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 GOODCOPY_LEN 128
105 #define FLT_EXACT_COUNT 8
107 unsigned int count; /* Number of addrs. Zero means disabled */
108 u32 mask[2]; /* Mask of the hashed addrs */
109 unsigned char addr[FLT_EXACT_COUNT][ETH_ALEN];
112 /* A tun_file connects an open character device to a tuntap netdevice. It
113 * also contains all socket related strctures (except sock_fprog and tap_filter)
114 * to serve as one transmit queue for tuntap device. The sock_fprog and
115 * tap_filter were kept in tun_struct since they were used for filtering for the
116 * netdevice not for a specific queue (at least I didn't see the reqirement for
121 struct socket socket;
124 struct tun_struct *tun;
126 struct fasync_struct *fasync;
127 /* only used for fasnyc */
131 /* Since the socket were moved to tun_file, to preserve the behavior of persist
132 * device, socket fileter, sndbuf and vnet header size were restore when the
133 * file were attached to a persist device.
136 struct tun_file *tfile;
141 struct net_device *dev;
142 netdev_features_t set_features;
143 #define TUN_USER_FEATURES (NETIF_F_HW_CSUM|NETIF_F_TSO_ECN|NETIF_F_TSO| \
144 NETIF_F_TSO6|NETIF_F_UFO)
148 struct tap_filter txflt;
149 struct sock_fprog fprog;
150 /* protected by rtnl lock */
151 bool filter_attached;
157 static int tun_attach(struct tun_struct *tun, struct file *file)
159 struct tun_file *tfile = file->private_data;
164 netif_tx_lock_bh(tun->dev);
176 /* Re-attach filter when attaching to a persist device */
177 if (tun->filter_attached == true) {
178 err = sk_attach_filter(&tun->fprog, tfile->socket.sk);
183 tfile->socket.sk->sk_sndbuf = tun->sndbuf;
185 netif_carrier_on(tun->dev);
187 sock_hold(&tfile->sk);
188 atomic_inc(&tfile->count);
191 netif_tx_unlock_bh(tun->dev);
195 static void __tun_detach(struct tun_struct *tun)
197 struct tun_file *tfile = tun->tfile;
198 /* Detach from net device */
199 netif_tx_lock_bh(tun->dev);
200 netif_carrier_off(tun->dev);
203 netif_tx_unlock_bh(tun->dev);
205 /* Drop read queue */
206 skb_queue_purge(&tfile->socket.sk->sk_receive_queue);
208 /* Drop the extra count on the net device */
212 static void tun_detach(struct tun_struct *tun)
219 static struct tun_struct *__tun_get(struct tun_file *tfile)
221 struct tun_struct *tun = NULL;
223 if (atomic_inc_not_zero(&tfile->count))
229 static struct tun_struct *tun_get(struct file *file)
231 return __tun_get(file->private_data);
234 static void tun_put(struct tun_struct *tun)
236 struct tun_file *tfile = tun->tfile;
238 if (atomic_dec_and_test(&tfile->count))
239 tun_detach(tfile->tun);
243 static void addr_hash_set(u32 *mask, const u8 *addr)
245 int n = ether_crc(ETH_ALEN, addr) >> 26;
246 mask[n >> 5] |= (1 << (n & 31));
249 static unsigned int addr_hash_test(const u32 *mask, const u8 *addr)
251 int n = ether_crc(ETH_ALEN, addr) >> 26;
252 return mask[n >> 5] & (1 << (n & 31));
255 static int update_filter(struct tap_filter *filter, void __user *arg)
257 struct { u8 u[ETH_ALEN]; } *addr;
258 struct tun_filter uf;
259 int err, alen, n, nexact;
261 if (copy_from_user(&uf, arg, sizeof(uf)))
270 alen = ETH_ALEN * uf.count;
271 addr = kmalloc(alen, GFP_KERNEL);
275 if (copy_from_user(addr, arg + sizeof(uf), alen)) {
280 /* The filter is updated without holding any locks. Which is
281 * perfectly safe. We disable it first and in the worst
282 * case we'll accept a few undesired packets. */
286 /* Use first set of addresses as an exact filter */
287 for (n = 0; n < uf.count && n < FLT_EXACT_COUNT; n++)
288 memcpy(filter->addr[n], addr[n].u, ETH_ALEN);
292 /* Remaining multicast addresses are hashed,
293 * unicast will leave the filter disabled. */
294 memset(filter->mask, 0, sizeof(filter->mask));
295 for (; n < uf.count; n++) {
296 if (!is_multicast_ether_addr(addr[n].u)) {
297 err = 0; /* no filter */
300 addr_hash_set(filter->mask, addr[n].u);
303 /* For ALLMULTI just set the mask to all ones.
304 * This overrides the mask populated above. */
305 if ((uf.flags & TUN_FLT_ALLMULTI))
306 memset(filter->mask, ~0, sizeof(filter->mask));
308 /* Now enable the filter */
310 filter->count = nexact;
312 /* Return the number of exact filters */
320 /* Returns: 0 - drop, !=0 - accept */
321 static int run_filter(struct tap_filter *filter, const struct sk_buff *skb)
323 /* Cannot use eth_hdr(skb) here because skb_mac_hdr() is incorrect
325 struct ethhdr *eh = (struct ethhdr *) skb->data;
329 for (i = 0; i < filter->count; i++)
330 if (ether_addr_equal(eh->h_dest, filter->addr[i]))
333 /* Inexact match (multicast only) */
334 if (is_multicast_ether_addr(eh->h_dest))
335 return addr_hash_test(filter->mask, eh->h_dest);
341 * Checks whether the packet is accepted or not.
342 * Returns: 0 - drop, !=0 - accept
344 static int check_filter(struct tap_filter *filter, const struct sk_buff *skb)
349 return run_filter(filter, skb);
352 /* Network device part of the driver */
354 static const struct ethtool_ops tun_ethtool_ops;
356 /* Net device detach from fd. */
357 static void tun_net_uninit(struct net_device *dev)
359 struct tun_struct *tun = netdev_priv(dev);
360 struct tun_file *tfile = tun->tfile;
362 /* Inform the methods they need to stop using the dev.
365 wake_up_all(&tfile->wq.wait);
366 if (atomic_dec_and_test(&tfile->count))
371 /* Net device open. */
372 static int tun_net_open(struct net_device *dev)
374 netif_start_queue(dev);
378 /* Net device close. */
379 static int tun_net_close(struct net_device *dev)
381 netif_stop_queue(dev);
385 /* Net device start xmit */
386 static netdev_tx_t tun_net_xmit(struct sk_buff *skb, struct net_device *dev)
388 struct tun_struct *tun = netdev_priv(dev);
389 struct tun_file *tfile = tun->tfile;
391 tun_debug(KERN_INFO, tun, "tun_net_xmit %d\n", skb->len);
393 /* Drop packet if interface is not attached */
397 /* Drop if the filter does not like it.
398 * This is a noop if the filter is disabled.
399 * Filter can be enabled only for the TAP devices. */
400 if (!check_filter(&tun->txflt, skb))
403 if (tfile->socket.sk->sk_filter &&
404 sk_filter(tfile->socket.sk, skb))
407 if (skb_queue_len(&tfile->socket.sk->sk_receive_queue)
408 >= dev->tx_queue_len) {
409 if (!(tun->flags & TUN_ONE_QUEUE)) {
410 /* Normal queueing mode. */
411 /* Packet scheduler handles dropping of further packets. */
412 netif_stop_queue(dev);
414 /* We won't see all dropped packets individually, so overrun
415 * error is more appropriate. */
416 dev->stats.tx_fifo_errors++;
418 /* Single queue mode.
419 * Driver handles dropping of all packets itself. */
424 /* Orphan the skb - required as we might hang on to it
425 * for indefinite time. */
426 if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC)))
431 skb_queue_tail(&tfile->socket.sk->sk_receive_queue, skb);
433 /* Notify and wake up reader process */
434 if (tfile->flags & TUN_FASYNC)
435 kill_fasync(&tfile->fasync, SIGIO, POLL_IN);
436 wake_up_interruptible_poll(&tfile->wq.wait, POLLIN |
437 POLLRDNORM | POLLRDBAND);
441 dev->stats.tx_dropped++;
446 static void tun_net_mclist(struct net_device *dev)
449 * This callback is supposed to deal with mc filter in
450 * _rx_ path and has nothing to do with the _tx_ path.
451 * In rx path we always accept everything userspace gives us.
456 #define MAX_MTU 65535
459 tun_net_change_mtu(struct net_device *dev, int new_mtu)
461 if (new_mtu < MIN_MTU || new_mtu + dev->hard_header_len > MAX_MTU)
467 static netdev_features_t tun_net_fix_features(struct net_device *dev,
468 netdev_features_t features)
470 struct tun_struct *tun = netdev_priv(dev);
472 return (features & tun->set_features) | (features & ~TUN_USER_FEATURES);
474 #ifdef CONFIG_NET_POLL_CONTROLLER
475 static void tun_poll_controller(struct net_device *dev)
478 * Tun only receives frames when:
479 * 1) the char device endpoint gets data from user space
480 * 2) the tun socket gets a sendmsg call from user space
481 * Since both of those are syncronous operations, we are guaranteed
482 * never to have pending data when we poll for it
483 * so theres nothing to do here but return.
484 * We need this though so netpoll recognizes us as an interface that
485 * supports polling, which enables bridge devices in virt setups to
486 * still use netconsole
491 static const struct net_device_ops tun_netdev_ops = {
492 .ndo_uninit = tun_net_uninit,
493 .ndo_open = tun_net_open,
494 .ndo_stop = tun_net_close,
495 .ndo_start_xmit = tun_net_xmit,
496 .ndo_change_mtu = tun_net_change_mtu,
497 .ndo_fix_features = tun_net_fix_features,
498 #ifdef CONFIG_NET_POLL_CONTROLLER
499 .ndo_poll_controller = tun_poll_controller,
503 static const struct net_device_ops tap_netdev_ops = {
504 .ndo_uninit = tun_net_uninit,
505 .ndo_open = tun_net_open,
506 .ndo_stop = tun_net_close,
507 .ndo_start_xmit = tun_net_xmit,
508 .ndo_change_mtu = tun_net_change_mtu,
509 .ndo_fix_features = tun_net_fix_features,
510 .ndo_set_rx_mode = tun_net_mclist,
511 .ndo_set_mac_address = eth_mac_addr,
512 .ndo_validate_addr = eth_validate_addr,
513 #ifdef CONFIG_NET_POLL_CONTROLLER
514 .ndo_poll_controller = tun_poll_controller,
518 /* Initialize net device. */
519 static void tun_net_init(struct net_device *dev)
521 struct tun_struct *tun = netdev_priv(dev);
523 switch (tun->flags & TUN_TYPE_MASK) {
525 dev->netdev_ops = &tun_netdev_ops;
527 /* Point-to-Point TUN Device */
528 dev->hard_header_len = 0;
532 /* Zero header length */
533 dev->type = ARPHRD_NONE;
534 dev->flags = IFF_POINTOPOINT | IFF_NOARP | IFF_MULTICAST;
535 dev->tx_queue_len = TUN_READQ_SIZE; /* We prefer our own queue length */
539 dev->netdev_ops = &tap_netdev_ops;
540 /* Ethernet TAP Device */
542 dev->priv_flags &= ~IFF_TX_SKB_SHARING;
544 eth_hw_addr_random(dev);
546 dev->tx_queue_len = TUN_READQ_SIZE; /* We prefer our own queue length */
551 /* Character device part */
554 static unsigned int tun_chr_poll(struct file *file, poll_table * wait)
556 struct tun_file *tfile = file->private_data;
557 struct tun_struct *tun = __tun_get(tfile);
559 unsigned int mask = 0;
564 sk = tfile->socket.sk;
566 tun_debug(KERN_INFO, tun, "tun_chr_poll\n");
568 poll_wait(file, &tfile->wq.wait, wait);
570 if (!skb_queue_empty(&sk->sk_receive_queue))
571 mask |= POLLIN | POLLRDNORM;
573 if (sock_writeable(sk) ||
574 (!test_and_set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags) &&
576 mask |= POLLOUT | POLLWRNORM;
578 if (tun->dev->reg_state != NETREG_REGISTERED)
585 /* prepad is the amount to reserve at front. len is length after that.
586 * linear is a hint as to how much to copy (usually headers). */
587 static struct sk_buff *tun_alloc_skb(struct tun_file *tfile,
588 size_t prepad, size_t len,
589 size_t linear, int noblock)
591 struct sock *sk = tfile->socket.sk;
595 /* Under a page? Don't bother with paged skb. */
596 if (prepad + len < PAGE_SIZE || !linear)
599 skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
604 skb_reserve(skb, prepad);
605 skb_put(skb, linear);
606 skb->data_len = len - linear;
607 skb->len += len - linear;
612 /* set skb frags from iovec, this can move to core network code for reuse */
613 static int zerocopy_sg_from_iovec(struct sk_buff *skb, const struct iovec *from,
614 int offset, size_t count)
616 int len = iov_length(from, count) - offset;
617 int copy = skb_headlen(skb);
618 int size, offset1 = 0;
621 /* Skip over from offset */
622 while (count && (offset >= from->iov_len)) {
623 offset -= from->iov_len;
628 /* copy up to skb headlen */
629 while (count && (copy > 0)) {
630 size = min_t(unsigned int, copy, from->iov_len - offset);
631 if (copy_from_user(skb->data + offset1, from->iov_base + offset,
648 struct page *page[MAX_SKB_FRAGS];
651 unsigned long truesize;
653 len = from->iov_len - offset;
659 base = (unsigned long)from->iov_base + offset;
660 size = ((base & ~PAGE_MASK) + len + ~PAGE_MASK) >> PAGE_SHIFT;
661 if (i + size > MAX_SKB_FRAGS)
663 num_pages = get_user_pages_fast(base, size, 0, &page[i]);
664 if (num_pages != size) {
665 for (i = 0; i < num_pages; i++)
669 truesize = size * PAGE_SIZE;
670 skb->data_len += len;
672 skb->truesize += truesize;
673 atomic_add(truesize, &skb->sk->sk_wmem_alloc);
675 int off = base & ~PAGE_MASK;
676 int size = min_t(int, len, PAGE_SIZE - off);
677 __skb_fill_page_desc(skb, i, page[i], off, size);
678 skb_shinfo(skb)->nr_frags++;
679 /* increase sk_wmem_alloc */
690 /* Get packet from user space buffer */
691 static ssize_t tun_get_user(struct tun_struct *tun, struct tun_file *tfile,
692 void *msg_control, const struct iovec *iv,
693 size_t total_len, size_t count, int noblock)
695 struct tun_pi pi = { 0, cpu_to_be16(ETH_P_IP) };
697 size_t len = total_len, align = NET_SKB_PAD;
698 struct virtio_net_hdr gso = { 0 };
701 bool zerocopy = false;
704 if (!(tun->flags & TUN_NO_PI)) {
705 if ((len -= sizeof(pi)) > total_len)
708 if (memcpy_fromiovecend((void *)&pi, iv, 0, sizeof(pi)))
710 offset += sizeof(pi);
713 if (tun->flags & TUN_VNET_HDR) {
714 if ((len -= tun->vnet_hdr_sz) > total_len)
717 if (memcpy_fromiovecend((void *)&gso, iv, offset, sizeof(gso)))
720 if ((gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
721 gso.csum_start + gso.csum_offset + 2 > gso.hdr_len)
722 gso.hdr_len = gso.csum_start + gso.csum_offset + 2;
724 if (gso.hdr_len > len)
726 offset += tun->vnet_hdr_sz;
729 if ((tun->flags & TUN_TYPE_MASK) == TUN_TAP_DEV) {
730 align += NET_IP_ALIGN;
731 if (unlikely(len < ETH_HLEN ||
732 (gso.hdr_len && gso.hdr_len < ETH_HLEN)))
740 /* Userspace may produce vectors with count greater than
741 * MAX_SKB_FRAGS, so we need to linearize parts of the skb
742 * to let the rest of data to be fit in the frags.
744 if (count > MAX_SKB_FRAGS) {
745 copylen = iov_length(iv, count - MAX_SKB_FRAGS);
746 if (copylen < offset)
752 /* There are 256 bytes to be copied in skb, so there is enough
753 * room for skb expand head in case it is used.
754 * The rest of the buffer is mapped from userspace.
756 if (copylen < gso.hdr_len)
757 copylen = gso.hdr_len;
759 copylen = GOODCOPY_LEN;
763 skb = tun_alloc_skb(tfile, align, copylen, gso.hdr_len, noblock);
765 if (PTR_ERR(skb) != -EAGAIN)
766 tun->dev->stats.rx_dropped++;
771 err = zerocopy_sg_from_iovec(skb, iv, offset, count);
773 err = skb_copy_datagram_from_iovec(skb, 0, iv, offset, len);
776 tun->dev->stats.rx_dropped++;
781 if (gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
782 if (!skb_partial_csum_set(skb, gso.csum_start,
784 tun->dev->stats.rx_frame_errors++;
790 switch (tun->flags & TUN_TYPE_MASK) {
792 if (tun->flags & TUN_NO_PI) {
793 switch (skb->data[0] & 0xf0) {
795 pi.proto = htons(ETH_P_IP);
798 pi.proto = htons(ETH_P_IPV6);
801 tun->dev->stats.rx_dropped++;
807 skb_reset_mac_header(skb);
808 skb->protocol = pi.proto;
812 skb->protocol = eth_type_trans(skb, tun->dev);
816 if (gso.gso_type != VIRTIO_NET_HDR_GSO_NONE) {
818 switch (gso.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
819 case VIRTIO_NET_HDR_GSO_TCPV4:
820 skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
822 case VIRTIO_NET_HDR_GSO_TCPV6:
823 skb_shinfo(skb)->gso_type = SKB_GSO_TCPV6;
825 case VIRTIO_NET_HDR_GSO_UDP:
826 skb_shinfo(skb)->gso_type = SKB_GSO_UDP;
829 tun->dev->stats.rx_frame_errors++;
834 if (gso.gso_type & VIRTIO_NET_HDR_GSO_ECN)
835 skb_shinfo(skb)->gso_type |= SKB_GSO_TCP_ECN;
837 skb_shinfo(skb)->gso_size = gso.gso_size;
838 if (skb_shinfo(skb)->gso_size == 0) {
839 tun->dev->stats.rx_frame_errors++;
844 /* Header must be checked, and gso_segs computed. */
845 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
846 skb_shinfo(skb)->gso_segs = 0;
849 /* copy skb_ubuf_info for callback when skb has no error */
851 skb_shinfo(skb)->destructor_arg = msg_control;
852 skb_shinfo(skb)->tx_flags |= SKBTX_DEV_ZEROCOPY;
857 tun->dev->stats.rx_packets++;
858 tun->dev->stats.rx_bytes += len;
863 static ssize_t tun_chr_aio_write(struct kiocb *iocb, const struct iovec *iv,
864 unsigned long count, loff_t pos)
866 struct file *file = iocb->ki_filp;
867 struct tun_struct *tun = tun_get(file);
868 struct tun_file *tfile = file->private_data;
874 tun_debug(KERN_INFO, tun, "tun_chr_write %ld\n", count);
876 result = tun_get_user(tun, tfile, NULL, iv, iov_length(iv, count),
877 count, file->f_flags & O_NONBLOCK);
883 /* Put packet to the user space buffer */
884 static ssize_t tun_put_user(struct tun_struct *tun,
885 struct tun_file *tfile,
887 const struct iovec *iv, int len)
889 struct tun_pi pi = { 0, skb->protocol };
892 if (!(tun->flags & TUN_NO_PI)) {
893 if ((len -= sizeof(pi)) < 0)
896 if (len < skb->len) {
897 /* Packet will be striped */
898 pi.flags |= TUN_PKT_STRIP;
901 if (memcpy_toiovecend(iv, (void *) &pi, 0, sizeof(pi)))
906 if (tun->flags & TUN_VNET_HDR) {
907 struct virtio_net_hdr gso = { 0 }; /* no info leak */
908 if ((len -= tun->vnet_hdr_sz) < 0)
911 if (skb_is_gso(skb)) {
912 struct skb_shared_info *sinfo = skb_shinfo(skb);
914 /* This is a hint as to how much should be linear. */
915 gso.hdr_len = skb_headlen(skb);
916 gso.gso_size = sinfo->gso_size;
917 if (sinfo->gso_type & SKB_GSO_TCPV4)
918 gso.gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
919 else if (sinfo->gso_type & SKB_GSO_TCPV6)
920 gso.gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
921 else if (sinfo->gso_type & SKB_GSO_UDP)
922 gso.gso_type = VIRTIO_NET_HDR_GSO_UDP;
924 pr_err("unexpected GSO type: "
925 "0x%x, gso_size %d, hdr_len %d\n",
926 sinfo->gso_type, gso.gso_size,
928 print_hex_dump(KERN_ERR, "tun: ",
931 min((int)gso.hdr_len, 64), true);
935 if (sinfo->gso_type & SKB_GSO_TCP_ECN)
936 gso.gso_type |= VIRTIO_NET_HDR_GSO_ECN;
938 gso.gso_type = VIRTIO_NET_HDR_GSO_NONE;
940 if (skb->ip_summed == CHECKSUM_PARTIAL) {
941 gso.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
942 gso.csum_start = skb_checksum_start_offset(skb);
943 gso.csum_offset = skb->csum_offset;
944 } else if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
945 gso.flags = VIRTIO_NET_HDR_F_DATA_VALID;
946 } /* else everything is zero */
948 if (unlikely(memcpy_toiovecend(iv, (void *)&gso, total,
951 total += tun->vnet_hdr_sz;
954 len = min_t(int, skb->len, len);
956 skb_copy_datagram_const_iovec(skb, 0, iv, total, len);
959 tun->dev->stats.tx_packets++;
960 tun->dev->stats.tx_bytes += len;
965 static ssize_t tun_do_read(struct tun_struct *tun, struct tun_file *tfile,
966 struct kiocb *iocb, const struct iovec *iv,
967 ssize_t len, int noblock)
969 DECLARE_WAITQUEUE(wait, current);
973 tun_debug(KERN_INFO, tun, "tun_chr_read\n");
975 if (unlikely(!noblock))
976 add_wait_queue(&tfile->wq.wait, &wait);
978 current->state = TASK_INTERRUPTIBLE;
980 /* Read frames from the queue */
981 if (!(skb = skb_dequeue(&tfile->socket.sk->sk_receive_queue))) {
986 if (signal_pending(current)) {
990 if (tun->dev->reg_state != NETREG_REGISTERED) {
995 /* Nothing to read, let's sleep */
999 netif_wake_queue(tun->dev);
1001 ret = tun_put_user(tun, tfile, skb, iv, len);
1006 current->state = TASK_RUNNING;
1007 if (unlikely(!noblock))
1008 remove_wait_queue(&tfile->wq.wait, &wait);
1013 static ssize_t tun_chr_aio_read(struct kiocb *iocb, const struct iovec *iv,
1014 unsigned long count, loff_t pos)
1016 struct file *file = iocb->ki_filp;
1017 struct tun_file *tfile = file->private_data;
1018 struct tun_struct *tun = __tun_get(tfile);
1023 len = iov_length(iv, count);
1029 ret = tun_do_read(tun, tfile, iocb, iv, len,
1030 file->f_flags & O_NONBLOCK);
1031 ret = min_t(ssize_t, ret, len);
1037 static void tun_setup(struct net_device *dev)
1039 struct tun_struct *tun = netdev_priv(dev);
1041 tun->owner = INVALID_UID;
1042 tun->group = INVALID_GID;
1044 dev->ethtool_ops = &tun_ethtool_ops;
1045 dev->destructor = free_netdev;
1048 /* Trivial set of netlink ops to allow deleting tun or tap
1049 * device with netlink.
1051 static int tun_validate(struct nlattr *tb[], struct nlattr *data[])
1056 static struct rtnl_link_ops tun_link_ops __read_mostly = {
1058 .priv_size = sizeof(struct tun_struct),
1060 .validate = tun_validate,
1063 static void tun_sock_write_space(struct sock *sk)
1065 struct tun_file *tfile;
1066 wait_queue_head_t *wqueue;
1068 if (!sock_writeable(sk))
1071 if (!test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags))
1074 wqueue = sk_sleep(sk);
1075 if (wqueue && waitqueue_active(wqueue))
1076 wake_up_interruptible_sync_poll(wqueue, POLLOUT |
1077 POLLWRNORM | POLLWRBAND);
1079 tfile = container_of(sk, struct tun_file, sk);
1080 kill_fasync(&tfile->fasync, SIGIO, POLL_OUT);
1083 static int tun_sendmsg(struct kiocb *iocb, struct socket *sock,
1084 struct msghdr *m, size_t total_len)
1087 struct tun_file *tfile = container_of(sock, struct tun_file, socket);
1088 struct tun_struct *tun = __tun_get(tfile);
1093 ret = tun_get_user(tun, tfile, m->msg_control, m->msg_iov, total_len,
1094 m->msg_iovlen, m->msg_flags & MSG_DONTWAIT);
1100 static int tun_recvmsg(struct kiocb *iocb, struct socket *sock,
1101 struct msghdr *m, size_t total_len,
1104 struct tun_file *tfile = container_of(sock, struct tun_file, socket);
1105 struct tun_struct *tun = __tun_get(tfile);
1111 if (flags & ~(MSG_DONTWAIT|MSG_TRUNC))
1113 ret = tun_do_read(tun, tfile, iocb, m->msg_iov, total_len,
1114 flags & MSG_DONTWAIT);
1115 if (ret > total_len) {
1116 m->msg_flags |= MSG_TRUNC;
1117 ret = flags & MSG_TRUNC ? ret : total_len;
1123 static int tun_release(struct socket *sock)
1130 /* Ops structure to mimic raw sockets with tun */
1131 static const struct proto_ops tun_socket_ops = {
1132 .sendmsg = tun_sendmsg,
1133 .recvmsg = tun_recvmsg,
1134 .release = tun_release,
1137 static struct proto tun_proto = {
1139 .owner = THIS_MODULE,
1140 .obj_size = sizeof(struct tun_file),
1143 static int tun_flags(struct tun_struct *tun)
1147 if (tun->flags & TUN_TUN_DEV)
1152 if (tun->flags & TUN_NO_PI)
1155 if (tun->flags & TUN_ONE_QUEUE)
1156 flags |= IFF_ONE_QUEUE;
1158 if (tun->flags & TUN_VNET_HDR)
1159 flags |= IFF_VNET_HDR;
1164 static ssize_t tun_show_flags(struct device *dev, struct device_attribute *attr,
1167 struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1168 return sprintf(buf, "0x%x\n", tun_flags(tun));
1171 static ssize_t tun_show_owner(struct device *dev, struct device_attribute *attr,
1174 struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1175 return uid_valid(tun->owner)?
1176 sprintf(buf, "%u\n",
1177 from_kuid_munged(current_user_ns(), tun->owner)):
1178 sprintf(buf, "-1\n");
1181 static ssize_t tun_show_group(struct device *dev, struct device_attribute *attr,
1184 struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1185 return gid_valid(tun->group) ?
1186 sprintf(buf, "%u\n",
1187 from_kgid_munged(current_user_ns(), tun->group)):
1188 sprintf(buf, "-1\n");
1191 static DEVICE_ATTR(tun_flags, 0444, tun_show_flags, NULL);
1192 static DEVICE_ATTR(owner, 0444, tun_show_owner, NULL);
1193 static DEVICE_ATTR(group, 0444, tun_show_group, NULL);
1195 static int tun_set_iff(struct net *net, struct file *file, struct ifreq *ifr)
1197 struct tun_struct *tun;
1198 struct tun_file *tfile = file->private_data;
1199 struct net_device *dev;
1202 dev = __dev_get_by_name(net, ifr->ifr_name);
1204 const struct cred *cred = current_cred();
1206 if (ifr->ifr_flags & IFF_TUN_EXCL)
1208 if ((ifr->ifr_flags & IFF_TUN) && dev->netdev_ops == &tun_netdev_ops)
1209 tun = netdev_priv(dev);
1210 else if ((ifr->ifr_flags & IFF_TAP) && dev->netdev_ops == &tap_netdev_ops)
1211 tun = netdev_priv(dev);
1215 if (((uid_valid(tun->owner) && !uid_eq(cred->euid, tun->owner)) ||
1216 (gid_valid(tun->group) && !in_egroup_p(tun->group))) &&
1217 !capable(CAP_NET_ADMIN))
1219 err = security_tun_dev_attach(tfile->socket.sk);
1223 err = tun_attach(tun, file);
1229 unsigned long flags = 0;
1231 if (!capable(CAP_NET_ADMIN))
1233 err = security_tun_dev_create();
1238 if (ifr->ifr_flags & IFF_TUN) {
1240 flags |= TUN_TUN_DEV;
1242 } else if (ifr->ifr_flags & IFF_TAP) {
1244 flags |= TUN_TAP_DEV;
1250 name = ifr->ifr_name;
1252 dev = alloc_netdev(sizeof(struct tun_struct), name,
1257 dev_net_set(dev, net);
1258 dev->rtnl_link_ops = &tun_link_ops;
1260 tun = netdev_priv(dev);
1263 tun->txflt.count = 0;
1264 tun->vnet_hdr_sz = sizeof(struct virtio_net_hdr);
1266 tun->filter_attached = false;
1267 tun->sndbuf = tfile->socket.sk->sk_sndbuf;
1269 security_tun_dev_post_create(&tfile->sk);
1273 dev->hw_features = NETIF_F_SG | NETIF_F_FRAGLIST |
1275 dev->features = dev->hw_features;
1277 err = register_netdevice(tun->dev);
1281 if (device_create_file(&tun->dev->dev, &dev_attr_tun_flags) ||
1282 device_create_file(&tun->dev->dev, &dev_attr_owner) ||
1283 device_create_file(&tun->dev->dev, &dev_attr_group))
1284 pr_err("Failed to create tun sysfs files\n");
1286 err = tun_attach(tun, file);
1291 tun_debug(KERN_INFO, tun, "tun_set_iff\n");
1293 if (ifr->ifr_flags & IFF_NO_PI)
1294 tun->flags |= TUN_NO_PI;
1296 tun->flags &= ~TUN_NO_PI;
1298 if (ifr->ifr_flags & IFF_ONE_QUEUE)
1299 tun->flags |= TUN_ONE_QUEUE;
1301 tun->flags &= ~TUN_ONE_QUEUE;
1303 if (ifr->ifr_flags & IFF_VNET_HDR)
1304 tun->flags |= TUN_VNET_HDR;
1306 tun->flags &= ~TUN_VNET_HDR;
1308 /* Make sure persistent devices do not get stuck in
1311 if (netif_running(tun->dev))
1312 netif_wake_queue(tun->dev);
1314 strcpy(ifr->ifr_name, tun->dev->name);
1323 static int tun_get_iff(struct net *net, struct tun_struct *tun,
1326 tun_debug(KERN_INFO, tun, "tun_get_iff\n");
1328 strcpy(ifr->ifr_name, tun->dev->name);
1330 ifr->ifr_flags = tun_flags(tun);
1335 /* This is like a cut-down ethtool ops, except done via tun fd so no
1336 * privs required. */
1337 static int set_offload(struct tun_struct *tun, unsigned long arg)
1339 netdev_features_t features = 0;
1341 if (arg & TUN_F_CSUM) {
1342 features |= NETIF_F_HW_CSUM;
1345 if (arg & (TUN_F_TSO4|TUN_F_TSO6)) {
1346 if (arg & TUN_F_TSO_ECN) {
1347 features |= NETIF_F_TSO_ECN;
1348 arg &= ~TUN_F_TSO_ECN;
1350 if (arg & TUN_F_TSO4)
1351 features |= NETIF_F_TSO;
1352 if (arg & TUN_F_TSO6)
1353 features |= NETIF_F_TSO6;
1354 arg &= ~(TUN_F_TSO4|TUN_F_TSO6);
1357 if (arg & TUN_F_UFO) {
1358 features |= NETIF_F_UFO;
1363 /* This gives the user a way to test for new features in future by
1364 * trying to set them. */
1368 tun->set_features = features;
1369 netdev_update_features(tun->dev);
1374 static long __tun_chr_ioctl(struct file *file, unsigned int cmd,
1375 unsigned long arg, int ifreq_len)
1377 struct tun_file *tfile = file->private_data;
1378 struct tun_struct *tun;
1379 void __user* argp = (void __user*)arg;
1387 if (cmd == TUNSETIFF || _IOC_TYPE(cmd) == 0x89) {
1388 if (copy_from_user(&ifr, argp, ifreq_len))
1391 memset(&ifr, 0, sizeof(ifr));
1393 if (cmd == TUNGETFEATURES) {
1394 /* Currently this just means: "what IFF flags are valid?".
1395 * This is needed because we never checked for invalid flags on
1397 return put_user(IFF_TUN | IFF_TAP | IFF_NO_PI | IFF_ONE_QUEUE |
1399 (unsigned int __user*)argp);
1404 tun = __tun_get(tfile);
1405 if (cmd == TUNSETIFF && !tun) {
1406 ifr.ifr_name[IFNAMSIZ-1] = '\0';
1408 ret = tun_set_iff(tfile->net, file, &ifr);
1413 if (copy_to_user(argp, &ifr, ifreq_len))
1422 tun_debug(KERN_INFO, tun, "tun_chr_ioctl cmd %u\n", cmd);
1427 ret = tun_get_iff(current->nsproxy->net_ns, tun, &ifr);
1431 if (copy_to_user(argp, &ifr, ifreq_len))
1436 /* Disable/Enable checksum */
1438 /* [unimplemented] */
1439 tun_debug(KERN_INFO, tun, "ignored: set checksum %s\n",
1440 arg ? "disabled" : "enabled");
1444 /* Disable/Enable persist mode. Keep an extra reference to the
1445 * module to prevent the module being unprobed.
1448 tun->flags |= TUN_PERSIST;
1449 __module_get(THIS_MODULE);
1451 tun->flags &= ~TUN_PERSIST;
1452 module_put(THIS_MODULE);
1455 tun_debug(KERN_INFO, tun, "persist %s\n",
1456 arg ? "enabled" : "disabled");
1460 /* Set owner of the device */
1461 owner = make_kuid(current_user_ns(), arg);
1462 if (!uid_valid(owner)) {
1467 tun_debug(KERN_INFO, tun, "owner set to %u\n",
1468 from_kuid(&init_user_ns, tun->owner));
1472 /* Set group of the device */
1473 group = make_kgid(current_user_ns(), arg);
1474 if (!gid_valid(group)) {
1479 tun_debug(KERN_INFO, tun, "group set to %u\n",
1480 from_kgid(&init_user_ns, tun->group));
1484 /* Only allow setting the type when the interface is down */
1485 if (tun->dev->flags & IFF_UP) {
1486 tun_debug(KERN_INFO, tun,
1487 "Linktype set failed because interface is up\n");
1490 tun->dev->type = (int) arg;
1491 tun_debug(KERN_INFO, tun, "linktype set to %d\n",
1503 ret = set_offload(tun, arg);
1506 case TUNSETTXFILTER:
1507 /* Can be set only for TAPs */
1509 if ((tun->flags & TUN_TYPE_MASK) != TUN_TAP_DEV)
1511 ret = update_filter(&tun->txflt, (void __user *)arg);
1515 /* Get hw address */
1516 memcpy(ifr.ifr_hwaddr.sa_data, tun->dev->dev_addr, ETH_ALEN);
1517 ifr.ifr_hwaddr.sa_family = tun->dev->type;
1518 if (copy_to_user(argp, &ifr, ifreq_len))
1523 /* Set hw address */
1524 tun_debug(KERN_DEBUG, tun, "set hw address: %pM\n",
1525 ifr.ifr_hwaddr.sa_data);
1527 ret = dev_set_mac_address(tun->dev, &ifr.ifr_hwaddr);
1531 sndbuf = tfile->socket.sk->sk_sndbuf;
1532 if (copy_to_user(argp, &sndbuf, sizeof(sndbuf)))
1537 if (copy_from_user(&sndbuf, argp, sizeof(sndbuf))) {
1542 tun->sndbuf = tfile->socket.sk->sk_sndbuf = sndbuf;
1545 case TUNGETVNETHDRSZ:
1546 vnet_hdr_sz = tun->vnet_hdr_sz;
1547 if (copy_to_user(argp, &vnet_hdr_sz, sizeof(vnet_hdr_sz)))
1551 case TUNSETVNETHDRSZ:
1552 if (copy_from_user(&vnet_hdr_sz, argp, sizeof(vnet_hdr_sz))) {
1556 if (vnet_hdr_sz < (int)sizeof(struct virtio_net_hdr)) {
1561 tun->vnet_hdr_sz = vnet_hdr_sz;
1564 case TUNATTACHFILTER:
1565 /* Can be set only for TAPs */
1567 if ((tun->flags & TUN_TYPE_MASK) != TUN_TAP_DEV)
1570 if (copy_from_user(&tun->fprog, argp, sizeof(tun->fprog)))
1573 ret = sk_attach_filter(&tun->fprog, tfile->socket.sk);
1575 tun->filter_attached = true;
1578 case TUNDETACHFILTER:
1579 /* Can be set only for TAPs */
1581 if ((tun->flags & TUN_TYPE_MASK) != TUN_TAP_DEV)
1583 ret = sk_detach_filter(tfile->socket.sk);
1585 tun->filter_attached = false;
1600 static long tun_chr_ioctl(struct file *file,
1601 unsigned int cmd, unsigned long arg)
1603 return __tun_chr_ioctl(file, cmd, arg, sizeof (struct ifreq));
1606 #ifdef CONFIG_COMPAT
1607 static long tun_chr_compat_ioctl(struct file *file,
1608 unsigned int cmd, unsigned long arg)
1613 case TUNSETTXFILTER:
1618 arg = (unsigned long)compat_ptr(arg);
1621 arg = (compat_ulong_t)arg;
1626 * compat_ifreq is shorter than ifreq, so we must not access beyond
1627 * the end of that structure. All fields that are used in this
1628 * driver are compatible though, we don't need to convert the
1631 return __tun_chr_ioctl(file, cmd, arg, sizeof(struct compat_ifreq));
1633 #endif /* CONFIG_COMPAT */
1635 static int tun_chr_fasync(int fd, struct file *file, int on)
1637 struct tun_file *tfile = file->private_data;
1640 if ((ret = fasync_helper(fd, file, on, &tfile->fasync)) < 0)
1644 ret = __f_setown(file, task_pid(current), PIDTYPE_PID, 0);
1647 tfile->flags |= TUN_FASYNC;
1649 tfile->flags &= ~TUN_FASYNC;
1655 static int tun_chr_open(struct inode *inode, struct file * file)
1657 struct tun_file *tfile;
1659 DBG1(KERN_INFO, "tunX: tun_chr_open\n");
1661 tfile = (struct tun_file *)sk_alloc(&init_net, AF_UNSPEC, GFP_KERNEL,
1665 atomic_set(&tfile->count, 0);
1667 tfile->net = get_net(current->nsproxy->net_ns);
1670 rcu_assign_pointer(tfile->socket.wq, &tfile->wq);
1671 init_waitqueue_head(&tfile->wq.wait);
1673 tfile->socket.file = file;
1674 tfile->socket.ops = &tun_socket_ops;
1676 sock_init_data(&tfile->socket, &tfile->sk);
1677 sk_change_net(&tfile->sk, tfile->net);
1679 tfile->sk.sk_write_space = tun_sock_write_space;
1680 tfile->sk.sk_sndbuf = INT_MAX;
1682 file->private_data = tfile;
1683 set_bit(SOCK_EXTERNALLY_ALLOCATED, &tfile->socket.flags);
1688 static int tun_chr_close(struct inode *inode, struct file *file)
1690 struct tun_file *tfile = file->private_data;
1691 struct tun_struct *tun;
1692 struct net *net = tfile->net;
1694 tun = __tun_get(tfile);
1696 struct net_device *dev = tun->dev;
1698 tun_debug(KERN_INFO, tun, "tun_chr_close\n");
1702 /* If desirable, unregister the netdevice. */
1703 if (!(tun->flags & TUN_PERSIST)) {
1705 if (dev->reg_state == NETREG_REGISTERED)
1706 unregister_netdevice(dev);
1710 /* drop the reference that netdevice holds */
1711 sock_put(&tfile->sk);
1714 /* drop the reference that file holds */
1715 BUG_ON(!test_bit(SOCK_EXTERNALLY_ALLOCATED,
1716 &tfile->socket.flags));
1717 sk_release_kernel(&tfile->sk);
1723 static const struct file_operations tun_fops = {
1724 .owner = THIS_MODULE,
1725 .llseek = no_llseek,
1726 .read = do_sync_read,
1727 .aio_read = tun_chr_aio_read,
1728 .write = do_sync_write,
1729 .aio_write = tun_chr_aio_write,
1730 .poll = tun_chr_poll,
1731 .unlocked_ioctl = tun_chr_ioctl,
1732 #ifdef CONFIG_COMPAT
1733 .compat_ioctl = tun_chr_compat_ioctl,
1735 .open = tun_chr_open,
1736 .release = tun_chr_close,
1737 .fasync = tun_chr_fasync
1740 static struct miscdevice tun_miscdev = {
1743 .nodename = "net/tun",
1747 /* ethtool interface */
1749 static int tun_get_settings(struct net_device *dev, struct ethtool_cmd *cmd)
1752 cmd->advertising = 0;
1753 ethtool_cmd_speed_set(cmd, SPEED_10);
1754 cmd->duplex = DUPLEX_FULL;
1755 cmd->port = PORT_TP;
1756 cmd->phy_address = 0;
1757 cmd->transceiver = XCVR_INTERNAL;
1758 cmd->autoneg = AUTONEG_DISABLE;
1764 static void tun_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
1766 struct tun_struct *tun = netdev_priv(dev);
1768 strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
1769 strlcpy(info->version, DRV_VERSION, sizeof(info->version));
1771 switch (tun->flags & TUN_TYPE_MASK) {
1773 strlcpy(info->bus_info, "tun", sizeof(info->bus_info));
1776 strlcpy(info->bus_info, "tap", sizeof(info->bus_info));
1781 static u32 tun_get_msglevel(struct net_device *dev)
1784 struct tun_struct *tun = netdev_priv(dev);
1791 static void tun_set_msglevel(struct net_device *dev, u32 value)
1794 struct tun_struct *tun = netdev_priv(dev);
1799 static const struct ethtool_ops tun_ethtool_ops = {
1800 .get_settings = tun_get_settings,
1801 .get_drvinfo = tun_get_drvinfo,
1802 .get_msglevel = tun_get_msglevel,
1803 .set_msglevel = tun_set_msglevel,
1804 .get_link = ethtool_op_get_link,
1808 static int __init tun_init(void)
1812 pr_info("%s, %s\n", DRV_DESCRIPTION, DRV_VERSION);
1813 pr_info("%s\n", DRV_COPYRIGHT);
1815 ret = rtnl_link_register(&tun_link_ops);
1817 pr_err("Can't register link_ops\n");
1821 ret = misc_register(&tun_miscdev);
1823 pr_err("Can't register misc device %d\n", TUN_MINOR);
1828 rtnl_link_unregister(&tun_link_ops);
1833 static void tun_cleanup(void)
1835 misc_deregister(&tun_miscdev);
1836 rtnl_link_unregister(&tun_link_ops);
1839 /* Get an underlying socket object from tun file. Returns error unless file is
1840 * attached to a device. The returned object works like a packet socket, it
1841 * can be used for sock_sendmsg/sock_recvmsg. The caller is responsible for
1842 * holding a reference to the file for as long as the socket is in use. */
1843 struct socket *tun_get_socket(struct file *file)
1845 struct tun_struct *tun;
1846 struct tun_file *tfile = file->private_data;
1847 if (file->f_op != &tun_fops)
1848 return ERR_PTR(-EINVAL);
1849 tun = tun_get(file);
1851 return ERR_PTR(-EBADFD);
1853 return &tfile->socket;
1855 EXPORT_SYMBOL_GPL(tun_get_socket);
1857 module_init(tun_init);
1858 module_exit(tun_cleanup);
1859 MODULE_DESCRIPTION(DRV_DESCRIPTION);
1860 MODULE_AUTHOR(DRV_COPYRIGHT);
1861 MODULE_LICENSE("GPL");
1862 MODULE_ALIAS_MISCDEV(TUN_MINOR);
1863 MODULE_ALIAS("devname:net/tun");