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 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.
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/system.h>
73 #include <asm/uaccess.h>
75 /* Uncomment to enable debugging */
76 /* #define TUN_DEBUG 1 */
81 #define tun_debug(level, tun, fmt, args...) \
84 netdev_printk(level, tun->dev, fmt, ##args); \
86 #define DBG1(level, fmt, args...) \
89 printk(level fmt, ##args); \
92 #define tun_debug(level, tun, fmt, args...) \
95 netdev_printk(level, tun->dev, fmt, ##args); \
97 #define DBG1(level, fmt, args...) \
100 printk(level fmt, ##args); \
104 #define FLT_EXACT_COUNT 8
106 unsigned int count; /* Number of addrs. Zero means disabled */
107 u32 mask[2]; /* Mask of the hashed addrs */
108 unsigned char addr[FLT_EXACT_COUNT][ETH_ALEN];
113 struct tun_struct *tun;
120 struct tun_file *tfile;
125 struct net_device *dev;
126 netdev_features_t set_features;
127 #define TUN_USER_FEATURES (NETIF_F_HW_CSUM|NETIF_F_TSO_ECN|NETIF_F_TSO| \
128 NETIF_F_TSO6|NETIF_F_UFO)
129 struct fasync_struct *fasync;
131 struct tap_filter txflt;
132 struct socket socket;
144 struct tun_struct *tun;
147 static inline struct tun_sock *tun_sk(struct sock *sk)
149 return container_of(sk, struct tun_sock, sk);
152 static int tun_attach(struct tun_struct *tun, struct file *file)
154 struct tun_file *tfile = file->private_data;
159 netif_tx_lock_bh(tun->dev);
172 tun->socket.file = file;
173 netif_carrier_on(tun->dev);
175 sock_hold(tun->socket.sk);
176 atomic_inc(&tfile->count);
179 netif_tx_unlock_bh(tun->dev);
183 static void __tun_detach(struct tun_struct *tun)
185 /* Detach from net device */
186 netif_tx_lock_bh(tun->dev);
187 netif_carrier_off(tun->dev);
189 tun->socket.file = NULL;
190 netif_tx_unlock_bh(tun->dev);
192 /* Drop read queue */
193 skb_queue_purge(&tun->socket.sk->sk_receive_queue);
195 /* Drop the extra count on the net device */
199 static void tun_detach(struct tun_struct *tun)
206 static struct tun_struct *__tun_get(struct tun_file *tfile)
208 struct tun_struct *tun = NULL;
210 if (atomic_inc_not_zero(&tfile->count))
216 static struct tun_struct *tun_get(struct file *file)
218 return __tun_get(file->private_data);
221 static void tun_put(struct tun_struct *tun)
223 struct tun_file *tfile = tun->tfile;
225 if (atomic_dec_and_test(&tfile->count))
226 tun_detach(tfile->tun);
230 static void addr_hash_set(u32 *mask, const u8 *addr)
232 int n = ether_crc(ETH_ALEN, addr) >> 26;
233 mask[n >> 5] |= (1 << (n & 31));
236 static unsigned int addr_hash_test(const u32 *mask, const u8 *addr)
238 int n = ether_crc(ETH_ALEN, addr) >> 26;
239 return mask[n >> 5] & (1 << (n & 31));
242 static int update_filter(struct tap_filter *filter, void __user *arg)
244 struct { u8 u[ETH_ALEN]; } *addr;
245 struct tun_filter uf;
246 int err, alen, n, nexact;
248 if (copy_from_user(&uf, arg, sizeof(uf)))
257 alen = ETH_ALEN * uf.count;
258 addr = kmalloc(alen, GFP_KERNEL);
262 if (copy_from_user(addr, arg + sizeof(uf), alen)) {
267 /* The filter is updated without holding any locks. Which is
268 * perfectly safe. We disable it first and in the worst
269 * case we'll accept a few undesired packets. */
273 /* Use first set of addresses as an exact filter */
274 for (n = 0; n < uf.count && n < FLT_EXACT_COUNT; n++)
275 memcpy(filter->addr[n], addr[n].u, ETH_ALEN);
279 /* Remaining multicast addresses are hashed,
280 * unicast will leave the filter disabled. */
281 memset(filter->mask, 0, sizeof(filter->mask));
282 for (; n < uf.count; n++) {
283 if (!is_multicast_ether_addr(addr[n].u)) {
284 err = 0; /* no filter */
287 addr_hash_set(filter->mask, addr[n].u);
290 /* For ALLMULTI just set the mask to all ones.
291 * This overrides the mask populated above. */
292 if ((uf.flags & TUN_FLT_ALLMULTI))
293 memset(filter->mask, ~0, sizeof(filter->mask));
295 /* Now enable the filter */
297 filter->count = nexact;
299 /* Return the number of exact filters */
307 /* Returns: 0 - drop, !=0 - accept */
308 static int run_filter(struct tap_filter *filter, const struct sk_buff *skb)
310 /* Cannot use eth_hdr(skb) here because skb_mac_hdr() is incorrect
312 struct ethhdr *eh = (struct ethhdr *) skb->data;
316 for (i = 0; i < filter->count; i++)
317 if (!compare_ether_addr(eh->h_dest, filter->addr[i]))
320 /* Inexact match (multicast only) */
321 if (is_multicast_ether_addr(eh->h_dest))
322 return addr_hash_test(filter->mask, eh->h_dest);
328 * Checks whether the packet is accepted or not.
329 * Returns: 0 - drop, !=0 - accept
331 static int check_filter(struct tap_filter *filter, const struct sk_buff *skb)
336 return run_filter(filter, skb);
339 /* Network device part of the driver */
341 static const struct ethtool_ops tun_ethtool_ops;
343 /* Net device detach from fd. */
344 static void tun_net_uninit(struct net_device *dev)
346 struct tun_struct *tun = netdev_priv(dev);
347 struct tun_file *tfile = tun->tfile;
349 /* Inform the methods they need to stop using the dev.
352 wake_up_all(&tun->wq.wait);
353 if (atomic_dec_and_test(&tfile->count))
358 static void tun_free_netdev(struct net_device *dev)
360 struct tun_struct *tun = netdev_priv(dev);
362 sk_release_kernel(tun->socket.sk);
365 /* Net device open. */
366 static int tun_net_open(struct net_device *dev)
368 netif_start_queue(dev);
372 /* Net device close. */
373 static int tun_net_close(struct net_device *dev)
375 netif_stop_queue(dev);
379 /* Net device start xmit */
380 static netdev_tx_t tun_net_xmit(struct sk_buff *skb, struct net_device *dev)
382 struct tun_struct *tun = netdev_priv(dev);
384 tun_debug(KERN_INFO, tun, "tun_net_xmit %d\n", skb->len);
386 /* Drop packet if interface is not attached */
390 /* Drop if the filter does not like it.
391 * This is a noop if the filter is disabled.
392 * Filter can be enabled only for the TAP devices. */
393 if (!check_filter(&tun->txflt, skb))
396 if (tun->socket.sk->sk_filter &&
397 sk_filter(tun->socket.sk, skb))
400 if (skb_queue_len(&tun->socket.sk->sk_receive_queue) >= dev->tx_queue_len) {
401 if (!(tun->flags & TUN_ONE_QUEUE)) {
402 /* Normal queueing mode. */
403 /* Packet scheduler handles dropping of further packets. */
404 netif_stop_queue(dev);
406 /* We won't see all dropped packets individually, so overrun
407 * error is more appropriate. */
408 dev->stats.tx_fifo_errors++;
410 /* Single queue mode.
411 * Driver handles dropping of all packets itself. */
416 /* Orphan the skb - required as we might hang on to it
417 * for indefinite time. */
421 skb_queue_tail(&tun->socket.sk->sk_receive_queue, skb);
423 /* Notify and wake up reader process */
424 if (tun->flags & TUN_FASYNC)
425 kill_fasync(&tun->fasync, SIGIO, POLL_IN);
426 wake_up_interruptible_poll(&tun->wq.wait, POLLIN |
427 POLLRDNORM | POLLRDBAND);
431 dev->stats.tx_dropped++;
436 static void tun_net_mclist(struct net_device *dev)
439 * This callback is supposed to deal with mc filter in
440 * _rx_ path and has nothing to do with the _tx_ path.
441 * In rx path we always accept everything userspace gives us.
446 #define MAX_MTU 65535
449 tun_net_change_mtu(struct net_device *dev, int new_mtu)
451 if (new_mtu < MIN_MTU || new_mtu + dev->hard_header_len > MAX_MTU)
457 static netdev_features_t tun_net_fix_features(struct net_device *dev,
458 netdev_features_t features)
460 struct tun_struct *tun = netdev_priv(dev);
462 return (features & tun->set_features) | (features & ~TUN_USER_FEATURES);
464 #ifdef CONFIG_NET_POLL_CONTROLLER
465 static void tun_poll_controller(struct net_device *dev)
468 * Tun only receives frames when:
469 * 1) the char device endpoint gets data from user space
470 * 2) the tun socket gets a sendmsg call from user space
471 * Since both of those are syncronous operations, we are guaranteed
472 * never to have pending data when we poll for it
473 * so theres nothing to do here but return.
474 * We need this though so netpoll recognizes us as an interface that
475 * supports polling, which enables bridge devices in virt setups to
476 * still use netconsole
481 static const struct net_device_ops tun_netdev_ops = {
482 .ndo_uninit = tun_net_uninit,
483 .ndo_open = tun_net_open,
484 .ndo_stop = tun_net_close,
485 .ndo_start_xmit = tun_net_xmit,
486 .ndo_change_mtu = tun_net_change_mtu,
487 .ndo_fix_features = tun_net_fix_features,
488 #ifdef CONFIG_NET_POLL_CONTROLLER
489 .ndo_poll_controller = tun_poll_controller,
493 static const struct net_device_ops tap_netdev_ops = {
494 .ndo_uninit = tun_net_uninit,
495 .ndo_open = tun_net_open,
496 .ndo_stop = tun_net_close,
497 .ndo_start_xmit = tun_net_xmit,
498 .ndo_change_mtu = tun_net_change_mtu,
499 .ndo_fix_features = tun_net_fix_features,
500 .ndo_set_rx_mode = tun_net_mclist,
501 .ndo_set_mac_address = eth_mac_addr,
502 .ndo_validate_addr = eth_validate_addr,
503 #ifdef CONFIG_NET_POLL_CONTROLLER
504 .ndo_poll_controller = tun_poll_controller,
508 /* Initialize net device. */
509 static void tun_net_init(struct net_device *dev)
511 struct tun_struct *tun = netdev_priv(dev);
513 switch (tun->flags & TUN_TYPE_MASK) {
515 dev->netdev_ops = &tun_netdev_ops;
517 /* Point-to-Point TUN Device */
518 dev->hard_header_len = 0;
522 /* Zero header length */
523 dev->type = ARPHRD_NONE;
524 dev->flags = IFF_POINTOPOINT | IFF_NOARP | IFF_MULTICAST;
525 dev->tx_queue_len = TUN_READQ_SIZE; /* We prefer our own queue length */
529 dev->netdev_ops = &tap_netdev_ops;
530 /* Ethernet TAP Device */
532 dev->priv_flags &= ~IFF_TX_SKB_SHARING;
534 eth_hw_addr_random(dev);
536 dev->tx_queue_len = TUN_READQ_SIZE; /* We prefer our own queue length */
541 /* Character device part */
544 static unsigned int tun_chr_poll(struct file *file, poll_table * wait)
546 struct tun_file *tfile = file->private_data;
547 struct tun_struct *tun = __tun_get(tfile);
549 unsigned int mask = 0;
556 tun_debug(KERN_INFO, tun, "tun_chr_poll\n");
558 poll_wait(file, &tun->wq.wait, wait);
560 if (!skb_queue_empty(&sk->sk_receive_queue))
561 mask |= POLLIN | POLLRDNORM;
563 if (sock_writeable(sk) ||
564 (!test_and_set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags) &&
566 mask |= POLLOUT | POLLWRNORM;
568 if (tun->dev->reg_state != NETREG_REGISTERED)
575 /* prepad is the amount to reserve at front. len is length after that.
576 * linear is a hint as to how much to copy (usually headers). */
577 static struct sk_buff *tun_alloc_skb(struct tun_struct *tun,
578 size_t prepad, size_t len,
579 size_t linear, int noblock)
581 struct sock *sk = tun->socket.sk;
585 sock_update_classid(sk);
587 /* Under a page? Don't bother with paged skb. */
588 if (prepad + len < PAGE_SIZE || !linear)
591 skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
596 skb_reserve(skb, prepad);
597 skb_put(skb, linear);
598 skb->data_len = len - linear;
599 skb->len += len - linear;
604 /* Get packet from user space buffer */
605 static ssize_t tun_get_user(struct tun_struct *tun,
606 const struct iovec *iv, size_t count,
609 struct tun_pi pi = { 0, cpu_to_be16(ETH_P_IP) };
611 size_t len = count, align = NET_SKB_PAD;
612 struct virtio_net_hdr gso = { 0 };
615 if (!(tun->flags & TUN_NO_PI)) {
616 if ((len -= sizeof(pi)) > count)
619 if (memcpy_fromiovecend((void *)&pi, iv, 0, sizeof(pi)))
621 offset += sizeof(pi);
624 if (tun->flags & TUN_VNET_HDR) {
625 if ((len -= tun->vnet_hdr_sz) > count)
628 if (memcpy_fromiovecend((void *)&gso, iv, offset, sizeof(gso)))
631 if ((gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
632 gso.csum_start + gso.csum_offset + 2 > gso.hdr_len)
633 gso.hdr_len = gso.csum_start + gso.csum_offset + 2;
635 if (gso.hdr_len > len)
637 offset += tun->vnet_hdr_sz;
640 if ((tun->flags & TUN_TYPE_MASK) == TUN_TAP_DEV) {
641 align += NET_IP_ALIGN;
642 if (unlikely(len < ETH_HLEN ||
643 (gso.hdr_len && gso.hdr_len < ETH_HLEN)))
647 skb = tun_alloc_skb(tun, align, len, gso.hdr_len, noblock);
649 if (PTR_ERR(skb) != -EAGAIN)
650 tun->dev->stats.rx_dropped++;
654 if (skb_copy_datagram_from_iovec(skb, 0, iv, offset, len)) {
655 tun->dev->stats.rx_dropped++;
660 if (gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
661 if (!skb_partial_csum_set(skb, gso.csum_start,
663 tun->dev->stats.rx_frame_errors++;
669 switch (tun->flags & TUN_TYPE_MASK) {
671 if (tun->flags & TUN_NO_PI) {
672 switch (skb->data[0] & 0xf0) {
674 pi.proto = htons(ETH_P_IP);
677 pi.proto = htons(ETH_P_IPV6);
680 tun->dev->stats.rx_dropped++;
686 skb_reset_mac_header(skb);
687 skb->protocol = pi.proto;
691 skb->protocol = eth_type_trans(skb, tun->dev);
695 if (gso.gso_type != VIRTIO_NET_HDR_GSO_NONE) {
697 switch (gso.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
698 case VIRTIO_NET_HDR_GSO_TCPV4:
699 skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
701 case VIRTIO_NET_HDR_GSO_TCPV6:
702 skb_shinfo(skb)->gso_type = SKB_GSO_TCPV6;
704 case VIRTIO_NET_HDR_GSO_UDP:
705 skb_shinfo(skb)->gso_type = SKB_GSO_UDP;
708 tun->dev->stats.rx_frame_errors++;
713 if (gso.gso_type & VIRTIO_NET_HDR_GSO_ECN)
714 skb_shinfo(skb)->gso_type |= SKB_GSO_TCP_ECN;
716 skb_shinfo(skb)->gso_size = gso.gso_size;
717 if (skb_shinfo(skb)->gso_size == 0) {
718 tun->dev->stats.rx_frame_errors++;
723 /* Header must be checked, and gso_segs computed. */
724 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
725 skb_shinfo(skb)->gso_segs = 0;
730 tun->dev->stats.rx_packets++;
731 tun->dev->stats.rx_bytes += len;
736 static ssize_t tun_chr_aio_write(struct kiocb *iocb, const struct iovec *iv,
737 unsigned long count, loff_t pos)
739 struct file *file = iocb->ki_filp;
740 struct tun_struct *tun = tun_get(file);
746 tun_debug(KERN_INFO, tun, "tun_chr_write %ld\n", count);
748 result = tun_get_user(tun, iv, iov_length(iv, count),
749 file->f_flags & O_NONBLOCK);
755 /* Put packet to the user space buffer */
756 static ssize_t tun_put_user(struct tun_struct *tun,
758 const struct iovec *iv, int len)
760 struct tun_pi pi = { 0, skb->protocol };
763 if (!(tun->flags & TUN_NO_PI)) {
764 if ((len -= sizeof(pi)) < 0)
767 if (len < skb->len) {
768 /* Packet will be striped */
769 pi.flags |= TUN_PKT_STRIP;
772 if (memcpy_toiovecend(iv, (void *) &pi, 0, sizeof(pi)))
777 if (tun->flags & TUN_VNET_HDR) {
778 struct virtio_net_hdr gso = { 0 }; /* no info leak */
779 if ((len -= tun->vnet_hdr_sz) < 0)
782 if (skb_is_gso(skb)) {
783 struct skb_shared_info *sinfo = skb_shinfo(skb);
785 /* This is a hint as to how much should be linear. */
786 gso.hdr_len = skb_headlen(skb);
787 gso.gso_size = sinfo->gso_size;
788 if (sinfo->gso_type & SKB_GSO_TCPV4)
789 gso.gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
790 else if (sinfo->gso_type & SKB_GSO_TCPV6)
791 gso.gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
792 else if (sinfo->gso_type & SKB_GSO_UDP)
793 gso.gso_type = VIRTIO_NET_HDR_GSO_UDP;
795 pr_err("unexpected GSO type: "
796 "0x%x, gso_size %d, hdr_len %d\n",
797 sinfo->gso_type, gso.gso_size,
799 print_hex_dump(KERN_ERR, "tun: ",
802 min((int)gso.hdr_len, 64), true);
806 if (sinfo->gso_type & SKB_GSO_TCP_ECN)
807 gso.gso_type |= VIRTIO_NET_HDR_GSO_ECN;
809 gso.gso_type = VIRTIO_NET_HDR_GSO_NONE;
811 if (skb->ip_summed == CHECKSUM_PARTIAL) {
812 gso.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
813 gso.csum_start = skb_checksum_start_offset(skb);
814 gso.csum_offset = skb->csum_offset;
815 } else if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
816 gso.flags = VIRTIO_NET_HDR_F_DATA_VALID;
817 } /* else everything is zero */
819 if (unlikely(memcpy_toiovecend(iv, (void *)&gso, total,
822 total += tun->vnet_hdr_sz;
825 len = min_t(int, skb->len, len);
827 skb_copy_datagram_const_iovec(skb, 0, iv, total, len);
830 tun->dev->stats.tx_packets++;
831 tun->dev->stats.tx_bytes += len;
836 static ssize_t tun_do_read(struct tun_struct *tun,
837 struct kiocb *iocb, const struct iovec *iv,
838 ssize_t len, int noblock)
840 DECLARE_WAITQUEUE(wait, current);
844 tun_debug(KERN_INFO, tun, "tun_chr_read\n");
846 if (unlikely(!noblock))
847 add_wait_queue(&tun->wq.wait, &wait);
849 current->state = TASK_INTERRUPTIBLE;
851 /* Read frames from the queue */
852 if (!(skb=skb_dequeue(&tun->socket.sk->sk_receive_queue))) {
857 if (signal_pending(current)) {
861 if (tun->dev->reg_state != NETREG_REGISTERED) {
866 /* Nothing to read, let's sleep */
870 netif_wake_queue(tun->dev);
872 ret = tun_put_user(tun, skb, iv, len);
877 current->state = TASK_RUNNING;
878 if (unlikely(!noblock))
879 remove_wait_queue(&tun->wq.wait, &wait);
884 static ssize_t tun_chr_aio_read(struct kiocb *iocb, const struct iovec *iv,
885 unsigned long count, loff_t pos)
887 struct file *file = iocb->ki_filp;
888 struct tun_file *tfile = file->private_data;
889 struct tun_struct *tun = __tun_get(tfile);
894 len = iov_length(iv, count);
900 ret = tun_do_read(tun, iocb, iv, len, file->f_flags & O_NONBLOCK);
901 ret = min_t(ssize_t, ret, len);
907 static void tun_setup(struct net_device *dev)
909 struct tun_struct *tun = netdev_priv(dev);
914 dev->ethtool_ops = &tun_ethtool_ops;
915 dev->destructor = tun_free_netdev;
918 /* Trivial set of netlink ops to allow deleting tun or tap
919 * device with netlink.
921 static int tun_validate(struct nlattr *tb[], struct nlattr *data[])
926 static struct rtnl_link_ops tun_link_ops __read_mostly = {
928 .priv_size = sizeof(struct tun_struct),
930 .validate = tun_validate,
933 static void tun_sock_write_space(struct sock *sk)
935 struct tun_struct *tun;
936 wait_queue_head_t *wqueue;
938 if (!sock_writeable(sk))
941 if (!test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags))
944 wqueue = sk_sleep(sk);
945 if (wqueue && waitqueue_active(wqueue))
946 wake_up_interruptible_sync_poll(wqueue, POLLOUT |
947 POLLWRNORM | POLLWRBAND);
949 tun = tun_sk(sk)->tun;
950 kill_fasync(&tun->fasync, SIGIO, POLL_OUT);
953 static void tun_sock_destruct(struct sock *sk)
955 free_netdev(tun_sk(sk)->tun->dev);
958 static int tun_sendmsg(struct kiocb *iocb, struct socket *sock,
959 struct msghdr *m, size_t total_len)
961 struct tun_struct *tun = container_of(sock, struct tun_struct, socket);
962 return tun_get_user(tun, m->msg_iov, total_len,
963 m->msg_flags & MSG_DONTWAIT);
966 static int tun_recvmsg(struct kiocb *iocb, struct socket *sock,
967 struct msghdr *m, size_t total_len,
970 struct tun_struct *tun = container_of(sock, struct tun_struct, socket);
972 if (flags & ~(MSG_DONTWAIT|MSG_TRUNC))
974 ret = tun_do_read(tun, iocb, m->msg_iov, total_len,
975 flags & MSG_DONTWAIT);
976 if (ret > total_len) {
977 m->msg_flags |= MSG_TRUNC;
978 ret = flags & MSG_TRUNC ? ret : total_len;
983 static int tun_release(struct socket *sock)
990 /* Ops structure to mimic raw sockets with tun */
991 static const struct proto_ops tun_socket_ops = {
992 .sendmsg = tun_sendmsg,
993 .recvmsg = tun_recvmsg,
994 .release = tun_release,
997 static struct proto tun_proto = {
999 .owner = THIS_MODULE,
1000 .obj_size = sizeof(struct tun_sock),
1003 static int tun_flags(struct tun_struct *tun)
1007 if (tun->flags & TUN_TUN_DEV)
1012 if (tun->flags & TUN_NO_PI)
1015 if (tun->flags & TUN_ONE_QUEUE)
1016 flags |= IFF_ONE_QUEUE;
1018 if (tun->flags & TUN_VNET_HDR)
1019 flags |= IFF_VNET_HDR;
1024 static ssize_t tun_show_flags(struct device *dev, struct device_attribute *attr,
1027 struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1028 return sprintf(buf, "0x%x\n", tun_flags(tun));
1031 static ssize_t tun_show_owner(struct device *dev, struct device_attribute *attr,
1034 struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1035 return sprintf(buf, "%d\n", tun->owner);
1038 static ssize_t tun_show_group(struct device *dev, struct device_attribute *attr,
1041 struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1042 return sprintf(buf, "%d\n", tun->group);
1045 static DEVICE_ATTR(tun_flags, 0444, tun_show_flags, NULL);
1046 static DEVICE_ATTR(owner, 0444, tun_show_owner, NULL);
1047 static DEVICE_ATTR(group, 0444, tun_show_group, NULL);
1049 static int tun_set_iff(struct net *net, struct file *file, struct ifreq *ifr)
1052 struct tun_struct *tun;
1053 struct net_device *dev;
1056 dev = __dev_get_by_name(net, ifr->ifr_name);
1058 const struct cred *cred = current_cred();
1060 if (ifr->ifr_flags & IFF_TUN_EXCL)
1062 if ((ifr->ifr_flags & IFF_TUN) && dev->netdev_ops == &tun_netdev_ops)
1063 tun = netdev_priv(dev);
1064 else if ((ifr->ifr_flags & IFF_TAP) && dev->netdev_ops == &tap_netdev_ops)
1065 tun = netdev_priv(dev);
1069 if (((tun->owner != -1 && cred->euid != tun->owner) ||
1070 (tun->group != -1 && !in_egroup_p(tun->group))) &&
1071 !capable(CAP_NET_ADMIN))
1073 err = security_tun_dev_attach(tun->socket.sk);
1077 err = tun_attach(tun, file);
1083 unsigned long flags = 0;
1085 if (!capable(CAP_NET_ADMIN))
1087 err = security_tun_dev_create();
1092 if (ifr->ifr_flags & IFF_TUN) {
1094 flags |= TUN_TUN_DEV;
1096 } else if (ifr->ifr_flags & IFF_TAP) {
1098 flags |= TUN_TAP_DEV;
1104 name = ifr->ifr_name;
1106 dev = alloc_netdev(sizeof(struct tun_struct), name,
1111 dev_net_set(dev, net);
1112 dev->rtnl_link_ops = &tun_link_ops;
1114 tun = netdev_priv(dev);
1117 tun->txflt.count = 0;
1118 tun->vnet_hdr_sz = sizeof(struct virtio_net_hdr);
1121 sk = sk_alloc(&init_net, AF_UNSPEC, GFP_KERNEL, &tun_proto);
1125 sk_change_net(sk, net);
1126 tun->socket.wq = &tun->wq;
1127 init_waitqueue_head(&tun->wq.wait);
1128 tun->socket.ops = &tun_socket_ops;
1129 sock_init_data(&tun->socket, sk);
1130 sk->sk_write_space = tun_sock_write_space;
1131 sk->sk_sndbuf = INT_MAX;
1133 tun_sk(sk)->tun = tun;
1135 security_tun_dev_post_create(sk);
1139 dev->hw_features = NETIF_F_SG | NETIF_F_FRAGLIST |
1141 dev->features = dev->hw_features;
1143 err = register_netdevice(tun->dev);
1147 if (device_create_file(&tun->dev->dev, &dev_attr_tun_flags) ||
1148 device_create_file(&tun->dev->dev, &dev_attr_owner) ||
1149 device_create_file(&tun->dev->dev, &dev_attr_group))
1150 pr_err("Failed to create tun sysfs files\n");
1152 sk->sk_destruct = tun_sock_destruct;
1154 err = tun_attach(tun, file);
1159 tun_debug(KERN_INFO, tun, "tun_set_iff\n");
1161 if (ifr->ifr_flags & IFF_NO_PI)
1162 tun->flags |= TUN_NO_PI;
1164 tun->flags &= ~TUN_NO_PI;
1166 if (ifr->ifr_flags & IFF_ONE_QUEUE)
1167 tun->flags |= TUN_ONE_QUEUE;
1169 tun->flags &= ~TUN_ONE_QUEUE;
1171 if (ifr->ifr_flags & IFF_VNET_HDR)
1172 tun->flags |= TUN_VNET_HDR;
1174 tun->flags &= ~TUN_VNET_HDR;
1176 /* Make sure persistent devices do not get stuck in
1179 if (netif_running(tun->dev))
1180 netif_wake_queue(tun->dev);
1182 strcpy(ifr->ifr_name, tun->dev->name);
1186 tun_free_netdev(dev);
1193 static int tun_get_iff(struct net *net, struct tun_struct *tun,
1196 tun_debug(KERN_INFO, tun, "tun_get_iff\n");
1198 strcpy(ifr->ifr_name, tun->dev->name);
1200 ifr->ifr_flags = tun_flags(tun);
1205 /* This is like a cut-down ethtool ops, except done via tun fd so no
1206 * privs required. */
1207 static int set_offload(struct tun_struct *tun, unsigned long arg)
1209 netdev_features_t features = 0;
1211 if (arg & TUN_F_CSUM) {
1212 features |= NETIF_F_HW_CSUM;
1215 if (arg & (TUN_F_TSO4|TUN_F_TSO6)) {
1216 if (arg & TUN_F_TSO_ECN) {
1217 features |= NETIF_F_TSO_ECN;
1218 arg &= ~TUN_F_TSO_ECN;
1220 if (arg & TUN_F_TSO4)
1221 features |= NETIF_F_TSO;
1222 if (arg & TUN_F_TSO6)
1223 features |= NETIF_F_TSO6;
1224 arg &= ~(TUN_F_TSO4|TUN_F_TSO6);
1227 if (arg & TUN_F_UFO) {
1228 features |= NETIF_F_UFO;
1233 /* This gives the user a way to test for new features in future by
1234 * trying to set them. */
1238 tun->set_features = features;
1239 netdev_update_features(tun->dev);
1244 static long __tun_chr_ioctl(struct file *file, unsigned int cmd,
1245 unsigned long arg, int ifreq_len)
1247 struct tun_file *tfile = file->private_data;
1248 struct tun_struct *tun;
1249 void __user* argp = (void __user*)arg;
1250 struct sock_fprog fprog;
1256 if (cmd == TUNSETIFF || _IOC_TYPE(cmd) == 0x89)
1257 if (copy_from_user(&ifr, argp, ifreq_len))
1260 if (cmd == TUNGETFEATURES) {
1261 /* Currently this just means: "what IFF flags are valid?".
1262 * This is needed because we never checked for invalid flags on
1264 return put_user(IFF_TUN | IFF_TAP | IFF_NO_PI | IFF_ONE_QUEUE |
1266 (unsigned int __user*)argp);
1271 tun = __tun_get(tfile);
1272 if (cmd == TUNSETIFF && !tun) {
1273 ifr.ifr_name[IFNAMSIZ-1] = '\0';
1275 ret = tun_set_iff(tfile->net, file, &ifr);
1280 if (copy_to_user(argp, &ifr, ifreq_len))
1289 tun_debug(KERN_INFO, tun, "tun_chr_ioctl cmd %d\n", cmd);
1294 ret = tun_get_iff(current->nsproxy->net_ns, tun, &ifr);
1298 if (copy_to_user(argp, &ifr, ifreq_len))
1303 /* Disable/Enable checksum */
1305 /* [unimplemented] */
1306 tun_debug(KERN_INFO, tun, "ignored: set checksum %s\n",
1307 arg ? "disabled" : "enabled");
1311 /* Disable/Enable persist mode */
1313 tun->flags |= TUN_PERSIST;
1315 tun->flags &= ~TUN_PERSIST;
1317 tun_debug(KERN_INFO, tun, "persist %s\n",
1318 arg ? "enabled" : "disabled");
1322 /* Set owner of the device */
1323 tun->owner = (uid_t) arg;
1325 tun_debug(KERN_INFO, tun, "owner set to %d\n", tun->owner);
1329 /* Set group of the device */
1330 tun->group= (gid_t) arg;
1332 tun_debug(KERN_INFO, tun, "group set to %d\n", tun->group);
1336 /* Only allow setting the type when the interface is down */
1337 if (tun->dev->flags & IFF_UP) {
1338 tun_debug(KERN_INFO, tun,
1339 "Linktype set failed because interface is up\n");
1342 tun->dev->type = (int) arg;
1343 tun_debug(KERN_INFO, tun, "linktype set to %d\n",
1355 ret = set_offload(tun, arg);
1358 case TUNSETTXFILTER:
1359 /* Can be set only for TAPs */
1361 if ((tun->flags & TUN_TYPE_MASK) != TUN_TAP_DEV)
1363 ret = update_filter(&tun->txflt, (void __user *)arg);
1367 /* Get hw address */
1368 memcpy(ifr.ifr_hwaddr.sa_data, tun->dev->dev_addr, ETH_ALEN);
1369 ifr.ifr_hwaddr.sa_family = tun->dev->type;
1370 if (copy_to_user(argp, &ifr, ifreq_len))
1375 /* Set hw address */
1376 tun_debug(KERN_DEBUG, tun, "set hw address: %pM\n",
1377 ifr.ifr_hwaddr.sa_data);
1379 ret = dev_set_mac_address(tun->dev, &ifr.ifr_hwaddr);
1383 sndbuf = tun->socket.sk->sk_sndbuf;
1384 if (copy_to_user(argp, &sndbuf, sizeof(sndbuf)))
1389 if (copy_from_user(&sndbuf, argp, sizeof(sndbuf))) {
1394 tun->socket.sk->sk_sndbuf = sndbuf;
1397 case TUNGETVNETHDRSZ:
1398 vnet_hdr_sz = tun->vnet_hdr_sz;
1399 if (copy_to_user(argp, &vnet_hdr_sz, sizeof(vnet_hdr_sz)))
1403 case TUNSETVNETHDRSZ:
1404 if (copy_from_user(&vnet_hdr_sz, argp, sizeof(vnet_hdr_sz))) {
1408 if (vnet_hdr_sz < (int)sizeof(struct virtio_net_hdr)) {
1413 tun->vnet_hdr_sz = vnet_hdr_sz;
1416 case TUNATTACHFILTER:
1417 /* Can be set only for TAPs */
1419 if ((tun->flags & TUN_TYPE_MASK) != TUN_TAP_DEV)
1422 if (copy_from_user(&fprog, argp, sizeof(fprog)))
1425 ret = sk_attach_filter(&fprog, tun->socket.sk);
1428 case TUNDETACHFILTER:
1429 /* Can be set only for TAPs */
1431 if ((tun->flags & TUN_TYPE_MASK) != TUN_TAP_DEV)
1433 ret = sk_detach_filter(tun->socket.sk);
1448 static long tun_chr_ioctl(struct file *file,
1449 unsigned int cmd, unsigned long arg)
1451 return __tun_chr_ioctl(file, cmd, arg, sizeof (struct ifreq));
1454 #ifdef CONFIG_COMPAT
1455 static long tun_chr_compat_ioctl(struct file *file,
1456 unsigned int cmd, unsigned long arg)
1461 case TUNSETTXFILTER:
1466 arg = (unsigned long)compat_ptr(arg);
1469 arg = (compat_ulong_t)arg;
1474 * compat_ifreq is shorter than ifreq, so we must not access beyond
1475 * the end of that structure. All fields that are used in this
1476 * driver are compatible though, we don't need to convert the
1479 return __tun_chr_ioctl(file, cmd, arg, sizeof(struct compat_ifreq));
1481 #endif /* CONFIG_COMPAT */
1483 static int tun_chr_fasync(int fd, struct file *file, int on)
1485 struct tun_struct *tun = tun_get(file);
1491 tun_debug(KERN_INFO, tun, "tun_chr_fasync %d\n", on);
1493 if ((ret = fasync_helper(fd, file, on, &tun->fasync)) < 0)
1497 ret = __f_setown(file, task_pid(current), PIDTYPE_PID, 0);
1500 tun->flags |= TUN_FASYNC;
1502 tun->flags &= ~TUN_FASYNC;
1509 static int tun_chr_open(struct inode *inode, struct file * file)
1511 struct tun_file *tfile;
1513 DBG1(KERN_INFO, "tunX: tun_chr_open\n");
1515 tfile = kmalloc(sizeof(*tfile), GFP_KERNEL);
1518 atomic_set(&tfile->count, 0);
1520 tfile->net = get_net(current->nsproxy->net_ns);
1521 file->private_data = tfile;
1525 static int tun_chr_close(struct inode *inode, struct file *file)
1527 struct tun_file *tfile = file->private_data;
1528 struct tun_struct *tun;
1530 tun = __tun_get(tfile);
1532 struct net_device *dev = tun->dev;
1534 tun_debug(KERN_INFO, tun, "tun_chr_close\n");
1538 /* If desirable, unregister the netdevice. */
1539 if (!(tun->flags & TUN_PERSIST)) {
1541 if (dev->reg_state == NETREG_REGISTERED)
1542 unregister_netdevice(dev);
1549 sock_put(tun->socket.sk);
1551 put_net(tfile->net);
1557 static const struct file_operations tun_fops = {
1558 .owner = THIS_MODULE,
1559 .llseek = no_llseek,
1560 .read = do_sync_read,
1561 .aio_read = tun_chr_aio_read,
1562 .write = do_sync_write,
1563 .aio_write = tun_chr_aio_write,
1564 .poll = tun_chr_poll,
1565 .unlocked_ioctl = tun_chr_ioctl,
1566 #ifdef CONFIG_COMPAT
1567 .compat_ioctl = tun_chr_compat_ioctl,
1569 .open = tun_chr_open,
1570 .release = tun_chr_close,
1571 .fasync = tun_chr_fasync
1574 static struct miscdevice tun_miscdev = {
1577 .nodename = "net/tun",
1581 /* ethtool interface */
1583 static int tun_get_settings(struct net_device *dev, struct ethtool_cmd *cmd)
1586 cmd->advertising = 0;
1587 ethtool_cmd_speed_set(cmd, SPEED_10);
1588 cmd->duplex = DUPLEX_FULL;
1589 cmd->port = PORT_TP;
1590 cmd->phy_address = 0;
1591 cmd->transceiver = XCVR_INTERNAL;
1592 cmd->autoneg = AUTONEG_DISABLE;
1598 static void tun_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
1600 struct tun_struct *tun = netdev_priv(dev);
1602 strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
1603 strlcpy(info->version, DRV_VERSION, sizeof(info->version));
1605 switch (tun->flags & TUN_TYPE_MASK) {
1607 strlcpy(info->bus_info, "tun", sizeof(info->bus_info));
1610 strlcpy(info->bus_info, "tap", sizeof(info->bus_info));
1615 static u32 tun_get_msglevel(struct net_device *dev)
1618 struct tun_struct *tun = netdev_priv(dev);
1625 static void tun_set_msglevel(struct net_device *dev, u32 value)
1628 struct tun_struct *tun = netdev_priv(dev);
1633 static const struct ethtool_ops tun_ethtool_ops = {
1634 .get_settings = tun_get_settings,
1635 .get_drvinfo = tun_get_drvinfo,
1636 .get_msglevel = tun_get_msglevel,
1637 .set_msglevel = tun_set_msglevel,
1638 .get_link = ethtool_op_get_link,
1642 static int __init tun_init(void)
1646 pr_info("%s, %s\n", DRV_DESCRIPTION, DRV_VERSION);
1647 pr_info("%s\n", DRV_COPYRIGHT);
1649 ret = rtnl_link_register(&tun_link_ops);
1651 pr_err("Can't register link_ops\n");
1655 ret = misc_register(&tun_miscdev);
1657 pr_err("Can't register misc device %d\n", TUN_MINOR);
1662 rtnl_link_unregister(&tun_link_ops);
1667 static void tun_cleanup(void)
1669 misc_deregister(&tun_miscdev);
1670 rtnl_link_unregister(&tun_link_ops);
1673 /* Get an underlying socket object from tun file. Returns error unless file is
1674 * attached to a device. The returned object works like a packet socket, it
1675 * can be used for sock_sendmsg/sock_recvmsg. The caller is responsible for
1676 * holding a reference to the file for as long as the socket is in use. */
1677 struct socket *tun_get_socket(struct file *file)
1679 struct tun_struct *tun;
1680 if (file->f_op != &tun_fops)
1681 return ERR_PTR(-EINVAL);
1682 tun = tun_get(file);
1684 return ERR_PTR(-EBADFD);
1686 return &tun->socket;
1688 EXPORT_SYMBOL_GPL(tun_get_socket);
1690 module_init(tun_init);
1691 module_exit(tun_cleanup);
1692 MODULE_DESCRIPTION(DRV_DESCRIPTION);
1693 MODULE_AUTHOR(DRV_COPYRIGHT);
1694 MODULE_LICENSE("GPL");
1695 MODULE_ALIAS_MISCDEV(TUN_MINOR);
1696 MODULE_ALIAS("devname:net/tun");