]> git.ipfire.org Git - people/arne_f/kernel.git/blob - drivers/net/tun.c
projects / people / arne_f / kernel.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
tun: fix rcu_read_lock imbalance in tun_build_skb
[people/arne_f/kernel.git] / drivers / net / tun.c
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 alloc_frag->offset = ALIGN((u64)alloc_frag->offset, SMP_CACHE_BYTES);
1290 if (unlikely(!skb_page_frag_refill(buflen, alloc_frag, GFP_KERNEL)))
1291 return ERR_PTR(-ENOMEM);
1292
1293 buf = (char *)page_address(alloc_frag->page) + alloc_frag->offset;
1294 copied = copy_page_from_iter(alloc_frag->page,
1295 alloc_frag->offset + pad,
1296 len, from);
1297 if (copied != len)
1298 return ERR_PTR(-EFAULT);
1299
1300 /* There's a small window that XDP may be set after the check
1301 * of xdp_prog above, this should be rare and for simplicity
1302 * we do XDP on skb in case the headroom is not enough.
1303 */
1304 if (hdr->gso_type || !xdp_prog)
1305 *skb_xdp = 1;
1306 else
1307 *skb_xdp = 0;
1308
1309 rcu_read_lock();
1310 xdp_prog = rcu_dereference(tun->xdp_prog);
1311 if (xdp_prog && !*skb_xdp) {
1312 struct xdp_buff xdp;
1313 void *orig_data;
1314 u32 act;
1315
1316 xdp.data_hard_start = buf;
1317 xdp.data = buf + pad;
1318 xdp.data_end = xdp.data + len;
1319 orig_data = xdp.data;
1320 act = bpf_prog_run_xdp(xdp_prog, &xdp);
1321
1322 switch (act) {
1323 case XDP_REDIRECT:
1324 get_page(alloc_frag->page);
1325 alloc_frag->offset += buflen;
1326 err = xdp_do_redirect(tun->dev, &xdp, xdp_prog);
1327 if (err)
1328 goto err_redirect;
1329 rcu_read_unlock();
1330 return NULL;
1331 case XDP_TX:
1332 xdp_xmit = true;
1333 /* fall through */
1334 case XDP_PASS:
1335 delta = orig_data - xdp.data;
1336 break;
1337 default:
1338 bpf_warn_invalid_xdp_action(act);
1339 /* fall through */
1340 case XDP_ABORTED:
1341 trace_xdp_exception(tun->dev, xdp_prog, act);
1342 /* fall through */
1343 case XDP_DROP:
1344 goto err_xdp;
1345 }
1346 }
1347
1348 skb = build_skb(buf, buflen);
1349 if (!skb) {
1350 rcu_read_unlock();
1351 return ERR_PTR(-ENOMEM);
1352 }
1353
1354 skb_reserve(skb, pad - delta);
1355 skb_put(skb, len + delta);
1356 get_page(alloc_frag->page);
1357 alloc_frag->offset += buflen;
1358
1359 if (xdp_xmit) {
1360 skb->dev = tun->dev;
1361 generic_xdp_tx(skb, xdp_prog);
1362 rcu_read_unlock();
1363 return NULL;
1364 }
1365
1366 rcu_read_unlock();
1367
1368 return skb;
1369
1370 err_redirect:
1371 put_page(alloc_frag->page);
1372 err_xdp:
1373 rcu_read_unlock();
1374 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1375 return NULL;
1376 }
1377
1378 /* Get packet from user space buffer */
1379 static ssize_t tun_get_user(struct tun_struct *tun, struct tun_file *tfile,
1380 void *msg_control, struct iov_iter *from,
1381 int noblock, bool more)
1382 {
1383 struct tun_pi pi = { 0, cpu_to_be16(ETH_P_IP) };
1384 struct sk_buff *skb;
1385 size_t total_len = iov_iter_count(from);
1386 size_t len = total_len, align = tun->align, linear;
1387 struct virtio_net_hdr gso = { 0 };
1388 struct tun_pcpu_stats *stats;
1389 int good_linear;
1390 int copylen;
1391 bool zerocopy = false;
1392 int err;
1393 u32 rxhash;
1394 int skb_xdp = 1;
1395
1396 if (!(tun->dev->flags & IFF_UP))
1397 return -EIO;
1398
1399 if (!(tun->flags & IFF_NO_PI)) {
1400 if (len < sizeof(pi))
1401 return -EINVAL;
1402 len -= sizeof(pi);
1403
1404 if (!copy_from_iter_full(&pi, sizeof(pi), from))
1405 return -EFAULT;
1406 }
1407
1408 if (tun->flags & IFF_VNET_HDR) {
1409 int vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
1410
1411 if (len < vnet_hdr_sz)
1412 return -EINVAL;
1413 len -= vnet_hdr_sz;
1414
1415 if (!copy_from_iter_full(&gso, sizeof(gso), from))
1416 return -EFAULT;
1417
1418 if ((gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
1419 tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2 > tun16_to_cpu(tun, gso.hdr_len))
1420 gso.hdr_len = cpu_to_tun16(tun, tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2);
1421
1422 if (tun16_to_cpu(tun, gso.hdr_len) > len)
1423 return -EINVAL;
1424 iov_iter_advance(from, vnet_hdr_sz - sizeof(gso));
1425 }
1426
1427 if ((tun->flags & TUN_TYPE_MASK) == IFF_TAP) {
1428 align += NET_IP_ALIGN;
1429 if (unlikely(len < ETH_HLEN ||
1430 (gso.hdr_len && tun16_to_cpu(tun, gso.hdr_len) < ETH_HLEN)))
1431 return -EINVAL;
1432 }
1433
1434 good_linear = SKB_MAX_HEAD(align);
1435
1436 if (msg_control) {
1437 struct iov_iter i = *from;
1438
1439 /* There are 256 bytes to be copied in skb, so there is
1440 * enough room for skb expand head in case it is used.
1441 * The rest of the buffer is mapped from userspace.
1442 */
1443 copylen = gso.hdr_len ? tun16_to_cpu(tun, gso.hdr_len) : GOODCOPY_LEN;
1444 if (copylen > good_linear)
1445 copylen = good_linear;
1446 linear = copylen;
1447 iov_iter_advance(&i, copylen);
1448 if (iov_iter_npages(&i, INT_MAX) <= MAX_SKB_FRAGS)
1449 zerocopy = true;
1450 }
1451
1452 if (tun_can_build_skb(tun, tfile, len, noblock, zerocopy)) {
1453 /* For the packet that is not easy to be processed
1454 * (e.g gso or jumbo packet), we will do it at after
1455 * skb was created with generic XDP routine.
1456 */
1457 skb = tun_build_skb(tun, tfile, from, &gso, len, &skb_xdp);
1458 if (IS_ERR(skb)) {
1459 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1460 return PTR_ERR(skb);
1461 }
1462 if (!skb)
1463 return total_len;
1464 } else {
1465 if (!zerocopy) {
1466 copylen = len;
1467 if (tun16_to_cpu(tun, gso.hdr_len) > good_linear)
1468 linear = good_linear;
1469 else
1470 linear = tun16_to_cpu(tun, gso.hdr_len);
1471 }
1472
1473 skb = tun_alloc_skb(tfile, align, copylen, linear, noblock);
1474 if (IS_ERR(skb)) {
1475 if (PTR_ERR(skb) != -EAGAIN)
1476 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1477 return PTR_ERR(skb);
1478 }
1479
1480 if (zerocopy)
1481 err = zerocopy_sg_from_iter(skb, from);
1482 else
1483 err = skb_copy_datagram_from_iter(skb, 0, from, len);
1484
1485 if (err) {
1486 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1487 kfree_skb(skb);
1488 return -EFAULT;
1489 }
1490 }
1491
1492 if (virtio_net_hdr_to_skb(skb, &gso, tun_is_little_endian(tun))) {
1493 this_cpu_inc(tun->pcpu_stats->rx_frame_errors);
1494 kfree_skb(skb);
1495 return -EINVAL;
1496 }
1497
1498 switch (tun->flags & TUN_TYPE_MASK) {
1499 case IFF_TUN:
1500 if (tun->flags & IFF_NO_PI) {
1501 u8 ip_version = skb->len ? (skb->data[0] >> 4) : 0;
1502
1503 switch (ip_version) {
1504 case 4:
1505 pi.proto = htons(ETH_P_IP);
1506 break;
1507 case 6:
1508 pi.proto = htons(ETH_P_IPV6);
1509 break;
1510 default:
1511 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1512 kfree_skb(skb);
1513 return -EINVAL;
1514 }
1515 }
1516
1517 skb_reset_mac_header(skb);
1518 skb->protocol = pi.proto;
1519 skb->dev = tun->dev;
1520 break;
1521 case IFF_TAP:
1522 skb->protocol = eth_type_trans(skb, tun->dev);
1523 break;
1524 }
1525
1526 /* copy skb_ubuf_info for callback when skb has no error */
1527 if (zerocopy) {
1528 skb_shinfo(skb)->destructor_arg = msg_control;
1529 skb_shinfo(skb)->tx_flags |= SKBTX_DEV_ZEROCOPY;
1530 skb_shinfo(skb)->tx_flags |= SKBTX_SHARED_FRAG;
1531 } else if (msg_control) {
1532 struct ubuf_info *uarg = msg_control;
1533 uarg->callback(uarg, false);
1534 }
1535
1536 skb_reset_network_header(skb);
1537 skb_probe_transport_header(skb, 0);
1538
1539 if (skb_xdp) {
1540 struct bpf_prog *xdp_prog;
1541 int ret;
1542
1543 rcu_read_lock();
1544 xdp_prog = rcu_dereference(tun->xdp_prog);
1545 if (xdp_prog) {
1546 ret = do_xdp_generic(xdp_prog, skb);
1547 if (ret != XDP_PASS) {
1548 rcu_read_unlock();
1549 return total_len;
1550 }
1551 }
1552 rcu_read_unlock();
1553 }
1554
1555 rxhash = __skb_get_hash_symmetric(skb);
1556 #ifndef CONFIG_4KSTACKS
1557 tun_rx_batched(tun, tfile, skb, more);
1558 #else
1559 netif_rx_ni(skb);
1560 #endif
1561
1562 stats = get_cpu_ptr(tun->pcpu_stats);
1563 u64_stats_update_begin(&stats->syncp);
1564 stats->rx_packets++;
1565 stats->rx_bytes += len;
1566 u64_stats_update_end(&stats->syncp);
1567 put_cpu_ptr(stats);
1568
1569 tun_flow_update(tun, rxhash, tfile);
1570 return total_len;
1571 }
1572
1573 static ssize_t tun_chr_write_iter(struct kiocb *iocb, struct iov_iter *from)
1574 {
1575 struct file *file = iocb->ki_filp;
1576 struct tun_struct *tun = tun_get(file);
1577 struct tun_file *tfile = file->private_data;
1578 ssize_t result;
1579
1580 if (!tun)
1581 return -EBADFD;
1582
1583 result = tun_get_user(tun, tfile, NULL, from,
1584 file->f_flags & O_NONBLOCK, false);
1585
1586 tun_put(tun);
1587 return result;
1588 }
1589
1590 /* Put packet to the user space buffer */
1591 static ssize_t tun_put_user(struct tun_struct *tun,
1592 struct tun_file *tfile,
1593 struct sk_buff *skb,
1594 struct iov_iter *iter)
1595 {
1596 struct tun_pi pi = { 0, skb->protocol };
1597 struct tun_pcpu_stats *stats;
1598 ssize_t total;
1599 int vlan_offset = 0;
1600 int vlan_hlen = 0;
1601 int vnet_hdr_sz = 0;
1602
1603 if (skb_vlan_tag_present(skb))
1604 vlan_hlen = VLAN_HLEN;
1605
1606 if (tun->flags & IFF_VNET_HDR)
1607 vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
1608
1609 total = skb->len + vlan_hlen + vnet_hdr_sz;
1610
1611 if (!(tun->flags & IFF_NO_PI)) {
1612 if (iov_iter_count(iter) < sizeof(pi))
1613 return -EINVAL;
1614
1615 total += sizeof(pi);
1616 if (iov_iter_count(iter) < total) {
1617 /* Packet will be striped */
1618 pi.flags |= TUN_PKT_STRIP;
1619 }
1620
1621 if (copy_to_iter(&pi, sizeof(pi), iter) != sizeof(pi))
1622 return -EFAULT;
1623 }
1624
1625 if (vnet_hdr_sz) {
1626 struct virtio_net_hdr gso;
1627
1628 if (iov_iter_count(iter) < vnet_hdr_sz)
1629 return -EINVAL;
1630
1631 if (virtio_net_hdr_from_skb(skb, &gso,
1632 tun_is_little_endian(tun), true)) {
1633 struct skb_shared_info *sinfo = skb_shinfo(skb);
1634 pr_err("unexpected GSO type: "
1635 "0x%x, gso_size %d, hdr_len %d\n",
1636 sinfo->gso_type, tun16_to_cpu(tun, gso.gso_size),
1637 tun16_to_cpu(tun, gso.hdr_len));
1638 print_hex_dump(KERN_ERR, "tun: ",
1639 DUMP_PREFIX_NONE,
1640 16, 1, skb->head,
1641 min((int)tun16_to_cpu(tun, gso.hdr_len), 64), true);
1642 WARN_ON_ONCE(1);
1643 return -EINVAL;
1644 }
1645
1646 if (copy_to_iter(&gso, sizeof(gso), iter) != sizeof(gso))
1647 return -EFAULT;
1648
1649 iov_iter_advance(iter, vnet_hdr_sz - sizeof(gso));
1650 }
1651
1652 if (vlan_hlen) {
1653 int ret;
1654 struct {
1655 __be16 h_vlan_proto;
1656 __be16 h_vlan_TCI;
1657 } veth;
1658
1659 veth.h_vlan_proto = skb->vlan_proto;
1660 veth.h_vlan_TCI = htons(skb_vlan_tag_get(skb));
1661
1662 vlan_offset = offsetof(struct vlan_ethhdr, h_vlan_proto);
1663
1664 ret = skb_copy_datagram_iter(skb, 0, iter, vlan_offset);
1665 if (ret || !iov_iter_count(iter))
1666 goto done;
1667
1668 ret = copy_to_iter(&veth, sizeof(veth), iter);
1669 if (ret != sizeof(veth) || !iov_iter_count(iter))
1670 goto done;
1671 }
1672
1673 skb_copy_datagram_iter(skb, vlan_offset, iter, skb->len - vlan_offset);
1674
1675 done:
1676 /* caller is in process context, */
1677 stats = get_cpu_ptr(tun->pcpu_stats);
1678 u64_stats_update_begin(&stats->syncp);
1679 stats->tx_packets++;
1680 stats->tx_bytes += skb->len + vlan_hlen;
1681 u64_stats_update_end(&stats->syncp);
1682 put_cpu_ptr(tun->pcpu_stats);
1683
1684 return total;
1685 }
1686
1687 static struct sk_buff *tun_ring_recv(struct tun_file *tfile, int noblock,
1688 int *err)
1689 {
1690 DECLARE_WAITQUEUE(wait, current);
1691 struct sk_buff *skb = NULL;
1692 int error = 0;
1693
1694 skb = skb_array_consume(&tfile->tx_array);
1695 if (skb)
1696 goto out;
1697 if (noblock) {
1698 error = -EAGAIN;
1699 goto out;
1700 }
1701
1702 add_wait_queue(&tfile->wq.wait, &wait);
1703 current->state = TASK_INTERRUPTIBLE;
1704
1705 while (1) {
1706 skb = skb_array_consume(&tfile->tx_array);
1707 if (skb)
1708 break;
1709 if (signal_pending(current)) {
1710 error = -ERESTARTSYS;
1711 break;
1712 }
1713 if (tfile->socket.sk->sk_shutdown & RCV_SHUTDOWN) {
1714 error = -EFAULT;
1715 break;
1716 }
1717
1718 schedule();
1719 }
1720
1721 current->state = TASK_RUNNING;
1722 remove_wait_queue(&tfile->wq.wait, &wait);
1723
1724 out:
1725 *err = error;
1726 return skb;
1727 }
1728
1729 static ssize_t tun_do_read(struct tun_struct *tun, struct tun_file *tfile,
1730 struct iov_iter *to,
1731 int noblock, struct sk_buff *skb)
1732 {
1733 ssize_t ret;
1734 int err;
1735
1736 tun_debug(KERN_INFO, tun, "tun_do_read\n");
1737
1738 if (!iov_iter_count(to)) {
1739 if (skb)
1740 kfree_skb(skb);
1741 return 0;
1742 }
1743
1744 if (!skb) {
1745 /* Read frames from ring */
1746 skb = tun_ring_recv(tfile, noblock, &err);
1747 if (!skb)
1748 return err;
1749 }
1750
1751 ret = tun_put_user(tun, tfile, skb, to);
1752 if (unlikely(ret < 0))
1753 kfree_skb(skb);
1754 else
1755 consume_skb(skb);
1756
1757 return ret;
1758 }
1759
1760 static ssize_t tun_chr_read_iter(struct kiocb *iocb, struct iov_iter *to)
1761 {
1762 struct file *file = iocb->ki_filp;
1763 struct tun_file *tfile = file->private_data;
1764 struct tun_struct *tun = __tun_get(tfile);
1765 ssize_t len = iov_iter_count(to), ret;
1766
1767 if (!tun)
1768 return -EBADFD;
1769 ret = tun_do_read(tun, tfile, to, file->f_flags & O_NONBLOCK, NULL);
1770 ret = min_t(ssize_t, ret, len);
1771 if (ret > 0)
1772 iocb->ki_pos = ret;
1773 tun_put(tun);
1774 return ret;
1775 }
1776
1777 static void tun_free_netdev(struct net_device *dev)
1778 {
1779 struct tun_struct *tun = netdev_priv(dev);
1780
1781 BUG_ON(!(list_empty(&tun->disabled)));
1782 free_percpu(tun->pcpu_stats);
1783 tun_flow_uninit(tun);
1784 security_tun_dev_free_security(tun->security);
1785 }
1786
1787 static void tun_setup(struct net_device *dev)
1788 {
1789 struct tun_struct *tun = netdev_priv(dev);
1790
1791 tun->owner = INVALID_UID;
1792 tun->group = INVALID_GID;
1793
1794 dev->ethtool_ops = &tun_ethtool_ops;
1795 dev->needs_free_netdev = true;
1796 dev->priv_destructor = tun_free_netdev;
1797 /* We prefer our own queue length */
1798 dev->tx_queue_len = TUN_READQ_SIZE;
1799 }
1800
1801 /* Trivial set of netlink ops to allow deleting tun or tap
1802 * device with netlink.
1803 */
1804 static int tun_validate(struct nlattr *tb[], struct nlattr *data[],
1805 struct netlink_ext_ack *extack)
1806 {
1807 return -EINVAL;
1808 }
1809
1810 static struct rtnl_link_ops tun_link_ops __read_mostly = {
1811 .kind = DRV_NAME,
1812 .priv_size = sizeof(struct tun_struct),
1813 .setup = tun_setup,
1814 .validate = tun_validate,
1815 };
1816
1817 static void tun_sock_write_space(struct sock *sk)
1818 {
1819 struct tun_file *tfile;
1820 wait_queue_head_t *wqueue;
1821
1822 if (!sock_writeable(sk))
1823 return;
1824
1825 if (!test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags))
1826 return;
1827
1828 wqueue = sk_sleep(sk);
1829 if (wqueue && waitqueue_active(wqueue))
1830 wake_up_interruptible_sync_poll(wqueue, POLLOUT |
1831 POLLWRNORM | POLLWRBAND);
1832
1833 tfile = container_of(sk, struct tun_file, sk);
1834 kill_fasync(&tfile->fasync, SIGIO, POLL_OUT);
1835 }
1836
1837 static int tun_sendmsg(struct socket *sock, struct msghdr *m, size_t total_len)
1838 {
1839 int ret;
1840 struct tun_file *tfile = container_of(sock, struct tun_file, socket);
1841 struct tun_struct *tun = __tun_get(tfile);
1842
1843 if (!tun)
1844 return -EBADFD;
1845
1846 ret = tun_get_user(tun, tfile, m->msg_control, &m->msg_iter,
1847 m->msg_flags & MSG_DONTWAIT,
1848 m->msg_flags & MSG_MORE);
1849 tun_put(tun);
1850 return ret;
1851 }
1852
1853 static int tun_recvmsg(struct socket *sock, struct msghdr *m, size_t total_len,
1854 int flags)
1855 {
1856 struct tun_file *tfile = container_of(sock, struct tun_file, socket);
1857 struct tun_struct *tun = __tun_get(tfile);
1858 struct sk_buff *skb = m->msg_control;
1859 int ret;
1860
1861 if (!tun) {
1862 ret = -EBADFD;
1863 goto out_free_skb;
1864 }
1865
1866 if (flags & ~(MSG_DONTWAIT|MSG_TRUNC|MSG_ERRQUEUE)) {
1867 ret = -EINVAL;
1868 goto out_put_tun;
1869 }
1870 if (flags & MSG_ERRQUEUE) {
1871 ret = sock_recv_errqueue(sock->sk, m, total_len,
1872 SOL_PACKET, TUN_TX_TIMESTAMP);
1873 goto out;
1874 }
1875 ret = tun_do_read(tun, tfile, &m->msg_iter, flags & MSG_DONTWAIT, skb);
1876 if (ret > (ssize_t)total_len) {
1877 m->msg_flags |= MSG_TRUNC;
1878 ret = flags & MSG_TRUNC ? ret : total_len;
1879 }
1880 out:
1881 tun_put(tun);
1882 return ret;
1883
1884 out_put_tun:
1885 tun_put(tun);
1886 out_free_skb:
1887 if (skb)
1888 kfree_skb(skb);
1889 return ret;
1890 }
1891
1892 static int tun_peek_len(struct socket *sock)
1893 {
1894 struct tun_file *tfile = container_of(sock, struct tun_file, socket);
1895 struct tun_struct *tun;
1896 int ret = 0;
1897
1898 tun = __tun_get(tfile);
1899 if (!tun)
1900 return 0;
1901
1902 ret = skb_array_peek_len(&tfile->tx_array);
1903 tun_put(tun);
1904
1905 return ret;
1906 }
1907
1908 /* Ops structure to mimic raw sockets with tun */
1909 static const struct proto_ops tun_socket_ops = {
1910 .peek_len = tun_peek_len,
1911 .sendmsg = tun_sendmsg,
1912 .recvmsg = tun_recvmsg,
1913 };
1914
1915 static struct proto tun_proto = {
1916 .name = "tun",
1917 .owner = THIS_MODULE,
1918 .obj_size = sizeof(struct tun_file),
1919 };
1920
1921 static int tun_flags(struct tun_struct *tun)
1922 {
1923 return tun->flags & (TUN_FEATURES | IFF_PERSIST | IFF_TUN | IFF_TAP);
1924 }
1925
1926 static ssize_t tun_show_flags(struct device *dev, struct device_attribute *attr,
1927 char *buf)
1928 {
1929 struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1930 return sprintf(buf, "0x%x\n", tun_flags(tun));
1931 }
1932
1933 static ssize_t tun_show_owner(struct device *dev, struct device_attribute *attr,
1934 char *buf)
1935 {
1936 struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1937 return uid_valid(tun->owner)?
1938 sprintf(buf, "%u\n",
1939 from_kuid_munged(current_user_ns(), tun->owner)):
1940 sprintf(buf, "-1\n");
1941 }
1942
1943 static ssize_t tun_show_group(struct device *dev, struct device_attribute *attr,
1944 char *buf)
1945 {
1946 struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1947 return gid_valid(tun->group) ?
1948 sprintf(buf, "%u\n",
1949 from_kgid_munged(current_user_ns(), tun->group)):
1950 sprintf(buf, "-1\n");
1951 }
1952
1953 static DEVICE_ATTR(tun_flags, 0444, tun_show_flags, NULL);
1954 static DEVICE_ATTR(owner, 0444, tun_show_owner, NULL);
1955 static DEVICE_ATTR(group, 0444, tun_show_group, NULL);
1956
1957 static struct attribute *tun_dev_attrs[] = {
1958 &dev_attr_tun_flags.attr,
1959 &dev_attr_owner.attr,
1960 &dev_attr_group.attr,
1961 NULL
1962 };
1963
1964 static const struct attribute_group tun_attr_group = {
1965 .attrs = tun_dev_attrs
1966 };
1967
1968 static int tun_set_iff(struct net *net, struct file *file, struct ifreq *ifr)
1969 {
1970 struct tun_struct *tun;
1971 struct tun_file *tfile = file->private_data;
1972 struct net_device *dev;
1973 int err;
1974
1975 if (tfile->detached)
1976 return -EINVAL;
1977
1978 dev = __dev_get_by_name(net, ifr->ifr_name);
1979 if (dev) {
1980 if (ifr->ifr_flags & IFF_TUN_EXCL)
1981 return -EBUSY;
1982 if ((ifr->ifr_flags & IFF_TUN) && dev->netdev_ops == &tun_netdev_ops)
1983 tun = netdev_priv(dev);
1984 else if ((ifr->ifr_flags & IFF_TAP) && dev->netdev_ops == &tap_netdev_ops)
1985 tun = netdev_priv(dev);
1986 else
1987 return -EINVAL;
1988
1989 if (!!(ifr->ifr_flags & IFF_MULTI_QUEUE) !=
1990 !!(tun->flags & IFF_MULTI_QUEUE))
1991 return -EINVAL;
1992
1993 if (tun_not_capable(tun))
1994 return -EPERM;
1995 err = security_tun_dev_open(tun->security);
1996 if (err < 0)
1997 return err;
1998
1999 err = tun_attach(tun, file, ifr->ifr_flags & IFF_NOFILTER);
2000 if (err < 0)
2001 return err;
2002
2003 if (tun->flags & IFF_MULTI_QUEUE &&
2004 (tun->numqueues + tun->numdisabled > 1)) {
2005 /* One or more queue has already been attached, no need
2006 * to initialize the device again.
2007 */
2008 return 0;
2009 }
2010 }
2011 else {
2012 char *name;
2013 unsigned long flags = 0;
2014 int queues = ifr->ifr_flags & IFF_MULTI_QUEUE ?
2015 MAX_TAP_QUEUES : 1;
2016
2017 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
2018 return -EPERM;
2019 err = security_tun_dev_create();
2020 if (err < 0)
2021 return err;
2022
2023 /* Set dev type */
2024 if (ifr->ifr_flags & IFF_TUN) {
2025 /* TUN device */
2026 flags |= IFF_TUN;
2027 name = "tun%d";
2028 } else if (ifr->ifr_flags & IFF_TAP) {
2029 /* TAP device */
2030 flags |= IFF_TAP;
2031 name = "tap%d";
2032 } else
2033 return -EINVAL;
2034
2035 if (*ifr->ifr_name)
2036 name = ifr->ifr_name;
2037
2038 dev = alloc_netdev_mqs(sizeof(struct tun_struct), name,
2039 NET_NAME_UNKNOWN, tun_setup, queues,
2040 queues);
2041
2042 if (!dev)
2043 return -ENOMEM;
2044 err = dev_get_valid_name(net, dev, name);
2045 if (err < 0)
2046 goto err_free_dev;
2047
2048 dev_net_set(dev, net);
2049 dev->rtnl_link_ops = &tun_link_ops;
2050 dev->ifindex = tfile->ifindex;
2051 dev->sysfs_groups[0] = &tun_attr_group;
2052
2053 tun = netdev_priv(dev);
2054 tun->dev = dev;
2055 tun->flags = flags;
2056 tun->txflt.count = 0;
2057 tun->vnet_hdr_sz = sizeof(struct virtio_net_hdr);
2058
2059 tun->align = NET_SKB_PAD;
2060 tun->filter_attached = false;
2061 tun->sndbuf = tfile->socket.sk->sk_sndbuf;
2062 tun->rx_batched = 0;
2063
2064 tun->pcpu_stats = netdev_alloc_pcpu_stats(struct tun_pcpu_stats);
2065 if (!tun->pcpu_stats) {
2066 err = -ENOMEM;
2067 goto err_free_dev;
2068 }
2069
2070 spin_lock_init(&tun->lock);
2071
2072 err = security_tun_dev_alloc_security(&tun->security);
2073 if (err < 0)
2074 goto err_free_stat;
2075
2076 tun_net_init(dev);
2077 tun_flow_init(tun);
2078
2079 dev->hw_features = NETIF_F_SG | NETIF_F_FRAGLIST |
2080 TUN_USER_FEATURES | NETIF_F_HW_VLAN_CTAG_TX |
2081 NETIF_F_HW_VLAN_STAG_TX;
2082 dev->features = dev->hw_features | NETIF_F_LLTX;
2083 dev->vlan_features = dev->features &
2084 ~(NETIF_F_HW_VLAN_CTAG_TX |
2085 NETIF_F_HW_VLAN_STAG_TX);
2086
2087 INIT_LIST_HEAD(&tun->disabled);
2088 err = tun_attach(tun, file, false);
2089 if (err < 0)
2090 goto err_free_flow;
2091
2092 err = register_netdevice(tun->dev);
2093 if (err < 0)
2094 goto err_detach;
2095 }
2096
2097 netif_carrier_on(tun->dev);
2098
2099 tun_debug(KERN_INFO, tun, "tun_set_iff\n");
2100
2101 tun->flags = (tun->flags & ~TUN_FEATURES) |
2102 (ifr->ifr_flags & TUN_FEATURES);
2103
2104 /* Make sure persistent devices do not get stuck in
2105 * xoff state.
2106 */
2107 if (netif_running(tun->dev))
2108 netif_tx_wake_all_queues(tun->dev);
2109
2110 strcpy(ifr->ifr_name, tun->dev->name);
2111 return 0;
2112
2113 err_detach:
2114 tun_detach_all(dev);
2115 /* register_netdevice() already called tun_free_netdev() */
2116 goto err_free_dev;
2117
2118 err_free_flow:
2119 tun_flow_uninit(tun);
2120 security_tun_dev_free_security(tun->security);
2121 err_free_stat:
2122 free_percpu(tun->pcpu_stats);
2123 err_free_dev:
2124 free_netdev(dev);
2125 return err;
2126 }
2127
2128 static void tun_get_iff(struct net *net, struct tun_struct *tun,
2129 struct ifreq *ifr)
2130 {
2131 tun_debug(KERN_INFO, tun, "tun_get_iff\n");
2132
2133 strcpy(ifr->ifr_name, tun->dev->name);
2134
2135 ifr->ifr_flags = tun_flags(tun);
2136
2137 }
2138
2139 /* This is like a cut-down ethtool ops, except done via tun fd so no
2140 * privs required. */
2141 static int set_offload(struct tun_struct *tun, unsigned long arg)
2142 {
2143 netdev_features_t features = 0;
2144
2145 if (arg & TUN_F_CSUM) {
2146 features |= NETIF_F_HW_CSUM;
2147 arg &= ~TUN_F_CSUM;
2148
2149 if (arg & (TUN_F_TSO4|TUN_F_TSO6)) {
2150 if (arg & TUN_F_TSO_ECN) {
2151 features |= NETIF_F_TSO_ECN;
2152 arg &= ~TUN_F_TSO_ECN;
2153 }
2154 if (arg & TUN_F_TSO4)
2155 features |= NETIF_F_TSO;
2156 if (arg & TUN_F_TSO6)
2157 features |= NETIF_F_TSO6;
2158 arg &= ~(TUN_F_TSO4|TUN_F_TSO6);
2159 }
2160 }
2161
2162 /* This gives the user a way to test for new features in future by
2163 * trying to set them. */
2164 if (arg)
2165 return -EINVAL;
2166
2167 tun->set_features = features;
2168 tun->dev->wanted_features &= ~TUN_USER_FEATURES;
2169 tun->dev->wanted_features |= features;
2170 netdev_update_features(tun->dev);
2171
2172 return 0;
2173 }
2174
2175 static void tun_detach_filter(struct tun_struct *tun, int n)
2176 {
2177 int i;
2178 struct tun_file *tfile;
2179
2180 for (i = 0; i < n; i++) {
2181 tfile = rtnl_dereference(tun->tfiles[i]);
2182 lock_sock(tfile->socket.sk);
2183 sk_detach_filter(tfile->socket.sk);
2184 release_sock(tfile->socket.sk);
2185 }
2186
2187 tun->filter_attached = false;
2188 }
2189
2190 static int tun_attach_filter(struct tun_struct *tun)
2191 {
2192 int i, ret = 0;
2193 struct tun_file *tfile;
2194
2195 for (i = 0; i < tun->numqueues; i++) {
2196 tfile = rtnl_dereference(tun->tfiles[i]);
2197 lock_sock(tfile->socket.sk);
2198 ret = sk_attach_filter(&tun->fprog, tfile->socket.sk);
2199 release_sock(tfile->socket.sk);
2200 if (ret) {
2201 tun_detach_filter(tun, i);
2202 return ret;
2203 }
2204 }
2205
2206 tun->filter_attached = true;
2207 return ret;
2208 }
2209
2210 static void tun_set_sndbuf(struct tun_struct *tun)
2211 {
2212 struct tun_file *tfile;
2213 int i;
2214
2215 for (i = 0; i < tun->numqueues; i++) {
2216 tfile = rtnl_dereference(tun->tfiles[i]);
2217 tfile->socket.sk->sk_sndbuf = tun->sndbuf;
2218 }
2219 }
2220
2221 static int tun_set_queue(struct file *file, struct ifreq *ifr)
2222 {
2223 struct tun_file *tfile = file->private_data;
2224 struct tun_struct *tun;
2225 int ret = 0;
2226
2227 rtnl_lock();
2228
2229 if (ifr->ifr_flags & IFF_ATTACH_QUEUE) {
2230 tun = tfile->detached;
2231 if (!tun) {
2232 ret = -EINVAL;
2233 goto unlock;
2234 }
2235 ret = security_tun_dev_attach_queue(tun->security);
2236 if (ret < 0)
2237 goto unlock;
2238 ret = tun_attach(tun, file, false);
2239 } else if (ifr->ifr_flags & IFF_DETACH_QUEUE) {
2240 tun = rtnl_dereference(tfile->tun);
2241 if (!tun || !(tun->flags & IFF_MULTI_QUEUE) || tfile->detached)
2242 ret = -EINVAL;
2243 else
2244 __tun_detach(tfile, false);
2245 } else
2246 ret = -EINVAL;
2247
2248 unlock:
2249 rtnl_unlock();
2250 return ret;
2251 }
2252
2253 static long __tun_chr_ioctl(struct file *file, unsigned int cmd,
2254 unsigned long arg, int ifreq_len)
2255 {
2256 struct tun_file *tfile = file->private_data;
2257 struct tun_struct *tun;
2258 void __user* argp = (void __user*)arg;
2259 struct ifreq ifr;
2260 kuid_t owner;
2261 kgid_t group;
2262 int sndbuf;
2263 int vnet_hdr_sz;
2264 unsigned int ifindex;
2265 int le;
2266 int ret;
2267
2268 if (cmd == TUNSETIFF || cmd == TUNSETQUEUE || _IOC_TYPE(cmd) == SOCK_IOC_TYPE) {
2269 if (copy_from_user(&ifr, argp, ifreq_len))
2270 return -EFAULT;
2271 } else {
2272 memset(&ifr, 0, sizeof(ifr));
2273 }
2274 if (cmd == TUNGETFEATURES) {
2275 /* Currently this just means: "what IFF flags are valid?".
2276 * This is needed because we never checked for invalid flags on
2277 * TUNSETIFF.
2278 */
2279 return put_user(IFF_TUN | IFF_TAP | TUN_FEATURES,
2280 (unsigned int __user*)argp);
2281 } else if (cmd == TUNSETQUEUE)
2282 return tun_set_queue(file, &ifr);
2283
2284 ret = 0;
2285 rtnl_lock();
2286
2287 tun = __tun_get(tfile);
2288 if (cmd == TUNSETIFF) {
2289 ret = -EEXIST;
2290 if (tun)
2291 goto unlock;
2292
2293 ifr.ifr_name[IFNAMSIZ-1] = '\0';
2294
2295 ret = tun_set_iff(sock_net(&tfile->sk), file, &ifr);
2296
2297 if (ret)
2298 goto unlock;
2299
2300 if (copy_to_user(argp, &ifr, ifreq_len))
2301 ret = -EFAULT;
2302 goto unlock;
2303 }
2304 if (cmd == TUNSETIFINDEX) {
2305 ret = -EPERM;
2306 if (tun)
2307 goto unlock;
2308
2309 ret = -EFAULT;
2310 if (copy_from_user(&ifindex, argp, sizeof(ifindex)))
2311 goto unlock;
2312
2313 ret = 0;
2314 tfile->ifindex = ifindex;
2315 goto unlock;
2316 }
2317
2318 ret = -EBADFD;
2319 if (!tun)
2320 goto unlock;
2321
2322 tun_debug(KERN_INFO, tun, "tun_chr_ioctl cmd %u\n", cmd);
2323
2324 ret = 0;
2325 switch (cmd) {
2326 case TUNGETIFF:
2327 tun_get_iff(current->nsproxy->net_ns, tun, &ifr);
2328
2329 if (tfile->detached)
2330 ifr.ifr_flags |= IFF_DETACH_QUEUE;
2331 if (!tfile->socket.sk->sk_filter)
2332 ifr.ifr_flags |= IFF_NOFILTER;
2333
2334 if (copy_to_user(argp, &ifr, ifreq_len))
2335 ret = -EFAULT;
2336 break;
2337
2338 case TUNSETNOCSUM:
2339 /* Disable/Enable checksum */
2340
2341 /* [unimplemented] */
2342 tun_debug(KERN_INFO, tun, "ignored: set checksum %s\n",
2343 arg ? "disabled" : "enabled");
2344 break;
2345
2346 case TUNSETPERSIST:
2347 /* Disable/Enable persist mode. Keep an extra reference to the
2348 * module to prevent the module being unprobed.
2349 */
2350 if (arg && !(tun->flags & IFF_PERSIST)) {
2351 tun->flags |= IFF_PERSIST;
2352 __module_get(THIS_MODULE);
2353 }
2354 if (!arg && (tun->flags & IFF_PERSIST)) {
2355 tun->flags &= ~IFF_PERSIST;
2356 module_put(THIS_MODULE);
2357 }
2358
2359 tun_debug(KERN_INFO, tun, "persist %s\n",
2360 arg ? "enabled" : "disabled");
2361 break;
2362
2363 case TUNSETOWNER:
2364 /* Set owner of the device */
2365 owner = make_kuid(current_user_ns(), arg);
2366 if (!uid_valid(owner)) {
2367 ret = -EINVAL;
2368 break;
2369 }
2370 tun->owner = owner;
2371 tun_debug(KERN_INFO, tun, "owner set to %u\n",
2372 from_kuid(&init_user_ns, tun->owner));
2373 break;
2374
2375 case TUNSETGROUP:
2376 /* Set group of the device */
2377 group = make_kgid(current_user_ns(), arg);
2378 if (!gid_valid(group)) {
2379 ret = -EINVAL;
2380 break;
2381 }
2382 tun->group = group;
2383 tun_debug(KERN_INFO, tun, "group set to %u\n",
2384 from_kgid(&init_user_ns, tun->group));
2385 break;
2386
2387 case TUNSETLINK:
2388 /* Only allow setting the type when the interface is down */
2389 if (tun->dev->flags & IFF_UP) {
2390 tun_debug(KERN_INFO, tun,
2391 "Linktype set failed because interface is up\n");
2392 ret = -EBUSY;
2393 } else {
2394 tun->dev->type = (int) arg;
2395 tun_debug(KERN_INFO, tun, "linktype set to %d\n",
2396 tun->dev->type);
2397 ret = 0;
2398 }
2399 break;
2400
2401 #ifdef TUN_DEBUG
2402 case TUNSETDEBUG:
2403 tun->debug = arg;
2404 break;
2405 #endif
2406 case TUNSETOFFLOAD:
2407 ret = set_offload(tun, arg);
2408 break;
2409
2410 case TUNSETTXFILTER:
2411 /* Can be set only for TAPs */
2412 ret = -EINVAL;
2413 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
2414 break;
2415 ret = update_filter(&tun->txflt, (void __user *)arg);
2416 break;
2417
2418 case SIOCGIFHWADDR:
2419 /* Get hw address */
2420 memcpy(ifr.ifr_hwaddr.sa_data, tun->dev->dev_addr, ETH_ALEN);
2421 ifr.ifr_hwaddr.sa_family = tun->dev->type;
2422 if (copy_to_user(argp, &ifr, ifreq_len))
2423 ret = -EFAULT;
2424 break;
2425
2426 case SIOCSIFHWADDR:
2427 /* Set hw address */
2428 tun_debug(KERN_DEBUG, tun, "set hw address: %pM\n",
2429 ifr.ifr_hwaddr.sa_data);
2430
2431 ret = dev_set_mac_address(tun->dev, &ifr.ifr_hwaddr);
2432 break;
2433
2434 case TUNGETSNDBUF:
2435 sndbuf = tfile->socket.sk->sk_sndbuf;
2436 if (copy_to_user(argp, &sndbuf, sizeof(sndbuf)))
2437 ret = -EFAULT;
2438 break;
2439
2440 case TUNSETSNDBUF:
2441 if (copy_from_user(&sndbuf, argp, sizeof(sndbuf))) {
2442 ret = -EFAULT;
2443 break;
2444 }
2445 if (sndbuf <= 0) {
2446 ret = -EINVAL;
2447 break;
2448 }
2449
2450 tun->sndbuf = sndbuf;
2451 tun_set_sndbuf(tun);
2452 break;
2453
2454 case TUNGETVNETHDRSZ:
2455 vnet_hdr_sz = tun->vnet_hdr_sz;
2456 if (copy_to_user(argp, &vnet_hdr_sz, sizeof(vnet_hdr_sz)))
2457 ret = -EFAULT;
2458 break;
2459
2460 case TUNSETVNETHDRSZ:
2461 if (copy_from_user(&vnet_hdr_sz, argp, sizeof(vnet_hdr_sz))) {
2462 ret = -EFAULT;
2463 break;
2464 }
2465 if (vnet_hdr_sz < (int)sizeof(struct virtio_net_hdr)) {
2466 ret = -EINVAL;
2467 break;
2468 }
2469
2470 tun->vnet_hdr_sz = vnet_hdr_sz;
2471 break;
2472
2473 case TUNGETVNETLE:
2474 le = !!(tun->flags & TUN_VNET_LE);
2475 if (put_user(le, (int __user *)argp))
2476 ret = -EFAULT;
2477 break;
2478
2479 case TUNSETVNETLE:
2480 if (get_user(le, (int __user *)argp)) {
2481 ret = -EFAULT;
2482 break;
2483 }
2484 if (le)
2485 tun->flags |= TUN_VNET_LE;
2486 else
2487 tun->flags &= ~TUN_VNET_LE;
2488 break;
2489
2490 case TUNGETVNETBE:
2491 ret = tun_get_vnet_be(tun, argp);
2492 break;
2493
2494 case TUNSETVNETBE:
2495 ret = tun_set_vnet_be(tun, argp);
2496 break;
2497
2498 case TUNATTACHFILTER:
2499 /* Can be set only for TAPs */
2500 ret = -EINVAL;
2501 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
2502 break;
2503 ret = -EFAULT;
2504 if (copy_from_user(&tun->fprog, argp, sizeof(tun->fprog)))
2505 break;
2506
2507 ret = tun_attach_filter(tun);
2508 break;
2509
2510 case TUNDETACHFILTER:
2511 /* Can be set only for TAPs */
2512 ret = -EINVAL;
2513 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
2514 break;
2515 ret = 0;
2516 tun_detach_filter(tun, tun->numqueues);
2517 break;
2518
2519 case TUNGETFILTER:
2520 ret = -EINVAL;
2521 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
2522 break;
2523 ret = -EFAULT;
2524 if (copy_to_user(argp, &tun->fprog, sizeof(tun->fprog)))
2525 break;
2526 ret = 0;
2527 break;
2528
2529 default:
2530 ret = -EINVAL;
2531 break;
2532 }
2533
2534 unlock:
2535 rtnl_unlock();
2536 if (tun)
2537 tun_put(tun);
2538 return ret;
2539 }
2540
2541 static long tun_chr_ioctl(struct file *file,
2542 unsigned int cmd, unsigned long arg)
2543 {
2544 return __tun_chr_ioctl(file, cmd, arg, sizeof (struct ifreq));
2545 }
2546
2547 #ifdef CONFIG_COMPAT
2548 static long tun_chr_compat_ioctl(struct file *file,
2549 unsigned int cmd, unsigned long arg)
2550 {
2551 switch (cmd) {
2552 case TUNSETIFF:
2553 case TUNGETIFF:
2554 case TUNSETTXFILTER:
2555 case TUNGETSNDBUF:
2556 case TUNSETSNDBUF:
2557 case SIOCGIFHWADDR:
2558 case SIOCSIFHWADDR:
2559 arg = (unsigned long)compat_ptr(arg);
2560 break;
2561 default:
2562 arg = (compat_ulong_t)arg;
2563 break;
2564 }
2565
2566 /*
2567 * compat_ifreq is shorter than ifreq, so we must not access beyond
2568 * the end of that structure. All fields that are used in this
2569 * driver are compatible though, we don't need to convert the
2570 * contents.
2571 */
2572 return __tun_chr_ioctl(file, cmd, arg, sizeof(struct compat_ifreq));
2573 }
2574 #endif /* CONFIG_COMPAT */
2575
2576 static int tun_chr_fasync(int fd, struct file *file, int on)
2577 {
2578 struct tun_file *tfile = file->private_data;
2579 int ret;
2580
2581 if ((ret = fasync_helper(fd, file, on, &tfile->fasync)) < 0)
2582 goto out;
2583
2584 if (on) {
2585 __f_setown(file, task_pid(current), PIDTYPE_PID, 0);
2586 tfile->flags |= TUN_FASYNC;
2587 } else
2588 tfile->flags &= ~TUN_FASYNC;
2589 ret = 0;
2590 out:
2591 return ret;
2592 }
2593
2594 static int tun_chr_open(struct inode *inode, struct file * file)
2595 {
2596 struct net *net = current->nsproxy->net_ns;
2597 struct tun_file *tfile;
2598
2599 DBG1(KERN_INFO, "tunX: tun_chr_open\n");
2600
2601 tfile = (struct tun_file *)sk_alloc(net, AF_UNSPEC, GFP_KERNEL,
2602 &tun_proto, 0);
2603 if (!tfile)
2604 return -ENOMEM;
2605 RCU_INIT_POINTER(tfile->tun, NULL);
2606 tfile->flags = 0;
2607 tfile->ifindex = 0;
2608
2609 init_waitqueue_head(&tfile->wq.wait);
2610 RCU_INIT_POINTER(tfile->socket.wq, &tfile->wq);
2611
2612 tfile->socket.file = file;
2613 tfile->socket.ops = &tun_socket_ops;
2614
2615 sock_init_data(&tfile->socket, &tfile->sk);
2616
2617 tfile->sk.sk_write_space = tun_sock_write_space;
2618 tfile->sk.sk_sndbuf = INT_MAX;
2619
2620 file->private_data = tfile;
2621 INIT_LIST_HEAD(&tfile->next);
2622
2623 sock_set_flag(&tfile->sk, SOCK_ZEROCOPY);
2624
2625 return 0;
2626 }
2627
2628 static int tun_chr_close(struct inode *inode, struct file *file)
2629 {
2630 struct tun_file *tfile = file->private_data;
2631
2632 tun_detach(tfile, true);
2633
2634 return 0;
2635 }
2636
2637 #ifdef CONFIG_PROC_FS
2638 static void tun_chr_show_fdinfo(struct seq_file *m, struct file *f)
2639 {
2640 struct tun_struct *tun;
2641 struct ifreq ifr;
2642
2643 memset(&ifr, 0, sizeof(ifr));
2644
2645 rtnl_lock();
2646 tun = tun_get(f);
2647 if (tun)
2648 tun_get_iff(current->nsproxy->net_ns, tun, &ifr);
2649 rtnl_unlock();
2650
2651 if (tun)
2652 tun_put(tun);
2653
2654 seq_printf(m, "iff:\t%s\n", ifr.ifr_name);
2655 }
2656 #endif
2657
2658 static const struct file_operations tun_fops = {
2659 .owner = THIS_MODULE,
2660 .llseek = no_llseek,
2661 .read_iter = tun_chr_read_iter,
2662 .write_iter = tun_chr_write_iter,
2663 .poll = tun_chr_poll,
2664 .unlocked_ioctl = tun_chr_ioctl,
2665 #ifdef CONFIG_COMPAT
2666 .compat_ioctl = tun_chr_compat_ioctl,
2667 #endif
2668 .open = tun_chr_open,
2669 .release = tun_chr_close,
2670 .fasync = tun_chr_fasync,
2671 #ifdef CONFIG_PROC_FS
2672 .show_fdinfo = tun_chr_show_fdinfo,
2673 #endif
2674 };
2675
2676 static struct miscdevice tun_miscdev = {
2677 .minor = TUN_MINOR,
2678 .name = "tun",
2679 .nodename = "net/tun",
2680 .fops = &tun_fops,
2681 };
2682
2683 /* ethtool interface */
2684
2685 static int tun_get_link_ksettings(struct net_device *dev,
2686 struct ethtool_link_ksettings *cmd)
2687 {
2688 ethtool_link_ksettings_zero_link_mode(cmd, supported);
2689 ethtool_link_ksettings_zero_link_mode(cmd, advertising);
2690 cmd->base.speed = SPEED_10;
2691 cmd->base.duplex = DUPLEX_FULL;
2692 cmd->base.port = PORT_TP;
2693 cmd->base.phy_address = 0;
2694 cmd->base.autoneg = AUTONEG_DISABLE;
2695 return 0;
2696 }
2697
2698 static void tun_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
2699 {
2700 struct tun_struct *tun = netdev_priv(dev);
2701
2702 strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
2703 strlcpy(info->version, DRV_VERSION, sizeof(info->version));
2704
2705 switch (tun->flags & TUN_TYPE_MASK) {
2706 case IFF_TUN:
2707 strlcpy(info->bus_info, "tun", sizeof(info->bus_info));
2708 break;
2709 case IFF_TAP:
2710 strlcpy(info->bus_info, "tap", sizeof(info->bus_info));
2711 break;
2712 }
2713 }
2714
2715 static u32 tun_get_msglevel(struct net_device *dev)
2716 {
2717 #ifdef TUN_DEBUG
2718 struct tun_struct *tun = netdev_priv(dev);
2719 return tun->debug;
2720 #else
2721 return -EOPNOTSUPP;
2722 #endif
2723 }
2724
2725 static void tun_set_msglevel(struct net_device *dev, u32 value)
2726 {
2727 #ifdef TUN_DEBUG
2728 struct tun_struct *tun = netdev_priv(dev);
2729 tun->debug = value;
2730 #endif
2731 }
2732
2733 static int tun_get_coalesce(struct net_device *dev,
2734 struct ethtool_coalesce *ec)
2735 {
2736 struct tun_struct *tun = netdev_priv(dev);
2737
2738 ec->rx_max_coalesced_frames = tun->rx_batched;
2739
2740 return 0;
2741 }
2742
2743 static int tun_set_coalesce(struct net_device *dev,
2744 struct ethtool_coalesce *ec)
2745 {
2746 struct tun_struct *tun = netdev_priv(dev);
2747
2748 if (ec->rx_max_coalesced_frames > NAPI_POLL_WEIGHT)
2749 tun->rx_batched = NAPI_POLL_WEIGHT;
2750 else
2751 tun->rx_batched = ec->rx_max_coalesced_frames;
2752
2753 return 0;
2754 }
2755
2756 static const struct ethtool_ops tun_ethtool_ops = {
2757 .get_drvinfo = tun_get_drvinfo,
2758 .get_msglevel = tun_get_msglevel,
2759 .set_msglevel = tun_set_msglevel,
2760 .get_link = ethtool_op_get_link,
2761 .get_ts_info = ethtool_op_get_ts_info,
2762 .get_coalesce = tun_get_coalesce,
2763 .set_coalesce = tun_set_coalesce,
2764 .get_link_ksettings = tun_get_link_ksettings,
2765 };
2766
2767 static int tun_queue_resize(struct tun_struct *tun)
2768 {
2769 struct net_device *dev = tun->dev;
2770 struct tun_file *tfile;
2771 struct skb_array **arrays;
2772 int n = tun->numqueues + tun->numdisabled;
2773 int ret, i;
2774
2775 arrays = kmalloc_array(n, sizeof(*arrays), GFP_KERNEL);
2776 if (!arrays)
2777 return -ENOMEM;
2778
2779 for (i = 0; i < tun->numqueues; i++) {
2780 tfile = rtnl_dereference(tun->tfiles[i]);
2781 arrays[i] = &tfile->tx_array;
2782 }
2783 list_for_each_entry(tfile, &tun->disabled, next)
2784 arrays[i++] = &tfile->tx_array;
2785
2786 ret = skb_array_resize_multiple(arrays, n,
2787 dev->tx_queue_len, GFP_KERNEL);
2788
2789 kfree(arrays);
2790 return ret;
2791 }
2792
2793 static int tun_device_event(struct notifier_block *unused,
2794 unsigned long event, void *ptr)
2795 {
2796 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
2797 struct tun_struct *tun = netdev_priv(dev);
2798
2799 if (dev->rtnl_link_ops != &tun_link_ops)
2800 return NOTIFY_DONE;
2801
2802 switch (event) {
2803 case NETDEV_CHANGE_TX_QUEUE_LEN:
2804 if (tun_queue_resize(tun))
2805 return NOTIFY_BAD;
2806 break;
2807 default:
2808 break;
2809 }
2810
2811 return NOTIFY_DONE;
2812 }
2813
2814 static struct notifier_block tun_notifier_block __read_mostly = {
2815 .notifier_call = tun_device_event,
2816 };
2817
2818 static int __init tun_init(void)
2819 {
2820 int ret = 0;
2821
2822 pr_info("%s, %s\n", DRV_DESCRIPTION, DRV_VERSION);
2823
2824 ret = rtnl_link_register(&tun_link_ops);
2825 if (ret) {
2826 pr_err("Can't register link_ops\n");
2827 goto err_linkops;
2828 }
2829
2830 ret = misc_register(&tun_miscdev);
2831 if (ret) {
2832 pr_err("Can't register misc device %d\n", TUN_MINOR);
2833 goto err_misc;
2834 }
2835
2836 ret = register_netdevice_notifier(&tun_notifier_block);
2837 if (ret) {
2838 pr_err("Can't register netdevice notifier\n");
2839 goto err_notifier;
2840 }
2841
2842 return 0;
2843
2844 err_notifier:
2845 misc_deregister(&tun_miscdev);
2846 err_misc:
2847 rtnl_link_unregister(&tun_link_ops);
2848 err_linkops:
2849 return ret;
2850 }
2851
2852 static void tun_cleanup(void)
2853 {
2854 misc_deregister(&tun_miscdev);
2855 rtnl_link_unregister(&tun_link_ops);
2856 unregister_netdevice_notifier(&tun_notifier_block);
2857 }
2858
2859 /* Get an underlying socket object from tun file. Returns error unless file is
2860 * attached to a device. The returned object works like a packet socket, it
2861 * can be used for sock_sendmsg/sock_recvmsg. The caller is responsible for
2862 * holding a reference to the file for as long as the socket is in use. */
2863 struct socket *tun_get_socket(struct file *file)
2864 {
2865 struct tun_file *tfile;
2866 if (file->f_op != &tun_fops)
2867 return ERR_PTR(-EINVAL);
2868 tfile = file->private_data;
2869 if (!tfile)
2870 return ERR_PTR(-EBADFD);
2871 return &tfile->socket;
2872 }
2873 EXPORT_SYMBOL_GPL(tun_get_socket);
2874
2875 struct skb_array *tun_get_skb_array(struct file *file)
2876 {
2877 struct tun_file *tfile;
2878
2879 if (file->f_op != &tun_fops)
2880 return ERR_PTR(-EINVAL);
2881 tfile = file->private_data;
2882 if (!tfile)
2883 return ERR_PTR(-EBADFD);
2884 return &tfile->tx_array;
2885 }
2886 EXPORT_SYMBOL_GPL(tun_get_skb_array);
2887
2888 module_init(tun_init);
2889 module_exit(tun_cleanup);
2890 MODULE_DESCRIPTION(DRV_DESCRIPTION);
2891 MODULE_AUTHOR(DRV_COPYRIGHT);
2892 MODULE_LICENSE("GPL");
2893 MODULE_ALIAS_MISCDEV(TUN_MINOR);
2894 MODULE_ALIAS("devname:net/tun");
Arne's tree of linux kernel.