]>
Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * TUN - Universal TUN/TAP device driver. | |
3 | * Copyright (C) 1999-2002 Maxim Krasnyansky <[email protected]> | |
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 | * | |
ff4cc3ac MK |
21 | * Mike Kershaw <[email protected]> 2005/08/14 |
22 | * Add TUNSETLINK ioctl to set the link encapsulation | |
23 | * | |
1da177e4 | 24 | * Mark Smith <[email protected]> |
f271b2cc | 25 | * Use random_ether_addr() for tap MAC address. |
1da177e4 LT |
26 | * |
27 | * Harald Roelle <[email protected]> 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 <[email protected]> | |
34 | * Modifications for 2.3.99-pre5 kernel. | |
35 | */ | |
36 | ||
6b8a66ee JP |
37 | #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt |
38 | ||
1da177e4 LT |
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 <[email protected]>" | |
43 | ||
1da177e4 LT |
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> | |
50857e2a | 58 | #include <linux/compat.h> |
1da177e4 LT |
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/crc32.h> | |
d647a591 | 64 | #include <linux/nsproxy.h> |
f43798c2 | 65 | #include <linux/virtio_net.h> |
99405162 | 66 | #include <linux/rcupdate.h> |
881d966b | 67 | #include <net/net_namespace.h> |
79d17604 | 68 | #include <net/netns/generic.h> |
f019a7a5 | 69 | #include <net/rtnetlink.h> |
33dccbb0 | 70 | #include <net/sock.h> |
1da177e4 LT |
71 | |
72 | #include <asm/system.h> | |
73 | #include <asm/uaccess.h> | |
74 | ||
14daa021 RR |
75 | /* Uncomment to enable debugging */ |
76 | /* #define TUN_DEBUG 1 */ | |
77 | ||
1da177e4 LT |
78 | #ifdef TUN_DEBUG |
79 | static int debug; | |
14daa021 | 80 | |
6b8a66ee JP |
81 | #define tun_debug(level, tun, fmt, args...) \ |
82 | do { \ | |
83 | if (tun->debug) \ | |
84 | netdev_printk(level, tun->dev, fmt, ##args); \ | |
85 | } while (0) | |
86 | #define DBG1(level, fmt, args...) \ | |
87 | do { \ | |
88 | if (debug == 2) \ | |
89 | printk(level fmt, ##args); \ | |
90 | } while (0) | |
14daa021 | 91 | #else |
6b8a66ee JP |
92 | #define tun_debug(level, tun, fmt, args...) \ |
93 | do { \ | |
94 | if (0) \ | |
95 | netdev_printk(level, tun->dev, fmt, ##args); \ | |
96 | } while (0) | |
97 | #define DBG1(level, fmt, args...) \ | |
98 | do { \ | |
99 | if (0) \ | |
100 | printk(level fmt, ##args); \ | |
101 | } while (0) | |
14daa021 RR |
102 | #endif |
103 | ||
f271b2cc MK |
104 | #define FLT_EXACT_COUNT 8 |
105 | struct tap_filter { | |
106 | unsigned int count; /* Number of addrs. Zero means disabled */ | |
107 | u32 mask[2]; /* Mask of the hashed addrs */ | |
108 | unsigned char addr[FLT_EXACT_COUNT][ETH_ALEN]; | |
109 | }; | |
110 | ||
631ab46b | 111 | struct tun_file { |
c70f1829 | 112 | atomic_t count; |
631ab46b | 113 | struct tun_struct *tun; |
36b50bab | 114 | struct net *net; |
631ab46b EB |
115 | }; |
116 | ||
33dccbb0 HX |
117 | struct tun_sock; |
118 | ||
14daa021 | 119 | struct tun_struct { |
631ab46b | 120 | struct tun_file *tfile; |
f271b2cc | 121 | unsigned int flags; |
14daa021 RR |
122 | uid_t owner; |
123 | gid_t group; | |
124 | ||
14daa021 | 125 | struct net_device *dev; |
c8f44aff | 126 | netdev_features_t set_features; |
88255375 MM |
127 | #define TUN_USER_FEATURES (NETIF_F_HW_CSUM|NETIF_F_TSO_ECN|NETIF_F_TSO| \ |
128 | NETIF_F_TSO6|NETIF_F_UFO) | |
f271b2cc | 129 | struct fasync_struct *fasync; |
14daa021 | 130 | |
f271b2cc | 131 | struct tap_filter txflt; |
33dccbb0 | 132 | struct socket socket; |
43815482 | 133 | struct socket_wq wq; |
d9d52b51 MT |
134 | |
135 | int vnet_hdr_sz; | |
136 | ||
14daa021 RR |
137 | #ifdef TUN_DEBUG |
138 | int debug; | |
1da177e4 | 139 | #endif |
14daa021 | 140 | }; |
1da177e4 | 141 | |
33dccbb0 HX |
142 | struct tun_sock { |
143 | struct sock sk; | |
144 | struct tun_struct *tun; | |
145 | }; | |
146 | ||
147 | static inline struct tun_sock *tun_sk(struct sock *sk) | |
148 | { | |
149 | return container_of(sk, struct tun_sock, sk); | |
150 | } | |
151 | ||
a7385ba2 EB |
152 | static int tun_attach(struct tun_struct *tun, struct file *file) |
153 | { | |
631ab46b | 154 | struct tun_file *tfile = file->private_data; |
38231b7a | 155 | int err; |
a7385ba2 EB |
156 | |
157 | ASSERT_RTNL(); | |
158 | ||
38231b7a EB |
159 | netif_tx_lock_bh(tun->dev); |
160 | ||
161 | err = -EINVAL; | |
162 | if (tfile->tun) | |
163 | goto out; | |
164 | ||
165 | err = -EBUSY; | |
166 | if (tun->tfile) | |
167 | goto out; | |
168 | ||
169 | err = 0; | |
631ab46b EB |
170 | tfile->tun = tun; |
171 | tun->tfile = tfile; | |
05c2828c | 172 | tun->socket.file = file; |
bee31369 | 173 | netif_carrier_on(tun->dev); |
c70f1829 | 174 | dev_hold(tun->dev); |
89f56d1e | 175 | sock_hold(tun->socket.sk); |
c70f1829 | 176 | atomic_inc(&tfile->count); |
a7385ba2 | 177 | |
38231b7a EB |
178 | out: |
179 | netif_tx_unlock_bh(tun->dev); | |
180 | return err; | |
a7385ba2 EB |
181 | } |
182 | ||
631ab46b EB |
183 | static void __tun_detach(struct tun_struct *tun) |
184 | { | |
631ab46b | 185 | /* Detach from net device */ |
38231b7a | 186 | netif_tx_lock_bh(tun->dev); |
bee31369 | 187 | netif_carrier_off(tun->dev); |
631ab46b | 188 | tun->tfile = NULL; |
05c2828c | 189 | tun->socket.file = NULL; |
38231b7a | 190 | netif_tx_unlock_bh(tun->dev); |
631ab46b EB |
191 | |
192 | /* Drop read queue */ | |
89f56d1e | 193 | skb_queue_purge(&tun->socket.sk->sk_receive_queue); |
c70f1829 EB |
194 | |
195 | /* Drop the extra count on the net device */ | |
196 | dev_put(tun->dev); | |
197 | } | |
198 | ||
199 | static void tun_detach(struct tun_struct *tun) | |
200 | { | |
201 | rtnl_lock(); | |
202 | __tun_detach(tun); | |
203 | rtnl_unlock(); | |
631ab46b EB |
204 | } |
205 | ||
206 | static struct tun_struct *__tun_get(struct tun_file *tfile) | |
207 | { | |
c70f1829 EB |
208 | struct tun_struct *tun = NULL; |
209 | ||
210 | if (atomic_inc_not_zero(&tfile->count)) | |
211 | tun = tfile->tun; | |
212 | ||
213 | return tun; | |
631ab46b EB |
214 | } |
215 | ||
216 | static struct tun_struct *tun_get(struct file *file) | |
217 | { | |
218 | return __tun_get(file->private_data); | |
219 | } | |
220 | ||
221 | static void tun_put(struct tun_struct *tun) | |
222 | { | |
c70f1829 EB |
223 | struct tun_file *tfile = tun->tfile; |
224 | ||
225 | if (atomic_dec_and_test(&tfile->count)) | |
226 | tun_detach(tfile->tun); | |
631ab46b EB |
227 | } |
228 | ||
6b8a66ee | 229 | /* TAP filtering */ |
f271b2cc MK |
230 | static void addr_hash_set(u32 *mask, const u8 *addr) |
231 | { | |
232 | int n = ether_crc(ETH_ALEN, addr) >> 26; | |
233 | mask[n >> 5] |= (1 << (n & 31)); | |
234 | } | |
235 | ||
236 | static unsigned int addr_hash_test(const u32 *mask, const u8 *addr) | |
237 | { | |
238 | int n = ether_crc(ETH_ALEN, addr) >> 26; | |
239 | return mask[n >> 5] & (1 << (n & 31)); | |
240 | } | |
241 | ||
242 | static int update_filter(struct tap_filter *filter, void __user *arg) | |
243 | { | |
244 | struct { u8 u[ETH_ALEN]; } *addr; | |
245 | struct tun_filter uf; | |
246 | int err, alen, n, nexact; | |
247 | ||
248 | if (copy_from_user(&uf, arg, sizeof(uf))) | |
249 | return -EFAULT; | |
250 | ||
251 | if (!uf.count) { | |
252 | /* Disabled */ | |
253 | filter->count = 0; | |
254 | return 0; | |
255 | } | |
256 | ||
257 | alen = ETH_ALEN * uf.count; | |
258 | addr = kmalloc(alen, GFP_KERNEL); | |
259 | if (!addr) | |
260 | return -ENOMEM; | |
261 | ||
262 | if (copy_from_user(addr, arg + sizeof(uf), alen)) { | |
263 | err = -EFAULT; | |
264 | goto done; | |
265 | } | |
266 | ||
267 | /* The filter is updated without holding any locks. Which is | |
268 | * perfectly safe. We disable it first and in the worst | |
269 | * case we'll accept a few undesired packets. */ | |
270 | filter->count = 0; | |
271 | wmb(); | |
272 | ||
273 | /* Use first set of addresses as an exact filter */ | |
274 | for (n = 0; n < uf.count && n < FLT_EXACT_COUNT; n++) | |
275 | memcpy(filter->addr[n], addr[n].u, ETH_ALEN); | |
276 | ||
277 | nexact = n; | |
278 | ||
cfbf84fc AW |
279 | /* Remaining multicast addresses are hashed, |
280 | * unicast will leave the filter disabled. */ | |
f271b2cc | 281 | memset(filter->mask, 0, sizeof(filter->mask)); |
cfbf84fc AW |
282 | for (; n < uf.count; n++) { |
283 | if (!is_multicast_ether_addr(addr[n].u)) { | |
284 | err = 0; /* no filter */ | |
285 | goto done; | |
286 | } | |
f271b2cc | 287 | addr_hash_set(filter->mask, addr[n].u); |
cfbf84fc | 288 | } |
f271b2cc MK |
289 | |
290 | /* For ALLMULTI just set the mask to all ones. | |
291 | * This overrides the mask populated above. */ | |
292 | if ((uf.flags & TUN_FLT_ALLMULTI)) | |
293 | memset(filter->mask, ~0, sizeof(filter->mask)); | |
294 | ||
295 | /* Now enable the filter */ | |
296 | wmb(); | |
297 | filter->count = nexact; | |
298 | ||
299 | /* Return the number of exact filters */ | |
300 | err = nexact; | |
301 | ||
302 | done: | |
303 | kfree(addr); | |
304 | return err; | |
305 | } | |
306 | ||
307 | /* Returns: 0 - drop, !=0 - accept */ | |
308 | static int run_filter(struct tap_filter *filter, const struct sk_buff *skb) | |
309 | { | |
310 | /* Cannot use eth_hdr(skb) here because skb_mac_hdr() is incorrect | |
311 | * at this point. */ | |
312 | struct ethhdr *eh = (struct ethhdr *) skb->data; | |
313 | int i; | |
314 | ||
315 | /* Exact match */ | |
316 | for (i = 0; i < filter->count; i++) | |
317 | if (!compare_ether_addr(eh->h_dest, filter->addr[i])) | |
318 | return 1; | |
319 | ||
320 | /* Inexact match (multicast only) */ | |
321 | if (is_multicast_ether_addr(eh->h_dest)) | |
322 | return addr_hash_test(filter->mask, eh->h_dest); | |
323 | ||
324 | return 0; | |
325 | } | |
326 | ||
327 | /* | |
328 | * Checks whether the packet is accepted or not. | |
329 | * Returns: 0 - drop, !=0 - accept | |
330 | */ | |
331 | static int check_filter(struct tap_filter *filter, const struct sk_buff *skb) | |
332 | { | |
333 | if (!filter->count) | |
334 | return 1; | |
335 | ||
336 | return run_filter(filter, skb); | |
337 | } | |
338 | ||
1da177e4 LT |
339 | /* Network device part of the driver */ |
340 | ||
7282d491 | 341 | static const struct ethtool_ops tun_ethtool_ops; |
1da177e4 | 342 | |
c70f1829 EB |
343 | /* Net device detach from fd. */ |
344 | static void tun_net_uninit(struct net_device *dev) | |
345 | { | |
346 | struct tun_struct *tun = netdev_priv(dev); | |
347 | struct tun_file *tfile = tun->tfile; | |
348 | ||
349 | /* Inform the methods they need to stop using the dev. | |
350 | */ | |
351 | if (tfile) { | |
43815482 | 352 | wake_up_all(&tun->wq.wait); |
c70f1829 EB |
353 | if (atomic_dec_and_test(&tfile->count)) |
354 | __tun_detach(tun); | |
355 | } | |
356 | } | |
357 | ||
9c3fea6a HX |
358 | static void tun_free_netdev(struct net_device *dev) |
359 | { | |
360 | struct tun_struct *tun = netdev_priv(dev); | |
361 | ||
1ab5ecb9 | 362 | sk_release_kernel(tun->socket.sk); |
9c3fea6a HX |
363 | } |
364 | ||
1da177e4 LT |
365 | /* Net device open. */ |
366 | static int tun_net_open(struct net_device *dev) | |
367 | { | |
368 | netif_start_queue(dev); | |
369 | return 0; | |
370 | } | |
371 | ||
372 | /* Net device close. */ | |
373 | static int tun_net_close(struct net_device *dev) | |
374 | { | |
375 | netif_stop_queue(dev); | |
376 | return 0; | |
377 | } | |
378 | ||
379 | /* Net device start xmit */ | |
424efe9c | 380 | static netdev_tx_t tun_net_xmit(struct sk_buff *skb, struct net_device *dev) |
1da177e4 LT |
381 | { |
382 | struct tun_struct *tun = netdev_priv(dev); | |
383 | ||
6b8a66ee | 384 | tun_debug(KERN_INFO, tun, "tun_net_xmit %d\n", skb->len); |
1da177e4 LT |
385 | |
386 | /* Drop packet if interface is not attached */ | |
631ab46b | 387 | if (!tun->tfile) |
1da177e4 LT |
388 | goto drop; |
389 | ||
f271b2cc MK |
390 | /* Drop if the filter does not like it. |
391 | * This is a noop if the filter is disabled. | |
392 | * Filter can be enabled only for the TAP devices. */ | |
393 | if (!check_filter(&tun->txflt, skb)) | |
394 | goto drop; | |
395 | ||
99405162 MT |
396 | if (tun->socket.sk->sk_filter && |
397 | sk_filter(tun->socket.sk, skb)) | |
398 | goto drop; | |
399 | ||
89f56d1e | 400 | if (skb_queue_len(&tun->socket.sk->sk_receive_queue) >= dev->tx_queue_len) { |
1da177e4 LT |
401 | if (!(tun->flags & TUN_ONE_QUEUE)) { |
402 | /* Normal queueing mode. */ | |
403 | /* Packet scheduler handles dropping of further packets. */ | |
404 | netif_stop_queue(dev); | |
405 | ||
406 | /* We won't see all dropped packets individually, so overrun | |
407 | * error is more appropriate. */ | |
09f75cd7 | 408 | dev->stats.tx_fifo_errors++; |
1da177e4 LT |
409 | } else { |
410 | /* Single queue mode. | |
411 | * Driver handles dropping of all packets itself. */ | |
412 | goto drop; | |
413 | } | |
414 | } | |
415 | ||
0110d6f2 MT |
416 | /* Orphan the skb - required as we might hang on to it |
417 | * for indefinite time. */ | |
418 | skb_orphan(skb); | |
419 | ||
f271b2cc | 420 | /* Enqueue packet */ |
89f56d1e | 421 | skb_queue_tail(&tun->socket.sk->sk_receive_queue, skb); |
1da177e4 LT |
422 | |
423 | /* Notify and wake up reader process */ | |
424 | if (tun->flags & TUN_FASYNC) | |
425 | kill_fasync(&tun->fasync, SIGIO, POLL_IN); | |
43815482 | 426 | wake_up_interruptible_poll(&tun->wq.wait, POLLIN | |
05c2828c | 427 | POLLRDNORM | POLLRDBAND); |
6ed10654 | 428 | return NETDEV_TX_OK; |
1da177e4 LT |
429 | |
430 | drop: | |
09f75cd7 | 431 | dev->stats.tx_dropped++; |
1da177e4 | 432 | kfree_skb(skb); |
6ed10654 | 433 | return NETDEV_TX_OK; |
1da177e4 LT |
434 | } |
435 | ||
f271b2cc | 436 | static void tun_net_mclist(struct net_device *dev) |
1da177e4 | 437 | { |
f271b2cc MK |
438 | /* |
439 | * This callback is supposed to deal with mc filter in | |
440 | * _rx_ path and has nothing to do with the _tx_ path. | |
441 | * In rx path we always accept everything userspace gives us. | |
442 | */ | |
1da177e4 LT |
443 | } |
444 | ||
4885a504 ES |
445 | #define MIN_MTU 68 |
446 | #define MAX_MTU 65535 | |
447 | ||
448 | static int | |
449 | tun_net_change_mtu(struct net_device *dev, int new_mtu) | |
450 | { | |
451 | if (new_mtu < MIN_MTU || new_mtu + dev->hard_header_len > MAX_MTU) | |
452 | return -EINVAL; | |
453 | dev->mtu = new_mtu; | |
454 | return 0; | |
455 | } | |
456 | ||
c8f44aff MM |
457 | static netdev_features_t tun_net_fix_features(struct net_device *dev, |
458 | netdev_features_t features) | |
88255375 MM |
459 | { |
460 | struct tun_struct *tun = netdev_priv(dev); | |
461 | ||
462 | return (features & tun->set_features) | (features & ~TUN_USER_FEATURES); | |
463 | } | |
bebd097a NH |
464 | #ifdef CONFIG_NET_POLL_CONTROLLER |
465 | static void tun_poll_controller(struct net_device *dev) | |
466 | { | |
467 | /* | |
468 | * Tun only receives frames when: | |
469 | * 1) the char device endpoint gets data from user space | |
470 | * 2) the tun socket gets a sendmsg call from user space | |
471 | * Since both of those are syncronous operations, we are guaranteed | |
472 | * never to have pending data when we poll for it | |
473 | * so theres nothing to do here but return. | |
474 | * We need this though so netpoll recognizes us as an interface that | |
475 | * supports polling, which enables bridge devices in virt setups to | |
476 | * still use netconsole | |
477 | */ | |
478 | return; | |
479 | } | |
480 | #endif | |
758e43b7 | 481 | static const struct net_device_ops tun_netdev_ops = { |
c70f1829 | 482 | .ndo_uninit = tun_net_uninit, |
758e43b7 SH |
483 | .ndo_open = tun_net_open, |
484 | .ndo_stop = tun_net_close, | |
00829823 | 485 | .ndo_start_xmit = tun_net_xmit, |
758e43b7 | 486 | .ndo_change_mtu = tun_net_change_mtu, |
88255375 | 487 | .ndo_fix_features = tun_net_fix_features, |
bebd097a NH |
488 | #ifdef CONFIG_NET_POLL_CONTROLLER |
489 | .ndo_poll_controller = tun_poll_controller, | |
490 | #endif | |
758e43b7 SH |
491 | }; |
492 | ||
493 | static const struct net_device_ops tap_netdev_ops = { | |
c70f1829 | 494 | .ndo_uninit = tun_net_uninit, |
758e43b7 SH |
495 | .ndo_open = tun_net_open, |
496 | .ndo_stop = tun_net_close, | |
00829823 | 497 | .ndo_start_xmit = tun_net_xmit, |
758e43b7 | 498 | .ndo_change_mtu = tun_net_change_mtu, |
88255375 | 499 | .ndo_fix_features = tun_net_fix_features, |
afc4b13d | 500 | .ndo_set_rx_mode = tun_net_mclist, |
758e43b7 SH |
501 | .ndo_set_mac_address = eth_mac_addr, |
502 | .ndo_validate_addr = eth_validate_addr, | |
bebd097a NH |
503 | #ifdef CONFIG_NET_POLL_CONTROLLER |
504 | .ndo_poll_controller = tun_poll_controller, | |
505 | #endif | |
758e43b7 SH |
506 | }; |
507 | ||
1da177e4 LT |
508 | /* Initialize net device. */ |
509 | static void tun_net_init(struct net_device *dev) | |
510 | { | |
511 | struct tun_struct *tun = netdev_priv(dev); | |
6aa20a22 | 512 | |
1da177e4 LT |
513 | switch (tun->flags & TUN_TYPE_MASK) { |
514 | case TUN_TUN_DEV: | |
758e43b7 SH |
515 | dev->netdev_ops = &tun_netdev_ops; |
516 | ||
1da177e4 LT |
517 | /* Point-to-Point TUN Device */ |
518 | dev->hard_header_len = 0; | |
519 | dev->addr_len = 0; | |
520 | dev->mtu = 1500; | |
521 | ||
522 | /* Zero header length */ | |
6aa20a22 | 523 | dev->type = ARPHRD_NONE; |
1da177e4 LT |
524 | dev->flags = IFF_POINTOPOINT | IFF_NOARP | IFF_MULTICAST; |
525 | dev->tx_queue_len = TUN_READQ_SIZE; /* We prefer our own queue length */ | |
526 | break; | |
527 | ||
528 | case TUN_TAP_DEV: | |
7a0a9608 | 529 | dev->netdev_ops = &tap_netdev_ops; |
1da177e4 | 530 | /* Ethernet TAP Device */ |
1da177e4 | 531 | ether_setup(dev); |
550fd08c | 532 | dev->priv_flags &= ~IFF_TX_SKB_SHARING; |
36226a8d | 533 | |
f2cedb63 | 534 | eth_hw_addr_random(dev); |
36226a8d | 535 | |
1da177e4 LT |
536 | dev->tx_queue_len = TUN_READQ_SIZE; /* We prefer our own queue length */ |
537 | break; | |
538 | } | |
539 | } | |
540 | ||
541 | /* Character device part */ | |
542 | ||
543 | /* Poll */ | |
544 | static unsigned int tun_chr_poll(struct file *file, poll_table * wait) | |
6aa20a22 | 545 | { |
b2430de3 EB |
546 | struct tun_file *tfile = file->private_data; |
547 | struct tun_struct *tun = __tun_get(tfile); | |
3c8a9c63 | 548 | struct sock *sk; |
33dccbb0 | 549 | unsigned int mask = 0; |
1da177e4 LT |
550 | |
551 | if (!tun) | |
eac9e902 | 552 | return POLLERR; |
1da177e4 | 553 | |
89f56d1e | 554 | sk = tun->socket.sk; |
3c8a9c63 | 555 | |
6b8a66ee | 556 | tun_debug(KERN_INFO, tun, "tun_chr_poll\n"); |
1da177e4 | 557 | |
43815482 | 558 | poll_wait(file, &tun->wq.wait, wait); |
6aa20a22 | 559 | |
89f56d1e | 560 | if (!skb_queue_empty(&sk->sk_receive_queue)) |
1da177e4 LT |
561 | mask |= POLLIN | POLLRDNORM; |
562 | ||
33dccbb0 HX |
563 | if (sock_writeable(sk) || |
564 | (!test_and_set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags) && | |
565 | sock_writeable(sk))) | |
566 | mask |= POLLOUT | POLLWRNORM; | |
567 | ||
c70f1829 EB |
568 | if (tun->dev->reg_state != NETREG_REGISTERED) |
569 | mask = POLLERR; | |
570 | ||
631ab46b | 571 | tun_put(tun); |
1da177e4 LT |
572 | return mask; |
573 | } | |
574 | ||
f42157cb RR |
575 | /* prepad is the amount to reserve at front. len is length after that. |
576 | * linear is a hint as to how much to copy (usually headers). */ | |
6f7c156c | 577 | static struct sk_buff *tun_alloc_skb(struct tun_struct *tun, |
578 | size_t prepad, size_t len, | |
579 | size_t linear, int noblock) | |
f42157cb | 580 | { |
89f56d1e | 581 | struct sock *sk = tun->socket.sk; |
f42157cb | 582 | struct sk_buff *skb; |
33dccbb0 | 583 | int err; |
f42157cb | 584 | |
82862742 HX |
585 | sock_update_classid(sk); |
586 | ||
f42157cb | 587 | /* Under a page? Don't bother with paged skb. */ |
0eca93bc | 588 | if (prepad + len < PAGE_SIZE || !linear) |
33dccbb0 | 589 | linear = len; |
f42157cb | 590 | |
33dccbb0 HX |
591 | skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock, |
592 | &err); | |
f42157cb | 593 | if (!skb) |
33dccbb0 | 594 | return ERR_PTR(err); |
f42157cb RR |
595 | |
596 | skb_reserve(skb, prepad); | |
597 | skb_put(skb, linear); | |
33dccbb0 HX |
598 | skb->data_len = len - linear; |
599 | skb->len += len - linear; | |
f42157cb RR |
600 | |
601 | return skb; | |
602 | } | |
603 | ||
1da177e4 | 604 | /* Get packet from user space buffer */ |
6f7c156c | 605 | static ssize_t tun_get_user(struct tun_struct *tun, |
606 | const struct iovec *iv, size_t count, | |
607 | int noblock) | |
1da177e4 | 608 | { |
09640e63 | 609 | struct tun_pi pi = { 0, cpu_to_be16(ETH_P_IP) }; |
1da177e4 | 610 | struct sk_buff *skb; |
a504b86e | 611 | size_t len = count, align = NET_SKB_PAD; |
f43798c2 | 612 | struct virtio_net_hdr gso = { 0 }; |
6f26c9a7 | 613 | int offset = 0; |
1da177e4 LT |
614 | |
615 | if (!(tun->flags & TUN_NO_PI)) { | |
616 | if ((len -= sizeof(pi)) > count) | |
617 | return -EINVAL; | |
618 | ||
6f26c9a7 | 619 | if (memcpy_fromiovecend((void *)&pi, iv, 0, sizeof(pi))) |
1da177e4 | 620 | return -EFAULT; |
6f26c9a7 | 621 | offset += sizeof(pi); |
1da177e4 LT |
622 | } |
623 | ||
f43798c2 | 624 | if (tun->flags & TUN_VNET_HDR) { |
d9d52b51 | 625 | if ((len -= tun->vnet_hdr_sz) > count) |
f43798c2 RR |
626 | return -EINVAL; |
627 | ||
6f26c9a7 | 628 | if (memcpy_fromiovecend((void *)&gso, iv, offset, sizeof(gso))) |
f43798c2 RR |
629 | return -EFAULT; |
630 | ||
4909122f HX |
631 | if ((gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) && |
632 | gso.csum_start + gso.csum_offset + 2 > gso.hdr_len) | |
633 | gso.hdr_len = gso.csum_start + gso.csum_offset + 2; | |
634 | ||
f43798c2 RR |
635 | if (gso.hdr_len > len) |
636 | return -EINVAL; | |
d9d52b51 | 637 | offset += tun->vnet_hdr_sz; |
f43798c2 RR |
638 | } |
639 | ||
e01bf1c8 | 640 | if ((tun->flags & TUN_TYPE_MASK) == TUN_TAP_DEV) { |
a504b86e | 641 | align += NET_IP_ALIGN; |
0eca93bc HX |
642 | if (unlikely(len < ETH_HLEN || |
643 | (gso.hdr_len && gso.hdr_len < ETH_HLEN))) | |
e01bf1c8 RR |
644 | return -EINVAL; |
645 | } | |
6aa20a22 | 646 | |
33dccbb0 HX |
647 | skb = tun_alloc_skb(tun, align, len, gso.hdr_len, noblock); |
648 | if (IS_ERR(skb)) { | |
649 | if (PTR_ERR(skb) != -EAGAIN) | |
650 | tun->dev->stats.rx_dropped++; | |
651 | return PTR_ERR(skb); | |
1da177e4 LT |
652 | } |
653 | ||
6f26c9a7 | 654 | if (skb_copy_datagram_from_iovec(skb, 0, iv, offset, len)) { |
09f75cd7 | 655 | tun->dev->stats.rx_dropped++; |
8f22757e | 656 | kfree_skb(skb); |
1da177e4 | 657 | return -EFAULT; |
8f22757e | 658 | } |
1da177e4 | 659 | |
f43798c2 RR |
660 | if (gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) { |
661 | if (!skb_partial_csum_set(skb, gso.csum_start, | |
662 | gso.csum_offset)) { | |
663 | tun->dev->stats.rx_frame_errors++; | |
664 | kfree_skb(skb); | |
665 | return -EINVAL; | |
666 | } | |
88255375 | 667 | } |
f43798c2 | 668 | |
1da177e4 LT |
669 | switch (tun->flags & TUN_TYPE_MASK) { |
670 | case TUN_TUN_DEV: | |
f09f7ee2 AWC |
671 | if (tun->flags & TUN_NO_PI) { |
672 | switch (skb->data[0] & 0xf0) { | |
673 | case 0x40: | |
674 | pi.proto = htons(ETH_P_IP); | |
675 | break; | |
676 | case 0x60: | |
677 | pi.proto = htons(ETH_P_IPV6); | |
678 | break; | |
679 | default: | |
680 | tun->dev->stats.rx_dropped++; | |
681 | kfree_skb(skb); | |
682 | return -EINVAL; | |
683 | } | |
684 | } | |
685 | ||
459a98ed | 686 | skb_reset_mac_header(skb); |
1da177e4 | 687 | skb->protocol = pi.proto; |
4c13eb66 | 688 | skb->dev = tun->dev; |
1da177e4 LT |
689 | break; |
690 | case TUN_TAP_DEV: | |
691 | skb->protocol = eth_type_trans(skb, tun->dev); | |
692 | break; | |
6403eab1 | 693 | } |
1da177e4 | 694 | |
f43798c2 RR |
695 | if (gso.gso_type != VIRTIO_NET_HDR_GSO_NONE) { |
696 | pr_debug("GSO!\n"); | |
697 | switch (gso.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) { | |
698 | case VIRTIO_NET_HDR_GSO_TCPV4: | |
699 | skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4; | |
700 | break; | |
701 | case VIRTIO_NET_HDR_GSO_TCPV6: | |
702 | skb_shinfo(skb)->gso_type = SKB_GSO_TCPV6; | |
703 | break; | |
e36aa25a SS |
704 | case VIRTIO_NET_HDR_GSO_UDP: |
705 | skb_shinfo(skb)->gso_type = SKB_GSO_UDP; | |
706 | break; | |
f43798c2 RR |
707 | default: |
708 | tun->dev->stats.rx_frame_errors++; | |
709 | kfree_skb(skb); | |
710 | return -EINVAL; | |
711 | } | |
712 | ||
713 | if (gso.gso_type & VIRTIO_NET_HDR_GSO_ECN) | |
714 | skb_shinfo(skb)->gso_type |= SKB_GSO_TCP_ECN; | |
715 | ||
716 | skb_shinfo(skb)->gso_size = gso.gso_size; | |
717 | if (skb_shinfo(skb)->gso_size == 0) { | |
718 | tun->dev->stats.rx_frame_errors++; | |
719 | kfree_skb(skb); | |
720 | return -EINVAL; | |
721 | } | |
722 | ||
723 | /* Header must be checked, and gso_segs computed. */ | |
724 | skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY; | |
725 | skb_shinfo(skb)->gso_segs = 0; | |
726 | } | |
6aa20a22 | 727 | |
1da177e4 | 728 | netif_rx_ni(skb); |
6aa20a22 | 729 | |
09f75cd7 JG |
730 | tun->dev->stats.rx_packets++; |
731 | tun->dev->stats.rx_bytes += len; | |
1da177e4 LT |
732 | |
733 | return count; | |
6aa20a22 | 734 | } |
1da177e4 | 735 | |
ee0b3e67 BP |
736 | static ssize_t tun_chr_aio_write(struct kiocb *iocb, const struct iovec *iv, |
737 | unsigned long count, loff_t pos) | |
1da177e4 | 738 | { |
33dccbb0 | 739 | struct file *file = iocb->ki_filp; |
ab46d779 | 740 | struct tun_struct *tun = tun_get(file); |
631ab46b | 741 | ssize_t result; |
1da177e4 LT |
742 | |
743 | if (!tun) | |
744 | return -EBADFD; | |
745 | ||
6b8a66ee | 746 | tun_debug(KERN_INFO, tun, "tun_chr_write %ld\n", count); |
1da177e4 | 747 | |
6f26c9a7 | 748 | result = tun_get_user(tun, iv, iov_length(iv, count), |
33dccbb0 | 749 | file->f_flags & O_NONBLOCK); |
631ab46b EB |
750 | |
751 | tun_put(tun); | |
752 | return result; | |
1da177e4 LT |
753 | } |
754 | ||
1da177e4 | 755 | /* Put packet to the user space buffer */ |
6f7c156c | 756 | static ssize_t tun_put_user(struct tun_struct *tun, |
757 | struct sk_buff *skb, | |
758 | const struct iovec *iv, int len) | |
1da177e4 LT |
759 | { |
760 | struct tun_pi pi = { 0, skb->protocol }; | |
761 | ssize_t total = 0; | |
762 | ||
763 | if (!(tun->flags & TUN_NO_PI)) { | |
764 | if ((len -= sizeof(pi)) < 0) | |
765 | return -EINVAL; | |
766 | ||
767 | if (len < skb->len) { | |
768 | /* Packet will be striped */ | |
769 | pi.flags |= TUN_PKT_STRIP; | |
770 | } | |
6aa20a22 | 771 | |
43b39dcd | 772 | if (memcpy_toiovecend(iv, (void *) &pi, 0, sizeof(pi))) |
1da177e4 LT |
773 | return -EFAULT; |
774 | total += sizeof(pi); | |
6aa20a22 | 775 | } |
1da177e4 | 776 | |
f43798c2 RR |
777 | if (tun->flags & TUN_VNET_HDR) { |
778 | struct virtio_net_hdr gso = { 0 }; /* no info leak */ | |
d9d52b51 | 779 | if ((len -= tun->vnet_hdr_sz) < 0) |
f43798c2 RR |
780 | return -EINVAL; |
781 | ||
782 | if (skb_is_gso(skb)) { | |
783 | struct skb_shared_info *sinfo = skb_shinfo(skb); | |
784 | ||
785 | /* This is a hint as to how much should be linear. */ | |
786 | gso.hdr_len = skb_headlen(skb); | |
787 | gso.gso_size = sinfo->gso_size; | |
788 | if (sinfo->gso_type & SKB_GSO_TCPV4) | |
789 | gso.gso_type = VIRTIO_NET_HDR_GSO_TCPV4; | |
790 | else if (sinfo->gso_type & SKB_GSO_TCPV6) | |
791 | gso.gso_type = VIRTIO_NET_HDR_GSO_TCPV6; | |
e36aa25a SS |
792 | else if (sinfo->gso_type & SKB_GSO_UDP) |
793 | gso.gso_type = VIRTIO_NET_HDR_GSO_UDP; | |
ef3db4a5 | 794 | else { |
6b8a66ee | 795 | pr_err("unexpected GSO type: " |
ef3db4a5 MT |
796 | "0x%x, gso_size %d, hdr_len %d\n", |
797 | sinfo->gso_type, gso.gso_size, | |
798 | gso.hdr_len); | |
799 | print_hex_dump(KERN_ERR, "tun: ", | |
800 | DUMP_PREFIX_NONE, | |
801 | 16, 1, skb->head, | |
802 | min((int)gso.hdr_len, 64), true); | |
803 | WARN_ON_ONCE(1); | |
804 | return -EINVAL; | |
805 | } | |
f43798c2 RR |
806 | if (sinfo->gso_type & SKB_GSO_TCP_ECN) |
807 | gso.gso_type |= VIRTIO_NET_HDR_GSO_ECN; | |
808 | } else | |
809 | gso.gso_type = VIRTIO_NET_HDR_GSO_NONE; | |
810 | ||
811 | if (skb->ip_summed == CHECKSUM_PARTIAL) { | |
812 | gso.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM; | |
55508d60 | 813 | gso.csum_start = skb_checksum_start_offset(skb); |
f43798c2 | 814 | gso.csum_offset = skb->csum_offset; |
10a8d94a JW |
815 | } else if (skb->ip_summed == CHECKSUM_UNNECESSARY) { |
816 | gso.flags = VIRTIO_NET_HDR_F_DATA_VALID; | |
f43798c2 RR |
817 | } /* else everything is zero */ |
818 | ||
43b39dcd MT |
819 | if (unlikely(memcpy_toiovecend(iv, (void *)&gso, total, |
820 | sizeof(gso)))) | |
f43798c2 | 821 | return -EFAULT; |
d9d52b51 | 822 | total += tun->vnet_hdr_sz; |
f43798c2 RR |
823 | } |
824 | ||
1da177e4 LT |
825 | len = min_t(int, skb->len, len); |
826 | ||
43b39dcd | 827 | skb_copy_datagram_const_iovec(skb, 0, iv, total, len); |
05c2828c | 828 | total += skb->len; |
1da177e4 | 829 | |
09f75cd7 JG |
830 | tun->dev->stats.tx_packets++; |
831 | tun->dev->stats.tx_bytes += len; | |
1da177e4 LT |
832 | |
833 | return total; | |
834 | } | |
835 | ||
05c2828c MT |
836 | static ssize_t tun_do_read(struct tun_struct *tun, |
837 | struct kiocb *iocb, const struct iovec *iv, | |
838 | ssize_t len, int noblock) | |
1da177e4 | 839 | { |
1da177e4 LT |
840 | DECLARE_WAITQUEUE(wait, current); |
841 | struct sk_buff *skb; | |
05c2828c | 842 | ssize_t ret = 0; |
1da177e4 | 843 | |
6b8a66ee | 844 | tun_debug(KERN_INFO, tun, "tun_chr_read\n"); |
1da177e4 | 845 | |
61a5ff15 AK |
846 | if (unlikely(!noblock)) |
847 | add_wait_queue(&tun->wq.wait, &wait); | |
1da177e4 | 848 | while (len) { |
1da177e4 LT |
849 | current->state = TASK_INTERRUPTIBLE; |
850 | ||
851 | /* Read frames from the queue */ | |
89f56d1e | 852 | if (!(skb=skb_dequeue(&tun->socket.sk->sk_receive_queue))) { |
05c2828c | 853 | if (noblock) { |
1da177e4 LT |
854 | ret = -EAGAIN; |
855 | break; | |
856 | } | |
857 | if (signal_pending(current)) { | |
858 | ret = -ERESTARTSYS; | |
859 | break; | |
860 | } | |
c70f1829 EB |
861 | if (tun->dev->reg_state != NETREG_REGISTERED) { |
862 | ret = -EIO; | |
863 | break; | |
864 | } | |
1da177e4 LT |
865 | |
866 | /* Nothing to read, let's sleep */ | |
867 | schedule(); | |
868 | continue; | |
869 | } | |
870 | netif_wake_queue(tun->dev); | |
871 | ||
43b39dcd | 872 | ret = tun_put_user(tun, skb, iv, len); |
f271b2cc MK |
873 | kfree_skb(skb); |
874 | break; | |
1da177e4 LT |
875 | } |
876 | ||
877 | current->state = TASK_RUNNING; | |
61a5ff15 AK |
878 | if (unlikely(!noblock)) |
879 | remove_wait_queue(&tun->wq.wait, &wait); | |
1da177e4 | 880 | |
05c2828c MT |
881 | return ret; |
882 | } | |
883 | ||
884 | static ssize_t tun_chr_aio_read(struct kiocb *iocb, const struct iovec *iv, | |
885 | unsigned long count, loff_t pos) | |
886 | { | |
887 | struct file *file = iocb->ki_filp; | |
888 | struct tun_file *tfile = file->private_data; | |
889 | struct tun_struct *tun = __tun_get(tfile); | |
890 | ssize_t len, ret; | |
891 | ||
892 | if (!tun) | |
893 | return -EBADFD; | |
894 | len = iov_length(iv, count); | |
895 | if (len < 0) { | |
896 | ret = -EINVAL; | |
897 | goto out; | |
898 | } | |
899 | ||
900 | ret = tun_do_read(tun, iocb, iv, len, file->f_flags & O_NONBLOCK); | |
901 | ret = min_t(ssize_t, ret, len); | |
631ab46b EB |
902 | out: |
903 | tun_put(tun); | |
1da177e4 LT |
904 | return ret; |
905 | } | |
906 | ||
1da177e4 LT |
907 | static void tun_setup(struct net_device *dev) |
908 | { | |
909 | struct tun_struct *tun = netdev_priv(dev); | |
910 | ||
1da177e4 | 911 | tun->owner = -1; |
8c644623 | 912 | tun->group = -1; |
1da177e4 | 913 | |
1da177e4 | 914 | dev->ethtool_ops = &tun_ethtool_ops; |
9c3fea6a | 915 | dev->destructor = tun_free_netdev; |
1da177e4 LT |
916 | } |
917 | ||
f019a7a5 EB |
918 | /* Trivial set of netlink ops to allow deleting tun or tap |
919 | * device with netlink. | |
920 | */ | |
921 | static int tun_validate(struct nlattr *tb[], struct nlattr *data[]) | |
922 | { | |
923 | return -EINVAL; | |
924 | } | |
925 | ||
926 | static struct rtnl_link_ops tun_link_ops __read_mostly = { | |
927 | .kind = DRV_NAME, | |
928 | .priv_size = sizeof(struct tun_struct), | |
929 | .setup = tun_setup, | |
930 | .validate = tun_validate, | |
931 | }; | |
932 | ||
33dccbb0 HX |
933 | static void tun_sock_write_space(struct sock *sk) |
934 | { | |
935 | struct tun_struct *tun; | |
43815482 | 936 | wait_queue_head_t *wqueue; |
33dccbb0 HX |
937 | |
938 | if (!sock_writeable(sk)) | |
939 | return; | |
940 | ||
33dccbb0 HX |
941 | if (!test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags)) |
942 | return; | |
943 | ||
43815482 ED |
944 | wqueue = sk_sleep(sk); |
945 | if (wqueue && waitqueue_active(wqueue)) | |
946 | wake_up_interruptible_sync_poll(wqueue, POLLOUT | | |
05c2828c | 947 | POLLWRNORM | POLLWRBAND); |
c722c625 | 948 | |
80924e5f | 949 | tun = tun_sk(sk)->tun; |
33dccbb0 HX |
950 | kill_fasync(&tun->fasync, SIGIO, POLL_OUT); |
951 | } | |
952 | ||
953 | static void tun_sock_destruct(struct sock *sk) | |
954 | { | |
80924e5f | 955 | free_netdev(tun_sk(sk)->tun->dev); |
33dccbb0 HX |
956 | } |
957 | ||
05c2828c MT |
958 | static int tun_sendmsg(struct kiocb *iocb, struct socket *sock, |
959 | struct msghdr *m, size_t total_len) | |
960 | { | |
961 | struct tun_struct *tun = container_of(sock, struct tun_struct, socket); | |
962 | return tun_get_user(tun, m->msg_iov, total_len, | |
963 | m->msg_flags & MSG_DONTWAIT); | |
964 | } | |
965 | ||
966 | static int tun_recvmsg(struct kiocb *iocb, struct socket *sock, | |
967 | struct msghdr *m, size_t total_len, | |
968 | int flags) | |
969 | { | |
970 | struct tun_struct *tun = container_of(sock, struct tun_struct, socket); | |
971 | int ret; | |
972 | if (flags & ~(MSG_DONTWAIT|MSG_TRUNC)) | |
973 | return -EINVAL; | |
974 | ret = tun_do_read(tun, iocb, m->msg_iov, total_len, | |
975 | flags & MSG_DONTWAIT); | |
976 | if (ret > total_len) { | |
977 | m->msg_flags |= MSG_TRUNC; | |
978 | ret = flags & MSG_TRUNC ? ret : total_len; | |
979 | } | |
980 | return ret; | |
981 | } | |
982 | ||
1ab5ecb9 SK |
983 | static int tun_release(struct socket *sock) |
984 | { | |
985 | if (sock->sk) | |
986 | sock_put(sock->sk); | |
987 | return 0; | |
988 | } | |
989 | ||
05c2828c MT |
990 | /* Ops structure to mimic raw sockets with tun */ |
991 | static const struct proto_ops tun_socket_ops = { | |
992 | .sendmsg = tun_sendmsg, | |
993 | .recvmsg = tun_recvmsg, | |
1ab5ecb9 | 994 | .release = tun_release, |
05c2828c MT |
995 | }; |
996 | ||
33dccbb0 HX |
997 | static struct proto tun_proto = { |
998 | .name = "tun", | |
999 | .owner = THIS_MODULE, | |
1000 | .obj_size = sizeof(struct tun_sock), | |
1001 | }; | |
f019a7a5 | 1002 | |
980c9e8c DW |
1003 | static int tun_flags(struct tun_struct *tun) |
1004 | { | |
1005 | int flags = 0; | |
1006 | ||
1007 | if (tun->flags & TUN_TUN_DEV) | |
1008 | flags |= IFF_TUN; | |
1009 | else | |
1010 | flags |= IFF_TAP; | |
1011 | ||
1012 | if (tun->flags & TUN_NO_PI) | |
1013 | flags |= IFF_NO_PI; | |
1014 | ||
1015 | if (tun->flags & TUN_ONE_QUEUE) | |
1016 | flags |= IFF_ONE_QUEUE; | |
1017 | ||
1018 | if (tun->flags & TUN_VNET_HDR) | |
1019 | flags |= IFF_VNET_HDR; | |
1020 | ||
1021 | return flags; | |
1022 | } | |
1023 | ||
1024 | static ssize_t tun_show_flags(struct device *dev, struct device_attribute *attr, | |
1025 | char *buf) | |
1026 | { | |
1027 | struct tun_struct *tun = netdev_priv(to_net_dev(dev)); | |
1028 | return sprintf(buf, "0x%x\n", tun_flags(tun)); | |
1029 | } | |
1030 | ||
1031 | static ssize_t tun_show_owner(struct device *dev, struct device_attribute *attr, | |
1032 | char *buf) | |
1033 | { | |
1034 | struct tun_struct *tun = netdev_priv(to_net_dev(dev)); | |
1035 | return sprintf(buf, "%d\n", tun->owner); | |
1036 | } | |
1037 | ||
1038 | static ssize_t tun_show_group(struct device *dev, struct device_attribute *attr, | |
1039 | char *buf) | |
1040 | { | |
1041 | struct tun_struct *tun = netdev_priv(to_net_dev(dev)); | |
1042 | return sprintf(buf, "%d\n", tun->group); | |
1043 | } | |
1044 | ||
1045 | static DEVICE_ATTR(tun_flags, 0444, tun_show_flags, NULL); | |
1046 | static DEVICE_ATTR(owner, 0444, tun_show_owner, NULL); | |
1047 | static DEVICE_ATTR(group, 0444, tun_show_group, NULL); | |
1048 | ||
d647a591 | 1049 | static int tun_set_iff(struct net *net, struct file *file, struct ifreq *ifr) |
1da177e4 | 1050 | { |
33dccbb0 | 1051 | struct sock *sk; |
1da177e4 LT |
1052 | struct tun_struct *tun; |
1053 | struct net_device *dev; | |
1054 | int err; | |
1055 | ||
74a3e5a7 EB |
1056 | dev = __dev_get_by_name(net, ifr->ifr_name); |
1057 | if (dev) { | |
2b980dbd PM |
1058 | const struct cred *cred = current_cred(); |
1059 | ||
f85ba780 DW |
1060 | if (ifr->ifr_flags & IFF_TUN_EXCL) |
1061 | return -EBUSY; | |
74a3e5a7 EB |
1062 | if ((ifr->ifr_flags & IFF_TUN) && dev->netdev_ops == &tun_netdev_ops) |
1063 | tun = netdev_priv(dev); | |
1064 | else if ((ifr->ifr_flags & IFF_TAP) && dev->netdev_ops == &tap_netdev_ops) | |
1065 | tun = netdev_priv(dev); | |
1066 | else | |
1067 | return -EINVAL; | |
1068 | ||
2b980dbd PM |
1069 | if (((tun->owner != -1 && cred->euid != tun->owner) || |
1070 | (tun->group != -1 && !in_egroup_p(tun->group))) && | |
1071 | !capable(CAP_NET_ADMIN)) | |
1072 | return -EPERM; | |
d7e9660a | 1073 | err = security_tun_dev_attach(tun->socket.sk); |
2b980dbd PM |
1074 | if (err < 0) |
1075 | return err; | |
1076 | ||
a7385ba2 EB |
1077 | err = tun_attach(tun, file); |
1078 | if (err < 0) | |
1079 | return err; | |
6aa20a22 | 1080 | } |
1da177e4 LT |
1081 | else { |
1082 | char *name; | |
1083 | unsigned long flags = 0; | |
1084 | ||
ca6bb5d7 DW |
1085 | if (!capable(CAP_NET_ADMIN)) |
1086 | return -EPERM; | |
2b980dbd PM |
1087 | err = security_tun_dev_create(); |
1088 | if (err < 0) | |
1089 | return err; | |
ca6bb5d7 | 1090 | |
1da177e4 LT |
1091 | /* Set dev type */ |
1092 | if (ifr->ifr_flags & IFF_TUN) { | |
1093 | /* TUN device */ | |
1094 | flags |= TUN_TUN_DEV; | |
1095 | name = "tun%d"; | |
1096 | } else if (ifr->ifr_flags & IFF_TAP) { | |
1097 | /* TAP device */ | |
1098 | flags |= TUN_TAP_DEV; | |
1099 | name = "tap%d"; | |
6aa20a22 | 1100 | } else |
36989b90 | 1101 | return -EINVAL; |
6aa20a22 | 1102 | |
1da177e4 LT |
1103 | if (*ifr->ifr_name) |
1104 | name = ifr->ifr_name; | |
1105 | ||
1106 | dev = alloc_netdev(sizeof(struct tun_struct), name, | |
1107 | tun_setup); | |
1108 | if (!dev) | |
1109 | return -ENOMEM; | |
1110 | ||
fc54c658 | 1111 | dev_net_set(dev, net); |
f019a7a5 | 1112 | dev->rtnl_link_ops = &tun_link_ops; |
758e43b7 | 1113 | |
1da177e4 LT |
1114 | tun = netdev_priv(dev); |
1115 | tun->dev = dev; | |
1116 | tun->flags = flags; | |
f271b2cc | 1117 | tun->txflt.count = 0; |
d9d52b51 | 1118 | tun->vnet_hdr_sz = sizeof(struct virtio_net_hdr); |
1da177e4 | 1119 | |
33dccbb0 | 1120 | err = -ENOMEM; |
1ab5ecb9 | 1121 | sk = sk_alloc(&init_net, AF_UNSPEC, GFP_KERNEL, &tun_proto); |
33dccbb0 HX |
1122 | if (!sk) |
1123 | goto err_free_dev; | |
1124 | ||
1ab5ecb9 | 1125 | sk_change_net(sk, net); |
43815482 ED |
1126 | tun->socket.wq = &tun->wq; |
1127 | init_waitqueue_head(&tun->wq.wait); | |
05c2828c | 1128 | tun->socket.ops = &tun_socket_ops; |
33dccbb0 HX |
1129 | sock_init_data(&tun->socket, sk); |
1130 | sk->sk_write_space = tun_sock_write_space; | |
33dccbb0 | 1131 | sk->sk_sndbuf = INT_MAX; |
33dccbb0 | 1132 | |
80924e5f | 1133 | tun_sk(sk)->tun = tun; |
33dccbb0 | 1134 | |
2b980dbd PM |
1135 | security_tun_dev_post_create(sk); |
1136 | ||
1da177e4 LT |
1137 | tun_net_init(dev); |
1138 | ||
88255375 MM |
1139 | dev->hw_features = NETIF_F_SG | NETIF_F_FRAGLIST | |
1140 | TUN_USER_FEATURES; | |
1141 | dev->features = dev->hw_features; | |
1142 | ||
1da177e4 LT |
1143 | err = register_netdevice(tun->dev); |
1144 | if (err < 0) | |
9c3fea6a HX |
1145 | goto err_free_sk; |
1146 | ||
980c9e8c DW |
1147 | if (device_create_file(&tun->dev->dev, &dev_attr_tun_flags) || |
1148 | device_create_file(&tun->dev->dev, &dev_attr_owner) || | |
1149 | device_create_file(&tun->dev->dev, &dev_attr_group)) | |
6b8a66ee | 1150 | pr_err("Failed to create tun sysfs files\n"); |
980c9e8c | 1151 | |
9c3fea6a | 1152 | sk->sk_destruct = tun_sock_destruct; |
a7385ba2 EB |
1153 | |
1154 | err = tun_attach(tun, file); | |
1155 | if (err < 0) | |
9c3fea6a | 1156 | goto failed; |
1da177e4 LT |
1157 | } |
1158 | ||
6b8a66ee | 1159 | tun_debug(KERN_INFO, tun, "tun_set_iff\n"); |
1da177e4 LT |
1160 | |
1161 | if (ifr->ifr_flags & IFF_NO_PI) | |
1162 | tun->flags |= TUN_NO_PI; | |
a26af1e0 NF |
1163 | else |
1164 | tun->flags &= ~TUN_NO_PI; | |
1da177e4 LT |
1165 | |
1166 | if (ifr->ifr_flags & IFF_ONE_QUEUE) | |
1167 | tun->flags |= TUN_ONE_QUEUE; | |
a26af1e0 NF |
1168 | else |
1169 | tun->flags &= ~TUN_ONE_QUEUE; | |
1da177e4 | 1170 | |
f43798c2 RR |
1171 | if (ifr->ifr_flags & IFF_VNET_HDR) |
1172 | tun->flags |= TUN_VNET_HDR; | |
1173 | else | |
1174 | tun->flags &= ~TUN_VNET_HDR; | |
1175 | ||
e35259a9 MK |
1176 | /* Make sure persistent devices do not get stuck in |
1177 | * xoff state. | |
1178 | */ | |
1179 | if (netif_running(tun->dev)) | |
1180 | netif_wake_queue(tun->dev); | |
1181 | ||
1da177e4 LT |
1182 | strcpy(ifr->ifr_name, tun->dev->name); |
1183 | return 0; | |
1184 | ||
33dccbb0 | 1185 | err_free_sk: |
1ab5ecb9 | 1186 | tun_free_netdev(dev); |
1da177e4 LT |
1187 | err_free_dev: |
1188 | free_netdev(dev); | |
1189 | failed: | |
1190 | return err; | |
1191 | } | |
1192 | ||
876bfd4d HX |
1193 | static int tun_get_iff(struct net *net, struct tun_struct *tun, |
1194 | struct ifreq *ifr) | |
e3b99556 | 1195 | { |
6b8a66ee | 1196 | tun_debug(KERN_INFO, tun, "tun_get_iff\n"); |
e3b99556 MM |
1197 | |
1198 | strcpy(ifr->ifr_name, tun->dev->name); | |
1199 | ||
980c9e8c | 1200 | ifr->ifr_flags = tun_flags(tun); |
e3b99556 MM |
1201 | |
1202 | return 0; | |
1203 | } | |
1204 | ||
5228ddc9 RR |
1205 | /* This is like a cut-down ethtool ops, except done via tun fd so no |
1206 | * privs required. */ | |
88255375 | 1207 | static int set_offload(struct tun_struct *tun, unsigned long arg) |
5228ddc9 | 1208 | { |
c8f44aff | 1209 | netdev_features_t features = 0; |
5228ddc9 RR |
1210 | |
1211 | if (arg & TUN_F_CSUM) { | |
88255375 | 1212 | features |= NETIF_F_HW_CSUM; |
5228ddc9 RR |
1213 | arg &= ~TUN_F_CSUM; |
1214 | ||
1215 | if (arg & (TUN_F_TSO4|TUN_F_TSO6)) { | |
1216 | if (arg & TUN_F_TSO_ECN) { | |
1217 | features |= NETIF_F_TSO_ECN; | |
1218 | arg &= ~TUN_F_TSO_ECN; | |
1219 | } | |
1220 | if (arg & TUN_F_TSO4) | |
1221 | features |= NETIF_F_TSO; | |
1222 | if (arg & TUN_F_TSO6) | |
1223 | features |= NETIF_F_TSO6; | |
1224 | arg &= ~(TUN_F_TSO4|TUN_F_TSO6); | |
1225 | } | |
e36aa25a SS |
1226 | |
1227 | if (arg & TUN_F_UFO) { | |
1228 | features |= NETIF_F_UFO; | |
1229 | arg &= ~TUN_F_UFO; | |
1230 | } | |
5228ddc9 RR |
1231 | } |
1232 | ||
1233 | /* This gives the user a way to test for new features in future by | |
1234 | * trying to set them. */ | |
1235 | if (arg) | |
1236 | return -EINVAL; | |
1237 | ||
88255375 MM |
1238 | tun->set_features = features; |
1239 | netdev_update_features(tun->dev); | |
5228ddc9 RR |
1240 | |
1241 | return 0; | |
1242 | } | |
1243 | ||
50857e2a AB |
1244 | static long __tun_chr_ioctl(struct file *file, unsigned int cmd, |
1245 | unsigned long arg, int ifreq_len) | |
1da177e4 | 1246 | { |
36b50bab | 1247 | struct tun_file *tfile = file->private_data; |
631ab46b | 1248 | struct tun_struct *tun; |
1da177e4 | 1249 | void __user* argp = (void __user*)arg; |
99405162 | 1250 | struct sock_fprog fprog; |
1da177e4 | 1251 | struct ifreq ifr; |
33dccbb0 | 1252 | int sndbuf; |
d9d52b51 | 1253 | int vnet_hdr_sz; |
f271b2cc | 1254 | int ret; |
1da177e4 LT |
1255 | |
1256 | if (cmd == TUNSETIFF || _IOC_TYPE(cmd) == 0x89) | |
50857e2a | 1257 | if (copy_from_user(&ifr, argp, ifreq_len)) |
1da177e4 LT |
1258 | return -EFAULT; |
1259 | ||
631ab46b EB |
1260 | if (cmd == TUNGETFEATURES) { |
1261 | /* Currently this just means: "what IFF flags are valid?". | |
1262 | * This is needed because we never checked for invalid flags on | |
1263 | * TUNSETIFF. */ | |
1264 | return put_user(IFF_TUN | IFF_TAP | IFF_NO_PI | IFF_ONE_QUEUE | | |
1265 | IFF_VNET_HDR, | |
1266 | (unsigned int __user*)argp); | |
1267 | } | |
1268 | ||
876bfd4d HX |
1269 | rtnl_lock(); |
1270 | ||
36b50bab | 1271 | tun = __tun_get(tfile); |
1da177e4 | 1272 | if (cmd == TUNSETIFF && !tun) { |
1da177e4 LT |
1273 | ifr.ifr_name[IFNAMSIZ-1] = '\0'; |
1274 | ||
876bfd4d | 1275 | ret = tun_set_iff(tfile->net, file, &ifr); |
1da177e4 | 1276 | |
876bfd4d HX |
1277 | if (ret) |
1278 | goto unlock; | |
1da177e4 | 1279 | |
50857e2a | 1280 | if (copy_to_user(argp, &ifr, ifreq_len)) |
876bfd4d HX |
1281 | ret = -EFAULT; |
1282 | goto unlock; | |
1da177e4 LT |
1283 | } |
1284 | ||
876bfd4d | 1285 | ret = -EBADFD; |
1da177e4 | 1286 | if (!tun) |
876bfd4d | 1287 | goto unlock; |
1da177e4 | 1288 | |
6b8a66ee | 1289 | tun_debug(KERN_INFO, tun, "tun_chr_ioctl cmd %d\n", cmd); |
1da177e4 | 1290 | |
631ab46b | 1291 | ret = 0; |
1da177e4 | 1292 | switch (cmd) { |
e3b99556 | 1293 | case TUNGETIFF: |
876bfd4d | 1294 | ret = tun_get_iff(current->nsproxy->net_ns, tun, &ifr); |
e3b99556 | 1295 | if (ret) |
631ab46b | 1296 | break; |
e3b99556 | 1297 | |
50857e2a | 1298 | if (copy_to_user(argp, &ifr, ifreq_len)) |
631ab46b | 1299 | ret = -EFAULT; |
e3b99556 MM |
1300 | break; |
1301 | ||
1da177e4 LT |
1302 | case TUNSETNOCSUM: |
1303 | /* Disable/Enable checksum */ | |
1da177e4 | 1304 | |
88255375 MM |
1305 | /* [unimplemented] */ |
1306 | tun_debug(KERN_INFO, tun, "ignored: set checksum %s\n", | |
6b8a66ee | 1307 | arg ? "disabled" : "enabled"); |
1da177e4 LT |
1308 | break; |
1309 | ||
1310 | case TUNSETPERSIST: | |
1311 | /* Disable/Enable persist mode */ | |
1312 | if (arg) | |
1313 | tun->flags |= TUN_PERSIST; | |
1314 | else | |
1315 | tun->flags &= ~TUN_PERSIST; | |
1316 | ||
6b8a66ee JP |
1317 | tun_debug(KERN_INFO, tun, "persist %s\n", |
1318 | arg ? "enabled" : "disabled"); | |
1da177e4 LT |
1319 | break; |
1320 | ||
1321 | case TUNSETOWNER: | |
1322 | /* Set owner of the device */ | |
1323 | tun->owner = (uid_t) arg; | |
1324 | ||
6b8a66ee | 1325 | tun_debug(KERN_INFO, tun, "owner set to %d\n", tun->owner); |
1da177e4 LT |
1326 | break; |
1327 | ||
8c644623 GG |
1328 | case TUNSETGROUP: |
1329 | /* Set group of the device */ | |
1330 | tun->group= (gid_t) arg; | |
1331 | ||
6b8a66ee | 1332 | tun_debug(KERN_INFO, tun, "group set to %d\n", tun->group); |
8c644623 GG |
1333 | break; |
1334 | ||
ff4cc3ac MK |
1335 | case TUNSETLINK: |
1336 | /* Only allow setting the type when the interface is down */ | |
1337 | if (tun->dev->flags & IFF_UP) { | |
6b8a66ee JP |
1338 | tun_debug(KERN_INFO, tun, |
1339 | "Linktype set failed because interface is up\n"); | |
48abfe05 | 1340 | ret = -EBUSY; |
ff4cc3ac MK |
1341 | } else { |
1342 | tun->dev->type = (int) arg; | |
6b8a66ee JP |
1343 | tun_debug(KERN_INFO, tun, "linktype set to %d\n", |
1344 | tun->dev->type); | |
48abfe05 | 1345 | ret = 0; |
ff4cc3ac | 1346 | } |
631ab46b | 1347 | break; |
ff4cc3ac | 1348 | |
1da177e4 LT |
1349 | #ifdef TUN_DEBUG |
1350 | case TUNSETDEBUG: | |
1351 | tun->debug = arg; | |
1352 | break; | |
1353 | #endif | |
5228ddc9 | 1354 | case TUNSETOFFLOAD: |
88255375 | 1355 | ret = set_offload(tun, arg); |
631ab46b | 1356 | break; |
5228ddc9 | 1357 | |
f271b2cc MK |
1358 | case TUNSETTXFILTER: |
1359 | /* Can be set only for TAPs */ | |
631ab46b | 1360 | ret = -EINVAL; |
f271b2cc | 1361 | if ((tun->flags & TUN_TYPE_MASK) != TUN_TAP_DEV) |
631ab46b | 1362 | break; |
c0e5a8c2 | 1363 | ret = update_filter(&tun->txflt, (void __user *)arg); |
631ab46b | 1364 | break; |
1da177e4 LT |
1365 | |
1366 | case SIOCGIFHWADDR: | |
b595076a | 1367 | /* Get hw address */ |
f271b2cc MK |
1368 | memcpy(ifr.ifr_hwaddr.sa_data, tun->dev->dev_addr, ETH_ALEN); |
1369 | ifr.ifr_hwaddr.sa_family = tun->dev->type; | |
50857e2a | 1370 | if (copy_to_user(argp, &ifr, ifreq_len)) |
631ab46b EB |
1371 | ret = -EFAULT; |
1372 | break; | |
1da177e4 LT |
1373 | |
1374 | case SIOCSIFHWADDR: | |
f271b2cc | 1375 | /* Set hw address */ |
6b8a66ee JP |
1376 | tun_debug(KERN_DEBUG, tun, "set hw address: %pM\n", |
1377 | ifr.ifr_hwaddr.sa_data); | |
40102371 | 1378 | |
40102371 | 1379 | ret = dev_set_mac_address(tun->dev, &ifr.ifr_hwaddr); |
631ab46b | 1380 | break; |
33dccbb0 HX |
1381 | |
1382 | case TUNGETSNDBUF: | |
89f56d1e | 1383 | sndbuf = tun->socket.sk->sk_sndbuf; |
33dccbb0 HX |
1384 | if (copy_to_user(argp, &sndbuf, sizeof(sndbuf))) |
1385 | ret = -EFAULT; | |
1386 | break; | |
1387 | ||
1388 | case TUNSETSNDBUF: | |
1389 | if (copy_from_user(&sndbuf, argp, sizeof(sndbuf))) { | |
1390 | ret = -EFAULT; | |
1391 | break; | |
1392 | } | |
1393 | ||
89f56d1e | 1394 | tun->socket.sk->sk_sndbuf = sndbuf; |
33dccbb0 HX |
1395 | break; |
1396 | ||
d9d52b51 MT |
1397 | case TUNGETVNETHDRSZ: |
1398 | vnet_hdr_sz = tun->vnet_hdr_sz; | |
1399 | if (copy_to_user(argp, &vnet_hdr_sz, sizeof(vnet_hdr_sz))) | |
1400 | ret = -EFAULT; | |
1401 | break; | |
1402 | ||
1403 | case TUNSETVNETHDRSZ: | |
1404 | if (copy_from_user(&vnet_hdr_sz, argp, sizeof(vnet_hdr_sz))) { | |
1405 | ret = -EFAULT; | |
1406 | break; | |
1407 | } | |
1408 | if (vnet_hdr_sz < (int)sizeof(struct virtio_net_hdr)) { | |
1409 | ret = -EINVAL; | |
1410 | break; | |
1411 | } | |
1412 | ||
1413 | tun->vnet_hdr_sz = vnet_hdr_sz; | |
1414 | break; | |
1415 | ||
99405162 MT |
1416 | case TUNATTACHFILTER: |
1417 | /* Can be set only for TAPs */ | |
1418 | ret = -EINVAL; | |
1419 | if ((tun->flags & TUN_TYPE_MASK) != TUN_TAP_DEV) | |
1420 | break; | |
1421 | ret = -EFAULT; | |
1422 | if (copy_from_user(&fprog, argp, sizeof(fprog))) | |
1423 | break; | |
1424 | ||
1425 | ret = sk_attach_filter(&fprog, tun->socket.sk); | |
1426 | break; | |
1427 | ||
1428 | case TUNDETACHFILTER: | |
1429 | /* Can be set only for TAPs */ | |
1430 | ret = -EINVAL; | |
1431 | if ((tun->flags & TUN_TYPE_MASK) != TUN_TAP_DEV) | |
1432 | break; | |
1433 | ret = sk_detach_filter(tun->socket.sk); | |
1434 | break; | |
1435 | ||
1da177e4 | 1436 | default: |
631ab46b EB |
1437 | ret = -EINVAL; |
1438 | break; | |
ee289b64 | 1439 | } |
1da177e4 | 1440 | |
876bfd4d HX |
1441 | unlock: |
1442 | rtnl_unlock(); | |
1443 | if (tun) | |
1444 | tun_put(tun); | |
631ab46b | 1445 | return ret; |
1da177e4 LT |
1446 | } |
1447 | ||
50857e2a AB |
1448 | static long tun_chr_ioctl(struct file *file, |
1449 | unsigned int cmd, unsigned long arg) | |
1450 | { | |
1451 | return __tun_chr_ioctl(file, cmd, arg, sizeof (struct ifreq)); | |
1452 | } | |
1453 | ||
1454 | #ifdef CONFIG_COMPAT | |
1455 | static long tun_chr_compat_ioctl(struct file *file, | |
1456 | unsigned int cmd, unsigned long arg) | |
1457 | { | |
1458 | switch (cmd) { | |
1459 | case TUNSETIFF: | |
1460 | case TUNGETIFF: | |
1461 | case TUNSETTXFILTER: | |
1462 | case TUNGETSNDBUF: | |
1463 | case TUNSETSNDBUF: | |
1464 | case SIOCGIFHWADDR: | |
1465 | case SIOCSIFHWADDR: | |
1466 | arg = (unsigned long)compat_ptr(arg); | |
1467 | break; | |
1468 | default: | |
1469 | arg = (compat_ulong_t)arg; | |
1470 | break; | |
1471 | } | |
1472 | ||
1473 | /* | |
1474 | * compat_ifreq is shorter than ifreq, so we must not access beyond | |
1475 | * the end of that structure. All fields that are used in this | |
1476 | * driver are compatible though, we don't need to convert the | |
1477 | * contents. | |
1478 | */ | |
1479 | return __tun_chr_ioctl(file, cmd, arg, sizeof(struct compat_ifreq)); | |
1480 | } | |
1481 | #endif /* CONFIG_COMPAT */ | |
1482 | ||
1da177e4 LT |
1483 | static int tun_chr_fasync(int fd, struct file *file, int on) |
1484 | { | |
631ab46b | 1485 | struct tun_struct *tun = tun_get(file); |
1da177e4 LT |
1486 | int ret; |
1487 | ||
1488 | if (!tun) | |
1489 | return -EBADFD; | |
1490 | ||
6b8a66ee | 1491 | tun_debug(KERN_INFO, tun, "tun_chr_fasync %d\n", on); |
1da177e4 LT |
1492 | |
1493 | if ((ret = fasync_helper(fd, file, on, &tun->fasync)) < 0) | |
9d319522 | 1494 | goto out; |
6aa20a22 | 1495 | |
1da177e4 | 1496 | if (on) { |
609d7fa9 | 1497 | ret = __f_setown(file, task_pid(current), PIDTYPE_PID, 0); |
1da177e4 | 1498 | if (ret) |
9d319522 | 1499 | goto out; |
1da177e4 | 1500 | tun->flags |= TUN_FASYNC; |
6aa20a22 | 1501 | } else |
1da177e4 | 1502 | tun->flags &= ~TUN_FASYNC; |
9d319522 JC |
1503 | ret = 0; |
1504 | out: | |
631ab46b | 1505 | tun_put(tun); |
9d319522 | 1506 | return ret; |
1da177e4 LT |
1507 | } |
1508 | ||
1509 | static int tun_chr_open(struct inode *inode, struct file * file) | |
1510 | { | |
631ab46b | 1511 | struct tun_file *tfile; |
deed49fb | 1512 | |
6b8a66ee | 1513 | DBG1(KERN_INFO, "tunX: tun_chr_open\n"); |
631ab46b EB |
1514 | |
1515 | tfile = kmalloc(sizeof(*tfile), GFP_KERNEL); | |
1516 | if (!tfile) | |
1517 | return -ENOMEM; | |
c70f1829 | 1518 | atomic_set(&tfile->count, 0); |
631ab46b | 1519 | tfile->tun = NULL; |
36b50bab | 1520 | tfile->net = get_net(current->nsproxy->net_ns); |
631ab46b | 1521 | file->private_data = tfile; |
1da177e4 LT |
1522 | return 0; |
1523 | } | |
1524 | ||
1525 | static int tun_chr_close(struct inode *inode, struct file *file) | |
1526 | { | |
631ab46b | 1527 | struct tun_file *tfile = file->private_data; |
f0a4d0e5 | 1528 | struct tun_struct *tun; |
1da177e4 | 1529 | |
f0a4d0e5 | 1530 | tun = __tun_get(tfile); |
631ab46b | 1531 | if (tun) { |
d23e4365 HX |
1532 | struct net_device *dev = tun->dev; |
1533 | ||
6b8a66ee | 1534 | tun_debug(KERN_INFO, tun, "tun_chr_close\n"); |
1da177e4 | 1535 | |
631ab46b | 1536 | __tun_detach(tun); |
1da177e4 | 1537 | |
3ad2f3fb | 1538 | /* If desirable, unregister the netdevice. */ |
d23e4365 HX |
1539 | if (!(tun->flags & TUN_PERSIST)) { |
1540 | rtnl_lock(); | |
1541 | if (dev->reg_state == NETREG_REGISTERED) | |
1542 | unregister_netdevice(dev); | |
1543 | rtnl_unlock(); | |
1544 | } | |
631ab46b | 1545 | } |
1da177e4 | 1546 | |
9c3fea6a HX |
1547 | tun = tfile->tun; |
1548 | if (tun) | |
89f56d1e | 1549 | sock_put(tun->socket.sk); |
9c3fea6a | 1550 | |
36b50bab | 1551 | put_net(tfile->net); |
631ab46b | 1552 | kfree(tfile); |
1da177e4 LT |
1553 | |
1554 | return 0; | |
1555 | } | |
1556 | ||
d54b1fdb | 1557 | static const struct file_operations tun_fops = { |
6aa20a22 | 1558 | .owner = THIS_MODULE, |
1da177e4 | 1559 | .llseek = no_llseek, |
ee0b3e67 BP |
1560 | .read = do_sync_read, |
1561 | .aio_read = tun_chr_aio_read, | |
1562 | .write = do_sync_write, | |
1563 | .aio_write = tun_chr_aio_write, | |
1da177e4 | 1564 | .poll = tun_chr_poll, |
50857e2a AB |
1565 | .unlocked_ioctl = tun_chr_ioctl, |
1566 | #ifdef CONFIG_COMPAT | |
1567 | .compat_ioctl = tun_chr_compat_ioctl, | |
1568 | #endif | |
1da177e4 LT |
1569 | .open = tun_chr_open, |
1570 | .release = tun_chr_close, | |
6aa20a22 | 1571 | .fasync = tun_chr_fasync |
1da177e4 LT |
1572 | }; |
1573 | ||
1574 | static struct miscdevice tun_miscdev = { | |
1575 | .minor = TUN_MINOR, | |
1576 | .name = "tun", | |
e454cea2 | 1577 | .nodename = "net/tun", |
1da177e4 | 1578 | .fops = &tun_fops, |
1da177e4 LT |
1579 | }; |
1580 | ||
1581 | /* ethtool interface */ | |
1582 | ||
1583 | static int tun_get_settings(struct net_device *dev, struct ethtool_cmd *cmd) | |
1584 | { | |
1585 | cmd->supported = 0; | |
1586 | cmd->advertising = 0; | |
70739497 | 1587 | ethtool_cmd_speed_set(cmd, SPEED_10); |
1da177e4 LT |
1588 | cmd->duplex = DUPLEX_FULL; |
1589 | cmd->port = PORT_TP; | |
1590 | cmd->phy_address = 0; | |
1591 | cmd->transceiver = XCVR_INTERNAL; | |
1592 | cmd->autoneg = AUTONEG_DISABLE; | |
1593 | cmd->maxtxpkt = 0; | |
1594 | cmd->maxrxpkt = 0; | |
1595 | return 0; | |
1596 | } | |
1597 | ||
1598 | static void tun_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info) | |
1599 | { | |
1600 | struct tun_struct *tun = netdev_priv(dev); | |
1601 | ||
33a5ba14 RJ |
1602 | strlcpy(info->driver, DRV_NAME, sizeof(info->driver)); |
1603 | strlcpy(info->version, DRV_VERSION, sizeof(info->version)); | |
1da177e4 LT |
1604 | |
1605 | switch (tun->flags & TUN_TYPE_MASK) { | |
1606 | case TUN_TUN_DEV: | |
33a5ba14 | 1607 | strlcpy(info->bus_info, "tun", sizeof(info->bus_info)); |
1da177e4 LT |
1608 | break; |
1609 | case TUN_TAP_DEV: | |
33a5ba14 | 1610 | strlcpy(info->bus_info, "tap", sizeof(info->bus_info)); |
1da177e4 LT |
1611 | break; |
1612 | } | |
1613 | } | |
1614 | ||
1615 | static u32 tun_get_msglevel(struct net_device *dev) | |
1616 | { | |
1617 | #ifdef TUN_DEBUG | |
1618 | struct tun_struct *tun = netdev_priv(dev); | |
1619 | return tun->debug; | |
1620 | #else | |
1621 | return -EOPNOTSUPP; | |
1622 | #endif | |
1623 | } | |
1624 | ||
1625 | static void tun_set_msglevel(struct net_device *dev, u32 value) | |
1626 | { | |
1627 | #ifdef TUN_DEBUG | |
1628 | struct tun_struct *tun = netdev_priv(dev); | |
1629 | tun->debug = value; | |
1630 | #endif | |
1631 | } | |
1632 | ||
7282d491 | 1633 | static const struct ethtool_ops tun_ethtool_ops = { |
1da177e4 LT |
1634 | .get_settings = tun_get_settings, |
1635 | .get_drvinfo = tun_get_drvinfo, | |
1636 | .get_msglevel = tun_get_msglevel, | |
1637 | .set_msglevel = tun_set_msglevel, | |
bee31369 | 1638 | .get_link = ethtool_op_get_link, |
1da177e4 LT |
1639 | }; |
1640 | ||
79d17604 | 1641 | |
1da177e4 LT |
1642 | static int __init tun_init(void) |
1643 | { | |
1644 | int ret = 0; | |
1645 | ||
6b8a66ee JP |
1646 | pr_info("%s, %s\n", DRV_DESCRIPTION, DRV_VERSION); |
1647 | pr_info("%s\n", DRV_COPYRIGHT); | |
1da177e4 | 1648 | |
f019a7a5 | 1649 | ret = rtnl_link_register(&tun_link_ops); |
79d17604 | 1650 | if (ret) { |
6b8a66ee | 1651 | pr_err("Can't register link_ops\n"); |
f019a7a5 | 1652 | goto err_linkops; |
79d17604 PE |
1653 | } |
1654 | ||
1da177e4 | 1655 | ret = misc_register(&tun_miscdev); |
79d17604 | 1656 | if (ret) { |
6b8a66ee | 1657 | pr_err("Can't register misc device %d\n", TUN_MINOR); |
79d17604 PE |
1658 | goto err_misc; |
1659 | } | |
f019a7a5 | 1660 | return 0; |
79d17604 | 1661 | err_misc: |
f019a7a5 EB |
1662 | rtnl_link_unregister(&tun_link_ops); |
1663 | err_linkops: | |
1da177e4 LT |
1664 | return ret; |
1665 | } | |
1666 | ||
1667 | static void tun_cleanup(void) | |
1668 | { | |
6aa20a22 | 1669 | misc_deregister(&tun_miscdev); |
f019a7a5 | 1670 | rtnl_link_unregister(&tun_link_ops); |
1da177e4 LT |
1671 | } |
1672 | ||
05c2828c MT |
1673 | /* Get an underlying socket object from tun file. Returns error unless file is |
1674 | * attached to a device. The returned object works like a packet socket, it | |
1675 | * can be used for sock_sendmsg/sock_recvmsg. The caller is responsible for | |
1676 | * holding a reference to the file for as long as the socket is in use. */ | |
1677 | struct socket *tun_get_socket(struct file *file) | |
1678 | { | |
1679 | struct tun_struct *tun; | |
1680 | if (file->f_op != &tun_fops) | |
1681 | return ERR_PTR(-EINVAL); | |
1682 | tun = tun_get(file); | |
1683 | if (!tun) | |
1684 | return ERR_PTR(-EBADFD); | |
1685 | tun_put(tun); | |
1686 | return &tun->socket; | |
1687 | } | |
1688 | EXPORT_SYMBOL_GPL(tun_get_socket); | |
1689 | ||
1da177e4 LT |
1690 | module_init(tun_init); |
1691 | module_exit(tun_cleanup); | |
1692 | MODULE_DESCRIPTION(DRV_DESCRIPTION); | |
1693 | MODULE_AUTHOR(DRV_COPYRIGHT); | |
1694 | MODULE_LICENSE("GPL"); | |
1695 | MODULE_ALIAS_MISCDEV(TUN_MINOR); | |
578454ff | 1696 | MODULE_ALIAS("devname:net/tun"); |