?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
drm/omap: Fix error handling path in 'omap_dmm_probe()'
[linux/fpc-iii.git] / drivers / net / tun.c
blob50bfded6d7efb228b5dcae2beea3fb8a80acb295
1 /*
2 * TUN - Universal TUN/TAP device driver.
3 * Copyright (C) 1999-2002 Maxim Krasnyansky <maxk@qualcomm.com>
4 *
5 * This program is free software; you can redistribute it and/or modify
6 * it under the terms of the GNU General Public License as published by
7 * the Free Software Foundation; either version 2 of the License, or
8 * (at your option) any later version.
9 *
10 * This program is distributed in the hope that it will be useful,
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
13 * GNU General Public License for more details.
14 *
15 * $Id: tun.c,v 1.15 2002/03/01 02:44:24 maxk Exp $
16 */
17
18 /*
19 * Changes:
20 *
21 * Mike Kershaw <dragorn@kismetwireless.net> 2005/08/14
22 * Add TUNSETLINK ioctl to set the link encapsulation
23 *
24 * Mark Smith <markzzzsmith@yahoo.com.au>
25 * Use eth_random_addr() for tap MAC address.
26 *
27 * Harald Roelle <harald.roelle@ifi.lmu.de> 2004/04/20
28 * Fixes in packet dropping, queue length setting and queue wakeup.
29 * Increased default tx queue length.
30 * Added ethtool API.
31 * Minor cleanups
32 *
33 * Daniel Podlejski <underley@underley.eu.org>
34 * Modifications for 2.3.99-pre5 kernel.
35 */
36
37 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
38
39 #define DRV_NAME "tun"
40 #define DRV_VERSION "1.6"
41 #define DRV_DESCRIPTION "Universal TUN/TAP device driver"
42 #define DRV_COPYRIGHT "(C) 1999-2004 Max Krasnyansky <maxk@qualcomm.com>"
43
44 #include <linux/module.h>
45 #include <linux/errno.h>
46 #include <linux/kernel.h>
47 #include <linux/major.h>
48 #include <linux/slab.h>
49 #include <linux/poll.h>
50 #include <linux/fcntl.h>
51 #include <linux/init.h>
52 #include <linux/skbuff.h>
53 #include <linux/netdevice.h>
54 #include <linux/etherdevice.h>
55 #include <linux/miscdevice.h>
56 #include <linux/ethtool.h>
57 #include <linux/rtnetlink.h>
58 #include <linux/compat.h>
59 #include <linux/if.h>
60 #include <linux/if_arp.h>
61 #include <linux/if_ether.h>
62 #include <linux/if_tun.h>
63 #include <linux/if_vlan.h>
64 #include <linux/crc32.h>
65 #include <linux/nsproxy.h>
66 #include <linux/virtio_net.h>
67 #include <linux/rcupdate.h>
68 #include <net/net_namespace.h>
69 #include <net/netns/generic.h>
70 #include <net/rtnetlink.h>
71 #include <net/sock.h>
72 #include <linux/seq_file.h>
73 #include <linux/uio.h>
74
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 /* TUN device flags */
107
108 /* IFF_ATTACH_QUEUE is never stored in device flags,
109 * overload it to mean fasync when stored there.
110 */
111 #define TUN_FASYNC IFF_ATTACH_QUEUE
112 /* High bits in flags field are unused. */
113 #define TUN_VNET_LE 0x80000000
114 #define TUN_VNET_BE 0x40000000
115
116 #define TUN_FEATURES (IFF_NO_PI | IFF_ONE_QUEUE | IFF_VNET_HDR | \
117 IFF_MULTI_QUEUE)
118 #define GOODCOPY_LEN 128
119
120 #define FLT_EXACT_COUNT 8
121 struct tap_filter {
122 unsigned int count; /* Number of addrs. Zero means disabled */
123 u32 mask[2]; /* Mask of the hashed addrs */
124 unsigned char addr[FLT_EXACT_COUNT][ETH_ALEN];
125 };
126
127 /* MAX_TAP_QUEUES 256 is chosen to allow rx/tx queues to be equal
128 * to max number of VCPUs in guest. */
129 #define MAX_TAP_QUEUES 256
130 #define MAX_TAP_FLOWS 4096
131
132 #define TUN_FLOW_EXPIRE (3 * HZ)
133
134 /* A tun_file connects an open character device to a tuntap netdevice. It
135 * also contains all socket related structures (except sock_fprog and tap_filter)
136 * to serve as one transmit queue for tuntap device. The sock_fprog and
137 * tap_filter were kept in tun_struct since they were used for filtering for the
138 * netdevice not for a specific queue (at least I didn't see the requirement for
139 * this).
140 *
141 * RCU usage:
142 * The tun_file and tun_struct are loosely coupled, the pointer from one to the
143 * other can only be read while rcu_read_lock or rtnl_lock is held.
144 */
145 struct tun_file {
146 struct sock sk;
147 struct socket socket;
148 struct socket_wq wq;
149 struct tun_struct __rcu *tun;
150 struct fasync_struct *fasync;
151 /* only used for fasnyc */
152 unsigned int flags;
153 union {
154 u16 queue_index;
155 unsigned int ifindex;
156 };
157 struct list_head next;
158 struct tun_struct *detached;
159 };
160
161 struct tun_flow_entry {
162 struct hlist_node hash_link;
163 struct rcu_head rcu;
164 struct tun_struct *tun;
165
166 u32 rxhash;
167 u32 rps_rxhash;
168 int queue_index;
169 unsigned long updated;
170 };
171
172 #define TUN_NUM_FLOW_ENTRIES 1024
173
174 /* Since the socket were moved to tun_file, to preserve the behavior of persist
175 * device, socket filter, sndbuf and vnet header size were restore when the
176 * file were attached to a persist device.
177 */
178 struct tun_struct {
179 struct tun_file __rcu *tfiles[MAX_TAP_QUEUES];
180 unsigned int numqueues;
181 unsigned int flags;
182 kuid_t owner;
183 kgid_t group;
184
185 struct net_device *dev;
186 netdev_features_t set_features;
187 #define TUN_USER_FEATURES (NETIF_F_HW_CSUM|NETIF_F_TSO_ECN|NETIF_F_TSO| \
188 NETIF_F_TSO6|NETIF_F_UFO)
189
190 int vnet_hdr_sz;
191 int sndbuf;
192 struct tap_filter txflt;
193 struct sock_fprog fprog;
194 /* protected by rtnl lock */
195 bool filter_attached;
196 #ifdef TUN_DEBUG
197 int debug;
198 #endif
199 spinlock_t lock;
200 struct hlist_head flows[TUN_NUM_FLOW_ENTRIES];
201 struct timer_list flow_gc_timer;
202 unsigned long ageing_time;
203 unsigned int numdisabled;
204 struct list_head disabled;
205 void *security;
206 u32 flow_count;
207 };
208
209 #ifdef CONFIG_TUN_VNET_CROSS_LE
210 static inline bool tun_legacy_is_little_endian(struct tun_struct *tun)
211 {
212 return tun->flags & TUN_VNET_BE ? false :
213 virtio_legacy_is_little_endian();
214 }
215
216 static long tun_get_vnet_be(struct tun_struct *tun, int __user *argp)
217 {
218 int be = !!(tun->flags & TUN_VNET_BE);
219
220 if (put_user(be, argp))
221 return -EFAULT;
222
223 return 0;
224 }
225
226 static long tun_set_vnet_be(struct tun_struct *tun, int __user *argp)
227 {
228 int be;
229
230 if (get_user(be, argp))
231 return -EFAULT;
232
233 if (be)
234 tun->flags |= TUN_VNET_BE;
235 else
236 tun->flags &= ~TUN_VNET_BE;
237
238 return 0;
239 }
240 #else
241 static inline bool tun_legacy_is_little_endian(struct tun_struct *tun)
242 {
243 return virtio_legacy_is_little_endian();
244 }
245
246 static long tun_get_vnet_be(struct tun_struct *tun, int __user *argp)
247 {
248 return -EINVAL;
249 }
250
251 static long tun_set_vnet_be(struct tun_struct *tun, int __user *argp)
252 {
253 return -EINVAL;
254 }
255 #endif /* CONFIG_TUN_VNET_CROSS_LE */
256
257 static inline bool tun_is_little_endian(struct tun_struct *tun)
258 {
259 return tun->flags & TUN_VNET_LE ||
260 tun_legacy_is_little_endian(tun);
261 }
262
263 static inline u16 tun16_to_cpu(struct tun_struct *tun, __virtio16 val)
264 {
265 return __virtio16_to_cpu(tun_is_little_endian(tun), val);
266 }
267
268 static inline __virtio16 cpu_to_tun16(struct tun_struct *tun, u16 val)
269 {
270 return __cpu_to_virtio16(tun_is_little_endian(tun), val);
271 }
272
273 static inline u32 tun_hashfn(u32 rxhash)
274 {
275 return rxhash & 0x3ff;
276 }
277
278 static struct tun_flow_entry *tun_flow_find(struct hlist_head *head, u32 rxhash)
279 {
280 struct tun_flow_entry *e;
281
282 hlist_for_each_entry_rcu(e, head, hash_link) {
283 if (e->rxhash == rxhash)
284 return e;
285 }
286 return NULL;
287 }
288
289 static struct tun_flow_entry *tun_flow_create(struct tun_struct *tun,
290 struct hlist_head *head,
291 u32 rxhash, u16 queue_index)
292 {
293 struct tun_flow_entry *e = kmalloc(sizeof(*e), GFP_ATOMIC);
294
295 if (e) {
296 tun_debug(KERN_INFO, tun, "create flow: hash %u index %u\n",
297 rxhash, queue_index);
298 e->updated = jiffies;
299 e->rxhash = rxhash;
300 e->rps_rxhash = 0;
301 e->queue_index = queue_index;
302 e->tun = tun;
303 hlist_add_head_rcu(&e->hash_link, head);
304 ++tun->flow_count;
305 }
306 return e;
307 }
308
309 static void tun_flow_delete(struct tun_struct *tun, struct tun_flow_entry *e)
310 {
311 tun_debug(KERN_INFO, tun, "delete flow: hash %u index %u\n",
312 e->rxhash, e->queue_index);
313 hlist_del_rcu(&e->hash_link);
314 kfree_rcu(e, rcu);
315 --tun->flow_count;
316 }
317
318 static void tun_flow_flush(struct tun_struct *tun)
319 {
320 int i;
321
322 spin_lock_bh(&tun->lock);
323 for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
324 struct tun_flow_entry *e;
325 struct hlist_node *n;
326
327 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link)
328 tun_flow_delete(tun, e);
329 }
330 spin_unlock_bh(&tun->lock);
331 }
332
333 static void tun_flow_delete_by_queue(struct tun_struct *tun, u16 queue_index)
334 {
335 int i;
336
337 spin_lock_bh(&tun->lock);
338 for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
339 struct tun_flow_entry *e;
340 struct hlist_node *n;
341
342 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
343 if (e->queue_index == queue_index)
344 tun_flow_delete(tun, e);
345 }
346 }
347 spin_unlock_bh(&tun->lock);
348 }
349
350 static void tun_flow_cleanup(unsigned long data)
351 {
352 struct tun_struct *tun = (struct tun_struct *)data;
353 unsigned long delay = tun->ageing_time;
354 unsigned long next_timer = jiffies + delay;
355 unsigned long count = 0;
356 int i;
357
358 tun_debug(KERN_INFO, tun, "tun_flow_cleanup\n");
359
360 spin_lock_bh(&tun->lock);
361 for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
362 struct tun_flow_entry *e;
363 struct hlist_node *n;
364
365 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
366 unsigned long this_timer;
367 count++;
368 this_timer = e->updated + delay;
369 if (time_before_eq(this_timer, jiffies))
370 tun_flow_delete(tun, e);
371 else if (time_before(this_timer, next_timer))
372 next_timer = this_timer;
373 }
374 }
375
376 if (count)
377 mod_timer(&tun->flow_gc_timer, round_jiffies_up(next_timer));
378 spin_unlock_bh(&tun->lock);
379 }
380
381 static void tun_flow_update(struct tun_struct *tun, u32 rxhash,
382 struct tun_file *tfile)
383 {
384 struct hlist_head *head;
385 struct tun_flow_entry *e;
386 unsigned long delay = tun->ageing_time;
387 u16 queue_index = tfile->queue_index;
388
389 if (!rxhash)
390 return;
391 else
392 head = &tun->flows[tun_hashfn(rxhash)];
393
394 rcu_read_lock();
395
396 /* We may get a very small possibility of OOO during switching, not
397 * worth to optimize.*/
398 if (tun->numqueues == 1 || tfile->detached)
399 goto unlock;
400
401 e = tun_flow_find(head, rxhash);
402 if (likely(e)) {
403 /* TODO: keep queueing to old queue until it's empty? */
404 e->queue_index = queue_index;
405 e->updated = jiffies;
406 sock_rps_record_flow_hash(e->rps_rxhash);
407 } else {
408 spin_lock_bh(&tun->lock);
409 if (!tun_flow_find(head, rxhash) &&
410 tun->flow_count < MAX_TAP_FLOWS)
411 tun_flow_create(tun, head, rxhash, queue_index);
412
413 if (!timer_pending(&tun->flow_gc_timer))
414 mod_timer(&tun->flow_gc_timer,
415 round_jiffies_up(jiffies + delay));
416 spin_unlock_bh(&tun->lock);
417 }
418
419 unlock:
420 rcu_read_unlock();
421 }
422
423 /**
424 * Save the hash received in the stack receive path and update the
425 * flow_hash table accordingly.
426 */
427 static inline void tun_flow_save_rps_rxhash(struct tun_flow_entry *e, u32 hash)
428 {
429 if (unlikely(e->rps_rxhash != hash))
430 e->rps_rxhash = hash;
431 }
432
433 /* We try to identify a flow through its rxhash first. The reason that
434 * we do not check rxq no. is because some cards(e.g 82599), chooses
435 * the rxq based on the txq where the last packet of the flow comes. As
436 * the userspace application move between processors, we may get a
437 * different rxq no. here. If we could not get rxhash, then we would
438 * hope the rxq no. may help here.
439 */
440 static u16 tun_select_queue(struct net_device *dev, struct sk_buff *skb,
441 void *accel_priv, select_queue_fallback_t fallback)
442 {
443 struct tun_struct *tun = netdev_priv(dev);
444 struct tun_flow_entry *e;
445 u32 txq = 0;
446 u32 numqueues = 0;
447
448 rcu_read_lock();
449 numqueues = ACCESS_ONCE(tun->numqueues);
450
451 txq = skb_get_hash(skb);
452 if (txq) {
453 e = tun_flow_find(&tun->flows[tun_hashfn(txq)], txq);
454 if (e) {
455 tun_flow_save_rps_rxhash(e, txq);
456 txq = e->queue_index;
457 } else
458 /* use multiply and shift instead of expensive divide */
459 txq = ((u64)txq * numqueues) >> 32;
460 } else if (likely(skb_rx_queue_recorded(skb))) {
461 txq = skb_get_rx_queue(skb);
462 while (unlikely(txq >= numqueues))
463 txq -= numqueues;
464 }
465
466 rcu_read_unlock();
467 return txq;
468 }
469
470 static inline bool tun_not_capable(struct tun_struct *tun)
471 {
472 const struct cred *cred = current_cred();
473 struct net *net = dev_net(tun->dev);
474
475 return ((uid_valid(tun->owner) && !uid_eq(cred->euid, tun->owner)) ||
476 (gid_valid(tun->group) && !in_egroup_p(tun->group))) &&
477 !ns_capable(net->user_ns, CAP_NET_ADMIN);
478 }
479
480 static void tun_set_real_num_queues(struct tun_struct *tun)
481 {
482 netif_set_real_num_tx_queues(tun->dev, tun->numqueues);
483 netif_set_real_num_rx_queues(tun->dev, tun->numqueues);
484 }
485
486 static void tun_disable_queue(struct tun_struct *tun, struct tun_file *tfile)
487 {
488 tfile->detached = tun;
489 list_add_tail(&tfile->next, &tun->disabled);
490 ++tun->numdisabled;
491 }
492
493 static struct tun_struct *tun_enable_queue(struct tun_file *tfile)
494 {
495 struct tun_struct *tun = tfile->detached;
496
497 tfile->detached = NULL;
498 list_del_init(&tfile->next);
499 --tun->numdisabled;
500 return tun;
501 }
502
503 static void tun_queue_purge(struct tun_file *tfile)
504 {
505 skb_queue_purge(&tfile->sk.sk_receive_queue);
506 skb_queue_purge(&tfile->sk.sk_error_queue);
507 }
508
509 static void __tun_detach(struct tun_file *tfile, bool clean)
510 {
511 struct tun_file *ntfile;
512 struct tun_struct *tun;
513
514 tun = rtnl_dereference(tfile->tun);
515
516 if (tun && !tfile->detached) {
517 u16 index = tfile->queue_index;
518 BUG_ON(index >= tun->numqueues);
519
520 rcu_assign_pointer(tun->tfiles[index],
521 tun->tfiles[tun->numqueues - 1]);
522 ntfile = rtnl_dereference(tun->tfiles[index]);
523 ntfile->queue_index = index;
524
525 --tun->numqueues;
526 if (clean) {
527 RCU_INIT_POINTER(tfile->tun, NULL);
528 sock_put(&tfile->sk);
529 } else
530 tun_disable_queue(tun, tfile);
531
532 synchronize_net();
533 tun_flow_delete_by_queue(tun, tun->numqueues + 1);
534 /* Drop read queue */
535 tun_queue_purge(tfile);
536 tun_set_real_num_queues(tun);
537 } else if (tfile->detached && clean) {
538 tun = tun_enable_queue(tfile);
539 sock_put(&tfile->sk);
540 }
541
542 if (clean) {
543 if (tun && tun->numqueues == 0 && tun->numdisabled == 0) {
544 netif_carrier_off(tun->dev);
545
546 if (!(tun->flags & IFF_PERSIST) &&
547 tun->dev->reg_state == NETREG_REGISTERED)
548 unregister_netdevice(tun->dev);
549 }
550 sock_put(&tfile->sk);
551 }
552 }
553
554 static void tun_detach(struct tun_file *tfile, bool clean)
555 {
556 rtnl_lock();
557 __tun_detach(tfile, clean);
558 rtnl_unlock();
559 }
560
561 static void tun_detach_all(struct net_device *dev)
562 {
563 struct tun_struct *tun = netdev_priv(dev);
564 struct tun_file *tfile, *tmp;
565 int i, n = tun->numqueues;
566
567 for (i = 0; i < n; i++) {
568 tfile = rtnl_dereference(tun->tfiles[i]);
569 BUG_ON(!tfile);
570 tfile->socket.sk->sk_shutdown = RCV_SHUTDOWN;
571 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
572 RCU_INIT_POINTER(tfile->tun, NULL);
573 --tun->numqueues;
574 }
575 list_for_each_entry(tfile, &tun->disabled, next) {
576 tfile->socket.sk->sk_shutdown = RCV_SHUTDOWN;
577 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
578 RCU_INIT_POINTER(tfile->tun, NULL);
579 }
580 BUG_ON(tun->numqueues != 0);
581
582 synchronize_net();
583 for (i = 0; i < n; i++) {
584 tfile = rtnl_dereference(tun->tfiles[i]);
585 /* Drop read queue */
586 tun_queue_purge(tfile);
587 sock_put(&tfile->sk);
588 }
589 list_for_each_entry_safe(tfile, tmp, &tun->disabled, next) {
590 tun_enable_queue(tfile);
591 tun_queue_purge(tfile);
592 sock_put(&tfile->sk);
593 }
594 BUG_ON(tun->numdisabled != 0);
595
596 if (tun->flags & IFF_PERSIST)
597 module_put(THIS_MODULE);
598 }
599
600 static int tun_attach(struct tun_struct *tun, struct file *file, bool skip_filter)
601 {
602 struct tun_file *tfile = file->private_data;
603 int err;
604
605 err = security_tun_dev_attach(tfile->socket.sk, tun->security);
606 if (err < 0)
607 goto out;
608
609 err = -EINVAL;
610 if (rtnl_dereference(tfile->tun) && !tfile->detached)
611 goto out;
612
613 err = -EBUSY;
614 if (!(tun->flags & IFF_MULTI_QUEUE) && tun->numqueues == 1)
615 goto out;
616
617 err = -E2BIG;
618 if (!tfile->detached &&
619 tun->numqueues + tun->numdisabled == MAX_TAP_QUEUES)
620 goto out;
621
622 err = 0;
623
624 /* Re-attach the filter to persist device */
625 if (!skip_filter && (tun->filter_attached == true)) {
626 err = __sk_attach_filter(&tun->fprog, tfile->socket.sk,
627 lockdep_rtnl_is_held());
628 if (!err)
629 goto out;
630 }
631 tfile->queue_index = tun->numqueues;
632 tfile->socket.sk->sk_shutdown &= ~RCV_SHUTDOWN;
633 rcu_assign_pointer(tfile->tun, tun);
634 rcu_assign_pointer(tun->tfiles[tun->numqueues], tfile);
635 tun->numqueues++;
636
637 if (tfile->detached)
638 tun_enable_queue(tfile);
639 else
640 sock_hold(&tfile->sk);
641
642 tun_set_real_num_queues(tun);
643
644 /* device is allowed to go away first, so no need to hold extra
645 * refcnt.
646 */
647
648 out:
649 return err;
650 }
651
652 static struct tun_struct *__tun_get(struct tun_file *tfile)
653 {
654 struct tun_struct *tun;
655
656 rcu_read_lock();
657 tun = rcu_dereference(tfile->tun);
658 if (tun)
659 dev_hold(tun->dev);
660 rcu_read_unlock();
661
662 return tun;
663 }
664
665 static struct tun_struct *tun_get(struct file *file)
666 {
667 return __tun_get(file->private_data);
668 }
669
670 static void tun_put(struct tun_struct *tun)
671 {
672 dev_put(tun->dev);
673 }
674
675 /* TAP filtering */
676 static void addr_hash_set(u32 *mask, const u8 *addr)
677 {
678 int n = ether_crc(ETH_ALEN, addr) >> 26;
679 mask[n >> 5] |= (1 << (n & 31));
680 }
681
682 static unsigned int addr_hash_test(const u32 *mask, const u8 *addr)
683 {
684 int n = ether_crc(ETH_ALEN, addr) >> 26;
685 return mask[n >> 5] & (1 << (n & 31));
686 }
687
688 static int update_filter(struct tap_filter *filter, void __user *arg)
689 {
690 struct { u8 u[ETH_ALEN]; } *addr;
691 struct tun_filter uf;
692 int err, alen, n, nexact;
693
694 if (copy_from_user(&uf, arg, sizeof(uf)))
695 return -EFAULT;
696
697 if (!uf.count) {
698 /* Disabled */
699 filter->count = 0;
700 return 0;
701 }
702
703 alen = ETH_ALEN * uf.count;
704 addr = kmalloc(alen, GFP_KERNEL);
705 if (!addr)
706 return -ENOMEM;
707
708 if (copy_from_user(addr, arg + sizeof(uf), alen)) {
709 err = -EFAULT;
710 goto done;
711 }
712
713 /* The filter is updated without holding any locks. Which is
714 * perfectly safe. We disable it first and in the worst
715 * case we'll accept a few undesired packets. */
716 filter->count = 0;
717 wmb();
718
719 /* Use first set of addresses as an exact filter */
720 for (n = 0; n < uf.count && n < FLT_EXACT_COUNT; n++)
721 memcpy(filter->addr[n], addr[n].u, ETH_ALEN);
722
723 nexact = n;
724
725 /* Remaining multicast addresses are hashed,
726 * unicast will leave the filter disabled. */
727 memset(filter->mask, 0, sizeof(filter->mask));
728 for (; n < uf.count; n++) {
729 if (!is_multicast_ether_addr(addr[n].u)) {
730 err = 0; /* no filter */
731 goto done;
732 }
733 addr_hash_set(filter->mask, addr[n].u);
734 }
735
736 /* For ALLMULTI just set the mask to all ones.
737 * This overrides the mask populated above. */
738 if ((uf.flags & TUN_FLT_ALLMULTI))
739 memset(filter->mask, ~0, sizeof(filter->mask));
740
741 /* Now enable the filter */
742 wmb();
743 filter->count = nexact;
744
745 /* Return the number of exact filters */
746 err = nexact;
747
748 done:
749 kfree(addr);
750 return err;
751 }
752
753 /* Returns: 0 - drop, !=0 - accept */
754 static int run_filter(struct tap_filter *filter, const struct sk_buff *skb)
755 {
756 /* Cannot use eth_hdr(skb) here because skb_mac_hdr() is incorrect
757 * at this point. */
758 struct ethhdr *eh = (struct ethhdr *) skb->data;
759 int i;
760
761 /* Exact match */
762 for (i = 0; i < filter->count; i++)
763 if (ether_addr_equal(eh->h_dest, filter->addr[i]))
764 return 1;
765
766 /* Inexact match (multicast only) */
767 if (is_multicast_ether_addr(eh->h_dest))
768 return addr_hash_test(filter->mask, eh->h_dest);
769
770 return 0;
771 }
772
773 /*
774 * Checks whether the packet is accepted or not.
775 * Returns: 0 - drop, !=0 - accept
776 */
777 static int check_filter(struct tap_filter *filter, const struct sk_buff *skb)
778 {
779 if (!filter->count)
780 return 1;
781
782 return run_filter(filter, skb);
783 }
784
785 /* Network device part of the driver */
786
787 static const struct ethtool_ops tun_ethtool_ops;
788
789 /* Net device detach from fd. */
790 static void tun_net_uninit(struct net_device *dev)
791 {
792 tun_detach_all(dev);
793 }
794
795 /* Net device open. */
796 static int tun_net_open(struct net_device *dev)
797 {
798 netif_tx_start_all_queues(dev);
799 return 0;
800 }
801
802 /* Net device close. */
803 static int tun_net_close(struct net_device *dev)
804 {
805 netif_tx_stop_all_queues(dev);
806 return 0;
807 }
808
809 /* Net device start xmit */
810 static netdev_tx_t tun_net_xmit(struct sk_buff *skb, struct net_device *dev)
811 {
812 struct tun_struct *tun = netdev_priv(dev);
813 int txq = skb->queue_mapping;
814 struct tun_file *tfile;
815 u32 numqueues = 0;
816
817 rcu_read_lock();
818 tfile = rcu_dereference(tun->tfiles[txq]);
819 numqueues = ACCESS_ONCE(tun->numqueues);
820
821 /* Drop packet if interface is not attached */
822 if (txq >= numqueues)
823 goto drop;
824
825 if (numqueues == 1) {
826 /* Select queue was not called for the skbuff, so we extract the
827 * RPS hash and save it into the flow_table here.
828 */
829 __u32 rxhash;
830
831 rxhash = skb_get_hash(skb);
832 if (rxhash) {
833 struct tun_flow_entry *e;
834 e = tun_flow_find(&tun->flows[tun_hashfn(rxhash)],
835 rxhash);
836 if (e)
837 tun_flow_save_rps_rxhash(e, rxhash);
838 }
839 }
840
841 tun_debug(KERN_INFO, tun, "tun_net_xmit %d\n", skb->len);
842
843 BUG_ON(!tfile);
844
845 /* Drop if the filter does not like it.
846 * This is a noop if the filter is disabled.
847 * Filter can be enabled only for the TAP devices. */
848 if (!check_filter(&tun->txflt, skb))
849 goto drop;
850
851 if (tfile->socket.sk->sk_filter &&
852 sk_filter(tfile->socket.sk, skb))
853 goto drop;
854
855 /* Limit the number of packets queued by dividing txq length with the
856 * number of queues.
857 */
858 if (skb_queue_len(&tfile->socket.sk->sk_receive_queue) * numqueues
859 >= dev->tx_queue_len)
860 goto drop;
861
862 if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC)))
863 goto drop;
864
865 if (skb->sk && sk_fullsock(skb->sk)) {
866 sock_tx_timestamp(skb->sk, &skb_shinfo(skb)->tx_flags);
867 sw_tx_timestamp(skb);
868 }
869
870 /* Orphan the skb - required as we might hang on to it
871 * for indefinite time.
872 */
873 skb_orphan(skb);
874
875 nf_reset(skb);
876
877 /* Enqueue packet */
878 skb_queue_tail(&tfile->socket.sk->sk_receive_queue, skb);
879
880 /* Notify and wake up reader process */
881 if (tfile->flags & TUN_FASYNC)
882 kill_fasync(&tfile->fasync, SIGIO, POLL_IN);
883 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
884
885 rcu_read_unlock();
886 return NETDEV_TX_OK;
887
888 drop:
889 dev->stats.tx_dropped++;
890 skb_tx_error(skb);
891 kfree_skb(skb);
892 rcu_read_unlock();
893 return NET_XMIT_DROP;
894 }
895
896 static void tun_net_mclist(struct net_device *dev)
897 {
898 /*
899 * This callback is supposed to deal with mc filter in
900 * _rx_ path and has nothing to do with the _tx_ path.
901 * In rx path we always accept everything userspace gives us.
902 */
903 }
904
905 #define MIN_MTU 68
906 #define MAX_MTU 65535
907
908 static int
909 tun_net_change_mtu(struct net_device *dev, int new_mtu)
910 {
911 if (new_mtu < MIN_MTU || new_mtu + dev->hard_header_len > MAX_MTU)
912 return -EINVAL;
913 dev->mtu = new_mtu;
914 return 0;
915 }
916
917 static netdev_features_t tun_net_fix_features(struct net_device *dev,
918 netdev_features_t features)
919 {
920 struct tun_struct *tun = netdev_priv(dev);
921
922 return (features & tun->set_features) | (features & ~TUN_USER_FEATURES);
923 }
924 #ifdef CONFIG_NET_POLL_CONTROLLER
925 static void tun_poll_controller(struct net_device *dev)
926 {
927 /*
928 * Tun only receives frames when:
929 * 1) the char device endpoint gets data from user space
930 * 2) the tun socket gets a sendmsg call from user space
931 * Since both of those are synchronous operations, we are guaranteed
932 * never to have pending data when we poll for it
933 * so there is nothing to do here but return.
934 * We need this though so netpoll recognizes us as an interface that
935 * supports polling, which enables bridge devices in virt setups to
936 * still use netconsole
937 */
938 return;
939 }
940 #endif
941 static const struct net_device_ops tun_netdev_ops = {
942 .ndo_uninit = tun_net_uninit,
943 .ndo_open = tun_net_open,
944 .ndo_stop = tun_net_close,
945 .ndo_start_xmit = tun_net_xmit,
946 .ndo_change_mtu = tun_net_change_mtu,
947 .ndo_fix_features = tun_net_fix_features,
948 .ndo_select_queue = tun_select_queue,
949 #ifdef CONFIG_NET_POLL_CONTROLLER
950 .ndo_poll_controller = tun_poll_controller,
951 #endif
952 };
953
954 static const struct net_device_ops tap_netdev_ops = {
955 .ndo_uninit = tun_net_uninit,
956 .ndo_open = tun_net_open,
957 .ndo_stop = tun_net_close,
958 .ndo_start_xmit = tun_net_xmit,
959 .ndo_change_mtu = tun_net_change_mtu,
960 .ndo_fix_features = tun_net_fix_features,
961 .ndo_set_rx_mode = tun_net_mclist,
962 .ndo_set_mac_address = eth_mac_addr,
963 .ndo_validate_addr = eth_validate_addr,
964 .ndo_select_queue = tun_select_queue,
965 #ifdef CONFIG_NET_POLL_CONTROLLER
966 .ndo_poll_controller = tun_poll_controller,
967 #endif
968 .ndo_features_check = passthru_features_check,
969 };
970
971 static void tun_flow_init(struct tun_struct *tun)
972 {
973 int i;
974
975 for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++)
976 INIT_HLIST_HEAD(&tun->flows[i]);
977
978 tun->ageing_time = TUN_FLOW_EXPIRE;
979 setup_timer(&tun->flow_gc_timer, tun_flow_cleanup, (unsigned long)tun);
980 mod_timer(&tun->flow_gc_timer,
981 round_jiffies_up(jiffies + tun->ageing_time));
982 }
983
984 static void tun_flow_uninit(struct tun_struct *tun)
985 {
986 del_timer_sync(&tun->flow_gc_timer);
987 tun_flow_flush(tun);
988 }
989
990 /* Initialize net device. */
991 static void tun_net_init(struct net_device *dev)
992 {
993 struct tun_struct *tun = netdev_priv(dev);
994
995 switch (tun->flags & TUN_TYPE_MASK) {
996 case IFF_TUN:
997 dev->netdev_ops = &tun_netdev_ops;
998
999 /* Point-to-Point TUN Device */
1000 dev->hard_header_len = 0;
1001 dev->addr_len = 0;
1002 dev->mtu = 1500;
1003
1004 /* Zero header length */
1005 dev->type = ARPHRD_NONE;
1006 dev->flags = IFF_POINTOPOINT | IFF_NOARP | IFF_MULTICAST;
1007 break;
1008
1009 case IFF_TAP:
1010 dev->netdev_ops = &tap_netdev_ops;
1011 /* Ethernet TAP Device */
1012 ether_setup(dev);
1013 dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1014 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
1015
1016 eth_hw_addr_random(dev);
1017
1018 break;
1019 }
1020 }
1021
1022 /* Character device part */
1023
1024 /* Poll */
1025 static unsigned int tun_chr_poll(struct file *file, poll_table *wait)
1026 {
1027 struct tun_file *tfile = file->private_data;
1028 struct tun_struct *tun = __tun_get(tfile);
1029 struct sock *sk;
1030 unsigned int mask = 0;
1031
1032 if (!tun)
1033 return POLLERR;
1034
1035 sk = tfile->socket.sk;
1036
1037 tun_debug(KERN_INFO, tun, "tun_chr_poll\n");
1038
1039 poll_wait(file, sk_sleep(sk), wait);
1040
1041 if (!skb_queue_empty(&sk->sk_receive_queue))
1042 mask |= POLLIN | POLLRDNORM;
1043
1044 if (sock_writeable(sk) ||
1045 (!test_and_set_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags) &&
1046 sock_writeable(sk)))
1047 mask |= POLLOUT | POLLWRNORM;
1048
1049 if (tun->dev->reg_state != NETREG_REGISTERED)
1050 mask = POLLERR;
1051
1052 tun_put(tun);
1053 return mask;
1054 }
1055
1056 /* prepad is the amount to reserve at front. len is length after that.
1057 * linear is a hint as to how much to copy (usually headers). */
1058 static struct sk_buff *tun_alloc_skb(struct tun_file *tfile,
1059 size_t prepad, size_t len,
1060 size_t linear, int noblock)
1061 {
1062 struct sock *sk = tfile->socket.sk;
1063 struct sk_buff *skb;
1064 int err;
1065
1066 /* Under a page? Don't bother with paged skb. */
1067 if (prepad + len < PAGE_SIZE || !linear)
1068 linear = len;
1069
1070 skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
1071 &err, 0);
1072 if (!skb)
1073 return ERR_PTR(err);
1074
1075 skb_reserve(skb, prepad);
1076 skb_put(skb, linear);
1077 skb->data_len = len - linear;
1078 skb->len += len - linear;
1079
1080 return skb;
1081 }
1082
1083 /* Get packet from user space buffer */
1084 static ssize_t tun_get_user(struct tun_struct *tun, struct tun_file *tfile,
1085 void *msg_control, struct iov_iter *from,
1086 int noblock)
1087 {
1088 struct tun_pi pi = { 0, cpu_to_be16(ETH_P_IP) };
1089 struct sk_buff *skb;
1090 size_t total_len = iov_iter_count(from);
1091 size_t len = total_len, align = NET_SKB_PAD, linear;
1092 struct virtio_net_hdr gso = { 0 };
1093 int good_linear;
1094 int copylen;
1095 bool zerocopy = false;
1096 int err;
1097 u32 rxhash;
1098 ssize_t n;
1099
1100 if (!(tun->flags & IFF_NO_PI)) {
1101 if (len < sizeof(pi))
1102 return -EINVAL;
1103 len -= sizeof(pi);
1104
1105 n = copy_from_iter(&pi, sizeof(pi), from);
1106 if (n != sizeof(pi))
1107 return -EFAULT;
1108 }
1109
1110 if (tun->flags & IFF_VNET_HDR) {
1111 int vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
1112
1113 if (len < vnet_hdr_sz)
1114 return -EINVAL;
1115 len -= vnet_hdr_sz;
1116
1117 n = copy_from_iter(&gso, sizeof(gso), from);
1118 if (n != sizeof(gso))
1119 return -EFAULT;
1120
1121 if ((gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
1122 tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2 > tun16_to_cpu(tun, gso.hdr_len))
1123 gso.hdr_len = cpu_to_tun16(tun, tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2);
1124
1125 if (tun16_to_cpu(tun, gso.hdr_len) > len)
1126 return -EINVAL;
1127 iov_iter_advance(from, vnet_hdr_sz - sizeof(gso));
1128 }
1129
1130 if ((tun->flags & TUN_TYPE_MASK) == IFF_TAP) {
1131 align += NET_IP_ALIGN;
1132 if (unlikely(len < ETH_HLEN ||
1133 (gso.hdr_len && tun16_to_cpu(tun, gso.hdr_len) < ETH_HLEN)))
1134 return -EINVAL;
1135 }
1136
1137 good_linear = SKB_MAX_HEAD(align);
1138
1139 if (msg_control) {
1140 struct iov_iter i = *from;
1141
1142 /* There are 256 bytes to be copied in skb, so there is
1143 * enough room for skb expand head in case it is used.
1144 * The rest of the buffer is mapped from userspace.
1145 */
1146 copylen = gso.hdr_len ? tun16_to_cpu(tun, gso.hdr_len) : GOODCOPY_LEN;
1147 if (copylen > good_linear)
1148 copylen = good_linear;
1149 linear = copylen;
1150 iov_iter_advance(&i, copylen);
1151 if (iov_iter_npages(&i, INT_MAX) <= MAX_SKB_FRAGS)
1152 zerocopy = true;
1153 }
1154
1155 if (!zerocopy) {
1156 copylen = len;
1157 if (tun16_to_cpu(tun, gso.hdr_len) > good_linear)
1158 linear = good_linear;
1159 else
1160 linear = tun16_to_cpu(tun, gso.hdr_len);
1161 }
1162
1163 skb = tun_alloc_skb(tfile, align, copylen, linear, noblock);
1164 if (IS_ERR(skb)) {
1165 if (PTR_ERR(skb) != -EAGAIN)
1166 tun->dev->stats.rx_dropped++;
1167 return PTR_ERR(skb);
1168 }
1169
1170 if (zerocopy)
1171 err = zerocopy_sg_from_iter(skb, from);
1172 else {
1173 err = skb_copy_datagram_from_iter(skb, 0, from, len);
1174 if (!err && msg_control) {
1175 struct ubuf_info *uarg = msg_control;
1176 uarg->callback(uarg, false);
1177 }
1178 }
1179
1180 if (err) {
1181 tun->dev->stats.rx_dropped++;
1182 kfree_skb(skb);
1183 return -EFAULT;
1184 }
1185
1186 if (gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
1187 if (!skb_partial_csum_set(skb, tun16_to_cpu(tun, gso.csum_start),
1188 tun16_to_cpu(tun, gso.csum_offset))) {
1189 tun->dev->stats.rx_frame_errors++;
1190 kfree_skb(skb);
1191 return -EINVAL;
1192 }
1193 }
1194
1195 switch (tun->flags & TUN_TYPE_MASK) {
1196 case IFF_TUN:
1197 if (tun->flags & IFF_NO_PI) {
1198 u8 ip_version = skb->len ? (skb->data[0] >> 4) : 0;
1199
1200 switch (ip_version) {
1201 case 4:
1202 pi.proto = htons(ETH_P_IP);
1203 break;
1204 case 6:
1205 pi.proto = htons(ETH_P_IPV6);
1206 break;
1207 default:
1208 tun->dev->stats.rx_dropped++;
1209 kfree_skb(skb);
1210 return -EINVAL;
1211 }
1212 }
1213
1214 skb_reset_mac_header(skb);
1215 skb->protocol = pi.proto;
1216 skb->dev = tun->dev;
1217 break;
1218 case IFF_TAP:
1219 skb->protocol = eth_type_trans(skb, tun->dev);
1220 break;
1221 }
1222
1223 if (gso.gso_type != VIRTIO_NET_HDR_GSO_NONE) {
1224 pr_debug("GSO!\n");
1225 switch (gso.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
1226 case VIRTIO_NET_HDR_GSO_TCPV4:
1227 skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
1228 break;
1229 case VIRTIO_NET_HDR_GSO_TCPV6:
1230 skb_shinfo(skb)->gso_type = SKB_GSO_TCPV6;
1231 break;
1232 case VIRTIO_NET_HDR_GSO_UDP:
1233 skb_shinfo(skb)->gso_type = SKB_GSO_UDP;
1234 break;
1235 default:
1236 tun->dev->stats.rx_frame_errors++;
1237 kfree_skb(skb);
1238 return -EINVAL;
1239 }
1240
1241 if (gso.gso_type & VIRTIO_NET_HDR_GSO_ECN)
1242 skb_shinfo(skb)->gso_type |= SKB_GSO_TCP_ECN;
1243
1244 skb_shinfo(skb)->gso_size = tun16_to_cpu(tun, gso.gso_size);
1245 if (skb_shinfo(skb)->gso_size == 0) {
1246 tun->dev->stats.rx_frame_errors++;
1247 kfree_skb(skb);
1248 return -EINVAL;
1249 }
1250
1251 /* Header must be checked, and gso_segs computed. */
1252 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
1253 skb_shinfo(skb)->gso_segs = 0;
1254 }
1255
1256 /* copy skb_ubuf_info for callback when skb has no error */
1257 if (zerocopy) {
1258 skb_shinfo(skb)->destructor_arg = msg_control;
1259 skb_shinfo(skb)->tx_flags |= SKBTX_DEV_ZEROCOPY;
1260 skb_shinfo(skb)->tx_flags |= SKBTX_SHARED_FRAG;
1261 }
1262
1263 skb_reset_network_header(skb);
1264 skb_probe_transport_header(skb, 0);
1265
1266 rxhash = skb_get_hash(skb);
1267 netif_rx_ni(skb);
1268
1269 tun->dev->stats.rx_packets++;
1270 tun->dev->stats.rx_bytes += len;
1271
1272 tun_flow_update(tun, rxhash, tfile);
1273 return total_len;
1274 }
1275
1276 static ssize_t tun_chr_write_iter(struct kiocb *iocb, struct iov_iter *from)
1277 {
1278 struct file *file = iocb->ki_filp;
1279 struct tun_struct *tun = tun_get(file);
1280 struct tun_file *tfile = file->private_data;
1281 ssize_t result;
1282
1283 if (!tun)
1284 return -EBADFD;
1285
1286 result = tun_get_user(tun, tfile, NULL, from, file->f_flags & O_NONBLOCK);
1287
1288 tun_put(tun);
1289 return result;
1290 }
1291
1292 /* Put packet to the user space buffer */
1293 static ssize_t tun_put_user(struct tun_struct *tun,
1294 struct tun_file *tfile,
1295 struct sk_buff *skb,
1296 struct iov_iter *iter)
1297 {
1298 struct tun_pi pi = { 0, skb->protocol };
1299 ssize_t total;
1300 int vlan_offset = 0;
1301 int vlan_hlen = 0;
1302 int vnet_hdr_sz = 0;
1303
1304 if (skb_vlan_tag_present(skb))
1305 vlan_hlen = VLAN_HLEN;
1306
1307 if (tun->flags & IFF_VNET_HDR)
1308 vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
1309
1310 total = skb->len + vlan_hlen + vnet_hdr_sz;
1311
1312 if (!(tun->flags & IFF_NO_PI)) {
1313 if (iov_iter_count(iter) < sizeof(pi))
1314 return -EINVAL;
1315
1316 total += sizeof(pi);
1317 if (iov_iter_count(iter) < total) {
1318 /* Packet will be striped */
1319 pi.flags |= TUN_PKT_STRIP;
1320 }
1321
1322 if (copy_to_iter(&pi, sizeof(pi), iter) != sizeof(pi))
1323 return -EFAULT;
1324 }
1325
1326 if (vnet_hdr_sz) {
1327 struct virtio_net_hdr gso = { 0 }; /* no info leak */
1328 if (iov_iter_count(iter) < vnet_hdr_sz)
1329 return -EINVAL;
1330
1331 if (skb_is_gso(skb)) {
1332 struct skb_shared_info *sinfo = skb_shinfo(skb);
1333
1334 /* This is a hint as to how much should be linear. */
1335 gso.hdr_len = cpu_to_tun16(tun, skb_headlen(skb));
1336 gso.gso_size = cpu_to_tun16(tun, sinfo->gso_size);
1337 if (sinfo->gso_type & SKB_GSO_TCPV4)
1338 gso.gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
1339 else if (sinfo->gso_type & SKB_GSO_TCPV6)
1340 gso.gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
1341 else if (sinfo->gso_type & SKB_GSO_UDP)
1342 gso.gso_type = VIRTIO_NET_HDR_GSO_UDP;
1343 else {
1344 pr_err("unexpected GSO type: "
1345 "0x%x, gso_size %d, hdr_len %d\n",
1346 sinfo->gso_type, tun16_to_cpu(tun, gso.gso_size),
1347 tun16_to_cpu(tun, gso.hdr_len));
1348 print_hex_dump(KERN_ERR, "tun: ",
1349 DUMP_PREFIX_NONE,
1350 16, 1, skb->head,
1351 min((int)tun16_to_cpu(tun, gso.hdr_len), 64), true);
1352 WARN_ON_ONCE(1);
1353 return -EINVAL;
1354 }
1355 if (sinfo->gso_type & SKB_GSO_TCP_ECN)
1356 gso.gso_type |= VIRTIO_NET_HDR_GSO_ECN;
1357 } else
1358 gso.gso_type = VIRTIO_NET_HDR_GSO_NONE;
1359
1360 if (skb->ip_summed == CHECKSUM_PARTIAL) {
1361 gso.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
1362 gso.csum_start = cpu_to_tun16(tun, skb_checksum_start_offset(skb) +
1363 vlan_hlen);
1364 gso.csum_offset = cpu_to_tun16(tun, skb->csum_offset);
1365 } else if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
1366 gso.flags = VIRTIO_NET_HDR_F_DATA_VALID;
1367 } /* else everything is zero */
1368
1369 if (copy_to_iter(&gso, sizeof(gso), iter) != sizeof(gso))
1370 return -EFAULT;
1371
1372 iov_iter_advance(iter, vnet_hdr_sz - sizeof(gso));
1373 }
1374
1375 if (vlan_hlen) {
1376 int ret;
1377 struct {
1378 __be16 h_vlan_proto;
1379 __be16 h_vlan_TCI;
1380 } veth;
1381
1382 veth.h_vlan_proto = skb->vlan_proto;
1383 veth.h_vlan_TCI = htons(skb_vlan_tag_get(skb));
1384
1385 vlan_offset = offsetof(struct vlan_ethhdr, h_vlan_proto);
1386
1387 ret = skb_copy_datagram_iter(skb, 0, iter, vlan_offset);
1388 if (ret || !iov_iter_count(iter))
1389 goto done;
1390
1391 ret = copy_to_iter(&veth, sizeof(veth), iter);
1392 if (ret != sizeof(veth) || !iov_iter_count(iter))
1393 goto done;
1394 }
1395
1396 skb_copy_datagram_iter(skb, vlan_offset, iter, skb->len - vlan_offset);
1397
1398 done:
1399 tun->dev->stats.tx_packets++;
1400 tun->dev->stats.tx_bytes += skb->len + vlan_hlen;
1401
1402 return total;
1403 }
1404
1405 static ssize_t tun_do_read(struct tun_struct *tun, struct tun_file *tfile,
1406 struct iov_iter *to,
1407 int noblock)
1408 {
1409 struct sk_buff *skb;
1410 ssize_t ret;
1411 int peeked, err, off = 0;
1412
1413 tun_debug(KERN_INFO, tun, "tun_do_read\n");
1414
1415 if (!iov_iter_count(to))
1416 return 0;
1417
1418 /* Read frames from queue */
1419 skb = __skb_recv_datagram(tfile->socket.sk, noblock ? MSG_DONTWAIT : 0,
1420 &peeked, &off, &err);
1421 if (!skb)
1422 return err;
1423
1424 ret = tun_put_user(tun, tfile, skb, to);
1425 if (unlikely(ret < 0))
1426 kfree_skb(skb);
1427 else
1428 consume_skb(skb);
1429
1430 return ret;
1431 }
1432
1433 static ssize_t tun_chr_read_iter(struct kiocb *iocb, struct iov_iter *to)
1434 {
1435 struct file *file = iocb->ki_filp;
1436 struct tun_file *tfile = file->private_data;
1437 struct tun_struct *tun = __tun_get(tfile);
1438 ssize_t len = iov_iter_count(to), ret;
1439
1440 if (!tun)
1441 return -EBADFD;
1442 ret = tun_do_read(tun, tfile, to, file->f_flags & O_NONBLOCK);
1443 ret = min_t(ssize_t, ret, len);
1444 if (ret > 0)
1445 iocb->ki_pos = ret;
1446 tun_put(tun);
1447 return ret;
1448 }
1449
1450 static void tun_free_netdev(struct net_device *dev)
1451 {
1452 struct tun_struct *tun = netdev_priv(dev);
1453
1454 BUG_ON(!(list_empty(&tun->disabled)));
1455 tun_flow_uninit(tun);
1456 security_tun_dev_free_security(tun->security);
1457 free_netdev(dev);
1458 }
1459
1460 static void tun_setup(struct net_device *dev)
1461 {
1462 struct tun_struct *tun = netdev_priv(dev);
1463
1464 tun->owner = INVALID_UID;
1465 tun->group = INVALID_GID;
1466
1467 dev->ethtool_ops = &tun_ethtool_ops;
1468 dev->destructor = tun_free_netdev;
1469 /* We prefer our own queue length */
1470 dev->tx_queue_len = TUN_READQ_SIZE;
1471 }
1472
1473 /* Trivial set of netlink ops to allow deleting tun or tap
1474 * device with netlink.
1475 */
1476 static int tun_validate(struct nlattr *tb[], struct nlattr *data[])
1477 {
1478 return -EINVAL;
1479 }
1480
1481 static struct rtnl_link_ops tun_link_ops __read_mostly = {
1482 .kind = DRV_NAME,
1483 .priv_size = sizeof(struct tun_struct),
1484 .setup = tun_setup,
1485 .validate = tun_validate,
1486 };
1487
1488 static void tun_sock_write_space(struct sock *sk)
1489 {
1490 struct tun_file *tfile;
1491 wait_queue_head_t *wqueue;
1492
1493 if (!sock_writeable(sk))
1494 return;
1495
1496 if (!test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags))
1497 return;
1498
1499 wqueue = sk_sleep(sk);
1500 if (wqueue && waitqueue_active(wqueue))
1501 wake_up_interruptible_sync_poll(wqueue, POLLOUT |
1502 POLLWRNORM | POLLWRBAND);
1503
1504 tfile = container_of(sk, struct tun_file, sk);
1505 kill_fasync(&tfile->fasync, SIGIO, POLL_OUT);
1506 }
1507
1508 static int tun_sendmsg(struct socket *sock, struct msghdr *m, size_t total_len)
1509 {
1510 int ret;
1511 struct tun_file *tfile = container_of(sock, struct tun_file, socket);
1512 struct tun_struct *tun = __tun_get(tfile);
1513
1514 if (!tun)
1515 return -EBADFD;
1516
1517 ret = tun_get_user(tun, tfile, m->msg_control, &m->msg_iter,
1518 m->msg_flags & MSG_DONTWAIT);
1519 tun_put(tun);
1520 return ret;
1521 }
1522
1523 static int tun_recvmsg(struct socket *sock, struct msghdr *m, size_t total_len,
1524 int flags)
1525 {
1526 struct tun_file *tfile = container_of(sock, struct tun_file, socket);
1527 struct tun_struct *tun = __tun_get(tfile);
1528 int ret;
1529
1530 if (!tun)
1531 return -EBADFD;
1532
1533 if (flags & ~(MSG_DONTWAIT|MSG_TRUNC|MSG_ERRQUEUE)) {
1534 ret = -EINVAL;
1535 goto out;
1536 }
1537 if (flags & MSG_ERRQUEUE) {
1538 ret = sock_recv_errqueue(sock->sk, m, total_len,
1539 SOL_PACKET, TUN_TX_TIMESTAMP);
1540 goto out;
1541 }
1542 ret = tun_do_read(tun, tfile, &m->msg_iter, flags & MSG_DONTWAIT);
1543 if (ret > (ssize_t)total_len) {
1544 m->msg_flags |= MSG_TRUNC;
1545 ret = flags & MSG_TRUNC ? ret : total_len;
1546 }
1547 out:
1548 tun_put(tun);
1549 return ret;
1550 }
1551
1552 /* Ops structure to mimic raw sockets with tun */
1553 static const struct proto_ops tun_socket_ops = {
1554 .sendmsg = tun_sendmsg,
1555 .recvmsg = tun_recvmsg,
1556 };
1557
1558 static struct proto tun_proto = {
1559 .name = "tun",
1560 .owner = THIS_MODULE,
1561 .obj_size = sizeof(struct tun_file),
1562 };
1563
1564 static int tun_flags(struct tun_struct *tun)
1565 {
1566 return tun->flags & (TUN_FEATURES | IFF_PERSIST | IFF_TUN | IFF_TAP);
1567 }
1568
1569 static ssize_t tun_show_flags(struct device *dev, struct device_attribute *attr,
1570 char *buf)
1571 {
1572 struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1573 return sprintf(buf, "0x%x\n", tun_flags(tun));
1574 }
1575
1576 static ssize_t tun_show_owner(struct device *dev, struct device_attribute *attr,
1577 char *buf)
1578 {
1579 struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1580 return uid_valid(tun->owner)?
1581 sprintf(buf, "%u\n",
1582 from_kuid_munged(current_user_ns(), tun->owner)):
1583 sprintf(buf, "-1\n");
1584 }
1585
1586 static ssize_t tun_show_group(struct device *dev, struct device_attribute *attr,
1587 char *buf)
1588 {
1589 struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1590 return gid_valid(tun->group) ?
1591 sprintf(buf, "%u\n",
1592 from_kgid_munged(current_user_ns(), tun->group)):
1593 sprintf(buf, "-1\n");
1594 }
1595
1596 static DEVICE_ATTR(tun_flags, 0444, tun_show_flags, NULL);
1597 static DEVICE_ATTR(owner, 0444, tun_show_owner, NULL);
1598 static DEVICE_ATTR(group, 0444, tun_show_group, NULL);
1599
1600 static struct attribute *tun_dev_attrs[] = {
1601 &dev_attr_tun_flags.attr,
1602 &dev_attr_owner.attr,
1603 &dev_attr_group.attr,
1604 NULL
1605 };
1606
1607 static const struct attribute_group tun_attr_group = {
1608 .attrs = tun_dev_attrs
1609 };
1610
1611 static int tun_set_iff(struct net *net, struct file *file, struct ifreq *ifr)
1612 {
1613 struct tun_struct *tun;
1614 struct tun_file *tfile = file->private_data;
1615 struct net_device *dev;
1616 int err;
1617
1618 if (tfile->detached)
1619 return -EINVAL;
1620
1621 dev = __dev_get_by_name(net, ifr->ifr_name);
1622 if (dev) {
1623 if (ifr->ifr_flags & IFF_TUN_EXCL)
1624 return -EBUSY;
1625 if ((ifr->ifr_flags & IFF_TUN) && dev->netdev_ops == &tun_netdev_ops)
1626 tun = netdev_priv(dev);
1627 else if ((ifr->ifr_flags & IFF_TAP) && dev->netdev_ops == &tap_netdev_ops)
1628 tun = netdev_priv(dev);
1629 else
1630 return -EINVAL;
1631
1632 if (!!(ifr->ifr_flags & IFF_MULTI_QUEUE) !=
1633 !!(tun->flags & IFF_MULTI_QUEUE))
1634 return -EINVAL;
1635
1636 if (tun_not_capable(tun))
1637 return -EPERM;
1638 err = security_tun_dev_open(tun->security);
1639 if (err < 0)
1640 return err;
1641
1642 err = tun_attach(tun, file, ifr->ifr_flags & IFF_NOFILTER);
1643 if (err < 0)
1644 return err;
1645
1646 if (tun->flags & IFF_MULTI_QUEUE &&
1647 (tun->numqueues + tun->numdisabled > 1)) {
1648 /* One or more queue has already been attached, no need
1649 * to initialize the device again.
1650 */
1651 return 0;
1652 }
1653 }
1654 else {
1655 char *name;
1656 unsigned long flags = 0;
1657 int queues = ifr->ifr_flags & IFF_MULTI_QUEUE ?
1658 MAX_TAP_QUEUES : 1;
1659
1660 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
1661 return -EPERM;
1662 err = security_tun_dev_create();
1663 if (err < 0)
1664 return err;
1665
1666 /* Set dev type */
1667 if (ifr->ifr_flags & IFF_TUN) {
1668 /* TUN device */
1669 flags |= IFF_TUN;
1670 name = "tun%d";
1671 } else if (ifr->ifr_flags & IFF_TAP) {
1672 /* TAP device */
1673 flags |= IFF_TAP;
1674 name = "tap%d";
1675 } else
1676 return -EINVAL;
1677
1678 if (*ifr->ifr_name)
1679 name = ifr->ifr_name;
1680
1681 dev = alloc_netdev_mqs(sizeof(struct tun_struct), name,
1682 NET_NAME_UNKNOWN, tun_setup, queues,
1683 queues);
1684
1685 if (!dev)
1686 return -ENOMEM;
1687 err = dev_get_valid_name(net, dev, name);
1688 if (err < 0)
1689 goto err_free_dev;
1690
1691 dev_net_set(dev, net);
1692 dev->rtnl_link_ops = &tun_link_ops;
1693 dev->ifindex = tfile->ifindex;
1694 dev->sysfs_groups[0] = &tun_attr_group;
1695
1696 tun = netdev_priv(dev);
1697 tun->dev = dev;
1698 tun->flags = flags;
1699 tun->txflt.count = 0;
1700 tun->vnet_hdr_sz = sizeof(struct virtio_net_hdr);
1701
1702 tun->filter_attached = false;
1703 tun->sndbuf = tfile->socket.sk->sk_sndbuf;
1704
1705 spin_lock_init(&tun->lock);
1706
1707 err = security_tun_dev_alloc_security(&tun->security);
1708 if (err < 0)
1709 goto err_free_dev;
1710
1711 tun_net_init(dev);
1712 tun_flow_init(tun);
1713
1714 dev->hw_features = NETIF_F_SG | NETIF_F_FRAGLIST |
1715 TUN_USER_FEATURES | NETIF_F_HW_VLAN_CTAG_TX |
1716 NETIF_F_HW_VLAN_STAG_TX;
1717 dev->features = dev->hw_features;
1718 dev->vlan_features = dev->features &
1719 ~(NETIF_F_HW_VLAN_CTAG_TX |
1720 NETIF_F_HW_VLAN_STAG_TX);
1721
1722 INIT_LIST_HEAD(&tun->disabled);
1723 err = tun_attach(tun, file, false);
1724 if (err < 0)
1725 goto err_free_flow;
1726
1727 err = register_netdevice(tun->dev);
1728 if (err < 0)
1729 goto err_detach;
1730 }
1731
1732 netif_carrier_on(tun->dev);
1733
1734 tun_debug(KERN_INFO, tun, "tun_set_iff\n");
1735
1736 tun->flags = (tun->flags & ~TUN_FEATURES) |
1737 (ifr->ifr_flags & TUN_FEATURES);
1738
1739 /* Make sure persistent devices do not get stuck in
1740 * xoff state.
1741 */
1742 if (netif_running(tun->dev))
1743 netif_tx_wake_all_queues(tun->dev);
1744
1745 strcpy(ifr->ifr_name, tun->dev->name);
1746 return 0;
1747
1748 err_detach:
1749 tun_detach_all(dev);
1750 err_free_flow:
1751 tun_flow_uninit(tun);
1752 security_tun_dev_free_security(tun->security);
1753 err_free_dev:
1754 free_netdev(dev);
1755 return err;
1756 }
1757
1758 static void tun_get_iff(struct net *net, struct tun_struct *tun,
1759 struct ifreq *ifr)
1760 {
1761 tun_debug(KERN_INFO, tun, "tun_get_iff\n");
1762
1763 strcpy(ifr->ifr_name, tun->dev->name);
1764
1765 ifr->ifr_flags = tun_flags(tun);
1766
1767 }
1768
1769 /* This is like a cut-down ethtool ops, except done via tun fd so no
1770 * privs required. */
1771 static int set_offload(struct tun_struct *tun, unsigned long arg)
1772 {
1773 netdev_features_t features = 0;
1774
1775 if (arg & TUN_F_CSUM) {
1776 features |= NETIF_F_HW_CSUM;
1777 arg &= ~TUN_F_CSUM;
1778
1779 if (arg & (TUN_F_TSO4|TUN_F_TSO6)) {
1780 if (arg & TUN_F_TSO_ECN) {
1781 features |= NETIF_F_TSO_ECN;
1782 arg &= ~TUN_F_TSO_ECN;
1783 }
1784 if (arg & TUN_F_TSO4)
1785 features |= NETIF_F_TSO;
1786 if (arg & TUN_F_TSO6)
1787 features |= NETIF_F_TSO6;
1788 arg &= ~(TUN_F_TSO4|TUN_F_TSO6);
1789 }
1790
1791 if (arg & TUN_F_UFO) {
1792 features |= NETIF_F_UFO;
1793 arg &= ~TUN_F_UFO;
1794 }
1795 }
1796
1797 /* This gives the user a way to test for new features in future by
1798 * trying to set them. */
1799 if (arg)
1800 return -EINVAL;
1801
1802 tun->set_features = features;
1803 netdev_update_features(tun->dev);
1804
1805 return 0;
1806 }
1807
1808 static void tun_detach_filter(struct tun_struct *tun, int n)
1809 {
1810 int i;
1811 struct tun_file *tfile;
1812
1813 for (i = 0; i < n; i++) {
1814 tfile = rtnl_dereference(tun->tfiles[i]);
1815 __sk_detach_filter(tfile->socket.sk, lockdep_rtnl_is_held());
1816 }
1817
1818 tun->filter_attached = false;
1819 }
1820
1821 static int tun_attach_filter(struct tun_struct *tun)
1822 {
1823 int i, ret = 0;
1824 struct tun_file *tfile;
1825
1826 for (i = 0; i < tun->numqueues; i++) {
1827 tfile = rtnl_dereference(tun->tfiles[i]);
1828 ret = __sk_attach_filter(&tun->fprog, tfile->socket.sk,
1829 lockdep_rtnl_is_held());
1830 if (ret) {
1831 tun_detach_filter(tun, i);
1832 return ret;
1833 }
1834 }
1835
1836 tun->filter_attached = true;
1837 return ret;
1838 }
1839
1840 static void tun_set_sndbuf(struct tun_struct *tun)
1841 {
1842 struct tun_file *tfile;
1843 int i;
1844
1845 for (i = 0; i < tun->numqueues; i++) {
1846 tfile = rtnl_dereference(tun->tfiles[i]);
1847 tfile->socket.sk->sk_sndbuf = tun->sndbuf;
1848 }
1849 }
1850
1851 static int tun_set_queue(struct file *file, struct ifreq *ifr)
1852 {
1853 struct tun_file *tfile = file->private_data;
1854 struct tun_struct *tun;
1855 int ret = 0;
1856
1857 rtnl_lock();
1858
1859 if (ifr->ifr_flags & IFF_ATTACH_QUEUE) {
1860 tun = tfile->detached;
1861 if (!tun) {
1862 ret = -EINVAL;
1863 goto unlock;
1864 }
1865 ret = security_tun_dev_attach_queue(tun->security);
1866 if (ret < 0)
1867 goto unlock;
1868 ret = tun_attach(tun, file, false);
1869 } else if (ifr->ifr_flags & IFF_DETACH_QUEUE) {
1870 tun = rtnl_dereference(tfile->tun);
1871 if (!tun || !(tun->flags & IFF_MULTI_QUEUE) || tfile->detached)
1872 ret = -EINVAL;
1873 else
1874 __tun_detach(tfile, false);
1875 } else
1876 ret = -EINVAL;
1877
1878 unlock:
1879 rtnl_unlock();
1880 return ret;
1881 }
1882
1883 static long __tun_chr_ioctl(struct file *file, unsigned int cmd,
1884 unsigned long arg, int ifreq_len)
1885 {
1886 struct tun_file *tfile = file->private_data;
1887 struct tun_struct *tun;
1888 void __user* argp = (void __user*)arg;
1889 struct ifreq ifr;
1890 kuid_t owner;
1891 kgid_t group;
1892 int sndbuf;
1893 int vnet_hdr_sz;
1894 unsigned int ifindex;
1895 int le;
1896 int ret;
1897
1898 if (cmd == TUNSETIFF || cmd == TUNSETQUEUE || _IOC_TYPE(cmd) == 0x89) {
1899 if (copy_from_user(&ifr, argp, ifreq_len))
1900 return -EFAULT;
1901 } else {
1902 memset(&ifr, 0, sizeof(ifr));
1903 }
1904 if (cmd == TUNGETFEATURES) {
1905 /* Currently this just means: "what IFF flags are valid?".
1906 * This is needed because we never checked for invalid flags on
1907 * TUNSETIFF.
1908 */
1909 return put_user(IFF_TUN | IFF_TAP | TUN_FEATURES,
1910 (unsigned int __user*)argp);
1911 } else if (cmd == TUNSETQUEUE)
1912 return tun_set_queue(file, &ifr);
1913
1914 ret = 0;
1915 rtnl_lock();
1916
1917 tun = __tun_get(tfile);
1918 if (cmd == TUNSETIFF && !tun) {
1919 ifr.ifr_name[IFNAMSIZ-1] = '\0';
1920
1921 ret = tun_set_iff(sock_net(&tfile->sk), file, &ifr);
1922
1923 if (ret)
1924 goto unlock;
1925
1926 if (copy_to_user(argp, &ifr, ifreq_len))
1927 ret = -EFAULT;
1928 goto unlock;
1929 }
1930 if (cmd == TUNSETIFINDEX) {
1931 ret = -EPERM;
1932 if (tun)
1933 goto unlock;
1934
1935 ret = -EFAULT;
1936 if (copy_from_user(&ifindex, argp, sizeof(ifindex)))
1937 goto unlock;
1938
1939 ret = 0;
1940 tfile->ifindex = ifindex;
1941 goto unlock;
1942 }
1943
1944 ret = -EBADFD;
1945 if (!tun)
1946 goto unlock;
1947
1948 tun_debug(KERN_INFO, tun, "tun_chr_ioctl cmd %u\n", cmd);
1949
1950 ret = 0;
1951 switch (cmd) {
1952 case TUNGETIFF:
1953 tun_get_iff(current->nsproxy->net_ns, tun, &ifr);
1954
1955 if (tfile->detached)
1956 ifr.ifr_flags |= IFF_DETACH_QUEUE;
1957 if (!tfile->socket.sk->sk_filter)
1958 ifr.ifr_flags |= IFF_NOFILTER;
1959
1960 if (copy_to_user(argp, &ifr, ifreq_len))
1961 ret = -EFAULT;
1962 break;
1963
1964 case TUNSETNOCSUM:
1965 /* Disable/Enable checksum */
1966
1967 /* [unimplemented] */
1968 tun_debug(KERN_INFO, tun, "ignored: set checksum %s\n",
1969 arg ? "disabled" : "enabled");
1970 break;
1971
1972 case TUNSETPERSIST:
1973 /* Disable/Enable persist mode. Keep an extra reference to the
1974 * module to prevent the module being unprobed.
1975 */
1976 if (arg && !(tun->flags & IFF_PERSIST)) {
1977 tun->flags |= IFF_PERSIST;
1978 __module_get(THIS_MODULE);
1979 }
1980 if (!arg && (tun->flags & IFF_PERSIST)) {
1981 tun->flags &= ~IFF_PERSIST;
1982 module_put(THIS_MODULE);
1983 }
1984
1985 tun_debug(KERN_INFO, tun, "persist %s\n",
1986 arg ? "enabled" : "disabled");
1987 break;
1988
1989 case TUNSETOWNER:
1990 /* Set owner of the device */
1991 owner = make_kuid(current_user_ns(), arg);
1992 if (!uid_valid(owner)) {
1993 ret = -EINVAL;
1994 break;
1995 }
1996 tun->owner = owner;
1997 tun_debug(KERN_INFO, tun, "owner set to %u\n",
1998 from_kuid(&init_user_ns, tun->owner));
1999 break;
2000
2001 case TUNSETGROUP:
2002 /* Set group of the device */
2003 group = make_kgid(current_user_ns(), arg);
2004 if (!gid_valid(group)) {
2005 ret = -EINVAL;
2006 break;
2007 }
2008 tun->group = group;
2009 tun_debug(KERN_INFO, tun, "group set to %u\n",
2010 from_kgid(&init_user_ns, tun->group));
2011 break;
2012
2013 case TUNSETLINK:
2014 /* Only allow setting the type when the interface is down */
2015 if (tun->dev->flags & IFF_UP) {
2016 tun_debug(KERN_INFO, tun,
2017 "Linktype set failed because interface is up\n");
2018 ret = -EBUSY;
2019 } else {
2020 tun->dev->type = (int) arg;
2021 tun_debug(KERN_INFO, tun, "linktype set to %d\n",
2022 tun->dev->type);
2023 ret = 0;
2024 }
2025 break;
2026
2027 #ifdef TUN_DEBUG
2028 case TUNSETDEBUG:
2029 tun->debug = arg;
2030 break;
2031 #endif
2032 case TUNSETOFFLOAD:
2033 ret = set_offload(tun, arg);
2034 break;
2035
2036 case TUNSETTXFILTER:
2037 /* Can be set only for TAPs */
2038 ret = -EINVAL;
2039 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
2040 break;
2041 ret = update_filter(&tun->txflt, (void __user *)arg);
2042 break;
2043
2044 case SIOCGIFHWADDR:
2045 /* Get hw address */
2046 memcpy(ifr.ifr_hwaddr.sa_data, tun->dev->dev_addr, ETH_ALEN);
2047 ifr.ifr_hwaddr.sa_family = tun->dev->type;
2048 if (copy_to_user(argp, &ifr, ifreq_len))
2049 ret = -EFAULT;
2050 break;
2051
2052 case SIOCSIFHWADDR:
2053 /* Set hw address */
2054 tun_debug(KERN_DEBUG, tun, "set hw address: %pM\n",
2055 ifr.ifr_hwaddr.sa_data);
2056
2057 ret = dev_set_mac_address(tun->dev, &ifr.ifr_hwaddr);
2058 break;
2059
2060 case TUNGETSNDBUF:
2061 sndbuf = tfile->socket.sk->sk_sndbuf;
2062 if (copy_to_user(argp, &sndbuf, sizeof(sndbuf)))
2063 ret = -EFAULT;
2064 break;
2065
2066 case TUNSETSNDBUF:
2067 if (copy_from_user(&sndbuf, argp, sizeof(sndbuf))) {
2068 ret = -EFAULT;
2069 break;
2070 }
2071 if (sndbuf <= 0) {
2072 ret = -EINVAL;
2073 break;
2074 }
2075
2076 tun->sndbuf = sndbuf;
2077 tun_set_sndbuf(tun);
2078 break;
2079
2080 case TUNGETVNETHDRSZ:
2081 vnet_hdr_sz = tun->vnet_hdr_sz;
2082 if (copy_to_user(argp, &vnet_hdr_sz, sizeof(vnet_hdr_sz)))
2083 ret = -EFAULT;
2084 break;
2085
2086 case TUNSETVNETHDRSZ:
2087 if (copy_from_user(&vnet_hdr_sz, argp, sizeof(vnet_hdr_sz))) {
2088 ret = -EFAULT;
2089 break;
2090 }
2091 if (vnet_hdr_sz < (int)sizeof(struct virtio_net_hdr)) {
2092 ret = -EINVAL;
2093 break;
2094 }
2095
2096 tun->vnet_hdr_sz = vnet_hdr_sz;
2097 break;
2098
2099 case TUNGETVNETLE:
2100 le = !!(tun->flags & TUN_VNET_LE);
2101 if (put_user(le, (int __user *)argp))
2102 ret = -EFAULT;
2103 break;
2104
2105 case TUNSETVNETLE:
2106 if (get_user(le, (int __user *)argp)) {
2107 ret = -EFAULT;
2108 break;
2109 }
2110 if (le)
2111 tun->flags |= TUN_VNET_LE;
2112 else
2113 tun->flags &= ~TUN_VNET_LE;
2114 break;
2115
2116 case TUNGETVNETBE:
2117 ret = tun_get_vnet_be(tun, argp);
2118 break;
2119
2120 case TUNSETVNETBE:
2121 ret = tun_set_vnet_be(tun, argp);
2122 break;
2123
2124 case TUNATTACHFILTER:
2125 /* Can be set only for TAPs */
2126 ret = -EINVAL;
2127 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
2128 break;
2129 ret = -EFAULT;
2130 if (copy_from_user(&tun->fprog, argp, sizeof(tun->fprog)))
2131 break;
2132
2133 ret = tun_attach_filter(tun);
2134 break;
2135
2136 case TUNDETACHFILTER:
2137 /* Can be set only for TAPs */
2138 ret = -EINVAL;
2139 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
2140 break;
2141 ret = 0;
2142 tun_detach_filter(tun, tun->numqueues);
2143 break;
2144
2145 case TUNGETFILTER:
2146 ret = -EINVAL;
2147 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
2148 break;
2149 ret = -EFAULT;
2150 if (copy_to_user(argp, &tun->fprog, sizeof(tun->fprog)))
2151 break;
2152 ret = 0;
2153 break;
2154
2155 default:
2156 ret = -EINVAL;
2157 break;
2158 }
2159
2160 unlock:
2161 rtnl_unlock();
2162 if (tun)
2163 tun_put(tun);
2164 return ret;
2165 }
2166
2167 static long tun_chr_ioctl(struct file *file,
2168 unsigned int cmd, unsigned long arg)
2169 {
2170 return __tun_chr_ioctl(file, cmd, arg, sizeof (struct ifreq));
2171 }
2172
2173 #ifdef CONFIG_COMPAT
2174 static long tun_chr_compat_ioctl(struct file *file,
2175 unsigned int cmd, unsigned long arg)
2176 {
2177 switch (cmd) {
2178 case TUNSETIFF:
2179 case TUNGETIFF:
2180 case TUNSETTXFILTER:
2181 case TUNGETSNDBUF:
2182 case TUNSETSNDBUF:
2183 case SIOCGIFHWADDR:
2184 case SIOCSIFHWADDR:
2185 arg = (unsigned long)compat_ptr(arg);
2186 break;
2187 default:
2188 arg = (compat_ulong_t)arg;
2189 break;
2190 }
2191
2192 /*
2193 * compat_ifreq is shorter than ifreq, so we must not access beyond
2194 * the end of that structure. All fields that are used in this
2195 * driver are compatible though, we don't need to convert the
2196 * contents.
2197 */
2198 return __tun_chr_ioctl(file, cmd, arg, sizeof(struct compat_ifreq));
2199 }
2200 #endif /* CONFIG_COMPAT */
2201
2202 static int tun_chr_fasync(int fd, struct file *file, int on)
2203 {
2204 struct tun_file *tfile = file->private_data;
2205 int ret;
2206
2207 if ((ret = fasync_helper(fd, file, on, &tfile->fasync)) < 0)
2208 goto out;
2209
2210 if (on) {
2211 __f_setown(file, task_pid(current), PIDTYPE_PID, 0);
2212 tfile->flags |= TUN_FASYNC;
2213 } else
2214 tfile->flags &= ~TUN_FASYNC;
2215 ret = 0;
2216 out:
2217 return ret;
2218 }
2219
2220 static int tun_chr_open(struct inode *inode, struct file * file)
2221 {
2222 struct net *net = current->nsproxy->net_ns;
2223 struct tun_file *tfile;
2224
2225 DBG1(KERN_INFO, "tunX: tun_chr_open\n");
2226
2227 tfile = (struct tun_file *)sk_alloc(net, AF_UNSPEC, GFP_KERNEL,
2228 &tun_proto, 0);
2229 if (!tfile)
2230 return -ENOMEM;
2231 RCU_INIT_POINTER(tfile->tun, NULL);
2232 tfile->flags = 0;
2233 tfile->ifindex = 0;
2234
2235 init_waitqueue_head(&tfile->wq.wait);
2236 RCU_INIT_POINTER(tfile->socket.wq, &tfile->wq);
2237
2238 tfile->socket.file = file;
2239 tfile->socket.ops = &tun_socket_ops;
2240
2241 sock_init_data(&tfile->socket, &tfile->sk);
2242
2243 tfile->sk.sk_write_space = tun_sock_write_space;
2244 tfile->sk.sk_sndbuf = INT_MAX;
2245
2246 file->private_data = tfile;
2247 INIT_LIST_HEAD(&tfile->next);
2248
2249 sock_set_flag(&tfile->sk, SOCK_ZEROCOPY);
2250
2251 return 0;
2252 }
2253
2254 static int tun_chr_close(struct inode *inode, struct file *file)
2255 {
2256 struct tun_file *tfile = file->private_data;
2257
2258 tun_detach(tfile, true);
2259
2260 return 0;
2261 }
2262
2263 #ifdef CONFIG_PROC_FS
2264 static void tun_chr_show_fdinfo(struct seq_file *m, struct file *f)
2265 {
2266 struct tun_struct *tun;
2267 struct ifreq ifr;
2268
2269 memset(&ifr, 0, sizeof(ifr));
2270
2271 rtnl_lock();
2272 tun = tun_get(f);
2273 if (tun)
2274 tun_get_iff(current->nsproxy->net_ns, tun, &ifr);
2275 rtnl_unlock();
2276
2277 if (tun)
2278 tun_put(tun);
2279
2280 seq_printf(m, "iff:\t%s\n", ifr.ifr_name);
2281 }
2282 #endif
2283
2284 static const struct file_operations tun_fops = {
2285 .owner = THIS_MODULE,
2286 .llseek = no_llseek,
2287 .read_iter = tun_chr_read_iter,
2288 .write_iter = tun_chr_write_iter,
2289 .poll = tun_chr_poll,
2290 .unlocked_ioctl = tun_chr_ioctl,
2291 #ifdef CONFIG_COMPAT
2292 .compat_ioctl = tun_chr_compat_ioctl,
2293 #endif
2294 .open = tun_chr_open,
2295 .release = tun_chr_close,
2296 .fasync = tun_chr_fasync,
2297 #ifdef CONFIG_PROC_FS
2298 .show_fdinfo = tun_chr_show_fdinfo,
2299 #endif
2300 };
2301
2302 static struct miscdevice tun_miscdev = {
2303 .minor = TUN_MINOR,
2304 .name = "tun",
2305 .nodename = "net/tun",
2306 .fops = &tun_fops,
2307 };
2308
2309 /* ethtool interface */
2310
2311 static int tun_get_settings(struct net_device *dev, struct ethtool_cmd *cmd)
2312 {
2313 cmd->supported = 0;
2314 cmd->advertising = 0;
2315 ethtool_cmd_speed_set(cmd, SPEED_10);
2316 cmd->duplex = DUPLEX_FULL;
2317 cmd->port = PORT_TP;
2318 cmd->phy_address = 0;
2319 cmd->transceiver = XCVR_INTERNAL;
2320 cmd->autoneg = AUTONEG_DISABLE;
2321 cmd->maxtxpkt = 0;
2322 cmd->maxrxpkt = 0;
2323 return 0;
2324 }
2325
2326 static void tun_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
2327 {
2328 struct tun_struct *tun = netdev_priv(dev);
2329
2330 strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
2331 strlcpy(info->version, DRV_VERSION, sizeof(info->version));
2332
2333 switch (tun->flags & TUN_TYPE_MASK) {
2334 case IFF_TUN:
2335 strlcpy(info->bus_info, "tun", sizeof(info->bus_info));
2336 break;
2337 case IFF_TAP:
2338 strlcpy(info->bus_info, "tap", sizeof(info->bus_info));
2339 break;
2340 }
2341 }
2342
2343 static u32 tun_get_msglevel(struct net_device *dev)
2344 {
2345 #ifdef TUN_DEBUG
2346 struct tun_struct *tun = netdev_priv(dev);
2347 return tun->debug;
2348 #else
2349 return -EOPNOTSUPP;
2350 #endif
2351 }
2352
2353 static void tun_set_msglevel(struct net_device *dev, u32 value)
2354 {
2355 #ifdef TUN_DEBUG
2356 struct tun_struct *tun = netdev_priv(dev);
2357 tun->debug = value;
2358 #endif
2359 }
2360
2361 static const struct ethtool_ops tun_ethtool_ops = {
2362 .get_settings = tun_get_settings,
2363 .get_drvinfo = tun_get_drvinfo,
2364 .get_msglevel = tun_get_msglevel,
2365 .set_msglevel = tun_set_msglevel,
2366 .get_link = ethtool_op_get_link,
2367 .get_ts_info = ethtool_op_get_ts_info,
2368 };
2369
2370
2371 static int __init tun_init(void)
2372 {
2373 int ret = 0;
2374
2375 pr_info("%s, %s\n", DRV_DESCRIPTION, DRV_VERSION);
2376 pr_info("%s\n", DRV_COPYRIGHT);
2377
2378 ret = rtnl_link_register(&tun_link_ops);
2379 if (ret) {
2380 pr_err("Can't register link_ops\n");
2381 goto err_linkops;
2382 }
2383
2384 ret = misc_register(&tun_miscdev);
2385 if (ret) {
2386 pr_err("Can't register misc device %d\n", TUN_MINOR);
2387 goto err_misc;
2388 }
2389 return 0;
2390 err_misc:
2391 rtnl_link_unregister(&tun_link_ops);
2392 err_linkops:
2393 return ret;
2394 }
2395
2396 static void tun_cleanup(void)
2397 {
2398 misc_deregister(&tun_miscdev);
2399 rtnl_link_unregister(&tun_link_ops);
2400 }
2401
2402 /* Get an underlying socket object from tun file. Returns error unless file is
2403 * attached to a device. The returned object works like a packet socket, it
2404 * can be used for sock_sendmsg/sock_recvmsg. The caller is responsible for
2405 * holding a reference to the file for as long as the socket is in use. */
2406 struct socket *tun_get_socket(struct file *file)
2407 {
2408 struct tun_file *tfile;
2409 if (file->f_op != &tun_fops)
2410 return ERR_PTR(-EINVAL);
2411 tfile = file->private_data;
2412 if (!tfile)
2413 return ERR_PTR(-EBADFD);
2414 return &tfile->socket;
2415 }
2416 EXPORT_SYMBOL_GPL(tun_get_socket);
2417
2418 module_init(tun_init);
2419 module_exit(tun_cleanup);
2420 MODULE_DESCRIPTION(DRV_DESCRIPTION);
2421 MODULE_AUTHOR(DRV_COPYRIGHT);
2422 MODULE_LICENSE("GPL");
2423 MODULE_ALIAS_MISCDEV(TUN_MINOR);
2424 MODULE_ALIAS("devname:net/tun");
Linux with added FPC-III support