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