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 $
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 eth_random_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.
33 * Daniel Podlejski <underley@underley.eu.org>
34 * Modifications for 2.3.99-pre5 kernel.
37 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
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>"
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>
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>
72 #include <asm/uaccess.h>
74 /* Uncomment to enable debugging */
75 /* #define TUN_DEBUG 1 */
80 #define tun_debug(level, tun, fmt, args...) \
83 netdev_printk(level, tun->dev, fmt, ##args); \
85 #define DBG1(level, fmt, args...) \
88 printk(level fmt, ##args); \
91 #define tun_debug(level, tun, fmt, args...) \
94 netdev_printk(level, tun->dev, fmt, ##args); \
96 #define DBG1(level, fmt, args...) \
99 printk(level fmt, ##args); \
103 #define FLT_EXACT_COUNT 8
105 unsigned int count; /* Number of addrs. Zero means disabled */
106 u32 mask[2]; /* Mask of the hashed addrs */
107 unsigned char addr[FLT_EXACT_COUNT][ETH_ALEN];
112 struct tun_struct *tun;
119 struct tun_file *tfile;
124 struct net_device *dev;
125 netdev_features_t set_features;
126 #define TUN_USER_FEATURES (NETIF_F_HW_CSUM|NETIF_F_TSO_ECN|NETIF_F_TSO| \
127 NETIF_F_TSO6|NETIF_F_UFO)
128 struct fasync_struct *fasync;
130 struct tap_filter txflt;
131 struct socket socket;
143 struct tun_struct *tun;
146 static inline struct tun_sock *tun_sk(struct sock *sk)
148 return container_of(sk, struct tun_sock, sk);
151 static int tun_attach(struct tun_struct *tun, struct file *file)
153 struct tun_file *tfile = file->private_data;
158 netif_tx_lock_bh(tun->dev);
171 tun->socket.file = file;
172 netif_carrier_on(tun->dev);
174 sock_hold(tun->socket.sk);
175 atomic_inc(&tfile->count);
178 netif_tx_unlock_bh(tun->dev);
182 static void __tun_detach(struct tun_struct *tun)
184 /* Detach from net device */
185 netif_tx_lock_bh(tun->dev);
186 netif_carrier_off(tun->dev);
188 tun->socket.file = NULL;
189 netif_tx_unlock_bh(tun->dev);
191 /* Drop read queue */
192 skb_queue_purge(&tun->socket.sk->sk_receive_queue);
194 /* Drop the extra count on the net device */
198 static void tun_detach(struct tun_struct *tun)
205 static struct tun_struct *__tun_get(struct tun_file *tfile)
207 struct tun_struct *tun = NULL;
209 if (atomic_inc_not_zero(&tfile->count))
215 static struct tun_struct *tun_get(struct file *file)
217 return __tun_get(file->private_data);
220 static void tun_put(struct tun_struct *tun)
222 struct tun_file *tfile = tun->tfile;
224 if (atomic_dec_and_test(&tfile->count))
225 tun_detach(tfile->tun);
229 static void addr_hash_set(u32 *mask, const u8 *addr)
231 int n = ether_crc(ETH_ALEN, addr) >> 26;
232 mask[n >> 5] |= (1 << (n & 31));
235 static unsigned int addr_hash_test(const u32 *mask, const u8 *addr)
237 int n = ether_crc(ETH_ALEN, addr) >> 26;
238 return mask[n >> 5] & (1 << (n & 31));
241 static int update_filter(struct tap_filter *filter, void __user *arg)
243 struct { u8 u[ETH_ALEN]; } *addr;
244 struct tun_filter uf;
245 int err, alen, n, nexact;
247 if (copy_from_user(&uf, arg, sizeof(uf)))
256 alen = ETH_ALEN * uf.count;
257 addr = kmalloc(alen, GFP_KERNEL);
261 if (copy_from_user(addr, arg + sizeof(uf), alen)) {
266 /* The filter is updated without holding any locks. Which is
267 * perfectly safe. We disable it first and in the worst
268 * case we'll accept a few undesired packets. */
272 /* Use first set of addresses as an exact filter */
273 for (n = 0; n < uf.count && n < FLT_EXACT_COUNT; n++)
274 memcpy(filter->addr[n], addr[n].u, ETH_ALEN);
278 /* Remaining multicast addresses are hashed,
279 * unicast will leave the filter disabled. */
280 memset(filter->mask, 0, sizeof(filter->mask));
281 for (; n < uf.count; n++) {
282 if (!is_multicast_ether_addr(addr[n].u)) {
283 err = 0; /* no filter */
286 addr_hash_set(filter->mask, addr[n].u);
289 /* For ALLMULTI just set the mask to all ones.
290 * This overrides the mask populated above. */
291 if ((uf.flags & TUN_FLT_ALLMULTI))
292 memset(filter->mask, ~0, sizeof(filter->mask));
294 /* Now enable the filter */
296 filter->count = nexact;
298 /* Return the number of exact filters */
306 /* Returns: 0 - drop, !=0 - accept */
307 static int run_filter(struct tap_filter *filter, const struct sk_buff *skb)
309 /* Cannot use eth_hdr(skb) here because skb_mac_hdr() is incorrect
311 struct ethhdr *eh = (struct ethhdr *) skb->data;
315 for (i = 0; i < filter->count; i++)
316 if (ether_addr_equal(eh->h_dest, filter->addr[i]))
319 /* Inexact match (multicast only) */
320 if (is_multicast_ether_addr(eh->h_dest))
321 return addr_hash_test(filter->mask, eh->h_dest);
327 * Checks whether the packet is accepted or not.
328 * Returns: 0 - drop, !=0 - accept
330 static int check_filter(struct tap_filter *filter, const struct sk_buff *skb)
335 return run_filter(filter, skb);
338 /* Network device part of the driver */
340 static const struct ethtool_ops tun_ethtool_ops;
342 /* Net device detach from fd. */
343 static void tun_net_uninit(struct net_device *dev)
345 struct tun_struct *tun = netdev_priv(dev);
346 struct tun_file *tfile = tun->tfile;
348 /* Inform the methods they need to stop using the dev.
351 wake_up_all(&tun->wq.wait);
352 if (atomic_dec_and_test(&tfile->count))
357 static void tun_free_netdev(struct net_device *dev)
359 struct tun_struct *tun = netdev_priv(dev);
361 sk_release_kernel(tun->socket.sk);
364 /* Net device open. */
365 static int tun_net_open(struct net_device *dev)
367 netif_start_queue(dev);
371 /* Net device close. */
372 static int tun_net_close(struct net_device *dev)
374 netif_stop_queue(dev);
378 /* Net device start xmit */
379 static netdev_tx_t tun_net_xmit(struct sk_buff *skb, struct net_device *dev)
381 struct tun_struct *tun = netdev_priv(dev);
383 tun_debug(KERN_INFO, tun, "tun_net_xmit %d\n", skb->len);
385 /* Drop packet if interface is not attached */
389 /* Drop if the filter does not like it.
390 * This is a noop if the filter is disabled.
391 * Filter can be enabled only for the TAP devices. */
392 if (!check_filter(&tun->txflt, skb))
395 if (tun->socket.sk->sk_filter &&
396 sk_filter(tun->socket.sk, skb))
399 if (skb_queue_len(&tun->socket.sk->sk_receive_queue) >= dev->tx_queue_len) {
400 if (!(tun->flags & TUN_ONE_QUEUE)) {
401 /* Normal queueing mode. */
402 /* Packet scheduler handles dropping of further packets. */
403 netif_stop_queue(dev);
405 /* We won't see all dropped packets individually, so overrun
406 * error is more appropriate. */
407 dev->stats.tx_fifo_errors++;
409 /* Single queue mode.
410 * Driver handles dropping of all packets itself. */
415 /* Orphan the skb - required as we might hang on to it
416 * for indefinite time. */
420 skb_queue_tail(&tun->socket.sk->sk_receive_queue, skb);
422 /* Notify and wake up reader process */
423 if (tun->flags & TUN_FASYNC)
424 kill_fasync(&tun->fasync, SIGIO, POLL_IN);
425 wake_up_interruptible_poll(&tun->wq.wait, POLLIN |
426 POLLRDNORM | POLLRDBAND);
430 dev->stats.tx_dropped++;
435 static void tun_net_mclist(struct net_device *dev)
438 * This callback is supposed to deal with mc filter in
439 * _rx_ path and has nothing to do with the _tx_ path.
440 * In rx path we always accept everything userspace gives us.
445 #define MAX_MTU 65535
448 tun_net_change_mtu(struct net_device *dev, int new_mtu)
450 if (new_mtu < MIN_MTU || new_mtu + dev->hard_header_len > MAX_MTU)
456 static netdev_features_t tun_net_fix_features(struct net_device *dev,
457 netdev_features_t features)
459 struct tun_struct *tun = netdev_priv(dev);
461 return (features & tun->set_features) | (features & ~TUN_USER_FEATURES);
463 #ifdef CONFIG_NET_POLL_CONTROLLER
464 static void tun_poll_controller(struct net_device *dev)
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
480 static 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,
492 static 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_rx_mode = 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,
507 /* Initialize net device. */
508 static void tun_net_init(struct net_device *dev)
510 struct tun_struct *tun = netdev_priv(dev);
512 switch (tun->flags & TUN_TYPE_MASK) {
514 dev->netdev_ops = &tun_netdev_ops;
516 /* Point-to-Point TUN Device */
517 dev->hard_header_len = 0;
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 */
528 dev->netdev_ops = &tap_netdev_ops;
529 /* Ethernet TAP Device */
531 dev->priv_flags &= ~IFF_TX_SKB_SHARING;
533 eth_hw_addr_random(dev);
535 dev->tx_queue_len = TUN_READQ_SIZE; /* We prefer our own queue length */
540 /* Character device part */
543 static unsigned int tun_chr_poll(struct file *file, poll_table * wait)
545 struct tun_file *tfile = file->private_data;
546 struct tun_struct *tun = __tun_get(tfile);
548 unsigned int mask = 0;
555 tun_debug(KERN_INFO, tun, "tun_chr_poll\n");
557 poll_wait(file, &tun->wq.wait, wait);
559 if (!skb_queue_empty(&sk->sk_receive_queue))
560 mask |= POLLIN | POLLRDNORM;
562 if (sock_writeable(sk) ||
563 (!test_and_set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags) &&
565 mask |= POLLOUT | POLLWRNORM;
567 if (tun->dev->reg_state != NETREG_REGISTERED)
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). */
576 static struct sk_buff *tun_alloc_skb(struct tun_struct *tun,
577 size_t prepad, size_t len,
578 size_t linear, int noblock)
580 struct sock *sk = tun->socket.sk;
584 sock_update_classid(sk);
586 /* Under a page? Don't bother with paged skb. */
587 if (prepad + len < PAGE_SIZE || !linear)
590 skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
595 skb_reserve(skb, prepad);
596 skb_put(skb, linear);
597 skb->data_len = len - linear;
598 skb->len += len - linear;
603 /* Get packet from user space buffer */
604 static ssize_t tun_get_user(struct tun_struct *tun,
605 const struct iovec *iv, size_t count,
608 struct tun_pi pi = { 0, cpu_to_be16(ETH_P_IP) };
610 size_t len = count, align = NET_SKB_PAD;
611 struct virtio_net_hdr gso = { 0 };
614 if (!(tun->flags & TUN_NO_PI)) {
615 if ((len -= sizeof(pi)) > count)
618 if (memcpy_fromiovecend((void *)&pi, iv, 0, sizeof(pi)))
620 offset += sizeof(pi);
623 if (tun->flags & TUN_VNET_HDR) {
624 if ((len -= tun->vnet_hdr_sz) > count)
627 if (memcpy_fromiovecend((void *)&gso, iv, offset, sizeof(gso)))
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;
634 if (gso.hdr_len > len)
636 offset += tun->vnet_hdr_sz;
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)))
646 skb = tun_alloc_skb(tun, align, len, gso.hdr_len, noblock);
648 if (PTR_ERR(skb) != -EAGAIN)
649 tun->dev->stats.rx_dropped++;
653 if (skb_copy_datagram_from_iovec(skb, 0, iv, offset, len)) {
654 tun->dev->stats.rx_dropped++;
659 if (gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
660 if (!skb_partial_csum_set(skb, gso.csum_start,
662 tun->dev->stats.rx_frame_errors++;
668 switch (tun->flags & TUN_TYPE_MASK) {
670 if (tun->flags & TUN_NO_PI) {
671 switch (skb->data[0] & 0xf0) {
673 pi.proto = htons(ETH_P_IP);
676 pi.proto = htons(ETH_P_IPV6);
679 tun->dev->stats.rx_dropped++;
685 skb_reset_mac_header(skb);
686 skb->protocol = pi.proto;
690 skb->protocol = eth_type_trans(skb, tun->dev);
694 if (gso.gso_type != VIRTIO_NET_HDR_GSO_NONE) {
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;
700 case VIRTIO_NET_HDR_GSO_TCPV6:
701 skb_shinfo(skb)->gso_type = SKB_GSO_TCPV6;
703 case VIRTIO_NET_HDR_GSO_UDP:
704 skb_shinfo(skb)->gso_type = SKB_GSO_UDP;
707 tun->dev->stats.rx_frame_errors++;
712 if (gso.gso_type & VIRTIO_NET_HDR_GSO_ECN)
713 skb_shinfo(skb)->gso_type |= SKB_GSO_TCP_ECN;
715 skb_shinfo(skb)->gso_size = gso.gso_size;
716 if (skb_shinfo(skb)->gso_size == 0) {
717 tun->dev->stats.rx_frame_errors++;
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;
729 tun->dev->stats.rx_packets++;
730 tun->dev->stats.rx_bytes += len;
735 static ssize_t tun_chr_aio_write(struct kiocb *iocb, const struct iovec *iv,
736 unsigned long count, loff_t pos)
738 struct file *file = iocb->ki_filp;
739 struct tun_struct *tun = tun_get(file);
745 tun_debug(KERN_INFO, tun, "tun_chr_write %ld\n", count);
747 result = tun_get_user(tun, iv, iov_length(iv, count),
748 file->f_flags & O_NONBLOCK);
754 /* Put packet to the user space buffer */
755 static ssize_t tun_put_user(struct tun_struct *tun,
757 const struct iovec *iv, int len)
759 struct tun_pi pi = { 0, skb->protocol };
762 if (!(tun->flags & TUN_NO_PI)) {
763 if ((len -= sizeof(pi)) < 0)
766 if (len < skb->len) {
767 /* Packet will be striped */
768 pi.flags |= TUN_PKT_STRIP;
771 if (memcpy_toiovecend(iv, (void *) &pi, 0, sizeof(pi)))
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)
781 if (skb_is_gso(skb)) {
782 struct skb_shared_info *sinfo = skb_shinfo(skb);
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;
794 pr_err("unexpected GSO type: "
795 "0x%x, gso_size %d, hdr_len %d\n",
796 sinfo->gso_type, gso.gso_size,
798 print_hex_dump(KERN_ERR, "tun: ",
801 min((int)gso.hdr_len, 64), true);
805 if (sinfo->gso_type & SKB_GSO_TCP_ECN)
806 gso.gso_type |= VIRTIO_NET_HDR_GSO_ECN;
808 gso.gso_type = VIRTIO_NET_HDR_GSO_NONE;
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 */
818 if (unlikely(memcpy_toiovecend(iv, (void *)&gso, total,
821 total += tun->vnet_hdr_sz;
824 len = min_t(int, skb->len, len);
826 skb_copy_datagram_const_iovec(skb, 0, iv, total, len);
829 tun->dev->stats.tx_packets++;
830 tun->dev->stats.tx_bytes += len;
835 static ssize_t tun_do_read(struct tun_struct *tun,
836 struct kiocb *iocb, const struct iovec *iv,
837 ssize_t len, int noblock)
839 DECLARE_WAITQUEUE(wait, current);
843 tun_debug(KERN_INFO, tun, "tun_chr_read\n");
845 if (unlikely(!noblock))
846 add_wait_queue(&tun->wq.wait, &wait);
848 current->state = TASK_INTERRUPTIBLE;
850 /* Read frames from the queue */
851 if (!(skb=skb_dequeue(&tun->socket.sk->sk_receive_queue))) {
856 if (signal_pending(current)) {
860 if (tun->dev->reg_state != NETREG_REGISTERED) {
865 /* Nothing to read, let's sleep */
869 netif_wake_queue(tun->dev);
871 ret = tun_put_user(tun, skb, iv, len);
876 current->state = TASK_RUNNING;
877 if (unlikely(!noblock))
878 remove_wait_queue(&tun->wq.wait, &wait);
883 static ssize_t tun_chr_aio_read(struct kiocb *iocb, const struct iovec *iv,
884 unsigned long count, loff_t pos)
886 struct file *file = iocb->ki_filp;
887 struct tun_file *tfile = file->private_data;
888 struct tun_struct *tun = __tun_get(tfile);
893 len = iov_length(iv, count);
899 ret = tun_do_read(tun, iocb, iv, len, file->f_flags & O_NONBLOCK);
900 ret = min_t(ssize_t, ret, len);
906 static void tun_setup(struct net_device *dev)
908 struct tun_struct *tun = netdev_priv(dev);
913 dev->ethtool_ops = &tun_ethtool_ops;
914 dev->destructor = tun_free_netdev;
917 /* Trivial set of netlink ops to allow deleting tun or tap
918 * device with netlink.
920 static int tun_validate(struct nlattr *tb[], struct nlattr *data[])
925 static struct rtnl_link_ops tun_link_ops __read_mostly = {
927 .priv_size = sizeof(struct tun_struct),
929 .validate = tun_validate,
932 static void tun_sock_write_space(struct sock *sk)
934 struct tun_struct *tun;
935 wait_queue_head_t *wqueue;
937 if (!sock_writeable(sk))
940 if (!test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags))
943 wqueue = sk_sleep(sk);
944 if (wqueue && waitqueue_active(wqueue))
945 wake_up_interruptible_sync_poll(wqueue, POLLOUT |
946 POLLWRNORM | POLLWRBAND);
948 tun = tun_sk(sk)->tun;
949 kill_fasync(&tun->fasync, SIGIO, POLL_OUT);
952 static void tun_sock_destruct(struct sock *sk)
954 free_netdev(tun_sk(sk)->tun->dev);
957 static int tun_sendmsg(struct kiocb *iocb, struct socket *sock,
958 struct msghdr *m, size_t total_len)
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);
965 static int tun_recvmsg(struct kiocb *iocb, struct socket *sock,
966 struct msghdr *m, size_t total_len,
969 struct tun_struct *tun = container_of(sock, struct tun_struct, socket);
971 if (flags & ~(MSG_DONTWAIT|MSG_TRUNC))
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;
982 static int tun_release(struct socket *sock)
989 /* Ops structure to mimic raw sockets with tun */
990 static const struct proto_ops tun_socket_ops = {
991 .sendmsg = tun_sendmsg,
992 .recvmsg = tun_recvmsg,
993 .release = tun_release,
996 static struct proto tun_proto = {
998 .owner = THIS_MODULE,
999 .obj_size = sizeof(struct tun_sock),
1002 static int tun_flags(struct tun_struct *tun)
1006 if (tun->flags & TUN_TUN_DEV)
1011 if (tun->flags & TUN_NO_PI)
1014 if (tun->flags & TUN_ONE_QUEUE)
1015 flags |= IFF_ONE_QUEUE;
1017 if (tun->flags & TUN_VNET_HDR)
1018 flags |= IFF_VNET_HDR;
1023 static ssize_t tun_show_flags(struct device *dev, struct device_attribute *attr,
1026 struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1027 return sprintf(buf, "0x%x\n", tun_flags(tun));
1030 static ssize_t tun_show_owner(struct device *dev, struct device_attribute *attr,
1033 struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1034 return sprintf(buf, "%d\n", tun->owner);
1037 static ssize_t tun_show_group(struct device *dev, struct device_attribute *attr,
1040 struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1041 return sprintf(buf, "%d\n", tun->group);
1044 static DEVICE_ATTR(tun_flags, 0444, tun_show_flags, NULL);
1045 static DEVICE_ATTR(owner, 0444, tun_show_owner, NULL);
1046 static DEVICE_ATTR(group, 0444, tun_show_group, NULL);
1048 static int tun_set_iff(struct net *net, struct file *file, struct ifreq *ifr)
1051 struct tun_struct *tun;
1052 struct net_device *dev;
1055 dev = __dev_get_by_name(net, ifr->ifr_name);
1057 const struct cred *cred = current_cred();
1059 if (ifr->ifr_flags & IFF_TUN_EXCL)
1061 if ((ifr->ifr_flags & IFF_TUN) && dev->netdev_ops == &tun_netdev_ops)
1062 tun = netdev_priv(dev);
1063 else if ((ifr->ifr_flags & IFF_TAP) && dev->netdev_ops == &tap_netdev_ops)
1064 tun = netdev_priv(dev);
1068 if (((tun->owner != -1 && cred->euid != tun->owner) ||
1069 (tun->group != -1 && !in_egroup_p(tun->group))) &&
1070 !capable(CAP_NET_ADMIN))
1072 err = security_tun_dev_attach(tun->socket.sk);
1076 err = tun_attach(tun, file);
1082 unsigned long flags = 0;
1084 if (!capable(CAP_NET_ADMIN))
1086 err = security_tun_dev_create();
1091 if (ifr->ifr_flags & IFF_TUN) {
1093 flags |= TUN_TUN_DEV;
1095 } else if (ifr->ifr_flags & IFF_TAP) {
1097 flags |= TUN_TAP_DEV;
1103 name = ifr->ifr_name;
1105 dev = alloc_netdev(sizeof(struct tun_struct), name,
1110 dev_net_set(dev, net);
1111 dev->rtnl_link_ops = &tun_link_ops;
1113 tun = netdev_priv(dev);
1116 tun->txflt.count = 0;
1117 tun->vnet_hdr_sz = sizeof(struct virtio_net_hdr);
1120 sk = sk_alloc(&init_net, AF_UNSPEC, GFP_KERNEL, &tun_proto);
1124 sk_change_net(sk, net);
1125 tun->socket.wq = &tun->wq;
1126 init_waitqueue_head(&tun->wq.wait);
1127 tun->socket.ops = &tun_socket_ops;
1128 sock_init_data(&tun->socket, sk);
1129 sk->sk_write_space = tun_sock_write_space;
1130 sk->sk_sndbuf = INT_MAX;
1132 tun_sk(sk)->tun = tun;
1134 security_tun_dev_post_create(sk);
1138 dev->hw_features = NETIF_F_SG | NETIF_F_FRAGLIST |
1140 dev->features = dev->hw_features;
1142 err = register_netdevice(tun->dev);
1146 if (device_create_file(&tun->dev->dev, &dev_attr_tun_flags) ||
1147 device_create_file(&tun->dev->dev, &dev_attr_owner) ||
1148 device_create_file(&tun->dev->dev, &dev_attr_group))
1149 pr_err("Failed to create tun sysfs files\n");
1151 sk->sk_destruct = tun_sock_destruct;
1153 err = tun_attach(tun, file);
1158 tun_debug(KERN_INFO, tun, "tun_set_iff\n");
1160 if (ifr->ifr_flags & IFF_NO_PI)
1161 tun->flags |= TUN_NO_PI;
1163 tun->flags &= ~TUN_NO_PI;
1165 if (ifr->ifr_flags & IFF_ONE_QUEUE)
1166 tun->flags |= TUN_ONE_QUEUE;
1168 tun->flags &= ~TUN_ONE_QUEUE;
1170 if (ifr->ifr_flags & IFF_VNET_HDR)
1171 tun->flags |= TUN_VNET_HDR;
1173 tun->flags &= ~TUN_VNET_HDR;
1175 /* Make sure persistent devices do not get stuck in
1178 if (netif_running(tun->dev))
1179 netif_wake_queue(tun->dev);
1181 strcpy(ifr->ifr_name, tun->dev->name);
1185 tun_free_netdev(dev);
1192 static int tun_get_iff(struct net *net, struct tun_struct *tun,
1195 tun_debug(KERN_INFO, tun, "tun_get_iff\n");
1197 strcpy(ifr->ifr_name, tun->dev->name);
1199 ifr->ifr_flags = tun_flags(tun);
1204 /* This is like a cut-down ethtool ops, except done via tun fd so no
1205 * privs required. */
1206 static int set_offload(struct tun_struct *tun, unsigned long arg)
1208 netdev_features_t features = 0;
1210 if (arg & TUN_F_CSUM) {
1211 features |= NETIF_F_HW_CSUM;
1214 if (arg & (TUN_F_TSO4|TUN_F_TSO6)) {
1215 if (arg & TUN_F_TSO_ECN) {
1216 features |= NETIF_F_TSO_ECN;
1217 arg &= ~TUN_F_TSO_ECN;
1219 if (arg & TUN_F_TSO4)
1220 features |= NETIF_F_TSO;
1221 if (arg & TUN_F_TSO6)
1222 features |= NETIF_F_TSO6;
1223 arg &= ~(TUN_F_TSO4|TUN_F_TSO6);
1226 if (arg & TUN_F_UFO) {
1227 features |= NETIF_F_UFO;
1232 /* This gives the user a way to test for new features in future by
1233 * trying to set them. */
1237 tun->set_features = features;
1238 netdev_update_features(tun->dev);
1243 static long __tun_chr_ioctl(struct file *file, unsigned int cmd,
1244 unsigned long arg, int ifreq_len)
1246 struct tun_file *tfile = file->private_data;
1247 struct tun_struct *tun;
1248 void __user* argp = (void __user*)arg;
1249 struct sock_fprog fprog;
1255 if (cmd == TUNSETIFF || _IOC_TYPE(cmd) == 0x89)
1256 if (copy_from_user(&ifr, argp, ifreq_len))
1259 if (cmd == TUNGETFEATURES) {
1260 /* Currently this just means: "what IFF flags are valid?".
1261 * This is needed because we never checked for invalid flags on
1263 return put_user(IFF_TUN | IFF_TAP | IFF_NO_PI | IFF_ONE_QUEUE |
1265 (unsigned int __user*)argp);
1270 tun = __tun_get(tfile);
1271 if (cmd == TUNSETIFF && !tun) {
1272 ifr.ifr_name[IFNAMSIZ-1] = '\0';
1274 ret = tun_set_iff(tfile->net, file, &ifr);
1279 if (copy_to_user(argp, &ifr, ifreq_len))
1288 tun_debug(KERN_INFO, tun, "tun_chr_ioctl cmd %d\n", cmd);
1293 ret = tun_get_iff(current->nsproxy->net_ns, tun, &ifr);
1297 if (copy_to_user(argp, &ifr, ifreq_len))
1302 /* Disable/Enable checksum */
1304 /* [unimplemented] */
1305 tun_debug(KERN_INFO, tun, "ignored: set checksum %s\n",
1306 arg ? "disabled" : "enabled");
1310 /* Disable/Enable persist mode */
1312 tun->flags |= TUN_PERSIST;
1314 tun->flags &= ~TUN_PERSIST;
1316 tun_debug(KERN_INFO, tun, "persist %s\n",
1317 arg ? "enabled" : "disabled");
1321 /* Set owner of the device */
1322 tun->owner = (uid_t) arg;
1324 tun_debug(KERN_INFO, tun, "owner set to %d\n", tun->owner);
1328 /* Set group of the device */
1329 tun->group= (gid_t) arg;
1331 tun_debug(KERN_INFO, tun, "group set to %d\n", tun->group);
1335 /* Only allow setting the type when the interface is down */
1336 if (tun->dev->flags & IFF_UP) {
1337 tun_debug(KERN_INFO, tun,
1338 "Linktype set failed because interface is up\n");
1341 tun->dev->type = (int) arg;
1342 tun_debug(KERN_INFO, tun, "linktype set to %d\n",
1354 ret = set_offload(tun, arg);
1357 case TUNSETTXFILTER:
1358 /* Can be set only for TAPs */
1360 if ((tun->flags & TUN_TYPE_MASK) != TUN_TAP_DEV)
1362 ret = update_filter(&tun->txflt, (void __user *)arg);
1366 /* Get hw address */
1367 memcpy(ifr.ifr_hwaddr.sa_data, tun->dev->dev_addr, ETH_ALEN);
1368 ifr.ifr_hwaddr.sa_family = tun->dev->type;
1369 if (copy_to_user(argp, &ifr, ifreq_len))
1374 /* Set hw address */
1375 tun_debug(KERN_DEBUG, tun, "set hw address: %pM\n",
1376 ifr.ifr_hwaddr.sa_data);
1378 ret = dev_set_mac_address(tun->dev, &ifr.ifr_hwaddr);
1382 sndbuf = tun->socket.sk->sk_sndbuf;
1383 if (copy_to_user(argp, &sndbuf, sizeof(sndbuf)))
1388 if (copy_from_user(&sndbuf, argp, sizeof(sndbuf))) {
1393 tun->socket.sk->sk_sndbuf = sndbuf;
1396 case TUNGETVNETHDRSZ:
1397 vnet_hdr_sz = tun->vnet_hdr_sz;
1398 if (copy_to_user(argp, &vnet_hdr_sz, sizeof(vnet_hdr_sz)))
1402 case TUNSETVNETHDRSZ:
1403 if (copy_from_user(&vnet_hdr_sz, argp, sizeof(vnet_hdr_sz))) {
1407 if (vnet_hdr_sz < (int)sizeof(struct virtio_net_hdr)) {
1412 tun->vnet_hdr_sz = vnet_hdr_sz;
1415 case TUNATTACHFILTER:
1416 /* Can be set only for TAPs */
1418 if ((tun->flags & TUN_TYPE_MASK) != TUN_TAP_DEV)
1421 if (copy_from_user(&fprog, argp, sizeof(fprog)))
1424 ret = sk_attach_filter(&fprog, tun->socket.sk);
1427 case TUNDETACHFILTER:
1428 /* Can be set only for TAPs */
1430 if ((tun->flags & TUN_TYPE_MASK) != TUN_TAP_DEV)
1432 ret = sk_detach_filter(tun->socket.sk);
1447 static long tun_chr_ioctl(struct file *file,
1448 unsigned int cmd, unsigned long arg)
1450 return __tun_chr_ioctl(file, cmd, arg, sizeof (struct ifreq));
1453 #ifdef CONFIG_COMPAT
1454 static long tun_chr_compat_ioctl(struct file *file,
1455 unsigned int cmd, unsigned long arg)
1460 case TUNSETTXFILTER:
1465 arg = (unsigned long)compat_ptr(arg);
1468 arg = (compat_ulong_t)arg;
1473 * compat_ifreq is shorter than ifreq, so we must not access beyond
1474 * the end of that structure. All fields that are used in this
1475 * driver are compatible though, we don't need to convert the
1478 return __tun_chr_ioctl(file, cmd, arg, sizeof(struct compat_ifreq));
1480 #endif /* CONFIG_COMPAT */
1482 static int tun_chr_fasync(int fd, struct file *file, int on)
1484 struct tun_struct *tun = tun_get(file);
1490 tun_debug(KERN_INFO, tun, "tun_chr_fasync %d\n", on);
1492 if ((ret = fasync_helper(fd, file, on, &tun->fasync)) < 0)
1496 ret = __f_setown(file, task_pid(current), PIDTYPE_PID, 0);
1499 tun->flags |= TUN_FASYNC;
1501 tun->flags &= ~TUN_FASYNC;
1508 static int tun_chr_open(struct inode *inode, struct file * file)
1510 struct tun_file *tfile;
1512 DBG1(KERN_INFO, "tunX: tun_chr_open\n");
1514 tfile = kmalloc(sizeof(*tfile), GFP_KERNEL);
1517 atomic_set(&tfile->count, 0);
1519 tfile->net = get_net(current->nsproxy->net_ns);
1520 file->private_data = tfile;
1524 static int tun_chr_close(struct inode *inode, struct file *file)
1526 struct tun_file *tfile = file->private_data;
1527 struct tun_struct *tun;
1529 tun = __tun_get(tfile);
1531 struct net_device *dev = tun->dev;
1533 tun_debug(KERN_INFO, tun, "tun_chr_close\n");
1537 /* If desirable, unregister the netdevice. */
1538 if (!(tun->flags & TUN_PERSIST)) {
1540 if (dev->reg_state == NETREG_REGISTERED)
1541 unregister_netdevice(dev);
1548 sock_put(tun->socket.sk);
1550 put_net(tfile->net);
1556 static const struct file_operations tun_fops = {
1557 .owner = THIS_MODULE,
1558 .llseek = no_llseek,
1559 .read = do_sync_read,
1560 .aio_read = tun_chr_aio_read,
1561 .write = do_sync_write,
1562 .aio_write = tun_chr_aio_write,
1563 .poll = tun_chr_poll,
1564 .unlocked_ioctl = tun_chr_ioctl,
1565 #ifdef CONFIG_COMPAT
1566 .compat_ioctl = tun_chr_compat_ioctl,
1568 .open = tun_chr_open,
1569 .release = tun_chr_close,
1570 .fasync = tun_chr_fasync
1573 static struct miscdevice tun_miscdev = {
1576 .nodename = "net/tun",
1580 /* ethtool interface */
1582 static int tun_get_settings(struct net_device *dev, struct ethtool_cmd *cmd)
1585 cmd->advertising = 0;
1586 ethtool_cmd_speed_set(cmd, SPEED_10);
1587 cmd->duplex = DUPLEX_FULL;
1588 cmd->port = PORT_TP;
1589 cmd->phy_address = 0;
1590 cmd->transceiver = XCVR_INTERNAL;
1591 cmd->autoneg = AUTONEG_DISABLE;
1597 static void tun_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
1599 struct tun_struct *tun = netdev_priv(dev);
1601 strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
1602 strlcpy(info->version, DRV_VERSION, sizeof(info->version));
1604 switch (tun->flags & TUN_TYPE_MASK) {
1606 strlcpy(info->bus_info, "tun", sizeof(info->bus_info));
1609 strlcpy(info->bus_info, "tap", sizeof(info->bus_info));
1614 static u32 tun_get_msglevel(struct net_device *dev)
1617 struct tun_struct *tun = netdev_priv(dev);
1624 static void tun_set_msglevel(struct net_device *dev, u32 value)
1627 struct tun_struct *tun = netdev_priv(dev);
1632 static const struct ethtool_ops tun_ethtool_ops = {
1633 .get_settings = tun_get_settings,
1634 .get_drvinfo = tun_get_drvinfo,
1635 .get_msglevel = tun_get_msglevel,
1636 .set_msglevel = tun_set_msglevel,
1637 .get_link = ethtool_op_get_link,
1641 static int __init tun_init(void)
1645 pr_info("%s, %s\n", DRV_DESCRIPTION, DRV_VERSION);
1646 pr_info("%s\n", DRV_COPYRIGHT);
1648 ret = rtnl_link_register(&tun_link_ops);
1650 pr_err("Can't register link_ops\n");
1654 ret = misc_register(&tun_miscdev);
1656 pr_err("Can't register misc device %d\n", TUN_MINOR);
1661 rtnl_link_unregister(&tun_link_ops);
1666 static void tun_cleanup(void)
1668 misc_deregister(&tun_miscdev);
1669 rtnl_link_unregister(&tun_link_ops);
1672 /* Get an underlying socket object from tun file. Returns error unless file is
1673 * attached to a device. The returned object works like a packet socket, it
1674 * can be used for sock_sendmsg/sock_recvmsg. The caller is responsible for
1675 * holding a reference to the file for as long as the socket is in use. */
1676 struct socket *tun_get_socket(struct file *file)
1678 struct tun_struct *tun;
1679 if (file->f_op != &tun_fops)
1680 return ERR_PTR(-EINVAL);
1681 tun = tun_get(file);
1683 return ERR_PTR(-EBADFD);
1685 return &tun->socket;
1687 EXPORT_SYMBOL_GPL(tun_get_socket);
1689 module_init(tun_init);
1690 module_exit(tun_cleanup);
1691 MODULE_DESCRIPTION(DRV_DESCRIPTION);
1692 MODULE_AUTHOR(DRV_COPYRIGHT);
1693 MODULE_LICENSE("GPL");
1694 MODULE_ALIAS_MISCDEV(TUN_MINOR);
1695 MODULE_ALIAS("devname:net/tun");