Loading...
1/*
2 * TUN - Universal TUN/TAP device driver.
3 * Copyright (C) 1999-2002 Maxim Krasnyansky <maxk@qualcomm.com>
4 *
5 * This program is free software; you can redistribute it and/or modify
6 * it under the terms of the GNU General Public License as published by
7 * the Free Software Foundation; either version 2 of the License, or
8 * (at your option) any later version.
9 *
10 * This program is distributed in the hope that it will be useful,
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
13 * GNU General Public License for more details.
14 *
15 * $Id: tun.c,v 1.15 2002/03/01 02:44:24 maxk Exp $
16 */
17
18/*
19 * Changes:
20 *
21 * Mike Kershaw <dragorn@kismetwireless.net> 2005/08/14
22 * Add TUNSETLINK ioctl to set the link encapsulation
23 *
24 * Mark Smith <markzzzsmith@yahoo.com.au>
25 * Use eth_random_addr() for tap MAC address.
26 *
27 * Harald Roelle <harald.roelle@ifi.lmu.de> 2004/04/20
28 * Fixes in packet dropping, queue length setting and queue wakeup.
29 * Increased default tx queue length.
30 * Added ethtool API.
31 * Minor cleanups
32 *
33 * Daniel Podlejski <underley@underley.eu.org>
34 * Modifications for 2.3.99-pre5 kernel.
35 */
36
37#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
38
39#define DRV_NAME "tun"
40#define DRV_VERSION "1.6"
41#define DRV_DESCRIPTION "Universal TUN/TAP device driver"
42#define DRV_COPYRIGHT "(C) 1999-2004 Max Krasnyansky <maxk@qualcomm.com>"
43
44#include <linux/module.h>
45#include <linux/errno.h>
46#include <linux/kernel.h>
47#include <linux/major.h>
48#include <linux/slab.h>
49#include <linux/poll.h>
50#include <linux/fcntl.h>
51#include <linux/init.h>
52#include <linux/skbuff.h>
53#include <linux/netdevice.h>
54#include <linux/etherdevice.h>
55#include <linux/miscdevice.h>
56#include <linux/ethtool.h>
57#include <linux/rtnetlink.h>
58#include <linux/compat.h>
59#include <linux/if.h>
60#include <linux/if_arp.h>
61#include <linux/if_ether.h>
62#include <linux/if_tun.h>
63#include <linux/if_vlan.h>
64#include <linux/crc32.h>
65#include <linux/nsproxy.h>
66#include <linux/virtio_net.h>
67#include <linux/rcupdate.h>
68#include <net/net_namespace.h>
69#include <net/netns/generic.h>
70#include <net/rtnetlink.h>
71#include <net/sock.h>
72#include <linux/seq_file.h>
73#include <linux/uio.h>
74
75#include <asm/uaccess.h>
76
77/* Uncomment to enable debugging */
78/* #define TUN_DEBUG 1 */
79
80#ifdef TUN_DEBUG
81static int debug;
82
83#define tun_debug(level, tun, fmt, args...) \
84do { \
85 if (tun->debug) \
86 netdev_printk(level, tun->dev, fmt, ##args); \
87} while (0)
88#define DBG1(level, fmt, args...) \
89do { \
90 if (debug == 2) \
91 printk(level fmt, ##args); \
92} while (0)
93#else
94#define tun_debug(level, tun, fmt, args...) \
95do { \
96 if (0) \
97 netdev_printk(level, tun->dev, fmt, ##args); \
98} while (0)
99#define DBG1(level, fmt, args...) \
100do { \
101 if (0) \
102 printk(level fmt, ##args); \
103} while (0)
104#endif
105
106/* TUN device flags */
107
108/* IFF_ATTACH_QUEUE is never stored in device flags,
109 * overload it to mean fasync when stored there.
110 */
111#define TUN_FASYNC IFF_ATTACH_QUEUE
112/* High bits in flags field are unused. */
113#define TUN_VNET_LE 0x80000000
114#define TUN_VNET_BE 0x40000000
115
116#define TUN_FEATURES (IFF_NO_PI | IFF_ONE_QUEUE | IFF_VNET_HDR | \
117 IFF_MULTI_QUEUE)
118#define GOODCOPY_LEN 128
119
120#define FLT_EXACT_COUNT 8
121struct tap_filter {
122 unsigned int count; /* Number of addrs. Zero means disabled */
123 u32 mask[2]; /* Mask of the hashed addrs */
124 unsigned char addr[FLT_EXACT_COUNT][ETH_ALEN];
125};
126
127/* MAX_TAP_QUEUES 256 is chosen to allow rx/tx queues to be equal
128 * to max number of VCPUs in guest. */
129#define MAX_TAP_QUEUES 256
130#define MAX_TAP_FLOWS 4096
131
132#define TUN_FLOW_EXPIRE (3 * HZ)
133
134/* A tun_file connects an open character device to a tuntap netdevice. It
135 * also contains all socket related structures (except sock_fprog and tap_filter)
136 * to serve as one transmit queue for tuntap device. The sock_fprog and
137 * tap_filter were kept in tun_struct since they were used for filtering for the
138 * netdevice not for a specific queue (at least I didn't see the requirement for
139 * this).
140 *
141 * RCU usage:
142 * The tun_file and tun_struct are loosely coupled, the pointer from one to the
143 * other can only be read while rcu_read_lock or rtnl_lock is held.
144 */
145struct tun_file {
146 struct sock sk;
147 struct socket socket;
148 struct socket_wq wq;
149 struct tun_struct __rcu *tun;
150 struct fasync_struct *fasync;
151 /* only used for fasnyc */
152 unsigned int flags;
153 union {
154 u16 queue_index;
155 unsigned int ifindex;
156 };
157 struct list_head next;
158 struct tun_struct *detached;
159};
160
161struct tun_flow_entry {
162 struct hlist_node hash_link;
163 struct rcu_head rcu;
164 struct tun_struct *tun;
165
166 u32 rxhash;
167 u32 rps_rxhash;
168 int queue_index;
169 unsigned long updated;
170};
171
172#define TUN_NUM_FLOW_ENTRIES 1024
173
174/* Since the socket were moved to tun_file, to preserve the behavior of persist
175 * device, socket filter, sndbuf and vnet header size were restore when the
176 * file were attached to a persist device.
177 */
178struct tun_struct {
179 struct tun_file __rcu *tfiles[MAX_TAP_QUEUES];
180 unsigned int numqueues;
181 unsigned int flags;
182 kuid_t owner;
183 kgid_t group;
184
185 struct net_device *dev;
186 netdev_features_t set_features;
187#define TUN_USER_FEATURES (NETIF_F_HW_CSUM|NETIF_F_TSO_ECN|NETIF_F_TSO| \
188 NETIF_F_TSO6|NETIF_F_UFO)
189
190 int align;
191 int vnet_hdr_sz;
192 int sndbuf;
193 struct tap_filter txflt;
194 struct sock_fprog fprog;
195 /* protected by rtnl lock */
196 bool filter_attached;
197#ifdef TUN_DEBUG
198 int debug;
199#endif
200 spinlock_t lock;
201 struct hlist_head flows[TUN_NUM_FLOW_ENTRIES];
202 struct timer_list flow_gc_timer;
203 unsigned long ageing_time;
204 unsigned int numdisabled;
205 struct list_head disabled;
206 void *security;
207 u32 flow_count;
208};
209
210#ifdef CONFIG_TUN_VNET_CROSS_LE
211static inline bool tun_legacy_is_little_endian(struct tun_struct *tun)
212{
213 return tun->flags & TUN_VNET_BE ? false :
214 virtio_legacy_is_little_endian();
215}
216
217static long tun_get_vnet_be(struct tun_struct *tun, int __user *argp)
218{
219 int be = !!(tun->flags & TUN_VNET_BE);
220
221 if (put_user(be, argp))
222 return -EFAULT;
223
224 return 0;
225}
226
227static long tun_set_vnet_be(struct tun_struct *tun, int __user *argp)
228{
229 int be;
230
231 if (get_user(be, argp))
232 return -EFAULT;
233
234 if (be)
235 tun->flags |= TUN_VNET_BE;
236 else
237 tun->flags &= ~TUN_VNET_BE;
238
239 return 0;
240}
241#else
242static inline bool tun_legacy_is_little_endian(struct tun_struct *tun)
243{
244 return virtio_legacy_is_little_endian();
245}
246
247static long tun_get_vnet_be(struct tun_struct *tun, int __user *argp)
248{
249 return -EINVAL;
250}
251
252static long tun_set_vnet_be(struct tun_struct *tun, int __user *argp)
253{
254 return -EINVAL;
255}
256#endif /* CONFIG_TUN_VNET_CROSS_LE */
257
258static inline bool tun_is_little_endian(struct tun_struct *tun)
259{
260 return tun->flags & TUN_VNET_LE ||
261 tun_legacy_is_little_endian(tun);
262}
263
264static inline u16 tun16_to_cpu(struct tun_struct *tun, __virtio16 val)
265{
266 return __virtio16_to_cpu(tun_is_little_endian(tun), val);
267}
268
269static inline __virtio16 cpu_to_tun16(struct tun_struct *tun, u16 val)
270{
271 return __cpu_to_virtio16(tun_is_little_endian(tun), val);
272}
273
274static inline u32 tun_hashfn(u32 rxhash)
275{
276 return rxhash & 0x3ff;
277}
278
279static struct tun_flow_entry *tun_flow_find(struct hlist_head *head, u32 rxhash)
280{
281 struct tun_flow_entry *e;
282
283 hlist_for_each_entry_rcu(e, head, hash_link) {
284 if (e->rxhash == rxhash)
285 return e;
286 }
287 return NULL;
288}
289
290static struct tun_flow_entry *tun_flow_create(struct tun_struct *tun,
291 struct hlist_head *head,
292 u32 rxhash, u16 queue_index)
293{
294 struct tun_flow_entry *e = kmalloc(sizeof(*e), GFP_ATOMIC);
295
296 if (e) {
297 tun_debug(KERN_INFO, tun, "create flow: hash %u index %u\n",
298 rxhash, queue_index);
299 e->updated = jiffies;
300 e->rxhash = rxhash;
301 e->rps_rxhash = 0;
302 e->queue_index = queue_index;
303 e->tun = tun;
304 hlist_add_head_rcu(&e->hash_link, head);
305 ++tun->flow_count;
306 }
307 return e;
308}
309
310static void tun_flow_delete(struct tun_struct *tun, struct tun_flow_entry *e)
311{
312 tun_debug(KERN_INFO, tun, "delete flow: hash %u index %u\n",
313 e->rxhash, e->queue_index);
314 hlist_del_rcu(&e->hash_link);
315 kfree_rcu(e, rcu);
316 --tun->flow_count;
317}
318
319static void tun_flow_flush(struct tun_struct *tun)
320{
321 int i;
322
323 spin_lock_bh(&tun->lock);
324 for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
325 struct tun_flow_entry *e;
326 struct hlist_node *n;
327
328 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link)
329 tun_flow_delete(tun, e);
330 }
331 spin_unlock_bh(&tun->lock);
332}
333
334static void tun_flow_delete_by_queue(struct tun_struct *tun, u16 queue_index)
335{
336 int i;
337
338 spin_lock_bh(&tun->lock);
339 for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
340 struct tun_flow_entry *e;
341 struct hlist_node *n;
342
343 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
344 if (e->queue_index == queue_index)
345 tun_flow_delete(tun, e);
346 }
347 }
348 spin_unlock_bh(&tun->lock);
349}
350
351static void tun_flow_cleanup(unsigned long data)
352{
353 struct tun_struct *tun = (struct tun_struct *)data;
354 unsigned long delay = tun->ageing_time;
355 unsigned long next_timer = jiffies + delay;
356 unsigned long count = 0;
357 int i;
358
359 tun_debug(KERN_INFO, tun, "tun_flow_cleanup\n");
360
361 spin_lock_bh(&tun->lock);
362 for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
363 struct tun_flow_entry *e;
364 struct hlist_node *n;
365
366 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
367 unsigned long this_timer;
368 count++;
369 this_timer = e->updated + delay;
370 if (time_before_eq(this_timer, jiffies))
371 tun_flow_delete(tun, e);
372 else if (time_before(this_timer, next_timer))
373 next_timer = this_timer;
374 }
375 }
376
377 if (count)
378 mod_timer(&tun->flow_gc_timer, round_jiffies_up(next_timer));
379 spin_unlock_bh(&tun->lock);
380}
381
382static void tun_flow_update(struct tun_struct *tun, u32 rxhash,
383 struct tun_file *tfile)
384{
385 struct hlist_head *head;
386 struct tun_flow_entry *e;
387 unsigned long delay = tun->ageing_time;
388 u16 queue_index = tfile->queue_index;
389
390 if (!rxhash)
391 return;
392 else
393 head = &tun->flows[tun_hashfn(rxhash)];
394
395 rcu_read_lock();
396
397 /* We may get a very small possibility of OOO during switching, not
398 * worth to optimize.*/
399 if (tun->numqueues == 1 || tfile->detached)
400 goto unlock;
401
402 e = tun_flow_find(head, rxhash);
403 if (likely(e)) {
404 /* TODO: keep queueing to old queue until it's empty? */
405 e->queue_index = queue_index;
406 e->updated = jiffies;
407 sock_rps_record_flow_hash(e->rps_rxhash);
408 } else {
409 spin_lock_bh(&tun->lock);
410 if (!tun_flow_find(head, rxhash) &&
411 tun->flow_count < MAX_TAP_FLOWS)
412 tun_flow_create(tun, head, rxhash, queue_index);
413
414 if (!timer_pending(&tun->flow_gc_timer))
415 mod_timer(&tun->flow_gc_timer,
416 round_jiffies_up(jiffies + delay));
417 spin_unlock_bh(&tun->lock);
418 }
419
420unlock:
421 rcu_read_unlock();
422}
423
424/**
425 * Save the hash received in the stack receive path and update the
426 * flow_hash table accordingly.
427 */
428static inline void tun_flow_save_rps_rxhash(struct tun_flow_entry *e, u32 hash)
429{
430 if (unlikely(e->rps_rxhash != hash))
431 e->rps_rxhash = hash;
432}
433
434/* We try to identify a flow through its rxhash first. The reason that
435 * we do not check rxq no. is because some cards(e.g 82599), chooses
436 * the rxq based on the txq where the last packet of the flow comes. As
437 * the userspace application move between processors, we may get a
438 * different rxq no. here. If we could not get rxhash, then we would
439 * hope the rxq no. may help here.
440 */
441static u16 tun_select_queue(struct net_device *dev, struct sk_buff *skb,
442 void *accel_priv, select_queue_fallback_t fallback)
443{
444 struct tun_struct *tun = netdev_priv(dev);
445 struct tun_flow_entry *e;
446 u32 txq = 0;
447 u32 numqueues = 0;
448
449 rcu_read_lock();
450 numqueues = ACCESS_ONCE(tun->numqueues);
451
452 txq = skb_get_hash(skb);
453 if (txq) {
454 e = tun_flow_find(&tun->flows[tun_hashfn(txq)], txq);
455 if (e) {
456 tun_flow_save_rps_rxhash(e, txq);
457 txq = e->queue_index;
458 } else
459 /* use multiply and shift instead of expensive divide */
460 txq = ((u64)txq * numqueues) >> 32;
461 } else if (likely(skb_rx_queue_recorded(skb))) {
462 txq = skb_get_rx_queue(skb);
463 while (unlikely(txq >= numqueues))
464 txq -= numqueues;
465 }
466
467 rcu_read_unlock();
468 return txq;
469}
470
471static inline bool tun_not_capable(struct tun_struct *tun)
472{
473 const struct cred *cred = current_cred();
474 struct net *net = dev_net(tun->dev);
475
476 return ((uid_valid(tun->owner) && !uid_eq(cred->euid, tun->owner)) ||
477 (gid_valid(tun->group) && !in_egroup_p(tun->group))) &&
478 !ns_capable(net->user_ns, CAP_NET_ADMIN);
479}
480
481static void tun_set_real_num_queues(struct tun_struct *tun)
482{
483 netif_set_real_num_tx_queues(tun->dev, tun->numqueues);
484 netif_set_real_num_rx_queues(tun->dev, tun->numqueues);
485}
486
487static void tun_disable_queue(struct tun_struct *tun, struct tun_file *tfile)
488{
489 tfile->detached = tun;
490 list_add_tail(&tfile->next, &tun->disabled);
491 ++tun->numdisabled;
492}
493
494static struct tun_struct *tun_enable_queue(struct tun_file *tfile)
495{
496 struct tun_struct *tun = tfile->detached;
497
498 tfile->detached = NULL;
499 list_del_init(&tfile->next);
500 --tun->numdisabled;
501 return tun;
502}
503
504static void tun_queue_purge(struct tun_file *tfile)
505{
506 skb_queue_purge(&tfile->sk.sk_receive_queue);
507 skb_queue_purge(&tfile->sk.sk_error_queue);
508}
509
510static void __tun_detach(struct tun_file *tfile, bool clean)
511{
512 struct tun_file *ntfile;
513 struct tun_struct *tun;
514
515 tun = rtnl_dereference(tfile->tun);
516
517 if (tun && !tfile->detached) {
518 u16 index = tfile->queue_index;
519 BUG_ON(index >= tun->numqueues);
520
521 rcu_assign_pointer(tun->tfiles[index],
522 tun->tfiles[tun->numqueues - 1]);
523 ntfile = rtnl_dereference(tun->tfiles[index]);
524 ntfile->queue_index = index;
525
526 --tun->numqueues;
527 if (clean) {
528 RCU_INIT_POINTER(tfile->tun, NULL);
529 sock_put(&tfile->sk);
530 } else
531 tun_disable_queue(tun, tfile);
532
533 synchronize_net();
534 tun_flow_delete_by_queue(tun, tun->numqueues + 1);
535 /* Drop read queue */
536 tun_queue_purge(tfile);
537 tun_set_real_num_queues(tun);
538 } else if (tfile->detached && clean) {
539 tun = tun_enable_queue(tfile);
540 sock_put(&tfile->sk);
541 }
542
543 if (clean) {
544 if (tun && tun->numqueues == 0 && tun->numdisabled == 0) {
545 netif_carrier_off(tun->dev);
546
547 if (!(tun->flags & IFF_PERSIST) &&
548 tun->dev->reg_state == NETREG_REGISTERED)
549 unregister_netdevice(tun->dev);
550 }
551 sock_put(&tfile->sk);
552 }
553}
554
555static void tun_detach(struct tun_file *tfile, bool clean)
556{
557 rtnl_lock();
558 __tun_detach(tfile, clean);
559 rtnl_unlock();
560}
561
562static void tun_detach_all(struct net_device *dev)
563{
564 struct tun_struct *tun = netdev_priv(dev);
565 struct tun_file *tfile, *tmp;
566 int i, n = tun->numqueues;
567
568 for (i = 0; i < n; i++) {
569 tfile = rtnl_dereference(tun->tfiles[i]);
570 BUG_ON(!tfile);
571 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
572 RCU_INIT_POINTER(tfile->tun, NULL);
573 --tun->numqueues;
574 }
575 list_for_each_entry(tfile, &tun->disabled, next) {
576 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
577 RCU_INIT_POINTER(tfile->tun, NULL);
578 }
579 BUG_ON(tun->numqueues != 0);
580
581 synchronize_net();
582 for (i = 0; i < n; i++) {
583 tfile = rtnl_dereference(tun->tfiles[i]);
584 /* Drop read queue */
585 tun_queue_purge(tfile);
586 sock_put(&tfile->sk);
587 }
588 list_for_each_entry_safe(tfile, tmp, &tun->disabled, next) {
589 tun_enable_queue(tfile);
590 tun_queue_purge(tfile);
591 sock_put(&tfile->sk);
592 }
593 BUG_ON(tun->numdisabled != 0);
594
595 if (tun->flags & IFF_PERSIST)
596 module_put(THIS_MODULE);
597}
598
599static int tun_attach(struct tun_struct *tun, struct file *file, bool skip_filter)
600{
601 struct tun_file *tfile = file->private_data;
602 int err;
603
604 err = security_tun_dev_attach(tfile->socket.sk, tun->security);
605 if (err < 0)
606 goto out;
607
608 err = -EINVAL;
609 if (rtnl_dereference(tfile->tun) && !tfile->detached)
610 goto out;
611
612 err = -EBUSY;
613 if (!(tun->flags & IFF_MULTI_QUEUE) && tun->numqueues == 1)
614 goto out;
615
616 err = -E2BIG;
617 if (!tfile->detached &&
618 tun->numqueues + tun->numdisabled == MAX_TAP_QUEUES)
619 goto out;
620
621 err = 0;
622
623 /* Re-attach the filter to persist device */
624 if (!skip_filter && (tun->filter_attached == true)) {
625 err = __sk_attach_filter(&tun->fprog, tfile->socket.sk,
626 lockdep_rtnl_is_held());
627 if (!err)
628 goto out;
629 }
630 tfile->queue_index = tun->numqueues;
631 rcu_assign_pointer(tfile->tun, tun);
632 rcu_assign_pointer(tun->tfiles[tun->numqueues], tfile);
633 tun->numqueues++;
634
635 if (tfile->detached)
636 tun_enable_queue(tfile);
637 else
638 sock_hold(&tfile->sk);
639
640 tun_set_real_num_queues(tun);
641
642 /* device is allowed to go away first, so no need to hold extra
643 * refcnt.
644 */
645
646out:
647 return err;
648}
649
650static struct tun_struct *__tun_get(struct tun_file *tfile)
651{
652 struct tun_struct *tun;
653
654 rcu_read_lock();
655 tun = rcu_dereference(tfile->tun);
656 if (tun)
657 dev_hold(tun->dev);
658 rcu_read_unlock();
659
660 return tun;
661}
662
663static struct tun_struct *tun_get(struct file *file)
664{
665 return __tun_get(file->private_data);
666}
667
668static void tun_put(struct tun_struct *tun)
669{
670 dev_put(tun->dev);
671}
672
673/* TAP filtering */
674static void addr_hash_set(u32 *mask, const u8 *addr)
675{
676 int n = ether_crc(ETH_ALEN, addr) >> 26;
677 mask[n >> 5] |= (1 << (n & 31));
678}
679
680static unsigned int addr_hash_test(const u32 *mask, const u8 *addr)
681{
682 int n = ether_crc(ETH_ALEN, addr) >> 26;
683 return mask[n >> 5] & (1 << (n & 31));
684}
685
686static int update_filter(struct tap_filter *filter, void __user *arg)
687{
688 struct { u8 u[ETH_ALEN]; } *addr;
689 struct tun_filter uf;
690 int err, alen, n, nexact;
691
692 if (copy_from_user(&uf, arg, sizeof(uf)))
693 return -EFAULT;
694
695 if (!uf.count) {
696 /* Disabled */
697 filter->count = 0;
698 return 0;
699 }
700
701 alen = ETH_ALEN * uf.count;
702 addr = kmalloc(alen, GFP_KERNEL);
703 if (!addr)
704 return -ENOMEM;
705
706 if (copy_from_user(addr, arg + sizeof(uf), alen)) {
707 err = -EFAULT;
708 goto done;
709 }
710
711 /* The filter is updated without holding any locks. Which is
712 * perfectly safe. We disable it first and in the worst
713 * case we'll accept a few undesired packets. */
714 filter->count = 0;
715 wmb();
716
717 /* Use first set of addresses as an exact filter */
718 for (n = 0; n < uf.count && n < FLT_EXACT_COUNT; n++)
719 memcpy(filter->addr[n], addr[n].u, ETH_ALEN);
720
721 nexact = n;
722
723 /* Remaining multicast addresses are hashed,
724 * unicast will leave the filter disabled. */
725 memset(filter->mask, 0, sizeof(filter->mask));
726 for (; n < uf.count; n++) {
727 if (!is_multicast_ether_addr(addr[n].u)) {
728 err = 0; /* no filter */
729 goto done;
730 }
731 addr_hash_set(filter->mask, addr[n].u);
732 }
733
734 /* For ALLMULTI just set the mask to all ones.
735 * This overrides the mask populated above. */
736 if ((uf.flags & TUN_FLT_ALLMULTI))
737 memset(filter->mask, ~0, sizeof(filter->mask));
738
739 /* Now enable the filter */
740 wmb();
741 filter->count = nexact;
742
743 /* Return the number of exact filters */
744 err = nexact;
745
746done:
747 kfree(addr);
748 return err;
749}
750
751/* Returns: 0 - drop, !=0 - accept */
752static int run_filter(struct tap_filter *filter, const struct sk_buff *skb)
753{
754 /* Cannot use eth_hdr(skb) here because skb_mac_hdr() is incorrect
755 * at this point. */
756 struct ethhdr *eh = (struct ethhdr *) skb->data;
757 int i;
758
759 /* Exact match */
760 for (i = 0; i < filter->count; i++)
761 if (ether_addr_equal(eh->h_dest, filter->addr[i]))
762 return 1;
763
764 /* Inexact match (multicast only) */
765 if (is_multicast_ether_addr(eh->h_dest))
766 return addr_hash_test(filter->mask, eh->h_dest);
767
768 return 0;
769}
770
771/*
772 * Checks whether the packet is accepted or not.
773 * Returns: 0 - drop, !=0 - accept
774 */
775static int check_filter(struct tap_filter *filter, const struct sk_buff *skb)
776{
777 if (!filter->count)
778 return 1;
779
780 return run_filter(filter, skb);
781}
782
783/* Network device part of the driver */
784
785static const struct ethtool_ops tun_ethtool_ops;
786
787/* Net device detach from fd. */
788static void tun_net_uninit(struct net_device *dev)
789{
790 tun_detach_all(dev);
791}
792
793/* Net device open. */
794static int tun_net_open(struct net_device *dev)
795{
796 netif_tx_start_all_queues(dev);
797 return 0;
798}
799
800/* Net device close. */
801static int tun_net_close(struct net_device *dev)
802{
803 netif_tx_stop_all_queues(dev);
804 return 0;
805}
806
807/* Net device start xmit */
808static netdev_tx_t tun_net_xmit(struct sk_buff *skb, struct net_device *dev)
809{
810 struct tun_struct *tun = netdev_priv(dev);
811 int txq = skb->queue_mapping;
812 struct tun_file *tfile;
813 u32 numqueues = 0;
814
815 rcu_read_lock();
816 tfile = rcu_dereference(tun->tfiles[txq]);
817 numqueues = ACCESS_ONCE(tun->numqueues);
818
819 /* Drop packet if interface is not attached */
820 if (txq >= numqueues)
821 goto drop;
822
823 if (numqueues == 1) {
824 /* Select queue was not called for the skbuff, so we extract the
825 * RPS hash and save it into the flow_table here.
826 */
827 __u32 rxhash;
828
829 rxhash = skb_get_hash(skb);
830 if (rxhash) {
831 struct tun_flow_entry *e;
832 e = tun_flow_find(&tun->flows[tun_hashfn(rxhash)],
833 rxhash);
834 if (e)
835 tun_flow_save_rps_rxhash(e, rxhash);
836 }
837 }
838
839 tun_debug(KERN_INFO, tun, "tun_net_xmit %d\n", skb->len);
840
841 BUG_ON(!tfile);
842
843 /* Drop if the filter does not like it.
844 * This is a noop if the filter is disabled.
845 * Filter can be enabled only for the TAP devices. */
846 if (!check_filter(&tun->txflt, skb))
847 goto drop;
848
849 if (tfile->socket.sk->sk_filter &&
850 sk_filter(tfile->socket.sk, skb))
851 goto drop;
852
853 /* Limit the number of packets queued by dividing txq length with the
854 * number of queues.
855 */
856 if (skb_queue_len(&tfile->socket.sk->sk_receive_queue) * numqueues
857 >= dev->tx_queue_len)
858 goto drop;
859
860 if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC)))
861 goto drop;
862
863 if (skb->sk && sk_fullsock(skb->sk)) {
864 sock_tx_timestamp(skb->sk, &skb_shinfo(skb)->tx_flags);
865 sw_tx_timestamp(skb);
866 }
867
868 /* Orphan the skb - required as we might hang on to it
869 * for indefinite time.
870 */
871 skb_orphan(skb);
872
873 nf_reset(skb);
874
875 /* Enqueue packet */
876 skb_queue_tail(&tfile->socket.sk->sk_receive_queue, skb);
877
878 /* Notify and wake up reader process */
879 if (tfile->flags & TUN_FASYNC)
880 kill_fasync(&tfile->fasync, SIGIO, POLL_IN);
881 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
882
883 rcu_read_unlock();
884 return NETDEV_TX_OK;
885
886drop:
887 dev->stats.tx_dropped++;
888 skb_tx_error(skb);
889 kfree_skb(skb);
890 rcu_read_unlock();
891 return NET_XMIT_DROP;
892}
893
894static void tun_net_mclist(struct net_device *dev)
895{
896 /*
897 * This callback is supposed to deal with mc filter in
898 * _rx_ path and has nothing to do with the _tx_ path.
899 * In rx path we always accept everything userspace gives us.
900 */
901}
902
903#define MIN_MTU 68
904#define MAX_MTU 65535
905
906static int
907tun_net_change_mtu(struct net_device *dev, int new_mtu)
908{
909 if (new_mtu < MIN_MTU || new_mtu + dev->hard_header_len > MAX_MTU)
910 return -EINVAL;
911 dev->mtu = new_mtu;
912 return 0;
913}
914
915static netdev_features_t tun_net_fix_features(struct net_device *dev,
916 netdev_features_t features)
917{
918 struct tun_struct *tun = netdev_priv(dev);
919
920 return (features & tun->set_features) | (features & ~TUN_USER_FEATURES);
921}
922#ifdef CONFIG_NET_POLL_CONTROLLER
923static void tun_poll_controller(struct net_device *dev)
924{
925 /*
926 * Tun only receives frames when:
927 * 1) the char device endpoint gets data from user space
928 * 2) the tun socket gets a sendmsg call from user space
929 * Since both of those are synchronous operations, we are guaranteed
930 * never to have pending data when we poll for it
931 * so there is nothing to do here but return.
932 * We need this though so netpoll recognizes us as an interface that
933 * supports polling, which enables bridge devices in virt setups to
934 * still use netconsole
935 */
936 return;
937}
938#endif
939
940static void tun_set_headroom(struct net_device *dev, int new_hr)
941{
942 struct tun_struct *tun = netdev_priv(dev);
943
944 if (new_hr < NET_SKB_PAD)
945 new_hr = NET_SKB_PAD;
946
947 tun->align = new_hr;
948}
949
950static const struct net_device_ops tun_netdev_ops = {
951 .ndo_uninit = tun_net_uninit,
952 .ndo_open = tun_net_open,
953 .ndo_stop = tun_net_close,
954 .ndo_start_xmit = tun_net_xmit,
955 .ndo_change_mtu = tun_net_change_mtu,
956 .ndo_fix_features = tun_net_fix_features,
957 .ndo_select_queue = tun_select_queue,
958#ifdef CONFIG_NET_POLL_CONTROLLER
959 .ndo_poll_controller = tun_poll_controller,
960#endif
961 .ndo_set_rx_headroom = tun_set_headroom,
962};
963
964static const struct net_device_ops tap_netdev_ops = {
965 .ndo_uninit = tun_net_uninit,
966 .ndo_open = tun_net_open,
967 .ndo_stop = tun_net_close,
968 .ndo_start_xmit = tun_net_xmit,
969 .ndo_change_mtu = tun_net_change_mtu,
970 .ndo_fix_features = tun_net_fix_features,
971 .ndo_set_rx_mode = tun_net_mclist,
972 .ndo_set_mac_address = eth_mac_addr,
973 .ndo_validate_addr = eth_validate_addr,
974 .ndo_select_queue = tun_select_queue,
975#ifdef CONFIG_NET_POLL_CONTROLLER
976 .ndo_poll_controller = tun_poll_controller,
977#endif
978 .ndo_features_check = passthru_features_check,
979 .ndo_set_rx_headroom = tun_set_headroom,
980};
981
982static void tun_flow_init(struct tun_struct *tun)
983{
984 int i;
985
986 for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++)
987 INIT_HLIST_HEAD(&tun->flows[i]);
988
989 tun->ageing_time = TUN_FLOW_EXPIRE;
990 setup_timer(&tun->flow_gc_timer, tun_flow_cleanup, (unsigned long)tun);
991 mod_timer(&tun->flow_gc_timer,
992 round_jiffies_up(jiffies + tun->ageing_time));
993}
994
995static void tun_flow_uninit(struct tun_struct *tun)
996{
997 del_timer_sync(&tun->flow_gc_timer);
998 tun_flow_flush(tun);
999}
1000
1001/* Initialize net device. */
1002static void tun_net_init(struct net_device *dev)
1003{
1004 struct tun_struct *tun = netdev_priv(dev);
1005
1006 switch (tun->flags & TUN_TYPE_MASK) {
1007 case IFF_TUN:
1008 dev->netdev_ops = &tun_netdev_ops;
1009
1010 /* Point-to-Point TUN Device */
1011 dev->hard_header_len = 0;
1012 dev->addr_len = 0;
1013 dev->mtu = 1500;
1014
1015 /* Zero header length */
1016 dev->type = ARPHRD_NONE;
1017 dev->flags = IFF_POINTOPOINT | IFF_NOARP | IFF_MULTICAST;
1018 break;
1019
1020 case IFF_TAP:
1021 dev->netdev_ops = &tap_netdev_ops;
1022 /* Ethernet TAP Device */
1023 ether_setup(dev);
1024 dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1025 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
1026
1027 eth_hw_addr_random(dev);
1028
1029 break;
1030 }
1031}
1032
1033/* Character device part */
1034
1035/* Poll */
1036static unsigned int tun_chr_poll(struct file *file, poll_table *wait)
1037{
1038 struct tun_file *tfile = file->private_data;
1039 struct tun_struct *tun = __tun_get(tfile);
1040 struct sock *sk;
1041 unsigned int mask = 0;
1042
1043 if (!tun)
1044 return POLLERR;
1045
1046 sk = tfile->socket.sk;
1047
1048 tun_debug(KERN_INFO, tun, "tun_chr_poll\n");
1049
1050 poll_wait(file, sk_sleep(sk), wait);
1051
1052 if (!skb_queue_empty(&sk->sk_receive_queue))
1053 mask |= POLLIN | POLLRDNORM;
1054
1055 if (sock_writeable(sk) ||
1056 (!test_and_set_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags) &&
1057 sock_writeable(sk)))
1058 mask |= POLLOUT | POLLWRNORM;
1059
1060 if (tun->dev->reg_state != NETREG_REGISTERED)
1061 mask = POLLERR;
1062
1063 tun_put(tun);
1064 return mask;
1065}
1066
1067/* prepad is the amount to reserve at front. len is length after that.
1068 * linear is a hint as to how much to copy (usually headers). */
1069static struct sk_buff *tun_alloc_skb(struct tun_file *tfile,
1070 size_t prepad, size_t len,
1071 size_t linear, int noblock)
1072{
1073 struct sock *sk = tfile->socket.sk;
1074 struct sk_buff *skb;
1075 int err;
1076
1077 /* Under a page? Don't bother with paged skb. */
1078 if (prepad + len < PAGE_SIZE || !linear)
1079 linear = len;
1080
1081 skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
1082 &err, 0);
1083 if (!skb)
1084 return ERR_PTR(err);
1085
1086 skb_reserve(skb, prepad);
1087 skb_put(skb, linear);
1088 skb->data_len = len - linear;
1089 skb->len += len - linear;
1090
1091 return skb;
1092}
1093
1094/* Get packet from user space buffer */
1095static ssize_t tun_get_user(struct tun_struct *tun, struct tun_file *tfile,
1096 void *msg_control, struct iov_iter *from,
1097 int noblock)
1098{
1099 struct tun_pi pi = { 0, cpu_to_be16(ETH_P_IP) };
1100 struct sk_buff *skb;
1101 size_t total_len = iov_iter_count(from);
1102 size_t len = total_len, align = tun->align, linear;
1103 struct virtio_net_hdr gso = { 0 };
1104 int good_linear;
1105 int copylen;
1106 bool zerocopy = false;
1107 int err;
1108 u32 rxhash;
1109 ssize_t n;
1110
1111 if (!(tun->dev->flags & IFF_UP))
1112 return -EIO;
1113
1114 if (!(tun->flags & IFF_NO_PI)) {
1115 if (len < sizeof(pi))
1116 return -EINVAL;
1117 len -= sizeof(pi);
1118
1119 n = copy_from_iter(&pi, sizeof(pi), from);
1120 if (n != sizeof(pi))
1121 return -EFAULT;
1122 }
1123
1124 if (tun->flags & IFF_VNET_HDR) {
1125 if (len < tun->vnet_hdr_sz)
1126 return -EINVAL;
1127 len -= tun->vnet_hdr_sz;
1128
1129 n = copy_from_iter(&gso, sizeof(gso), from);
1130 if (n != sizeof(gso))
1131 return -EFAULT;
1132
1133 if ((gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
1134 tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2 > tun16_to_cpu(tun, gso.hdr_len))
1135 gso.hdr_len = cpu_to_tun16(tun, tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2);
1136
1137 if (tun16_to_cpu(tun, gso.hdr_len) > len)
1138 return -EINVAL;
1139 iov_iter_advance(from, tun->vnet_hdr_sz - sizeof(gso));
1140 }
1141
1142 if ((tun->flags & TUN_TYPE_MASK) == IFF_TAP) {
1143 align += NET_IP_ALIGN;
1144 if (unlikely(len < ETH_HLEN ||
1145 (gso.hdr_len && tun16_to_cpu(tun, gso.hdr_len) < ETH_HLEN)))
1146 return -EINVAL;
1147 }
1148
1149 good_linear = SKB_MAX_HEAD(align);
1150
1151 if (msg_control) {
1152 struct iov_iter i = *from;
1153
1154 /* There are 256 bytes to be copied in skb, so there is
1155 * enough room for skb expand head in case it is used.
1156 * The rest of the buffer is mapped from userspace.
1157 */
1158 copylen = gso.hdr_len ? tun16_to_cpu(tun, gso.hdr_len) : GOODCOPY_LEN;
1159 if (copylen > good_linear)
1160 copylen = good_linear;
1161 linear = copylen;
1162 iov_iter_advance(&i, copylen);
1163 if (iov_iter_npages(&i, INT_MAX) <= MAX_SKB_FRAGS)
1164 zerocopy = true;
1165 }
1166
1167 if (!zerocopy) {
1168 copylen = len;
1169 if (tun16_to_cpu(tun, gso.hdr_len) > good_linear)
1170 linear = good_linear;
1171 else
1172 linear = tun16_to_cpu(tun, gso.hdr_len);
1173 }
1174
1175 skb = tun_alloc_skb(tfile, align, copylen, linear, noblock);
1176 if (IS_ERR(skb)) {
1177 if (PTR_ERR(skb) != -EAGAIN)
1178 tun->dev->stats.rx_dropped++;
1179 return PTR_ERR(skb);
1180 }
1181
1182 if (zerocopy)
1183 err = zerocopy_sg_from_iter(skb, from);
1184 else {
1185 err = skb_copy_datagram_from_iter(skb, 0, from, len);
1186 if (!err && msg_control) {
1187 struct ubuf_info *uarg = msg_control;
1188 uarg->callback(uarg, false);
1189 }
1190 }
1191
1192 if (err) {
1193 tun->dev->stats.rx_dropped++;
1194 kfree_skb(skb);
1195 return -EFAULT;
1196 }
1197
1198 if (gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
1199 if (!skb_partial_csum_set(skb, tun16_to_cpu(tun, gso.csum_start),
1200 tun16_to_cpu(tun, gso.csum_offset))) {
1201 tun->dev->stats.rx_frame_errors++;
1202 kfree_skb(skb);
1203 return -EINVAL;
1204 }
1205 }
1206
1207 switch (tun->flags & TUN_TYPE_MASK) {
1208 case IFF_TUN:
1209 if (tun->flags & IFF_NO_PI) {
1210 switch (skb->data[0] & 0xf0) {
1211 case 0x40:
1212 pi.proto = htons(ETH_P_IP);
1213 break;
1214 case 0x60:
1215 pi.proto = htons(ETH_P_IPV6);
1216 break;
1217 default:
1218 tun->dev->stats.rx_dropped++;
1219 kfree_skb(skb);
1220 return -EINVAL;
1221 }
1222 }
1223
1224 skb_reset_mac_header(skb);
1225 skb->protocol = pi.proto;
1226 skb->dev = tun->dev;
1227 break;
1228 case IFF_TAP:
1229 skb->protocol = eth_type_trans(skb, tun->dev);
1230 break;
1231 }
1232
1233 if (gso.gso_type != VIRTIO_NET_HDR_GSO_NONE) {
1234 pr_debug("GSO!\n");
1235 switch (gso.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
1236 case VIRTIO_NET_HDR_GSO_TCPV4:
1237 skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
1238 break;
1239 case VIRTIO_NET_HDR_GSO_TCPV6:
1240 skb_shinfo(skb)->gso_type = SKB_GSO_TCPV6;
1241 break;
1242 case VIRTIO_NET_HDR_GSO_UDP:
1243 skb_shinfo(skb)->gso_type = SKB_GSO_UDP;
1244 break;
1245 default:
1246 tun->dev->stats.rx_frame_errors++;
1247 kfree_skb(skb);
1248 return -EINVAL;
1249 }
1250
1251 if (gso.gso_type & VIRTIO_NET_HDR_GSO_ECN)
1252 skb_shinfo(skb)->gso_type |= SKB_GSO_TCP_ECN;
1253
1254 skb_shinfo(skb)->gso_size = tun16_to_cpu(tun, gso.gso_size);
1255 if (skb_shinfo(skb)->gso_size == 0) {
1256 tun->dev->stats.rx_frame_errors++;
1257 kfree_skb(skb);
1258 return -EINVAL;
1259 }
1260
1261 /* Header must be checked, and gso_segs computed. */
1262 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
1263 skb_shinfo(skb)->gso_segs = 0;
1264 }
1265
1266 /* copy skb_ubuf_info for callback when skb has no error */
1267 if (zerocopy) {
1268 skb_shinfo(skb)->destructor_arg = msg_control;
1269 skb_shinfo(skb)->tx_flags |= SKBTX_DEV_ZEROCOPY;
1270 skb_shinfo(skb)->tx_flags |= SKBTX_SHARED_FRAG;
1271 }
1272
1273 skb_reset_network_header(skb);
1274 skb_probe_transport_header(skb, 0);
1275
1276 rxhash = skb_get_hash(skb);
1277 netif_rx_ni(skb);
1278
1279 tun->dev->stats.rx_packets++;
1280 tun->dev->stats.rx_bytes += len;
1281
1282 tun_flow_update(tun, rxhash, tfile);
1283 return total_len;
1284}
1285
1286static ssize_t tun_chr_write_iter(struct kiocb *iocb, struct iov_iter *from)
1287{
1288 struct file *file = iocb->ki_filp;
1289 struct tun_struct *tun = tun_get(file);
1290 struct tun_file *tfile = file->private_data;
1291 ssize_t result;
1292
1293 if (!tun)
1294 return -EBADFD;
1295
1296 result = tun_get_user(tun, tfile, NULL, from, file->f_flags & O_NONBLOCK);
1297
1298 tun_put(tun);
1299 return result;
1300}
1301
1302/* Put packet to the user space buffer */
1303static ssize_t tun_put_user(struct tun_struct *tun,
1304 struct tun_file *tfile,
1305 struct sk_buff *skb,
1306 struct iov_iter *iter)
1307{
1308 struct tun_pi pi = { 0, skb->protocol };
1309 ssize_t total;
1310 int vlan_offset = 0;
1311 int vlan_hlen = 0;
1312 int vnet_hdr_sz = 0;
1313
1314 if (skb_vlan_tag_present(skb))
1315 vlan_hlen = VLAN_HLEN;
1316
1317 if (tun->flags & IFF_VNET_HDR)
1318 vnet_hdr_sz = tun->vnet_hdr_sz;
1319
1320 total = skb->len + vlan_hlen + vnet_hdr_sz;
1321
1322 if (!(tun->flags & IFF_NO_PI)) {
1323 if (iov_iter_count(iter) < sizeof(pi))
1324 return -EINVAL;
1325
1326 total += sizeof(pi);
1327 if (iov_iter_count(iter) < total) {
1328 /* Packet will be striped */
1329 pi.flags |= TUN_PKT_STRIP;
1330 }
1331
1332 if (copy_to_iter(&pi, sizeof(pi), iter) != sizeof(pi))
1333 return -EFAULT;
1334 }
1335
1336 if (vnet_hdr_sz) {
1337 struct virtio_net_hdr gso = { 0 }; /* no info leak */
1338 if (iov_iter_count(iter) < vnet_hdr_sz)
1339 return -EINVAL;
1340
1341 if (skb_is_gso(skb)) {
1342 struct skb_shared_info *sinfo = skb_shinfo(skb);
1343
1344 /* This is a hint as to how much should be linear. */
1345 gso.hdr_len = cpu_to_tun16(tun, skb_headlen(skb));
1346 gso.gso_size = cpu_to_tun16(tun, sinfo->gso_size);
1347 if (sinfo->gso_type & SKB_GSO_TCPV4)
1348 gso.gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
1349 else if (sinfo->gso_type & SKB_GSO_TCPV6)
1350 gso.gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
1351 else if (sinfo->gso_type & SKB_GSO_UDP)
1352 gso.gso_type = VIRTIO_NET_HDR_GSO_UDP;
1353 else {
1354 pr_err("unexpected GSO type: "
1355 "0x%x, gso_size %d, hdr_len %d\n",
1356 sinfo->gso_type, tun16_to_cpu(tun, gso.gso_size),
1357 tun16_to_cpu(tun, gso.hdr_len));
1358 print_hex_dump(KERN_ERR, "tun: ",
1359 DUMP_PREFIX_NONE,
1360 16, 1, skb->head,
1361 min((int)tun16_to_cpu(tun, gso.hdr_len), 64), true);
1362 WARN_ON_ONCE(1);
1363 return -EINVAL;
1364 }
1365 if (sinfo->gso_type & SKB_GSO_TCP_ECN)
1366 gso.gso_type |= VIRTIO_NET_HDR_GSO_ECN;
1367 } else
1368 gso.gso_type = VIRTIO_NET_HDR_GSO_NONE;
1369
1370 if (skb->ip_summed == CHECKSUM_PARTIAL) {
1371 gso.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
1372 gso.csum_start = cpu_to_tun16(tun, skb_checksum_start_offset(skb) +
1373 vlan_hlen);
1374 gso.csum_offset = cpu_to_tun16(tun, skb->csum_offset);
1375 } else if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
1376 gso.flags = VIRTIO_NET_HDR_F_DATA_VALID;
1377 } /* else everything is zero */
1378
1379 if (copy_to_iter(&gso, sizeof(gso), iter) != sizeof(gso))
1380 return -EFAULT;
1381
1382 iov_iter_advance(iter, vnet_hdr_sz - sizeof(gso));
1383 }
1384
1385 if (vlan_hlen) {
1386 int ret;
1387 struct {
1388 __be16 h_vlan_proto;
1389 __be16 h_vlan_TCI;
1390 } veth;
1391
1392 veth.h_vlan_proto = skb->vlan_proto;
1393 veth.h_vlan_TCI = htons(skb_vlan_tag_get(skb));
1394
1395 vlan_offset = offsetof(struct vlan_ethhdr, h_vlan_proto);
1396
1397 ret = skb_copy_datagram_iter(skb, 0, iter, vlan_offset);
1398 if (ret || !iov_iter_count(iter))
1399 goto done;
1400
1401 ret = copy_to_iter(&veth, sizeof(veth), iter);
1402 if (ret != sizeof(veth) || !iov_iter_count(iter))
1403 goto done;
1404 }
1405
1406 skb_copy_datagram_iter(skb, vlan_offset, iter, skb->len - vlan_offset);
1407
1408done:
1409 tun->dev->stats.tx_packets++;
1410 tun->dev->stats.tx_bytes += skb->len + vlan_hlen;
1411
1412 return total;
1413}
1414
1415static ssize_t tun_do_read(struct tun_struct *tun, struct tun_file *tfile,
1416 struct iov_iter *to,
1417 int noblock)
1418{
1419 struct sk_buff *skb;
1420 ssize_t ret;
1421 int peeked, err, off = 0;
1422
1423 tun_debug(KERN_INFO, tun, "tun_do_read\n");
1424
1425 if (!iov_iter_count(to))
1426 return 0;
1427
1428 if (tun->dev->reg_state != NETREG_REGISTERED)
1429 return -EIO;
1430
1431 /* Read frames from queue */
1432 skb = __skb_recv_datagram(tfile->socket.sk, noblock ? MSG_DONTWAIT : 0,
1433 &peeked, &off, &err);
1434 if (!skb)
1435 return err;
1436
1437 ret = tun_put_user(tun, tfile, skb, to);
1438 if (unlikely(ret < 0))
1439 kfree_skb(skb);
1440 else
1441 consume_skb(skb);
1442
1443 return ret;
1444}
1445
1446static ssize_t tun_chr_read_iter(struct kiocb *iocb, struct iov_iter *to)
1447{
1448 struct file *file = iocb->ki_filp;
1449 struct tun_file *tfile = file->private_data;
1450 struct tun_struct *tun = __tun_get(tfile);
1451 ssize_t len = iov_iter_count(to), ret;
1452
1453 if (!tun)
1454 return -EBADFD;
1455 ret = tun_do_read(tun, tfile, to, file->f_flags & O_NONBLOCK);
1456 ret = min_t(ssize_t, ret, len);
1457 if (ret > 0)
1458 iocb->ki_pos = ret;
1459 tun_put(tun);
1460 return ret;
1461}
1462
1463static void tun_free_netdev(struct net_device *dev)
1464{
1465 struct tun_struct *tun = netdev_priv(dev);
1466
1467 BUG_ON(!(list_empty(&tun->disabled)));
1468 tun_flow_uninit(tun);
1469 security_tun_dev_free_security(tun->security);
1470 free_netdev(dev);
1471}
1472
1473static void tun_setup(struct net_device *dev)
1474{
1475 struct tun_struct *tun = netdev_priv(dev);
1476
1477 tun->owner = INVALID_UID;
1478 tun->group = INVALID_GID;
1479
1480 dev->ethtool_ops = &tun_ethtool_ops;
1481 dev->destructor = tun_free_netdev;
1482 /* We prefer our own queue length */
1483 dev->tx_queue_len = TUN_READQ_SIZE;
1484}
1485
1486/* Trivial set of netlink ops to allow deleting tun or tap
1487 * device with netlink.
1488 */
1489static int tun_validate(struct nlattr *tb[], struct nlattr *data[])
1490{
1491 return -EINVAL;
1492}
1493
1494static struct rtnl_link_ops tun_link_ops __read_mostly = {
1495 .kind = DRV_NAME,
1496 .priv_size = sizeof(struct tun_struct),
1497 .setup = tun_setup,
1498 .validate = tun_validate,
1499};
1500
1501static void tun_sock_write_space(struct sock *sk)
1502{
1503 struct tun_file *tfile;
1504 wait_queue_head_t *wqueue;
1505
1506 if (!sock_writeable(sk))
1507 return;
1508
1509 if (!test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags))
1510 return;
1511
1512 wqueue = sk_sleep(sk);
1513 if (wqueue && waitqueue_active(wqueue))
1514 wake_up_interruptible_sync_poll(wqueue, POLLOUT |
1515 POLLWRNORM | POLLWRBAND);
1516
1517 tfile = container_of(sk, struct tun_file, sk);
1518 kill_fasync(&tfile->fasync, SIGIO, POLL_OUT);
1519}
1520
1521static int tun_sendmsg(struct socket *sock, struct msghdr *m, size_t total_len)
1522{
1523 int ret;
1524 struct tun_file *tfile = container_of(sock, struct tun_file, socket);
1525 struct tun_struct *tun = __tun_get(tfile);
1526
1527 if (!tun)
1528 return -EBADFD;
1529
1530 ret = tun_get_user(tun, tfile, m->msg_control, &m->msg_iter,
1531 m->msg_flags & MSG_DONTWAIT);
1532 tun_put(tun);
1533 return ret;
1534}
1535
1536static int tun_recvmsg(struct socket *sock, struct msghdr *m, size_t total_len,
1537 int flags)
1538{
1539 struct tun_file *tfile = container_of(sock, struct tun_file, socket);
1540 struct tun_struct *tun = __tun_get(tfile);
1541 int ret;
1542
1543 if (!tun)
1544 return -EBADFD;
1545
1546 if (flags & ~(MSG_DONTWAIT|MSG_TRUNC|MSG_ERRQUEUE)) {
1547 ret = -EINVAL;
1548 goto out;
1549 }
1550 if (flags & MSG_ERRQUEUE) {
1551 ret = sock_recv_errqueue(sock->sk, m, total_len,
1552 SOL_PACKET, TUN_TX_TIMESTAMP);
1553 goto out;
1554 }
1555 ret = tun_do_read(tun, tfile, &m->msg_iter, flags & MSG_DONTWAIT);
1556 if (ret > (ssize_t)total_len) {
1557 m->msg_flags |= MSG_TRUNC;
1558 ret = flags & MSG_TRUNC ? ret : total_len;
1559 }
1560out:
1561 tun_put(tun);
1562 return ret;
1563}
1564
1565/* Ops structure to mimic raw sockets with tun */
1566static const struct proto_ops tun_socket_ops = {
1567 .sendmsg = tun_sendmsg,
1568 .recvmsg = tun_recvmsg,
1569};
1570
1571static struct proto tun_proto = {
1572 .name = "tun",
1573 .owner = THIS_MODULE,
1574 .obj_size = sizeof(struct tun_file),
1575};
1576
1577static int tun_flags(struct tun_struct *tun)
1578{
1579 return tun->flags & (TUN_FEATURES | IFF_PERSIST | IFF_TUN | IFF_TAP);
1580}
1581
1582static ssize_t tun_show_flags(struct device *dev, struct device_attribute *attr,
1583 char *buf)
1584{
1585 struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1586 return sprintf(buf, "0x%x\n", tun_flags(tun));
1587}
1588
1589static ssize_t tun_show_owner(struct device *dev, struct device_attribute *attr,
1590 char *buf)
1591{
1592 struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1593 return uid_valid(tun->owner)?
1594 sprintf(buf, "%u\n",
1595 from_kuid_munged(current_user_ns(), tun->owner)):
1596 sprintf(buf, "-1\n");
1597}
1598
1599static ssize_t tun_show_group(struct device *dev, struct device_attribute *attr,
1600 char *buf)
1601{
1602 struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1603 return gid_valid(tun->group) ?
1604 sprintf(buf, "%u\n",
1605 from_kgid_munged(current_user_ns(), tun->group)):
1606 sprintf(buf, "-1\n");
1607}
1608
1609static DEVICE_ATTR(tun_flags, 0444, tun_show_flags, NULL);
1610static DEVICE_ATTR(owner, 0444, tun_show_owner, NULL);
1611static DEVICE_ATTR(group, 0444, tun_show_group, NULL);
1612
1613static struct attribute *tun_dev_attrs[] = {
1614 &dev_attr_tun_flags.attr,
1615 &dev_attr_owner.attr,
1616 &dev_attr_group.attr,
1617 NULL
1618};
1619
1620static const struct attribute_group tun_attr_group = {
1621 .attrs = tun_dev_attrs
1622};
1623
1624static int tun_set_iff(struct net *net, struct file *file, struct ifreq *ifr)
1625{
1626 struct tun_struct *tun;
1627 struct tun_file *tfile = file->private_data;
1628 struct net_device *dev;
1629 int err;
1630
1631 if (tfile->detached)
1632 return -EINVAL;
1633
1634 dev = __dev_get_by_name(net, ifr->ifr_name);
1635 if (dev) {
1636 if (ifr->ifr_flags & IFF_TUN_EXCL)
1637 return -EBUSY;
1638 if ((ifr->ifr_flags & IFF_TUN) && dev->netdev_ops == &tun_netdev_ops)
1639 tun = netdev_priv(dev);
1640 else if ((ifr->ifr_flags & IFF_TAP) && dev->netdev_ops == &tap_netdev_ops)
1641 tun = netdev_priv(dev);
1642 else
1643 return -EINVAL;
1644
1645 if (!!(ifr->ifr_flags & IFF_MULTI_QUEUE) !=
1646 !!(tun->flags & IFF_MULTI_QUEUE))
1647 return -EINVAL;
1648
1649 if (tun_not_capable(tun))
1650 return -EPERM;
1651 err = security_tun_dev_open(tun->security);
1652 if (err < 0)
1653 return err;
1654
1655 err = tun_attach(tun, file, ifr->ifr_flags & IFF_NOFILTER);
1656 if (err < 0)
1657 return err;
1658
1659 if (tun->flags & IFF_MULTI_QUEUE &&
1660 (tun->numqueues + tun->numdisabled > 1)) {
1661 /* One or more queue has already been attached, no need
1662 * to initialize the device again.
1663 */
1664 return 0;
1665 }
1666 }
1667 else {
1668 char *name;
1669 unsigned long flags = 0;
1670 int queues = ifr->ifr_flags & IFF_MULTI_QUEUE ?
1671 MAX_TAP_QUEUES : 1;
1672
1673 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
1674 return -EPERM;
1675 err = security_tun_dev_create();
1676 if (err < 0)
1677 return err;
1678
1679 /* Set dev type */
1680 if (ifr->ifr_flags & IFF_TUN) {
1681 /* TUN device */
1682 flags |= IFF_TUN;
1683 name = "tun%d";
1684 } else if (ifr->ifr_flags & IFF_TAP) {
1685 /* TAP device */
1686 flags |= IFF_TAP;
1687 name = "tap%d";
1688 } else
1689 return -EINVAL;
1690
1691 if (*ifr->ifr_name)
1692 name = ifr->ifr_name;
1693
1694 dev = alloc_netdev_mqs(sizeof(struct tun_struct), name,
1695 NET_NAME_UNKNOWN, tun_setup, queues,
1696 queues);
1697
1698 if (!dev)
1699 return -ENOMEM;
1700
1701 dev_net_set(dev, net);
1702 dev->rtnl_link_ops = &tun_link_ops;
1703 dev->ifindex = tfile->ifindex;
1704 dev->sysfs_groups[0] = &tun_attr_group;
1705
1706 tun = netdev_priv(dev);
1707 tun->dev = dev;
1708 tun->flags = flags;
1709 tun->txflt.count = 0;
1710 tun->vnet_hdr_sz = sizeof(struct virtio_net_hdr);
1711
1712 tun->align = NET_SKB_PAD;
1713 tun->filter_attached = false;
1714 tun->sndbuf = tfile->socket.sk->sk_sndbuf;
1715
1716 spin_lock_init(&tun->lock);
1717
1718 err = security_tun_dev_alloc_security(&tun->security);
1719 if (err < 0)
1720 goto err_free_dev;
1721
1722 tun_net_init(dev);
1723 tun_flow_init(tun);
1724
1725 dev->hw_features = NETIF_F_SG | NETIF_F_FRAGLIST |
1726 TUN_USER_FEATURES | NETIF_F_HW_VLAN_CTAG_TX |
1727 NETIF_F_HW_VLAN_STAG_TX;
1728 dev->features = dev->hw_features;
1729 dev->vlan_features = dev->features &
1730 ~(NETIF_F_HW_VLAN_CTAG_TX |
1731 NETIF_F_HW_VLAN_STAG_TX);
1732
1733 INIT_LIST_HEAD(&tun->disabled);
1734 err = tun_attach(tun, file, false);
1735 if (err < 0)
1736 goto err_free_flow;
1737
1738 err = register_netdevice(tun->dev);
1739 if (err < 0)
1740 goto err_detach;
1741 }
1742
1743 netif_carrier_on(tun->dev);
1744
1745 tun_debug(KERN_INFO, tun, "tun_set_iff\n");
1746
1747 tun->flags = (tun->flags & ~TUN_FEATURES) |
1748 (ifr->ifr_flags & TUN_FEATURES);
1749
1750 /* Make sure persistent devices do not get stuck in
1751 * xoff state.
1752 */
1753 if (netif_running(tun->dev))
1754 netif_tx_wake_all_queues(tun->dev);
1755
1756 strcpy(ifr->ifr_name, tun->dev->name);
1757 return 0;
1758
1759err_detach:
1760 tun_detach_all(dev);
1761err_free_flow:
1762 tun_flow_uninit(tun);
1763 security_tun_dev_free_security(tun->security);
1764err_free_dev:
1765 free_netdev(dev);
1766 return err;
1767}
1768
1769static void tun_get_iff(struct net *net, struct tun_struct *tun,
1770 struct ifreq *ifr)
1771{
1772 tun_debug(KERN_INFO, tun, "tun_get_iff\n");
1773
1774 strcpy(ifr->ifr_name, tun->dev->name);
1775
1776 ifr->ifr_flags = tun_flags(tun);
1777
1778}
1779
1780/* This is like a cut-down ethtool ops, except done via tun fd so no
1781 * privs required. */
1782static int set_offload(struct tun_struct *tun, unsigned long arg)
1783{
1784 netdev_features_t features = 0;
1785
1786 if (arg & TUN_F_CSUM) {
1787 features |= NETIF_F_HW_CSUM;
1788 arg &= ~TUN_F_CSUM;
1789
1790 if (arg & (TUN_F_TSO4|TUN_F_TSO6)) {
1791 if (arg & TUN_F_TSO_ECN) {
1792 features |= NETIF_F_TSO_ECN;
1793 arg &= ~TUN_F_TSO_ECN;
1794 }
1795 if (arg & TUN_F_TSO4)
1796 features |= NETIF_F_TSO;
1797 if (arg & TUN_F_TSO6)
1798 features |= NETIF_F_TSO6;
1799 arg &= ~(TUN_F_TSO4|TUN_F_TSO6);
1800 }
1801
1802 if (arg & TUN_F_UFO) {
1803 features |= NETIF_F_UFO;
1804 arg &= ~TUN_F_UFO;
1805 }
1806 }
1807
1808 /* This gives the user a way to test for new features in future by
1809 * trying to set them. */
1810 if (arg)
1811 return -EINVAL;
1812
1813 tun->set_features = features;
1814 netdev_update_features(tun->dev);
1815
1816 return 0;
1817}
1818
1819static void tun_detach_filter(struct tun_struct *tun, int n)
1820{
1821 int i;
1822 struct tun_file *tfile;
1823
1824 for (i = 0; i < n; i++) {
1825 tfile = rtnl_dereference(tun->tfiles[i]);
1826 __sk_detach_filter(tfile->socket.sk, lockdep_rtnl_is_held());
1827 }
1828
1829 tun->filter_attached = false;
1830}
1831
1832static int tun_attach_filter(struct tun_struct *tun)
1833{
1834 int i, ret = 0;
1835 struct tun_file *tfile;
1836
1837 for (i = 0; i < tun->numqueues; i++) {
1838 tfile = rtnl_dereference(tun->tfiles[i]);
1839 ret = __sk_attach_filter(&tun->fprog, tfile->socket.sk,
1840 lockdep_rtnl_is_held());
1841 if (ret) {
1842 tun_detach_filter(tun, i);
1843 return ret;
1844 }
1845 }
1846
1847 tun->filter_attached = true;
1848 return ret;
1849}
1850
1851static void tun_set_sndbuf(struct tun_struct *tun)
1852{
1853 struct tun_file *tfile;
1854 int i;
1855
1856 for (i = 0; i < tun->numqueues; i++) {
1857 tfile = rtnl_dereference(tun->tfiles[i]);
1858 tfile->socket.sk->sk_sndbuf = tun->sndbuf;
1859 }
1860}
1861
1862static int tun_set_queue(struct file *file, struct ifreq *ifr)
1863{
1864 struct tun_file *tfile = file->private_data;
1865 struct tun_struct *tun;
1866 int ret = 0;
1867
1868 rtnl_lock();
1869
1870 if (ifr->ifr_flags & IFF_ATTACH_QUEUE) {
1871 tun = tfile->detached;
1872 if (!tun) {
1873 ret = -EINVAL;
1874 goto unlock;
1875 }
1876 ret = security_tun_dev_attach_queue(tun->security);
1877 if (ret < 0)
1878 goto unlock;
1879 ret = tun_attach(tun, file, false);
1880 } else if (ifr->ifr_flags & IFF_DETACH_QUEUE) {
1881 tun = rtnl_dereference(tfile->tun);
1882 if (!tun || !(tun->flags & IFF_MULTI_QUEUE) || tfile->detached)
1883 ret = -EINVAL;
1884 else
1885 __tun_detach(tfile, false);
1886 } else
1887 ret = -EINVAL;
1888
1889unlock:
1890 rtnl_unlock();
1891 return ret;
1892}
1893
1894static long __tun_chr_ioctl(struct file *file, unsigned int cmd,
1895 unsigned long arg, int ifreq_len)
1896{
1897 struct tun_file *tfile = file->private_data;
1898 struct tun_struct *tun;
1899 void __user* argp = (void __user*)arg;
1900 struct ifreq ifr;
1901 kuid_t owner;
1902 kgid_t group;
1903 int sndbuf;
1904 int vnet_hdr_sz;
1905 unsigned int ifindex;
1906 int le;
1907 int ret;
1908
1909 if (cmd == TUNSETIFF || cmd == TUNSETQUEUE || _IOC_TYPE(cmd) == 0x89) {
1910 if (copy_from_user(&ifr, argp, ifreq_len))
1911 return -EFAULT;
1912 } else {
1913 memset(&ifr, 0, sizeof(ifr));
1914 }
1915 if (cmd == TUNGETFEATURES) {
1916 /* Currently this just means: "what IFF flags are valid?".
1917 * This is needed because we never checked for invalid flags on
1918 * TUNSETIFF.
1919 */
1920 return put_user(IFF_TUN | IFF_TAP | TUN_FEATURES,
1921 (unsigned int __user*)argp);
1922 } else if (cmd == TUNSETQUEUE)
1923 return tun_set_queue(file, &ifr);
1924
1925 ret = 0;
1926 rtnl_lock();
1927
1928 tun = __tun_get(tfile);
1929 if (cmd == TUNSETIFF && !tun) {
1930 ifr.ifr_name[IFNAMSIZ-1] = '\0';
1931
1932 ret = tun_set_iff(sock_net(&tfile->sk), file, &ifr);
1933
1934 if (ret)
1935 goto unlock;
1936
1937 if (copy_to_user(argp, &ifr, ifreq_len))
1938 ret = -EFAULT;
1939 goto unlock;
1940 }
1941 if (cmd == TUNSETIFINDEX) {
1942 ret = -EPERM;
1943 if (tun)
1944 goto unlock;
1945
1946 ret = -EFAULT;
1947 if (copy_from_user(&ifindex, argp, sizeof(ifindex)))
1948 goto unlock;
1949
1950 ret = 0;
1951 tfile->ifindex = ifindex;
1952 goto unlock;
1953 }
1954
1955 ret = -EBADFD;
1956 if (!tun)
1957 goto unlock;
1958
1959 tun_debug(KERN_INFO, tun, "tun_chr_ioctl cmd %u\n", cmd);
1960
1961 ret = 0;
1962 switch (cmd) {
1963 case TUNGETIFF:
1964 tun_get_iff(current->nsproxy->net_ns, tun, &ifr);
1965
1966 if (tfile->detached)
1967 ifr.ifr_flags |= IFF_DETACH_QUEUE;
1968 if (!tfile->socket.sk->sk_filter)
1969 ifr.ifr_flags |= IFF_NOFILTER;
1970
1971 if (copy_to_user(argp, &ifr, ifreq_len))
1972 ret = -EFAULT;
1973 break;
1974
1975 case TUNSETNOCSUM:
1976 /* Disable/Enable checksum */
1977
1978 /* [unimplemented] */
1979 tun_debug(KERN_INFO, tun, "ignored: set checksum %s\n",
1980 arg ? "disabled" : "enabled");
1981 break;
1982
1983 case TUNSETPERSIST:
1984 /* Disable/Enable persist mode. Keep an extra reference to the
1985 * module to prevent the module being unprobed.
1986 */
1987 if (arg && !(tun->flags & IFF_PERSIST)) {
1988 tun->flags |= IFF_PERSIST;
1989 __module_get(THIS_MODULE);
1990 }
1991 if (!arg && (tun->flags & IFF_PERSIST)) {
1992 tun->flags &= ~IFF_PERSIST;
1993 module_put(THIS_MODULE);
1994 }
1995
1996 tun_debug(KERN_INFO, tun, "persist %s\n",
1997 arg ? "enabled" : "disabled");
1998 break;
1999
2000 case TUNSETOWNER:
2001 /* Set owner of the device */
2002 owner = make_kuid(current_user_ns(), arg);
2003 if (!uid_valid(owner)) {
2004 ret = -EINVAL;
2005 break;
2006 }
2007 tun->owner = owner;
2008 tun_debug(KERN_INFO, tun, "owner set to %u\n",
2009 from_kuid(&init_user_ns, tun->owner));
2010 break;
2011
2012 case TUNSETGROUP:
2013 /* Set group of the device */
2014 group = make_kgid(current_user_ns(), arg);
2015 if (!gid_valid(group)) {
2016 ret = -EINVAL;
2017 break;
2018 }
2019 tun->group = group;
2020 tun_debug(KERN_INFO, tun, "group set to %u\n",
2021 from_kgid(&init_user_ns, tun->group));
2022 break;
2023
2024 case TUNSETLINK:
2025 /* Only allow setting the type when the interface is down */
2026 if (tun->dev->flags & IFF_UP) {
2027 tun_debug(KERN_INFO, tun,
2028 "Linktype set failed because interface is up\n");
2029 ret = -EBUSY;
2030 } else {
2031 tun->dev->type = (int) arg;
2032 tun_debug(KERN_INFO, tun, "linktype set to %d\n",
2033 tun->dev->type);
2034 ret = 0;
2035 }
2036 break;
2037
2038#ifdef TUN_DEBUG
2039 case TUNSETDEBUG:
2040 tun->debug = arg;
2041 break;
2042#endif
2043 case TUNSETOFFLOAD:
2044 ret = set_offload(tun, arg);
2045 break;
2046
2047 case TUNSETTXFILTER:
2048 /* Can be set only for TAPs */
2049 ret = -EINVAL;
2050 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
2051 break;
2052 ret = update_filter(&tun->txflt, (void __user *)arg);
2053 break;
2054
2055 case SIOCGIFHWADDR:
2056 /* Get hw address */
2057 memcpy(ifr.ifr_hwaddr.sa_data, tun->dev->dev_addr, ETH_ALEN);
2058 ifr.ifr_hwaddr.sa_family = tun->dev->type;
2059 if (copy_to_user(argp, &ifr, ifreq_len))
2060 ret = -EFAULT;
2061 break;
2062
2063 case SIOCSIFHWADDR:
2064 /* Set hw address */
2065 tun_debug(KERN_DEBUG, tun, "set hw address: %pM\n",
2066 ifr.ifr_hwaddr.sa_data);
2067
2068 ret = dev_set_mac_address(tun->dev, &ifr.ifr_hwaddr);
2069 break;
2070
2071 case TUNGETSNDBUF:
2072 sndbuf = tfile->socket.sk->sk_sndbuf;
2073 if (copy_to_user(argp, &sndbuf, sizeof(sndbuf)))
2074 ret = -EFAULT;
2075 break;
2076
2077 case TUNSETSNDBUF:
2078 if (copy_from_user(&sndbuf, argp, sizeof(sndbuf))) {
2079 ret = -EFAULT;
2080 break;
2081 }
2082
2083 tun->sndbuf = sndbuf;
2084 tun_set_sndbuf(tun);
2085 break;
2086
2087 case TUNGETVNETHDRSZ:
2088 vnet_hdr_sz = tun->vnet_hdr_sz;
2089 if (copy_to_user(argp, &vnet_hdr_sz, sizeof(vnet_hdr_sz)))
2090 ret = -EFAULT;
2091 break;
2092
2093 case TUNSETVNETHDRSZ:
2094 if (copy_from_user(&vnet_hdr_sz, argp, sizeof(vnet_hdr_sz))) {
2095 ret = -EFAULT;
2096 break;
2097 }
2098 if (vnet_hdr_sz < (int)sizeof(struct virtio_net_hdr)) {
2099 ret = -EINVAL;
2100 break;
2101 }
2102
2103 tun->vnet_hdr_sz = vnet_hdr_sz;
2104 break;
2105
2106 case TUNGETVNETLE:
2107 le = !!(tun->flags & TUN_VNET_LE);
2108 if (put_user(le, (int __user *)argp))
2109 ret = -EFAULT;
2110 break;
2111
2112 case TUNSETVNETLE:
2113 if (get_user(le, (int __user *)argp)) {
2114 ret = -EFAULT;
2115 break;
2116 }
2117 if (le)
2118 tun->flags |= TUN_VNET_LE;
2119 else
2120 tun->flags &= ~TUN_VNET_LE;
2121 break;
2122
2123 case TUNGETVNETBE:
2124 ret = tun_get_vnet_be(tun, argp);
2125 break;
2126
2127 case TUNSETVNETBE:
2128 ret = tun_set_vnet_be(tun, argp);
2129 break;
2130
2131 case TUNATTACHFILTER:
2132 /* Can be set only for TAPs */
2133 ret = -EINVAL;
2134 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
2135 break;
2136 ret = -EFAULT;
2137 if (copy_from_user(&tun->fprog, argp, sizeof(tun->fprog)))
2138 break;
2139
2140 ret = tun_attach_filter(tun);
2141 break;
2142
2143 case TUNDETACHFILTER:
2144 /* Can be set only for TAPs */
2145 ret = -EINVAL;
2146 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
2147 break;
2148 ret = 0;
2149 tun_detach_filter(tun, tun->numqueues);
2150 break;
2151
2152 case TUNGETFILTER:
2153 ret = -EINVAL;
2154 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
2155 break;
2156 ret = -EFAULT;
2157 if (copy_to_user(argp, &tun->fprog, sizeof(tun->fprog)))
2158 break;
2159 ret = 0;
2160 break;
2161
2162 default:
2163 ret = -EINVAL;
2164 break;
2165 }
2166
2167unlock:
2168 rtnl_unlock();
2169 if (tun)
2170 tun_put(tun);
2171 return ret;
2172}
2173
2174static long tun_chr_ioctl(struct file *file,
2175 unsigned int cmd, unsigned long arg)
2176{
2177 return __tun_chr_ioctl(file, cmd, arg, sizeof (struct ifreq));
2178}
2179
2180#ifdef CONFIG_COMPAT
2181static long tun_chr_compat_ioctl(struct file *file,
2182 unsigned int cmd, unsigned long arg)
2183{
2184 switch (cmd) {
2185 case TUNSETIFF:
2186 case TUNGETIFF:
2187 case TUNSETTXFILTER:
2188 case TUNGETSNDBUF:
2189 case TUNSETSNDBUF:
2190 case SIOCGIFHWADDR:
2191 case SIOCSIFHWADDR:
2192 arg = (unsigned long)compat_ptr(arg);
2193 break;
2194 default:
2195 arg = (compat_ulong_t)arg;
2196 break;
2197 }
2198
2199 /*
2200 * compat_ifreq is shorter than ifreq, so we must not access beyond
2201 * the end of that structure. All fields that are used in this
2202 * driver are compatible though, we don't need to convert the
2203 * contents.
2204 */
2205 return __tun_chr_ioctl(file, cmd, arg, sizeof(struct compat_ifreq));
2206}
2207#endif /* CONFIG_COMPAT */
2208
2209static int tun_chr_fasync(int fd, struct file *file, int on)
2210{
2211 struct tun_file *tfile = file->private_data;
2212 int ret;
2213
2214 if ((ret = fasync_helper(fd, file, on, &tfile->fasync)) < 0)
2215 goto out;
2216
2217 if (on) {
2218 __f_setown(file, task_pid(current), PIDTYPE_PID, 0);
2219 tfile->flags |= TUN_FASYNC;
2220 } else
2221 tfile->flags &= ~TUN_FASYNC;
2222 ret = 0;
2223out:
2224 return ret;
2225}
2226
2227static int tun_chr_open(struct inode *inode, struct file * file)
2228{
2229 struct net *net = current->nsproxy->net_ns;
2230 struct tun_file *tfile;
2231
2232 DBG1(KERN_INFO, "tunX: tun_chr_open\n");
2233
2234 tfile = (struct tun_file *)sk_alloc(net, AF_UNSPEC, GFP_KERNEL,
2235 &tun_proto, 0);
2236 if (!tfile)
2237 return -ENOMEM;
2238 RCU_INIT_POINTER(tfile->tun, NULL);
2239 tfile->flags = 0;
2240 tfile->ifindex = 0;
2241
2242 init_waitqueue_head(&tfile->wq.wait);
2243 RCU_INIT_POINTER(tfile->socket.wq, &tfile->wq);
2244
2245 tfile->socket.file = file;
2246 tfile->socket.ops = &tun_socket_ops;
2247
2248 sock_init_data(&tfile->socket, &tfile->sk);
2249
2250 tfile->sk.sk_write_space = tun_sock_write_space;
2251 tfile->sk.sk_sndbuf = INT_MAX;
2252
2253 file->private_data = tfile;
2254 INIT_LIST_HEAD(&tfile->next);
2255
2256 sock_set_flag(&tfile->sk, SOCK_ZEROCOPY);
2257
2258 return 0;
2259}
2260
2261static int tun_chr_close(struct inode *inode, struct file *file)
2262{
2263 struct tun_file *tfile = file->private_data;
2264
2265 tun_detach(tfile, true);
2266
2267 return 0;
2268}
2269
2270#ifdef CONFIG_PROC_FS
2271static void tun_chr_show_fdinfo(struct seq_file *m, struct file *f)
2272{
2273 struct tun_struct *tun;
2274 struct ifreq ifr;
2275
2276 memset(&ifr, 0, sizeof(ifr));
2277
2278 rtnl_lock();
2279 tun = tun_get(f);
2280 if (tun)
2281 tun_get_iff(current->nsproxy->net_ns, tun, &ifr);
2282 rtnl_unlock();
2283
2284 if (tun)
2285 tun_put(tun);
2286
2287 seq_printf(m, "iff:\t%s\n", ifr.ifr_name);
2288}
2289#endif
2290
2291static const struct file_operations tun_fops = {
2292 .owner = THIS_MODULE,
2293 .llseek = no_llseek,
2294 .read_iter = tun_chr_read_iter,
2295 .write_iter = tun_chr_write_iter,
2296 .poll = tun_chr_poll,
2297 .unlocked_ioctl = tun_chr_ioctl,
2298#ifdef CONFIG_COMPAT
2299 .compat_ioctl = tun_chr_compat_ioctl,
2300#endif
2301 .open = tun_chr_open,
2302 .release = tun_chr_close,
2303 .fasync = tun_chr_fasync,
2304#ifdef CONFIG_PROC_FS
2305 .show_fdinfo = tun_chr_show_fdinfo,
2306#endif
2307};
2308
2309static struct miscdevice tun_miscdev = {
2310 .minor = TUN_MINOR,
2311 .name = "tun",
2312 .nodename = "net/tun",
2313 .fops = &tun_fops,
2314};
2315
2316/* ethtool interface */
2317
2318static int tun_get_settings(struct net_device *dev, struct ethtool_cmd *cmd)
2319{
2320 cmd->supported = 0;
2321 cmd->advertising = 0;
2322 ethtool_cmd_speed_set(cmd, SPEED_10);
2323 cmd->duplex = DUPLEX_FULL;
2324 cmd->port = PORT_TP;
2325 cmd->phy_address = 0;
2326 cmd->transceiver = XCVR_INTERNAL;
2327 cmd->autoneg = AUTONEG_DISABLE;
2328 cmd->maxtxpkt = 0;
2329 cmd->maxrxpkt = 0;
2330 return 0;
2331}
2332
2333static void tun_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
2334{
2335 struct tun_struct *tun = netdev_priv(dev);
2336
2337 strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
2338 strlcpy(info->version, DRV_VERSION, sizeof(info->version));
2339
2340 switch (tun->flags & TUN_TYPE_MASK) {
2341 case IFF_TUN:
2342 strlcpy(info->bus_info, "tun", sizeof(info->bus_info));
2343 break;
2344 case IFF_TAP:
2345 strlcpy(info->bus_info, "tap", sizeof(info->bus_info));
2346 break;
2347 }
2348}
2349
2350static u32 tun_get_msglevel(struct net_device *dev)
2351{
2352#ifdef TUN_DEBUG
2353 struct tun_struct *tun = netdev_priv(dev);
2354 return tun->debug;
2355#else
2356 return -EOPNOTSUPP;
2357#endif
2358}
2359
2360static void tun_set_msglevel(struct net_device *dev, u32 value)
2361{
2362#ifdef TUN_DEBUG
2363 struct tun_struct *tun = netdev_priv(dev);
2364 tun->debug = value;
2365#endif
2366}
2367
2368static const struct ethtool_ops tun_ethtool_ops = {
2369 .get_settings = tun_get_settings,
2370 .get_drvinfo = tun_get_drvinfo,
2371 .get_msglevel = tun_get_msglevel,
2372 .set_msglevel = tun_set_msglevel,
2373 .get_link = ethtool_op_get_link,
2374 .get_ts_info = ethtool_op_get_ts_info,
2375};
2376
2377
2378static int __init tun_init(void)
2379{
2380 int ret = 0;
2381
2382 pr_info("%s, %s\n", DRV_DESCRIPTION, DRV_VERSION);
2383 pr_info("%s\n", DRV_COPYRIGHT);
2384
2385 ret = rtnl_link_register(&tun_link_ops);
2386 if (ret) {
2387 pr_err("Can't register link_ops\n");
2388 goto err_linkops;
2389 }
2390
2391 ret = misc_register(&tun_miscdev);
2392 if (ret) {
2393 pr_err("Can't register misc device %d\n", TUN_MINOR);
2394 goto err_misc;
2395 }
2396 return 0;
2397err_misc:
2398 rtnl_link_unregister(&tun_link_ops);
2399err_linkops:
2400 return ret;
2401}
2402
2403static void tun_cleanup(void)
2404{
2405 misc_deregister(&tun_miscdev);
2406 rtnl_link_unregister(&tun_link_ops);
2407}
2408
2409/* Get an underlying socket object from tun file. Returns error unless file is
2410 * attached to a device. The returned object works like a packet socket, it
2411 * can be used for sock_sendmsg/sock_recvmsg. The caller is responsible for
2412 * holding a reference to the file for as long as the socket is in use. */
2413struct socket *tun_get_socket(struct file *file)
2414{
2415 struct tun_file *tfile;
2416 if (file->f_op != &tun_fops)
2417 return ERR_PTR(-EINVAL);
2418 tfile = file->private_data;
2419 if (!tfile)
2420 return ERR_PTR(-EBADFD);
2421 return &tfile->socket;
2422}
2423EXPORT_SYMBOL_GPL(tun_get_socket);
2424
2425module_init(tun_init);
2426module_exit(tun_cleanup);
2427MODULE_DESCRIPTION(DRV_DESCRIPTION);
2428MODULE_AUTHOR(DRV_COPYRIGHT);
2429MODULE_LICENSE("GPL");
2430MODULE_ALIAS_MISCDEV(TUN_MINOR);
2431MODULE_ALIAS("devname:net/tun");
1/*
2 * TUN - Universal TUN/TAP device driver.
3 * Copyright (C) 1999-2002 Maxim Krasnyansky <maxk@qualcomm.com>
4 *
5 * This program is free software; you can redistribute it and/or modify
6 * it under the terms of the GNU General Public License as published by
7 * the Free Software Foundation; either version 2 of the License, or
8 * (at your option) any later version.
9 *
10 * This program is distributed in the hope that it will be useful,
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
13 * GNU General Public License for more details.
14 *
15 * $Id: tun.c,v 1.15 2002/03/01 02:44:24 maxk Exp $
16 */
17
18/*
19 * Changes:
20 *
21 * Mike Kershaw <dragorn@kismetwireless.net> 2005/08/14
22 * Add TUNSETLINK ioctl to set the link encapsulation
23 *
24 * Mark Smith <markzzzsmith@yahoo.com.au>
25 * Use eth_random_addr() for tap MAC address.
26 *
27 * Harald Roelle <harald.roelle@ifi.lmu.de> 2004/04/20
28 * Fixes in packet dropping, queue length setting and queue wakeup.
29 * Increased default tx queue length.
30 * Added ethtool API.
31 * Minor cleanups
32 *
33 * Daniel Podlejski <underley@underley.eu.org>
34 * Modifications for 2.3.99-pre5 kernel.
35 */
36
37#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
38
39#define DRV_NAME "tun"
40#define DRV_VERSION "1.6"
41#define DRV_DESCRIPTION "Universal TUN/TAP device driver"
42#define DRV_COPYRIGHT "(C) 1999-2004 Max Krasnyansky <maxk@qualcomm.com>"
43
44#include <linux/module.h>
45#include <linux/errno.h>
46#include <linux/kernel.h>
47#include <linux/major.h>
48#include <linux/slab.h>
49#include <linux/poll.h>
50#include <linux/fcntl.h>
51#include <linux/init.h>
52#include <linux/skbuff.h>
53#include <linux/netdevice.h>
54#include <linux/etherdevice.h>
55#include <linux/miscdevice.h>
56#include <linux/ethtool.h>
57#include <linux/rtnetlink.h>
58#include <linux/compat.h>
59#include <linux/if.h>
60#include <linux/if_arp.h>
61#include <linux/if_ether.h>
62#include <linux/if_tun.h>
63#include <linux/if_vlan.h>
64#include <linux/crc32.h>
65#include <linux/nsproxy.h>
66#include <linux/virtio_net.h>
67#include <linux/rcupdate.h>
68#include <net/net_namespace.h>
69#include <net/netns/generic.h>
70#include <net/rtnetlink.h>
71#include <net/sock.h>
72#include <linux/seq_file.h>
73
74#include <asm/uaccess.h>
75
76/* Uncomment to enable debugging */
77/* #define TUN_DEBUG 1 */
78
79#ifdef TUN_DEBUG
80static int debug;
81
82#define tun_debug(level, tun, fmt, args...) \
83do { \
84 if (tun->debug) \
85 netdev_printk(level, tun->dev, fmt, ##args); \
86} while (0)
87#define DBG1(level, fmt, args...) \
88do { \
89 if (debug == 2) \
90 printk(level fmt, ##args); \
91} while (0)
92#else
93#define tun_debug(level, tun, fmt, args...) \
94do { \
95 if (0) \
96 netdev_printk(level, tun->dev, fmt, ##args); \
97} while (0)
98#define DBG1(level, fmt, args...) \
99do { \
100 if (0) \
101 printk(level fmt, ##args); \
102} while (0)
103#endif
104
105#define GOODCOPY_LEN 128
106
107#define FLT_EXACT_COUNT 8
108struct tap_filter {
109 unsigned int count; /* Number of addrs. Zero means disabled */
110 u32 mask[2]; /* Mask of the hashed addrs */
111 unsigned char addr[FLT_EXACT_COUNT][ETH_ALEN];
112};
113
114/* DEFAULT_MAX_NUM_RSS_QUEUES were chosen to let the rx/tx queues allocated for
115 * the netdevice to be fit in one page. So we can make sure the success of
116 * memory allocation. TODO: increase the limit. */
117#define MAX_TAP_QUEUES DEFAULT_MAX_NUM_RSS_QUEUES
118#define MAX_TAP_FLOWS 4096
119
120#define TUN_FLOW_EXPIRE (3 * HZ)
121
122/* A tun_file connects an open character device to a tuntap netdevice. It
123 * also contains all socket related structures (except sock_fprog and tap_filter)
124 * to serve as one transmit queue for tuntap device. The sock_fprog and
125 * tap_filter were kept in tun_struct since they were used for filtering for the
126 * netdevice not for a specific queue (at least I didn't see the requirement for
127 * this).
128 *
129 * RCU usage:
130 * The tun_file and tun_struct are loosely coupled, the pointer from one to the
131 * other can only be read while rcu_read_lock or rtnl_lock is held.
132 */
133struct tun_file {
134 struct sock sk;
135 struct socket socket;
136 struct socket_wq wq;
137 struct tun_struct __rcu *tun;
138 struct net *net;
139 struct fasync_struct *fasync;
140 /* only used for fasnyc */
141 unsigned int flags;
142 union {
143 u16 queue_index;
144 unsigned int ifindex;
145 };
146 struct list_head next;
147 struct tun_struct *detached;
148};
149
150struct tun_flow_entry {
151 struct hlist_node hash_link;
152 struct rcu_head rcu;
153 struct tun_struct *tun;
154
155 u32 rxhash;
156 u32 rps_rxhash;
157 int queue_index;
158 unsigned long updated;
159};
160
161#define TUN_NUM_FLOW_ENTRIES 1024
162
163/* Since the socket were moved to tun_file, to preserve the behavior of persist
164 * device, socket filter, sndbuf and vnet header size were restore when the
165 * file were attached to a persist device.
166 */
167struct tun_struct {
168 struct tun_file __rcu *tfiles[MAX_TAP_QUEUES];
169 unsigned int numqueues;
170 unsigned int flags;
171 kuid_t owner;
172 kgid_t group;
173
174 struct net_device *dev;
175 netdev_features_t set_features;
176#define TUN_USER_FEATURES (NETIF_F_HW_CSUM|NETIF_F_TSO_ECN|NETIF_F_TSO| \
177 NETIF_F_TSO6|NETIF_F_UFO)
178
179 int vnet_hdr_sz;
180 int sndbuf;
181 struct tap_filter txflt;
182 struct sock_fprog fprog;
183 /* protected by rtnl lock */
184 bool filter_attached;
185#ifdef TUN_DEBUG
186 int debug;
187#endif
188 spinlock_t lock;
189 struct hlist_head flows[TUN_NUM_FLOW_ENTRIES];
190 struct timer_list flow_gc_timer;
191 unsigned long ageing_time;
192 unsigned int numdisabled;
193 struct list_head disabled;
194 void *security;
195 u32 flow_count;
196};
197
198static inline u32 tun_hashfn(u32 rxhash)
199{
200 return rxhash & 0x3ff;
201}
202
203static struct tun_flow_entry *tun_flow_find(struct hlist_head *head, u32 rxhash)
204{
205 struct tun_flow_entry *e;
206
207 hlist_for_each_entry_rcu(e, head, hash_link) {
208 if (e->rxhash == rxhash)
209 return e;
210 }
211 return NULL;
212}
213
214static struct tun_flow_entry *tun_flow_create(struct tun_struct *tun,
215 struct hlist_head *head,
216 u32 rxhash, u16 queue_index)
217{
218 struct tun_flow_entry *e = kmalloc(sizeof(*e), GFP_ATOMIC);
219
220 if (e) {
221 tun_debug(KERN_INFO, tun, "create flow: hash %u index %u\n",
222 rxhash, queue_index);
223 e->updated = jiffies;
224 e->rxhash = rxhash;
225 e->rps_rxhash = 0;
226 e->queue_index = queue_index;
227 e->tun = tun;
228 hlist_add_head_rcu(&e->hash_link, head);
229 ++tun->flow_count;
230 }
231 return e;
232}
233
234static void tun_flow_delete(struct tun_struct *tun, struct tun_flow_entry *e)
235{
236 tun_debug(KERN_INFO, tun, "delete flow: hash %u index %u\n",
237 e->rxhash, e->queue_index);
238 sock_rps_reset_flow_hash(e->rps_rxhash);
239 hlist_del_rcu(&e->hash_link);
240 kfree_rcu(e, rcu);
241 --tun->flow_count;
242}
243
244static void tun_flow_flush(struct tun_struct *tun)
245{
246 int i;
247
248 spin_lock_bh(&tun->lock);
249 for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
250 struct tun_flow_entry *e;
251 struct hlist_node *n;
252
253 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link)
254 tun_flow_delete(tun, e);
255 }
256 spin_unlock_bh(&tun->lock);
257}
258
259static void tun_flow_delete_by_queue(struct tun_struct *tun, u16 queue_index)
260{
261 int i;
262
263 spin_lock_bh(&tun->lock);
264 for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
265 struct tun_flow_entry *e;
266 struct hlist_node *n;
267
268 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
269 if (e->queue_index == queue_index)
270 tun_flow_delete(tun, e);
271 }
272 }
273 spin_unlock_bh(&tun->lock);
274}
275
276static void tun_flow_cleanup(unsigned long data)
277{
278 struct tun_struct *tun = (struct tun_struct *)data;
279 unsigned long delay = tun->ageing_time;
280 unsigned long next_timer = jiffies + delay;
281 unsigned long count = 0;
282 int i;
283
284 tun_debug(KERN_INFO, tun, "tun_flow_cleanup\n");
285
286 spin_lock_bh(&tun->lock);
287 for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
288 struct tun_flow_entry *e;
289 struct hlist_node *n;
290
291 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
292 unsigned long this_timer;
293 count++;
294 this_timer = e->updated + delay;
295 if (time_before_eq(this_timer, jiffies))
296 tun_flow_delete(tun, e);
297 else if (time_before(this_timer, next_timer))
298 next_timer = this_timer;
299 }
300 }
301
302 if (count)
303 mod_timer(&tun->flow_gc_timer, round_jiffies_up(next_timer));
304 spin_unlock_bh(&tun->lock);
305}
306
307static void tun_flow_update(struct tun_struct *tun, u32 rxhash,
308 struct tun_file *tfile)
309{
310 struct hlist_head *head;
311 struct tun_flow_entry *e;
312 unsigned long delay = tun->ageing_time;
313 u16 queue_index = tfile->queue_index;
314
315 if (!rxhash)
316 return;
317 else
318 head = &tun->flows[tun_hashfn(rxhash)];
319
320 rcu_read_lock();
321
322 /* We may get a very small possibility of OOO during switching, not
323 * worth to optimize.*/
324 if (tun->numqueues == 1 || tfile->detached)
325 goto unlock;
326
327 e = tun_flow_find(head, rxhash);
328 if (likely(e)) {
329 /* TODO: keep queueing to old queue until it's empty? */
330 e->queue_index = queue_index;
331 e->updated = jiffies;
332 sock_rps_record_flow_hash(e->rps_rxhash);
333 } else {
334 spin_lock_bh(&tun->lock);
335 if (!tun_flow_find(head, rxhash) &&
336 tun->flow_count < MAX_TAP_FLOWS)
337 tun_flow_create(tun, head, rxhash, queue_index);
338
339 if (!timer_pending(&tun->flow_gc_timer))
340 mod_timer(&tun->flow_gc_timer,
341 round_jiffies_up(jiffies + delay));
342 spin_unlock_bh(&tun->lock);
343 }
344
345unlock:
346 rcu_read_unlock();
347}
348
349/**
350 * Save the hash received in the stack receive path and update the
351 * flow_hash table accordingly.
352 */
353static inline void tun_flow_save_rps_rxhash(struct tun_flow_entry *e, u32 hash)
354{
355 if (unlikely(e->rps_rxhash != hash)) {
356 sock_rps_reset_flow_hash(e->rps_rxhash);
357 e->rps_rxhash = hash;
358 }
359}
360
361/* We try to identify a flow through its rxhash first. The reason that
362 * we do not check rxq no. is because some cards(e.g 82599), chooses
363 * the rxq based on the txq where the last packet of the flow comes. As
364 * the userspace application move between processors, we may get a
365 * different rxq no. here. If we could not get rxhash, then we would
366 * hope the rxq no. may help here.
367 */
368static u16 tun_select_queue(struct net_device *dev, struct sk_buff *skb,
369 void *accel_priv, select_queue_fallback_t fallback)
370{
371 struct tun_struct *tun = netdev_priv(dev);
372 struct tun_flow_entry *e;
373 u32 txq = 0;
374 u32 numqueues = 0;
375
376 rcu_read_lock();
377 numqueues = ACCESS_ONCE(tun->numqueues);
378
379 txq = skb_get_hash(skb);
380 if (txq) {
381 e = tun_flow_find(&tun->flows[tun_hashfn(txq)], txq);
382 if (e) {
383 tun_flow_save_rps_rxhash(e, txq);
384 txq = e->queue_index;
385 } else
386 /* use multiply and shift instead of expensive divide */
387 txq = ((u64)txq * numqueues) >> 32;
388 } else if (likely(skb_rx_queue_recorded(skb))) {
389 txq = skb_get_rx_queue(skb);
390 while (unlikely(txq >= numqueues))
391 txq -= numqueues;
392 }
393
394 rcu_read_unlock();
395 return txq;
396}
397
398static inline bool tun_not_capable(struct tun_struct *tun)
399{
400 const struct cred *cred = current_cred();
401 struct net *net = dev_net(tun->dev);
402
403 return ((uid_valid(tun->owner) && !uid_eq(cred->euid, tun->owner)) ||
404 (gid_valid(tun->group) && !in_egroup_p(tun->group))) &&
405 !ns_capable(net->user_ns, CAP_NET_ADMIN);
406}
407
408static void tun_set_real_num_queues(struct tun_struct *tun)
409{
410 netif_set_real_num_tx_queues(tun->dev, tun->numqueues);
411 netif_set_real_num_rx_queues(tun->dev, tun->numqueues);
412}
413
414static void tun_disable_queue(struct tun_struct *tun, struct tun_file *tfile)
415{
416 tfile->detached = tun;
417 list_add_tail(&tfile->next, &tun->disabled);
418 ++tun->numdisabled;
419}
420
421static struct tun_struct *tun_enable_queue(struct tun_file *tfile)
422{
423 struct tun_struct *tun = tfile->detached;
424
425 tfile->detached = NULL;
426 list_del_init(&tfile->next);
427 --tun->numdisabled;
428 return tun;
429}
430
431static void tun_queue_purge(struct tun_file *tfile)
432{
433 skb_queue_purge(&tfile->sk.sk_receive_queue);
434 skb_queue_purge(&tfile->sk.sk_error_queue);
435}
436
437static void __tun_detach(struct tun_file *tfile, bool clean)
438{
439 struct tun_file *ntfile;
440 struct tun_struct *tun;
441
442 tun = rtnl_dereference(tfile->tun);
443
444 if (tun && !tfile->detached) {
445 u16 index = tfile->queue_index;
446 BUG_ON(index >= tun->numqueues);
447
448 rcu_assign_pointer(tun->tfiles[index],
449 tun->tfiles[tun->numqueues - 1]);
450 ntfile = rtnl_dereference(tun->tfiles[index]);
451 ntfile->queue_index = index;
452
453 --tun->numqueues;
454 if (clean) {
455 RCU_INIT_POINTER(tfile->tun, NULL);
456 sock_put(&tfile->sk);
457 } else
458 tun_disable_queue(tun, tfile);
459
460 synchronize_net();
461 tun_flow_delete_by_queue(tun, tun->numqueues + 1);
462 /* Drop read queue */
463 tun_queue_purge(tfile);
464 tun_set_real_num_queues(tun);
465 } else if (tfile->detached && clean) {
466 tun = tun_enable_queue(tfile);
467 sock_put(&tfile->sk);
468 }
469
470 if (clean) {
471 if (tun && tun->numqueues == 0 && tun->numdisabled == 0) {
472 netif_carrier_off(tun->dev);
473
474 if (!(tun->flags & TUN_PERSIST) &&
475 tun->dev->reg_state == NETREG_REGISTERED)
476 unregister_netdevice(tun->dev);
477 }
478
479 BUG_ON(!test_bit(SOCK_EXTERNALLY_ALLOCATED,
480 &tfile->socket.flags));
481 sk_release_kernel(&tfile->sk);
482 }
483}
484
485static void tun_detach(struct tun_file *tfile, bool clean)
486{
487 rtnl_lock();
488 __tun_detach(tfile, clean);
489 rtnl_unlock();
490}
491
492static void tun_detach_all(struct net_device *dev)
493{
494 struct tun_struct *tun = netdev_priv(dev);
495 struct tun_file *tfile, *tmp;
496 int i, n = tun->numqueues;
497
498 for (i = 0; i < n; i++) {
499 tfile = rtnl_dereference(tun->tfiles[i]);
500 BUG_ON(!tfile);
501 wake_up_all(&tfile->wq.wait);
502 RCU_INIT_POINTER(tfile->tun, NULL);
503 --tun->numqueues;
504 }
505 list_for_each_entry(tfile, &tun->disabled, next) {
506 wake_up_all(&tfile->wq.wait);
507 RCU_INIT_POINTER(tfile->tun, NULL);
508 }
509 BUG_ON(tun->numqueues != 0);
510
511 synchronize_net();
512 for (i = 0; i < n; i++) {
513 tfile = rtnl_dereference(tun->tfiles[i]);
514 /* Drop read queue */
515 tun_queue_purge(tfile);
516 sock_put(&tfile->sk);
517 }
518 list_for_each_entry_safe(tfile, tmp, &tun->disabled, next) {
519 tun_enable_queue(tfile);
520 tun_queue_purge(tfile);
521 sock_put(&tfile->sk);
522 }
523 BUG_ON(tun->numdisabled != 0);
524
525 if (tun->flags & TUN_PERSIST)
526 module_put(THIS_MODULE);
527}
528
529static int tun_attach(struct tun_struct *tun, struct file *file, bool skip_filter)
530{
531 struct tun_file *tfile = file->private_data;
532 int err;
533
534 err = security_tun_dev_attach(tfile->socket.sk, tun->security);
535 if (err < 0)
536 goto out;
537
538 err = -EINVAL;
539 if (rtnl_dereference(tfile->tun) && !tfile->detached)
540 goto out;
541
542 err = -EBUSY;
543 if (!(tun->flags & TUN_TAP_MQ) && tun->numqueues == 1)
544 goto out;
545
546 err = -E2BIG;
547 if (!tfile->detached &&
548 tun->numqueues + tun->numdisabled == MAX_TAP_QUEUES)
549 goto out;
550
551 err = 0;
552
553 /* Re-attach the filter to persist device */
554 if (!skip_filter && (tun->filter_attached == true)) {
555 err = sk_attach_filter(&tun->fprog, tfile->socket.sk);
556 if (!err)
557 goto out;
558 }
559 tfile->queue_index = tun->numqueues;
560 rcu_assign_pointer(tfile->tun, tun);
561 rcu_assign_pointer(tun->tfiles[tun->numqueues], tfile);
562 tun->numqueues++;
563
564 if (tfile->detached)
565 tun_enable_queue(tfile);
566 else
567 sock_hold(&tfile->sk);
568
569 tun_set_real_num_queues(tun);
570
571 /* device is allowed to go away first, so no need to hold extra
572 * refcnt.
573 */
574
575out:
576 return err;
577}
578
579static struct tun_struct *__tun_get(struct tun_file *tfile)
580{
581 struct tun_struct *tun;
582
583 rcu_read_lock();
584 tun = rcu_dereference(tfile->tun);
585 if (tun)
586 dev_hold(tun->dev);
587 rcu_read_unlock();
588
589 return tun;
590}
591
592static struct tun_struct *tun_get(struct file *file)
593{
594 return __tun_get(file->private_data);
595}
596
597static void tun_put(struct tun_struct *tun)
598{
599 dev_put(tun->dev);
600}
601
602/* TAP filtering */
603static void addr_hash_set(u32 *mask, const u8 *addr)
604{
605 int n = ether_crc(ETH_ALEN, addr) >> 26;
606 mask[n >> 5] |= (1 << (n & 31));
607}
608
609static unsigned int addr_hash_test(const u32 *mask, const u8 *addr)
610{
611 int n = ether_crc(ETH_ALEN, addr) >> 26;
612 return mask[n >> 5] & (1 << (n & 31));
613}
614
615static int update_filter(struct tap_filter *filter, void __user *arg)
616{
617 struct { u8 u[ETH_ALEN]; } *addr;
618 struct tun_filter uf;
619 int err, alen, n, nexact;
620
621 if (copy_from_user(&uf, arg, sizeof(uf)))
622 return -EFAULT;
623
624 if (!uf.count) {
625 /* Disabled */
626 filter->count = 0;
627 return 0;
628 }
629
630 alen = ETH_ALEN * uf.count;
631 addr = kmalloc(alen, GFP_KERNEL);
632 if (!addr)
633 return -ENOMEM;
634
635 if (copy_from_user(addr, arg + sizeof(uf), alen)) {
636 err = -EFAULT;
637 goto done;
638 }
639
640 /* The filter is updated without holding any locks. Which is
641 * perfectly safe. We disable it first and in the worst
642 * case we'll accept a few undesired packets. */
643 filter->count = 0;
644 wmb();
645
646 /* Use first set of addresses as an exact filter */
647 for (n = 0; n < uf.count && n < FLT_EXACT_COUNT; n++)
648 memcpy(filter->addr[n], addr[n].u, ETH_ALEN);
649
650 nexact = n;
651
652 /* Remaining multicast addresses are hashed,
653 * unicast will leave the filter disabled. */
654 memset(filter->mask, 0, sizeof(filter->mask));
655 for (; n < uf.count; n++) {
656 if (!is_multicast_ether_addr(addr[n].u)) {
657 err = 0; /* no filter */
658 goto done;
659 }
660 addr_hash_set(filter->mask, addr[n].u);
661 }
662
663 /* For ALLMULTI just set the mask to all ones.
664 * This overrides the mask populated above. */
665 if ((uf.flags & TUN_FLT_ALLMULTI))
666 memset(filter->mask, ~0, sizeof(filter->mask));
667
668 /* Now enable the filter */
669 wmb();
670 filter->count = nexact;
671
672 /* Return the number of exact filters */
673 err = nexact;
674
675done:
676 kfree(addr);
677 return err;
678}
679
680/* Returns: 0 - drop, !=0 - accept */
681static int run_filter(struct tap_filter *filter, const struct sk_buff *skb)
682{
683 /* Cannot use eth_hdr(skb) here because skb_mac_hdr() is incorrect
684 * at this point. */
685 struct ethhdr *eh = (struct ethhdr *) skb->data;
686 int i;
687
688 /* Exact match */
689 for (i = 0; i < filter->count; i++)
690 if (ether_addr_equal(eh->h_dest, filter->addr[i]))
691 return 1;
692
693 /* Inexact match (multicast only) */
694 if (is_multicast_ether_addr(eh->h_dest))
695 return addr_hash_test(filter->mask, eh->h_dest);
696
697 return 0;
698}
699
700/*
701 * Checks whether the packet is accepted or not.
702 * Returns: 0 - drop, !=0 - accept
703 */
704static int check_filter(struct tap_filter *filter, const struct sk_buff *skb)
705{
706 if (!filter->count)
707 return 1;
708
709 return run_filter(filter, skb);
710}
711
712/* Network device part of the driver */
713
714static const struct ethtool_ops tun_ethtool_ops;
715
716/* Net device detach from fd. */
717static void tun_net_uninit(struct net_device *dev)
718{
719 tun_detach_all(dev);
720}
721
722/* Net device open. */
723static int tun_net_open(struct net_device *dev)
724{
725 netif_tx_start_all_queues(dev);
726 return 0;
727}
728
729/* Net device close. */
730static int tun_net_close(struct net_device *dev)
731{
732 netif_tx_stop_all_queues(dev);
733 return 0;
734}
735
736/* Net device start xmit */
737static netdev_tx_t tun_net_xmit(struct sk_buff *skb, struct net_device *dev)
738{
739 struct tun_struct *tun = netdev_priv(dev);
740 int txq = skb->queue_mapping;
741 struct tun_file *tfile;
742 u32 numqueues = 0;
743
744 rcu_read_lock();
745 tfile = rcu_dereference(tun->tfiles[txq]);
746 numqueues = ACCESS_ONCE(tun->numqueues);
747
748 /* Drop packet if interface is not attached */
749 if (txq >= numqueues)
750 goto drop;
751
752 if (numqueues == 1) {
753 /* Select queue was not called for the skbuff, so we extract the
754 * RPS hash and save it into the flow_table here.
755 */
756 __u32 rxhash;
757
758 rxhash = skb_get_hash(skb);
759 if (rxhash) {
760 struct tun_flow_entry *e;
761 e = tun_flow_find(&tun->flows[tun_hashfn(rxhash)],
762 rxhash);
763 if (e)
764 tun_flow_save_rps_rxhash(e, rxhash);
765 }
766 }
767
768 tun_debug(KERN_INFO, tun, "tun_net_xmit %d\n", skb->len);
769
770 BUG_ON(!tfile);
771
772 /* Drop if the filter does not like it.
773 * This is a noop if the filter is disabled.
774 * Filter can be enabled only for the TAP devices. */
775 if (!check_filter(&tun->txflt, skb))
776 goto drop;
777
778 if (tfile->socket.sk->sk_filter &&
779 sk_filter(tfile->socket.sk, skb))
780 goto drop;
781
782 /* Limit the number of packets queued by dividing txq length with the
783 * number of queues.
784 */
785 if (skb_queue_len(&tfile->socket.sk->sk_receive_queue) * numqueues
786 >= dev->tx_queue_len)
787 goto drop;
788
789 if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC)))
790 goto drop;
791
792 if (skb->sk) {
793 sock_tx_timestamp(skb->sk, &skb_shinfo(skb)->tx_flags);
794 sw_tx_timestamp(skb);
795 }
796
797 /* Orphan the skb - required as we might hang on to it
798 * for indefinite time.
799 */
800 skb_orphan(skb);
801
802 nf_reset(skb);
803
804 /* Enqueue packet */
805 skb_queue_tail(&tfile->socket.sk->sk_receive_queue, skb);
806
807 /* Notify and wake up reader process */
808 if (tfile->flags & TUN_FASYNC)
809 kill_fasync(&tfile->fasync, SIGIO, POLL_IN);
810 wake_up_interruptible_poll(&tfile->wq.wait, POLLIN |
811 POLLRDNORM | POLLRDBAND);
812
813 rcu_read_unlock();
814 return NETDEV_TX_OK;
815
816drop:
817 dev->stats.tx_dropped++;
818 skb_tx_error(skb);
819 kfree_skb(skb);
820 rcu_read_unlock();
821 return NETDEV_TX_OK;
822}
823
824static void tun_net_mclist(struct net_device *dev)
825{
826 /*
827 * This callback is supposed to deal with mc filter in
828 * _rx_ path and has nothing to do with the _tx_ path.
829 * In rx path we always accept everything userspace gives us.
830 */
831}
832
833#define MIN_MTU 68
834#define MAX_MTU 65535
835
836static int
837tun_net_change_mtu(struct net_device *dev, int new_mtu)
838{
839 if (new_mtu < MIN_MTU || new_mtu + dev->hard_header_len > MAX_MTU)
840 return -EINVAL;
841 dev->mtu = new_mtu;
842 return 0;
843}
844
845static netdev_features_t tun_net_fix_features(struct net_device *dev,
846 netdev_features_t features)
847{
848 struct tun_struct *tun = netdev_priv(dev);
849
850 return (features & tun->set_features) | (features & ~TUN_USER_FEATURES);
851}
852#ifdef CONFIG_NET_POLL_CONTROLLER
853static void tun_poll_controller(struct net_device *dev)
854{
855 /*
856 * Tun only receives frames when:
857 * 1) the char device endpoint gets data from user space
858 * 2) the tun socket gets a sendmsg call from user space
859 * Since both of those are synchronous operations, we are guaranteed
860 * never to have pending data when we poll for it
861 * so there is nothing to do here but return.
862 * We need this though so netpoll recognizes us as an interface that
863 * supports polling, which enables bridge devices in virt setups to
864 * still use netconsole
865 */
866 return;
867}
868#endif
869static const struct net_device_ops tun_netdev_ops = {
870 .ndo_uninit = tun_net_uninit,
871 .ndo_open = tun_net_open,
872 .ndo_stop = tun_net_close,
873 .ndo_start_xmit = tun_net_xmit,
874 .ndo_change_mtu = tun_net_change_mtu,
875 .ndo_fix_features = tun_net_fix_features,
876 .ndo_select_queue = tun_select_queue,
877#ifdef CONFIG_NET_POLL_CONTROLLER
878 .ndo_poll_controller = tun_poll_controller,
879#endif
880};
881
882static const struct net_device_ops tap_netdev_ops = {
883 .ndo_uninit = tun_net_uninit,
884 .ndo_open = tun_net_open,
885 .ndo_stop = tun_net_close,
886 .ndo_start_xmit = tun_net_xmit,
887 .ndo_change_mtu = tun_net_change_mtu,
888 .ndo_fix_features = tun_net_fix_features,
889 .ndo_set_rx_mode = tun_net_mclist,
890 .ndo_set_mac_address = eth_mac_addr,
891 .ndo_validate_addr = eth_validate_addr,
892 .ndo_select_queue = tun_select_queue,
893#ifdef CONFIG_NET_POLL_CONTROLLER
894 .ndo_poll_controller = tun_poll_controller,
895#endif
896};
897
898static void tun_flow_init(struct tun_struct *tun)
899{
900 int i;
901
902 for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++)
903 INIT_HLIST_HEAD(&tun->flows[i]);
904
905 tun->ageing_time = TUN_FLOW_EXPIRE;
906 setup_timer(&tun->flow_gc_timer, tun_flow_cleanup, (unsigned long)tun);
907 mod_timer(&tun->flow_gc_timer,
908 round_jiffies_up(jiffies + tun->ageing_time));
909}
910
911static void tun_flow_uninit(struct tun_struct *tun)
912{
913 del_timer_sync(&tun->flow_gc_timer);
914 tun_flow_flush(tun);
915}
916
917/* Initialize net device. */
918static void tun_net_init(struct net_device *dev)
919{
920 struct tun_struct *tun = netdev_priv(dev);
921
922 switch (tun->flags & TUN_TYPE_MASK) {
923 case TUN_TUN_DEV:
924 dev->netdev_ops = &tun_netdev_ops;
925
926 /* Point-to-Point TUN Device */
927 dev->hard_header_len = 0;
928 dev->addr_len = 0;
929 dev->mtu = 1500;
930
931 /* Zero header length */
932 dev->type = ARPHRD_NONE;
933 dev->flags = IFF_POINTOPOINT | IFF_NOARP | IFF_MULTICAST;
934 dev->tx_queue_len = TUN_READQ_SIZE; /* We prefer our own queue length */
935 break;
936
937 case TUN_TAP_DEV:
938 dev->netdev_ops = &tap_netdev_ops;
939 /* Ethernet TAP Device */
940 ether_setup(dev);
941 dev->priv_flags &= ~IFF_TX_SKB_SHARING;
942 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
943
944 eth_hw_addr_random(dev);
945
946 dev->tx_queue_len = TUN_READQ_SIZE; /* We prefer our own queue length */
947 break;
948 }
949}
950
951/* Character device part */
952
953/* Poll */
954static unsigned int tun_chr_poll(struct file *file, poll_table *wait)
955{
956 struct tun_file *tfile = file->private_data;
957 struct tun_struct *tun = __tun_get(tfile);
958 struct sock *sk;
959 unsigned int mask = 0;
960
961 if (!tun)
962 return POLLERR;
963
964 sk = tfile->socket.sk;
965
966 tun_debug(KERN_INFO, tun, "tun_chr_poll\n");
967
968 poll_wait(file, &tfile->wq.wait, wait);
969
970 if (!skb_queue_empty(&sk->sk_receive_queue))
971 mask |= POLLIN | POLLRDNORM;
972
973 if (sock_writeable(sk) ||
974 (!test_and_set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags) &&
975 sock_writeable(sk)))
976 mask |= POLLOUT | POLLWRNORM;
977
978 if (tun->dev->reg_state != NETREG_REGISTERED)
979 mask = POLLERR;
980
981 tun_put(tun);
982 return mask;
983}
984
985/* prepad is the amount to reserve at front. len is length after that.
986 * linear is a hint as to how much to copy (usually headers). */
987static struct sk_buff *tun_alloc_skb(struct tun_file *tfile,
988 size_t prepad, size_t len,
989 size_t linear, int noblock)
990{
991 struct sock *sk = tfile->socket.sk;
992 struct sk_buff *skb;
993 int err;
994
995 /* Under a page? Don't bother with paged skb. */
996 if (prepad + len < PAGE_SIZE || !linear)
997 linear = len;
998
999 skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
1000 &err, 0);
1001 if (!skb)
1002 return ERR_PTR(err);
1003
1004 skb_reserve(skb, prepad);
1005 skb_put(skb, linear);
1006 skb->data_len = len - linear;
1007 skb->len += len - linear;
1008
1009 return skb;
1010}
1011
1012/* Get packet from user space buffer */
1013static ssize_t tun_get_user(struct tun_struct *tun, struct tun_file *tfile,
1014 void *msg_control, const struct iovec *iv,
1015 size_t total_len, size_t count, int noblock)
1016{
1017 struct tun_pi pi = { 0, cpu_to_be16(ETH_P_IP) };
1018 struct sk_buff *skb;
1019 size_t len = total_len, align = NET_SKB_PAD, linear;
1020 struct virtio_net_hdr gso = { 0 };
1021 int good_linear;
1022 int offset = 0;
1023 int copylen;
1024 bool zerocopy = false;
1025 int err;
1026 u32 rxhash;
1027
1028 if (!(tun->flags & TUN_NO_PI)) {
1029 if (len < sizeof(pi))
1030 return -EINVAL;
1031 len -= sizeof(pi);
1032
1033 if (memcpy_fromiovecend((void *)&pi, iv, 0, sizeof(pi)))
1034 return -EFAULT;
1035 offset += sizeof(pi);
1036 }
1037
1038 if (tun->flags & TUN_VNET_HDR) {
1039 if (len < tun->vnet_hdr_sz)
1040 return -EINVAL;
1041 len -= tun->vnet_hdr_sz;
1042
1043 if (memcpy_fromiovecend((void *)&gso, iv, offset, sizeof(gso)))
1044 return -EFAULT;
1045
1046 if ((gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
1047 gso.csum_start + gso.csum_offset + 2 > gso.hdr_len)
1048 gso.hdr_len = gso.csum_start + gso.csum_offset + 2;
1049
1050 if (gso.hdr_len > len)
1051 return -EINVAL;
1052 offset += tun->vnet_hdr_sz;
1053 }
1054
1055 if ((tun->flags & TUN_TYPE_MASK) == TUN_TAP_DEV) {
1056 align += NET_IP_ALIGN;
1057 if (unlikely(len < ETH_HLEN ||
1058 (gso.hdr_len && gso.hdr_len < ETH_HLEN)))
1059 return -EINVAL;
1060 }
1061
1062 good_linear = SKB_MAX_HEAD(align);
1063
1064 if (msg_control) {
1065 /* There are 256 bytes to be copied in skb, so there is
1066 * enough room for skb expand head in case it is used.
1067 * The rest of the buffer is mapped from userspace.
1068 */
1069 copylen = gso.hdr_len ? gso.hdr_len : GOODCOPY_LEN;
1070 if (copylen > good_linear)
1071 copylen = good_linear;
1072 linear = copylen;
1073 if (iov_pages(iv, offset + copylen, count) <= MAX_SKB_FRAGS)
1074 zerocopy = true;
1075 }
1076
1077 if (!zerocopy) {
1078 copylen = len;
1079 if (gso.hdr_len > good_linear)
1080 linear = good_linear;
1081 else
1082 linear = gso.hdr_len;
1083 }
1084
1085 skb = tun_alloc_skb(tfile, align, copylen, linear, noblock);
1086 if (IS_ERR(skb)) {
1087 if (PTR_ERR(skb) != -EAGAIN)
1088 tun->dev->stats.rx_dropped++;
1089 return PTR_ERR(skb);
1090 }
1091
1092 if (zerocopy)
1093 err = zerocopy_sg_from_iovec(skb, iv, offset, count);
1094 else {
1095 err = skb_copy_datagram_from_iovec(skb, 0, iv, offset, len);
1096 if (!err && msg_control) {
1097 struct ubuf_info *uarg = msg_control;
1098 uarg->callback(uarg, false);
1099 }
1100 }
1101
1102 if (err) {
1103 tun->dev->stats.rx_dropped++;
1104 kfree_skb(skb);
1105 return -EFAULT;
1106 }
1107
1108 if (gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
1109 if (!skb_partial_csum_set(skb, gso.csum_start,
1110 gso.csum_offset)) {
1111 tun->dev->stats.rx_frame_errors++;
1112 kfree_skb(skb);
1113 return -EINVAL;
1114 }
1115 }
1116
1117 switch (tun->flags & TUN_TYPE_MASK) {
1118 case TUN_TUN_DEV:
1119 if (tun->flags & TUN_NO_PI) {
1120 switch (skb->data[0] & 0xf0) {
1121 case 0x40:
1122 pi.proto = htons(ETH_P_IP);
1123 break;
1124 case 0x60:
1125 pi.proto = htons(ETH_P_IPV6);
1126 break;
1127 default:
1128 tun->dev->stats.rx_dropped++;
1129 kfree_skb(skb);
1130 return -EINVAL;
1131 }
1132 }
1133
1134 skb_reset_mac_header(skb);
1135 skb->protocol = pi.proto;
1136 skb->dev = tun->dev;
1137 break;
1138 case TUN_TAP_DEV:
1139 skb->protocol = eth_type_trans(skb, tun->dev);
1140 break;
1141 }
1142
1143 if (gso.gso_type != VIRTIO_NET_HDR_GSO_NONE) {
1144 pr_debug("GSO!\n");
1145 switch (gso.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
1146 case VIRTIO_NET_HDR_GSO_TCPV4:
1147 skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
1148 break;
1149 case VIRTIO_NET_HDR_GSO_TCPV6:
1150 skb_shinfo(skb)->gso_type = SKB_GSO_TCPV6;
1151 break;
1152 case VIRTIO_NET_HDR_GSO_UDP:
1153 skb_shinfo(skb)->gso_type = SKB_GSO_UDP;
1154 break;
1155 default:
1156 tun->dev->stats.rx_frame_errors++;
1157 kfree_skb(skb);
1158 return -EINVAL;
1159 }
1160
1161 if (gso.gso_type & VIRTIO_NET_HDR_GSO_ECN)
1162 skb_shinfo(skb)->gso_type |= SKB_GSO_TCP_ECN;
1163
1164 skb_shinfo(skb)->gso_size = gso.gso_size;
1165 if (skb_shinfo(skb)->gso_size == 0) {
1166 tun->dev->stats.rx_frame_errors++;
1167 kfree_skb(skb);
1168 return -EINVAL;
1169 }
1170
1171 /* Header must be checked, and gso_segs computed. */
1172 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
1173 skb_shinfo(skb)->gso_segs = 0;
1174 }
1175
1176 /* copy skb_ubuf_info for callback when skb has no error */
1177 if (zerocopy) {
1178 skb_shinfo(skb)->destructor_arg = msg_control;
1179 skb_shinfo(skb)->tx_flags |= SKBTX_DEV_ZEROCOPY;
1180 skb_shinfo(skb)->tx_flags |= SKBTX_SHARED_FRAG;
1181 }
1182
1183 skb_reset_network_header(skb);
1184 skb_probe_transport_header(skb, 0);
1185
1186 rxhash = skb_get_hash(skb);
1187 netif_rx_ni(skb);
1188
1189 tun->dev->stats.rx_packets++;
1190 tun->dev->stats.rx_bytes += len;
1191
1192 tun_flow_update(tun, rxhash, tfile);
1193 return total_len;
1194}
1195
1196static ssize_t tun_chr_aio_write(struct kiocb *iocb, const struct iovec *iv,
1197 unsigned long count, loff_t pos)
1198{
1199 struct file *file = iocb->ki_filp;
1200 struct tun_struct *tun = tun_get(file);
1201 struct tun_file *tfile = file->private_data;
1202 ssize_t result;
1203
1204 if (!tun)
1205 return -EBADFD;
1206
1207 tun_debug(KERN_INFO, tun, "tun_chr_write %ld\n", count);
1208
1209 result = tun_get_user(tun, tfile, NULL, iv, iov_length(iv, count),
1210 count, file->f_flags & O_NONBLOCK);
1211
1212 tun_put(tun);
1213 return result;
1214}
1215
1216/* Put packet to the user space buffer */
1217static ssize_t tun_put_user(struct tun_struct *tun,
1218 struct tun_file *tfile,
1219 struct sk_buff *skb,
1220 const struct iovec *iv, int len)
1221{
1222 struct tun_pi pi = { 0, skb->protocol };
1223 ssize_t total = 0;
1224 int vlan_offset = 0, copied;
1225
1226 if (!(tun->flags & TUN_NO_PI)) {
1227 if ((len -= sizeof(pi)) < 0)
1228 return -EINVAL;
1229
1230 if (len < skb->len) {
1231 /* Packet will be striped */
1232 pi.flags |= TUN_PKT_STRIP;
1233 }
1234
1235 if (memcpy_toiovecend(iv, (void *) &pi, 0, sizeof(pi)))
1236 return -EFAULT;
1237 total += sizeof(pi);
1238 }
1239
1240 if (tun->flags & TUN_VNET_HDR) {
1241 struct virtio_net_hdr gso = { 0 }; /* no info leak */
1242 if ((len -= tun->vnet_hdr_sz) < 0)
1243 return -EINVAL;
1244
1245 if (skb_is_gso(skb)) {
1246 struct skb_shared_info *sinfo = skb_shinfo(skb);
1247
1248 /* This is a hint as to how much should be linear. */
1249 gso.hdr_len = skb_headlen(skb);
1250 gso.gso_size = sinfo->gso_size;
1251 if (sinfo->gso_type & SKB_GSO_TCPV4)
1252 gso.gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
1253 else if (sinfo->gso_type & SKB_GSO_TCPV6)
1254 gso.gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
1255 else if (sinfo->gso_type & SKB_GSO_UDP)
1256 gso.gso_type = VIRTIO_NET_HDR_GSO_UDP;
1257 else {
1258 pr_err("unexpected GSO type: "
1259 "0x%x, gso_size %d, hdr_len %d\n",
1260 sinfo->gso_type, gso.gso_size,
1261 gso.hdr_len);
1262 print_hex_dump(KERN_ERR, "tun: ",
1263 DUMP_PREFIX_NONE,
1264 16, 1, skb->head,
1265 min((int)gso.hdr_len, 64), true);
1266 WARN_ON_ONCE(1);
1267 return -EINVAL;
1268 }
1269 if (sinfo->gso_type & SKB_GSO_TCP_ECN)
1270 gso.gso_type |= VIRTIO_NET_HDR_GSO_ECN;
1271 } else
1272 gso.gso_type = VIRTIO_NET_HDR_GSO_NONE;
1273
1274 if (skb->ip_summed == CHECKSUM_PARTIAL) {
1275 gso.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
1276 gso.csum_start = skb_checksum_start_offset(skb);
1277 gso.csum_offset = skb->csum_offset;
1278 } else if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
1279 gso.flags = VIRTIO_NET_HDR_F_DATA_VALID;
1280 } /* else everything is zero */
1281
1282 if (unlikely(memcpy_toiovecend(iv, (void *)&gso, total,
1283 sizeof(gso))))
1284 return -EFAULT;
1285 total += tun->vnet_hdr_sz;
1286 }
1287
1288 copied = total;
1289 total += skb->len;
1290 if (!vlan_tx_tag_present(skb)) {
1291 len = min_t(int, skb->len, len);
1292 } else {
1293 int copy, ret;
1294 struct {
1295 __be16 h_vlan_proto;
1296 __be16 h_vlan_TCI;
1297 } veth;
1298
1299 veth.h_vlan_proto = skb->vlan_proto;
1300 veth.h_vlan_TCI = htons(vlan_tx_tag_get(skb));
1301
1302 vlan_offset = offsetof(struct vlan_ethhdr, h_vlan_proto);
1303 len = min_t(int, skb->len + VLAN_HLEN, len);
1304 total += VLAN_HLEN;
1305
1306 copy = min_t(int, vlan_offset, len);
1307 ret = skb_copy_datagram_const_iovec(skb, 0, iv, copied, copy);
1308 len -= copy;
1309 copied += copy;
1310 if (ret || !len)
1311 goto done;
1312
1313 copy = min_t(int, sizeof(veth), len);
1314 ret = memcpy_toiovecend(iv, (void *)&veth, copied, copy);
1315 len -= copy;
1316 copied += copy;
1317 if (ret || !len)
1318 goto done;
1319 }
1320
1321 skb_copy_datagram_const_iovec(skb, vlan_offset, iv, copied, len);
1322
1323done:
1324 tun->dev->stats.tx_packets++;
1325 tun->dev->stats.tx_bytes += len;
1326
1327 return total;
1328}
1329
1330static ssize_t tun_do_read(struct tun_struct *tun, struct tun_file *tfile,
1331 const struct iovec *iv, ssize_t len, int noblock)
1332{
1333 DECLARE_WAITQUEUE(wait, current);
1334 struct sk_buff *skb;
1335 ssize_t ret = 0;
1336
1337 tun_debug(KERN_INFO, tun, "tun_do_read\n");
1338
1339 if (unlikely(!noblock))
1340 add_wait_queue(&tfile->wq.wait, &wait);
1341 while (len) {
1342 if (unlikely(!noblock))
1343 current->state = TASK_INTERRUPTIBLE;
1344
1345 /* Read frames from the queue */
1346 if (!(skb = skb_dequeue(&tfile->socket.sk->sk_receive_queue))) {
1347 if (noblock) {
1348 ret = -EAGAIN;
1349 break;
1350 }
1351 if (signal_pending(current)) {
1352 ret = -ERESTARTSYS;
1353 break;
1354 }
1355 if (tun->dev->reg_state != NETREG_REGISTERED) {
1356 ret = -EIO;
1357 break;
1358 }
1359
1360 /* Nothing to read, let's sleep */
1361 schedule();
1362 continue;
1363 }
1364
1365 ret = tun_put_user(tun, tfile, skb, iv, len);
1366 kfree_skb(skb);
1367 break;
1368 }
1369
1370 if (unlikely(!noblock)) {
1371 current->state = TASK_RUNNING;
1372 remove_wait_queue(&tfile->wq.wait, &wait);
1373 }
1374
1375 return ret;
1376}
1377
1378static ssize_t tun_chr_aio_read(struct kiocb *iocb, const struct iovec *iv,
1379 unsigned long count, loff_t pos)
1380{
1381 struct file *file = iocb->ki_filp;
1382 struct tun_file *tfile = file->private_data;
1383 struct tun_struct *tun = __tun_get(tfile);
1384 ssize_t len, ret;
1385
1386 if (!tun)
1387 return -EBADFD;
1388 len = iov_length(iv, count);
1389 if (len < 0) {
1390 ret = -EINVAL;
1391 goto out;
1392 }
1393
1394 ret = tun_do_read(tun, tfile, iv, len,
1395 file->f_flags & O_NONBLOCK);
1396 ret = min_t(ssize_t, ret, len);
1397 if (ret > 0)
1398 iocb->ki_pos = ret;
1399out:
1400 tun_put(tun);
1401 return ret;
1402}
1403
1404static void tun_free_netdev(struct net_device *dev)
1405{
1406 struct tun_struct *tun = netdev_priv(dev);
1407
1408 BUG_ON(!(list_empty(&tun->disabled)));
1409 tun_flow_uninit(tun);
1410 security_tun_dev_free_security(tun->security);
1411 free_netdev(dev);
1412}
1413
1414static void tun_setup(struct net_device *dev)
1415{
1416 struct tun_struct *tun = netdev_priv(dev);
1417
1418 tun->owner = INVALID_UID;
1419 tun->group = INVALID_GID;
1420
1421 dev->ethtool_ops = &tun_ethtool_ops;
1422 dev->destructor = tun_free_netdev;
1423}
1424
1425/* Trivial set of netlink ops to allow deleting tun or tap
1426 * device with netlink.
1427 */
1428static int tun_validate(struct nlattr *tb[], struct nlattr *data[])
1429{
1430 return -EINVAL;
1431}
1432
1433static struct rtnl_link_ops tun_link_ops __read_mostly = {
1434 .kind = DRV_NAME,
1435 .priv_size = sizeof(struct tun_struct),
1436 .setup = tun_setup,
1437 .validate = tun_validate,
1438};
1439
1440static void tun_sock_write_space(struct sock *sk)
1441{
1442 struct tun_file *tfile;
1443 wait_queue_head_t *wqueue;
1444
1445 if (!sock_writeable(sk))
1446 return;
1447
1448 if (!test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags))
1449 return;
1450
1451 wqueue = sk_sleep(sk);
1452 if (wqueue && waitqueue_active(wqueue))
1453 wake_up_interruptible_sync_poll(wqueue, POLLOUT |
1454 POLLWRNORM | POLLWRBAND);
1455
1456 tfile = container_of(sk, struct tun_file, sk);
1457 kill_fasync(&tfile->fasync, SIGIO, POLL_OUT);
1458}
1459
1460static int tun_sendmsg(struct kiocb *iocb, struct socket *sock,
1461 struct msghdr *m, size_t total_len)
1462{
1463 int ret;
1464 struct tun_file *tfile = container_of(sock, struct tun_file, socket);
1465 struct tun_struct *tun = __tun_get(tfile);
1466
1467 if (!tun)
1468 return -EBADFD;
1469 ret = tun_get_user(tun, tfile, m->msg_control, m->msg_iov, total_len,
1470 m->msg_iovlen, m->msg_flags & MSG_DONTWAIT);
1471 tun_put(tun);
1472 return ret;
1473}
1474
1475static int tun_recvmsg(struct kiocb *iocb, struct socket *sock,
1476 struct msghdr *m, size_t total_len,
1477 int flags)
1478{
1479 struct tun_file *tfile = container_of(sock, struct tun_file, socket);
1480 struct tun_struct *tun = __tun_get(tfile);
1481 int ret;
1482
1483 if (!tun)
1484 return -EBADFD;
1485
1486 if (flags & ~(MSG_DONTWAIT|MSG_TRUNC|MSG_ERRQUEUE)) {
1487 ret = -EINVAL;
1488 goto out;
1489 }
1490 if (flags & MSG_ERRQUEUE) {
1491 ret = sock_recv_errqueue(sock->sk, m, total_len,
1492 SOL_PACKET, TUN_TX_TIMESTAMP);
1493 goto out;
1494 }
1495 ret = tun_do_read(tun, tfile, m->msg_iov, total_len,
1496 flags & MSG_DONTWAIT);
1497 if (ret > total_len) {
1498 m->msg_flags |= MSG_TRUNC;
1499 ret = flags & MSG_TRUNC ? ret : total_len;
1500 }
1501out:
1502 tun_put(tun);
1503 return ret;
1504}
1505
1506static int tun_release(struct socket *sock)
1507{
1508 if (sock->sk)
1509 sock_put(sock->sk);
1510 return 0;
1511}
1512
1513/* Ops structure to mimic raw sockets with tun */
1514static const struct proto_ops tun_socket_ops = {
1515 .sendmsg = tun_sendmsg,
1516 .recvmsg = tun_recvmsg,
1517 .release = tun_release,
1518};
1519
1520static struct proto tun_proto = {
1521 .name = "tun",
1522 .owner = THIS_MODULE,
1523 .obj_size = sizeof(struct tun_file),
1524};
1525
1526static int tun_flags(struct tun_struct *tun)
1527{
1528 int flags = 0;
1529
1530 if (tun->flags & TUN_TUN_DEV)
1531 flags |= IFF_TUN;
1532 else
1533 flags |= IFF_TAP;
1534
1535 if (tun->flags & TUN_NO_PI)
1536 flags |= IFF_NO_PI;
1537
1538 /* This flag has no real effect. We track the value for backwards
1539 * compatibility.
1540 */
1541 if (tun->flags & TUN_ONE_QUEUE)
1542 flags |= IFF_ONE_QUEUE;
1543
1544 if (tun->flags & TUN_VNET_HDR)
1545 flags |= IFF_VNET_HDR;
1546
1547 if (tun->flags & TUN_TAP_MQ)
1548 flags |= IFF_MULTI_QUEUE;
1549
1550 if (tun->flags & TUN_PERSIST)
1551 flags |= IFF_PERSIST;
1552
1553 return flags;
1554}
1555
1556static ssize_t tun_show_flags(struct device *dev, struct device_attribute *attr,
1557 char *buf)
1558{
1559 struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1560 return sprintf(buf, "0x%x\n", tun_flags(tun));
1561}
1562
1563static ssize_t tun_show_owner(struct device *dev, struct device_attribute *attr,
1564 char *buf)
1565{
1566 struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1567 return uid_valid(tun->owner)?
1568 sprintf(buf, "%u\n",
1569 from_kuid_munged(current_user_ns(), tun->owner)):
1570 sprintf(buf, "-1\n");
1571}
1572
1573static ssize_t tun_show_group(struct device *dev, struct device_attribute *attr,
1574 char *buf)
1575{
1576 struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1577 return gid_valid(tun->group) ?
1578 sprintf(buf, "%u\n",
1579 from_kgid_munged(current_user_ns(), tun->group)):
1580 sprintf(buf, "-1\n");
1581}
1582
1583static DEVICE_ATTR(tun_flags, 0444, tun_show_flags, NULL);
1584static DEVICE_ATTR(owner, 0444, tun_show_owner, NULL);
1585static DEVICE_ATTR(group, 0444, tun_show_group, NULL);
1586
1587static int tun_set_iff(struct net *net, struct file *file, struct ifreq *ifr)
1588{
1589 struct tun_struct *tun;
1590 struct tun_file *tfile = file->private_data;
1591 struct net_device *dev;
1592 int err;
1593
1594 if (tfile->detached)
1595 return -EINVAL;
1596
1597 dev = __dev_get_by_name(net, ifr->ifr_name);
1598 if (dev) {
1599 if (ifr->ifr_flags & IFF_TUN_EXCL)
1600 return -EBUSY;
1601 if ((ifr->ifr_flags & IFF_TUN) && dev->netdev_ops == &tun_netdev_ops)
1602 tun = netdev_priv(dev);
1603 else if ((ifr->ifr_flags & IFF_TAP) && dev->netdev_ops == &tap_netdev_ops)
1604 tun = netdev_priv(dev);
1605 else
1606 return -EINVAL;
1607
1608 if (!!(ifr->ifr_flags & IFF_MULTI_QUEUE) !=
1609 !!(tun->flags & TUN_TAP_MQ))
1610 return -EINVAL;
1611
1612 if (tun_not_capable(tun))
1613 return -EPERM;
1614 err = security_tun_dev_open(tun->security);
1615 if (err < 0)
1616 return err;
1617
1618 err = tun_attach(tun, file, ifr->ifr_flags & IFF_NOFILTER);
1619 if (err < 0)
1620 return err;
1621
1622 if (tun->flags & TUN_TAP_MQ &&
1623 (tun->numqueues + tun->numdisabled > 1)) {
1624 /* One or more queue has already been attached, no need
1625 * to initialize the device again.
1626 */
1627 return 0;
1628 }
1629 }
1630 else {
1631 char *name;
1632 unsigned long flags = 0;
1633 int queues = ifr->ifr_flags & IFF_MULTI_QUEUE ?
1634 MAX_TAP_QUEUES : 1;
1635
1636 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
1637 return -EPERM;
1638 err = security_tun_dev_create();
1639 if (err < 0)
1640 return err;
1641
1642 /* Set dev type */
1643 if (ifr->ifr_flags & IFF_TUN) {
1644 /* TUN device */
1645 flags |= TUN_TUN_DEV;
1646 name = "tun%d";
1647 } else if (ifr->ifr_flags & IFF_TAP) {
1648 /* TAP device */
1649 flags |= TUN_TAP_DEV;
1650 name = "tap%d";
1651 } else
1652 return -EINVAL;
1653
1654 if (*ifr->ifr_name)
1655 name = ifr->ifr_name;
1656
1657 dev = alloc_netdev_mqs(sizeof(struct tun_struct), name,
1658 tun_setup, queues, queues);
1659
1660 if (!dev)
1661 return -ENOMEM;
1662
1663 dev_net_set(dev, net);
1664 dev->rtnl_link_ops = &tun_link_ops;
1665 dev->ifindex = tfile->ifindex;
1666
1667 tun = netdev_priv(dev);
1668 tun->dev = dev;
1669 tun->flags = flags;
1670 tun->txflt.count = 0;
1671 tun->vnet_hdr_sz = sizeof(struct virtio_net_hdr);
1672
1673 tun->filter_attached = false;
1674 tun->sndbuf = tfile->socket.sk->sk_sndbuf;
1675
1676 spin_lock_init(&tun->lock);
1677
1678 err = security_tun_dev_alloc_security(&tun->security);
1679 if (err < 0)
1680 goto err_free_dev;
1681
1682 tun_net_init(dev);
1683 tun_flow_init(tun);
1684
1685 dev->hw_features = NETIF_F_SG | NETIF_F_FRAGLIST |
1686 TUN_USER_FEATURES | NETIF_F_HW_VLAN_CTAG_TX |
1687 NETIF_F_HW_VLAN_STAG_TX;
1688 dev->features = dev->hw_features;
1689 dev->vlan_features = dev->features &
1690 ~(NETIF_F_HW_VLAN_CTAG_TX |
1691 NETIF_F_HW_VLAN_STAG_TX);
1692
1693 INIT_LIST_HEAD(&tun->disabled);
1694 err = tun_attach(tun, file, false);
1695 if (err < 0)
1696 goto err_free_flow;
1697
1698 err = register_netdevice(tun->dev);
1699 if (err < 0)
1700 goto err_detach;
1701
1702 if (device_create_file(&tun->dev->dev, &dev_attr_tun_flags) ||
1703 device_create_file(&tun->dev->dev, &dev_attr_owner) ||
1704 device_create_file(&tun->dev->dev, &dev_attr_group))
1705 pr_err("Failed to create tun sysfs files\n");
1706 }
1707
1708 netif_carrier_on(tun->dev);
1709
1710 tun_debug(KERN_INFO, tun, "tun_set_iff\n");
1711
1712 if (ifr->ifr_flags & IFF_NO_PI)
1713 tun->flags |= TUN_NO_PI;
1714 else
1715 tun->flags &= ~TUN_NO_PI;
1716
1717 /* This flag has no real effect. We track the value for backwards
1718 * compatibility.
1719 */
1720 if (ifr->ifr_flags & IFF_ONE_QUEUE)
1721 tun->flags |= TUN_ONE_QUEUE;
1722 else
1723 tun->flags &= ~TUN_ONE_QUEUE;
1724
1725 if (ifr->ifr_flags & IFF_VNET_HDR)
1726 tun->flags |= TUN_VNET_HDR;
1727 else
1728 tun->flags &= ~TUN_VNET_HDR;
1729
1730 if (ifr->ifr_flags & IFF_MULTI_QUEUE)
1731 tun->flags |= TUN_TAP_MQ;
1732 else
1733 tun->flags &= ~TUN_TAP_MQ;
1734
1735 /* Make sure persistent devices do not get stuck in
1736 * xoff state.
1737 */
1738 if (netif_running(tun->dev))
1739 netif_tx_wake_all_queues(tun->dev);
1740
1741 strcpy(ifr->ifr_name, tun->dev->name);
1742 return 0;
1743
1744err_detach:
1745 tun_detach_all(dev);
1746err_free_flow:
1747 tun_flow_uninit(tun);
1748 security_tun_dev_free_security(tun->security);
1749err_free_dev:
1750 free_netdev(dev);
1751 return err;
1752}
1753
1754static void tun_get_iff(struct net *net, struct tun_struct *tun,
1755 struct ifreq *ifr)
1756{
1757 tun_debug(KERN_INFO, tun, "tun_get_iff\n");
1758
1759 strcpy(ifr->ifr_name, tun->dev->name);
1760
1761 ifr->ifr_flags = tun_flags(tun);
1762
1763}
1764
1765/* This is like a cut-down ethtool ops, except done via tun fd so no
1766 * privs required. */
1767static int set_offload(struct tun_struct *tun, unsigned long arg)
1768{
1769 netdev_features_t features = 0;
1770
1771 if (arg & TUN_F_CSUM) {
1772 features |= NETIF_F_HW_CSUM;
1773 arg &= ~TUN_F_CSUM;
1774
1775 if (arg & (TUN_F_TSO4|TUN_F_TSO6)) {
1776 if (arg & TUN_F_TSO_ECN) {
1777 features |= NETIF_F_TSO_ECN;
1778 arg &= ~TUN_F_TSO_ECN;
1779 }
1780 if (arg & TUN_F_TSO4)
1781 features |= NETIF_F_TSO;
1782 if (arg & TUN_F_TSO6)
1783 features |= NETIF_F_TSO6;
1784 arg &= ~(TUN_F_TSO4|TUN_F_TSO6);
1785 }
1786
1787 if (arg & TUN_F_UFO) {
1788 features |= NETIF_F_UFO;
1789 arg &= ~TUN_F_UFO;
1790 }
1791 }
1792
1793 /* This gives the user a way to test for new features in future by
1794 * trying to set them. */
1795 if (arg)
1796 return -EINVAL;
1797
1798 tun->set_features = features;
1799 netdev_update_features(tun->dev);
1800
1801 return 0;
1802}
1803
1804static void tun_detach_filter(struct tun_struct *tun, int n)
1805{
1806 int i;
1807 struct tun_file *tfile;
1808
1809 for (i = 0; i < n; i++) {
1810 tfile = rtnl_dereference(tun->tfiles[i]);
1811 sk_detach_filter(tfile->socket.sk);
1812 }
1813
1814 tun->filter_attached = false;
1815}
1816
1817static int tun_attach_filter(struct tun_struct *tun)
1818{
1819 int i, ret = 0;
1820 struct tun_file *tfile;
1821
1822 for (i = 0; i < tun->numqueues; i++) {
1823 tfile = rtnl_dereference(tun->tfiles[i]);
1824 ret = sk_attach_filter(&tun->fprog, tfile->socket.sk);
1825 if (ret) {
1826 tun_detach_filter(tun, i);
1827 return ret;
1828 }
1829 }
1830
1831 tun->filter_attached = true;
1832 return ret;
1833}
1834
1835static void tun_set_sndbuf(struct tun_struct *tun)
1836{
1837 struct tun_file *tfile;
1838 int i;
1839
1840 for (i = 0; i < tun->numqueues; i++) {
1841 tfile = rtnl_dereference(tun->tfiles[i]);
1842 tfile->socket.sk->sk_sndbuf = tun->sndbuf;
1843 }
1844}
1845
1846static int tun_set_queue(struct file *file, struct ifreq *ifr)
1847{
1848 struct tun_file *tfile = file->private_data;
1849 struct tun_struct *tun;
1850 int ret = 0;
1851
1852 rtnl_lock();
1853
1854 if (ifr->ifr_flags & IFF_ATTACH_QUEUE) {
1855 tun = tfile->detached;
1856 if (!tun) {
1857 ret = -EINVAL;
1858 goto unlock;
1859 }
1860 ret = security_tun_dev_attach_queue(tun->security);
1861 if (ret < 0)
1862 goto unlock;
1863 ret = tun_attach(tun, file, false);
1864 } else if (ifr->ifr_flags & IFF_DETACH_QUEUE) {
1865 tun = rtnl_dereference(tfile->tun);
1866 if (!tun || !(tun->flags & TUN_TAP_MQ) || tfile->detached)
1867 ret = -EINVAL;
1868 else
1869 __tun_detach(tfile, false);
1870 } else
1871 ret = -EINVAL;
1872
1873unlock:
1874 rtnl_unlock();
1875 return ret;
1876}
1877
1878static long __tun_chr_ioctl(struct file *file, unsigned int cmd,
1879 unsigned long arg, int ifreq_len)
1880{
1881 struct tun_file *tfile = file->private_data;
1882 struct tun_struct *tun;
1883 void __user* argp = (void __user*)arg;
1884 struct ifreq ifr;
1885 kuid_t owner;
1886 kgid_t group;
1887 int sndbuf;
1888 int vnet_hdr_sz;
1889 unsigned int ifindex;
1890 int ret;
1891
1892 if (cmd == TUNSETIFF || cmd == TUNSETQUEUE || _IOC_TYPE(cmd) == 0x89) {
1893 if (copy_from_user(&ifr, argp, ifreq_len))
1894 return -EFAULT;
1895 } else {
1896 memset(&ifr, 0, sizeof(ifr));
1897 }
1898 if (cmd == TUNGETFEATURES) {
1899 /* Currently this just means: "what IFF flags are valid?".
1900 * This is needed because we never checked for invalid flags on
1901 * TUNSETIFF. */
1902 return put_user(IFF_TUN | IFF_TAP | IFF_NO_PI | IFF_ONE_QUEUE |
1903 IFF_VNET_HDR | IFF_MULTI_QUEUE,
1904 (unsigned int __user*)argp);
1905 } else if (cmd == TUNSETQUEUE)
1906 return tun_set_queue(file, &ifr);
1907
1908 ret = 0;
1909 rtnl_lock();
1910
1911 tun = __tun_get(tfile);
1912 if (cmd == TUNSETIFF && !tun) {
1913 ifr.ifr_name[IFNAMSIZ-1] = '\0';
1914
1915 ret = tun_set_iff(tfile->net, file, &ifr);
1916
1917 if (ret)
1918 goto unlock;
1919
1920 if (copy_to_user(argp, &ifr, ifreq_len))
1921 ret = -EFAULT;
1922 goto unlock;
1923 }
1924 if (cmd == TUNSETIFINDEX) {
1925 ret = -EPERM;
1926 if (tun)
1927 goto unlock;
1928
1929 ret = -EFAULT;
1930 if (copy_from_user(&ifindex, argp, sizeof(ifindex)))
1931 goto unlock;
1932
1933 ret = 0;
1934 tfile->ifindex = ifindex;
1935 goto unlock;
1936 }
1937
1938 ret = -EBADFD;
1939 if (!tun)
1940 goto unlock;
1941
1942 tun_debug(KERN_INFO, tun, "tun_chr_ioctl cmd %u\n", cmd);
1943
1944 ret = 0;
1945 switch (cmd) {
1946 case TUNGETIFF:
1947 tun_get_iff(current->nsproxy->net_ns, tun, &ifr);
1948
1949 if (tfile->detached)
1950 ifr.ifr_flags |= IFF_DETACH_QUEUE;
1951 if (!tfile->socket.sk->sk_filter)
1952 ifr.ifr_flags |= IFF_NOFILTER;
1953
1954 if (copy_to_user(argp, &ifr, ifreq_len))
1955 ret = -EFAULT;
1956 break;
1957
1958 case TUNSETNOCSUM:
1959 /* Disable/Enable checksum */
1960
1961 /* [unimplemented] */
1962 tun_debug(KERN_INFO, tun, "ignored: set checksum %s\n",
1963 arg ? "disabled" : "enabled");
1964 break;
1965
1966 case TUNSETPERSIST:
1967 /* Disable/Enable persist mode. Keep an extra reference to the
1968 * module to prevent the module being unprobed.
1969 */
1970 if (arg && !(tun->flags & TUN_PERSIST)) {
1971 tun->flags |= TUN_PERSIST;
1972 __module_get(THIS_MODULE);
1973 }
1974 if (!arg && (tun->flags & TUN_PERSIST)) {
1975 tun->flags &= ~TUN_PERSIST;
1976 module_put(THIS_MODULE);
1977 }
1978
1979 tun_debug(KERN_INFO, tun, "persist %s\n",
1980 arg ? "enabled" : "disabled");
1981 break;
1982
1983 case TUNSETOWNER:
1984 /* Set owner of the device */
1985 owner = make_kuid(current_user_ns(), arg);
1986 if (!uid_valid(owner)) {
1987 ret = -EINVAL;
1988 break;
1989 }
1990 tun->owner = owner;
1991 tun_debug(KERN_INFO, tun, "owner set to %u\n",
1992 from_kuid(&init_user_ns, tun->owner));
1993 break;
1994
1995 case TUNSETGROUP:
1996 /* Set group of the device */
1997 group = make_kgid(current_user_ns(), arg);
1998 if (!gid_valid(group)) {
1999 ret = -EINVAL;
2000 break;
2001 }
2002 tun->group = group;
2003 tun_debug(KERN_INFO, tun, "group set to %u\n",
2004 from_kgid(&init_user_ns, tun->group));
2005 break;
2006
2007 case TUNSETLINK:
2008 /* Only allow setting the type when the interface is down */
2009 if (tun->dev->flags & IFF_UP) {
2010 tun_debug(KERN_INFO, tun,
2011 "Linktype set failed because interface is up\n");
2012 ret = -EBUSY;
2013 } else {
2014 tun->dev->type = (int) arg;
2015 tun_debug(KERN_INFO, tun, "linktype set to %d\n",
2016 tun->dev->type);
2017 ret = 0;
2018 }
2019 break;
2020
2021#ifdef TUN_DEBUG
2022 case TUNSETDEBUG:
2023 tun->debug = arg;
2024 break;
2025#endif
2026 case TUNSETOFFLOAD:
2027 ret = set_offload(tun, arg);
2028 break;
2029
2030 case TUNSETTXFILTER:
2031 /* Can be set only for TAPs */
2032 ret = -EINVAL;
2033 if ((tun->flags & TUN_TYPE_MASK) != TUN_TAP_DEV)
2034 break;
2035 ret = update_filter(&tun->txflt, (void __user *)arg);
2036 break;
2037
2038 case SIOCGIFHWADDR:
2039 /* Get hw address */
2040 memcpy(ifr.ifr_hwaddr.sa_data, tun->dev->dev_addr, ETH_ALEN);
2041 ifr.ifr_hwaddr.sa_family = tun->dev->type;
2042 if (copy_to_user(argp, &ifr, ifreq_len))
2043 ret = -EFAULT;
2044 break;
2045
2046 case SIOCSIFHWADDR:
2047 /* Set hw address */
2048 tun_debug(KERN_DEBUG, tun, "set hw address: %pM\n",
2049 ifr.ifr_hwaddr.sa_data);
2050
2051 ret = dev_set_mac_address(tun->dev, &ifr.ifr_hwaddr);
2052 break;
2053
2054 case TUNGETSNDBUF:
2055 sndbuf = tfile->socket.sk->sk_sndbuf;
2056 if (copy_to_user(argp, &sndbuf, sizeof(sndbuf)))
2057 ret = -EFAULT;
2058 break;
2059
2060 case TUNSETSNDBUF:
2061 if (copy_from_user(&sndbuf, argp, sizeof(sndbuf))) {
2062 ret = -EFAULT;
2063 break;
2064 }
2065
2066 tun->sndbuf = sndbuf;
2067 tun_set_sndbuf(tun);
2068 break;
2069
2070 case TUNGETVNETHDRSZ:
2071 vnet_hdr_sz = tun->vnet_hdr_sz;
2072 if (copy_to_user(argp, &vnet_hdr_sz, sizeof(vnet_hdr_sz)))
2073 ret = -EFAULT;
2074 break;
2075
2076 case TUNSETVNETHDRSZ:
2077 if (copy_from_user(&vnet_hdr_sz, argp, sizeof(vnet_hdr_sz))) {
2078 ret = -EFAULT;
2079 break;
2080 }
2081 if (vnet_hdr_sz < (int)sizeof(struct virtio_net_hdr)) {
2082 ret = -EINVAL;
2083 break;
2084 }
2085
2086 tun->vnet_hdr_sz = vnet_hdr_sz;
2087 break;
2088
2089 case TUNATTACHFILTER:
2090 /* Can be set only for TAPs */
2091 ret = -EINVAL;
2092 if ((tun->flags & TUN_TYPE_MASK) != TUN_TAP_DEV)
2093 break;
2094 ret = -EFAULT;
2095 if (copy_from_user(&tun->fprog, argp, sizeof(tun->fprog)))
2096 break;
2097
2098 ret = tun_attach_filter(tun);
2099 break;
2100
2101 case TUNDETACHFILTER:
2102 /* Can be set only for TAPs */
2103 ret = -EINVAL;
2104 if ((tun->flags & TUN_TYPE_MASK) != TUN_TAP_DEV)
2105 break;
2106 ret = 0;
2107 tun_detach_filter(tun, tun->numqueues);
2108 break;
2109
2110 case TUNGETFILTER:
2111 ret = -EINVAL;
2112 if ((tun->flags & TUN_TYPE_MASK) != TUN_TAP_DEV)
2113 break;
2114 ret = -EFAULT;
2115 if (copy_to_user(argp, &tun->fprog, sizeof(tun->fprog)))
2116 break;
2117 ret = 0;
2118 break;
2119
2120 default:
2121 ret = -EINVAL;
2122 break;
2123 }
2124
2125unlock:
2126 rtnl_unlock();
2127 if (tun)
2128 tun_put(tun);
2129 return ret;
2130}
2131
2132static long tun_chr_ioctl(struct file *file,
2133 unsigned int cmd, unsigned long arg)
2134{
2135 return __tun_chr_ioctl(file, cmd, arg, sizeof (struct ifreq));
2136}
2137
2138#ifdef CONFIG_COMPAT
2139static long tun_chr_compat_ioctl(struct file *file,
2140 unsigned int cmd, unsigned long arg)
2141{
2142 switch (cmd) {
2143 case TUNSETIFF:
2144 case TUNGETIFF:
2145 case TUNSETTXFILTER:
2146 case TUNGETSNDBUF:
2147 case TUNSETSNDBUF:
2148 case SIOCGIFHWADDR:
2149 case SIOCSIFHWADDR:
2150 arg = (unsigned long)compat_ptr(arg);
2151 break;
2152 default:
2153 arg = (compat_ulong_t)arg;
2154 break;
2155 }
2156
2157 /*
2158 * compat_ifreq is shorter than ifreq, so we must not access beyond
2159 * the end of that structure. All fields that are used in this
2160 * driver are compatible though, we don't need to convert the
2161 * contents.
2162 */
2163 return __tun_chr_ioctl(file, cmd, arg, sizeof(struct compat_ifreq));
2164}
2165#endif /* CONFIG_COMPAT */
2166
2167static int tun_chr_fasync(int fd, struct file *file, int on)
2168{
2169 struct tun_file *tfile = file->private_data;
2170 int ret;
2171
2172 if ((ret = fasync_helper(fd, file, on, &tfile->fasync)) < 0)
2173 goto out;
2174
2175 if (on) {
2176 ret = __f_setown(file, task_pid(current), PIDTYPE_PID, 0);
2177 if (ret)
2178 goto out;
2179 tfile->flags |= TUN_FASYNC;
2180 } else
2181 tfile->flags &= ~TUN_FASYNC;
2182 ret = 0;
2183out:
2184 return ret;
2185}
2186
2187static int tun_chr_open(struct inode *inode, struct file * file)
2188{
2189 struct tun_file *tfile;
2190
2191 DBG1(KERN_INFO, "tunX: tun_chr_open\n");
2192
2193 tfile = (struct tun_file *)sk_alloc(&init_net, AF_UNSPEC, GFP_KERNEL,
2194 &tun_proto);
2195 if (!tfile)
2196 return -ENOMEM;
2197 RCU_INIT_POINTER(tfile->tun, NULL);
2198 tfile->net = get_net(current->nsproxy->net_ns);
2199 tfile->flags = 0;
2200 tfile->ifindex = 0;
2201
2202 rcu_assign_pointer(tfile->socket.wq, &tfile->wq);
2203 init_waitqueue_head(&tfile->wq.wait);
2204
2205 tfile->socket.file = file;
2206 tfile->socket.ops = &tun_socket_ops;
2207
2208 sock_init_data(&tfile->socket, &tfile->sk);
2209 sk_change_net(&tfile->sk, tfile->net);
2210
2211 tfile->sk.sk_write_space = tun_sock_write_space;
2212 tfile->sk.sk_sndbuf = INT_MAX;
2213
2214 file->private_data = tfile;
2215 set_bit(SOCK_EXTERNALLY_ALLOCATED, &tfile->socket.flags);
2216 INIT_LIST_HEAD(&tfile->next);
2217
2218 sock_set_flag(&tfile->sk, SOCK_ZEROCOPY);
2219
2220 return 0;
2221}
2222
2223static int tun_chr_close(struct inode *inode, struct file *file)
2224{
2225 struct tun_file *tfile = file->private_data;
2226 struct net *net = tfile->net;
2227
2228 tun_detach(tfile, true);
2229 put_net(net);
2230
2231 return 0;
2232}
2233
2234#ifdef CONFIG_PROC_FS
2235static int tun_chr_show_fdinfo(struct seq_file *m, struct file *f)
2236{
2237 struct tun_struct *tun;
2238 struct ifreq ifr;
2239
2240 memset(&ifr, 0, sizeof(ifr));
2241
2242 rtnl_lock();
2243 tun = tun_get(f);
2244 if (tun)
2245 tun_get_iff(current->nsproxy->net_ns, tun, &ifr);
2246 rtnl_unlock();
2247
2248 if (tun)
2249 tun_put(tun);
2250
2251 return seq_printf(m, "iff:\t%s\n", ifr.ifr_name);
2252}
2253#endif
2254
2255static const struct file_operations tun_fops = {
2256 .owner = THIS_MODULE,
2257 .llseek = no_llseek,
2258 .read = do_sync_read,
2259 .aio_read = tun_chr_aio_read,
2260 .write = do_sync_write,
2261 .aio_write = tun_chr_aio_write,
2262 .poll = tun_chr_poll,
2263 .unlocked_ioctl = tun_chr_ioctl,
2264#ifdef CONFIG_COMPAT
2265 .compat_ioctl = tun_chr_compat_ioctl,
2266#endif
2267 .open = tun_chr_open,
2268 .release = tun_chr_close,
2269 .fasync = tun_chr_fasync,
2270#ifdef CONFIG_PROC_FS
2271 .show_fdinfo = tun_chr_show_fdinfo,
2272#endif
2273};
2274
2275static struct miscdevice tun_miscdev = {
2276 .minor = TUN_MINOR,
2277 .name = "tun",
2278 .nodename = "net/tun",
2279 .fops = &tun_fops,
2280};
2281
2282/* ethtool interface */
2283
2284static int tun_get_settings(struct net_device *dev, struct ethtool_cmd *cmd)
2285{
2286 cmd->supported = 0;
2287 cmd->advertising = 0;
2288 ethtool_cmd_speed_set(cmd, SPEED_10);
2289 cmd->duplex = DUPLEX_FULL;
2290 cmd->port = PORT_TP;
2291 cmd->phy_address = 0;
2292 cmd->transceiver = XCVR_INTERNAL;
2293 cmd->autoneg = AUTONEG_DISABLE;
2294 cmd->maxtxpkt = 0;
2295 cmd->maxrxpkt = 0;
2296 return 0;
2297}
2298
2299static void tun_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
2300{
2301 struct tun_struct *tun = netdev_priv(dev);
2302
2303 strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
2304 strlcpy(info->version, DRV_VERSION, sizeof(info->version));
2305
2306 switch (tun->flags & TUN_TYPE_MASK) {
2307 case TUN_TUN_DEV:
2308 strlcpy(info->bus_info, "tun", sizeof(info->bus_info));
2309 break;
2310 case TUN_TAP_DEV:
2311 strlcpy(info->bus_info, "tap", sizeof(info->bus_info));
2312 break;
2313 }
2314}
2315
2316static u32 tun_get_msglevel(struct net_device *dev)
2317{
2318#ifdef TUN_DEBUG
2319 struct tun_struct *tun = netdev_priv(dev);
2320 return tun->debug;
2321#else
2322 return -EOPNOTSUPP;
2323#endif
2324}
2325
2326static void tun_set_msglevel(struct net_device *dev, u32 value)
2327{
2328#ifdef TUN_DEBUG
2329 struct tun_struct *tun = netdev_priv(dev);
2330 tun->debug = value;
2331#endif
2332}
2333
2334static const struct ethtool_ops tun_ethtool_ops = {
2335 .get_settings = tun_get_settings,
2336 .get_drvinfo = tun_get_drvinfo,
2337 .get_msglevel = tun_get_msglevel,
2338 .set_msglevel = tun_set_msglevel,
2339 .get_link = ethtool_op_get_link,
2340 .get_ts_info = ethtool_op_get_ts_info,
2341};
2342
2343
2344static int __init tun_init(void)
2345{
2346 int ret = 0;
2347
2348 pr_info("%s, %s\n", DRV_DESCRIPTION, DRV_VERSION);
2349 pr_info("%s\n", DRV_COPYRIGHT);
2350
2351 ret = rtnl_link_register(&tun_link_ops);
2352 if (ret) {
2353 pr_err("Can't register link_ops\n");
2354 goto err_linkops;
2355 }
2356
2357 ret = misc_register(&tun_miscdev);
2358 if (ret) {
2359 pr_err("Can't register misc device %d\n", TUN_MINOR);
2360 goto err_misc;
2361 }
2362 return 0;
2363err_misc:
2364 rtnl_link_unregister(&tun_link_ops);
2365err_linkops:
2366 return ret;
2367}
2368
2369static void tun_cleanup(void)
2370{
2371 misc_deregister(&tun_miscdev);
2372 rtnl_link_unregister(&tun_link_ops);
2373}
2374
2375/* Get an underlying socket object from tun file. Returns error unless file is
2376 * attached to a device. The returned object works like a packet socket, it
2377 * can be used for sock_sendmsg/sock_recvmsg. The caller is responsible for
2378 * holding a reference to the file for as long as the socket is in use. */
2379struct socket *tun_get_socket(struct file *file)
2380{
2381 struct tun_file *tfile;
2382 if (file->f_op != &tun_fops)
2383 return ERR_PTR(-EINVAL);
2384 tfile = file->private_data;
2385 if (!tfile)
2386 return ERR_PTR(-EBADFD);
2387 return &tfile->socket;
2388}
2389EXPORT_SYMBOL_GPL(tun_get_socket);
2390
2391module_init(tun_init);
2392module_exit(tun_cleanup);
2393MODULE_DESCRIPTION(DRV_DESCRIPTION);
2394MODULE_AUTHOR(DRV_COPYRIGHT);
2395MODULE_LICENSE("GPL");
2396MODULE_ALIAS_MISCDEV(TUN_MINOR);
2397MODULE_ALIAS("devname:net/tun");