x86/amd-iommu: Add per IOMMU reference counting
[linux/fpc-iii.git] / drivers / net / tun.c
blob4fdfa2ae5418e6c030be80b5568e3c12dcbbf3f9
1 /*
2 * TUN - Universal TUN/TAP device driver.
3 * Copyright (C) 1999-2002 Maxim Krasnyansky <maxk@qualcomm.com>
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.
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.
15 * $Id: tun.c,v 1.15 2002/03/01 02:44:24 maxk Exp $
19 * Changes:
21 * Mike Kershaw <dragorn@kismetwireless.net> 2005/08/14
22 * Add TUNSETLINK ioctl to set the link encapsulation
24 * Mark Smith <markzzzsmith@yahoo.com.au>
25 * Use random_ether_addr() for tap MAC address.
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
33 * Daniel Podlejski <underley@underley.eu.org>
34 * Modifications for 2.3.99-pre5 kernel.
37 #define DRV_NAME "tun"
38 #define DRV_VERSION "1.6"
39 #define DRV_DESCRIPTION "Universal TUN/TAP device driver"
40 #define DRV_COPYRIGHT "(C) 1999-2004 Max Krasnyansky <maxk@qualcomm.com>"
42 #include <linux/module.h>
43 #include <linux/errno.h>
44 #include <linux/kernel.h>
45 #include <linux/major.h>
46 #include <linux/slab.h>
47 #include <linux/smp_lock.h>
48 #include <linux/poll.h>
49 #include <linux/fcntl.h>
50 #include <linux/init.h>
51 #include <linux/skbuff.h>
52 #include <linux/netdevice.h>
53 #include <linux/etherdevice.h>
54 #include <linux/miscdevice.h>
55 #include <linux/ethtool.h>
56 #include <linux/rtnetlink.h>
57 #include <linux/if.h>
58 #include <linux/if_arp.h>
59 #include <linux/if_ether.h>
60 #include <linux/if_tun.h>
61 #include <linux/crc32.h>
62 #include <linux/nsproxy.h>
63 #include <linux/virtio_net.h>
64 #include <net/net_namespace.h>
65 #include <net/netns/generic.h>
66 #include <net/rtnetlink.h>
67 #include <net/sock.h>
69 #include <asm/system.h>
70 #include <asm/uaccess.h>
72 /* Uncomment to enable debugging */
73 /* #define TUN_DEBUG 1 */
75 #ifdef TUN_DEBUG
76 static int debug;
78 #define DBG if(tun->debug)printk
79 #define DBG1 if(debug==2)printk
80 #else
81 #define DBG( a... )
82 #define DBG1( a... )
83 #endif
85 #define FLT_EXACT_COUNT 8
86 struct tap_filter {
87 unsigned int count; /* Number of addrs. Zero means disabled */
88 u32 mask[2]; /* Mask of the hashed addrs */
89 unsigned char addr[FLT_EXACT_COUNT][ETH_ALEN];
92 struct tun_file {
93 atomic_t count;
94 struct tun_struct *tun;
95 struct net *net;
98 struct tun_sock;
100 struct tun_struct {
101 struct tun_file *tfile;
102 unsigned int flags;
103 uid_t owner;
104 gid_t group;
106 struct net_device *dev;
107 struct fasync_struct *fasync;
109 struct tap_filter txflt;
110 struct socket socket;
112 #ifdef TUN_DEBUG
113 int debug;
114 #endif
117 struct tun_sock {
118 struct sock sk;
119 struct tun_struct *tun;
122 static inline struct tun_sock *tun_sk(struct sock *sk)
124 return container_of(sk, struct tun_sock, sk);
127 static int tun_attach(struct tun_struct *tun, struct file *file)
129 struct tun_file *tfile = file->private_data;
130 int err;
132 ASSERT_RTNL();
134 netif_tx_lock_bh(tun->dev);
136 err = -EINVAL;
137 if (tfile->tun)
138 goto out;
140 err = -EBUSY;
141 if (tun->tfile)
142 goto out;
144 err = 0;
145 tfile->tun = tun;
146 tun->tfile = tfile;
147 dev_hold(tun->dev);
148 sock_hold(tun->socket.sk);
149 atomic_inc(&tfile->count);
151 out:
152 netif_tx_unlock_bh(tun->dev);
153 return err;
156 static void __tun_detach(struct tun_struct *tun)
158 /* Detach from net device */
159 netif_tx_lock_bh(tun->dev);
160 tun->tfile = NULL;
161 netif_tx_unlock_bh(tun->dev);
163 /* Drop read queue */
164 skb_queue_purge(&tun->socket.sk->sk_receive_queue);
166 /* Drop the extra count on the net device */
167 dev_put(tun->dev);
170 static void tun_detach(struct tun_struct *tun)
172 rtnl_lock();
173 __tun_detach(tun);
174 rtnl_unlock();
177 static struct tun_struct *__tun_get(struct tun_file *tfile)
179 struct tun_struct *tun = NULL;
181 if (atomic_inc_not_zero(&tfile->count))
182 tun = tfile->tun;
184 return tun;
187 static struct tun_struct *tun_get(struct file *file)
189 return __tun_get(file->private_data);
192 static void tun_put(struct tun_struct *tun)
194 struct tun_file *tfile = tun->tfile;
196 if (atomic_dec_and_test(&tfile->count))
197 tun_detach(tfile->tun);
200 /* TAP filterting */
201 static void addr_hash_set(u32 *mask, const u8 *addr)
203 int n = ether_crc(ETH_ALEN, addr) >> 26;
204 mask[n >> 5] |= (1 << (n & 31));
207 static unsigned int addr_hash_test(const u32 *mask, const u8 *addr)
209 int n = ether_crc(ETH_ALEN, addr) >> 26;
210 return mask[n >> 5] & (1 << (n & 31));
213 static int update_filter(struct tap_filter *filter, void __user *arg)
215 struct { u8 u[ETH_ALEN]; } *addr;
216 struct tun_filter uf;
217 int err, alen, n, nexact;
219 if (copy_from_user(&uf, arg, sizeof(uf)))
220 return -EFAULT;
222 if (!uf.count) {
223 /* Disabled */
224 filter->count = 0;
225 return 0;
228 alen = ETH_ALEN * uf.count;
229 addr = kmalloc(alen, GFP_KERNEL);
230 if (!addr)
231 return -ENOMEM;
233 if (copy_from_user(addr, arg + sizeof(uf), alen)) {
234 err = -EFAULT;
235 goto done;
238 /* The filter is updated without holding any locks. Which is
239 * perfectly safe. We disable it first and in the worst
240 * case we'll accept a few undesired packets. */
241 filter->count = 0;
242 wmb();
244 /* Use first set of addresses as an exact filter */
245 for (n = 0; n < uf.count && n < FLT_EXACT_COUNT; n++)
246 memcpy(filter->addr[n], addr[n].u, ETH_ALEN);
248 nexact = n;
250 /* Remaining multicast addresses are hashed,
251 * unicast will leave the filter disabled. */
252 memset(filter->mask, 0, sizeof(filter->mask));
253 for (; n < uf.count; n++) {
254 if (!is_multicast_ether_addr(addr[n].u)) {
255 err = 0; /* no filter */
256 goto done;
258 addr_hash_set(filter->mask, addr[n].u);
261 /* For ALLMULTI just set the mask to all ones.
262 * This overrides the mask populated above. */
263 if ((uf.flags & TUN_FLT_ALLMULTI))
264 memset(filter->mask, ~0, sizeof(filter->mask));
266 /* Now enable the filter */
267 wmb();
268 filter->count = nexact;
270 /* Return the number of exact filters */
271 err = nexact;
273 done:
274 kfree(addr);
275 return err;
278 /* Returns: 0 - drop, !=0 - accept */
279 static int run_filter(struct tap_filter *filter, const struct sk_buff *skb)
281 /* Cannot use eth_hdr(skb) here because skb_mac_hdr() is incorrect
282 * at this point. */
283 struct ethhdr *eh = (struct ethhdr *) skb->data;
284 int i;
286 /* Exact match */
287 for (i = 0; i < filter->count; i++)
288 if (!compare_ether_addr(eh->h_dest, filter->addr[i]))
289 return 1;
291 /* Inexact match (multicast only) */
292 if (is_multicast_ether_addr(eh->h_dest))
293 return addr_hash_test(filter->mask, eh->h_dest);
295 return 0;
299 * Checks whether the packet is accepted or not.
300 * Returns: 0 - drop, !=0 - accept
302 static int check_filter(struct tap_filter *filter, const struct sk_buff *skb)
304 if (!filter->count)
305 return 1;
307 return run_filter(filter, skb);
310 /* Network device part of the driver */
312 static const struct ethtool_ops tun_ethtool_ops;
314 /* Net device detach from fd. */
315 static void tun_net_uninit(struct net_device *dev)
317 struct tun_struct *tun = netdev_priv(dev);
318 struct tun_file *tfile = tun->tfile;
320 /* Inform the methods they need to stop using the dev.
322 if (tfile) {
323 wake_up_all(&tun->socket.wait);
324 if (atomic_dec_and_test(&tfile->count))
325 __tun_detach(tun);
329 static void tun_free_netdev(struct net_device *dev)
331 struct tun_struct *tun = netdev_priv(dev);
333 sock_put(tun->socket.sk);
336 /* Net device open. */
337 static int tun_net_open(struct net_device *dev)
339 netif_start_queue(dev);
340 return 0;
343 /* Net device close. */
344 static int tun_net_close(struct net_device *dev)
346 netif_stop_queue(dev);
347 return 0;
350 /* Net device start xmit */
351 static netdev_tx_t tun_net_xmit(struct sk_buff *skb, struct net_device *dev)
353 struct tun_struct *tun = netdev_priv(dev);
355 DBG(KERN_INFO "%s: tun_net_xmit %d\n", tun->dev->name, skb->len);
357 /* Drop packet if interface is not attached */
358 if (!tun->tfile)
359 goto drop;
361 /* Drop if the filter does not like it.
362 * This is a noop if the filter is disabled.
363 * Filter can be enabled only for the TAP devices. */
364 if (!check_filter(&tun->txflt, skb))
365 goto drop;
367 if (skb_queue_len(&tun->socket.sk->sk_receive_queue) >= dev->tx_queue_len) {
368 if (!(tun->flags & TUN_ONE_QUEUE)) {
369 /* Normal queueing mode. */
370 /* Packet scheduler handles dropping of further packets. */
371 netif_stop_queue(dev);
373 /* We won't see all dropped packets individually, so overrun
374 * error is more appropriate. */
375 dev->stats.tx_fifo_errors++;
376 } else {
377 /* Single queue mode.
378 * Driver handles dropping of all packets itself. */
379 goto drop;
383 /* Enqueue packet */
384 skb_queue_tail(&tun->socket.sk->sk_receive_queue, skb);
385 dev->trans_start = jiffies;
387 /* Notify and wake up reader process */
388 if (tun->flags & TUN_FASYNC)
389 kill_fasync(&tun->fasync, SIGIO, POLL_IN);
390 wake_up_interruptible(&tun->socket.wait);
391 return NETDEV_TX_OK;
393 drop:
394 dev->stats.tx_dropped++;
395 kfree_skb(skb);
396 return NETDEV_TX_OK;
399 static void tun_net_mclist(struct net_device *dev)
402 * This callback is supposed to deal with mc filter in
403 * _rx_ path and has nothing to do with the _tx_ path.
404 * In rx path we always accept everything userspace gives us.
406 return;
409 #define MIN_MTU 68
410 #define MAX_MTU 65535
412 static int
413 tun_net_change_mtu(struct net_device *dev, int new_mtu)
415 if (new_mtu < MIN_MTU || new_mtu + dev->hard_header_len > MAX_MTU)
416 return -EINVAL;
417 dev->mtu = new_mtu;
418 return 0;
421 static const struct net_device_ops tun_netdev_ops = {
422 .ndo_uninit = tun_net_uninit,
423 .ndo_open = tun_net_open,
424 .ndo_stop = tun_net_close,
425 .ndo_start_xmit = tun_net_xmit,
426 .ndo_change_mtu = tun_net_change_mtu,
429 static const struct net_device_ops tap_netdev_ops = {
430 .ndo_uninit = tun_net_uninit,
431 .ndo_open = tun_net_open,
432 .ndo_stop = tun_net_close,
433 .ndo_start_xmit = tun_net_xmit,
434 .ndo_change_mtu = tun_net_change_mtu,
435 .ndo_set_multicast_list = tun_net_mclist,
436 .ndo_set_mac_address = eth_mac_addr,
437 .ndo_validate_addr = eth_validate_addr,
440 /* Initialize net device. */
441 static void tun_net_init(struct net_device *dev)
443 struct tun_struct *tun = netdev_priv(dev);
445 switch (tun->flags & TUN_TYPE_MASK) {
446 case TUN_TUN_DEV:
447 dev->netdev_ops = &tun_netdev_ops;
449 /* Point-to-Point TUN Device */
450 dev->hard_header_len = 0;
451 dev->addr_len = 0;
452 dev->mtu = 1500;
454 /* Zero header length */
455 dev->type = ARPHRD_NONE;
456 dev->flags = IFF_POINTOPOINT | IFF_NOARP | IFF_MULTICAST;
457 dev->tx_queue_len = TUN_READQ_SIZE; /* We prefer our own queue length */
458 break;
460 case TUN_TAP_DEV:
461 dev->netdev_ops = &tap_netdev_ops;
462 /* Ethernet TAP Device */
463 ether_setup(dev);
465 random_ether_addr(dev->dev_addr);
467 dev->tx_queue_len = TUN_READQ_SIZE; /* We prefer our own queue length */
468 break;
472 /* Character device part */
474 /* Poll */
475 static unsigned int tun_chr_poll(struct file *file, poll_table * wait)
477 struct tun_file *tfile = file->private_data;
478 struct tun_struct *tun = __tun_get(tfile);
479 struct sock *sk;
480 unsigned int mask = 0;
482 if (!tun)
483 return POLLERR;
485 sk = tun->socket.sk;
487 DBG(KERN_INFO "%s: tun_chr_poll\n", tun->dev->name);
489 poll_wait(file, &tun->socket.wait, wait);
491 if (!skb_queue_empty(&sk->sk_receive_queue))
492 mask |= POLLIN | POLLRDNORM;
494 if (sock_writeable(sk) ||
495 (!test_and_set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags) &&
496 sock_writeable(sk)))
497 mask |= POLLOUT | POLLWRNORM;
499 if (tun->dev->reg_state != NETREG_REGISTERED)
500 mask = POLLERR;
502 tun_put(tun);
503 return mask;
506 /* prepad is the amount to reserve at front. len is length after that.
507 * linear is a hint as to how much to copy (usually headers). */
508 static inline struct sk_buff *tun_alloc_skb(struct tun_struct *tun,
509 size_t prepad, size_t len,
510 size_t linear, int noblock)
512 struct sock *sk = tun->socket.sk;
513 struct sk_buff *skb;
514 int err;
516 /* Under a page? Don't bother with paged skb. */
517 if (prepad + len < PAGE_SIZE || !linear)
518 linear = len;
520 skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
521 &err);
522 if (!skb)
523 return ERR_PTR(err);
525 skb_reserve(skb, prepad);
526 skb_put(skb, linear);
527 skb->data_len = len - linear;
528 skb->len += len - linear;
530 return skb;
533 /* Get packet from user space buffer */
534 static __inline__ ssize_t tun_get_user(struct tun_struct *tun,
535 const struct iovec *iv, size_t count,
536 int noblock)
538 struct tun_pi pi = { 0, cpu_to_be16(ETH_P_IP) };
539 struct sk_buff *skb;
540 size_t len = count, align = 0;
541 struct virtio_net_hdr gso = { 0 };
542 int offset = 0;
544 if (!(tun->flags & TUN_NO_PI)) {
545 if ((len -= sizeof(pi)) > count)
546 return -EINVAL;
548 if (memcpy_fromiovecend((void *)&pi, iv, 0, sizeof(pi)))
549 return -EFAULT;
550 offset += sizeof(pi);
553 if (tun->flags & TUN_VNET_HDR) {
554 if ((len -= sizeof(gso)) > count)
555 return -EINVAL;
557 if (memcpy_fromiovecend((void *)&gso, iv, offset, sizeof(gso)))
558 return -EFAULT;
560 if ((gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
561 gso.csum_start + gso.csum_offset + 2 > gso.hdr_len)
562 gso.hdr_len = gso.csum_start + gso.csum_offset + 2;
564 if (gso.hdr_len > len)
565 return -EINVAL;
566 offset += sizeof(gso);
569 if ((tun->flags & TUN_TYPE_MASK) == TUN_TAP_DEV) {
570 align = NET_IP_ALIGN;
571 if (unlikely(len < ETH_HLEN ||
572 (gso.hdr_len && gso.hdr_len < ETH_HLEN)))
573 return -EINVAL;
576 skb = tun_alloc_skb(tun, align, len, gso.hdr_len, noblock);
577 if (IS_ERR(skb)) {
578 if (PTR_ERR(skb) != -EAGAIN)
579 tun->dev->stats.rx_dropped++;
580 return PTR_ERR(skb);
583 if (skb_copy_datagram_from_iovec(skb, 0, iv, offset, len)) {
584 tun->dev->stats.rx_dropped++;
585 kfree_skb(skb);
586 return -EFAULT;
589 if (gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
590 if (!skb_partial_csum_set(skb, gso.csum_start,
591 gso.csum_offset)) {
592 tun->dev->stats.rx_frame_errors++;
593 kfree_skb(skb);
594 return -EINVAL;
596 } else if (tun->flags & TUN_NOCHECKSUM)
597 skb->ip_summed = CHECKSUM_UNNECESSARY;
599 switch (tun->flags & TUN_TYPE_MASK) {
600 case TUN_TUN_DEV:
601 if (tun->flags & TUN_NO_PI) {
602 switch (skb->data[0] & 0xf0) {
603 case 0x40:
604 pi.proto = htons(ETH_P_IP);
605 break;
606 case 0x60:
607 pi.proto = htons(ETH_P_IPV6);
608 break;
609 default:
610 tun->dev->stats.rx_dropped++;
611 kfree_skb(skb);
612 return -EINVAL;
616 skb_reset_mac_header(skb);
617 skb->protocol = pi.proto;
618 skb->dev = tun->dev;
619 break;
620 case TUN_TAP_DEV:
621 skb->protocol = eth_type_trans(skb, tun->dev);
622 break;
625 if (gso.gso_type != VIRTIO_NET_HDR_GSO_NONE) {
626 pr_debug("GSO!\n");
627 switch (gso.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
628 case VIRTIO_NET_HDR_GSO_TCPV4:
629 skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
630 break;
631 case VIRTIO_NET_HDR_GSO_TCPV6:
632 skb_shinfo(skb)->gso_type = SKB_GSO_TCPV6;
633 break;
634 case VIRTIO_NET_HDR_GSO_UDP:
635 skb_shinfo(skb)->gso_type = SKB_GSO_UDP;
636 break;
637 default:
638 tun->dev->stats.rx_frame_errors++;
639 kfree_skb(skb);
640 return -EINVAL;
643 if (gso.gso_type & VIRTIO_NET_HDR_GSO_ECN)
644 skb_shinfo(skb)->gso_type |= SKB_GSO_TCP_ECN;
646 skb_shinfo(skb)->gso_size = gso.gso_size;
647 if (skb_shinfo(skb)->gso_size == 0) {
648 tun->dev->stats.rx_frame_errors++;
649 kfree_skb(skb);
650 return -EINVAL;
653 /* Header must be checked, and gso_segs computed. */
654 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
655 skb_shinfo(skb)->gso_segs = 0;
658 netif_rx_ni(skb);
660 tun->dev->stats.rx_packets++;
661 tun->dev->stats.rx_bytes += len;
663 return count;
666 static ssize_t tun_chr_aio_write(struct kiocb *iocb, const struct iovec *iv,
667 unsigned long count, loff_t pos)
669 struct file *file = iocb->ki_filp;
670 struct tun_struct *tun = tun_get(file);
671 ssize_t result;
673 if (!tun)
674 return -EBADFD;
676 DBG(KERN_INFO "%s: tun_chr_write %ld\n", tun->dev->name, count);
678 result = tun_get_user(tun, iv, iov_length(iv, count),
679 file->f_flags & O_NONBLOCK);
681 tun_put(tun);
682 return result;
685 /* Put packet to the user space buffer */
686 static __inline__ ssize_t tun_put_user(struct tun_struct *tun,
687 struct sk_buff *skb,
688 const struct iovec *iv, int len)
690 struct tun_pi pi = { 0, skb->protocol };
691 ssize_t total = 0;
693 if (!(tun->flags & TUN_NO_PI)) {
694 if ((len -= sizeof(pi)) < 0)
695 return -EINVAL;
697 if (len < skb->len) {
698 /* Packet will be striped */
699 pi.flags |= TUN_PKT_STRIP;
702 if (memcpy_toiovecend(iv, (void *) &pi, 0, sizeof(pi)))
703 return -EFAULT;
704 total += sizeof(pi);
707 if (tun->flags & TUN_VNET_HDR) {
708 struct virtio_net_hdr gso = { 0 }; /* no info leak */
709 if ((len -= sizeof(gso)) < 0)
710 return -EINVAL;
712 if (skb_is_gso(skb)) {
713 struct skb_shared_info *sinfo = skb_shinfo(skb);
715 /* This is a hint as to how much should be linear. */
716 gso.hdr_len = skb_headlen(skb);
717 gso.gso_size = sinfo->gso_size;
718 if (sinfo->gso_type & SKB_GSO_TCPV4)
719 gso.gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
720 else if (sinfo->gso_type & SKB_GSO_TCPV6)
721 gso.gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
722 else if (sinfo->gso_type & SKB_GSO_UDP)
723 gso.gso_type = VIRTIO_NET_HDR_GSO_UDP;
724 else
725 BUG();
726 if (sinfo->gso_type & SKB_GSO_TCP_ECN)
727 gso.gso_type |= VIRTIO_NET_HDR_GSO_ECN;
728 } else
729 gso.gso_type = VIRTIO_NET_HDR_GSO_NONE;
731 if (skb->ip_summed == CHECKSUM_PARTIAL) {
732 gso.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
733 gso.csum_start = skb->csum_start - skb_headroom(skb);
734 gso.csum_offset = skb->csum_offset;
735 } /* else everything is zero */
737 if (unlikely(memcpy_toiovecend(iv, (void *)&gso, total,
738 sizeof(gso))))
739 return -EFAULT;
740 total += sizeof(gso);
743 len = min_t(int, skb->len, len);
745 skb_copy_datagram_const_iovec(skb, 0, iv, total, len);
746 total += len;
748 tun->dev->stats.tx_packets++;
749 tun->dev->stats.tx_bytes += len;
751 return total;
754 static ssize_t tun_chr_aio_read(struct kiocb *iocb, const struct iovec *iv,
755 unsigned long count, loff_t pos)
757 struct file *file = iocb->ki_filp;
758 struct tun_file *tfile = file->private_data;
759 struct tun_struct *tun = __tun_get(tfile);
760 DECLARE_WAITQUEUE(wait, current);
761 struct sk_buff *skb;
762 ssize_t len, ret = 0;
764 if (!tun)
765 return -EBADFD;
767 DBG(KERN_INFO "%s: tun_chr_read\n", tun->dev->name);
769 len = iov_length(iv, count);
770 if (len < 0) {
771 ret = -EINVAL;
772 goto out;
775 add_wait_queue(&tun->socket.wait, &wait);
776 while (len) {
777 current->state = TASK_INTERRUPTIBLE;
779 /* Read frames from the queue */
780 if (!(skb=skb_dequeue(&tun->socket.sk->sk_receive_queue))) {
781 if (file->f_flags & O_NONBLOCK) {
782 ret = -EAGAIN;
783 break;
785 if (signal_pending(current)) {
786 ret = -ERESTARTSYS;
787 break;
789 if (tun->dev->reg_state != NETREG_REGISTERED) {
790 ret = -EIO;
791 break;
794 /* Nothing to read, let's sleep */
795 schedule();
796 continue;
798 netif_wake_queue(tun->dev);
800 ret = tun_put_user(tun, skb, iv, len);
801 kfree_skb(skb);
802 break;
805 current->state = TASK_RUNNING;
806 remove_wait_queue(&tun->socket.wait, &wait);
808 out:
809 tun_put(tun);
810 return ret;
813 static void tun_setup(struct net_device *dev)
815 struct tun_struct *tun = netdev_priv(dev);
817 tun->owner = -1;
818 tun->group = -1;
820 dev->ethtool_ops = &tun_ethtool_ops;
821 dev->destructor = tun_free_netdev;
824 /* Trivial set of netlink ops to allow deleting tun or tap
825 * device with netlink.
827 static int tun_validate(struct nlattr *tb[], struct nlattr *data[])
829 return -EINVAL;
832 static struct rtnl_link_ops tun_link_ops __read_mostly = {
833 .kind = DRV_NAME,
834 .priv_size = sizeof(struct tun_struct),
835 .setup = tun_setup,
836 .validate = tun_validate,
839 static void tun_sock_write_space(struct sock *sk)
841 struct tun_struct *tun;
843 if (!sock_writeable(sk))
844 return;
846 if (!test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags))
847 return;
849 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
850 wake_up_interruptible_sync(sk->sk_sleep);
852 tun = container_of(sk, struct tun_sock, sk)->tun;
853 kill_fasync(&tun->fasync, SIGIO, POLL_OUT);
856 static void tun_sock_destruct(struct sock *sk)
858 free_netdev(container_of(sk, struct tun_sock, sk)->tun->dev);
861 static struct proto tun_proto = {
862 .name = "tun",
863 .owner = THIS_MODULE,
864 .obj_size = sizeof(struct tun_sock),
867 static int tun_flags(struct tun_struct *tun)
869 int flags = 0;
871 if (tun->flags & TUN_TUN_DEV)
872 flags |= IFF_TUN;
873 else
874 flags |= IFF_TAP;
876 if (tun->flags & TUN_NO_PI)
877 flags |= IFF_NO_PI;
879 if (tun->flags & TUN_ONE_QUEUE)
880 flags |= IFF_ONE_QUEUE;
882 if (tun->flags & TUN_VNET_HDR)
883 flags |= IFF_VNET_HDR;
885 return flags;
888 static ssize_t tun_show_flags(struct device *dev, struct device_attribute *attr,
889 char *buf)
891 struct tun_struct *tun = netdev_priv(to_net_dev(dev));
892 return sprintf(buf, "0x%x\n", tun_flags(tun));
895 static ssize_t tun_show_owner(struct device *dev, struct device_attribute *attr,
896 char *buf)
898 struct tun_struct *tun = netdev_priv(to_net_dev(dev));
899 return sprintf(buf, "%d\n", tun->owner);
902 static ssize_t tun_show_group(struct device *dev, struct device_attribute *attr,
903 char *buf)
905 struct tun_struct *tun = netdev_priv(to_net_dev(dev));
906 return sprintf(buf, "%d\n", tun->group);
909 static DEVICE_ATTR(tun_flags, 0444, tun_show_flags, NULL);
910 static DEVICE_ATTR(owner, 0444, tun_show_owner, NULL);
911 static DEVICE_ATTR(group, 0444, tun_show_group, NULL);
913 static int tun_set_iff(struct net *net, struct file *file, struct ifreq *ifr)
915 struct sock *sk;
916 struct tun_struct *tun;
917 struct net_device *dev;
918 int err;
920 dev = __dev_get_by_name(net, ifr->ifr_name);
921 if (dev) {
922 const struct cred *cred = current_cred();
924 if (ifr->ifr_flags & IFF_TUN_EXCL)
925 return -EBUSY;
926 if ((ifr->ifr_flags & IFF_TUN) && dev->netdev_ops == &tun_netdev_ops)
927 tun = netdev_priv(dev);
928 else if ((ifr->ifr_flags & IFF_TAP) && dev->netdev_ops == &tap_netdev_ops)
929 tun = netdev_priv(dev);
930 else
931 return -EINVAL;
933 if (((tun->owner != -1 && cred->euid != tun->owner) ||
934 (tun->group != -1 && !in_egroup_p(tun->group))) &&
935 !capable(CAP_NET_ADMIN))
936 return -EPERM;
937 err = security_tun_dev_attach(tun->socket.sk);
938 if (err < 0)
939 return err;
941 err = tun_attach(tun, file);
942 if (err < 0)
943 return err;
945 else {
946 char *name;
947 unsigned long flags = 0;
949 if (!capable(CAP_NET_ADMIN))
950 return -EPERM;
951 err = security_tun_dev_create();
952 if (err < 0)
953 return err;
955 /* Set dev type */
956 if (ifr->ifr_flags & IFF_TUN) {
957 /* TUN device */
958 flags |= TUN_TUN_DEV;
959 name = "tun%d";
960 } else if (ifr->ifr_flags & IFF_TAP) {
961 /* TAP device */
962 flags |= TUN_TAP_DEV;
963 name = "tap%d";
964 } else
965 return -EINVAL;
967 if (*ifr->ifr_name)
968 name = ifr->ifr_name;
970 dev = alloc_netdev(sizeof(struct tun_struct), name,
971 tun_setup);
972 if (!dev)
973 return -ENOMEM;
975 dev_net_set(dev, net);
976 dev->rtnl_link_ops = &tun_link_ops;
978 tun = netdev_priv(dev);
979 tun->dev = dev;
980 tun->flags = flags;
981 tun->txflt.count = 0;
983 err = -ENOMEM;
984 sk = sk_alloc(net, AF_UNSPEC, GFP_KERNEL, &tun_proto);
985 if (!sk)
986 goto err_free_dev;
988 init_waitqueue_head(&tun->socket.wait);
989 sock_init_data(&tun->socket, sk);
990 sk->sk_write_space = tun_sock_write_space;
991 sk->sk_sndbuf = INT_MAX;
993 container_of(sk, struct tun_sock, sk)->tun = tun;
995 security_tun_dev_post_create(sk);
997 tun_net_init(dev);
999 if (strchr(dev->name, '%')) {
1000 err = dev_alloc_name(dev, dev->name);
1001 if (err < 0)
1002 goto err_free_sk;
1005 err = register_netdevice(tun->dev);
1006 if (err < 0)
1007 goto err_free_sk;
1009 if (device_create_file(&tun->dev->dev, &dev_attr_tun_flags) ||
1010 device_create_file(&tun->dev->dev, &dev_attr_owner) ||
1011 device_create_file(&tun->dev->dev, &dev_attr_group))
1012 printk(KERN_ERR "Failed to create tun sysfs files\n");
1014 sk->sk_destruct = tun_sock_destruct;
1016 err = tun_attach(tun, file);
1017 if (err < 0)
1018 goto failed;
1021 DBG(KERN_INFO "%s: tun_set_iff\n", tun->dev->name);
1023 if (ifr->ifr_flags & IFF_NO_PI)
1024 tun->flags |= TUN_NO_PI;
1025 else
1026 tun->flags &= ~TUN_NO_PI;
1028 if (ifr->ifr_flags & IFF_ONE_QUEUE)
1029 tun->flags |= TUN_ONE_QUEUE;
1030 else
1031 tun->flags &= ~TUN_ONE_QUEUE;
1033 if (ifr->ifr_flags & IFF_VNET_HDR)
1034 tun->flags |= TUN_VNET_HDR;
1035 else
1036 tun->flags &= ~TUN_VNET_HDR;
1038 /* Make sure persistent devices do not get stuck in
1039 * xoff state.
1041 if (netif_running(tun->dev))
1042 netif_wake_queue(tun->dev);
1044 strcpy(ifr->ifr_name, tun->dev->name);
1045 return 0;
1047 err_free_sk:
1048 sock_put(sk);
1049 err_free_dev:
1050 free_netdev(dev);
1051 failed:
1052 return err;
1055 static int tun_get_iff(struct net *net, struct tun_struct *tun,
1056 struct ifreq *ifr)
1058 DBG(KERN_INFO "%s: tun_get_iff\n", tun->dev->name);
1060 strcpy(ifr->ifr_name, tun->dev->name);
1062 ifr->ifr_flags = tun_flags(tun);
1064 return 0;
1067 /* This is like a cut-down ethtool ops, except done via tun fd so no
1068 * privs required. */
1069 static int set_offload(struct net_device *dev, unsigned long arg)
1071 unsigned int old_features, features;
1073 old_features = dev->features;
1074 /* Unset features, set them as we chew on the arg. */
1075 features = (old_features & ~(NETIF_F_HW_CSUM|NETIF_F_SG|NETIF_F_FRAGLIST
1076 |NETIF_F_TSO_ECN|NETIF_F_TSO|NETIF_F_TSO6
1077 |NETIF_F_UFO));
1079 if (arg & TUN_F_CSUM) {
1080 features |= NETIF_F_HW_CSUM|NETIF_F_SG|NETIF_F_FRAGLIST;
1081 arg &= ~TUN_F_CSUM;
1083 if (arg & (TUN_F_TSO4|TUN_F_TSO6)) {
1084 if (arg & TUN_F_TSO_ECN) {
1085 features |= NETIF_F_TSO_ECN;
1086 arg &= ~TUN_F_TSO_ECN;
1088 if (arg & TUN_F_TSO4)
1089 features |= NETIF_F_TSO;
1090 if (arg & TUN_F_TSO6)
1091 features |= NETIF_F_TSO6;
1092 arg &= ~(TUN_F_TSO4|TUN_F_TSO6);
1095 if (arg & TUN_F_UFO) {
1096 features |= NETIF_F_UFO;
1097 arg &= ~TUN_F_UFO;
1101 /* This gives the user a way to test for new features in future by
1102 * trying to set them. */
1103 if (arg)
1104 return -EINVAL;
1106 dev->features = features;
1107 if (old_features != dev->features)
1108 netdev_features_change(dev);
1110 return 0;
1113 static long tun_chr_ioctl(struct file *file, unsigned int cmd,
1114 unsigned long arg)
1116 struct tun_file *tfile = file->private_data;
1117 struct tun_struct *tun;
1118 void __user* argp = (void __user*)arg;
1119 struct ifreq ifr;
1120 int sndbuf;
1121 int ret;
1123 if (cmd == TUNSETIFF || _IOC_TYPE(cmd) == 0x89)
1124 if (copy_from_user(&ifr, argp, sizeof ifr))
1125 return -EFAULT;
1127 if (cmd == TUNGETFEATURES) {
1128 /* Currently this just means: "what IFF flags are valid?".
1129 * This is needed because we never checked for invalid flags on
1130 * TUNSETIFF. */
1131 return put_user(IFF_TUN | IFF_TAP | IFF_NO_PI | IFF_ONE_QUEUE |
1132 IFF_VNET_HDR,
1133 (unsigned int __user*)argp);
1136 rtnl_lock();
1138 tun = __tun_get(tfile);
1139 if (cmd == TUNSETIFF && !tun) {
1140 ifr.ifr_name[IFNAMSIZ-1] = '\0';
1142 ret = tun_set_iff(tfile->net, file, &ifr);
1144 if (ret)
1145 goto unlock;
1147 if (copy_to_user(argp, &ifr, sizeof(ifr)))
1148 ret = -EFAULT;
1149 goto unlock;
1152 ret = -EBADFD;
1153 if (!tun)
1154 goto unlock;
1156 DBG(KERN_INFO "%s: tun_chr_ioctl cmd %d\n", tun->dev->name, cmd);
1158 ret = 0;
1159 switch (cmd) {
1160 case TUNGETIFF:
1161 ret = tun_get_iff(current->nsproxy->net_ns, tun, &ifr);
1162 if (ret)
1163 break;
1165 if (copy_to_user(argp, &ifr, sizeof(ifr)))
1166 ret = -EFAULT;
1167 break;
1169 case TUNSETNOCSUM:
1170 /* Disable/Enable checksum */
1171 if (arg)
1172 tun->flags |= TUN_NOCHECKSUM;
1173 else
1174 tun->flags &= ~TUN_NOCHECKSUM;
1176 DBG(KERN_INFO "%s: checksum %s\n",
1177 tun->dev->name, arg ? "disabled" : "enabled");
1178 break;
1180 case TUNSETPERSIST:
1181 /* Disable/Enable persist mode */
1182 if (arg)
1183 tun->flags |= TUN_PERSIST;
1184 else
1185 tun->flags &= ~TUN_PERSIST;
1187 DBG(KERN_INFO "%s: persist %s\n",
1188 tun->dev->name, arg ? "enabled" : "disabled");
1189 break;
1191 case TUNSETOWNER:
1192 /* Set owner of the device */
1193 tun->owner = (uid_t) arg;
1195 DBG(KERN_INFO "%s: owner set to %d\n", tun->dev->name, tun->owner);
1196 break;
1198 case TUNSETGROUP:
1199 /* Set group of the device */
1200 tun->group= (gid_t) arg;
1202 DBG(KERN_INFO "%s: group set to %d\n", tun->dev->name, tun->group);
1203 break;
1205 case TUNSETLINK:
1206 /* Only allow setting the type when the interface is down */
1207 if (tun->dev->flags & IFF_UP) {
1208 DBG(KERN_INFO "%s: Linktype set failed because interface is up\n",
1209 tun->dev->name);
1210 ret = -EBUSY;
1211 } else {
1212 tun->dev->type = (int) arg;
1213 DBG(KERN_INFO "%s: linktype set to %d\n", tun->dev->name, tun->dev->type);
1214 ret = 0;
1216 break;
1218 #ifdef TUN_DEBUG
1219 case TUNSETDEBUG:
1220 tun->debug = arg;
1221 break;
1222 #endif
1223 case TUNSETOFFLOAD:
1224 ret = set_offload(tun->dev, arg);
1225 break;
1227 case TUNSETTXFILTER:
1228 /* Can be set only for TAPs */
1229 ret = -EINVAL;
1230 if ((tun->flags & TUN_TYPE_MASK) != TUN_TAP_DEV)
1231 break;
1232 ret = update_filter(&tun->txflt, (void __user *)arg);
1233 break;
1235 case SIOCGIFHWADDR:
1236 /* Get hw addres */
1237 memcpy(ifr.ifr_hwaddr.sa_data, tun->dev->dev_addr, ETH_ALEN);
1238 ifr.ifr_hwaddr.sa_family = tun->dev->type;
1239 if (copy_to_user(argp, &ifr, sizeof ifr))
1240 ret = -EFAULT;
1241 break;
1243 case SIOCSIFHWADDR:
1244 /* Set hw address */
1245 DBG(KERN_DEBUG "%s: set hw address: %pM\n",
1246 tun->dev->name, ifr.ifr_hwaddr.sa_data);
1248 ret = dev_set_mac_address(tun->dev, &ifr.ifr_hwaddr);
1249 break;
1251 case TUNGETSNDBUF:
1252 sndbuf = tun->socket.sk->sk_sndbuf;
1253 if (copy_to_user(argp, &sndbuf, sizeof(sndbuf)))
1254 ret = -EFAULT;
1255 break;
1257 case TUNSETSNDBUF:
1258 if (copy_from_user(&sndbuf, argp, sizeof(sndbuf))) {
1259 ret = -EFAULT;
1260 break;
1263 tun->socket.sk->sk_sndbuf = sndbuf;
1264 break;
1266 default:
1267 ret = -EINVAL;
1268 break;
1271 unlock:
1272 rtnl_unlock();
1273 if (tun)
1274 tun_put(tun);
1275 return ret;
1278 static int tun_chr_fasync(int fd, struct file *file, int on)
1280 struct tun_struct *tun = tun_get(file);
1281 int ret;
1283 if (!tun)
1284 return -EBADFD;
1286 DBG(KERN_INFO "%s: tun_chr_fasync %d\n", tun->dev->name, on);
1288 lock_kernel();
1289 if ((ret = fasync_helper(fd, file, on, &tun->fasync)) < 0)
1290 goto out;
1292 if (on) {
1293 ret = __f_setown(file, task_pid(current), PIDTYPE_PID, 0);
1294 if (ret)
1295 goto out;
1296 tun->flags |= TUN_FASYNC;
1297 } else
1298 tun->flags &= ~TUN_FASYNC;
1299 ret = 0;
1300 out:
1301 unlock_kernel();
1302 tun_put(tun);
1303 return ret;
1306 static int tun_chr_open(struct inode *inode, struct file * file)
1308 struct tun_file *tfile;
1309 cycle_kernel_lock();
1310 DBG1(KERN_INFO "tunX: tun_chr_open\n");
1312 tfile = kmalloc(sizeof(*tfile), GFP_KERNEL);
1313 if (!tfile)
1314 return -ENOMEM;
1315 atomic_set(&tfile->count, 0);
1316 tfile->tun = NULL;
1317 tfile->net = get_net(current->nsproxy->net_ns);
1318 file->private_data = tfile;
1319 return 0;
1322 static int tun_chr_close(struct inode *inode, struct file *file)
1324 struct tun_file *tfile = file->private_data;
1325 struct tun_struct *tun;
1327 tun = __tun_get(tfile);
1328 if (tun) {
1329 struct net_device *dev = tun->dev;
1331 DBG(KERN_INFO "%s: tun_chr_close\n", dev->name);
1333 __tun_detach(tun);
1335 /* If desireable, unregister the netdevice. */
1336 if (!(tun->flags & TUN_PERSIST)) {
1337 rtnl_lock();
1338 if (dev->reg_state == NETREG_REGISTERED)
1339 unregister_netdevice(dev);
1340 rtnl_unlock();
1344 tun = tfile->tun;
1345 if (tun)
1346 sock_put(tun->socket.sk);
1348 put_net(tfile->net);
1349 kfree(tfile);
1351 return 0;
1354 static const struct file_operations tun_fops = {
1355 .owner = THIS_MODULE,
1356 .llseek = no_llseek,
1357 .read = do_sync_read,
1358 .aio_read = tun_chr_aio_read,
1359 .write = do_sync_write,
1360 .aio_write = tun_chr_aio_write,
1361 .poll = tun_chr_poll,
1362 .unlocked_ioctl = tun_chr_ioctl,
1363 .open = tun_chr_open,
1364 .release = tun_chr_close,
1365 .fasync = tun_chr_fasync
1368 static struct miscdevice tun_miscdev = {
1369 .minor = TUN_MINOR,
1370 .name = "tun",
1371 .nodename = "net/tun",
1372 .fops = &tun_fops,
1375 /* ethtool interface */
1377 static int tun_get_settings(struct net_device *dev, struct ethtool_cmd *cmd)
1379 cmd->supported = 0;
1380 cmd->advertising = 0;
1381 cmd->speed = SPEED_10;
1382 cmd->duplex = DUPLEX_FULL;
1383 cmd->port = PORT_TP;
1384 cmd->phy_address = 0;
1385 cmd->transceiver = XCVR_INTERNAL;
1386 cmd->autoneg = AUTONEG_DISABLE;
1387 cmd->maxtxpkt = 0;
1388 cmd->maxrxpkt = 0;
1389 return 0;
1392 static void tun_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
1394 struct tun_struct *tun = netdev_priv(dev);
1396 strcpy(info->driver, DRV_NAME);
1397 strcpy(info->version, DRV_VERSION);
1398 strcpy(info->fw_version, "N/A");
1400 switch (tun->flags & TUN_TYPE_MASK) {
1401 case TUN_TUN_DEV:
1402 strcpy(info->bus_info, "tun");
1403 break;
1404 case TUN_TAP_DEV:
1405 strcpy(info->bus_info, "tap");
1406 break;
1410 static u32 tun_get_msglevel(struct net_device *dev)
1412 #ifdef TUN_DEBUG
1413 struct tun_struct *tun = netdev_priv(dev);
1414 return tun->debug;
1415 #else
1416 return -EOPNOTSUPP;
1417 #endif
1420 static void tun_set_msglevel(struct net_device *dev, u32 value)
1422 #ifdef TUN_DEBUG
1423 struct tun_struct *tun = netdev_priv(dev);
1424 tun->debug = value;
1425 #endif
1428 static u32 tun_get_link(struct net_device *dev)
1430 struct tun_struct *tun = netdev_priv(dev);
1431 return !!tun->tfile;
1434 static u32 tun_get_rx_csum(struct net_device *dev)
1436 struct tun_struct *tun = netdev_priv(dev);
1437 return (tun->flags & TUN_NOCHECKSUM) == 0;
1440 static int tun_set_rx_csum(struct net_device *dev, u32 data)
1442 struct tun_struct *tun = netdev_priv(dev);
1443 if (data)
1444 tun->flags &= ~TUN_NOCHECKSUM;
1445 else
1446 tun->flags |= TUN_NOCHECKSUM;
1447 return 0;
1450 static const struct ethtool_ops tun_ethtool_ops = {
1451 .get_settings = tun_get_settings,
1452 .get_drvinfo = tun_get_drvinfo,
1453 .get_msglevel = tun_get_msglevel,
1454 .set_msglevel = tun_set_msglevel,
1455 .get_link = tun_get_link,
1456 .get_rx_csum = tun_get_rx_csum,
1457 .set_rx_csum = tun_set_rx_csum
1461 static int __init tun_init(void)
1463 int ret = 0;
1465 printk(KERN_INFO "tun: %s, %s\n", DRV_DESCRIPTION, DRV_VERSION);
1466 printk(KERN_INFO "tun: %s\n", DRV_COPYRIGHT);
1468 ret = rtnl_link_register(&tun_link_ops);
1469 if (ret) {
1470 printk(KERN_ERR "tun: Can't register link_ops\n");
1471 goto err_linkops;
1474 ret = misc_register(&tun_miscdev);
1475 if (ret) {
1476 printk(KERN_ERR "tun: Can't register misc device %d\n", TUN_MINOR);
1477 goto err_misc;
1479 return 0;
1480 err_misc:
1481 rtnl_link_unregister(&tun_link_ops);
1482 err_linkops:
1483 return ret;
1486 static void tun_cleanup(void)
1488 misc_deregister(&tun_miscdev);
1489 rtnl_link_unregister(&tun_link_ops);
1492 module_init(tun_init);
1493 module_exit(tun_cleanup);
1494 MODULE_DESCRIPTION(DRV_DESCRIPTION);
1495 MODULE_AUTHOR(DRV_COPYRIGHT);
1496 MODULE_LICENSE("GPL");
1497 MODULE_ALIAS_MISCDEV(TUN_MINOR);