 }; /* no info leak */
1250 if ((len -= tun->vnet_hdr_sz) < 0)
1251 return -EINVAL;
1252
1253 if (skb_is_gso(skb)) {
1254 struct skb_shared_info *sinfo = skb_shinfo(skb);
1255
1256 /* This is a hint as to how much should be linear. */
1257 gso.hdr_len = skb_headlen(skb);
1258 gso.gso_size = sinfo->gso_size;
1259 if (sinfo->gso_type & SKB_GSO_TCPV4)
1260 gso.gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
1261 else if (sinfo->gso_type & SKB_GSO_TCPV6)
1262 gso.gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
1263 else if (sinfo->gso_type & SKB_GSO_UDP)
1264 gso.gso_type = VIRTIO_NET_HDR_GSO_UDP;
1265 else {
1266 pr_err("unexpected GSO type: "
1267 "0x%x, gso_size %d, hdr_len %d\n",
1268 sinfo->gso_type, gso.gso_size,
1269 gso.hdr_len);
1270 print_hex_dump(KERN_ERR, "tun: ",
1271 DUMP_PREFIX_NONE,
1272 16, 1, skb->head,
1273 min((int)gso.hdr_len, 64), true);
1274 WARN_ON_ONCE(1);
1275 return -EINVAL;
1276 }
1277 if (sinfo->gso_type & SKB_GSO_TCP_ECN)
1278 gso.gso_type |= VIRTIO_NET_HDR_GSO_ECN;
1279 } else
1280 gso.gso_type = VIRTIO_NET_HDR_GSO_NONE;
1281
1282 if (skb->ip_summed == CHECKSUM_PARTIAL) {
1283 gso.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
1284 gso.csum_start = skb_checksum_start_offset(skb) +
1285 vlan_hlen;
1286 gso.csum_offset = skb->csum_offset;
1287 } else if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
1288 gso.flags = VIRTIO_NET_HDR_F_DATA_VALID;
1289 } /* else everything is zero */
1290
1291 if (unlikely(memcpy_toiovecend(iv, (void *)&gso, total,
1292 sizeof(gso))))
1293 return -EFAULT;
1294 total += tun->vnet_hdr_sz;
1295 }
1296
1297 copied = total;
1298 len = min_t(int, skb->len + vlan_hlen, len);
1299 total += skb->len + vlan_hlen;
1300 if (vlan_hlen) {
1301 int copy, ret;
1302 struct {
1303 __be16 h_vlan_proto;
1304 __be16 h_vlan_TCI;
1305 } veth;
1306
1307 veth.h_vlan_proto = skb->vlan_proto;
1308 veth.h_vlan_TCI = htons(vlan_tx_tag_get(skb));
1309
1310 vlan_offset = offsetof(struct vlan_ethhdr, h_vlan_proto);
1311
1312 copy = min_t(int, vlan_offset, len);
1313 ret = skb_copy_datagram_const_iovec(skb, 0, iv, copied, copy);
1314 len -= copy;
1315 copied += copy;
1316 if (ret || !len)
1317 goto done;
1318
1319 copy = min_t(int, sizeof(veth), len);
1320 ret = memcpy_toiovecend(iv, (void *)&veth, copied, copy);
1321 len -= copy;
1322 copied += copy;
1323 if (ret || !len)
1324 goto done;
1325 }
1326
1327 skb_copy_datagram_const_iovec(skb, vlan_offset, iv, copied, len);
1328
1329 done:
1330 tun->dev->stats.tx_packets++;
1331 tun->dev->stats.tx_bytes += len;
1332
1333 return total;
1334 }
1335
1336 static ssize_t tun_do_read(struct tun_struct *tun, struct tun_file *tfile,
1337 const struct iovec *iv, ssize_t len, int noblock)
1338 {
1339 struct sk_buff *skb;
1340 ssize_t ret = 0;
1341 int peeked, err, off = 0;
1342
1343 tun_debug(KERN_INFO, tun, "tun_do_read\n");
1344
1345 if (!len)
1346 return ret;
1347
1348 if (tun->dev->reg_state != NETREG_REGISTERED)
1349 return -EIO;
1350
1351 /* Read frames from queue */
1352 skb = __skb_recv_datagram(tfile->socket.sk, noblock ? MSG_DONTWAIT : 0,
1353 &peeked, &off, &err);
1354 if (skb) {
1355 ret = tun_put_user(tun, tfile, skb, iv, len);
1356 kfree_skb(skb);
1357 } else
1358 ret = err;
1359
1360 return ret;
1361 }
1362
1363 static ssize_t tun_chr_aio_read(struct kiocb *iocb, const struct iovec *iv,
1364 unsigned long count, loff_t pos)
1365 {
1366 struct file *file = iocb->ki_filp;
1367 struct tun_file *tfile = file->private_data;
1368 struct tun_struct *tun = __tun_get(tfile);
1369 ssize_t len, ret;
1370
1371 if (!tun)
1372 return -EBADFD;
1373 len = iov_length(iv, count);
1374 if (len < 0) {
1375 ret = -EINVAL;
1376 goto out;
1377 }
1378
1379 ret = tun_do_read(tun, tfile, iv, len,
1380 file->f_flags & O_NONBLOCK);
1381 ret = min_t(ssize_t, ret, len);
1382 if (ret > 0)
1383 iocb->ki_pos = ret;
1384 out:
1385 tun_put(tun);
1386 return ret;
1387 }
1388
1389 static void tun_free_netdev(struct net_device *dev)
1390 {
1391 struct tun_struct *tun = netdev_priv(dev);
1392
1393 BUG_ON(!(list_empty(&tun->disabled)));
1394 tun_flow_uninit(tun);
1395 security_tun_dev_free_security(tun->security);
1396 free_netdev(dev);
1397 }
1398
1399 static void tun_setup(struct net_device *dev)
1400 {
1401 struct tun_struct *tun = netdev_priv(dev);
1402
1403 tun->owner = INVALID_UID;
1404 tun->group = INVALID_GID;
1405
1406 dev->ethtool_ops = &tun_ethtool_ops;
1407 dev->destructor = tun_free_netdev;
1408 }
1409
1410 /* Trivial set of netlink ops to allow deleting tun or tap
1411 * device with netlink.
1412 */
1413 static int tun_validate(struct nlattr *tb[], struct nlattr *data[])
1414 {
1415 return -EINVAL;
1416 }
1417
1418 static struct rtnl_link_ops tun_link_ops __read_mostly = {
1419 .kind = DRV_NAME,
1420 .priv_size = sizeof(struct tun_struct),
1421 .setup = tun_setup,
1422 .validate = tun_validate,
1423 };
1424
1425 static void tun_sock_write_space(struct sock *sk)
1426 {
1427 struct tun_file *tfile;
1428 wait_queue_head_t *wqueue;
1429
1430 if (!sock_writeable(sk))
1431 return;
1432
1433 if (!test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags))
1434 return;
1435
1436 wqueue = sk_sleep(sk);
1437 if (wqueue && waitqueue_active(wqueue))
1438 wake_up_interruptible_sync_poll(wqueue, POLLOUT |
1439 POLLWRNORM | POLLWRBAND);
1440
1441 tfile = container_of(sk, struct tun_file, sk);
1442 kill_fasync(&tfile->fasync, SIGIO, POLL_OUT);
1443 }
1444
1445 static int tun_sendmsg(struct kiocb *iocb, struct socket *sock,
1446 struct msghdr *m, size_t total_len)
1447 {
1448 int ret;
1449 struct tun_file *tfile = container_of(sock, struct tun_file, socket);
1450 struct tun_struct *tun = __tun_get(tfile);
1451
1452 if (!tun)
1453 return -EBADFD;
1454 ret = tun_get_user(tun, tfile, m->msg_control, m->msg_iov, total_len,
1455 m->msg_iovlen, m->msg_flags & MSG_DONTWAIT);
1456 tun_put(tun);
1457 return ret;
1458 }
1459
1460 static int tun_recvmsg(struct kiocb *iocb, struct socket *sock,
1461 struct msghdr *m, size_t total_len,
1462 int flags)
1463 {
1464 struct tun_file *tfile = container_of(sock, struct tun_file, socket);
1465 struct tun_struct *tun = __tun_get(tfile);
1466 int ret;
1467
1468 if (!tun)
1469 return -EBADFD;
1470
1471 if (flags & ~(MSG_DONTWAIT|MSG_TRUNC|MSG_ERRQUEUE)) {
1472 ret = -EINVAL;
1473 goto out;
1474 }
1475 if (flags & MSG_ERRQUEUE) {
1476 ret = sock_recv_errqueue(sock->sk, m, total_len,
1477 SOL_PACKET, TUN_TX_TIMESTAMP);
1478 goto out;
1479 }
1480 ret = tun_do_read(tun, tfile, m->msg_iov, total_len,
1481 flags & MSG_DONTWAIT);
1482 if (ret > total_len) {
1483 m->msg_flags |= MSG_TRUNC;
1484 ret = flags & MSG_TRUNC ? ret : total_len;
1485 }
1486 out:
1487 tun_put(tun);
1488 return ret;
1489 }
1490
1491 static int tun_release(struct socket *sock)
1492 {
1493 if (sock->sk)
1494 sock_put(sock->sk);
1495 return 0;
1496 }
1497
1498 /* Ops structure to mimic raw sockets with tun */
1499 static const struct proto_ops tun_socket_ops = {
1500 .sendmsg = tun_sendmsg,
1501 .recvmsg = tun_recvmsg,
1502 .release = tun_release,
1503 };
1504
1505 static struct proto tun_proto = {
1506 .name = "tun",
1507 .owner = THIS_MODULE,
1508 .obj_size = sizeof(struct tun_file),
1509 };
1510
1511 static int tun_flags(struct tun_struct *tun)
1512 {
1513 int flags = 0;
1514
1515 if (tun->flags & TUN_TUN_DEV)
1516 flags |= IFF_TUN;
1517 else
1518 flags |= IFF_TAP;
1519
1520 if (tun->flags & TUN_NO_PI)
1521 flags |= IFF_NO_PI;
1522
1523 /* This flag has no real effect. We track the value for backwards
1524 * compatibility.
1525 */
1526 if (tun->flags & TUN_ONE_QUEUE)
1527 flags |= IFF_ONE_QUEUE;
1528
1529 if (tun->flags & TUN_VNET_HDR)
1530 flags |= IFF_VNET_HDR;
1531
1532 if (tun->flags & TUN_TAP_MQ)
1533 flags |= IFF_MULTI_QUEUE;
1534
1535 if (tun->flags & TUN_PERSIST)
1536 flags |= IFF_PERSIST;
1537
1538 return flags;
1539 }
1540
1541 static ssize_t tun_show_flags(struct device *dev, struct device_attribute *attr,
1542 char *buf)
1543 {
1544 struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1545 return sprintf(buf, "0x%x\n", tun_flags(tun));
1546 }
1547
1548 static ssize_t tun_show_owner(struct device *dev, struct device_attribute *attr,
1549 char *buf)
1550 {
1551 struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1552 return uid_valid(tun->owner)?
1553 sprintf(buf, "%u\n",
1554 from_kuid_munged(current_user_ns(), tun->owner)):
1555 sprintf(buf, "-1\n");
1556 }
1557
1558 static ssize_t tun_show_group(struct device *dev, struct device_attribute *attr,
1559 char *buf)
1560 {
1561 struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1562 return gid_valid(tun->group) ?
1563 sprintf(buf, "%u\n",
1564 from_kgid_munged(current_user_ns(), tun->group)):
1565 sprintf(buf, "-1\n");
1566 }
1567
1568 static DEVICE_ATTR(tun_flags, 0444, tun_show_flags, NULL);
1569 static DEVICE_ATTR(owner, 0444, tun_show_owner, NULL);
1570 static DEVICE_ATTR(group, 0444, tun_show_group, NULL);
1571
1572 static int tun_set_iff(struct net *net, struct file *file, struct ifreq *ifr)
1573 {
1574 struct tun_struct *tun;
1575 struct tun_file *tfile = file->private_data;
1576 struct net_device *dev;
1577 int err;
1578
1579 if (tfile->detached)
1580 return -EINVAL;
1581
1582 dev = __dev_get_by_name(net, ifr->ifr_name);
1583 if (dev) {
1584 if (ifr->ifr_flags & IFF_TUN_EXCL)
1585 return -EBUSY;
1586 if ((ifr->ifr_flags & IFF_TUN) && dev->netdev_ops == &tun_netdev_ops)
1587 tun = netdev_priv(dev);
1588 else if ((ifr->ifr_flags & IFF_TAP) && dev->netdev_ops == &tap_netdev_ops)
1589 tun = netdev_priv(dev);
1590 else
1591 return -EINVAL;
1592
1593 if (!!(ifr->ifr_flags & IFF_MULTI_QUEUE) !=
1594 !!(tun->flags & TUN_TAP_MQ))
1595 return -EINVAL;
1596
1597 if (tun_not_capable(tun))
1598 return -EPERM;
1599 err = security_tun_dev_open(tun->security);
1600 if (err < 0)
1601 return err;
1602
1603 err = tun_attach(tun, file, ifr->ifr_flags & IFF_NOFILTER);
1604 if (err < 0)
1605 return err;
1606
1607 if (tun->flags & TUN_TAP_MQ &&
1608 (tun->numqueues + tun->numdisabled > 1)) {
1609 /* One or more queue has already been attached, no need
1610 * to initialize the device again.
1611 */
1612 return 0;
1613 }
1614 }
1615 else {
1616 char *name;
1617 unsigned long flags = 0;
1618 int queues = ifr->ifr_flags & IFF_MULTI_QUEUE ?
1619 MAX_TAP_QUEUES : 1;
1620
1621 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
1622 return -EPERM;
1623 err = security_tun_dev_create();
1624 if (err < 0)
1625 return err;
1626
1627 /* Set dev type */
1628 if (ifr->ifr_flags & IFF_TUN) {
1629 /* TUN device */
1630 flags |= TUN_TUN_DEV;
1631 name = "tun%d";
1632 } else if (ifr->ifr_flags & IFF_TAP) {
1633 /* TAP device */
1634 flags |= TUN_TAP_DEV;
1635 name = "tap%d";
1636 } else
1637 return -EINVAL;
1638
1639 if (*ifr->ifr_name)
1640 name = ifr->ifr_name;
1641
1642 dev = alloc_netdev_mqs(sizeof(struct tun_struct), name,
1643 tun_setup, queues, queues);
1644
1645 if (!dev)
1646 return -ENOMEM;
1647
1648 dev_net_set(dev, net);
1649 dev->rtnl_link_ops = &tun_link_ops;
1650 dev->ifindex = tfile->ifindex;
1651
1652 tun = netdev_priv(dev);
1653 tun->dev = dev;
1654 tun->flags = flags;
1655 tun->txflt.count = 0;
1656 tun->vnet_hdr_sz = sizeof(struct virtio_net_hdr);
1657
1658 tun->filter_attached = false;
1659 tun->sndbuf = tfile->socket.sk->sk_sndbuf;
1660
1661 spin_lock_init(&tun->lock);
1662
1663 err = security_tun_dev_alloc_security(&tun->security);
1664 if (err < 0)
1665 goto err_free_dev;
1666
1667 tun_net_init(dev);
1668 tun_flow_init(tun);
1669
1670 dev->hw_features = NETIF_F_SG | NETIF_F_FRAGLIST |
1671 TUN_USER_FEATURES | NETIF_F_HW_VLAN_CTAG_TX |
1672 NETIF_F_HW_VLAN_STAG_TX;
1673 dev->features = dev->hw_features;
1674 dev->vlan_features = dev->features &
1675 ~(NETIF_F_HW_VLAN_CTAG_TX |
1676 NETIF_F_HW_VLAN_STAG_TX);
1677
1678 INIT_LIST_HEAD(&tun->disabled);
1679 err = tun_attach(tun, file, false);
1680 if (err < 0)
1681 goto err_free_flow;
1682
1683 err = register_netdevice(tun->dev);
1684 if (err < 0)
1685 goto err_detach;
1686
1687 if (device_create_file(&tun->dev->dev, &dev_attr_tun_flags) ||
1688 device_create_file(&tun->dev->dev, &dev_attr_owner) ||
1689 device_create_file(&tun->dev->dev, &dev_attr_group))
1690 pr_err("Failed to create tun sysfs files\n");
1691 }
1692
1693 netif_carrier_on(tun->dev);
1694
1695 tun_debug(KERN_INFO, tun, "tun_set_iff\n");
1696
1697 if (ifr->ifr_flags & IFF_NO_PI)
1698 tun->flags |= TUN_NO_PI;
1699 else
1700 tun->flags &= ~TUN_NO_PI;
1701
1702 /* This flag has no real effect. We track the value for backwards
1703 * compatibility.
1704 */
1705 if (ifr->ifr_flags & IFF_ONE_QUEUE)
1706 tun->flags |= TUN_ONE_QUEUE;
1707 else
1708 tun->flags &= ~TUN_ONE_QUEUE;
1709
1710 if (ifr->ifr_flags & IFF_VNET_HDR)
1711 tun->flags |= TUN_VNET_HDR;
1712 else
1713 tun->flags &= ~TUN_VNET_HDR;
1714
1715 if (ifr->ifr_flags & IFF_MULTI_QUEUE)
1716 tun->flags |= TUN_TAP_MQ;
1717 else
1718 tun->flags &= ~TUN_TAP_MQ;
1719
1720 /* Make sure persistent devices do not get stuck in
1721 * xoff state.
1722 */
1723 if (netif_running(tun->dev))
1724 netif_tx_wake_all_queues(tun->dev);
1725
1726 strcpy(ifr->ifr_name, tun->dev->name);
1727 return 0;
1728
1729 err_detach:
1730 tun_detach_all(dev);
1731 err_free_flow:
1732 tun_flow_uninit(tun);
1733 security_tun_dev_free_security(tun->security);
1734 err_free_dev:
1735 free_netdev(dev);
1736 return err;
1737 }
1738
1739 static void tun_get_iff(struct net *net, struct tun_struct *tun,
1740 struct ifreq *ifr)
1741 {
1742 tun_debug(KERN_INFO, tun, "tun_get_iff\n");
1743
1744 strcpy(ifr->ifr_name, tun->dev->name);
1745
1746 ifr->ifr_flags = tun_flags(tun);
1747
1748 }
1749
1750 /* This is like a cut-down ethtool ops, except done via tun fd so no
1751 * privs required. */
1752 static int set_offload(struct tun_struct *tun, unsigned long arg)
1753 {
1754 netdev_features_t features = 0;
1755
1756 if (arg & TUN_F_CSUM) {
1757 features |= NETIF_F_HW_CSUM;
1758 arg &= ~TUN_F_CSUM;
1759
1760 if (arg & (TUN_F_TSO4|TUN_F_TSO6)) {
1761 if (arg & TUN_F_TSO_ECN) {
1762 features |= NETIF_F_TSO_ECN;
1763 arg &= ~TUN_F_TSO_ECN;
1764 }
1765 if (arg & TUN_F_TSO4)
1766 features |= NETIF_F_TSO;
1767 if (arg & TUN_F_TSO6)
1768 features |= NETIF_F_TSO6;
1769 arg &= ~(TUN_F_TSO4|TUN_F_TSO6);
1770 }
1771
1772 if (arg & TUN_F_UFO) {
1773 features |= NETIF_F_UFO;
1774 arg &= ~TUN_F_UFO;
1775 }
1776 }
1777
1778 /* This gives the user a way to test for new features in future by
1779 * trying to set them. */
1780 if (arg)
1781 return -EINVAL;
1782
1783 tun->set_features = features;
1784 netdev_update_features(tun->dev);
1785
1786 return 0;
1787 }
1788
1789 static void tun_detach_filter(struct tun_struct *tun, int n)
1790 {
1791 int i;
1792 struct tun_file *tfile;
1793
1794 for (i = 0; i < n; i++) {
1795 tfile = rtnl_dereference(tun->tfiles[i]);
1796 __sk_detach_filter(tfile->socket.sk, lockdep_rtnl_is_held());
1797 }
1798
1799 tun->filter_attached = false;
1800 }
1801
1802 static int tun_attach_filter(struct tun_struct *tun)
1803 {
1804 int i, ret = 0;
1805 struct tun_file *tfile;
1806
1807 for (i = 0; i < tun->numqueues; i++) {
1808 tfile = rtnl_dereference(tun->tfiles[i]);
1809 ret = __sk_attach_filter(&tun->fprog, tfile->socket.sk,
1810 lockdep_rtnl_is_held());
1811 if (ret) {
1812 tun_detach_filter(tun, i);
1813 return ret;
1814 }
1815 }
1816
1817 tun->filter_attached = true;
1818 return ret;
1819 }
1820
1821 static void tun_set_sndbuf(struct tun_struct *tun)
1822 {
1823 struct tun_file *tfile;
1824 int i;
1825
1826 for (i = 0; i < tun->numqueues; i++) {
1827 tfile = rtnl_dereference(tun->tfiles[i]);
1828 tfile->socket.sk->sk_sndbuf = tun->sndbuf;
1829 }
1830 }
1831
1832 static int tun_set_queue(struct file *file, struct ifreq *ifr)
1833 {
1834 struct tun_file *tfile = file->private_data;
1835 struct tun_struct *tun;
1836 int ret = 0;
1837
1838 rtnl_lock();
1839
1840 if (ifr->ifr_flags & IFF_ATTACH_QUEUE) {
1841 tun = tfile->detached;
1842 if (!tun) {
1843 ret = -EINVAL;
1844 goto unlock;
1845 }
1846 ret = security_tun_dev_attach_queue(tun->security);
1847 if (ret < 0)
1848 goto unlock;
1849 ret = tun_attach(tun, file, false);
1850 } else if (ifr->ifr_flags & IFF_DETACH_QUEUE) {
1851 tun = rtnl_dereference(tfile->tun);
1852 if (!tun || !(tun->flags & TUN_TAP_MQ) || tfile->detached)
1853 ret = -EINVAL;
1854 else
1855 __tun_detach(tfile, false);
1856 } else
1857 ret = -EINVAL;
1858
1859 unlock:
1860 rtnl_unlock();
1861 return ret;
1862 }
1863
1864 static long __tun_chr_ioctl(struct file *file, unsigned int cmd,
1865 unsigned long arg, int ifreq_len)
1866 {
1867 struct tun_file *tfile = file->private_data;
1868 struct tun_struct *tun;
1869 void __user* argp = (void __user*)arg;
1870 struct ifreq ifr;
1871 kuid_t owner;
1872 kgid_t group;
1873 int sndbuf;
1874 int vnet_hdr_sz;
1875 unsigned int ifindex;
1876 int ret;
1877
1878 if (cmd == TUNSETIFF || cmd == TUNSETQUEUE || _IOC_TYPE(cmd) == 0x89) {
1879 if (copy_from_user(&ifr, argp, ifreq_len))
1880 return -EFAULT;
1881 } else {
1882 memset(&ifr, 0, sizeof(ifr));
1883 }
1884 if (cmd == TUNGETFEATURES) {
1885 /* Currently this just means: "what IFF flags are valid?".
1886 * This is needed because we never checked for invalid flags on
1887 * TUNSETIFF. */
1888 return put_user(IFF_TUN | IFF_TAP | IFF_NO_PI | IFF_ONE_QUEUE |
1889 IFF_VNET_HDR | IFF_MULTI_QUEUE,
1890 (unsigned int __user*)argp);
1891 } else if (cmd == TUNSETQUEUE)
1892 return tun_set_queue(file, &ifr);
1893
1894 ret = 0;
1895 rtnl_lock();
1896
1897 tun = __tun_get(tfile);
1898 if (cmd == TUNSETIFF && !tun) {
1899 ifr.ifr_name[IFNAMSIZ-1] = '\0';
1900
1901 ret = tun_set_iff(tfile->net, file, &ifr);
1902
1903 if (ret)
1904 goto unlock;
1905
1906 if (copy_to_user(argp, &ifr, ifreq_len))
1907 ret = -EFAULT;
1908 goto unlock;
1909 }
1910 if (cmd == TUNSETIFINDEX) {
1911 ret = -EPERM;
1912 if (tun)
1913 goto unlock;
1914
1915 ret = -EFAULT;
1916 if (copy_from_user(&ifindex, argp, sizeof(ifindex)))
1917 goto unlock;
1918
1919 ret = 0;
1920 tfile->ifindex = ifindex;
1921 goto unlock;
1922 }
1923
1924 ret = -EBADFD;
1925 if (!tun)
1926 goto unlock;
1927
1928 tun_debug(KERN_INFO, tun, "tun_chr_ioctl cmd %u\n", cmd);
1929
1930 ret = 0;
1931 switch (cmd) {
1932 case TUNGETIFF:
1933 tun_get_iff(current->nsproxy->net_ns, tun, &ifr);
1934
1935 if (tfile->detached)
1936 ifr.ifr_flags |= IFF_DETACH_QUEUE;
1937 if (!tfile->socket.sk->sk_filter)
1938 ifr.ifr_flags |= IFF_NOFILTER;
1939
1940 if (copy_to_user(argp, &ifr, ifreq_len))
1941 ret = -EFAULT;
1942 break;
1943
1944 case TUNSETNOCSUM:
1945 /* Disable/Enable checksum */
1946
1947 /* [unimplemented] */
1948 tun_debug(KERN_INFO, tun, "ignored: set checksum %s\n",
1949 arg ? "disabled" : "enabled");
1950 break;
1951
1952 case TUNSETPERSIST:
1953 /* Disable/Enable persist mode. Keep an extra reference to the
1954 * module to prevent the module being unprobed.
1955 */
1956 if (arg && !(tun->flags & TUN_PERSIST)) {
1957 tun->flags |= TUN_PERSIST;
1958 __module_get(THIS_MODULE);
1959 }
1960 if (!arg && (tun->flags & TUN_PERSIST)) {
1961 tun->flags &= ~TUN_PERSIST;
1962 module_put(THIS_MODULE);
1963 }
1964
1965 tun_debug(KERN_INFO, tun, "persist %s\n",
1966 arg ? "enabled" : "disabled");
1967 break;
1968
1969 case TUNSETOWNER:
1970 /* Set owner of the device */
1971 owner = make_kuid(current_user_ns(), arg);
1972 if (!uid_valid(owner)) {
1973 ret = -EINVAL;
1974 break;
1975 }
1976 tun->owner = owner;
1977 tun_debug(KERN_INFO, tun, "owner set to %u\n",
1978 from_kuid(&init_user_ns, tun->owner));
1979 break;
1980
1981 case TUNSETGROUP:
1982 /* Set group of the device */
1983 group = make_kgid(current_user_ns(), arg);
1984 if (!gid_valid(group)) {
1985 ret = -EINVAL;
1986 break;
1987 }
1988 tun->group = group;
1989 tun_debug(KERN_INFO, tun, "group set to %u\n",
1990 from_kgid(&init_user_ns, tun->group));
1991 break;
1992
1993 case TUNSETLINK:
1994 /* Only allow setting the type when the interface is down */
1995 if (tun->dev->flags & IFF_UP) {
1996 tun_debug(KERN_INFO, tun,
1997 "Linktype set failed because interface is up\n");
1998 ret = -EBUSY;
1999 } else {
2000 tun->dev->type = (int) arg;
2001 tun_debug(KERN_INFO, tun, "linktype set to %d\n",
2002 tun->dev->type);
2003 ret = 0;
2004 }
2005 break;
2006
2007 #ifdef TUN_DEBUG
2008 case TUNSETDEBUG:
2009 tun->debug = arg;
2010 break;
2011 #endif
2012 case TUNSETOFFLOAD:
2013 ret = set_offload(tun, arg);
2014 break;
2015
2016 case TUNSETTXFILTER:
2017 /* Can be set only for TAPs */
2018 ret = -EINVAL;
2019 if ((tun->flags & TUN_TYPE_MASK) != TUN_TAP_DEV)
2020 break;
2021 ret = update_filter(&tun->txflt, (void __user *)arg);
2022 break;
2023
2024 case SIOCGIFHWADDR:
2025 /* Get hw address */
2026 memcpy(ifr.ifr_hwaddr.sa_data, tun->dev->dev_addr, ETH_ALEN);
2027 ifr.ifr_hwaddr.sa_family = tun->dev->type;
2028 if (copy_to_user(argp, &ifr, ifreq_len))
2029 ret = -EFAULT;
2030 break;
2031
2032 case SIOCSIFHWADDR:
2033 /* Set hw address */
2034 tun_debug(KERN_DEBUG, tun, "set hw address: %pM\n",
2035 ifr.ifr_hwaddr.sa_data);
2036
2037 ret = dev_set_mac_address(tun->dev, &ifr.ifr_hwaddr);
2038 break;
2039
2040 case TUNGETSNDBUF:
2041 sndbuf = tfile->socket.sk->sk_sndbuf;
2042 if (copy_to_user(argp, &sndbuf, sizeof(sndbuf)))
2043 ret = -EFAULT;
2044 break;
2045
2046 case TUNSETSNDBUF:
2047 if (copy_from_user(&sndbuf, argp, sizeof(sndbuf))) {
2048 ret = -EFAULT;
2049 break;
2050 }
2051
2052 tun->sndbuf = sndbuf;
2053 tun_set_sndbuf(tun);
2054 break;
2055
2056 case TUNGETVNETHDRSZ:
2057 vnet_hdr_sz = tun->vnet_hdr_sz;
2058 if (copy_to_user(argp, &vnet_hdr_sz, sizeof(vnet_hdr_sz)))
2059 ret = -EFAULT;
2060 break;
2061
2062 case TUNSETVNETHDRSZ:
2063 if (copy_from_user(&vnet_hdr_sz, argp, sizeof(vnet_hdr_sz))) {
2064 ret = -EFAULT;
2065 break;
2066 }
2067 if (vnet_hdr_sz < (int)sizeof(struct virtio_net_hdr)) {
2068 ret = -EINVAL;
2069 break;
2070 }
2071
2072 tun->vnet_hdr_sz = vnet_hdr_sz;
2073 break;
2074
2075 case TUNATTACHFILTER:
2076 /* Can be set only for TAPs */
2077 ret = -EINVAL;
2078 if ((tun->flags & TUN_TYPE_MASK) != TUN_TAP_DEV)
2079 break;
2080 ret = -EFAULT;
2081 if (copy_from_user(&tun->fprog, argp, sizeof(tun->fprog)))
2082 break;
2083
2084 ret = tun_attach_filter(tun);
2085 break;
2086
2087 case TUNDETACHFILTER:
2088 /* Can be set only for TAPs */
2089 ret = -EINVAL;
2090 if ((tun->flags & TUN_TYPE_MASK) != TUN_TAP_DEV)
2091 break;
2092 ret = 0;
2093 tun_detach_filter(tun, tun->numqueues);
2094 break;
2095
2096 case TUNGETFILTER:
2097 ret = -EINVAL;
2098 if ((tun->flags & TUN_TYPE_MASK) != TUN_TAP_DEV)
2099 break;
2100 ret = -EFAULT;
2101 if (copy_to_user(argp, &tun->fprog, sizeof(tun->fprog)))
2102 break;
2103 ret = 0;
2104 break;
2105
2106 default:
2107 ret = -EINVAL;
2108 break;
2109 }
2110
2111 unlock:
2112 rtnl_unlock();
2113 if (tun)
2114 tun_put(tun);
2115 return ret;
2116 }
2117
2118 static long tun_chr_ioctl(struct file *file,
2119 unsigned int cmd, unsigned long arg)
2120 {
2121 return __tun_chr_ioctl(file, cmd, arg, sizeof (struct ifreq));
2122 }
2123
2124 #ifdef CONFIG_COMPAT
2125 static long tun_chr_compat_ioctl(struct file *file,
2126 unsigned int cmd, unsigned long arg)
2127 {
2128 switch (cmd) {
2129 case TUNSETIFF:
2130 case TUNGETIFF:
2131 case TUNSETTXFILTER:
2132 case TUNGETSNDBUF:
2133 case TUNSETSNDBUF:
2134 case SIOCGIFHWADDR:
2135 case SIOCSIFHWADDR:
2136 arg = (unsigned long)compat_ptr(arg);
2137 break;
2138 default:
2139 arg = (compat_ulong_t)arg;
2140 break;
2141 }
2142
2143 /*
2144 * compat_ifreq is shorter than ifreq, so we must not access beyond
2145 * the end of that structure. All fields that are used in this
2146 * driver are compatible though, we don't need to convert the
2147 * contents.
2148 */
2149 return __tun_chr_ioctl(file, cmd, arg, sizeof(struct compat_ifreq));
2150 }
2151 #endif /* CONFIG_COMPAT */
2152
2153 static int tun_chr_fasync(int fd, struct file *file, int on)
2154 {
2155 struct tun_file *tfile = file->private_data;
2156 int ret;
2157
2158 if ((ret = fasync_helper(fd, file, on, &tfile->fasync)) < 0)
2159 goto out;
2160
2161 if (on) {
2162 ret = __f_setown(file, task_pid(current), PIDTYPE_PID, 0);
2163 if (ret)
2164 goto out;
2165 tfile->flags |= TUN_FASYNC;
2166 } else
2167 tfile->flags &= ~TUN_FASYNC;
2168 ret = 0;
2169 out:
2170 return ret;
2171 }
2172
2173 static int tun_chr_open(struct inode *inode, struct file * file)
2174 {
2175 struct tun_file *tfile;
2176
2177 DBG1(KERN_INFO, "tunX: tun_chr_open\n");
2178
2179 tfile = (struct tun_file *)sk_alloc(&init_net, AF_UNSPEC, GFP_KERNEL,
2180 &tun_proto);
2181 if (!tfile)
2182 return -ENOMEM;
2183 RCU_INIT_POINTER(tfile->tun, NULL);
2184 tfile->net = get_net(current->nsproxy->net_ns);
2185 tfile->flags = 0;
2186 tfile->ifindex = 0;
2187
2188 init_waitqueue_head(&tfile->wq.wait);
2189 RCU_INIT_POINTER(tfile->socket.wq, &tfile->wq);
2190
2191 tfile->socket.file = file;
2192 tfile->socket.ops = &tun_socket_ops;
2193
2194 sock_init_data(&tfile->socket, &tfile->sk);
2195 sk_change_net(&tfile->sk, tfile->net);
2196
2197 tfile->sk.sk_write_space = tun_sock_write_space;
2198 tfile->sk.sk_sndbuf = INT_MAX;
2199
2200 file->private_data = tfile;
2201 set_bit(SOCK_EXTERNALLY_ALLOCATED, &tfile->socket.flags);
2202 INIT_LIST_HEAD(&tfile->next);
2203
2204 sock_set_flag(&tfile->sk, SOCK_ZEROCOPY);
2205
2206 return 0;
2207 }
2208
2209 static int tun_chr_close(struct inode *inode, struct file *file)
2210 {
2211 struct tun_file *tfile = file->private_data;
2212 struct net *net = tfile->net;
2213
2214 tun_detach(tfile, true);
2215 put_net(net);
2216
2217 return 0;
2218 }
2219
2220 #ifdef CONFIG_PROC_FS
2221 static int tun_chr_show_fdinfo(struct seq_file *m, struct file *f)
2222 {
2223 struct tun_struct *tun;
2224 struct ifreq ifr;
2225
2226 memset(&ifr, 0, sizeof(ifr));
2227
2228 rtnl_lock();
2229 tun = tun_get(f);
2230 if (tun)
2231 tun_get_iff(current->nsproxy->net_ns, tun, &ifr);
2232 rtnl_unlock();
2233
2234 if (tun)
2235 tun_put(tun);
2236
2237 return seq_printf(m, "iff:\t%s\n", ifr.ifr_name);
2238 }
2239 #endif
2240
2241 static const struct file_operations tun_fops = {
2242 .owner = THIS_MODULE,
2243 .llseek = no_llseek,
2244 .read = do_sync_read,
2245 .aio_read = tun_chr_aio_read,
2246 .write = do_sync_write,
2247 .aio_write = tun_chr_aio_write,
2248 .poll = tun_chr_poll,
2249 .unlocked_ioctl = tun_chr_ioctl,
2250 #ifdef CONFIG_COMPAT
2251 .compat_ioctl = tun_chr_compat_ioctl,
2252 #endif
2253 .open = tun_chr_open,
2254 .release = tun_chr_close,
2255 .fasync = tun_chr_fasync,
2256 #ifdef CONFIG_PROC_FS
2257 .show_fdinfo = tun_chr_show_fdinfo,
2258 #endif
2259 };
2260
2261 static struct miscdevice tun_miscdev = {
2262 .minor = TUN_MINOR,
2263 .name = "tun",
2264 .nodename = "net/tun",
2265 .fops = &tun_fops,
2266 };
2267
2268 /* ethtool interface */
2269
2270 static int tun_get_settings(struct net_device *dev, struct ethtool_cmd *cmd)
2271 {
2272 cmd->supported = 0;
2273 cmd->advertising = 0;
2274 ethtool_cmd_speed_set(cmd, SPEED_10);
2275 cmd->duplex = DUPLEX_FULL;
2276 cmd->port = PORT_TP;
2277 cmd->phy_address = 0;
2278 cmd->transceiver = XCVR_INTERNAL;
2279 cmd->autoneg = AUTONEG_DISABLE;
2280 cmd->maxtxpkt = 0;
2281 cmd->maxrxpkt = 0;
2282 return 0;
2283 }
2284
2285 static void tun_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
2286 {
2287 struct tun_struct *tun = netdev_priv(dev);
2288
2289 strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
2290 strlcpy(info->version, DRV_VERSION, sizeof(info->version));
2291
2292 switch (tun->flags & TUN_TYPE_MASK) {
2293 case TUN_TUN_DEV:
2294 strlcpy(info->bus_info, "tun", sizeof(info->bus_info));
2295 break;
2296 case TUN_TAP_DEV:
2297 strlcpy(info->bus_info, "tap", sizeof(info->bus_info));
2298 break;
2299 }
2300 }
2301
2302 static u32 tun_get_msglevel(struct net_device *dev)
2303 {
2304 #ifdef TUN_DEBUG
2305 struct tun_struct *tun = netdev_priv(dev);
2306 return tun->debug;
2307 #else
2308 return -EOPNOTSUPP;
2309 #endif
2310 }
2311
2312 static void tun_set_msglevel(struct net_device *dev, u32 value)
2313 {
2314 #ifdef TUN_DEBUG
2315 struct tun_struct *tun = netdev_priv(dev);
2316 tun->debug = value;
2317 #endif
2318 }
2319
2320 static const struct ethtool_ops tun_ethtool_ops = {
2321 .get_settings = tun_get_settings,
2322 .get_drvinfo = tun_get_drvinfo,
2323 .get_msglevel = tun_get_msglevel,
2324 .set_msglevel = tun_set_msglevel,
2325 .get_link = ethtool_op_get_link,
2326 .get_ts_info = ethtool_op_get_ts_info,
2327 };
2328
2329
2330 static int __init tun_init(void)
2331 {
2332 int ret = 0;
2333
2334 pr_info("%s, %s\n", DRV_DESCRIPTION, DRV_VERSION);
2335 pr_info("%s\n", DRV_COPYRIGHT);
2336
2337 ret = rtnl_link_register(&tun_link_ops);
2338 if (ret) {
2339 pr_err("Can't register link_ops\n");
2340 goto err_linkops;
2341 }
2342
2343 ret = misc_register(&tun_miscdev);
2344 if (ret) {
2345 pr_err("Can't register misc device %d\n", TUN_MINOR);
2346 goto err_misc;
2347 }
2348 return 0;
2349 err_misc:
2350 rtnl_link_unregister(&tun_link_ops);
2351 err_linkops:
2352 return ret;
2353 }
2354
2355 static void tun_cleanup(void)
2356 {
2357 misc_deregister(&tun_miscdev);
2358 rtnl_link_unregister(&tun_link_ops);
2359 }
2360
2361 /* Get an underlying socket object from tun file. Returns error unless file is
2362 * attached to a device. The returned object works like a packet socket, it
2363 * can be used for sock_sendmsg/sock_recvmsg. The caller is responsible for
2364 * holding a reference to the file for as long as the socket is in use. */
2365 struct socket *tun_get_socket(struct file *file)
2366 {
2367 struct tun_file *tfile;
2368 if (file->f_op != &tun_fops)
2369 return ERR_PTR(-EINVAL);
2370 tfile = file->private_data;
2371 if (!tfile)
2372 return ERR_PTR(-EBADFD);
2373 return &tfile->socket;
2374 }
2375 EXPORT_SYMBOL_GPL(tun_get_socket);
2376
2377 module_init(tun_init);
2378 module_exit(tun_cleanup);
2379 MODULE_DESCRIPTION(DRV_DESCRIPTION);
2380 MODULE_AUTHOR(DRV_COPYRIGHT);
2381 MODULE_LICENSE("GPL");
2382 MODULE_ALIAS_MISCDEV(TUN_MINOR);
2383 MODULE_ALIAS("devname:net/tun");
Linux with added FPC-III support