?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
pvrusb2: reduce stack usage pvr2_eeprom_analyze()
[linux/fpc-iii.git] / drivers / net / tun.c
bloba931b73393c84cc37098dc2dd00b81d9ef613443
1 /*
2 * TUN - Universal TUN/TAP device driver.
3 * Copyright (C) 1999-2002 Maxim Krasnyansky <maxk@qualcomm.com>
4 *
5 * This program is free software; you can redistribute it and/or modify
6 * it under the terms of the GNU General Public License as published by
7 * the Free Software Foundation; either version 2 of the License, or
8 * (at your option) any later version.
9 *
10 * This program is distributed in the hope that it will be useful,
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
13 * GNU General Public License for more details.
14 *
15 * $Id: tun.c,v 1.15 2002/03/01 02:44:24 maxk Exp $
16 */
17
18 /*
19 * Changes:
20 *
21 * Mike Kershaw <dragorn@kismetwireless.net> 2005/08/14
22 * Add TUNSETLINK ioctl to set the link encapsulation
23 *
24 * Mark Smith <markzzzsmith@yahoo.com.au>
25 * Use eth_random_addr() for tap MAC address.
26 *
27 * Harald Roelle <harald.roelle@ifi.lmu.de> 2004/04/20
28 * Fixes in packet dropping, queue length setting and queue wakeup.
29 * Increased default tx queue length.
30 * Added ethtool API.
31 * Minor cleanups
32 *
33 * Daniel Podlejski <underley@underley.eu.org>
34 * Modifications for 2.3.99-pre5 kernel.
35 */
36
37 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
38
39 #define DRV_NAME "tun"
40 #define DRV_VERSION "1.6"
41 #define DRV_DESCRIPTION "Universal TUN/TAP device driver"
42 #define DRV_COPYRIGHT "(C) 1999-2004 Max Krasnyansky <maxk@qualcomm.com>"
43
44 #include <linux/module.h>
45 #include <linux/errno.h>
46 #include <linux/kernel.h>
47 #include <linux/major.h>
48 #include <linux/slab.h>
49 #include <linux/poll.h>
50 #include <linux/fcntl.h>
51 #include <linux/init.h>
52 #include <linux/skbuff.h>
53 #include <linux/netdevice.h>
54 #include <linux/etherdevice.h>
55 #include <linux/miscdevice.h>
56 #include <linux/ethtool.h>
57 #include <linux/rtnetlink.h>
58 #include <linux/compat.h>
59 #include <linux/if.h>
60 #include <linux/if_arp.h>
61 #include <linux/if_ether.h>
62 #include <linux/if_tun.h>
63 #include <linux/if_vlan.h>
64 #include <linux/crc32.h>
65 #include <linux/nsproxy.h>
66 #include <linux/virtio_net.h>
67 #include <linux/rcupdate.h>
68 #include <net/net_namespace.h>
69 #include <net/netns/generic.h>
70 #include <net/rtnetlink.h>
71 #include <net/sock.h>
72 #include <linux/seq_file.h>
73 #include <linux/uio.h>
74 #include <linux/skb_array.h>
75
76 #include <asm/uaccess.h>
77
78 /* Uncomment to enable debugging */
79 /* #define TUN_DEBUG 1 */
80
81 #ifdef TUN_DEBUG
82 static int debug;
83
84 #define tun_debug(level, tun, fmt, args...) \
85 do { \
86 if (tun->debug) \
87 netdev_printk(level, tun->dev, fmt, ##args); \
88 } while (0)
89 #define DBG1(level, fmt, args...) \
90 do { \
91 if (debug == 2) \
92 printk(level fmt, ##args); \
93 } while (0)
94 #else
95 #define tun_debug(level, tun, fmt, args...) \
96 do { \
97 if (0) \
98 netdev_printk(level, tun->dev, fmt, ##args); \
99 } while (0)
100 #define DBG1(level, fmt, args...) \
101 do { \
102 if (0) \
103 printk(level fmt, ##args); \
104 } while (0)
105 #endif
106
107 /* TUN device flags */
108
109 /* IFF_ATTACH_QUEUE is never stored in device flags,
110 * overload it to mean fasync when stored there.
111 */
112 #define TUN_FASYNC IFF_ATTACH_QUEUE
113 /* High bits in flags field are unused. */
114 #define TUN_VNET_LE 0x80000000
115 #define TUN_VNET_BE 0x40000000
116
117 #define TUN_FEATURES (IFF_NO_PI | IFF_ONE_QUEUE | IFF_VNET_HDR | \
118 IFF_MULTI_QUEUE)
119 #define GOODCOPY_LEN 128
120
121 #define FLT_EXACT_COUNT 8
122 struct tap_filter {
123 unsigned int count; /* Number of addrs. Zero means disabled */
124 u32 mask[2]; /* Mask of the hashed addrs */
125 unsigned char addr[FLT_EXACT_COUNT][ETH_ALEN];
126 };
127
128 /* MAX_TAP_QUEUES 256 is chosen to allow rx/tx queues to be equal
129 * to max number of VCPUs in guest. */
130 #define MAX_TAP_QUEUES 256
131 #define MAX_TAP_FLOWS 4096
132
133 #define TUN_FLOW_EXPIRE (3 * HZ)
134
135 struct tun_pcpu_stats {
136 u64 rx_packets;
137 u64 rx_bytes;
138 u64 tx_packets;
139 u64 tx_bytes;
140 struct u64_stats_sync syncp;
141 u32 rx_dropped;
142 u32 tx_dropped;
143 u32 rx_frame_errors;
144 };
145
146 /* A tun_file connects an open character device to a tuntap netdevice. It
147 * also contains all socket related structures (except sock_fprog and tap_filter)
148 * to serve as one transmit queue for tuntap device. The sock_fprog and
149 * tap_filter were kept in tun_struct since they were used for filtering for the
150 * netdevice not for a specific queue (at least I didn't see the requirement for
151 * this).
152 *
153 * RCU usage:
154 * The tun_file and tun_struct are loosely coupled, the pointer from one to the
155 * other can only be read while rcu_read_lock or rtnl_lock is held.
156 */
157 struct tun_file {
158 struct sock sk;
159 struct socket socket;
160 struct socket_wq wq;
161 struct tun_struct __rcu *tun;
162 struct fasync_struct *fasync;
163 /* only used for fasnyc */
164 unsigned int flags;
165 union {
166 u16 queue_index;
167 unsigned int ifindex;
168 };
169 struct list_head next;
170 struct tun_struct *detached;
171 struct skb_array tx_array;
172 };
173
174 struct tun_flow_entry {
175 struct hlist_node hash_link;
176 struct rcu_head rcu;
177 struct tun_struct *tun;
178
179 u32 rxhash;
180 u32 rps_rxhash;
181 int queue_index;
182 unsigned long updated;
183 };
184
185 #define TUN_NUM_FLOW_ENTRIES 1024
186
187 /* Since the socket were moved to tun_file, to preserve the behavior of persist
188 * device, socket filter, sndbuf and vnet header size were restore when the
189 * file were attached to a persist device.
190 */
191 struct tun_struct {
192 struct tun_file __rcu *tfiles[MAX_TAP_QUEUES];
193 unsigned int numqueues;
194 unsigned int flags;
195 kuid_t owner;
196 kgid_t group;
197
198 struct net_device *dev;
199 netdev_features_t set_features;
200 #define TUN_USER_FEATURES (NETIF_F_HW_CSUM|NETIF_F_TSO_ECN|NETIF_F_TSO| \
201 NETIF_F_TSO6|NETIF_F_UFO)
202
203 int align;
204 int vnet_hdr_sz;
205 int sndbuf;
206 struct tap_filter txflt;
207 struct sock_fprog fprog;
208 /* protected by rtnl lock */
209 bool filter_attached;
210 #ifdef TUN_DEBUG
211 int debug;
212 #endif
213 spinlock_t lock;
214 struct hlist_head flows[TUN_NUM_FLOW_ENTRIES];
215 struct timer_list flow_gc_timer;
216 unsigned long ageing_time;
217 unsigned int numdisabled;
218 struct list_head disabled;
219 void *security;
220 u32 flow_count;
221 struct tun_pcpu_stats __percpu *pcpu_stats;
222 };
223
224 #ifdef CONFIG_TUN_VNET_CROSS_LE
225 static inline bool tun_legacy_is_little_endian(struct tun_struct *tun)
226 {
227 return tun->flags & TUN_VNET_BE ? false :
228 virtio_legacy_is_little_endian();
229 }
230
231 static long tun_get_vnet_be(struct tun_struct *tun, int __user *argp)
232 {
233 int be = !!(tun->flags & TUN_VNET_BE);
234
235 if (put_user(be, argp))
236 return -EFAULT;
237
238 return 0;
239 }
240
241 static long tun_set_vnet_be(struct tun_struct *tun, int __user *argp)
242 {
243 int be;
244
245 if (get_user(be, argp))
246 return -EFAULT;
247
248 if (be)
249 tun->flags |= TUN_VNET_BE;
250 else
251 tun->flags &= ~TUN_VNET_BE;
252
253 return 0;
254 }
255 #else
256 static inline bool tun_legacy_is_little_endian(struct tun_struct *tun)
257 {
258 return virtio_legacy_is_little_endian();
259 }
260
261 static long tun_get_vnet_be(struct tun_struct *tun, int __user *argp)
262 {
263 return -EINVAL;
264 }
265
266 static long tun_set_vnet_be(struct tun_struct *tun, int __user *argp)
267 {
268 return -EINVAL;
269 }
270 #endif /* CONFIG_TUN_VNET_CROSS_LE */
271
272 static inline bool tun_is_little_endian(struct tun_struct *tun)
273 {
274 return tun->flags & TUN_VNET_LE ||
275 tun_legacy_is_little_endian(tun);
276 }
277
278 static inline u16 tun16_to_cpu(struct tun_struct *tun, __virtio16 val)
279 {
280 return __virtio16_to_cpu(tun_is_little_endian(tun), val);
281 }
282
283 static inline __virtio16 cpu_to_tun16(struct tun_struct *tun, u16 val)
284 {
285 return __cpu_to_virtio16(tun_is_little_endian(tun), val);
286 }
287
288 static inline u32 tun_hashfn(u32 rxhash)
289 {
290 return rxhash & 0x3ff;
291 }
292
293 static struct tun_flow_entry *tun_flow_find(struct hlist_head *head, u32 rxhash)
294 {
295 struct tun_flow_entry *e;
296
297 hlist_for_each_entry_rcu(e, head, hash_link) {
298 if (e->rxhash == rxhash)
299 return e;
300 }
301 return NULL;
302 }
303
304 static struct tun_flow_entry *tun_flow_create(struct tun_struct *tun,
305 struct hlist_head *head,
306 u32 rxhash, u16 queue_index)
307 {
308 struct tun_flow_entry *e = kmalloc(sizeof(*e), GFP_ATOMIC);
309
310 if (e) {
311 tun_debug(KERN_INFO, tun, "create flow: hash %u index %u\n",
312 rxhash, queue_index);
313 e->updated = jiffies;
314 e->rxhash = rxhash;
315 e->rps_rxhash = 0;
316 e->queue_index = queue_index;
317 e->tun = tun;
318 hlist_add_head_rcu(&e->hash_link, head);
319 ++tun->flow_count;
320 }
321 return e;
322 }
323
324 static void tun_flow_delete(struct tun_struct *tun, struct tun_flow_entry *e)
325 {
326 tun_debug(KERN_INFO, tun, "delete flow: hash %u index %u\n",
327 e->rxhash, e->queue_index);
328 hlist_del_rcu(&e->hash_link);
329 kfree_rcu(e, rcu);
330 --tun->flow_count;
331 }
332
333 static void tun_flow_flush(struct tun_struct *tun)
334 {
335 int i;
336
337 spin_lock_bh(&tun->lock);
338 for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
339 struct tun_flow_entry *e;
340 struct hlist_node *n;
341
342 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link)
343 tun_flow_delete(tun, e);
344 }
345 spin_unlock_bh(&tun->lock);
346 }
347
348 static void tun_flow_delete_by_queue(struct tun_struct *tun, u16 queue_index)
349 {
350 int i;
351
352 spin_lock_bh(&tun->lock);
353 for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
354 struct tun_flow_entry *e;
355 struct hlist_node *n;
356
357 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
358 if (e->queue_index == queue_index)
359 tun_flow_delete(tun, e);
360 }
361 }
362 spin_unlock_bh(&tun->lock);
363 }
364
365 static void tun_flow_cleanup(unsigned long data)
366 {
367 struct tun_struct *tun = (struct tun_struct *)data;
368 unsigned long delay = tun->ageing_time;
369 unsigned long next_timer = jiffies + delay;
370 unsigned long count = 0;
371 int i;
372
373 tun_debug(KERN_INFO, tun, "tun_flow_cleanup\n");
374
375 spin_lock_bh(&tun->lock);
376 for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
377 struct tun_flow_entry *e;
378 struct hlist_node *n;
379
380 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
381 unsigned long this_timer;
382 count++;
383 this_timer = e->updated + delay;
384 if (time_before_eq(this_timer, jiffies))
385 tun_flow_delete(tun, e);
386 else if (time_before(this_timer, next_timer))
387 next_timer = this_timer;
388 }
389 }
390
391 if (count)
392 mod_timer(&tun->flow_gc_timer, round_jiffies_up(next_timer));
393 spin_unlock_bh(&tun->lock);
394 }
395
396 static void tun_flow_update(struct tun_struct *tun, u32 rxhash,
397 struct tun_file *tfile)
398 {
399 struct hlist_head *head;
400 struct tun_flow_entry *e;
401 unsigned long delay = tun->ageing_time;
402 u16 queue_index = tfile->queue_index;
403
404 if (!rxhash)
405 return;
406 else
407 head = &tun->flows[tun_hashfn(rxhash)];
408
409 rcu_read_lock();
410
411 /* We may get a very small possibility of OOO during switching, not
412 * worth to optimize.*/
413 if (tun->numqueues == 1 || tfile->detached)
414 goto unlock;
415
416 e = tun_flow_find(head, rxhash);
417 if (likely(e)) {
418 /* TODO: keep queueing to old queue until it's empty? */
419 e->queue_index = queue_index;
420 e->updated = jiffies;
421 sock_rps_record_flow_hash(e->rps_rxhash);
422 } else {
423 spin_lock_bh(&tun->lock);
424 if (!tun_flow_find(head, rxhash) &&
425 tun->flow_count < MAX_TAP_FLOWS)
426 tun_flow_create(tun, head, rxhash, queue_index);
427
428 if (!timer_pending(&tun->flow_gc_timer))
429 mod_timer(&tun->flow_gc_timer,
430 round_jiffies_up(jiffies + delay));
431 spin_unlock_bh(&tun->lock);
432 }
433
434 unlock:
435 rcu_read_unlock();
436 }
437
438 /**
439 * Save the hash received in the stack receive path and update the
440 * flow_hash table accordingly.
441 */
442 static inline void tun_flow_save_rps_rxhash(struct tun_flow_entry *e, u32 hash)
443 {
444 if (unlikely(e->rps_rxhash != hash))
445 e->rps_rxhash = hash;
446 }
447
448 /* We try to identify a flow through its rxhash first. The reason that
449 * we do not check rxq no. is because some cards(e.g 82599), chooses
450 * the rxq based on the txq where the last packet of the flow comes. As
451 * the userspace application move between processors, we may get a
452 * different rxq no. here. If we could not get rxhash, then we would
453 * hope the rxq no. may help here.
454 */
455 static u16 tun_select_queue(struct net_device *dev, struct sk_buff *skb,
456 void *accel_priv, select_queue_fallback_t fallback)
457 {
458 struct tun_struct *tun = netdev_priv(dev);
459 struct tun_flow_entry *e;
460 u32 txq = 0;
461 u32 numqueues = 0;
462
463 rcu_read_lock();
464 numqueues = ACCESS_ONCE(tun->numqueues);
465
466 txq = skb_get_hash(skb);
467 if (txq) {
468 e = tun_flow_find(&tun->flows[tun_hashfn(txq)], txq);
469 if (e) {
470 tun_flow_save_rps_rxhash(e, txq);
471 txq = e->queue_index;
472 } else
473 /* use multiply and shift instead of expensive divide */
474 txq = ((u64)txq * numqueues) >> 32;
475 } else if (likely(skb_rx_queue_recorded(skb))) {
476 txq = skb_get_rx_queue(skb);
477 while (unlikely(txq >= numqueues))
478 txq -= numqueues;
479 }
480
481 rcu_read_unlock();
482 return txq;
483 }
484
485 static inline bool tun_not_capable(struct tun_struct *tun)
486 {
487 const struct cred *cred = current_cred();
488 struct net *net = dev_net(tun->dev);
489
490 return ((uid_valid(tun->owner) && !uid_eq(cred->euid, tun->owner)) ||
491 (gid_valid(tun->group) && !in_egroup_p(tun->group))) &&
492 !ns_capable(net->user_ns, CAP_NET_ADMIN);
493 }
494
495 static void tun_set_real_num_queues(struct tun_struct *tun)
496 {
497 netif_set_real_num_tx_queues(tun->dev, tun->numqueues);
498 netif_set_real_num_rx_queues(tun->dev, tun->numqueues);
499 }
500
501 static void tun_disable_queue(struct tun_struct *tun, struct tun_file *tfile)
502 {
503 tfile->detached = tun;
504 list_add_tail(&tfile->next, &tun->disabled);
505 ++tun->numdisabled;
506 }
507
508 static struct tun_struct *tun_enable_queue(struct tun_file *tfile)
509 {
510 struct tun_struct *tun = tfile->detached;
511
512 tfile->detached = NULL;
513 list_del_init(&tfile->next);
514 --tun->numdisabled;
515 return tun;
516 }
517
518 static void tun_queue_purge(struct tun_file *tfile)
519 {
520 struct sk_buff *skb;
521
522 while ((skb = skb_array_consume(&tfile->tx_array)) != NULL)
523 kfree_skb(skb);
524
525 skb_queue_purge(&tfile->sk.sk_error_queue);
526 }
527
528 static void __tun_detach(struct tun_file *tfile, bool clean)
529 {
530 struct tun_file *ntfile;
531 struct tun_struct *tun;
532
533 tun = rtnl_dereference(tfile->tun);
534
535 if (tun && !tfile->detached) {
536 u16 index = tfile->queue_index;
537 BUG_ON(index >= tun->numqueues);
538
539 rcu_assign_pointer(tun->tfiles[index],
540 tun->tfiles[tun->numqueues - 1]);
541 ntfile = rtnl_dereference(tun->tfiles[index]);
542 ntfile->queue_index = index;
543
544 --tun->numqueues;
545 if (clean) {
546 RCU_INIT_POINTER(tfile->tun, NULL);
547 sock_put(&tfile->sk);
548 } else
549 tun_disable_queue(tun, tfile);
550
551 synchronize_net();
552 tun_flow_delete_by_queue(tun, tun->numqueues + 1);
553 /* Drop read queue */
554 tun_queue_purge(tfile);
555 tun_set_real_num_queues(tun);
556 } else if (tfile->detached && clean) {
557 tun = tun_enable_queue(tfile);
558 sock_put(&tfile->sk);
559 }
560
561 if (clean) {
562 if (tun && tun->numqueues == 0 && tun->numdisabled == 0) {
563 netif_carrier_off(tun->dev);
564
565 if (!(tun->flags & IFF_PERSIST) &&
566 tun->dev->reg_state == NETREG_REGISTERED)
567 unregister_netdevice(tun->dev);
568 }
569 if (tun)
570 skb_array_cleanup(&tfile->tx_array);
571 sock_put(&tfile->sk);
572 }
573 }
574
575 static void tun_detach(struct tun_file *tfile, bool clean)
576 {
577 rtnl_lock();
578 __tun_detach(tfile, clean);
579 rtnl_unlock();
580 }
581
582 static void tun_detach_all(struct net_device *dev)
583 {
584 struct tun_struct *tun = netdev_priv(dev);
585 struct tun_file *tfile, *tmp;
586 int i, n = tun->numqueues;
587
588 for (i = 0; i < n; i++) {
589 tfile = rtnl_dereference(tun->tfiles[i]);
590 BUG_ON(!tfile);
591 tfile->socket.sk->sk_shutdown = RCV_SHUTDOWN;
592 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
593 RCU_INIT_POINTER(tfile->tun, NULL);
594 --tun->numqueues;
595 }
596 list_for_each_entry(tfile, &tun->disabled, next) {
597 tfile->socket.sk->sk_shutdown = RCV_SHUTDOWN;
598 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
599 RCU_INIT_POINTER(tfile->tun, NULL);
600 }
601 BUG_ON(tun->numqueues != 0);
602
603 synchronize_net();
604 for (i = 0; i < n; i++) {
605 tfile = rtnl_dereference(tun->tfiles[i]);
606 /* Drop read queue */
607 tun_queue_purge(tfile);
608 sock_put(&tfile->sk);
609 }
610 list_for_each_entry_safe(tfile, tmp, &tun->disabled, next) {
611 tun_enable_queue(tfile);
612 tun_queue_purge(tfile);
613 sock_put(&tfile->sk);
614 }
615 BUG_ON(tun->numdisabled != 0);
616
617 if (tun->flags & IFF_PERSIST)
618 module_put(THIS_MODULE);
619 }
620
621 static int tun_attach(struct tun_struct *tun, struct file *file, bool skip_filter)
622 {
623 struct tun_file *tfile = file->private_data;
624 struct net_device *dev = tun->dev;
625 int err;
626
627 err = security_tun_dev_attach(tfile->socket.sk, tun->security);
628 if (err < 0)
629 goto out;
630
631 err = -EINVAL;
632 if (rtnl_dereference(tfile->tun) && !tfile->detached)
633 goto out;
634
635 err = -EBUSY;
636 if (!(tun->flags & IFF_MULTI_QUEUE) && tun->numqueues == 1)
637 goto out;
638
639 err = -E2BIG;
640 if (!tfile->detached &&
641 tun->numqueues + tun->numdisabled == MAX_TAP_QUEUES)
642 goto out;
643
644 err = 0;
645
646 /* Re-attach the filter to persist device */
647 if (!skip_filter && (tun->filter_attached == true)) {
648 lock_sock(tfile->socket.sk);
649 err = sk_attach_filter(&tun->fprog, tfile->socket.sk);
650 release_sock(tfile->socket.sk);
651 if (!err)
652 goto out;
653 }
654
655 if (!tfile->detached &&
656 skb_array_init(&tfile->tx_array, dev->tx_queue_len, GFP_KERNEL)) {
657 err = -ENOMEM;
658 goto out;
659 }
660
661 tfile->queue_index = tun->numqueues;
662 tfile->socket.sk->sk_shutdown &= ~RCV_SHUTDOWN;
663 rcu_assign_pointer(tfile->tun, tun);
664 rcu_assign_pointer(tun->tfiles[tun->numqueues], tfile);
665 tun->numqueues++;
666
667 if (tfile->detached)
668 tun_enable_queue(tfile);
669 else
670 sock_hold(&tfile->sk);
671
672 tun_set_real_num_queues(tun);
673
674 /* device is allowed to go away first, so no need to hold extra
675 * refcnt.
676 */
677
678 out:
679 return err;
680 }
681
682 static struct tun_struct *__tun_get(struct tun_file *tfile)
683 {
684 struct tun_struct *tun;
685
686 rcu_read_lock();
687 tun = rcu_dereference(tfile->tun);
688 if (tun)
689 dev_hold(tun->dev);
690 rcu_read_unlock();
691
692 return tun;
693 }
694
695 static struct tun_struct *tun_get(struct file *file)
696 {
697 return __tun_get(file->private_data);
698 }
699
700 static void tun_put(struct tun_struct *tun)
701 {
702 dev_put(tun->dev);
703 }
704
705 /* TAP filtering */
706 static void addr_hash_set(u32 *mask, const u8 *addr)
707 {
708 int n = ether_crc(ETH_ALEN, addr) >> 26;
709 mask[n >> 5] |= (1 << (n & 31));
710 }
711
712 static unsigned int addr_hash_test(const u32 *mask, const u8 *addr)
713 {
714 int n = ether_crc(ETH_ALEN, addr) >> 26;
715 return mask[n >> 5] & (1 << (n & 31));
716 }
717
718 static int update_filter(struct tap_filter *filter, void __user *arg)
719 {
720 struct { u8 u[ETH_ALEN]; } *addr;
721 struct tun_filter uf;
722 int err, alen, n, nexact;
723
724 if (copy_from_user(&uf, arg, sizeof(uf)))
725 return -EFAULT;
726
727 if (!uf.count) {
728 /* Disabled */
729 filter->count = 0;
730 return 0;
731 }
732
733 alen = ETH_ALEN * uf.count;
734 addr = memdup_user(arg + sizeof(uf), alen);
735 if (IS_ERR(addr))
736 return PTR_ERR(addr);
737
738 /* The filter is updated without holding any locks. Which is
739 * perfectly safe. We disable it first and in the worst
740 * case we'll accept a few undesired packets. */
741 filter->count = 0;
742 wmb();
743
744 /* Use first set of addresses as an exact filter */
745 for (n = 0; n < uf.count && n < FLT_EXACT_COUNT; n++)
746 memcpy(filter->addr[n], addr[n].u, ETH_ALEN);
747
748 nexact = n;
749
750 /* Remaining multicast addresses are hashed,
751 * unicast will leave the filter disabled. */
752 memset(filter->mask, 0, sizeof(filter->mask));
753 for (; n < uf.count; n++) {
754 if (!is_multicast_ether_addr(addr[n].u)) {
755 err = 0; /* no filter */
756 goto free_addr;
757 }
758 addr_hash_set(filter->mask, addr[n].u);
759 }
760
761 /* For ALLMULTI just set the mask to all ones.
762 * This overrides the mask populated above. */
763 if ((uf.flags & TUN_FLT_ALLMULTI))
764 memset(filter->mask, ~0, sizeof(filter->mask));
765
766 /* Now enable the filter */
767 wmb();
768 filter->count = nexact;
769
770 /* Return the number of exact filters */
771 err = nexact;
772 free_addr:
773 kfree(addr);
774 return err;
775 }
776
777 /* Returns: 0 - drop, !=0 - accept */
778 static int run_filter(struct tap_filter *filter, const struct sk_buff *skb)
779 {
780 /* Cannot use eth_hdr(skb) here because skb_mac_hdr() is incorrect
781 * at this point. */
782 struct ethhdr *eh = (struct ethhdr *) skb->data;
783 int i;
784
785 /* Exact match */
786 for (i = 0; i < filter->count; i++)
787 if (ether_addr_equal(eh->h_dest, filter->addr[i]))
788 return 1;
789
790 /* Inexact match (multicast only) */
791 if (is_multicast_ether_addr(eh->h_dest))
792 return addr_hash_test(filter->mask, eh->h_dest);
793
794 return 0;
795 }
796
797 /*
798 * Checks whether the packet is accepted or not.
799 * Returns: 0 - drop, !=0 - accept
800 */
801 static int check_filter(struct tap_filter *filter, const struct sk_buff *skb)
802 {
803 if (!filter->count)
804 return 1;
805
806 return run_filter(filter, skb);
807 }
808
809 /* Network device part of the driver */
810
811 static const struct ethtool_ops tun_ethtool_ops;
812
813 /* Net device detach from fd. */
814 static void tun_net_uninit(struct net_device *dev)
815 {
816 tun_detach_all(dev);
817 }
818
819 /* Net device open. */
820 static int tun_net_open(struct net_device *dev)
821 {
822 struct tun_struct *tun = netdev_priv(dev);
823 int i;
824
825 netif_tx_start_all_queues(dev);
826
827 for (i = 0; i < tun->numqueues; i++) {
828 struct tun_file *tfile;
829
830 tfile = rtnl_dereference(tun->tfiles[i]);
831 tfile->socket.sk->sk_write_space(tfile->socket.sk);
832 }
833
834 return 0;
835 }
836
837 /* Net device close. */
838 static int tun_net_close(struct net_device *dev)
839 {
840 netif_tx_stop_all_queues(dev);
841 return 0;
842 }
843
844 /* Net device start xmit */
845 static netdev_tx_t tun_net_xmit(struct sk_buff *skb, struct net_device *dev)
846 {
847 struct tun_struct *tun = netdev_priv(dev);
848 int txq = skb->queue_mapping;
849 struct tun_file *tfile;
850 u32 numqueues = 0;
851
852 rcu_read_lock();
853 tfile = rcu_dereference(tun->tfiles[txq]);
854 numqueues = ACCESS_ONCE(tun->numqueues);
855
856 /* Drop packet if interface is not attached */
857 if (txq >= numqueues)
858 goto drop;
859
860 #ifdef CONFIG_RPS
861 if (numqueues == 1 && static_key_false(&rps_needed)) {
862 /* Select queue was not called for the skbuff, so we extract the
863 * RPS hash and save it into the flow_table here.
864 */
865 __u32 rxhash;
866
867 rxhash = skb_get_hash(skb);
868 if (rxhash) {
869 struct tun_flow_entry *e;
870 e = tun_flow_find(&tun->flows[tun_hashfn(rxhash)],
871 rxhash);
872 if (e)
873 tun_flow_save_rps_rxhash(e, rxhash);
874 }
875 }
876 #endif
877
878 tun_debug(KERN_INFO, tun, "tun_net_xmit %d\n", skb->len);
879
880 BUG_ON(!tfile);
881
882 /* Drop if the filter does not like it.
883 * This is a noop if the filter is disabled.
884 * Filter can be enabled only for the TAP devices. */
885 if (!check_filter(&tun->txflt, skb))
886 goto drop;
887
888 if (tfile->socket.sk->sk_filter &&
889 sk_filter(tfile->socket.sk, skb))
890 goto drop;
891
892 /* Limit the number of packets queued by dividing txq length with the
893 * number of queues.
894 */
895 if (skb_queue_len(&tfile->socket.sk->sk_receive_queue) * numqueues
896 >= dev->tx_queue_len)
897 goto drop;
898
899 if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC)))
900 goto drop;
901
902 skb_tx_timestamp(skb);
903
904 /* Orphan the skb - required as we might hang on to it
905 * for indefinite time.
906 */
907 skb_orphan(skb);
908
909 nf_reset(skb);
910
911 if (skb_array_produce(&tfile->tx_array, skb))
912 goto drop;
913
914 /* Notify and wake up reader process */
915 if (tfile->flags & TUN_FASYNC)
916 kill_fasync(&tfile->fasync, SIGIO, POLL_IN);
917 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
918
919 rcu_read_unlock();
920 return NETDEV_TX_OK;
921
922 drop:
923 this_cpu_inc(tun->pcpu_stats->tx_dropped);
924 skb_tx_error(skb);
925 kfree_skb(skb);
926 rcu_read_unlock();
927 return NET_XMIT_DROP;
928 }
929
930 static void tun_net_mclist(struct net_device *dev)
931 {
932 /*
933 * This callback is supposed to deal with mc filter in
934 * _rx_ path and has nothing to do with the _tx_ path.
935 * In rx path we always accept everything userspace gives us.
936 */
937 }
938
939 #define MIN_MTU 68
940 #define MAX_MTU 65535
941
942 static int
943 tun_net_change_mtu(struct net_device *dev, int new_mtu)
944 {
945 if (new_mtu < MIN_MTU || new_mtu + dev->hard_header_len > MAX_MTU)
946 return -EINVAL;
947 dev->mtu = new_mtu;
948 return 0;
949 }
950
951 static netdev_features_t tun_net_fix_features(struct net_device *dev,
952 netdev_features_t features)
953 {
954 struct tun_struct *tun = netdev_priv(dev);
955
956 return (features & tun->set_features) | (features & ~TUN_USER_FEATURES);
957 }
958 #ifdef CONFIG_NET_POLL_CONTROLLER
959 static void tun_poll_controller(struct net_device *dev)
960 {
961 /*
962 * Tun only receives frames when:
963 * 1) the char device endpoint gets data from user space
964 * 2) the tun socket gets a sendmsg call from user space
965 * Since both of those are synchronous operations, we are guaranteed
966 * never to have pending data when we poll for it
967 * so there is nothing to do here but return.
968 * We need this though so netpoll recognizes us as an interface that
969 * supports polling, which enables bridge devices in virt setups to
970 * still use netconsole
971 */
972 return;
973 }
974 #endif
975
976 static void tun_set_headroom(struct net_device *dev, int new_hr)
977 {
978 struct tun_struct *tun = netdev_priv(dev);
979
980 if (new_hr < NET_SKB_PAD)
981 new_hr = NET_SKB_PAD;
982
983 tun->align = new_hr;
984 }
985
986 static struct rtnl_link_stats64 *
987 tun_net_get_stats64(struct net_device *dev, struct rtnl_link_stats64 *stats)
988 {
989 u32 rx_dropped = 0, tx_dropped = 0, rx_frame_errors = 0;
990 struct tun_struct *tun = netdev_priv(dev);
991 struct tun_pcpu_stats *p;
992 int i;
993
994 for_each_possible_cpu(i) {
995 u64 rxpackets, rxbytes, txpackets, txbytes;
996 unsigned int start;
997
998 p = per_cpu_ptr(tun->pcpu_stats, i);
999 do {
1000 start = u64_stats_fetch_begin(&p->syncp);
1001 rxpackets = p->rx_packets;
1002 rxbytes = p->rx_bytes;
1003 txpackets = p->tx_packets;
1004 txbytes = p->tx_bytes;
1005 } while (u64_stats_fetch_retry(&p->syncp, start));
1006
1007 stats->rx_packets += rxpackets;
1008 stats->rx_bytes += rxbytes;
1009 stats->tx_packets += txpackets;
1010 stats->tx_bytes += txbytes;
1011
1012 /* u32 counters */
1013 rx_dropped += p->rx_dropped;
1014 rx_frame_errors += p->rx_frame_errors;
1015 tx_dropped += p->tx_dropped;
1016 }
1017 stats->rx_dropped = rx_dropped;
1018 stats->rx_frame_errors = rx_frame_errors;
1019 stats->tx_dropped = tx_dropped;
1020 return stats;
1021 }
1022
1023 static const struct net_device_ops tun_netdev_ops = {
1024 .ndo_uninit = tun_net_uninit,
1025 .ndo_open = tun_net_open,
1026 .ndo_stop = tun_net_close,
1027 .ndo_start_xmit = tun_net_xmit,
1028 .ndo_change_mtu = tun_net_change_mtu,
1029 .ndo_fix_features = tun_net_fix_features,
1030 .ndo_select_queue = tun_select_queue,
1031 #ifdef CONFIG_NET_POLL_CONTROLLER
1032 .ndo_poll_controller = tun_poll_controller,
1033 #endif
1034 .ndo_set_rx_headroom = tun_set_headroom,
1035 .ndo_get_stats64 = tun_net_get_stats64,
1036 };
1037
1038 static const struct net_device_ops tap_netdev_ops = {
1039 .ndo_uninit = tun_net_uninit,
1040 .ndo_open = tun_net_open,
1041 .ndo_stop = tun_net_close,
1042 .ndo_start_xmit = tun_net_xmit,
1043 .ndo_change_mtu = tun_net_change_mtu,
1044 .ndo_fix_features = tun_net_fix_features,
1045 .ndo_set_rx_mode = tun_net_mclist,
1046 .ndo_set_mac_address = eth_mac_addr,
1047 .ndo_validate_addr = eth_validate_addr,
1048 .ndo_select_queue = tun_select_queue,
1049 #ifdef CONFIG_NET_POLL_CONTROLLER
1050 .ndo_poll_controller = tun_poll_controller,
1051 #endif
1052 .ndo_features_check = passthru_features_check,
1053 .ndo_set_rx_headroom = tun_set_headroom,
1054 .ndo_get_stats64 = tun_net_get_stats64,
1055 };
1056
1057 static void tun_flow_init(struct tun_struct *tun)
1058 {
1059 int i;
1060
1061 for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++)
1062 INIT_HLIST_HEAD(&tun->flows[i]);
1063
1064 tun->ageing_time = TUN_FLOW_EXPIRE;
1065 setup_timer(&tun->flow_gc_timer, tun_flow_cleanup, (unsigned long)tun);
1066 mod_timer(&tun->flow_gc_timer,
1067 round_jiffies_up(jiffies + tun->ageing_time));
1068 }
1069
1070 static void tun_flow_uninit(struct tun_struct *tun)
1071 {
1072 del_timer_sync(&tun->flow_gc_timer);
1073 tun_flow_flush(tun);
1074 }
1075
1076 /* Initialize net device. */
1077 static void tun_net_init(struct net_device *dev)
1078 {
1079 struct tun_struct *tun = netdev_priv(dev);
1080
1081 switch (tun->flags & TUN_TYPE_MASK) {
1082 case IFF_TUN:
1083 dev->netdev_ops = &tun_netdev_ops;
1084
1085 /* Point-to-Point TUN Device */
1086 dev->hard_header_len = 0;
1087 dev->addr_len = 0;
1088 dev->mtu = 1500;
1089
1090 /* Zero header length */
1091 dev->type = ARPHRD_NONE;
1092 dev->flags = IFF_POINTOPOINT | IFF_NOARP | IFF_MULTICAST;
1093 break;
1094
1095 case IFF_TAP:
1096 dev->netdev_ops = &tap_netdev_ops;
1097 /* Ethernet TAP Device */
1098 ether_setup(dev);
1099 dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1100 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
1101
1102 eth_hw_addr_random(dev);
1103
1104 break;
1105 }
1106 }
1107
1108 /* Character device part */
1109
1110 /* Poll */
1111 static unsigned int tun_chr_poll(struct file *file, poll_table *wait)
1112 {
1113 struct tun_file *tfile = file->private_data;
1114 struct tun_struct *tun = __tun_get(tfile);
1115 struct sock *sk;
1116 unsigned int mask = 0;
1117
1118 if (!tun)
1119 return POLLERR;
1120
1121 sk = tfile->socket.sk;
1122
1123 tun_debug(KERN_INFO, tun, "tun_chr_poll\n");
1124
1125 poll_wait(file, sk_sleep(sk), wait);
1126
1127 if (!skb_array_empty(&tfile->tx_array))
1128 mask |= POLLIN | POLLRDNORM;
1129
1130 if (tun->dev->flags & IFF_UP &&
1131 (sock_writeable(sk) ||
1132 (!test_and_set_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags) &&
1133 sock_writeable(sk))))
1134 mask |= POLLOUT | POLLWRNORM;
1135
1136 if (tun->dev->reg_state != NETREG_REGISTERED)
1137 mask = POLLERR;
1138
1139 tun_put(tun);
1140 return mask;
1141 }
1142
1143 /* prepad is the amount to reserve at front. len is length after that.
1144 * linear is a hint as to how much to copy (usually headers). */
1145 static struct sk_buff *tun_alloc_skb(struct tun_file *tfile,
1146 size_t prepad, size_t len,
1147 size_t linear, int noblock)
1148 {
1149 struct sock *sk = tfile->socket.sk;
1150 struct sk_buff *skb;
1151 int err;
1152
1153 /* Under a page? Don't bother with paged skb. */
1154 if (prepad + len < PAGE_SIZE || !linear)
1155 linear = len;
1156
1157 skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
1158 &err, 0);
1159 if (!skb)
1160 return ERR_PTR(err);
1161
1162 skb_reserve(skb, prepad);
1163 skb_put(skb, linear);
1164 skb->data_len = len - linear;
1165 skb->len += len - linear;
1166
1167 return skb;
1168 }
1169
1170 /* Get packet from user space buffer */
1171 static ssize_t tun_get_user(struct tun_struct *tun, struct tun_file *tfile,
1172 void *msg_control, struct iov_iter *from,
1173 int noblock)
1174 {
1175 struct tun_pi pi = { 0, cpu_to_be16(ETH_P_IP) };
1176 struct sk_buff *skb;
1177 size_t total_len = iov_iter_count(from);
1178 size_t len = total_len, align = tun->align, linear;
1179 struct virtio_net_hdr gso = { 0 };
1180 struct tun_pcpu_stats *stats;
1181 int good_linear;
1182 int copylen;
1183 bool zerocopy = false;
1184 int err;
1185 u32 rxhash;
1186 ssize_t n;
1187
1188 if (!(tun->dev->flags & IFF_UP))
1189 return -EIO;
1190
1191 if (!(tun->flags & IFF_NO_PI)) {
1192 if (len < sizeof(pi))
1193 return -EINVAL;
1194 len -= sizeof(pi);
1195
1196 n = copy_from_iter(&pi, sizeof(pi), from);
1197 if (n != sizeof(pi))
1198 return -EFAULT;
1199 }
1200
1201 if (tun->flags & IFF_VNET_HDR) {
1202 int vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
1203
1204 if (len < vnet_hdr_sz)
1205 return -EINVAL;
1206 len -= vnet_hdr_sz;
1207
1208 n = copy_from_iter(&gso, sizeof(gso), from);
1209 if (n != sizeof(gso))
1210 return -EFAULT;
1211
1212 if ((gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
1213 tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2 > tun16_to_cpu(tun, gso.hdr_len))
1214 gso.hdr_len = cpu_to_tun16(tun, tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2);
1215
1216 if (tun16_to_cpu(tun, gso.hdr_len) > len)
1217 return -EINVAL;
1218 iov_iter_advance(from, vnet_hdr_sz - sizeof(gso));
1219 }
1220
1221 if ((tun->flags & TUN_TYPE_MASK) == IFF_TAP) {
1222 align += NET_IP_ALIGN;
1223 if (unlikely(len < ETH_HLEN ||
1224 (gso.hdr_len && tun16_to_cpu(tun, gso.hdr_len) < ETH_HLEN)))
1225 return -EINVAL;
1226 }
1227
1228 good_linear = SKB_MAX_HEAD(align);
1229
1230 if (msg_control) {
1231 struct iov_iter i = *from;
1232
1233 /* There are 256 bytes to be copied in skb, so there is
1234 * enough room for skb expand head in case it is used.
1235 * The rest of the buffer is mapped from userspace.
1236 */
1237 copylen = gso.hdr_len ? tun16_to_cpu(tun, gso.hdr_len) : GOODCOPY_LEN;
1238 if (copylen > good_linear)
1239 copylen = good_linear;
1240 linear = copylen;
1241 iov_iter_advance(&i, copylen);
1242 if (iov_iter_npages(&i, INT_MAX) <= MAX_SKB_FRAGS)
1243 zerocopy = true;
1244 }
1245
1246 if (!zerocopy) {
1247 copylen = len;
1248 if (tun16_to_cpu(tun, gso.hdr_len) > good_linear)
1249 linear = good_linear;
1250 else
1251 linear = tun16_to_cpu(tun, gso.hdr_len);
1252 }
1253
1254 skb = tun_alloc_skb(tfile, align, copylen, linear, noblock);
1255 if (IS_ERR(skb)) {
1256 if (PTR_ERR(skb) != -EAGAIN)
1257 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1258 return PTR_ERR(skb);
1259 }
1260
1261 if (zerocopy)
1262 err = zerocopy_sg_from_iter(skb, from);
1263 else
1264 err = skb_copy_datagram_from_iter(skb, 0, from, len);
1265
1266 if (err) {
1267 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1268 kfree_skb(skb);
1269 return -EFAULT;
1270 }
1271
1272 err = virtio_net_hdr_to_skb(skb, &gso, tun_is_little_endian(tun));
1273 if (err) {
1274 this_cpu_inc(tun->pcpu_stats->rx_frame_errors);
1275 kfree_skb(skb);
1276 return -EINVAL;
1277 }
1278
1279 switch (tun->flags & TUN_TYPE_MASK) {
1280 case IFF_TUN:
1281 if (tun->flags & IFF_NO_PI) {
1282 switch (skb->data[0] & 0xf0) {
1283 case 0x40:
1284 pi.proto = htons(ETH_P_IP);
1285 break;
1286 case 0x60:
1287 pi.proto = htons(ETH_P_IPV6);
1288 break;
1289 default:
1290 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1291 kfree_skb(skb);
1292 return -EINVAL;
1293 }
1294 }
1295
1296 skb_reset_mac_header(skb);
1297 skb->protocol = pi.proto;
1298 skb->dev = tun->dev;
1299 break;
1300 case IFF_TAP:
1301 skb->protocol = eth_type_trans(skb, tun->dev);
1302 break;
1303 }
1304
1305 /* copy skb_ubuf_info for callback when skb has no error */
1306 if (zerocopy) {
1307 skb_shinfo(skb)->destructor_arg = msg_control;
1308 skb_shinfo(skb)->tx_flags |= SKBTX_DEV_ZEROCOPY;
1309 skb_shinfo(skb)->tx_flags |= SKBTX_SHARED_FRAG;
1310 } else if (msg_control) {
1311 struct ubuf_info *uarg = msg_control;
1312 uarg->callback(uarg, false);
1313 }
1314
1315 skb_reset_network_header(skb);
1316 skb_probe_transport_header(skb, 0);
1317
1318 rxhash = skb_get_hash(skb);
1319 netif_rx_ni(skb);
1320
1321 stats = get_cpu_ptr(tun->pcpu_stats);
1322 u64_stats_update_begin(&stats->syncp);
1323 stats->rx_packets++;
1324 stats->rx_bytes += len;
1325 u64_stats_update_end(&stats->syncp);
1326 put_cpu_ptr(stats);
1327
1328 tun_flow_update(tun, rxhash, tfile);
1329 return total_len;
1330 }
1331
1332 static ssize_t tun_chr_write_iter(struct kiocb *iocb, struct iov_iter *from)
1333 {
1334 struct file *file = iocb->ki_filp;
1335 struct tun_struct *tun = tun_get(file);
1336 struct tun_file *tfile = file->private_data;
1337 ssize_t result;
1338
1339 if (!tun)
1340 return -EBADFD;
1341
1342 result = tun_get_user(tun, tfile, NULL, from, file->f_flags & O_NONBLOCK);
1343
1344 tun_put(tun);
1345 return result;
1346 }
1347
1348 /* Put packet to the user space buffer */
1349 static ssize_t tun_put_user(struct tun_struct *tun,
1350 struct tun_file *tfile,
1351 struct sk_buff *skb,
1352 struct iov_iter *iter)
1353 {
1354 struct tun_pi pi = { 0, skb->protocol };
1355 struct tun_pcpu_stats *stats;
1356 ssize_t total;
1357 int vlan_offset = 0;
1358 int vlan_hlen = 0;
1359 int vnet_hdr_sz = 0;
1360
1361 if (skb_vlan_tag_present(skb))
1362 vlan_hlen = VLAN_HLEN;
1363
1364 if (tun->flags & IFF_VNET_HDR)
1365 vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
1366
1367 total = skb->len + vlan_hlen + vnet_hdr_sz;
1368
1369 if (!(tun->flags & IFF_NO_PI)) {
1370 if (iov_iter_count(iter) < sizeof(pi))
1371 return -EINVAL;
1372
1373 total += sizeof(pi);
1374 if (iov_iter_count(iter) < total) {
1375 /* Packet will be striped */
1376 pi.flags |= TUN_PKT_STRIP;
1377 }
1378
1379 if (copy_to_iter(&pi, sizeof(pi), iter) != sizeof(pi))
1380 return -EFAULT;
1381 }
1382
1383 if (vnet_hdr_sz) {
1384 struct virtio_net_hdr gso = { 0 }; /* no info leak */
1385 int ret;
1386
1387 if (iov_iter_count(iter) < vnet_hdr_sz)
1388 return -EINVAL;
1389
1390 ret = virtio_net_hdr_from_skb(skb, &gso,
1391 tun_is_little_endian(tun), true);
1392 if (ret) {
1393 struct skb_shared_info *sinfo = skb_shinfo(skb);
1394 pr_err("unexpected GSO type: "
1395 "0x%x, gso_size %d, hdr_len %d\n",
1396 sinfo->gso_type, tun16_to_cpu(tun, gso.gso_size),
1397 tun16_to_cpu(tun, gso.hdr_len));
1398 print_hex_dump(KERN_ERR, "tun: ",
1399 DUMP_PREFIX_NONE,
1400 16, 1, skb->head,
1401 min((int)tun16_to_cpu(tun, gso.hdr_len), 64), true);
1402 WARN_ON_ONCE(1);
1403 return -EINVAL;
1404 }
1405
1406 if (copy_to_iter(&gso, sizeof(gso), iter) != sizeof(gso))
1407 return -EFAULT;
1408
1409 iov_iter_advance(iter, vnet_hdr_sz - sizeof(gso));
1410 }
1411
1412 if (vlan_hlen) {
1413 int ret;
1414 struct {
1415 __be16 h_vlan_proto;
1416 __be16 h_vlan_TCI;
1417 } veth;
1418
1419 veth.h_vlan_proto = skb->vlan_proto;
1420 veth.h_vlan_TCI = htons(skb_vlan_tag_get(skb));
1421
1422 vlan_offset = offsetof(struct vlan_ethhdr, h_vlan_proto);
1423
1424 ret = skb_copy_datagram_iter(skb, 0, iter, vlan_offset);
1425 if (ret || !iov_iter_count(iter))
1426 goto done;
1427
1428 ret = copy_to_iter(&veth, sizeof(veth), iter);
1429 if (ret != sizeof(veth) || !iov_iter_count(iter))
1430 goto done;
1431 }
1432
1433 skb_copy_datagram_iter(skb, vlan_offset, iter, skb->len - vlan_offset);
1434
1435 done:
1436 /* caller is in process context, */
1437 stats = get_cpu_ptr(tun->pcpu_stats);
1438 u64_stats_update_begin(&stats->syncp);
1439 stats->tx_packets++;
1440 stats->tx_bytes += skb->len + vlan_hlen;
1441 u64_stats_update_end(&stats->syncp);
1442 put_cpu_ptr(tun->pcpu_stats);
1443
1444 return total;
1445 }
1446
1447 static struct sk_buff *tun_ring_recv(struct tun_file *tfile, int noblock,
1448 int *err)
1449 {
1450 DECLARE_WAITQUEUE(wait, current);
1451 struct sk_buff *skb = NULL;
1452 int error = 0;
1453
1454 skb = skb_array_consume(&tfile->tx_array);
1455 if (skb)
1456 goto out;
1457 if (noblock) {
1458 error = -EAGAIN;
1459 goto out;
1460 }
1461
1462 add_wait_queue(&tfile->wq.wait, &wait);
1463 current->state = TASK_INTERRUPTIBLE;
1464
1465 while (1) {
1466 skb = skb_array_consume(&tfile->tx_array);
1467 if (skb)
1468 break;
1469 if (signal_pending(current)) {
1470 error = -ERESTARTSYS;
1471 break;
1472 }
1473 if (tfile->socket.sk->sk_shutdown & RCV_SHUTDOWN) {
1474 error = -EFAULT;
1475 break;
1476 }
1477
1478 schedule();
1479 }
1480
1481 current->state = TASK_RUNNING;
1482 remove_wait_queue(&tfile->wq.wait, &wait);
1483
1484 out:
1485 *err = error;
1486 return skb;
1487 }
1488
1489 static ssize_t tun_do_read(struct tun_struct *tun, struct tun_file *tfile,
1490 struct iov_iter *to,
1491 int noblock)
1492 {
1493 struct sk_buff *skb;
1494 ssize_t ret;
1495 int err;
1496
1497 tun_debug(KERN_INFO, tun, "tun_do_read\n");
1498
1499 if (!iov_iter_count(to))
1500 return 0;
1501
1502 /* Read frames from ring */
1503 skb = tun_ring_recv(tfile, noblock, &err);
1504 if (!skb)
1505 return err;
1506
1507 ret = tun_put_user(tun, tfile, skb, to);
1508 if (unlikely(ret < 0))
1509 kfree_skb(skb);
1510 else
1511 consume_skb(skb);
1512
1513 return ret;
1514 }
1515
1516 static ssize_t tun_chr_read_iter(struct kiocb *iocb, struct iov_iter *to)
1517 {
1518 struct file *file = iocb->ki_filp;
1519 struct tun_file *tfile = file->private_data;
1520 struct tun_struct *tun = __tun_get(tfile);
1521 ssize_t len = iov_iter_count(to), ret;
1522
1523 if (!tun)
1524 return -EBADFD;
1525 ret = tun_do_read(tun, tfile, to, file->f_flags & O_NONBLOCK);
1526 ret = min_t(ssize_t, ret, len);
1527 if (ret > 0)
1528 iocb->ki_pos = ret;
1529 tun_put(tun);
1530 return ret;
1531 }
1532
1533 static void tun_free_netdev(struct net_device *dev)
1534 {
1535 struct tun_struct *tun = netdev_priv(dev);
1536
1537 BUG_ON(!(list_empty(&tun->disabled)));
1538 free_percpu(tun->pcpu_stats);
1539 tun_flow_uninit(tun);
1540 security_tun_dev_free_security(tun->security);
1541 free_netdev(dev);
1542 }
1543
1544 static void tun_setup(struct net_device *dev)
1545 {
1546 struct tun_struct *tun = netdev_priv(dev);
1547
1548 tun->owner = INVALID_UID;
1549 tun->group = INVALID_GID;
1550
1551 dev->ethtool_ops = &tun_ethtool_ops;
1552 dev->destructor = tun_free_netdev;
1553 /* We prefer our own queue length */
1554 dev->tx_queue_len = TUN_READQ_SIZE;
1555 }
1556
1557 /* Trivial set of netlink ops to allow deleting tun or tap
1558 * device with netlink.
1559 */
1560 static int tun_validate(struct nlattr *tb[], struct nlattr *data[])
1561 {
1562 return -EINVAL;
1563 }
1564
1565 static struct rtnl_link_ops tun_link_ops __read_mostly = {
1566 .kind = DRV_NAME,
1567 .priv_size = sizeof(struct tun_struct),
1568 .setup = tun_setup,
1569 .validate = tun_validate,
1570 };
1571
1572 static void tun_sock_write_space(struct sock *sk)
1573 {
1574 struct tun_file *tfile;
1575 wait_queue_head_t *wqueue;
1576
1577 if (!sock_writeable(sk))
1578 return;
1579
1580 if (!test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags))
1581 return;
1582
1583 wqueue = sk_sleep(sk);
1584 if (wqueue && waitqueue_active(wqueue))
1585 wake_up_interruptible_sync_poll(wqueue, POLLOUT |
1586 POLLWRNORM | POLLWRBAND);
1587
1588 tfile = container_of(sk, struct tun_file, sk);
1589 kill_fasync(&tfile->fasync, SIGIO, POLL_OUT);
1590 }
1591
1592 static int tun_sendmsg(struct socket *sock, struct msghdr *m, size_t total_len)
1593 {
1594 int ret;
1595 struct tun_file *tfile = container_of(sock, struct tun_file, socket);
1596 struct tun_struct *tun = __tun_get(tfile);
1597
1598 if (!tun)
1599 return -EBADFD;
1600
1601 ret = tun_get_user(tun, tfile, m->msg_control, &m->msg_iter,
1602 m->msg_flags & MSG_DONTWAIT);
1603 tun_put(tun);
1604 return ret;
1605 }
1606
1607 static int tun_recvmsg(struct socket *sock, struct msghdr *m, size_t total_len,
1608 int flags)
1609 {
1610 struct tun_file *tfile = container_of(sock, struct tun_file, socket);
1611 struct tun_struct *tun = __tun_get(tfile);
1612 int ret;
1613
1614 if (!tun)
1615 return -EBADFD;
1616
1617 if (flags & ~(MSG_DONTWAIT|MSG_TRUNC|MSG_ERRQUEUE)) {
1618 ret = -EINVAL;
1619 goto out;
1620 }
1621 if (flags & MSG_ERRQUEUE) {
1622 ret = sock_recv_errqueue(sock->sk, m, total_len,
1623 SOL_PACKET, TUN_TX_TIMESTAMP);
1624 goto out;
1625 }
1626 ret = tun_do_read(tun, tfile, &m->msg_iter, flags & MSG_DONTWAIT);
1627 if (ret > (ssize_t)total_len) {
1628 m->msg_flags |= MSG_TRUNC;
1629 ret = flags & MSG_TRUNC ? ret : total_len;
1630 }
1631 out:
1632 tun_put(tun);
1633 return ret;
1634 }
1635
1636 static int tun_peek_len(struct socket *sock)
1637 {
1638 struct tun_file *tfile = container_of(sock, struct tun_file, socket);
1639 struct tun_struct *tun;
1640 int ret = 0;
1641
1642 tun = __tun_get(tfile);
1643 if (!tun)
1644 return 0;
1645
1646 ret = skb_array_peek_len(&tfile->tx_array);
1647 tun_put(tun);
1648
1649 return ret;
1650 }
1651
1652 /* Ops structure to mimic raw sockets with tun */
1653 static const struct proto_ops tun_socket_ops = {
1654 .peek_len = tun_peek_len,
1655 .sendmsg = tun_sendmsg,
1656 .recvmsg = tun_recvmsg,
1657 };
1658
1659 static struct proto tun_proto = {
1660 .name = "tun",
1661 .owner = THIS_MODULE,
1662 .obj_size = sizeof(struct tun_file),
1663 };
1664
1665 static int tun_flags(struct tun_struct *tun)
1666 {
1667 return tun->flags & (TUN_FEATURES | IFF_PERSIST | IFF_TUN | IFF_TAP);
1668 }
1669
1670 static ssize_t tun_show_flags(struct device *dev, struct device_attribute *attr,
1671 char *buf)
1672 {
1673 struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1674 return sprintf(buf, "0x%x\n", tun_flags(tun));
1675 }
1676
1677 static ssize_t tun_show_owner(struct device *dev, struct device_attribute *attr,
1678 char *buf)
1679 {
1680 struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1681 return uid_valid(tun->owner)?
1682 sprintf(buf, "%u\n",
1683 from_kuid_munged(current_user_ns(), tun->owner)):
1684 sprintf(buf, "-1\n");
1685 }
1686
1687 static ssize_t tun_show_group(struct device *dev, struct device_attribute *attr,
1688 char *buf)
1689 {
1690 struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1691 return gid_valid(tun->group) ?
1692 sprintf(buf, "%u\n",
1693 from_kgid_munged(current_user_ns(), tun->group)):
1694 sprintf(buf, "-1\n");
1695 }
1696
1697 static DEVICE_ATTR(tun_flags, 0444, tun_show_flags, NULL);
1698 static DEVICE_ATTR(owner, 0444, tun_show_owner, NULL);
1699 static DEVICE_ATTR(group, 0444, tun_show_group, NULL);
1700
1701 static struct attribute *tun_dev_attrs[] = {
1702 &dev_attr_tun_flags.attr,
1703 &dev_attr_owner.attr,
1704 &dev_attr_group.attr,
1705 NULL
1706 };
1707
1708 static const struct attribute_group tun_attr_group = {
1709 .attrs = tun_dev_attrs
1710 };
1711
1712 static int tun_set_iff(struct net *net, struct file *file, struct ifreq *ifr)
1713 {
1714 struct tun_struct *tun;
1715 struct tun_file *tfile = file->private_data;
1716 struct net_device *dev;
1717 int err;
1718
1719 if (tfile->detached)
1720 return -EINVAL;
1721
1722 dev = __dev_get_by_name(net, ifr->ifr_name);
1723 if (dev) {
1724 if (ifr->ifr_flags & IFF_TUN_EXCL)
1725 return -EBUSY;
1726 if ((ifr->ifr_flags & IFF_TUN) && dev->netdev_ops == &tun_netdev_ops)
1727 tun = netdev_priv(dev);
1728 else if ((ifr->ifr_flags & IFF_TAP) && dev->netdev_ops == &tap_netdev_ops)
1729 tun = netdev_priv(dev);
1730 else
1731 return -EINVAL;
1732
1733 if (!!(ifr->ifr_flags & IFF_MULTI_QUEUE) !=
1734 !!(tun->flags & IFF_MULTI_QUEUE))
1735 return -EINVAL;
1736
1737 if (tun_not_capable(tun))
1738 return -EPERM;
1739 err = security_tun_dev_open(tun->security);
1740 if (err < 0)
1741 return err;
1742
1743 err = tun_attach(tun, file, ifr->ifr_flags & IFF_NOFILTER);
1744 if (err < 0)
1745 return err;
1746
1747 if (tun->flags & IFF_MULTI_QUEUE &&
1748 (tun->numqueues + tun->numdisabled > 1)) {
1749 /* One or more queue has already been attached, no need
1750 * to initialize the device again.
1751 */
1752 return 0;
1753 }
1754 }
1755 else {
1756 char *name;
1757 unsigned long flags = 0;
1758 int queues = ifr->ifr_flags & IFF_MULTI_QUEUE ?
1759 MAX_TAP_QUEUES : 1;
1760
1761 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
1762 return -EPERM;
1763 err = security_tun_dev_create();
1764 if (err < 0)
1765 return err;
1766
1767 /* Set dev type */
1768 if (ifr->ifr_flags & IFF_TUN) {
1769 /* TUN device */
1770 flags |= IFF_TUN;
1771 name = "tun%d";
1772 } else if (ifr->ifr_flags & IFF_TAP) {
1773 /* TAP device */
1774 flags |= IFF_TAP;
1775 name = "tap%d";
1776 } else
1777 return -EINVAL;
1778
1779 if (*ifr->ifr_name)
1780 name = ifr->ifr_name;
1781
1782 dev = alloc_netdev_mqs(sizeof(struct tun_struct), name,
1783 NET_NAME_UNKNOWN, tun_setup, queues,
1784 queues);
1785
1786 if (!dev)
1787 return -ENOMEM;
1788
1789 dev_net_set(dev, net);
1790 dev->rtnl_link_ops = &tun_link_ops;
1791 dev->ifindex = tfile->ifindex;
1792 dev->sysfs_groups[0] = &tun_attr_group;
1793
1794 tun = netdev_priv(dev);
1795 tun->dev = dev;
1796 tun->flags = flags;
1797 tun->txflt.count = 0;
1798 tun->vnet_hdr_sz = sizeof(struct virtio_net_hdr);
1799
1800 tun->align = NET_SKB_PAD;
1801 tun->filter_attached = false;
1802 tun->sndbuf = tfile->socket.sk->sk_sndbuf;
1803
1804 tun->pcpu_stats = netdev_alloc_pcpu_stats(struct tun_pcpu_stats);
1805 if (!tun->pcpu_stats) {
1806 err = -ENOMEM;
1807 goto err_free_dev;
1808 }
1809
1810 spin_lock_init(&tun->lock);
1811
1812 err = security_tun_dev_alloc_security(&tun->security);
1813 if (err < 0)
1814 goto err_free_stat;
1815
1816 tun_net_init(dev);
1817 tun_flow_init(tun);
1818
1819 dev->hw_features = NETIF_F_SG | NETIF_F_FRAGLIST |
1820 TUN_USER_FEATURES | NETIF_F_HW_VLAN_CTAG_TX |
1821 NETIF_F_HW_VLAN_STAG_TX;
1822 dev->features = dev->hw_features | NETIF_F_LLTX;
1823 dev->vlan_features = dev->features &
1824 ~(NETIF_F_HW_VLAN_CTAG_TX |
1825 NETIF_F_HW_VLAN_STAG_TX);
1826
1827 INIT_LIST_HEAD(&tun->disabled);
1828 err = tun_attach(tun, file, false);
1829 if (err < 0)
1830 goto err_free_flow;
1831
1832 err = register_netdevice(tun->dev);
1833 if (err < 0)
1834 goto err_detach;
1835 }
1836
1837 netif_carrier_on(tun->dev);
1838
1839 tun_debug(KERN_INFO, tun, "tun_set_iff\n");
1840
1841 tun->flags = (tun->flags & ~TUN_FEATURES) |
1842 (ifr->ifr_flags & TUN_FEATURES);
1843
1844 /* Make sure persistent devices do not get stuck in
1845 * xoff state.
1846 */
1847 if (netif_running(tun->dev))
1848 netif_tx_wake_all_queues(tun->dev);
1849
1850 strcpy(ifr->ifr_name, tun->dev->name);
1851 return 0;
1852
1853 err_detach:
1854 tun_detach_all(dev);
1855 err_free_flow:
1856 tun_flow_uninit(tun);
1857 security_tun_dev_free_security(tun->security);
1858 err_free_stat:
1859 free_percpu(tun->pcpu_stats);
1860 err_free_dev:
1861 free_netdev(dev);
1862 return err;
1863 }
1864
1865 static void tun_get_iff(struct net *net, struct tun_struct *tun,
1866 struct ifreq *ifr)
1867 {
1868 tun_debug(KERN_INFO, tun, "tun_get_iff\n");
1869
1870 strcpy(ifr->ifr_name, tun->dev->name);
1871
1872 ifr->ifr_flags = tun_flags(tun);
1873
1874 }
1875
1876 /* This is like a cut-down ethtool ops, except done via tun fd so no
1877 * privs required. */
1878 static int set_offload(struct tun_struct *tun, unsigned long arg)
1879 {
1880 netdev_features_t features = 0;
1881
1882 if (arg & TUN_F_CSUM) {
1883 features |= NETIF_F_HW_CSUM;
1884 arg &= ~TUN_F_CSUM;
1885
1886 if (arg & (TUN_F_TSO4|TUN_F_TSO6)) {
1887 if (arg & TUN_F_TSO_ECN) {
1888 features |= NETIF_F_TSO_ECN;
1889 arg &= ~TUN_F_TSO_ECN;
1890 }
1891 if (arg & TUN_F_TSO4)
1892 features |= NETIF_F_TSO;
1893 if (arg & TUN_F_TSO6)
1894 features |= NETIF_F_TSO6;
1895 arg &= ~(TUN_F_TSO4|TUN_F_TSO6);
1896 }
1897
1898 if (arg & TUN_F_UFO) {
1899 features |= NETIF_F_UFO;
1900 arg &= ~TUN_F_UFO;
1901 }
1902 }
1903
1904 /* This gives the user a way to test for new features in future by
1905 * trying to set them. */
1906 if (arg)
1907 return -EINVAL;
1908
1909 tun->set_features = features;
1910 netdev_update_features(tun->dev);
1911
1912 return 0;
1913 }
1914
1915 static void tun_detach_filter(struct tun_struct *tun, int n)
1916 {
1917 int i;
1918 struct tun_file *tfile;
1919
1920 for (i = 0; i < n; i++) {
1921 tfile = rtnl_dereference(tun->tfiles[i]);
1922 lock_sock(tfile->socket.sk);
1923 sk_detach_filter(tfile->socket.sk);
1924 release_sock(tfile->socket.sk);
1925 }
1926
1927 tun->filter_attached = false;
1928 }
1929
1930 static int tun_attach_filter(struct tun_struct *tun)
1931 {
1932 int i, ret = 0;
1933 struct tun_file *tfile;
1934
1935 for (i = 0; i < tun->numqueues; i++) {
1936 tfile = rtnl_dereference(tun->tfiles[i]);
1937 lock_sock(tfile->socket.sk);
1938 ret = sk_attach_filter(&tun->fprog, tfile->socket.sk);
1939 release_sock(tfile->socket.sk);
1940 if (ret) {
1941 tun_detach_filter(tun, i);
1942 return ret;
1943 }
1944 }
1945
1946 tun->filter_attached = true;
1947 return ret;
1948 }
1949
1950 static void tun_set_sndbuf(struct tun_struct *tun)
1951 {
1952 struct tun_file *tfile;
1953 int i;
1954
1955 for (i = 0; i < tun->numqueues; i++) {
1956 tfile = rtnl_dereference(tun->tfiles[i]);
1957 tfile->socket.sk->sk_sndbuf = tun->sndbuf;
1958 }
1959 }
1960
1961 static int tun_set_queue(struct file *file, struct ifreq *ifr)
1962 {
1963 struct tun_file *tfile = file->private_data;
1964 struct tun_struct *tun;
1965 int ret = 0;
1966
1967 rtnl_lock();
1968
1969 if (ifr->ifr_flags & IFF_ATTACH_QUEUE) {
1970 tun = tfile->detached;
1971 if (!tun) {
1972 ret = -EINVAL;
1973 goto unlock;
1974 }
1975 ret = security_tun_dev_attach_queue(tun->security);
1976 if (ret < 0)
1977 goto unlock;
1978 ret = tun_attach(tun, file, false);
1979 } else if (ifr->ifr_flags & IFF_DETACH_QUEUE) {
1980 tun = rtnl_dereference(tfile->tun);
1981 if (!tun || !(tun->flags & IFF_MULTI_QUEUE) || tfile->detached)
1982 ret = -EINVAL;
1983 else
1984 __tun_detach(tfile, false);
1985 } else
1986 ret = -EINVAL;
1987
1988 unlock:
1989 rtnl_unlock();
1990 return ret;
1991 }
1992
1993 static long __tun_chr_ioctl(struct file *file, unsigned int cmd,
1994 unsigned long arg, int ifreq_len)
1995 {
1996 struct tun_file *tfile = file->private_data;
1997 struct tun_struct *tun;
1998 void __user* argp = (void __user*)arg;
1999 struct ifreq ifr;
2000 kuid_t owner;
2001 kgid_t group;
2002 int sndbuf;
2003 int vnet_hdr_sz;
2004 unsigned int ifindex;
2005 int le;
2006 int ret;
2007
2008 if (cmd == TUNSETIFF || cmd == TUNSETQUEUE || _IOC_TYPE(cmd) == 0x89) {
2009 if (copy_from_user(&ifr, argp, ifreq_len))
2010 return -EFAULT;
2011 } else {
2012 memset(&ifr, 0, sizeof(ifr));
2013 }
2014 if (cmd == TUNGETFEATURES) {
2015 /* Currently this just means: "what IFF flags are valid?".
2016 * This is needed because we never checked for invalid flags on
2017 * TUNSETIFF.
2018 */
2019 return put_user(IFF_TUN | IFF_TAP | TUN_FEATURES,
2020 (unsigned int __user*)argp);
2021 } else if (cmd == TUNSETQUEUE)
2022 return tun_set_queue(file, &ifr);
2023
2024 ret = 0;
2025 rtnl_lock();
2026
2027 tun = __tun_get(tfile);
2028 if (cmd == TUNSETIFF && !tun) {
2029 ifr.ifr_name[IFNAMSIZ-1] = '\0';
2030
2031 ret = tun_set_iff(sock_net(&tfile->sk), file, &ifr);
2032
2033 if (ret)
2034 goto unlock;
2035
2036 if (copy_to_user(argp, &ifr, ifreq_len))
2037 ret = -EFAULT;
2038 goto unlock;
2039 }
2040 if (cmd == TUNSETIFINDEX) {
2041 ret = -EPERM;
2042 if (tun)
2043 goto unlock;
2044
2045 ret = -EFAULT;
2046 if (copy_from_user(&ifindex, argp, sizeof(ifindex)))
2047 goto unlock;
2048
2049 ret = 0;
2050 tfile->ifindex = ifindex;
2051 goto unlock;
2052 }
2053
2054 ret = -EBADFD;
2055 if (!tun)
2056 goto unlock;
2057
2058 tun_debug(KERN_INFO, tun, "tun_chr_ioctl cmd %u\n", cmd);
2059
2060 ret = 0;
2061 switch (cmd) {
2062 case TUNGETIFF:
2063 tun_get_iff(current->nsproxy->net_ns, tun, &ifr);
2064
2065 if (tfile->detached)
2066 ifr.ifr_flags |= IFF_DETACH_QUEUE;
2067 if (!tfile->socket.sk->sk_filter)
2068 ifr.ifr_flags |= IFF_NOFILTER;
2069
2070 if (copy_to_user(argp, &ifr, ifreq_len))
2071 ret = -EFAULT;
2072 break;
2073
2074 case TUNSETNOCSUM:
2075 /* Disable/Enable checksum */
2076
2077 /* [unimplemented] */
2078 tun_debug(KERN_INFO, tun, "ignored: set checksum %s\n",
2079 arg ? "disabled" : "enabled");
2080 break;
2081
2082 case TUNSETPERSIST:
2083 /* Disable/Enable persist mode. Keep an extra reference to the
2084 * module to prevent the module being unprobed.
2085 */
2086 if (arg && !(tun->flags & IFF_PERSIST)) {
2087 tun->flags |= IFF_PERSIST;
2088 __module_get(THIS_MODULE);
2089 }
2090 if (!arg && (tun->flags & IFF_PERSIST)) {
2091 tun->flags &= ~IFF_PERSIST;
2092 module_put(THIS_MODULE);
2093 }
2094
2095 tun_debug(KERN_INFO, tun, "persist %s\n",
2096 arg ? "enabled" : "disabled");
2097 break;
2098
2099 case TUNSETOWNER:
2100 /* Set owner of the device */
2101 owner = make_kuid(current_user_ns(), arg);
2102 if (!uid_valid(owner)) {
2103 ret = -EINVAL;
2104 break;
2105 }
2106 tun->owner = owner;
2107 tun_debug(KERN_INFO, tun, "owner set to %u\n",
2108 from_kuid(&init_user_ns, tun->owner));
2109 break;
2110
2111 case TUNSETGROUP:
2112 /* Set group of the device */
2113 group = make_kgid(current_user_ns(), arg);
2114 if (!gid_valid(group)) {
2115 ret = -EINVAL;
2116 break;
2117 }
2118 tun->group = group;
2119 tun_debug(KERN_INFO, tun, "group set to %u\n",
2120 from_kgid(&init_user_ns, tun->group));
2121 break;
2122
2123 case TUNSETLINK:
2124 /* Only allow setting the type when the interface is down */
2125 if (tun->dev->flags & IFF_UP) {
2126 tun_debug(KERN_INFO, tun,
2127 "Linktype set failed because interface is up\n");
2128 ret = -EBUSY;
2129 } else {
2130 tun->dev->type = (int) arg;
2131 tun_debug(KERN_INFO, tun, "linktype set to %d\n",
2132 tun->dev->type);
2133 ret = 0;
2134 }
2135 break;
2136
2137 #ifdef TUN_DEBUG
2138 case TUNSETDEBUG:
2139 tun->debug = arg;
2140 break;
2141 #endif
2142 case TUNSETOFFLOAD:
2143 ret = set_offload(tun, arg);
2144 break;
2145
2146 case TUNSETTXFILTER:
2147 /* Can be set only for TAPs */
2148 ret = -EINVAL;
2149 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
2150 break;
2151 ret = update_filter(&tun->txflt, (void __user *)arg);
2152 break;
2153
2154 case SIOCGIFHWADDR:
2155 /* Get hw address */
2156 memcpy(ifr.ifr_hwaddr.sa_data, tun->dev->dev_addr, ETH_ALEN);
2157 ifr.ifr_hwaddr.sa_family = tun->dev->type;
2158 if (copy_to_user(argp, &ifr, ifreq_len))
2159 ret = -EFAULT;
2160 break;
2161
2162 case SIOCSIFHWADDR:
2163 /* Set hw address */
2164 tun_debug(KERN_DEBUG, tun, "set hw address: %pM\n",
2165 ifr.ifr_hwaddr.sa_data);
2166
2167 ret = dev_set_mac_address(tun->dev, &ifr.ifr_hwaddr);
2168 break;
2169
2170 case TUNGETSNDBUF:
2171 sndbuf = tfile->socket.sk->sk_sndbuf;
2172 if (copy_to_user(argp, &sndbuf, sizeof(sndbuf)))
2173 ret = -EFAULT;
2174 break;
2175
2176 case TUNSETSNDBUF:
2177 if (copy_from_user(&sndbuf, argp, sizeof(sndbuf))) {
2178 ret = -EFAULT;
2179 break;
2180 }
2181
2182 tun->sndbuf = sndbuf;
2183 tun_set_sndbuf(tun);
2184 break;
2185
2186 case TUNGETVNETHDRSZ:
2187 vnet_hdr_sz = tun->vnet_hdr_sz;
2188 if (copy_to_user(argp, &vnet_hdr_sz, sizeof(vnet_hdr_sz)))
2189 ret = -EFAULT;
2190 break;
2191
2192 case TUNSETVNETHDRSZ:
2193 if (copy_from_user(&vnet_hdr_sz, argp, sizeof(vnet_hdr_sz))) {
2194 ret = -EFAULT;
2195 break;
2196 }
2197 if (vnet_hdr_sz < (int)sizeof(struct virtio_net_hdr)) {
2198 ret = -EINVAL;
2199 break;
2200 }
2201
2202 tun->vnet_hdr_sz = vnet_hdr_sz;
2203 break;
2204
2205 case TUNGETVNETLE:
2206 le = !!(tun->flags & TUN_VNET_LE);
2207 if (put_user(le, (int __user *)argp))
2208 ret = -EFAULT;
2209 break;
2210
2211 case TUNSETVNETLE:
2212 if (get_user(le, (int __user *)argp)) {
2213 ret = -EFAULT;
2214 break;
2215 }
2216 if (le)
2217 tun->flags |= TUN_VNET_LE;
2218 else
2219 tun->flags &= ~TUN_VNET_LE;
2220 break;
2221
2222 case TUNGETVNETBE:
2223 ret = tun_get_vnet_be(tun, argp);
2224 break;
2225
2226 case TUNSETVNETBE:
2227 ret = tun_set_vnet_be(tun, argp);
2228 break;
2229
2230 case TUNATTACHFILTER:
2231 /* Can be set only for TAPs */
2232 ret = -EINVAL;
2233 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
2234 break;
2235 ret = -EFAULT;
2236 if (copy_from_user(&tun->fprog, argp, sizeof(tun->fprog)))
2237 break;
2238
2239 ret = tun_attach_filter(tun);
2240 break;
2241
2242 case TUNDETACHFILTER:
2243 /* Can be set only for TAPs */
2244 ret = -EINVAL;
2245 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
2246 break;
2247 ret = 0;
2248 tun_detach_filter(tun, tun->numqueues);
2249 break;
2250
2251 case TUNGETFILTER:
2252 ret = -EINVAL;
2253 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
2254 break;
2255 ret = -EFAULT;
2256 if (copy_to_user(argp, &tun->fprog, sizeof(tun->fprog)))
2257 break;
2258 ret = 0;
2259 break;
2260
2261 default:
2262 ret = -EINVAL;
2263 break;
2264 }
2265
2266 unlock:
2267 rtnl_unlock();
2268 if (tun)
2269 tun_put(tun);
2270 return ret;
2271 }
2272
2273 static long tun_chr_ioctl(struct file *file,
2274 unsigned int cmd, unsigned long arg)
2275 {
2276 return __tun_chr_ioctl(file, cmd, arg, sizeof (struct ifreq));
2277 }
2278
2279 #ifdef CONFIG_COMPAT
2280 static long tun_chr_compat_ioctl(struct file *file,
2281 unsigned int cmd, unsigned long arg)
2282 {
2283 switch (cmd) {
2284 case TUNSETIFF:
2285 case TUNGETIFF:
2286 case TUNSETTXFILTER:
2287 case TUNGETSNDBUF:
2288 case TUNSETSNDBUF:
2289 case SIOCGIFHWADDR:
2290 case SIOCSIFHWADDR:
2291 arg = (unsigned long)compat_ptr(arg);
2292 break;
2293 default:
2294 arg = (compat_ulong_t)arg;
2295 break;
2296 }
2297
2298 /*
2299 * compat_ifreq is shorter than ifreq, so we must not access beyond
2300 * the end of that structure. All fields that are used in this
2301 * driver are compatible though, we don't need to convert the
2302 * contents.
2303 */
2304 return __tun_chr_ioctl(file, cmd, arg, sizeof(struct compat_ifreq));
2305 }
2306 #endif /* CONFIG_COMPAT */
2307
2308 static int tun_chr_fasync(int fd, struct file *file, int on)
2309 {
2310 struct tun_file *tfile = file->private_data;
2311 int ret;
2312
2313 if ((ret = fasync_helper(fd, file, on, &tfile->fasync)) < 0)
2314 goto out;
2315
2316 if (on) {
2317 __f_setown(file, task_pid(current), PIDTYPE_PID, 0);
2318 tfile->flags |= TUN_FASYNC;
2319 } else
2320 tfile->flags &= ~TUN_FASYNC;
2321 ret = 0;
2322 out:
2323 return ret;
2324 }
2325
2326 static int tun_chr_open(struct inode *inode, struct file * file)
2327 {
2328 struct net *net = current->nsproxy->net_ns;
2329 struct tun_file *tfile;
2330
2331 DBG1(KERN_INFO, "tunX: tun_chr_open\n");
2332
2333 tfile = (struct tun_file *)sk_alloc(net, AF_UNSPEC, GFP_KERNEL,
2334 &tun_proto, 0);
2335 if (!tfile)
2336 return -ENOMEM;
2337 RCU_INIT_POINTER(tfile->tun, NULL);
2338 tfile->flags = 0;
2339 tfile->ifindex = 0;
2340
2341 init_waitqueue_head(&tfile->wq.wait);
2342 RCU_INIT_POINTER(tfile->socket.wq, &tfile->wq);
2343
2344 tfile->socket.file = file;
2345 tfile->socket.ops = &tun_socket_ops;
2346
2347 sock_init_data(&tfile->socket, &tfile->sk);
2348
2349 tfile->sk.sk_write_space = tun_sock_write_space;
2350 tfile->sk.sk_sndbuf = INT_MAX;
2351
2352 file->private_data = tfile;
2353 INIT_LIST_HEAD(&tfile->next);
2354
2355 sock_set_flag(&tfile->sk, SOCK_ZEROCOPY);
2356
2357 return 0;
2358 }
2359
2360 static int tun_chr_close(struct inode *inode, struct file *file)
2361 {
2362 struct tun_file *tfile = file->private_data;
2363
2364 tun_detach(tfile, true);
2365
2366 return 0;
2367 }
2368
2369 #ifdef CONFIG_PROC_FS
2370 static void tun_chr_show_fdinfo(struct seq_file *m, struct file *f)
2371 {
2372 struct tun_struct *tun;
2373 struct ifreq ifr;
2374
2375 memset(&ifr, 0, sizeof(ifr));
2376
2377 rtnl_lock();
2378 tun = tun_get(f);
2379 if (tun)
2380 tun_get_iff(current->nsproxy->net_ns, tun, &ifr);
2381 rtnl_unlock();
2382
2383 if (tun)
2384 tun_put(tun);
2385
2386 seq_printf(m, "iff:\t%s\n", ifr.ifr_name);
2387 }
2388 #endif
2389
2390 static const struct file_operations tun_fops = {
2391 .owner = THIS_MODULE,
2392 .llseek = no_llseek,
2393 .read_iter = tun_chr_read_iter,
2394 .write_iter = tun_chr_write_iter,
2395 .poll = tun_chr_poll,
2396 .unlocked_ioctl = tun_chr_ioctl,
2397 #ifdef CONFIG_COMPAT
2398 .compat_ioctl = tun_chr_compat_ioctl,
2399 #endif
2400 .open = tun_chr_open,
2401 .release = tun_chr_close,
2402 .fasync = tun_chr_fasync,
2403 #ifdef CONFIG_PROC_FS
2404 .show_fdinfo = tun_chr_show_fdinfo,
2405 #endif
2406 };
2407
2408 static struct miscdevice tun_miscdev = {
2409 .minor = TUN_MINOR,
2410 .name = "tun",
2411 .nodename = "net/tun",
2412 .fops = &tun_fops,
2413 };
2414
2415 /* ethtool interface */
2416
2417 static int tun_get_settings(struct net_device *dev, struct ethtool_cmd *cmd)
2418 {
2419 cmd->supported = 0;
2420 cmd->advertising = 0;
2421 ethtool_cmd_speed_set(cmd, SPEED_10);
2422 cmd->duplex = DUPLEX_FULL;
2423 cmd->port = PORT_TP;
2424 cmd->phy_address = 0;
2425 cmd->transceiver = XCVR_INTERNAL;
2426 cmd->autoneg = AUTONEG_DISABLE;
2427 cmd->maxtxpkt = 0;
2428 cmd->maxrxpkt = 0;
2429 return 0;
2430 }
2431
2432 static void tun_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
2433 {
2434 struct tun_struct *tun = netdev_priv(dev);
2435
2436 strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
2437 strlcpy(info->version, DRV_VERSION, sizeof(info->version));
2438
2439 switch (tun->flags & TUN_TYPE_MASK) {
2440 case IFF_TUN:
2441 strlcpy(info->bus_info, "tun", sizeof(info->bus_info));
2442 break;
2443 case IFF_TAP:
2444 strlcpy(info->bus_info, "tap", sizeof(info->bus_info));
2445 break;
2446 }
2447 }
2448
2449 static u32 tun_get_msglevel(struct net_device *dev)
2450 {
2451 #ifdef TUN_DEBUG
2452 struct tun_struct *tun = netdev_priv(dev);
2453 return tun->debug;
2454 #else
2455 return -EOPNOTSUPP;
2456 #endif
2457 }
2458
2459 static void tun_set_msglevel(struct net_device *dev, u32 value)
2460 {
2461 #ifdef TUN_DEBUG
2462 struct tun_struct *tun = netdev_priv(dev);
2463 tun->debug = value;
2464 #endif
2465 }
2466
2467 static const struct ethtool_ops tun_ethtool_ops = {
2468 .get_settings = tun_get_settings,
2469 .get_drvinfo = tun_get_drvinfo,
2470 .get_msglevel = tun_get_msglevel,
2471 .set_msglevel = tun_set_msglevel,
2472 .get_link = ethtool_op_get_link,
2473 .get_ts_info = ethtool_op_get_ts_info,
2474 };
2475
2476 static int tun_queue_resize(struct tun_struct *tun)
2477 {
2478 struct net_device *dev = tun->dev;
2479 struct tun_file *tfile;
2480 struct skb_array **arrays;
2481 int n = tun->numqueues + tun->numdisabled;
2482 int ret, i;
2483
2484 arrays = kmalloc(sizeof *arrays * n, GFP_KERNEL);
2485 if (!arrays)
2486 return -ENOMEM;
2487
2488 for (i = 0; i < tun->numqueues; i++) {
2489 tfile = rtnl_dereference(tun->tfiles[i]);
2490 arrays[i] = &tfile->tx_array;
2491 }
2492 list_for_each_entry(tfile, &tun->disabled, next)
2493 arrays[i++] = &tfile->tx_array;
2494
2495 ret = skb_array_resize_multiple(arrays, n,
2496 dev->tx_queue_len, GFP_KERNEL);
2497
2498 kfree(arrays);
2499 return ret;
2500 }
2501
2502 static int tun_device_event(struct notifier_block *unused,
2503 unsigned long event, void *ptr)
2504 {
2505 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
2506 struct tun_struct *tun = netdev_priv(dev);
2507
2508 if (dev->rtnl_link_ops != &tun_link_ops)
2509 return NOTIFY_DONE;
2510
2511 switch (event) {
2512 case NETDEV_CHANGE_TX_QUEUE_LEN:
2513 if (tun_queue_resize(tun))
2514 return NOTIFY_BAD;
2515 break;
2516 default:
2517 break;
2518 }
2519
2520 return NOTIFY_DONE;
2521 }
2522
2523 static struct notifier_block tun_notifier_block __read_mostly = {
2524 .notifier_call = tun_device_event,
2525 };
2526
2527 static int __init tun_init(void)
2528 {
2529 int ret = 0;
2530
2531 pr_info("%s, %s\n", DRV_DESCRIPTION, DRV_VERSION);
2532 pr_info("%s\n", DRV_COPYRIGHT);
2533
2534 ret = rtnl_link_register(&tun_link_ops);
2535 if (ret) {
2536 pr_err("Can't register link_ops\n");
2537 goto err_linkops;
2538 }
2539
2540 ret = misc_register(&tun_miscdev);
2541 if (ret) {
2542 pr_err("Can't register misc device %d\n", TUN_MINOR);
2543 goto err_misc;
2544 }
2545
2546 register_netdevice_notifier(&tun_notifier_block);
2547 return 0;
2548 err_misc:
2549 rtnl_link_unregister(&tun_link_ops);
2550 err_linkops:
2551 return ret;
2552 }
2553
2554 static void tun_cleanup(void)
2555 {
2556 misc_deregister(&tun_miscdev);
2557 rtnl_link_unregister(&tun_link_ops);
2558 unregister_netdevice_notifier(&tun_notifier_block);
2559 }
2560
2561 /* Get an underlying socket object from tun file. Returns error unless file is
2562 * attached to a device. The returned object works like a packet socket, it
2563 * can be used for sock_sendmsg/sock_recvmsg. The caller is responsible for
2564 * holding a reference to the file for as long as the socket is in use. */
2565 struct socket *tun_get_socket(struct file *file)
2566 {
2567 struct tun_file *tfile;
2568 if (file->f_op != &tun_fops)
2569 return ERR_PTR(-EINVAL);
2570 tfile = file->private_data;
2571 if (!tfile)
2572 return ERR_PTR(-EBADFD);
2573 return &tfile->socket;
2574 }
2575 EXPORT_SYMBOL_GPL(tun_get_socket);
2576
2577 module_init(tun_init);
2578 module_exit(tun_cleanup);
2579 MODULE_DESCRIPTION(DRV_DESCRIPTION);
2580 MODULE_AUTHOR(DRV_COPYRIGHT);
2581 MODULE_LICENSE("GPL");
2582 MODULE_ALIAS_MISCDEV(TUN_MINOR);
2583 MODULE_ALIAS("devname:net/tun");
Linux with added FPC-III support