?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
watchdog/core: Rename some softlockup_* functions
[linux/fpc-iii.git] / drivers / net / tun.c
blob3c9985f299503ea65dad7eb3b47e2ab3bef87800
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/sched/signal.h>
48 #include <linux/major.h>
49 #include <linux/slab.h>
50 #include <linux/poll.h>
51 #include <linux/fcntl.h>
52 #include <linux/init.h>
53 #include <linux/skbuff.h>
54 #include <linux/netdevice.h>
55 #include <linux/etherdevice.h>
56 #include <linux/miscdevice.h>
57 #include <linux/ethtool.h>
58 #include <linux/rtnetlink.h>
59 #include <linux/compat.h>
60 #include <linux/if.h>
61 #include <linux/if_arp.h>
62 #include <linux/if_ether.h>
63 #include <linux/if_tun.h>
64 #include <linux/if_vlan.h>
65 #include <linux/crc32.h>
66 #include <linux/nsproxy.h>
67 #include <linux/virtio_net.h>
68 #include <linux/rcupdate.h>
69 #include <net/net_namespace.h>
70 #include <net/netns/generic.h>
71 #include <net/rtnetlink.h>
72 #include <net/sock.h>
73 #include <linux/seq_file.h>
74 #include <linux/uio.h>
75 #include <linux/skb_array.h>
76 #include <linux/bpf.h>
77 #include <linux/bpf_trace.h>
78
79 #include <linux/uaccess.h>
80
81 /* Uncomment to enable debugging */
82 /* #define TUN_DEBUG 1 */
83
84 #ifdef TUN_DEBUG
85 static int debug;
86
87 #define tun_debug(level, tun, fmt, args...) \
88 do { \
89 if (tun->debug) \
90 netdev_printk(level, tun->dev, fmt, ##args); \
91 } while (0)
92 #define DBG1(level, fmt, args...) \
93 do { \
94 if (debug == 2) \
95 printk(level fmt, ##args); \
96 } while (0)
97 #else
98 #define tun_debug(level, tun, fmt, args...) \
99 do { \
100 if (0) \
101 netdev_printk(level, tun->dev, fmt, ##args); \
102 } while (0)
103 #define DBG1(level, fmt, args...) \
104 do { \
105 if (0) \
106 printk(level fmt, ##args); \
107 } while (0)
108 #endif
109
110 #define TUN_HEADROOM 256
111 #define TUN_RX_PAD (NET_IP_ALIGN + NET_SKB_PAD)
112
113 /* TUN device flags */
114
115 /* IFF_ATTACH_QUEUE is never stored in device flags,
116 * overload it to mean fasync when stored there.
117 */
118 #define TUN_FASYNC IFF_ATTACH_QUEUE
119 /* High bits in flags field are unused. */
120 #define TUN_VNET_LE 0x80000000
121 #define TUN_VNET_BE 0x40000000
122
123 #define TUN_FEATURES (IFF_NO_PI | IFF_ONE_QUEUE | IFF_VNET_HDR | \
124 IFF_MULTI_QUEUE)
125 #define GOODCOPY_LEN 128
126
127 #define FLT_EXACT_COUNT 8
128 struct tap_filter {
129 unsigned int count; /* Number of addrs. Zero means disabled */
130 u32 mask[2]; /* Mask of the hashed addrs */
131 unsigned char addr[FLT_EXACT_COUNT][ETH_ALEN];
132 };
133
134 /* MAX_TAP_QUEUES 256 is chosen to allow rx/tx queues to be equal
135 * to max number of VCPUs in guest. */
136 #define MAX_TAP_QUEUES 256
137 #define MAX_TAP_FLOWS 4096
138
139 #define TUN_FLOW_EXPIRE (3 * HZ)
140
141 struct tun_pcpu_stats {
142 u64 rx_packets;
143 u64 rx_bytes;
144 u64 tx_packets;
145 u64 tx_bytes;
146 struct u64_stats_sync syncp;
147 u32 rx_dropped;
148 u32 tx_dropped;
149 u32 rx_frame_errors;
150 };
151
152 /* A tun_file connects an open character device to a tuntap netdevice. It
153 * also contains all socket related structures (except sock_fprog and tap_filter)
154 * to serve as one transmit queue for tuntap device. The sock_fprog and
155 * tap_filter were kept in tun_struct since they were used for filtering for the
156 * netdevice not for a specific queue (at least I didn't see the requirement for
157 * this).
158 *
159 * RCU usage:
160 * The tun_file and tun_struct are loosely coupled, the pointer from one to the
161 * other can only be read while rcu_read_lock or rtnl_lock is held.
162 */
163 struct tun_file {
164 struct sock sk;
165 struct socket socket;
166 struct socket_wq wq;
167 struct tun_struct __rcu *tun;
168 struct fasync_struct *fasync;
169 /* only used for fasnyc */
170 unsigned int flags;
171 union {
172 u16 queue_index;
173 unsigned int ifindex;
174 };
175 struct list_head next;
176 struct tun_struct *detached;
177 struct skb_array tx_array;
178 };
179
180 struct tun_flow_entry {
181 struct hlist_node hash_link;
182 struct rcu_head rcu;
183 struct tun_struct *tun;
184
185 u32 rxhash;
186 u32 rps_rxhash;
187 int queue_index;
188 unsigned long updated;
189 };
190
191 #define TUN_NUM_FLOW_ENTRIES 1024
192
193 /* Since the socket were moved to tun_file, to preserve the behavior of persist
194 * device, socket filter, sndbuf and vnet header size were restore when the
195 * file were attached to a persist device.
196 */
197 struct tun_struct {
198 struct tun_file __rcu *tfiles[MAX_TAP_QUEUES];
199 unsigned int numqueues;
200 unsigned int flags;
201 kuid_t owner;
202 kgid_t group;
203
204 struct net_device *dev;
205 netdev_features_t set_features;
206 #define TUN_USER_FEATURES (NETIF_F_HW_CSUM|NETIF_F_TSO_ECN|NETIF_F_TSO| \
207 NETIF_F_TSO6)
208
209 int align;
210 int vnet_hdr_sz;
211 int sndbuf;
212 struct tap_filter txflt;
213 struct sock_fprog fprog;
214 /* protected by rtnl lock */
215 bool filter_attached;
216 #ifdef TUN_DEBUG
217 int debug;
218 #endif
219 spinlock_t lock;
220 struct hlist_head flows[TUN_NUM_FLOW_ENTRIES];
221 struct timer_list flow_gc_timer;
222 unsigned long ageing_time;
223 unsigned int numdisabled;
224 struct list_head disabled;
225 void *security;
226 u32 flow_count;
227 u32 rx_batched;
228 struct tun_pcpu_stats __percpu *pcpu_stats;
229 struct bpf_prog __rcu *xdp_prog;
230 };
231
232 #ifdef CONFIG_TUN_VNET_CROSS_LE
233 static inline bool tun_legacy_is_little_endian(struct tun_struct *tun)
234 {
235 return tun->flags & TUN_VNET_BE ? false :
236 virtio_legacy_is_little_endian();
237 }
238
239 static long tun_get_vnet_be(struct tun_struct *tun, int __user *argp)
240 {
241 int be = !!(tun->flags & TUN_VNET_BE);
242
243 if (put_user(be, argp))
244 return -EFAULT;
245
246 return 0;
247 }
248
249 static long tun_set_vnet_be(struct tun_struct *tun, int __user *argp)
250 {
251 int be;
252
253 if (get_user(be, argp))
254 return -EFAULT;
255
256 if (be)
257 tun->flags |= TUN_VNET_BE;
258 else
259 tun->flags &= ~TUN_VNET_BE;
260
261 return 0;
262 }
263 #else
264 static inline bool tun_legacy_is_little_endian(struct tun_struct *tun)
265 {
266 return virtio_legacy_is_little_endian();
267 }
268
269 static long tun_get_vnet_be(struct tun_struct *tun, int __user *argp)
270 {
271 return -EINVAL;
272 }
273
274 static long tun_set_vnet_be(struct tun_struct *tun, int __user *argp)
275 {
276 return -EINVAL;
277 }
278 #endif /* CONFIG_TUN_VNET_CROSS_LE */
279
280 static inline bool tun_is_little_endian(struct tun_struct *tun)
281 {
282 return tun->flags & TUN_VNET_LE ||
283 tun_legacy_is_little_endian(tun);
284 }
285
286 static inline u16 tun16_to_cpu(struct tun_struct *tun, __virtio16 val)
287 {
288 return __virtio16_to_cpu(tun_is_little_endian(tun), val);
289 }
290
291 static inline __virtio16 cpu_to_tun16(struct tun_struct *tun, u16 val)
292 {
293 return __cpu_to_virtio16(tun_is_little_endian(tun), val);
294 }
295
296 static inline u32 tun_hashfn(u32 rxhash)
297 {
298 return rxhash & 0x3ff;
299 }
300
301 static struct tun_flow_entry *tun_flow_find(struct hlist_head *head, u32 rxhash)
302 {
303 struct tun_flow_entry *e;
304
305 hlist_for_each_entry_rcu(e, head, hash_link) {
306 if (e->rxhash == rxhash)
307 return e;
308 }
309 return NULL;
310 }
311
312 static struct tun_flow_entry *tun_flow_create(struct tun_struct *tun,
313 struct hlist_head *head,
314 u32 rxhash, u16 queue_index)
315 {
316 struct tun_flow_entry *e = kmalloc(sizeof(*e), GFP_ATOMIC);
317
318 if (e) {
319 tun_debug(KERN_INFO, tun, "create flow: hash %u index %u\n",
320 rxhash, queue_index);
321 e->updated = jiffies;
322 e->rxhash = rxhash;
323 e->rps_rxhash = 0;
324 e->queue_index = queue_index;
325 e->tun = tun;
326 hlist_add_head_rcu(&e->hash_link, head);
327 ++tun->flow_count;
328 }
329 return e;
330 }
331
332 static void tun_flow_delete(struct tun_struct *tun, struct tun_flow_entry *e)
333 {
334 tun_debug(KERN_INFO, tun, "delete flow: hash %u index %u\n",
335 e->rxhash, e->queue_index);
336 hlist_del_rcu(&e->hash_link);
337 kfree_rcu(e, rcu);
338 --tun->flow_count;
339 }
340
341 static void tun_flow_flush(struct tun_struct *tun)
342 {
343 int i;
344
345 spin_lock_bh(&tun->lock);
346 for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
347 struct tun_flow_entry *e;
348 struct hlist_node *n;
349
350 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link)
351 tun_flow_delete(tun, e);
352 }
353 spin_unlock_bh(&tun->lock);
354 }
355
356 static void tun_flow_delete_by_queue(struct tun_struct *tun, u16 queue_index)
357 {
358 int i;
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 if (e->queue_index == queue_index)
367 tun_flow_delete(tun, e);
368 }
369 }
370 spin_unlock_bh(&tun->lock);
371 }
372
373 static void tun_flow_cleanup(unsigned long data)
374 {
375 struct tun_struct *tun = (struct tun_struct *)data;
376 unsigned long delay = tun->ageing_time;
377 unsigned long next_timer = jiffies + delay;
378 unsigned long count = 0;
379 int i;
380
381 tun_debug(KERN_INFO, tun, "tun_flow_cleanup\n");
382
383 spin_lock_bh(&tun->lock);
384 for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
385 struct tun_flow_entry *e;
386 struct hlist_node *n;
387
388 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
389 unsigned long this_timer;
390 count++;
391 this_timer = e->updated + delay;
392 if (time_before_eq(this_timer, jiffies))
393 tun_flow_delete(tun, e);
394 else if (time_before(this_timer, next_timer))
395 next_timer = this_timer;
396 }
397 }
398
399 if (count)
400 mod_timer(&tun->flow_gc_timer, round_jiffies_up(next_timer));
401 spin_unlock_bh(&tun->lock);
402 }
403
404 static void tun_flow_update(struct tun_struct *tun, u32 rxhash,
405 struct tun_file *tfile)
406 {
407 struct hlist_head *head;
408 struct tun_flow_entry *e;
409 unsigned long delay = tun->ageing_time;
410 u16 queue_index = tfile->queue_index;
411
412 if (!rxhash)
413 return;
414 else
415 head = &tun->flows[tun_hashfn(rxhash)];
416
417 rcu_read_lock();
418
419 /* We may get a very small possibility of OOO during switching, not
420 * worth to optimize.*/
421 if (tun->numqueues == 1 || tfile->detached)
422 goto unlock;
423
424 e = tun_flow_find(head, rxhash);
425 if (likely(e)) {
426 /* TODO: keep queueing to old queue until it's empty? */
427 e->queue_index = queue_index;
428 e->updated = jiffies;
429 sock_rps_record_flow_hash(e->rps_rxhash);
430 } else {
431 spin_lock_bh(&tun->lock);
432 if (!tun_flow_find(head, rxhash) &&
433 tun->flow_count < MAX_TAP_FLOWS)
434 tun_flow_create(tun, head, rxhash, queue_index);
435
436 if (!timer_pending(&tun->flow_gc_timer))
437 mod_timer(&tun->flow_gc_timer,
438 round_jiffies_up(jiffies + delay));
439 spin_unlock_bh(&tun->lock);
440 }
441
442 unlock:
443 rcu_read_unlock();
444 }
445
446 /**
447 * Save the hash received in the stack receive path and update the
448 * flow_hash table accordingly.
449 */
450 static inline void tun_flow_save_rps_rxhash(struct tun_flow_entry *e, u32 hash)
451 {
452 if (unlikely(e->rps_rxhash != hash))
453 e->rps_rxhash = hash;
454 }
455
456 /* We try to identify a flow through its rxhash first. The reason that
457 * we do not check rxq no. is because some cards(e.g 82599), chooses
458 * the rxq based on the txq where the last packet of the flow comes. As
459 * the userspace application move between processors, we may get a
460 * different rxq no. here. If we could not get rxhash, then we would
461 * hope the rxq no. may help here.
462 */
463 static u16 tun_select_queue(struct net_device *dev, struct sk_buff *skb,
464 void *accel_priv, select_queue_fallback_t fallback)
465 {
466 struct tun_struct *tun = netdev_priv(dev);
467 struct tun_flow_entry *e;
468 u32 txq = 0;
469 u32 numqueues = 0;
470
471 rcu_read_lock();
472 numqueues = ACCESS_ONCE(tun->numqueues);
473
474 txq = __skb_get_hash_symmetric(skb);
475 if (txq) {
476 e = tun_flow_find(&tun->flows[tun_hashfn(txq)], txq);
477 if (e) {
478 tun_flow_save_rps_rxhash(e, txq);
479 txq = e->queue_index;
480 } else
481 /* use multiply and shift instead of expensive divide */
482 txq = ((u64)txq * numqueues) >> 32;
483 } else if (likely(skb_rx_queue_recorded(skb))) {
484 txq = skb_get_rx_queue(skb);
485 while (unlikely(txq >= numqueues))
486 txq -= numqueues;
487 }
488
489 rcu_read_unlock();
490 return txq;
491 }
492
493 static inline bool tun_not_capable(struct tun_struct *tun)
494 {
495 const struct cred *cred = current_cred();
496 struct net *net = dev_net(tun->dev);
497
498 return ((uid_valid(tun->owner) && !uid_eq(cred->euid, tun->owner)) ||
499 (gid_valid(tun->group) && !in_egroup_p(tun->group))) &&
500 !ns_capable(net->user_ns, CAP_NET_ADMIN);
501 }
502
503 static void tun_set_real_num_queues(struct tun_struct *tun)
504 {
505 netif_set_real_num_tx_queues(tun->dev, tun->numqueues);
506 netif_set_real_num_rx_queues(tun->dev, tun->numqueues);
507 }
508
509 static void tun_disable_queue(struct tun_struct *tun, struct tun_file *tfile)
510 {
511 tfile->detached = tun;
512 list_add_tail(&tfile->next, &tun->disabled);
513 ++tun->numdisabled;
514 }
515
516 static struct tun_struct *tun_enable_queue(struct tun_file *tfile)
517 {
518 struct tun_struct *tun = tfile->detached;
519
520 tfile->detached = NULL;
521 list_del_init(&tfile->next);
522 --tun->numdisabled;
523 return tun;
524 }
525
526 static void tun_queue_purge(struct tun_file *tfile)
527 {
528 struct sk_buff *skb;
529
530 while ((skb = skb_array_consume(&tfile->tx_array)) != NULL)
531 kfree_skb(skb);
532
533 skb_queue_purge(&tfile->sk.sk_write_queue);
534 skb_queue_purge(&tfile->sk.sk_error_queue);
535 }
536
537 static void __tun_detach(struct tun_file *tfile, bool clean)
538 {
539 struct tun_file *ntfile;
540 struct tun_struct *tun;
541
542 tun = rtnl_dereference(tfile->tun);
543
544 if (tun && !tfile->detached) {
545 u16 index = tfile->queue_index;
546 BUG_ON(index >= tun->numqueues);
547
548 rcu_assign_pointer(tun->tfiles[index],
549 tun->tfiles[tun->numqueues - 1]);
550 ntfile = rtnl_dereference(tun->tfiles[index]);
551 ntfile->queue_index = index;
552
553 --tun->numqueues;
554 if (clean) {
555 RCU_INIT_POINTER(tfile->tun, NULL);
556 sock_put(&tfile->sk);
557 } else
558 tun_disable_queue(tun, tfile);
559
560 synchronize_net();
561 tun_flow_delete_by_queue(tun, tun->numqueues + 1);
562 /* Drop read queue */
563 tun_queue_purge(tfile);
564 tun_set_real_num_queues(tun);
565 } else if (tfile->detached && clean) {
566 tun = tun_enable_queue(tfile);
567 sock_put(&tfile->sk);
568 }
569
570 if (clean) {
571 if (tun && tun->numqueues == 0 && tun->numdisabled == 0) {
572 netif_carrier_off(tun->dev);
573
574 if (!(tun->flags & IFF_PERSIST) &&
575 tun->dev->reg_state == NETREG_REGISTERED)
576 unregister_netdevice(tun->dev);
577 }
578 if (tun)
579 skb_array_cleanup(&tfile->tx_array);
580 sock_put(&tfile->sk);
581 }
582 }
583
584 static void tun_detach(struct tun_file *tfile, bool clean)
585 {
586 rtnl_lock();
587 __tun_detach(tfile, clean);
588 rtnl_unlock();
589 }
590
591 static void tun_detach_all(struct net_device *dev)
592 {
593 struct tun_struct *tun = netdev_priv(dev);
594 struct bpf_prog *xdp_prog = rtnl_dereference(tun->xdp_prog);
595 struct tun_file *tfile, *tmp;
596 int i, n = tun->numqueues;
597
598 for (i = 0; i < n; i++) {
599 tfile = rtnl_dereference(tun->tfiles[i]);
600 BUG_ON(!tfile);
601 tfile->socket.sk->sk_shutdown = RCV_SHUTDOWN;
602 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
603 RCU_INIT_POINTER(tfile->tun, NULL);
604 --tun->numqueues;
605 }
606 list_for_each_entry(tfile, &tun->disabled, next) {
607 tfile->socket.sk->sk_shutdown = RCV_SHUTDOWN;
608 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
609 RCU_INIT_POINTER(tfile->tun, NULL);
610 }
611 BUG_ON(tun->numqueues != 0);
612
613 synchronize_net();
614 for (i = 0; i < n; i++) {
615 tfile = rtnl_dereference(tun->tfiles[i]);
616 /* Drop read queue */
617 tun_queue_purge(tfile);
618 sock_put(&tfile->sk);
619 }
620 list_for_each_entry_safe(tfile, tmp, &tun->disabled, next) {
621 tun_enable_queue(tfile);
622 tun_queue_purge(tfile);
623 sock_put(&tfile->sk);
624 }
625 BUG_ON(tun->numdisabled != 0);
626
627 if (xdp_prog)
628 bpf_prog_put(xdp_prog);
629
630 if (tun->flags & IFF_PERSIST)
631 module_put(THIS_MODULE);
632 }
633
634 static int tun_attach(struct tun_struct *tun, struct file *file, bool skip_filter)
635 {
636 struct tun_file *tfile = file->private_data;
637 struct net_device *dev = tun->dev;
638 int err;
639
640 err = security_tun_dev_attach(tfile->socket.sk, tun->security);
641 if (err < 0)
642 goto out;
643
644 err = -EINVAL;
645 if (rtnl_dereference(tfile->tun) && !tfile->detached)
646 goto out;
647
648 err = -EBUSY;
649 if (!(tun->flags & IFF_MULTI_QUEUE) && tun->numqueues == 1)
650 goto out;
651
652 err = -E2BIG;
653 if (!tfile->detached &&
654 tun->numqueues + tun->numdisabled == MAX_TAP_QUEUES)
655 goto out;
656
657 err = 0;
658
659 /* Re-attach the filter to persist device */
660 if (!skip_filter && (tun->filter_attached == true)) {
661 lock_sock(tfile->socket.sk);
662 err = sk_attach_filter(&tun->fprog, tfile->socket.sk);
663 release_sock(tfile->socket.sk);
664 if (!err)
665 goto out;
666 }
667
668 if (!tfile->detached &&
669 skb_array_init(&tfile->tx_array, dev->tx_queue_len, GFP_KERNEL)) {
670 err = -ENOMEM;
671 goto out;
672 }
673
674 tfile->queue_index = tun->numqueues;
675 tfile->socket.sk->sk_shutdown &= ~RCV_SHUTDOWN;
676 rcu_assign_pointer(tfile->tun, tun);
677 rcu_assign_pointer(tun->tfiles[tun->numqueues], tfile);
678 tun->numqueues++;
679
680 if (tfile->detached)
681 tun_enable_queue(tfile);
682 else
683 sock_hold(&tfile->sk);
684
685 tun_set_real_num_queues(tun);
686
687 /* device is allowed to go away first, so no need to hold extra
688 * refcnt.
689 */
690
691 out:
692 return err;
693 }
694
695 static struct tun_struct *__tun_get(struct tun_file *tfile)
696 {
697 struct tun_struct *tun;
698
699 rcu_read_lock();
700 tun = rcu_dereference(tfile->tun);
701 if (tun)
702 dev_hold(tun->dev);
703 rcu_read_unlock();
704
705 return tun;
706 }
707
708 static struct tun_struct *tun_get(struct file *file)
709 {
710 return __tun_get(file->private_data);
711 }
712
713 static void tun_put(struct tun_struct *tun)
714 {
715 dev_put(tun->dev);
716 }
717
718 /* TAP filtering */
719 static void addr_hash_set(u32 *mask, const u8 *addr)
720 {
721 int n = ether_crc(ETH_ALEN, addr) >> 26;
722 mask[n >> 5] |= (1 << (n & 31));
723 }
724
725 static unsigned int addr_hash_test(const u32 *mask, const u8 *addr)
726 {
727 int n = ether_crc(ETH_ALEN, addr) >> 26;
728 return mask[n >> 5] & (1 << (n & 31));
729 }
730
731 static int update_filter(struct tap_filter *filter, void __user *arg)
732 {
733 struct { u8 u[ETH_ALEN]; } *addr;
734 struct tun_filter uf;
735 int err, alen, n, nexact;
736
737 if (copy_from_user(&uf, arg, sizeof(uf)))
738 return -EFAULT;
739
740 if (!uf.count) {
741 /* Disabled */
742 filter->count = 0;
743 return 0;
744 }
745
746 alen = ETH_ALEN * uf.count;
747 addr = memdup_user(arg + sizeof(uf), alen);
748 if (IS_ERR(addr))
749 return PTR_ERR(addr);
750
751 /* The filter is updated without holding any locks. Which is
752 * perfectly safe. We disable it first and in the worst
753 * case we'll accept a few undesired packets. */
754 filter->count = 0;
755 wmb();
756
757 /* Use first set of addresses as an exact filter */
758 for (n = 0; n < uf.count && n < FLT_EXACT_COUNT; n++)
759 memcpy(filter->addr[n], addr[n].u, ETH_ALEN);
760
761 nexact = n;
762
763 /* Remaining multicast addresses are hashed,
764 * unicast will leave the filter disabled. */
765 memset(filter->mask, 0, sizeof(filter->mask));
766 for (; n < uf.count; n++) {
767 if (!is_multicast_ether_addr(addr[n].u)) {
768 err = 0; /* no filter */
769 goto free_addr;
770 }
771 addr_hash_set(filter->mask, addr[n].u);
772 }
773
774 /* For ALLMULTI just set the mask to all ones.
775 * This overrides the mask populated above. */
776 if ((uf.flags & TUN_FLT_ALLMULTI))
777 memset(filter->mask, ~0, sizeof(filter->mask));
778
779 /* Now enable the filter */
780 wmb();
781 filter->count = nexact;
782
783 /* Return the number of exact filters */
784 err = nexact;
785 free_addr:
786 kfree(addr);
787 return err;
788 }
789
790 /* Returns: 0 - drop, !=0 - accept */
791 static int run_filter(struct tap_filter *filter, const struct sk_buff *skb)
792 {
793 /* Cannot use eth_hdr(skb) here because skb_mac_hdr() is incorrect
794 * at this point. */
795 struct ethhdr *eh = (struct ethhdr *) skb->data;
796 int i;
797
798 /* Exact match */
799 for (i = 0; i < filter->count; i++)
800 if (ether_addr_equal(eh->h_dest, filter->addr[i]))
801 return 1;
802
803 /* Inexact match (multicast only) */
804 if (is_multicast_ether_addr(eh->h_dest))
805 return addr_hash_test(filter->mask, eh->h_dest);
806
807 return 0;
808 }
809
810 /*
811 * Checks whether the packet is accepted or not.
812 * Returns: 0 - drop, !=0 - accept
813 */
814 static int check_filter(struct tap_filter *filter, const struct sk_buff *skb)
815 {
816 if (!filter->count)
817 return 1;
818
819 return run_filter(filter, skb);
820 }
821
822 /* Network device part of the driver */
823
824 static const struct ethtool_ops tun_ethtool_ops;
825
826 /* Net device detach from fd. */
827 static void tun_net_uninit(struct net_device *dev)
828 {
829 tun_detach_all(dev);
830 }
831
832 /* Net device open. */
833 static int tun_net_open(struct net_device *dev)
834 {
835 struct tun_struct *tun = netdev_priv(dev);
836 int i;
837
838 netif_tx_start_all_queues(dev);
839
840 for (i = 0; i < tun->numqueues; i++) {
841 struct tun_file *tfile;
842
843 tfile = rtnl_dereference(tun->tfiles[i]);
844 tfile->socket.sk->sk_write_space(tfile->socket.sk);
845 }
846
847 return 0;
848 }
849
850 /* Net device close. */
851 static int tun_net_close(struct net_device *dev)
852 {
853 netif_tx_stop_all_queues(dev);
854 return 0;
855 }
856
857 /* Net device start xmit */
858 static netdev_tx_t tun_net_xmit(struct sk_buff *skb, struct net_device *dev)
859 {
860 struct tun_struct *tun = netdev_priv(dev);
861 int txq = skb->queue_mapping;
862 struct tun_file *tfile;
863 u32 numqueues = 0;
864
865 rcu_read_lock();
866 tfile = rcu_dereference(tun->tfiles[txq]);
867 numqueues = ACCESS_ONCE(tun->numqueues);
868
869 /* Drop packet if interface is not attached */
870 if (txq >= numqueues)
871 goto drop;
872
873 #ifdef CONFIG_RPS
874 if (numqueues == 1 && static_key_false(&rps_needed)) {
875 /* Select queue was not called for the skbuff, so we extract the
876 * RPS hash and save it into the flow_table here.
877 */
878 __u32 rxhash;
879
880 rxhash = __skb_get_hash_symmetric(skb);
881 if (rxhash) {
882 struct tun_flow_entry *e;
883 e = tun_flow_find(&tun->flows[tun_hashfn(rxhash)],
884 rxhash);
885 if (e)
886 tun_flow_save_rps_rxhash(e, rxhash);
887 }
888 }
889 #endif
890
891 tun_debug(KERN_INFO, tun, "tun_net_xmit %d\n", skb->len);
892
893 BUG_ON(!tfile);
894
895 /* Drop if the filter does not like it.
896 * This is a noop if the filter is disabled.
897 * Filter can be enabled only for the TAP devices. */
898 if (!check_filter(&tun->txflt, skb))
899 goto drop;
900
901 if (tfile->socket.sk->sk_filter &&
902 sk_filter(tfile->socket.sk, skb))
903 goto drop;
904
905 if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC)))
906 goto drop;
907
908 skb_tx_timestamp(skb);
909
910 /* Orphan the skb - required as we might hang on to it
911 * for indefinite time.
912 */
913 skb_orphan(skb);
914
915 nf_reset(skb);
916
917 if (skb_array_produce(&tfile->tx_array, skb))
918 goto drop;
919
920 /* Notify and wake up reader process */
921 if (tfile->flags & TUN_FASYNC)
922 kill_fasync(&tfile->fasync, SIGIO, POLL_IN);
923 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
924
925 rcu_read_unlock();
926 return NETDEV_TX_OK;
927
928 drop:
929 this_cpu_inc(tun->pcpu_stats->tx_dropped);
930 skb_tx_error(skb);
931 kfree_skb(skb);
932 rcu_read_unlock();
933 return NET_XMIT_DROP;
934 }
935
936 static void tun_net_mclist(struct net_device *dev)
937 {
938 /*
939 * This callback is supposed to deal with mc filter in
940 * _rx_ path and has nothing to do with the _tx_ path.
941 * In rx path we always accept everything userspace gives us.
942 */
943 }
944
945 static netdev_features_t tun_net_fix_features(struct net_device *dev,
946 netdev_features_t features)
947 {
948 struct tun_struct *tun = netdev_priv(dev);
949
950 return (features & tun->set_features) | (features & ~TUN_USER_FEATURES);
951 }
952 #ifdef CONFIG_NET_POLL_CONTROLLER
953 static void tun_poll_controller(struct net_device *dev)
954 {
955 /*
956 * Tun only receives frames when:
957 * 1) the char device endpoint gets data from user space
958 * 2) the tun socket gets a sendmsg call from user space
959 * Since both of those are synchronous operations, we are guaranteed
960 * never to have pending data when we poll for it
961 * so there is nothing to do here but return.
962 * We need this though so netpoll recognizes us as an interface that
963 * supports polling, which enables bridge devices in virt setups to
964 * still use netconsole
965 */
966 return;
967 }
968 #endif
969
970 static void tun_set_headroom(struct net_device *dev, int new_hr)
971 {
972 struct tun_struct *tun = netdev_priv(dev);
973
974 if (new_hr < NET_SKB_PAD)
975 new_hr = NET_SKB_PAD;
976
977 tun->align = new_hr;
978 }
979
980 static void
981 tun_net_get_stats64(struct net_device *dev, struct rtnl_link_stats64 *stats)
982 {
983 u32 rx_dropped = 0, tx_dropped = 0, rx_frame_errors = 0;
984 struct tun_struct *tun = netdev_priv(dev);
985 struct tun_pcpu_stats *p;
986 int i;
987
988 for_each_possible_cpu(i) {
989 u64 rxpackets, rxbytes, txpackets, txbytes;
990 unsigned int start;
991
992 p = per_cpu_ptr(tun->pcpu_stats, i);
993 do {
994 start = u64_stats_fetch_begin(&p->syncp);
995 rxpackets = p->rx_packets;
996 rxbytes = p->rx_bytes;
997 txpackets = p->tx_packets;
998 txbytes = p->tx_bytes;
999 } while (u64_stats_fetch_retry(&p->syncp, start));
1000
1001 stats->rx_packets += rxpackets;
1002 stats->rx_bytes += rxbytes;
1003 stats->tx_packets += txpackets;
1004 stats->tx_bytes += txbytes;
1005
1006 /* u32 counters */
1007 rx_dropped += p->rx_dropped;
1008 rx_frame_errors += p->rx_frame_errors;
1009 tx_dropped += p->tx_dropped;
1010 }
1011 stats->rx_dropped = rx_dropped;
1012 stats->rx_frame_errors = rx_frame_errors;
1013 stats->tx_dropped = tx_dropped;
1014 }
1015
1016 static int tun_xdp_set(struct net_device *dev, struct bpf_prog *prog,
1017 struct netlink_ext_ack *extack)
1018 {
1019 struct tun_struct *tun = netdev_priv(dev);
1020 struct bpf_prog *old_prog;
1021
1022 old_prog = rtnl_dereference(tun->xdp_prog);
1023 rcu_assign_pointer(tun->xdp_prog, prog);
1024 if (old_prog)
1025 bpf_prog_put(old_prog);
1026
1027 return 0;
1028 }
1029
1030 static u32 tun_xdp_query(struct net_device *dev)
1031 {
1032 struct tun_struct *tun = netdev_priv(dev);
1033 const struct bpf_prog *xdp_prog;
1034
1035 xdp_prog = rtnl_dereference(tun->xdp_prog);
1036 if (xdp_prog)
1037 return xdp_prog->aux->id;
1038
1039 return 0;
1040 }
1041
1042 static int tun_xdp(struct net_device *dev, struct netdev_xdp *xdp)
1043 {
1044 switch (xdp->command) {
1045 case XDP_SETUP_PROG:
1046 return tun_xdp_set(dev, xdp->prog, xdp->extack);
1047 case XDP_QUERY_PROG:
1048 xdp->prog_id = tun_xdp_query(dev);
1049 xdp->prog_attached = !!xdp->prog_id;
1050 return 0;
1051 default:
1052 return -EINVAL;
1053 }
1054 }
1055
1056 static const struct net_device_ops tun_netdev_ops = {
1057 .ndo_uninit = tun_net_uninit,
1058 .ndo_open = tun_net_open,
1059 .ndo_stop = tun_net_close,
1060 .ndo_start_xmit = tun_net_xmit,
1061 .ndo_fix_features = tun_net_fix_features,
1062 .ndo_select_queue = tun_select_queue,
1063 #ifdef CONFIG_NET_POLL_CONTROLLER
1064 .ndo_poll_controller = tun_poll_controller,
1065 #endif
1066 .ndo_set_rx_headroom = tun_set_headroom,
1067 .ndo_get_stats64 = tun_net_get_stats64,
1068 };
1069
1070 static const struct net_device_ops tap_netdev_ops = {
1071 .ndo_uninit = tun_net_uninit,
1072 .ndo_open = tun_net_open,
1073 .ndo_stop = tun_net_close,
1074 .ndo_start_xmit = tun_net_xmit,
1075 .ndo_fix_features = tun_net_fix_features,
1076 .ndo_set_rx_mode = tun_net_mclist,
1077 .ndo_set_mac_address = eth_mac_addr,
1078 .ndo_validate_addr = eth_validate_addr,
1079 .ndo_select_queue = tun_select_queue,
1080 #ifdef CONFIG_NET_POLL_CONTROLLER
1081 .ndo_poll_controller = tun_poll_controller,
1082 #endif
1083 .ndo_features_check = passthru_features_check,
1084 .ndo_set_rx_headroom = tun_set_headroom,
1085 .ndo_get_stats64 = tun_net_get_stats64,
1086 .ndo_xdp = tun_xdp,
1087 };
1088
1089 static void tun_flow_init(struct tun_struct *tun)
1090 {
1091 int i;
1092
1093 for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++)
1094 INIT_HLIST_HEAD(&tun->flows[i]);
1095
1096 tun->ageing_time = TUN_FLOW_EXPIRE;
1097 setup_timer(&tun->flow_gc_timer, tun_flow_cleanup, (unsigned long)tun);
1098 mod_timer(&tun->flow_gc_timer,
1099 round_jiffies_up(jiffies + tun->ageing_time));
1100 }
1101
1102 static void tun_flow_uninit(struct tun_struct *tun)
1103 {
1104 del_timer_sync(&tun->flow_gc_timer);
1105 tun_flow_flush(tun);
1106 }
1107
1108 #define MIN_MTU 68
1109 #define MAX_MTU 65535
1110
1111 /* Initialize net device. */
1112 static void tun_net_init(struct net_device *dev)
1113 {
1114 struct tun_struct *tun = netdev_priv(dev);
1115
1116 switch (tun->flags & TUN_TYPE_MASK) {
1117 case IFF_TUN:
1118 dev->netdev_ops = &tun_netdev_ops;
1119
1120 /* Point-to-Point TUN Device */
1121 dev->hard_header_len = 0;
1122 dev->addr_len = 0;
1123 dev->mtu = 1500;
1124
1125 /* Zero header length */
1126 dev->type = ARPHRD_NONE;
1127 dev->flags = IFF_POINTOPOINT | IFF_NOARP | IFF_MULTICAST;
1128 break;
1129
1130 case IFF_TAP:
1131 dev->netdev_ops = &tap_netdev_ops;
1132 /* Ethernet TAP Device */
1133 ether_setup(dev);
1134 dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1135 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
1136
1137 eth_hw_addr_random(dev);
1138
1139 break;
1140 }
1141
1142 dev->min_mtu = MIN_MTU;
1143 dev->max_mtu = MAX_MTU - dev->hard_header_len;
1144 }
1145
1146 /* Character device part */
1147
1148 /* Poll */
1149 static unsigned int tun_chr_poll(struct file *file, poll_table *wait)
1150 {
1151 struct tun_file *tfile = file->private_data;
1152 struct tun_struct *tun = __tun_get(tfile);
1153 struct sock *sk;
1154 unsigned int mask = 0;
1155
1156 if (!tun)
1157 return POLLERR;
1158
1159 sk = tfile->socket.sk;
1160
1161 tun_debug(KERN_INFO, tun, "tun_chr_poll\n");
1162
1163 poll_wait(file, sk_sleep(sk), wait);
1164
1165 if (!skb_array_empty(&tfile->tx_array))
1166 mask |= POLLIN | POLLRDNORM;
1167
1168 if (tun->dev->flags & IFF_UP &&
1169 (sock_writeable(sk) ||
1170 (!test_and_set_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags) &&
1171 sock_writeable(sk))))
1172 mask |= POLLOUT | POLLWRNORM;
1173
1174 if (tun->dev->reg_state != NETREG_REGISTERED)
1175 mask = POLLERR;
1176
1177 tun_put(tun);
1178 return mask;
1179 }
1180
1181 /* prepad is the amount to reserve at front. len is length after that.
1182 * linear is a hint as to how much to copy (usually headers). */
1183 static struct sk_buff *tun_alloc_skb(struct tun_file *tfile,
1184 size_t prepad, size_t len,
1185 size_t linear, int noblock)
1186 {
1187 struct sock *sk = tfile->socket.sk;
1188 struct sk_buff *skb;
1189 int err;
1190
1191 /* Under a page? Don't bother with paged skb. */
1192 if (prepad + len < PAGE_SIZE || !linear)
1193 linear = len;
1194
1195 skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
1196 &err, 0);
1197 if (!skb)
1198 return ERR_PTR(err);
1199
1200 skb_reserve(skb, prepad);
1201 skb_put(skb, linear);
1202 skb->data_len = len - linear;
1203 skb->len += len - linear;
1204
1205 return skb;
1206 }
1207
1208 static void tun_rx_batched(struct tun_struct *tun, struct tun_file *tfile,
1209 struct sk_buff *skb, int more)
1210 {
1211 struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
1212 struct sk_buff_head process_queue;
1213 u32 rx_batched = tun->rx_batched;
1214 bool rcv = false;
1215
1216 if (!rx_batched || (!more && skb_queue_empty(queue))) {
1217 local_bh_disable();
1218 netif_receive_skb(skb);
1219 local_bh_enable();
1220 return;
1221 }
1222
1223 spin_lock(&queue->lock);
1224 if (!more || skb_queue_len(queue) == rx_batched) {
1225 __skb_queue_head_init(&process_queue);
1226 skb_queue_splice_tail_init(queue, &process_queue);
1227 rcv = true;
1228 } else {
1229 __skb_queue_tail(queue, skb);
1230 }
1231 spin_unlock(&queue->lock);
1232
1233 if (rcv) {
1234 struct sk_buff *nskb;
1235
1236 local_bh_disable();
1237 while ((nskb = __skb_dequeue(&process_queue)))
1238 netif_receive_skb(nskb);
1239 netif_receive_skb(skb);
1240 local_bh_enable();
1241 }
1242 }
1243
1244 static bool tun_can_build_skb(struct tun_struct *tun, struct tun_file *tfile,
1245 int len, int noblock, bool zerocopy)
1246 {
1247 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
1248 return false;
1249
1250 if (tfile->socket.sk->sk_sndbuf != INT_MAX)
1251 return false;
1252
1253 if (!noblock)
1254 return false;
1255
1256 if (zerocopy)
1257 return false;
1258
1259 if (SKB_DATA_ALIGN(len + TUN_RX_PAD) +
1260 SKB_DATA_ALIGN(sizeof(struct skb_shared_info)) > PAGE_SIZE)
1261 return false;
1262
1263 return true;
1264 }
1265
1266 static struct sk_buff *tun_build_skb(struct tun_struct *tun,
1267 struct tun_file *tfile,
1268 struct iov_iter *from,
1269 struct virtio_net_hdr *hdr,
1270 int len, int *skb_xdp)
1271 {
1272 struct page_frag *alloc_frag = &current->task_frag;
1273 struct sk_buff *skb;
1274 struct bpf_prog *xdp_prog;
1275 int buflen = SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
1276 unsigned int delta = 0;
1277 char *buf;
1278 size_t copied;
1279 bool xdp_xmit = false;
1280 int err, pad = TUN_RX_PAD;
1281
1282 rcu_read_lock();
1283 xdp_prog = rcu_dereference(tun->xdp_prog);
1284 if (xdp_prog)
1285 pad += TUN_HEADROOM;
1286 buflen += SKB_DATA_ALIGN(len + pad);
1287 rcu_read_unlock();
1288
1289 if (unlikely(!skb_page_frag_refill(buflen, alloc_frag, GFP_KERNEL)))
1290 return ERR_PTR(-ENOMEM);
1291
1292 buf = (char *)page_address(alloc_frag->page) + alloc_frag->offset;
1293 copied = copy_page_from_iter(alloc_frag->page,
1294 alloc_frag->offset + pad,
1295 len, from);
1296 if (copied != len)
1297 return ERR_PTR(-EFAULT);
1298
1299 /* There's a small window that XDP may be set after the check
1300 * of xdp_prog above, this should be rare and for simplicity
1301 * we do XDP on skb in case the headroom is not enough.
1302 */
1303 if (hdr->gso_type || !xdp_prog)
1304 *skb_xdp = 1;
1305 else
1306 *skb_xdp = 0;
1307
1308 rcu_read_lock();
1309 xdp_prog = rcu_dereference(tun->xdp_prog);
1310 if (xdp_prog && !*skb_xdp) {
1311 struct xdp_buff xdp;
1312 void *orig_data;
1313 u32 act;
1314
1315 xdp.data_hard_start = buf;
1316 xdp.data = buf + pad;
1317 xdp.data_end = xdp.data + len;
1318 orig_data = xdp.data;
1319 act = bpf_prog_run_xdp(xdp_prog, &xdp);
1320
1321 switch (act) {
1322 case XDP_REDIRECT:
1323 get_page(alloc_frag->page);
1324 alloc_frag->offset += buflen;
1325 err = xdp_do_redirect(tun->dev, &xdp, xdp_prog);
1326 if (err)
1327 goto err_redirect;
1328 return NULL;
1329 case XDP_TX:
1330 xdp_xmit = true;
1331 /* fall through */
1332 case XDP_PASS:
1333 delta = orig_data - xdp.data;
1334 break;
1335 default:
1336 bpf_warn_invalid_xdp_action(act);
1337 /* fall through */
1338 case XDP_ABORTED:
1339 trace_xdp_exception(tun->dev, xdp_prog, act);
1340 /* fall through */
1341 case XDP_DROP:
1342 goto err_xdp;
1343 }
1344 }
1345
1346 skb = build_skb(buf, buflen);
1347 if (!skb) {
1348 rcu_read_unlock();
1349 return ERR_PTR(-ENOMEM);
1350 }
1351
1352 skb_reserve(skb, pad - delta);
1353 skb_put(skb, len + delta);
1354 get_page(alloc_frag->page);
1355 alloc_frag->offset += buflen;
1356
1357 if (xdp_xmit) {
1358 skb->dev = tun->dev;
1359 generic_xdp_tx(skb, xdp_prog);
1360 rcu_read_lock();
1361 return NULL;
1362 }
1363
1364 rcu_read_unlock();
1365
1366 return skb;
1367
1368 err_redirect:
1369 put_page(alloc_frag->page);
1370 err_xdp:
1371 rcu_read_unlock();
1372 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1373 return NULL;
1374 }
1375
1376 /* Get packet from user space buffer */
1377 static ssize_t tun_get_user(struct tun_struct *tun, struct tun_file *tfile,
1378 void *msg_control, struct iov_iter *from,
1379 int noblock, bool more)
1380 {
1381 struct tun_pi pi = { 0, cpu_to_be16(ETH_P_IP) };
1382 struct sk_buff *skb;
1383 size_t total_len = iov_iter_count(from);
1384 size_t len = total_len, align = tun->align, linear;
1385 struct virtio_net_hdr gso = { 0 };
1386 struct tun_pcpu_stats *stats;
1387 int good_linear;
1388 int copylen;
1389 bool zerocopy = false;
1390 int err;
1391 u32 rxhash;
1392 int skb_xdp = 1;
1393
1394 if (!(tun->dev->flags & IFF_UP))
1395 return -EIO;
1396
1397 if (!(tun->flags & IFF_NO_PI)) {
1398 if (len < sizeof(pi))
1399 return -EINVAL;
1400 len -= sizeof(pi);
1401
1402 if (!copy_from_iter_full(&pi, sizeof(pi), from))
1403 return -EFAULT;
1404 }
1405
1406 if (tun->flags & IFF_VNET_HDR) {
1407 int vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
1408
1409 if (len < vnet_hdr_sz)
1410 return -EINVAL;
1411 len -= vnet_hdr_sz;
1412
1413 if (!copy_from_iter_full(&gso, sizeof(gso), from))
1414 return -EFAULT;
1415
1416 if ((gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
1417 tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2 > tun16_to_cpu(tun, gso.hdr_len))
1418 gso.hdr_len = cpu_to_tun16(tun, tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2);
1419
1420 if (tun16_to_cpu(tun, gso.hdr_len) > len)
1421 return -EINVAL;
1422 iov_iter_advance(from, vnet_hdr_sz - sizeof(gso));
1423 }
1424
1425 if ((tun->flags & TUN_TYPE_MASK) == IFF_TAP) {
1426 align += NET_IP_ALIGN;
1427 if (unlikely(len < ETH_HLEN ||
1428 (gso.hdr_len && tun16_to_cpu(tun, gso.hdr_len) < ETH_HLEN)))
1429 return -EINVAL;
1430 }
1431
1432 good_linear = SKB_MAX_HEAD(align);
1433
1434 if (msg_control) {
1435 struct iov_iter i = *from;
1436
1437 /* There are 256 bytes to be copied in skb, so there is
1438 * enough room for skb expand head in case it is used.
1439 * The rest of the buffer is mapped from userspace.
1440 */
1441 copylen = gso.hdr_len ? tun16_to_cpu(tun, gso.hdr_len) : GOODCOPY_LEN;
1442 if (copylen > good_linear)
1443 copylen = good_linear;
1444 linear = copylen;
1445 iov_iter_advance(&i, copylen);
1446 if (iov_iter_npages(&i, INT_MAX) <= MAX_SKB_FRAGS)
1447 zerocopy = true;
1448 }
1449
1450 if (tun_can_build_skb(tun, tfile, len, noblock, zerocopy)) {
1451 /* For the packet that is not easy to be processed
1452 * (e.g gso or jumbo packet), we will do it at after
1453 * skb was created with generic XDP routine.
1454 */
1455 skb = tun_build_skb(tun, tfile, from, &gso, len, &skb_xdp);
1456 if (IS_ERR(skb)) {
1457 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1458 return PTR_ERR(skb);
1459 }
1460 if (!skb)
1461 return total_len;
1462 } else {
1463 if (!zerocopy) {
1464 copylen = len;
1465 if (tun16_to_cpu(tun, gso.hdr_len) > good_linear)
1466 linear = good_linear;
1467 else
1468 linear = tun16_to_cpu(tun, gso.hdr_len);
1469 }
1470
1471 skb = tun_alloc_skb(tfile, align, copylen, linear, noblock);
1472 if (IS_ERR(skb)) {
1473 if (PTR_ERR(skb) != -EAGAIN)
1474 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1475 return PTR_ERR(skb);
1476 }
1477
1478 if (zerocopy)
1479 err = zerocopy_sg_from_iter(skb, from);
1480 else
1481 err = skb_copy_datagram_from_iter(skb, 0, from, len);
1482
1483 if (err) {
1484 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1485 kfree_skb(skb);
1486 return -EFAULT;
1487 }
1488 }
1489
1490 if (virtio_net_hdr_to_skb(skb, &gso, tun_is_little_endian(tun))) {
1491 this_cpu_inc(tun->pcpu_stats->rx_frame_errors);
1492 kfree_skb(skb);
1493 return -EINVAL;
1494 }
1495
1496 switch (tun->flags & TUN_TYPE_MASK) {
1497 case IFF_TUN:
1498 if (tun->flags & IFF_NO_PI) {
1499 switch (skb->data[0] & 0xf0) {
1500 case 0x40:
1501 pi.proto = htons(ETH_P_IP);
1502 break;
1503 case 0x60:
1504 pi.proto = htons(ETH_P_IPV6);
1505 break;
1506 default:
1507 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1508 kfree_skb(skb);
1509 return -EINVAL;
1510 }
1511 }
1512
1513 skb_reset_mac_header(skb);
1514 skb->protocol = pi.proto;
1515 skb->dev = tun->dev;
1516 break;
1517 case IFF_TAP:
1518 skb->protocol = eth_type_trans(skb, tun->dev);
1519 break;
1520 }
1521
1522 /* copy skb_ubuf_info for callback when skb has no error */
1523 if (zerocopy) {
1524 skb_shinfo(skb)->destructor_arg = msg_control;
1525 skb_shinfo(skb)->tx_flags |= SKBTX_DEV_ZEROCOPY;
1526 skb_shinfo(skb)->tx_flags |= SKBTX_SHARED_FRAG;
1527 } else if (msg_control) {
1528 struct ubuf_info *uarg = msg_control;
1529 uarg->callback(uarg, false);
1530 }
1531
1532 skb_reset_network_header(skb);
1533 skb_probe_transport_header(skb, 0);
1534
1535 if (skb_xdp) {
1536 struct bpf_prog *xdp_prog;
1537 int ret;
1538
1539 rcu_read_lock();
1540 xdp_prog = rcu_dereference(tun->xdp_prog);
1541 if (xdp_prog) {
1542 ret = do_xdp_generic(xdp_prog, skb);
1543 if (ret != XDP_PASS) {
1544 rcu_read_unlock();
1545 return total_len;
1546 }
1547 }
1548 rcu_read_unlock();
1549 }
1550
1551 rxhash = __skb_get_hash_symmetric(skb);
1552 #ifndef CONFIG_4KSTACKS
1553 tun_rx_batched(tun, tfile, skb, more);
1554 #else
1555 netif_rx_ni(skb);
1556 #endif
1557
1558 stats = get_cpu_ptr(tun->pcpu_stats);
1559 u64_stats_update_begin(&stats->syncp);
1560 stats->rx_packets++;
1561 stats->rx_bytes += len;
1562 u64_stats_update_end(&stats->syncp);
1563 put_cpu_ptr(stats);
1564
1565 tun_flow_update(tun, rxhash, tfile);
1566 return total_len;
1567 }
1568
1569 static ssize_t tun_chr_write_iter(struct kiocb *iocb, struct iov_iter *from)
1570 {
1571 struct file *file = iocb->ki_filp;
1572 struct tun_struct *tun = tun_get(file);
1573 struct tun_file *tfile = file->private_data;
1574 ssize_t result;
1575
1576 if (!tun)
1577 return -EBADFD;
1578
1579 result = tun_get_user(tun, tfile, NULL, from,
1580 file->f_flags & O_NONBLOCK, false);
1581
1582 tun_put(tun);
1583 return result;
1584 }
1585
1586 /* Put packet to the user space buffer */
1587 static ssize_t tun_put_user(struct tun_struct *tun,
1588 struct tun_file *tfile,
1589 struct sk_buff *skb,
1590 struct iov_iter *iter)
1591 {
1592 struct tun_pi pi = { 0, skb->protocol };
1593 struct tun_pcpu_stats *stats;
1594 ssize_t total;
1595 int vlan_offset = 0;
1596 int vlan_hlen = 0;
1597 int vnet_hdr_sz = 0;
1598
1599 if (skb_vlan_tag_present(skb))
1600 vlan_hlen = VLAN_HLEN;
1601
1602 if (tun->flags & IFF_VNET_HDR)
1603 vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
1604
1605 total = skb->len + vlan_hlen + vnet_hdr_sz;
1606
1607 if (!(tun->flags & IFF_NO_PI)) {
1608 if (iov_iter_count(iter) < sizeof(pi))
1609 return -EINVAL;
1610
1611 total += sizeof(pi);
1612 if (iov_iter_count(iter) < total) {
1613 /* Packet will be striped */
1614 pi.flags |= TUN_PKT_STRIP;
1615 }
1616
1617 if (copy_to_iter(&pi, sizeof(pi), iter) != sizeof(pi))
1618 return -EFAULT;
1619 }
1620
1621 if (vnet_hdr_sz) {
1622 struct virtio_net_hdr gso;
1623
1624 if (iov_iter_count(iter) < vnet_hdr_sz)
1625 return -EINVAL;
1626
1627 if (virtio_net_hdr_from_skb(skb, &gso,
1628 tun_is_little_endian(tun), true)) {
1629 struct skb_shared_info *sinfo = skb_shinfo(skb);
1630 pr_err("unexpected GSO type: "
1631 "0x%x, gso_size %d, hdr_len %d\n",
1632 sinfo->gso_type, tun16_to_cpu(tun, gso.gso_size),
1633 tun16_to_cpu(tun, gso.hdr_len));
1634 print_hex_dump(KERN_ERR, "tun: ",
1635 DUMP_PREFIX_NONE,
1636 16, 1, skb->head,
1637 min((int)tun16_to_cpu(tun, gso.hdr_len), 64), true);
1638 WARN_ON_ONCE(1);
1639 return -EINVAL;
1640 }
1641
1642 if (copy_to_iter(&gso, sizeof(gso), iter) != sizeof(gso))
1643 return -EFAULT;
1644
1645 iov_iter_advance(iter, vnet_hdr_sz - sizeof(gso));
1646 }
1647
1648 if (vlan_hlen) {
1649 int ret;
1650 struct {
1651 __be16 h_vlan_proto;
1652 __be16 h_vlan_TCI;
1653 } veth;
1654
1655 veth.h_vlan_proto = skb->vlan_proto;
1656 veth.h_vlan_TCI = htons(skb_vlan_tag_get(skb));
1657
1658 vlan_offset = offsetof(struct vlan_ethhdr, h_vlan_proto);
1659
1660 ret = skb_copy_datagram_iter(skb, 0, iter, vlan_offset);
1661 if (ret || !iov_iter_count(iter))
1662 goto done;
1663
1664 ret = copy_to_iter(&veth, sizeof(veth), iter);
1665 if (ret != sizeof(veth) || !iov_iter_count(iter))
1666 goto done;
1667 }
1668
1669 skb_copy_datagram_iter(skb, vlan_offset, iter, skb->len - vlan_offset);
1670
1671 done:
1672 /* caller is in process context, */
1673 stats = get_cpu_ptr(tun->pcpu_stats);
1674 u64_stats_update_begin(&stats->syncp);
1675 stats->tx_packets++;
1676 stats->tx_bytes += skb->len + vlan_hlen;
1677 u64_stats_update_end(&stats->syncp);
1678 put_cpu_ptr(tun->pcpu_stats);
1679
1680 return total;
1681 }
1682
1683 static struct sk_buff *tun_ring_recv(struct tun_file *tfile, int noblock,
1684 int *err)
1685 {
1686 DECLARE_WAITQUEUE(wait, current);
1687 struct sk_buff *skb = NULL;
1688 int error = 0;
1689
1690 skb = skb_array_consume(&tfile->tx_array);
1691 if (skb)
1692 goto out;
1693 if (noblock) {
1694 error = -EAGAIN;
1695 goto out;
1696 }
1697
1698 add_wait_queue(&tfile->wq.wait, &wait);
1699 current->state = TASK_INTERRUPTIBLE;
1700
1701 while (1) {
1702 skb = skb_array_consume(&tfile->tx_array);
1703 if (skb)
1704 break;
1705 if (signal_pending(current)) {
1706 error = -ERESTARTSYS;
1707 break;
1708 }
1709 if (tfile->socket.sk->sk_shutdown & RCV_SHUTDOWN) {
1710 error = -EFAULT;
1711 break;
1712 }
1713
1714 schedule();
1715 }
1716
1717 current->state = TASK_RUNNING;
1718 remove_wait_queue(&tfile->wq.wait, &wait);
1719
1720 out:
1721 *err = error;
1722 return skb;
1723 }
1724
1725 static ssize_t tun_do_read(struct tun_struct *tun, struct tun_file *tfile,
1726 struct iov_iter *to,
1727 int noblock, struct sk_buff *skb)
1728 {
1729 ssize_t ret;
1730 int err;
1731
1732 tun_debug(KERN_INFO, tun, "tun_do_read\n");
1733
1734 if (!iov_iter_count(to))
1735 return 0;
1736
1737 if (!skb) {
1738 /* Read frames from ring */
1739 skb = tun_ring_recv(tfile, noblock, &err);
1740 if (!skb)
1741 return err;
1742 }
1743
1744 ret = tun_put_user(tun, tfile, skb, to);
1745 if (unlikely(ret < 0))
1746 kfree_skb(skb);
1747 else
1748 consume_skb(skb);
1749
1750 return ret;
1751 }
1752
1753 static ssize_t tun_chr_read_iter(struct kiocb *iocb, struct iov_iter *to)
1754 {
1755 struct file *file = iocb->ki_filp;
1756 struct tun_file *tfile = file->private_data;
1757 struct tun_struct *tun = __tun_get(tfile);
1758 ssize_t len = iov_iter_count(to), ret;
1759
1760 if (!tun)
1761 return -EBADFD;
1762 ret = tun_do_read(tun, tfile, to, file->f_flags & O_NONBLOCK, NULL);
1763 ret = min_t(ssize_t, ret, len);
1764 if (ret > 0)
1765 iocb->ki_pos = ret;
1766 tun_put(tun);
1767 return ret;
1768 }
1769
1770 static void tun_free_netdev(struct net_device *dev)
1771 {
1772 struct tun_struct *tun = netdev_priv(dev);
1773
1774 BUG_ON(!(list_empty(&tun->disabled)));
1775 free_percpu(tun->pcpu_stats);
1776 tun_flow_uninit(tun);
1777 security_tun_dev_free_security(tun->security);
1778 }
1779
1780 static void tun_setup(struct net_device *dev)
1781 {
1782 struct tun_struct *tun = netdev_priv(dev);
1783
1784 tun->owner = INVALID_UID;
1785 tun->group = INVALID_GID;
1786
1787 dev->ethtool_ops = &tun_ethtool_ops;
1788 dev->needs_free_netdev = true;
1789 dev->priv_destructor = tun_free_netdev;
1790 /* We prefer our own queue length */
1791 dev->tx_queue_len = TUN_READQ_SIZE;
1792 }
1793
1794 /* Trivial set of netlink ops to allow deleting tun or tap
1795 * device with netlink.
1796 */
1797 static int tun_validate(struct nlattr *tb[], struct nlattr *data[],
1798 struct netlink_ext_ack *extack)
1799 {
1800 return -EINVAL;
1801 }
1802
1803 static struct rtnl_link_ops tun_link_ops __read_mostly = {
1804 .kind = DRV_NAME,
1805 .priv_size = sizeof(struct tun_struct),
1806 .setup = tun_setup,
1807 .validate = tun_validate,
1808 };
1809
1810 static void tun_sock_write_space(struct sock *sk)
1811 {
1812 struct tun_file *tfile;
1813 wait_queue_head_t *wqueue;
1814
1815 if (!sock_writeable(sk))
1816 return;
1817
1818 if (!test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags))
1819 return;
1820
1821 wqueue = sk_sleep(sk);
1822 if (wqueue && waitqueue_active(wqueue))
1823 wake_up_interruptible_sync_poll(wqueue, POLLOUT |
1824 POLLWRNORM | POLLWRBAND);
1825
1826 tfile = container_of(sk, struct tun_file, sk);
1827 kill_fasync(&tfile->fasync, SIGIO, POLL_OUT);
1828 }
1829
1830 static int tun_sendmsg(struct socket *sock, struct msghdr *m, size_t total_len)
1831 {
1832 int ret;
1833 struct tun_file *tfile = container_of(sock, struct tun_file, socket);
1834 struct tun_struct *tun = __tun_get(tfile);
1835
1836 if (!tun)
1837 return -EBADFD;
1838
1839 ret = tun_get_user(tun, tfile, m->msg_control, &m->msg_iter,
1840 m->msg_flags & MSG_DONTWAIT,
1841 m->msg_flags & MSG_MORE);
1842 tun_put(tun);
1843 return ret;
1844 }
1845
1846 static int tun_recvmsg(struct socket *sock, struct msghdr *m, size_t total_len,
1847 int flags)
1848 {
1849 struct tun_file *tfile = container_of(sock, struct tun_file, socket);
1850 struct tun_struct *tun = __tun_get(tfile);
1851 int ret;
1852
1853 if (!tun)
1854 return -EBADFD;
1855
1856 if (flags & ~(MSG_DONTWAIT|MSG_TRUNC|MSG_ERRQUEUE)) {
1857 ret = -EINVAL;
1858 goto out;
1859 }
1860 if (flags & MSG_ERRQUEUE) {
1861 ret = sock_recv_errqueue(sock->sk, m, total_len,
1862 SOL_PACKET, TUN_TX_TIMESTAMP);
1863 goto out;
1864 }
1865 ret = tun_do_read(tun, tfile, &m->msg_iter, flags & MSG_DONTWAIT,
1866 m->msg_control);
1867 if (ret > (ssize_t)total_len) {
1868 m->msg_flags |= MSG_TRUNC;
1869 ret = flags & MSG_TRUNC ? ret : total_len;
1870 }
1871 out:
1872 tun_put(tun);
1873 return ret;
1874 }
1875
1876 static int tun_peek_len(struct socket *sock)
1877 {
1878 struct tun_file *tfile = container_of(sock, struct tun_file, socket);
1879 struct tun_struct *tun;
1880 int ret = 0;
1881
1882 tun = __tun_get(tfile);
1883 if (!tun)
1884 return 0;
1885
1886 ret = skb_array_peek_len(&tfile->tx_array);
1887 tun_put(tun);
1888
1889 return ret;
1890 }
1891
1892 /* Ops structure to mimic raw sockets with tun */
1893 static const struct proto_ops tun_socket_ops = {
1894 .peek_len = tun_peek_len,
1895 .sendmsg = tun_sendmsg,
1896 .recvmsg = tun_recvmsg,
1897 };
1898
1899 static struct proto tun_proto = {
1900 .name = "tun",
1901 .owner = THIS_MODULE,
1902 .obj_size = sizeof(struct tun_file),
1903 };
1904
1905 static int tun_flags(struct tun_struct *tun)
1906 {
1907 return tun->flags & (TUN_FEATURES | IFF_PERSIST | IFF_TUN | IFF_TAP);
1908 }
1909
1910 static ssize_t tun_show_flags(struct device *dev, struct device_attribute *attr,
1911 char *buf)
1912 {
1913 struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1914 return sprintf(buf, "0x%x\n", tun_flags(tun));
1915 }
1916
1917 static ssize_t tun_show_owner(struct device *dev, struct device_attribute *attr,
1918 char *buf)
1919 {
1920 struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1921 return uid_valid(tun->owner)?
1922 sprintf(buf, "%u\n",
1923 from_kuid_munged(current_user_ns(), tun->owner)):
1924 sprintf(buf, "-1\n");
1925 }
1926
1927 static ssize_t tun_show_group(struct device *dev, struct device_attribute *attr,
1928 char *buf)
1929 {
1930 struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1931 return gid_valid(tun->group) ?
1932 sprintf(buf, "%u\n",
1933 from_kgid_munged(current_user_ns(), tun->group)):
1934 sprintf(buf, "-1\n");
1935 }
1936
1937 static DEVICE_ATTR(tun_flags, 0444, tun_show_flags, NULL);
1938 static DEVICE_ATTR(owner, 0444, tun_show_owner, NULL);
1939 static DEVICE_ATTR(group, 0444, tun_show_group, NULL);
1940
1941 static struct attribute *tun_dev_attrs[] = {
1942 &dev_attr_tun_flags.attr,
1943 &dev_attr_owner.attr,
1944 &dev_attr_group.attr,
1945 NULL
1946 };
1947
1948 static const struct attribute_group tun_attr_group = {
1949 .attrs = tun_dev_attrs
1950 };
1951
1952 static int tun_set_iff(struct net *net, struct file *file, struct ifreq *ifr)
1953 {
1954 struct tun_struct *tun;
1955 struct tun_file *tfile = file->private_data;
1956 struct net_device *dev;
1957 int err;
1958
1959 if (tfile->detached)
1960 return -EINVAL;
1961
1962 dev = __dev_get_by_name(net, ifr->ifr_name);
1963 if (dev) {
1964 if (ifr->ifr_flags & IFF_TUN_EXCL)
1965 return -EBUSY;
1966 if ((ifr->ifr_flags & IFF_TUN) && dev->netdev_ops == &tun_netdev_ops)
1967 tun = netdev_priv(dev);
1968 else if ((ifr->ifr_flags & IFF_TAP) && dev->netdev_ops == &tap_netdev_ops)
1969 tun = netdev_priv(dev);
1970 else
1971 return -EINVAL;
1972
1973 if (!!(ifr->ifr_flags & IFF_MULTI_QUEUE) !=
1974 !!(tun->flags & IFF_MULTI_QUEUE))
1975 return -EINVAL;
1976
1977 if (tun_not_capable(tun))
1978 return -EPERM;
1979 err = security_tun_dev_open(tun->security);
1980 if (err < 0)
1981 return err;
1982
1983 err = tun_attach(tun, file, ifr->ifr_flags & IFF_NOFILTER);
1984 if (err < 0)
1985 return err;
1986
1987 if (tun->flags & IFF_MULTI_QUEUE &&
1988 (tun->numqueues + tun->numdisabled > 1)) {
1989 /* One or more queue has already been attached, no need
1990 * to initialize the device again.
1991 */
1992 return 0;
1993 }
1994 }
1995 else {
1996 char *name;
1997 unsigned long flags = 0;
1998 int queues = ifr->ifr_flags & IFF_MULTI_QUEUE ?
1999 MAX_TAP_QUEUES : 1;
2000
2001 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
2002 return -EPERM;
2003 err = security_tun_dev_create();
2004 if (err < 0)
2005 return err;
2006
2007 /* Set dev type */
2008 if (ifr->ifr_flags & IFF_TUN) {
2009 /* TUN device */
2010 flags |= IFF_TUN;
2011 name = "tun%d";
2012 } else if (ifr->ifr_flags & IFF_TAP) {
2013 /* TAP device */
2014 flags |= IFF_TAP;
2015 name = "tap%d";
2016 } else
2017 return -EINVAL;
2018
2019 if (*ifr->ifr_name)
2020 name = ifr->ifr_name;
2021
2022 dev = alloc_netdev_mqs(sizeof(struct tun_struct), name,
2023 NET_NAME_UNKNOWN, tun_setup, queues,
2024 queues);
2025
2026 if (!dev)
2027 return -ENOMEM;
2028
2029 dev_net_set(dev, net);
2030 dev->rtnl_link_ops = &tun_link_ops;
2031 dev->ifindex = tfile->ifindex;
2032 dev->sysfs_groups[0] = &tun_attr_group;
2033
2034 tun = netdev_priv(dev);
2035 tun->dev = dev;
2036 tun->flags = flags;
2037 tun->txflt.count = 0;
2038 tun->vnet_hdr_sz = sizeof(struct virtio_net_hdr);
2039
2040 tun->align = NET_SKB_PAD;
2041 tun->filter_attached = false;
2042 tun->sndbuf = tfile->socket.sk->sk_sndbuf;
2043 tun->rx_batched = 0;
2044
2045 tun->pcpu_stats = netdev_alloc_pcpu_stats(struct tun_pcpu_stats);
2046 if (!tun->pcpu_stats) {
2047 err = -ENOMEM;
2048 goto err_free_dev;
2049 }
2050
2051 spin_lock_init(&tun->lock);
2052
2053 err = security_tun_dev_alloc_security(&tun->security);
2054 if (err < 0)
2055 goto err_free_stat;
2056
2057 tun_net_init(dev);
2058 tun_flow_init(tun);
2059
2060 dev->hw_features = NETIF_F_SG | NETIF_F_FRAGLIST |
2061 TUN_USER_FEATURES | NETIF_F_HW_VLAN_CTAG_TX |
2062 NETIF_F_HW_VLAN_STAG_TX;
2063 dev->features = dev->hw_features | NETIF_F_LLTX;
2064 dev->vlan_features = dev->features &
2065 ~(NETIF_F_HW_VLAN_CTAG_TX |
2066 NETIF_F_HW_VLAN_STAG_TX);
2067
2068 INIT_LIST_HEAD(&tun->disabled);
2069 err = tun_attach(tun, file, false);
2070 if (err < 0)
2071 goto err_free_flow;
2072
2073 err = register_netdevice(tun->dev);
2074 if (err < 0)
2075 goto err_detach;
2076 }
2077
2078 netif_carrier_on(tun->dev);
2079
2080 tun_debug(KERN_INFO, tun, "tun_set_iff\n");
2081
2082 tun->flags = (tun->flags & ~TUN_FEATURES) |
2083 (ifr->ifr_flags & TUN_FEATURES);
2084
2085 /* Make sure persistent devices do not get stuck in
2086 * xoff state.
2087 */
2088 if (netif_running(tun->dev))
2089 netif_tx_wake_all_queues(tun->dev);
2090
2091 strcpy(ifr->ifr_name, tun->dev->name);
2092 return 0;
2093
2094 err_detach:
2095 tun_detach_all(dev);
2096 /* register_netdevice() already called tun_free_netdev() */
2097 goto err_free_dev;
2098
2099 err_free_flow:
2100 tun_flow_uninit(tun);
2101 security_tun_dev_free_security(tun->security);
2102 err_free_stat:
2103 free_percpu(tun->pcpu_stats);
2104 err_free_dev:
2105 free_netdev(dev);
2106 return err;
2107 }
2108
2109 static void tun_get_iff(struct net *net, struct tun_struct *tun,
2110 struct ifreq *ifr)
2111 {
2112 tun_debug(KERN_INFO, tun, "tun_get_iff\n");
2113
2114 strcpy(ifr->ifr_name, tun->dev->name);
2115
2116 ifr->ifr_flags = tun_flags(tun);
2117
2118 }
2119
2120 /* This is like a cut-down ethtool ops, except done via tun fd so no
2121 * privs required. */
2122 static int set_offload(struct tun_struct *tun, unsigned long arg)
2123 {
2124 netdev_features_t features = 0;
2125
2126 if (arg & TUN_F_CSUM) {
2127 features |= NETIF_F_HW_CSUM;
2128 arg &= ~TUN_F_CSUM;
2129
2130 if (arg & (TUN_F_TSO4|TUN_F_TSO6)) {
2131 if (arg & TUN_F_TSO_ECN) {
2132 features |= NETIF_F_TSO_ECN;
2133 arg &= ~TUN_F_TSO_ECN;
2134 }
2135 if (arg & TUN_F_TSO4)
2136 features |= NETIF_F_TSO;
2137 if (arg & TUN_F_TSO6)
2138 features |= NETIF_F_TSO6;
2139 arg &= ~(TUN_F_TSO4|TUN_F_TSO6);
2140 }
2141 }
2142
2143 /* This gives the user a way to test for new features in future by
2144 * trying to set them. */
2145 if (arg)
2146 return -EINVAL;
2147
2148 tun->set_features = features;
2149 tun->dev->wanted_features &= ~TUN_USER_FEATURES;
2150 tun->dev->wanted_features |= features;
2151 netdev_update_features(tun->dev);
2152
2153 return 0;
2154 }
2155
2156 static void tun_detach_filter(struct tun_struct *tun, int n)
2157 {
2158 int i;
2159 struct tun_file *tfile;
2160
2161 for (i = 0; i < n; i++) {
2162 tfile = rtnl_dereference(tun->tfiles[i]);
2163 lock_sock(tfile->socket.sk);
2164 sk_detach_filter(tfile->socket.sk);
2165 release_sock(tfile->socket.sk);
2166 }
2167
2168 tun->filter_attached = false;
2169 }
2170
2171 static int tun_attach_filter(struct tun_struct *tun)
2172 {
2173 int i, ret = 0;
2174 struct tun_file *tfile;
2175
2176 for (i = 0; i < tun->numqueues; i++) {
2177 tfile = rtnl_dereference(tun->tfiles[i]);
2178 lock_sock(tfile->socket.sk);
2179 ret = sk_attach_filter(&tun->fprog, tfile->socket.sk);
2180 release_sock(tfile->socket.sk);
2181 if (ret) {
2182 tun_detach_filter(tun, i);
2183 return ret;
2184 }
2185 }
2186
2187 tun->filter_attached = true;
2188 return ret;
2189 }
2190
2191 static void tun_set_sndbuf(struct tun_struct *tun)
2192 {
2193 struct tun_file *tfile;
2194 int i;
2195
2196 for (i = 0; i < tun->numqueues; i++) {
2197 tfile = rtnl_dereference(tun->tfiles[i]);
2198 tfile->socket.sk->sk_sndbuf = tun->sndbuf;
2199 }
2200 }
2201
2202 static int tun_set_queue(struct file *file, struct ifreq *ifr)
2203 {
2204 struct tun_file *tfile = file->private_data;
2205 struct tun_struct *tun;
2206 int ret = 0;
2207
2208 rtnl_lock();
2209
2210 if (ifr->ifr_flags & IFF_ATTACH_QUEUE) {
2211 tun = tfile->detached;
2212 if (!tun) {
2213 ret = -EINVAL;
2214 goto unlock;
2215 }
2216 ret = security_tun_dev_attach_queue(tun->security);
2217 if (ret < 0)
2218 goto unlock;
2219 ret = tun_attach(tun, file, false);
2220 } else if (ifr->ifr_flags & IFF_DETACH_QUEUE) {
2221 tun = rtnl_dereference(tfile->tun);
2222 if (!tun || !(tun->flags & IFF_MULTI_QUEUE) || tfile->detached)
2223 ret = -EINVAL;
2224 else
2225 __tun_detach(tfile, false);
2226 } else
2227 ret = -EINVAL;
2228
2229 unlock:
2230 rtnl_unlock();
2231 return ret;
2232 }
2233
2234 static long __tun_chr_ioctl(struct file *file, unsigned int cmd,
2235 unsigned long arg, int ifreq_len)
2236 {
2237 struct tun_file *tfile = file->private_data;
2238 struct tun_struct *tun;
2239 void __user* argp = (void __user*)arg;
2240 struct ifreq ifr;
2241 kuid_t owner;
2242 kgid_t group;
2243 int sndbuf;
2244 int vnet_hdr_sz;
2245 unsigned int ifindex;
2246 int le;
2247 int ret;
2248
2249 if (cmd == TUNSETIFF || cmd == TUNSETQUEUE || _IOC_TYPE(cmd) == SOCK_IOC_TYPE) {
2250 if (copy_from_user(&ifr, argp, ifreq_len))
2251 return -EFAULT;
2252 } else {
2253 memset(&ifr, 0, sizeof(ifr));
2254 }
2255 if (cmd == TUNGETFEATURES) {
2256 /* Currently this just means: "what IFF flags are valid?".
2257 * This is needed because we never checked for invalid flags on
2258 * TUNSETIFF.
2259 */
2260 return put_user(IFF_TUN | IFF_TAP | TUN_FEATURES,
2261 (unsigned int __user*)argp);
2262 } else if (cmd == TUNSETQUEUE)
2263 return tun_set_queue(file, &ifr);
2264
2265 ret = 0;
2266 rtnl_lock();
2267
2268 tun = __tun_get(tfile);
2269 if (cmd == TUNSETIFF) {
2270 ret = -EEXIST;
2271 if (tun)
2272 goto unlock;
2273
2274 ifr.ifr_name[IFNAMSIZ-1] = '\0';
2275
2276 ret = tun_set_iff(sock_net(&tfile->sk), file, &ifr);
2277
2278 if (ret)
2279 goto unlock;
2280
2281 if (copy_to_user(argp, &ifr, ifreq_len))
2282 ret = -EFAULT;
2283 goto unlock;
2284 }
2285 if (cmd == TUNSETIFINDEX) {
2286 ret = -EPERM;
2287 if (tun)
2288 goto unlock;
2289
2290 ret = -EFAULT;
2291 if (copy_from_user(&ifindex, argp, sizeof(ifindex)))
2292 goto unlock;
2293
2294 ret = 0;
2295 tfile->ifindex = ifindex;
2296 goto unlock;
2297 }
2298
2299 ret = -EBADFD;
2300 if (!tun)
2301 goto unlock;
2302
2303 tun_debug(KERN_INFO, tun, "tun_chr_ioctl cmd %u\n", cmd);
2304
2305 ret = 0;
2306 switch (cmd) {
2307 case TUNGETIFF:
2308 tun_get_iff(current->nsproxy->net_ns, tun, &ifr);
2309
2310 if (tfile->detached)
2311 ifr.ifr_flags |= IFF_DETACH_QUEUE;
2312 if (!tfile->socket.sk->sk_filter)
2313 ifr.ifr_flags |= IFF_NOFILTER;
2314
2315 if (copy_to_user(argp, &ifr, ifreq_len))
2316 ret = -EFAULT;
2317 break;
2318
2319 case TUNSETNOCSUM:
2320 /* Disable/Enable checksum */
2321
2322 /* [unimplemented] */
2323 tun_debug(KERN_INFO, tun, "ignored: set checksum %s\n",
2324 arg ? "disabled" : "enabled");
2325 break;
2326
2327 case TUNSETPERSIST:
2328 /* Disable/Enable persist mode. Keep an extra reference to the
2329 * module to prevent the module being unprobed.
2330 */
2331 if (arg && !(tun->flags & IFF_PERSIST)) {
2332 tun->flags |= IFF_PERSIST;
2333 __module_get(THIS_MODULE);
2334 }
2335 if (!arg && (tun->flags & IFF_PERSIST)) {
2336 tun->flags &= ~IFF_PERSIST;
2337 module_put(THIS_MODULE);
2338 }
2339
2340 tun_debug(KERN_INFO, tun, "persist %s\n",
2341 arg ? "enabled" : "disabled");
2342 break;
2343
2344 case TUNSETOWNER:
2345 /* Set owner of the device */
2346 owner = make_kuid(current_user_ns(), arg);
2347 if (!uid_valid(owner)) {
2348 ret = -EINVAL;
2349 break;
2350 }
2351 tun->owner = owner;
2352 tun_debug(KERN_INFO, tun, "owner set to %u\n",
2353 from_kuid(&init_user_ns, tun->owner));
2354 break;
2355
2356 case TUNSETGROUP:
2357 /* Set group of the device */
2358 group = make_kgid(current_user_ns(), arg);
2359 if (!gid_valid(group)) {
2360 ret = -EINVAL;
2361 break;
2362 }
2363 tun->group = group;
2364 tun_debug(KERN_INFO, tun, "group set to %u\n",
2365 from_kgid(&init_user_ns, tun->group));
2366 break;
2367
2368 case TUNSETLINK:
2369 /* Only allow setting the type when the interface is down */
2370 if (tun->dev->flags & IFF_UP) {
2371 tun_debug(KERN_INFO, tun,
2372 "Linktype set failed because interface is up\n");
2373 ret = -EBUSY;
2374 } else {
2375 tun->dev->type = (int) arg;
2376 tun_debug(KERN_INFO, tun, "linktype set to %d\n",
2377 tun->dev->type);
2378 ret = 0;
2379 }
2380 break;
2381
2382 #ifdef TUN_DEBUG
2383 case TUNSETDEBUG:
2384 tun->debug = arg;
2385 break;
2386 #endif
2387 case TUNSETOFFLOAD:
2388 ret = set_offload(tun, arg);
2389 break;
2390
2391 case TUNSETTXFILTER:
2392 /* Can be set only for TAPs */
2393 ret = -EINVAL;
2394 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
2395 break;
2396 ret = update_filter(&tun->txflt, (void __user *)arg);
2397 break;
2398
2399 case SIOCGIFHWADDR:
2400 /* Get hw address */
2401 memcpy(ifr.ifr_hwaddr.sa_data, tun->dev->dev_addr, ETH_ALEN);
2402 ifr.ifr_hwaddr.sa_family = tun->dev->type;
2403 if (copy_to_user(argp, &ifr, ifreq_len))
2404 ret = -EFAULT;
2405 break;
2406
2407 case SIOCSIFHWADDR:
2408 /* Set hw address */
2409 tun_debug(KERN_DEBUG, tun, "set hw address: %pM\n",
2410 ifr.ifr_hwaddr.sa_data);
2411
2412 ret = dev_set_mac_address(tun->dev, &ifr.ifr_hwaddr);
2413 break;
2414
2415 case TUNGETSNDBUF:
2416 sndbuf = tfile->socket.sk->sk_sndbuf;
2417 if (copy_to_user(argp, &sndbuf, sizeof(sndbuf)))
2418 ret = -EFAULT;
2419 break;
2420
2421 case TUNSETSNDBUF:
2422 if (copy_from_user(&sndbuf, argp, sizeof(sndbuf))) {
2423 ret = -EFAULT;
2424 break;
2425 }
2426
2427 tun->sndbuf = sndbuf;
2428 tun_set_sndbuf(tun);
2429 break;
2430
2431 case TUNGETVNETHDRSZ:
2432 vnet_hdr_sz = tun->vnet_hdr_sz;
2433 if (copy_to_user(argp, &vnet_hdr_sz, sizeof(vnet_hdr_sz)))
2434 ret = -EFAULT;
2435 break;
2436
2437 case TUNSETVNETHDRSZ:
2438 if (copy_from_user(&vnet_hdr_sz, argp, sizeof(vnet_hdr_sz))) {
2439 ret = -EFAULT;
2440 break;
2441 }
2442 if (vnet_hdr_sz < (int)sizeof(struct virtio_net_hdr)) {
2443 ret = -EINVAL;
2444 break;
2445 }
2446
2447 tun->vnet_hdr_sz = vnet_hdr_sz;
2448 break;
2449
2450 case TUNGETVNETLE:
2451 le = !!(tun->flags & TUN_VNET_LE);
2452 if (put_user(le, (int __user *)argp))
2453 ret = -EFAULT;
2454 break;
2455
2456 case TUNSETVNETLE:
2457 if (get_user(le, (int __user *)argp)) {
2458 ret = -EFAULT;
2459 break;
2460 }
2461 if (le)
2462 tun->flags |= TUN_VNET_LE;
2463 else
2464 tun->flags &= ~TUN_VNET_LE;
2465 break;
2466
2467 case TUNGETVNETBE:
2468 ret = tun_get_vnet_be(tun, argp);
2469 break;
2470
2471 case TUNSETVNETBE:
2472 ret = tun_set_vnet_be(tun, argp);
2473 break;
2474
2475 case TUNATTACHFILTER:
2476 /* Can be set only for TAPs */
2477 ret = -EINVAL;
2478 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
2479 break;
2480 ret = -EFAULT;
2481 if (copy_from_user(&tun->fprog, argp, sizeof(tun->fprog)))
2482 break;
2483
2484 ret = tun_attach_filter(tun);
2485 break;
2486
2487 case TUNDETACHFILTER:
2488 /* Can be set only for TAPs */
2489 ret = -EINVAL;
2490 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
2491 break;
2492 ret = 0;
2493 tun_detach_filter(tun, tun->numqueues);
2494 break;
2495
2496 case TUNGETFILTER:
2497 ret = -EINVAL;
2498 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
2499 break;
2500 ret = -EFAULT;
2501 if (copy_to_user(argp, &tun->fprog, sizeof(tun->fprog)))
2502 break;
2503 ret = 0;
2504 break;
2505
2506 default:
2507 ret = -EINVAL;
2508 break;
2509 }
2510
2511 unlock:
2512 rtnl_unlock();
2513 if (tun)
2514 tun_put(tun);
2515 return ret;
2516 }
2517
2518 static long tun_chr_ioctl(struct file *file,
2519 unsigned int cmd, unsigned long arg)
2520 {
2521 return __tun_chr_ioctl(file, cmd, arg, sizeof (struct ifreq));
2522 }
2523
2524 #ifdef CONFIG_COMPAT
2525 static long tun_chr_compat_ioctl(struct file *file,
2526 unsigned int cmd, unsigned long arg)
2527 {
2528 switch (cmd) {
2529 case TUNSETIFF:
2530 case TUNGETIFF:
2531 case TUNSETTXFILTER:
2532 case TUNGETSNDBUF:
2533 case TUNSETSNDBUF:
2534 case SIOCGIFHWADDR:
2535 case SIOCSIFHWADDR:
2536 arg = (unsigned long)compat_ptr(arg);
2537 break;
2538 default:
2539 arg = (compat_ulong_t)arg;
2540 break;
2541 }
2542
2543 /*
2544 * compat_ifreq is shorter than ifreq, so we must not access beyond
2545 * the end of that structure. All fields that are used in this
2546 * driver are compatible though, we don't need to convert the
2547 * contents.
2548 */
2549 return __tun_chr_ioctl(file, cmd, arg, sizeof(struct compat_ifreq));
2550 }
2551 #endif /* CONFIG_COMPAT */
2552
2553 static int tun_chr_fasync(int fd, struct file *file, int on)
2554 {
2555 struct tun_file *tfile = file->private_data;
2556 int ret;
2557
2558 if ((ret = fasync_helper(fd, file, on, &tfile->fasync)) < 0)
2559 goto out;
2560
2561 if (on) {
2562 __f_setown(file, task_pid(current), PIDTYPE_PID, 0);
2563 tfile->flags |= TUN_FASYNC;
2564 } else
2565 tfile->flags &= ~TUN_FASYNC;
2566 ret = 0;
2567 out:
2568 return ret;
2569 }
2570
2571 static int tun_chr_open(struct inode *inode, struct file * file)
2572 {
2573 struct net *net = current->nsproxy->net_ns;
2574 struct tun_file *tfile;
2575
2576 DBG1(KERN_INFO, "tunX: tun_chr_open\n");
2577
2578 tfile = (struct tun_file *)sk_alloc(net, AF_UNSPEC, GFP_KERNEL,
2579 &tun_proto, 0);
2580 if (!tfile)
2581 return -ENOMEM;
2582 RCU_INIT_POINTER(tfile->tun, NULL);
2583 tfile->flags = 0;
2584 tfile->ifindex = 0;
2585
2586 init_waitqueue_head(&tfile->wq.wait);
2587 RCU_INIT_POINTER(tfile->socket.wq, &tfile->wq);
2588
2589 tfile->socket.file = file;
2590 tfile->socket.ops = &tun_socket_ops;
2591
2592 sock_init_data(&tfile->socket, &tfile->sk);
2593
2594 tfile->sk.sk_write_space = tun_sock_write_space;
2595 tfile->sk.sk_sndbuf = INT_MAX;
2596
2597 file->private_data = tfile;
2598 INIT_LIST_HEAD(&tfile->next);
2599
2600 sock_set_flag(&tfile->sk, SOCK_ZEROCOPY);
2601
2602 return 0;
2603 }
2604
2605 static int tun_chr_close(struct inode *inode, struct file *file)
2606 {
2607 struct tun_file *tfile = file->private_data;
2608
2609 tun_detach(tfile, true);
2610
2611 return 0;
2612 }
2613
2614 #ifdef CONFIG_PROC_FS
2615 static void tun_chr_show_fdinfo(struct seq_file *m, struct file *f)
2616 {
2617 struct tun_struct *tun;
2618 struct ifreq ifr;
2619
2620 memset(&ifr, 0, sizeof(ifr));
2621
2622 rtnl_lock();
2623 tun = tun_get(f);
2624 if (tun)
2625 tun_get_iff(current->nsproxy->net_ns, tun, &ifr);
2626 rtnl_unlock();
2627
2628 if (tun)
2629 tun_put(tun);
2630
2631 seq_printf(m, "iff:\t%s\n", ifr.ifr_name);
2632 }
2633 #endif
2634
2635 static const struct file_operations tun_fops = {
2636 .owner = THIS_MODULE,
2637 .llseek = no_llseek,
2638 .read_iter = tun_chr_read_iter,
2639 .write_iter = tun_chr_write_iter,
2640 .poll = tun_chr_poll,
2641 .unlocked_ioctl = tun_chr_ioctl,
2642 #ifdef CONFIG_COMPAT
2643 .compat_ioctl = tun_chr_compat_ioctl,
2644 #endif
2645 .open = tun_chr_open,
2646 .release = tun_chr_close,
2647 .fasync = tun_chr_fasync,
2648 #ifdef CONFIG_PROC_FS
2649 .show_fdinfo = tun_chr_show_fdinfo,
2650 #endif
2651 };
2652
2653 static struct miscdevice tun_miscdev = {
2654 .minor = TUN_MINOR,
2655 .name = "tun",
2656 .nodename = "net/tun",
2657 .fops = &tun_fops,
2658 };
2659
2660 /* ethtool interface */
2661
2662 static int tun_get_link_ksettings(struct net_device *dev,
2663 struct ethtool_link_ksettings *cmd)
2664 {
2665 ethtool_link_ksettings_zero_link_mode(cmd, supported);
2666 ethtool_link_ksettings_zero_link_mode(cmd, advertising);
2667 cmd->base.speed = SPEED_10;
2668 cmd->base.duplex = DUPLEX_FULL;
2669 cmd->base.port = PORT_TP;
2670 cmd->base.phy_address = 0;
2671 cmd->base.autoneg = AUTONEG_DISABLE;
2672 return 0;
2673 }
2674
2675 static void tun_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
2676 {
2677 struct tun_struct *tun = netdev_priv(dev);
2678
2679 strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
2680 strlcpy(info->version, DRV_VERSION, sizeof(info->version));
2681
2682 switch (tun->flags & TUN_TYPE_MASK) {
2683 case IFF_TUN:
2684 strlcpy(info->bus_info, "tun", sizeof(info->bus_info));
2685 break;
2686 case IFF_TAP:
2687 strlcpy(info->bus_info, "tap", sizeof(info->bus_info));
2688 break;
2689 }
2690 }
2691
2692 static u32 tun_get_msglevel(struct net_device *dev)
2693 {
2694 #ifdef TUN_DEBUG
2695 struct tun_struct *tun = netdev_priv(dev);
2696 return tun->debug;
2697 #else
2698 return -EOPNOTSUPP;
2699 #endif
2700 }
2701
2702 static void tun_set_msglevel(struct net_device *dev, u32 value)
2703 {
2704 #ifdef TUN_DEBUG
2705 struct tun_struct *tun = netdev_priv(dev);
2706 tun->debug = value;
2707 #endif
2708 }
2709
2710 static int tun_get_coalesce(struct net_device *dev,
2711 struct ethtool_coalesce *ec)
2712 {
2713 struct tun_struct *tun = netdev_priv(dev);
2714
2715 ec->rx_max_coalesced_frames = tun->rx_batched;
2716
2717 return 0;
2718 }
2719
2720 static int tun_set_coalesce(struct net_device *dev,
2721 struct ethtool_coalesce *ec)
2722 {
2723 struct tun_struct *tun = netdev_priv(dev);
2724
2725 if (ec->rx_max_coalesced_frames > NAPI_POLL_WEIGHT)
2726 tun->rx_batched = NAPI_POLL_WEIGHT;
2727 else
2728 tun->rx_batched = ec->rx_max_coalesced_frames;
2729
2730 return 0;
2731 }
2732
2733 static const struct ethtool_ops tun_ethtool_ops = {
2734 .get_drvinfo = tun_get_drvinfo,
2735 .get_msglevel = tun_get_msglevel,
2736 .set_msglevel = tun_set_msglevel,
2737 .get_link = ethtool_op_get_link,
2738 .get_ts_info = ethtool_op_get_ts_info,
2739 .get_coalesce = tun_get_coalesce,
2740 .set_coalesce = tun_set_coalesce,
2741 .get_link_ksettings = tun_get_link_ksettings,
2742 };
2743
2744 static int tun_queue_resize(struct tun_struct *tun)
2745 {
2746 struct net_device *dev = tun->dev;
2747 struct tun_file *tfile;
2748 struct skb_array **arrays;
2749 int n = tun->numqueues + tun->numdisabled;
2750 int ret, i;
2751
2752 arrays = kmalloc_array(n, sizeof(*arrays), GFP_KERNEL);
2753 if (!arrays)
2754 return -ENOMEM;
2755
2756 for (i = 0; i < tun->numqueues; i++) {
2757 tfile = rtnl_dereference(tun->tfiles[i]);
2758 arrays[i] = &tfile->tx_array;
2759 }
2760 list_for_each_entry(tfile, &tun->disabled, next)
2761 arrays[i++] = &tfile->tx_array;
2762
2763 ret = skb_array_resize_multiple(arrays, n,
2764 dev->tx_queue_len, GFP_KERNEL);
2765
2766 kfree(arrays);
2767 return ret;
2768 }
2769
2770 static int tun_device_event(struct notifier_block *unused,
2771 unsigned long event, void *ptr)
2772 {
2773 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
2774 struct tun_struct *tun = netdev_priv(dev);
2775
2776 if (dev->rtnl_link_ops != &tun_link_ops)
2777 return NOTIFY_DONE;
2778
2779 switch (event) {
2780 case NETDEV_CHANGE_TX_QUEUE_LEN:
2781 if (tun_queue_resize(tun))
2782 return NOTIFY_BAD;
2783 break;
2784 default:
2785 break;
2786 }
2787
2788 return NOTIFY_DONE;
2789 }
2790
2791 static struct notifier_block tun_notifier_block __read_mostly = {
2792 .notifier_call = tun_device_event,
2793 };
2794
2795 static int __init tun_init(void)
2796 {
2797 int ret = 0;
2798
2799 pr_info("%s, %s\n", DRV_DESCRIPTION, DRV_VERSION);
2800
2801 ret = rtnl_link_register(&tun_link_ops);
2802 if (ret) {
2803 pr_err("Can't register link_ops\n");
2804 goto err_linkops;
2805 }
2806
2807 ret = misc_register(&tun_miscdev);
2808 if (ret) {
2809 pr_err("Can't register misc device %d\n", TUN_MINOR);
2810 goto err_misc;
2811 }
2812
2813 ret = register_netdevice_notifier(&tun_notifier_block);
2814 if (ret) {
2815 pr_err("Can't register netdevice notifier\n");
2816 goto err_notifier;
2817 }
2818
2819 return 0;
2820
2821 err_notifier:
2822 misc_deregister(&tun_miscdev);
2823 err_misc:
2824 rtnl_link_unregister(&tun_link_ops);
2825 err_linkops:
2826 return ret;
2827 }
2828
2829 static void tun_cleanup(void)
2830 {
2831 misc_deregister(&tun_miscdev);
2832 rtnl_link_unregister(&tun_link_ops);
2833 unregister_netdevice_notifier(&tun_notifier_block);
2834 }
2835
2836 /* Get an underlying socket object from tun file. Returns error unless file is
2837 * attached to a device. The returned object works like a packet socket, it
2838 * can be used for sock_sendmsg/sock_recvmsg. The caller is responsible for
2839 * holding a reference to the file for as long as the socket is in use. */
2840 struct socket *tun_get_socket(struct file *file)
2841 {
2842 struct tun_file *tfile;
2843 if (file->f_op != &tun_fops)
2844 return ERR_PTR(-EINVAL);
2845 tfile = file->private_data;
2846 if (!tfile)
2847 return ERR_PTR(-EBADFD);
2848 return &tfile->socket;
2849 }
2850 EXPORT_SYMBOL_GPL(tun_get_socket);
2851
2852 struct skb_array *tun_get_skb_array(struct file *file)
2853 {
2854 struct tun_file *tfile;
2855
2856 if (file->f_op != &tun_fops)
2857 return ERR_PTR(-EINVAL);
2858 tfile = file->private_data;
2859 if (!tfile)
2860 return ERR_PTR(-EBADFD);
2861 return &tfile->tx_array;
2862 }
2863 EXPORT_SYMBOL_GPL(tun_get_skb_array);
2864
2865 module_init(tun_init);
2866 module_exit(tun_cleanup);
2867 MODULE_DESCRIPTION(DRV_DESCRIPTION);
2868 MODULE_AUTHOR(DRV_COPYRIGHT);
2869 MODULE_LICENSE("GPL");
2870 MODULE_ALIAS_MISCDEV(TUN_MINOR);
2871 MODULE_ALIAS("devname:net/tun");
Linux with added FPC-III support