x86, delay: tsc based udelay should have rdtsc_barrier
[linux-2.6/mini2440.git] / drivers / net / tun.c
blob11a0ba47b67782d920f3bf5cb36d3e24d4874b63
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 sk_buff_head readq;
108 struct net_device *dev;
109 struct fasync_struct *fasync;
111 struct tap_filter txflt;
112 struct sock *sk;
113 struct socket socket;
115 #ifdef TUN_DEBUG
116 int debug;
117 #endif
120 struct tun_sock {
121 struct sock sk;
122 struct tun_struct *tun;
125 static inline struct tun_sock *tun_sk(struct sock *sk)
127 return container_of(sk, struct tun_sock, sk);
130 static int tun_attach(struct tun_struct *tun, struct file *file)
132 struct tun_file *tfile = file->private_data;
133 const struct cred *cred = current_cred();
134 int err;
136 ASSERT_RTNL();
138 /* Check permissions */
139 if (((tun->owner != -1 && cred->euid != tun->owner) ||
140 (tun->group != -1 && !in_egroup_p(tun->group))) &&
141 !capable(CAP_NET_ADMIN))
142 return -EPERM;
144 netif_tx_lock_bh(tun->dev);
146 err = -EINVAL;
147 if (tfile->tun)
148 goto out;
150 err = -EBUSY;
151 if (tun->tfile)
152 goto out;
154 err = 0;
155 tfile->tun = tun;
156 tun->tfile = tfile;
157 dev_hold(tun->dev);
158 sock_hold(tun->sk);
159 atomic_inc(&tfile->count);
161 out:
162 netif_tx_unlock_bh(tun->dev);
163 return err;
166 static void __tun_detach(struct tun_struct *tun)
168 /* Detach from net device */
169 netif_tx_lock_bh(tun->dev);
170 tun->tfile = NULL;
171 netif_tx_unlock_bh(tun->dev);
173 /* Drop read queue */
174 skb_queue_purge(&tun->readq);
176 /* Drop the extra count on the net device */
177 dev_put(tun->dev);
180 static void tun_detach(struct tun_struct *tun)
182 rtnl_lock();
183 __tun_detach(tun);
184 rtnl_unlock();
187 static struct tun_struct *__tun_get(struct tun_file *tfile)
189 struct tun_struct *tun = NULL;
191 if (atomic_inc_not_zero(&tfile->count))
192 tun = tfile->tun;
194 return tun;
197 static struct tun_struct *tun_get(struct file *file)
199 return __tun_get(file->private_data);
202 static void tun_put(struct tun_struct *tun)
204 struct tun_file *tfile = tun->tfile;
206 if (atomic_dec_and_test(&tfile->count))
207 tun_detach(tfile->tun);
210 /* TAP filterting */
211 static void addr_hash_set(u32 *mask, const u8 *addr)
213 int n = ether_crc(ETH_ALEN, addr) >> 26;
214 mask[n >> 5] |= (1 << (n & 31));
217 static unsigned int addr_hash_test(const u32 *mask, const u8 *addr)
219 int n = ether_crc(ETH_ALEN, addr) >> 26;
220 return mask[n >> 5] & (1 << (n & 31));
223 static int update_filter(struct tap_filter *filter, void __user *arg)
225 struct { u8 u[ETH_ALEN]; } *addr;
226 struct tun_filter uf;
227 int err, alen, n, nexact;
229 if (copy_from_user(&uf, arg, sizeof(uf)))
230 return -EFAULT;
232 if (!uf.count) {
233 /* Disabled */
234 filter->count = 0;
235 return 0;
238 alen = ETH_ALEN * uf.count;
239 addr = kmalloc(alen, GFP_KERNEL);
240 if (!addr)
241 return -ENOMEM;
243 if (copy_from_user(addr, arg + sizeof(uf), alen)) {
244 err = -EFAULT;
245 goto done;
248 /* The filter is updated without holding any locks. Which is
249 * perfectly safe. We disable it first and in the worst
250 * case we'll accept a few undesired packets. */
251 filter->count = 0;
252 wmb();
254 /* Use first set of addresses as an exact filter */
255 for (n = 0; n < uf.count && n < FLT_EXACT_COUNT; n++)
256 memcpy(filter->addr[n], addr[n].u, ETH_ALEN);
258 nexact = n;
260 /* Remaining multicast addresses are hashed,
261 * unicast will leave the filter disabled. */
262 memset(filter->mask, 0, sizeof(filter->mask));
263 for (; n < uf.count; n++) {
264 if (!is_multicast_ether_addr(addr[n].u)) {
265 err = 0; /* no filter */
266 goto done;
268 addr_hash_set(filter->mask, addr[n].u);
271 /* For ALLMULTI just set the mask to all ones.
272 * This overrides the mask populated above. */
273 if ((uf.flags & TUN_FLT_ALLMULTI))
274 memset(filter->mask, ~0, sizeof(filter->mask));
276 /* Now enable the filter */
277 wmb();
278 filter->count = nexact;
280 /* Return the number of exact filters */
281 err = nexact;
283 done:
284 kfree(addr);
285 return err;
288 /* Returns: 0 - drop, !=0 - accept */
289 static int run_filter(struct tap_filter *filter, const struct sk_buff *skb)
291 /* Cannot use eth_hdr(skb) here because skb_mac_hdr() is incorrect
292 * at this point. */
293 struct ethhdr *eh = (struct ethhdr *) skb->data;
294 int i;
296 /* Exact match */
297 for (i = 0; i < filter->count; i++)
298 if (!compare_ether_addr(eh->h_dest, filter->addr[i]))
299 return 1;
301 /* Inexact match (multicast only) */
302 if (is_multicast_ether_addr(eh->h_dest))
303 return addr_hash_test(filter->mask, eh->h_dest);
305 return 0;
309 * Checks whether the packet is accepted or not.
310 * Returns: 0 - drop, !=0 - accept
312 static int check_filter(struct tap_filter *filter, const struct sk_buff *skb)
314 if (!filter->count)
315 return 1;
317 return run_filter(filter, skb);
320 /* Network device part of the driver */
322 static const struct ethtool_ops tun_ethtool_ops;
324 /* Net device detach from fd. */
325 static void tun_net_uninit(struct net_device *dev)
327 struct tun_struct *tun = netdev_priv(dev);
328 struct tun_file *tfile = tun->tfile;
330 /* Inform the methods they need to stop using the dev.
332 if (tfile) {
333 wake_up_all(&tun->socket.wait);
334 if (atomic_dec_and_test(&tfile->count))
335 __tun_detach(tun);
339 static void tun_free_netdev(struct net_device *dev)
341 struct tun_struct *tun = netdev_priv(dev);
343 sock_put(tun->sk);
346 /* Net device open. */
347 static int tun_net_open(struct net_device *dev)
349 netif_start_queue(dev);
350 return 0;
353 /* Net device close. */
354 static int tun_net_close(struct net_device *dev)
356 netif_stop_queue(dev);
357 return 0;
360 /* Net device start xmit */
361 static int tun_net_xmit(struct sk_buff *skb, struct net_device *dev)
363 struct tun_struct *tun = netdev_priv(dev);
365 DBG(KERN_INFO "%s: tun_net_xmit %d\n", tun->dev->name, skb->len);
367 /* Drop packet if interface is not attached */
368 if (!tun->tfile)
369 goto drop;
371 /* Drop if the filter does not like it.
372 * This is a noop if the filter is disabled.
373 * Filter can be enabled only for the TAP devices. */
374 if (!check_filter(&tun->txflt, skb))
375 goto drop;
377 if (skb_queue_len(&tun->readq) >= dev->tx_queue_len) {
378 if (!(tun->flags & TUN_ONE_QUEUE)) {
379 /* Normal queueing mode. */
380 /* Packet scheduler handles dropping of further packets. */
381 netif_stop_queue(dev);
383 /* We won't see all dropped packets individually, so overrun
384 * error is more appropriate. */
385 dev->stats.tx_fifo_errors++;
386 } else {
387 /* Single queue mode.
388 * Driver handles dropping of all packets itself. */
389 goto drop;
393 /* Enqueue packet */
394 skb_queue_tail(&tun->readq, skb);
395 dev->trans_start = jiffies;
397 /* Notify and wake up reader process */
398 if (tun->flags & TUN_FASYNC)
399 kill_fasync(&tun->fasync, SIGIO, POLL_IN);
400 wake_up_interruptible(&tun->socket.wait);
401 return 0;
403 drop:
404 dev->stats.tx_dropped++;
405 kfree_skb(skb);
406 return 0;
409 static void tun_net_mclist(struct net_device *dev)
412 * This callback is supposed to deal with mc filter in
413 * _rx_ path and has nothing to do with the _tx_ path.
414 * In rx path we always accept everything userspace gives us.
416 return;
419 #define MIN_MTU 68
420 #define MAX_MTU 65535
422 static int
423 tun_net_change_mtu(struct net_device *dev, int new_mtu)
425 if (new_mtu < MIN_MTU || new_mtu + dev->hard_header_len > MAX_MTU)
426 return -EINVAL;
427 dev->mtu = new_mtu;
428 return 0;
431 static const struct net_device_ops tun_netdev_ops = {
432 .ndo_uninit = tun_net_uninit,
433 .ndo_open = tun_net_open,
434 .ndo_stop = tun_net_close,
435 .ndo_start_xmit = tun_net_xmit,
436 .ndo_change_mtu = tun_net_change_mtu,
439 static const struct net_device_ops tap_netdev_ops = {
440 .ndo_uninit = tun_net_uninit,
441 .ndo_open = tun_net_open,
442 .ndo_stop = tun_net_close,
443 .ndo_start_xmit = tun_net_xmit,
444 .ndo_change_mtu = tun_net_change_mtu,
445 .ndo_set_multicast_list = tun_net_mclist,
446 .ndo_set_mac_address = eth_mac_addr,
447 .ndo_validate_addr = eth_validate_addr,
450 /* Initialize net device. */
451 static void tun_net_init(struct net_device *dev)
453 struct tun_struct *tun = netdev_priv(dev);
455 switch (tun->flags & TUN_TYPE_MASK) {
456 case TUN_TUN_DEV:
457 dev->netdev_ops = &tun_netdev_ops;
459 /* Point-to-Point TUN Device */
460 dev->hard_header_len = 0;
461 dev->addr_len = 0;
462 dev->mtu = 1500;
464 /* Zero header length */
465 dev->type = ARPHRD_NONE;
466 dev->flags = IFF_POINTOPOINT | IFF_NOARP | IFF_MULTICAST;
467 dev->tx_queue_len = TUN_READQ_SIZE; /* We prefer our own queue length */
468 break;
470 case TUN_TAP_DEV:
471 dev->netdev_ops = &tap_netdev_ops;
472 /* Ethernet TAP Device */
473 ether_setup(dev);
475 random_ether_addr(dev->dev_addr);
477 dev->tx_queue_len = TUN_READQ_SIZE; /* We prefer our own queue length */
478 break;
482 /* Character device part */
484 /* Poll */
485 static unsigned int tun_chr_poll(struct file *file, poll_table * wait)
487 struct tun_file *tfile = file->private_data;
488 struct tun_struct *tun = __tun_get(tfile);
489 struct sock *sk = tun->sk;
490 unsigned int mask = 0;
492 if (!tun)
493 return POLLERR;
495 DBG(KERN_INFO "%s: tun_chr_poll\n", tun->dev->name);
497 poll_wait(file, &tun->socket.wait, wait);
499 if (!skb_queue_empty(&tun->readq))
500 mask |= POLLIN | POLLRDNORM;
502 if (sock_writeable(sk) ||
503 (!test_and_set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags) &&
504 sock_writeable(sk)))
505 mask |= POLLOUT | POLLWRNORM;
507 if (tun->dev->reg_state != NETREG_REGISTERED)
508 mask = POLLERR;
510 tun_put(tun);
511 return mask;
514 /* prepad is the amount to reserve at front. len is length after that.
515 * linear is a hint as to how much to copy (usually headers). */
516 static inline struct sk_buff *tun_alloc_skb(struct tun_struct *tun,
517 size_t prepad, size_t len,
518 size_t linear, int noblock)
520 struct sock *sk = tun->sk;
521 struct sk_buff *skb;
522 int err;
524 /* Under a page? Don't bother with paged skb. */
525 if (prepad + len < PAGE_SIZE || !linear)
526 linear = len;
528 skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
529 &err);
530 if (!skb)
531 return ERR_PTR(err);
533 skb_reserve(skb, prepad);
534 skb_put(skb, linear);
535 skb->data_len = len - linear;
536 skb->len += len - linear;
538 return skb;
541 /* Get packet from user space buffer */
542 static __inline__ ssize_t tun_get_user(struct tun_struct *tun,
543 const struct iovec *iv, size_t count,
544 int noblock)
546 struct tun_pi pi = { 0, cpu_to_be16(ETH_P_IP) };
547 struct sk_buff *skb;
548 size_t len = count, align = 0;
549 struct virtio_net_hdr gso = { 0 };
550 int offset = 0;
552 if (!(tun->flags & TUN_NO_PI)) {
553 if ((len -= sizeof(pi)) > count)
554 return -EINVAL;
556 if (memcpy_fromiovecend((void *)&pi, iv, 0, sizeof(pi)))
557 return -EFAULT;
558 offset += sizeof(pi);
561 if (tun->flags & TUN_VNET_HDR) {
562 if ((len -= sizeof(gso)) > count)
563 return -EINVAL;
565 if (memcpy_fromiovecend((void *)&gso, iv, offset, sizeof(gso)))
566 return -EFAULT;
568 if ((gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
569 gso.csum_start + gso.csum_offset + 2 > gso.hdr_len)
570 gso.hdr_len = gso.csum_start + gso.csum_offset + 2;
572 if (gso.hdr_len > len)
573 return -EINVAL;
574 offset += sizeof(gso);
577 if ((tun->flags & TUN_TYPE_MASK) == TUN_TAP_DEV) {
578 align = NET_IP_ALIGN;
579 if (unlikely(len < ETH_HLEN ||
580 (gso.hdr_len && gso.hdr_len < ETH_HLEN)))
581 return -EINVAL;
584 skb = tun_alloc_skb(tun, align, len, gso.hdr_len, noblock);
585 if (IS_ERR(skb)) {
586 if (PTR_ERR(skb) != -EAGAIN)
587 tun->dev->stats.rx_dropped++;
588 return PTR_ERR(skb);
591 if (skb_copy_datagram_from_iovec(skb, 0, iv, offset, len)) {
592 tun->dev->stats.rx_dropped++;
593 kfree_skb(skb);
594 return -EFAULT;
597 if (gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
598 if (!skb_partial_csum_set(skb, gso.csum_start,
599 gso.csum_offset)) {
600 tun->dev->stats.rx_frame_errors++;
601 kfree_skb(skb);
602 return -EINVAL;
604 } else if (tun->flags & TUN_NOCHECKSUM)
605 skb->ip_summed = CHECKSUM_UNNECESSARY;
607 switch (tun->flags & TUN_TYPE_MASK) {
608 case TUN_TUN_DEV:
609 if (tun->flags & TUN_NO_PI) {
610 switch (skb->data[0] & 0xf0) {
611 case 0x40:
612 pi.proto = htons(ETH_P_IP);
613 break;
614 case 0x60:
615 pi.proto = htons(ETH_P_IPV6);
616 break;
617 default:
618 tun->dev->stats.rx_dropped++;
619 kfree_skb(skb);
620 return -EINVAL;
624 skb_reset_mac_header(skb);
625 skb->protocol = pi.proto;
626 skb->dev = tun->dev;
627 break;
628 case TUN_TAP_DEV:
629 skb->protocol = eth_type_trans(skb, tun->dev);
630 break;
633 if (gso.gso_type != VIRTIO_NET_HDR_GSO_NONE) {
634 pr_debug("GSO!\n");
635 switch (gso.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
636 case VIRTIO_NET_HDR_GSO_TCPV4:
637 skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
638 break;
639 case VIRTIO_NET_HDR_GSO_TCPV6:
640 skb_shinfo(skb)->gso_type = SKB_GSO_TCPV6;
641 break;
642 default:
643 tun->dev->stats.rx_frame_errors++;
644 kfree_skb(skb);
645 return -EINVAL;
648 if (gso.gso_type & VIRTIO_NET_HDR_GSO_ECN)
649 skb_shinfo(skb)->gso_type |= SKB_GSO_TCP_ECN;
651 skb_shinfo(skb)->gso_size = gso.gso_size;
652 if (skb_shinfo(skb)->gso_size == 0) {
653 tun->dev->stats.rx_frame_errors++;
654 kfree_skb(skb);
655 return -EINVAL;
658 /* Header must be checked, and gso_segs computed. */
659 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
660 skb_shinfo(skb)->gso_segs = 0;
663 netif_rx_ni(skb);
665 tun->dev->stats.rx_packets++;
666 tun->dev->stats.rx_bytes += len;
668 return count;
671 static ssize_t tun_chr_aio_write(struct kiocb *iocb, const struct iovec *iv,
672 unsigned long count, loff_t pos)
674 struct file *file = iocb->ki_filp;
675 struct tun_struct *tun = tun_get(file);
676 ssize_t result;
678 if (!tun)
679 return -EBADFD;
681 DBG(KERN_INFO "%s: tun_chr_write %ld\n", tun->dev->name, count);
683 result = tun_get_user(tun, iv, iov_length(iv, count),
684 file->f_flags & O_NONBLOCK);
686 tun_put(tun);
687 return result;
690 /* Put packet to the user space buffer */
691 static __inline__ ssize_t tun_put_user(struct tun_struct *tun,
692 struct sk_buff *skb,
693 const struct iovec *iv, int len)
695 struct tun_pi pi = { 0, skb->protocol };
696 ssize_t total = 0;
698 if (!(tun->flags & TUN_NO_PI)) {
699 if ((len -= sizeof(pi)) < 0)
700 return -EINVAL;
702 if (len < skb->len) {
703 /* Packet will be striped */
704 pi.flags |= TUN_PKT_STRIP;
707 if (memcpy_toiovecend(iv, (void *) &pi, 0, sizeof(pi)))
708 return -EFAULT;
709 total += sizeof(pi);
712 if (tun->flags & TUN_VNET_HDR) {
713 struct virtio_net_hdr gso = { 0 }; /* no info leak */
714 if ((len -= sizeof(gso)) < 0)
715 return -EINVAL;
717 if (skb_is_gso(skb)) {
718 struct skb_shared_info *sinfo = skb_shinfo(skb);
720 /* This is a hint as to how much should be linear. */
721 gso.hdr_len = skb_headlen(skb);
722 gso.gso_size = sinfo->gso_size;
723 if (sinfo->gso_type & SKB_GSO_TCPV4)
724 gso.gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
725 else if (sinfo->gso_type & SKB_GSO_TCPV6)
726 gso.gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
727 else
728 BUG();
729 if (sinfo->gso_type & SKB_GSO_TCP_ECN)
730 gso.gso_type |= VIRTIO_NET_HDR_GSO_ECN;
731 } else
732 gso.gso_type = VIRTIO_NET_HDR_GSO_NONE;
734 if (skb->ip_summed == CHECKSUM_PARTIAL) {
735 gso.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
736 gso.csum_start = skb->csum_start - skb_headroom(skb);
737 gso.csum_offset = skb->csum_offset;
738 } /* else everything is zero */
740 if (unlikely(memcpy_toiovecend(iv, (void *)&gso, total,
741 sizeof(gso))))
742 return -EFAULT;
743 total += sizeof(gso);
746 len = min_t(int, skb->len, len);
748 skb_copy_datagram_const_iovec(skb, 0, iv, total, len);
749 total += len;
751 tun->dev->stats.tx_packets++;
752 tun->dev->stats.tx_bytes += len;
754 return total;
757 static ssize_t tun_chr_aio_read(struct kiocb *iocb, const struct iovec *iv,
758 unsigned long count, loff_t pos)
760 struct file *file = iocb->ki_filp;
761 struct tun_file *tfile = file->private_data;
762 struct tun_struct *tun = __tun_get(tfile);
763 DECLARE_WAITQUEUE(wait, current);
764 struct sk_buff *skb;
765 ssize_t len, ret = 0;
767 if (!tun)
768 return -EBADFD;
770 DBG(KERN_INFO "%s: tun_chr_read\n", tun->dev->name);
772 len = iov_length(iv, count);
773 if (len < 0) {
774 ret = -EINVAL;
775 goto out;
778 add_wait_queue(&tun->socket.wait, &wait);
779 while (len) {
780 current->state = TASK_INTERRUPTIBLE;
782 /* Read frames from the queue */
783 if (!(skb=skb_dequeue(&tun->readq))) {
784 if (file->f_flags & O_NONBLOCK) {
785 ret = -EAGAIN;
786 break;
788 if (signal_pending(current)) {
789 ret = -ERESTARTSYS;
790 break;
792 if (tun->dev->reg_state != NETREG_REGISTERED) {
793 ret = -EIO;
794 break;
797 /* Nothing to read, let's sleep */
798 schedule();
799 continue;
801 netif_wake_queue(tun->dev);
803 ret = tun_put_user(tun, skb, iv, len);
804 kfree_skb(skb);
805 break;
808 current->state = TASK_RUNNING;
809 remove_wait_queue(&tun->socket.wait, &wait);
811 out:
812 tun_put(tun);
813 return ret;
816 static void tun_setup(struct net_device *dev)
818 struct tun_struct *tun = netdev_priv(dev);
820 skb_queue_head_init(&tun->readq);
822 tun->owner = -1;
823 tun->group = -1;
825 dev->ethtool_ops = &tun_ethtool_ops;
826 dev->destructor = tun_free_netdev;
829 /* Trivial set of netlink ops to allow deleting tun or tap
830 * device with netlink.
832 static int tun_validate(struct nlattr *tb[], struct nlattr *data[])
834 return -EINVAL;
837 static struct rtnl_link_ops tun_link_ops __read_mostly = {
838 .kind = DRV_NAME,
839 .priv_size = sizeof(struct tun_struct),
840 .setup = tun_setup,
841 .validate = tun_validate,
844 static void tun_sock_write_space(struct sock *sk)
846 struct tun_struct *tun;
848 if (!sock_writeable(sk))
849 return;
851 if (!test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags))
852 return;
854 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
855 wake_up_interruptible_sync(sk->sk_sleep);
857 tun = container_of(sk, struct tun_sock, sk)->tun;
858 kill_fasync(&tun->fasync, SIGIO, POLL_OUT);
861 static void tun_sock_destruct(struct sock *sk)
863 free_netdev(container_of(sk, struct tun_sock, sk)->tun->dev);
866 static struct proto tun_proto = {
867 .name = "tun",
868 .owner = THIS_MODULE,
869 .obj_size = sizeof(struct tun_sock),
872 static int tun_flags(struct tun_struct *tun)
874 int flags = 0;
876 if (tun->flags & TUN_TUN_DEV)
877 flags |= IFF_TUN;
878 else
879 flags |= IFF_TAP;
881 if (tun->flags & TUN_NO_PI)
882 flags |= IFF_NO_PI;
884 if (tun->flags & TUN_ONE_QUEUE)
885 flags |= IFF_ONE_QUEUE;
887 if (tun->flags & TUN_VNET_HDR)
888 flags |= IFF_VNET_HDR;
890 return flags;
893 static ssize_t tun_show_flags(struct device *dev, struct device_attribute *attr,
894 char *buf)
896 struct tun_struct *tun = netdev_priv(to_net_dev(dev));
897 return sprintf(buf, "0x%x\n", tun_flags(tun));
900 static ssize_t tun_show_owner(struct device *dev, struct device_attribute *attr,
901 char *buf)
903 struct tun_struct *tun = netdev_priv(to_net_dev(dev));
904 return sprintf(buf, "%d\n", tun->owner);
907 static ssize_t tun_show_group(struct device *dev, struct device_attribute *attr,
908 char *buf)
910 struct tun_struct *tun = netdev_priv(to_net_dev(dev));
911 return sprintf(buf, "%d\n", tun->group);
914 static DEVICE_ATTR(tun_flags, 0444, tun_show_flags, NULL);
915 static DEVICE_ATTR(owner, 0444, tun_show_owner, NULL);
916 static DEVICE_ATTR(group, 0444, tun_show_group, NULL);
918 static int tun_set_iff(struct net *net, struct file *file, struct ifreq *ifr)
920 struct sock *sk;
921 struct tun_struct *tun;
922 struct net_device *dev;
923 int err;
925 dev = __dev_get_by_name(net, ifr->ifr_name);
926 if (dev) {
927 if (ifr->ifr_flags & IFF_TUN_EXCL)
928 return -EBUSY;
929 if ((ifr->ifr_flags & IFF_TUN) && dev->netdev_ops == &tun_netdev_ops)
930 tun = netdev_priv(dev);
931 else if ((ifr->ifr_flags & IFF_TAP) && dev->netdev_ops == &tap_netdev_ops)
932 tun = netdev_priv(dev);
933 else
934 return -EINVAL;
936 err = tun_attach(tun, file);
937 if (err < 0)
938 return err;
940 else {
941 char *name;
942 unsigned long flags = 0;
944 err = -EINVAL;
946 if (!capable(CAP_NET_ADMIN))
947 return -EPERM;
949 /* Set dev type */
950 if (ifr->ifr_flags & IFF_TUN) {
951 /* TUN device */
952 flags |= TUN_TUN_DEV;
953 name = "tun%d";
954 } else if (ifr->ifr_flags & IFF_TAP) {
955 /* TAP device */
956 flags |= TUN_TAP_DEV;
957 name = "tap%d";
958 } else
959 goto failed;
961 if (*ifr->ifr_name)
962 name = ifr->ifr_name;
964 dev = alloc_netdev(sizeof(struct tun_struct), name,
965 tun_setup);
966 if (!dev)
967 return -ENOMEM;
969 dev_net_set(dev, net);
970 dev->rtnl_link_ops = &tun_link_ops;
972 tun = netdev_priv(dev);
973 tun->dev = dev;
974 tun->flags = flags;
975 tun->txflt.count = 0;
977 err = -ENOMEM;
978 sk = sk_alloc(net, AF_UNSPEC, GFP_KERNEL, &tun_proto);
979 if (!sk)
980 goto err_free_dev;
982 init_waitqueue_head(&tun->socket.wait);
983 sock_init_data(&tun->socket, sk);
984 sk->sk_write_space = tun_sock_write_space;
985 sk->sk_sndbuf = INT_MAX;
987 tun->sk = sk;
988 container_of(sk, struct tun_sock, sk)->tun = tun;
990 tun_net_init(dev);
992 if (strchr(dev->name, '%')) {
993 err = dev_alloc_name(dev, dev->name);
994 if (err < 0)
995 goto err_free_sk;
998 err = -EINVAL;
999 err = register_netdevice(tun->dev);
1000 if (err < 0)
1001 goto err_free_sk;
1003 if (device_create_file(&tun->dev->dev, &dev_attr_tun_flags) ||
1004 device_create_file(&tun->dev->dev, &dev_attr_owner) ||
1005 device_create_file(&tun->dev->dev, &dev_attr_group))
1006 printk(KERN_ERR "Failed to create tun sysfs files\n");
1008 sk->sk_destruct = tun_sock_destruct;
1010 err = tun_attach(tun, file);
1011 if (err < 0)
1012 goto failed;
1015 DBG(KERN_INFO "%s: tun_set_iff\n", tun->dev->name);
1017 if (ifr->ifr_flags & IFF_NO_PI)
1018 tun->flags |= TUN_NO_PI;
1019 else
1020 tun->flags &= ~TUN_NO_PI;
1022 if (ifr->ifr_flags & IFF_ONE_QUEUE)
1023 tun->flags |= TUN_ONE_QUEUE;
1024 else
1025 tun->flags &= ~TUN_ONE_QUEUE;
1027 if (ifr->ifr_flags & IFF_VNET_HDR)
1028 tun->flags |= TUN_VNET_HDR;
1029 else
1030 tun->flags &= ~TUN_VNET_HDR;
1032 /* Make sure persistent devices do not get stuck in
1033 * xoff state.
1035 if (netif_running(tun->dev))
1036 netif_wake_queue(tun->dev);
1038 strcpy(ifr->ifr_name, tun->dev->name);
1039 return 0;
1041 err_free_sk:
1042 sock_put(sk);
1043 err_free_dev:
1044 free_netdev(dev);
1045 failed:
1046 return err;
1049 static int tun_get_iff(struct net *net, struct file *file, struct ifreq *ifr)
1051 struct tun_struct *tun = tun_get(file);
1053 if (!tun)
1054 return -EBADFD;
1056 DBG(KERN_INFO "%s: tun_get_iff\n", tun->dev->name);
1058 strcpy(ifr->ifr_name, tun->dev->name);
1060 ifr->ifr_flags = tun_flags(tun);
1062 tun_put(tun);
1063 return 0;
1066 /* This is like a cut-down ethtool ops, except done via tun fd so no
1067 * privs required. */
1068 static int set_offload(struct net_device *dev, unsigned long arg)
1070 unsigned int old_features, features;
1072 old_features = dev->features;
1073 /* Unset features, set them as we chew on the arg. */
1074 features = (old_features & ~(NETIF_F_HW_CSUM|NETIF_F_SG|NETIF_F_FRAGLIST
1075 |NETIF_F_TSO_ECN|NETIF_F_TSO|NETIF_F_TSO6));
1077 if (arg & TUN_F_CSUM) {
1078 features |= NETIF_F_HW_CSUM|NETIF_F_SG|NETIF_F_FRAGLIST;
1079 arg &= ~TUN_F_CSUM;
1081 if (arg & (TUN_F_TSO4|TUN_F_TSO6)) {
1082 if (arg & TUN_F_TSO_ECN) {
1083 features |= NETIF_F_TSO_ECN;
1084 arg &= ~TUN_F_TSO_ECN;
1086 if (arg & TUN_F_TSO4)
1087 features |= NETIF_F_TSO;
1088 if (arg & TUN_F_TSO6)
1089 features |= NETIF_F_TSO6;
1090 arg &= ~(TUN_F_TSO4|TUN_F_TSO6);
1094 /* This gives the user a way to test for new features in future by
1095 * trying to set them. */
1096 if (arg)
1097 return -EINVAL;
1099 dev->features = features;
1100 if (old_features != dev->features)
1101 netdev_features_change(dev);
1103 return 0;
1106 static int tun_chr_ioctl(struct inode *inode, struct file *file,
1107 unsigned int cmd, unsigned long arg)
1109 struct tun_file *tfile = file->private_data;
1110 struct tun_struct *tun;
1111 void __user* argp = (void __user*)arg;
1112 struct ifreq ifr;
1113 int sndbuf;
1114 int ret;
1116 if (cmd == TUNSETIFF || _IOC_TYPE(cmd) == 0x89)
1117 if (copy_from_user(&ifr, argp, sizeof ifr))
1118 return -EFAULT;
1120 if (cmd == TUNGETFEATURES) {
1121 /* Currently this just means: "what IFF flags are valid?".
1122 * This is needed because we never checked for invalid flags on
1123 * TUNSETIFF. */
1124 return put_user(IFF_TUN | IFF_TAP | IFF_NO_PI | IFF_ONE_QUEUE |
1125 IFF_VNET_HDR,
1126 (unsigned int __user*)argp);
1129 tun = __tun_get(tfile);
1130 if (cmd == TUNSETIFF && !tun) {
1131 int err;
1133 ifr.ifr_name[IFNAMSIZ-1] = '\0';
1135 rtnl_lock();
1136 err = tun_set_iff(tfile->net, file, &ifr);
1137 rtnl_unlock();
1139 if (err)
1140 return err;
1142 if (copy_to_user(argp, &ifr, sizeof(ifr)))
1143 return -EFAULT;
1144 return 0;
1148 if (!tun)
1149 return -EBADFD;
1151 DBG(KERN_INFO "%s: tun_chr_ioctl cmd %d\n", tun->dev->name, cmd);
1153 ret = 0;
1154 switch (cmd) {
1155 case TUNGETIFF:
1156 ret = tun_get_iff(current->nsproxy->net_ns, file, &ifr);
1157 if (ret)
1158 break;
1160 if (copy_to_user(argp, &ifr, sizeof(ifr)))
1161 ret = -EFAULT;
1162 break;
1164 case TUNSETNOCSUM:
1165 /* Disable/Enable checksum */
1166 if (arg)
1167 tun->flags |= TUN_NOCHECKSUM;
1168 else
1169 tun->flags &= ~TUN_NOCHECKSUM;
1171 DBG(KERN_INFO "%s: checksum %s\n",
1172 tun->dev->name, arg ? "disabled" : "enabled");
1173 break;
1175 case TUNSETPERSIST:
1176 /* Disable/Enable persist mode */
1177 if (arg)
1178 tun->flags |= TUN_PERSIST;
1179 else
1180 tun->flags &= ~TUN_PERSIST;
1182 DBG(KERN_INFO "%s: persist %s\n",
1183 tun->dev->name, arg ? "enabled" : "disabled");
1184 break;
1186 case TUNSETOWNER:
1187 /* Set owner of the device */
1188 tun->owner = (uid_t) arg;
1190 DBG(KERN_INFO "%s: owner set to %d\n", tun->dev->name, tun->owner);
1191 break;
1193 case TUNSETGROUP:
1194 /* Set group of the device */
1195 tun->group= (gid_t) arg;
1197 DBG(KERN_INFO "%s: group set to %d\n", tun->dev->name, tun->group);
1198 break;
1200 case TUNSETLINK:
1201 /* Only allow setting the type when the interface is down */
1202 rtnl_lock();
1203 if (tun->dev->flags & IFF_UP) {
1204 DBG(KERN_INFO "%s: Linktype set failed because interface is up\n",
1205 tun->dev->name);
1206 ret = -EBUSY;
1207 } else {
1208 tun->dev->type = (int) arg;
1209 DBG(KERN_INFO "%s: linktype set to %d\n", tun->dev->name, tun->dev->type);
1210 ret = 0;
1212 rtnl_unlock();
1213 break;
1215 #ifdef TUN_DEBUG
1216 case TUNSETDEBUG:
1217 tun->debug = arg;
1218 break;
1219 #endif
1220 case TUNSETOFFLOAD:
1221 rtnl_lock();
1222 ret = set_offload(tun->dev, arg);
1223 rtnl_unlock();
1224 break;
1226 case TUNSETTXFILTER:
1227 /* Can be set only for TAPs */
1228 ret = -EINVAL;
1229 if ((tun->flags & TUN_TYPE_MASK) != TUN_TAP_DEV)
1230 break;
1231 rtnl_lock();
1232 ret = update_filter(&tun->txflt, (void __user *)arg);
1233 rtnl_unlock();
1234 break;
1236 case SIOCGIFHWADDR:
1237 /* Get hw addres */
1238 memcpy(ifr.ifr_hwaddr.sa_data, tun->dev->dev_addr, ETH_ALEN);
1239 ifr.ifr_hwaddr.sa_family = tun->dev->type;
1240 if (copy_to_user(argp, &ifr, sizeof ifr))
1241 ret = -EFAULT;
1242 break;
1244 case SIOCSIFHWADDR:
1245 /* Set hw address */
1246 DBG(KERN_DEBUG "%s: set hw address: %pM\n",
1247 tun->dev->name, ifr.ifr_hwaddr.sa_data);
1249 rtnl_lock();
1250 ret = dev_set_mac_address(tun->dev, &ifr.ifr_hwaddr);
1251 rtnl_unlock();
1252 break;
1254 case TUNGETSNDBUF:
1255 sndbuf = tun->sk->sk_sndbuf;
1256 if (copy_to_user(argp, &sndbuf, sizeof(sndbuf)))
1257 ret = -EFAULT;
1258 break;
1260 case TUNSETSNDBUF:
1261 if (copy_from_user(&sndbuf, argp, sizeof(sndbuf))) {
1262 ret = -EFAULT;
1263 break;
1266 tun->sk->sk_sndbuf = sndbuf;
1267 break;
1269 default:
1270 ret = -EINVAL;
1271 break;
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;
1328 rtnl_lock();
1329 tun = __tun_get(tfile);
1330 if (tun) {
1331 DBG(KERN_INFO "%s: tun_chr_close\n", tun->dev->name);
1333 __tun_detach(tun);
1335 /* If desireable, unregister the netdevice. */
1336 if (!(tun->flags & TUN_PERSIST))
1337 unregister_netdevice(tun->dev);
1340 rtnl_unlock();
1342 tun = tfile->tun;
1343 if (tun)
1344 sock_put(tun->sk);
1346 put_net(tfile->net);
1347 kfree(tfile);
1349 return 0;
1352 static const struct file_operations tun_fops = {
1353 .owner = THIS_MODULE,
1354 .llseek = no_llseek,
1355 .read = do_sync_read,
1356 .aio_read = tun_chr_aio_read,
1357 .write = do_sync_write,
1358 .aio_write = tun_chr_aio_write,
1359 .poll = tun_chr_poll,
1360 .ioctl = tun_chr_ioctl,
1361 .open = tun_chr_open,
1362 .release = tun_chr_close,
1363 .fasync = tun_chr_fasync
1366 static struct miscdevice tun_miscdev = {
1367 .minor = TUN_MINOR,
1368 .name = "tun",
1369 .devnode = "net/tun",
1370 .fops = &tun_fops,
1373 /* ethtool interface */
1375 static int tun_get_settings(struct net_device *dev, struct ethtool_cmd *cmd)
1377 cmd->supported = 0;
1378 cmd->advertising = 0;
1379 cmd->speed = SPEED_10;
1380 cmd->duplex = DUPLEX_FULL;
1381 cmd->port = PORT_TP;
1382 cmd->phy_address = 0;
1383 cmd->transceiver = XCVR_INTERNAL;
1384 cmd->autoneg = AUTONEG_DISABLE;
1385 cmd->maxtxpkt = 0;
1386 cmd->maxrxpkt = 0;
1387 return 0;
1390 static void tun_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
1392 struct tun_struct *tun = netdev_priv(dev);
1394 strcpy(info->driver, DRV_NAME);
1395 strcpy(info->version, DRV_VERSION);
1396 strcpy(info->fw_version, "N/A");
1398 switch (tun->flags & TUN_TYPE_MASK) {
1399 case TUN_TUN_DEV:
1400 strcpy(info->bus_info, "tun");
1401 break;
1402 case TUN_TAP_DEV:
1403 strcpy(info->bus_info, "tap");
1404 break;
1408 static u32 tun_get_msglevel(struct net_device *dev)
1410 #ifdef TUN_DEBUG
1411 struct tun_struct *tun = netdev_priv(dev);
1412 return tun->debug;
1413 #else
1414 return -EOPNOTSUPP;
1415 #endif
1418 static void tun_set_msglevel(struct net_device *dev, u32 value)
1420 #ifdef TUN_DEBUG
1421 struct tun_struct *tun = netdev_priv(dev);
1422 tun->debug = value;
1423 #endif
1426 static u32 tun_get_link(struct net_device *dev)
1428 struct tun_struct *tun = netdev_priv(dev);
1429 return !!tun->tfile;
1432 static u32 tun_get_rx_csum(struct net_device *dev)
1434 struct tun_struct *tun = netdev_priv(dev);
1435 return (tun->flags & TUN_NOCHECKSUM) == 0;
1438 static int tun_set_rx_csum(struct net_device *dev, u32 data)
1440 struct tun_struct *tun = netdev_priv(dev);
1441 if (data)
1442 tun->flags &= ~TUN_NOCHECKSUM;
1443 else
1444 tun->flags |= TUN_NOCHECKSUM;
1445 return 0;
1448 static const struct ethtool_ops tun_ethtool_ops = {
1449 .get_settings = tun_get_settings,
1450 .get_drvinfo = tun_get_drvinfo,
1451 .get_msglevel = tun_get_msglevel,
1452 .set_msglevel = tun_set_msglevel,
1453 .get_link = tun_get_link,
1454 .get_rx_csum = tun_get_rx_csum,
1455 .set_rx_csum = tun_set_rx_csum
1459 static int __init tun_init(void)
1461 int ret = 0;
1463 printk(KERN_INFO "tun: %s, %s\n", DRV_DESCRIPTION, DRV_VERSION);
1464 printk(KERN_INFO "tun: %s\n", DRV_COPYRIGHT);
1466 ret = rtnl_link_register(&tun_link_ops);
1467 if (ret) {
1468 printk(KERN_ERR "tun: Can't register link_ops\n");
1469 goto err_linkops;
1472 ret = misc_register(&tun_miscdev);
1473 if (ret) {
1474 printk(KERN_ERR "tun: Can't register misc device %d\n", TUN_MINOR);
1475 goto err_misc;
1477 return 0;
1478 err_misc:
1479 rtnl_link_unregister(&tun_link_ops);
1480 err_linkops:
1481 return ret;
1484 static void tun_cleanup(void)
1486 misc_deregister(&tun_miscdev);
1487 rtnl_link_unregister(&tun_link_ops);
1490 module_init(tun_init);
1491 module_exit(tun_cleanup);
1492 MODULE_DESCRIPTION(DRV_DESCRIPTION);
1493 MODULE_AUTHOR(DRV_COPYRIGHT);
1494 MODULE_LICENSE("GPL");
1495 MODULE_ALIAS_MISCDEV(TUN_MINOR);