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>
71 #include <net/cls_cgroup.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 GOODCOPY_LEN 128
106 #define FLT_EXACT_COUNT 8
108 unsigned int count; /* Number of addrs. Zero means disabled */
109 u32 mask[2]; /* Mask of the hashed addrs */
110 unsigned char addr[FLT_EXACT_COUNT][ETH_ALEN];
115 struct tun_struct *tun;
122 struct tun_file *tfile;
127 struct net_device *dev;
128 netdev_features_t set_features;
129 #define TUN_USER_FEATURES (NETIF_F_HW_CSUM|NETIF_F_TSO_ECN|NETIF_F_TSO| \
130 NETIF_F_TSO6|NETIF_F_UFO)
131 struct fasync_struct *fasync;
133 struct tap_filter txflt;
134 struct socket socket;
146 struct tun_struct *tun;
149 static inline struct tun_sock *tun_sk(struct sock *sk)
151 return container_of(sk, struct tun_sock, sk);
154 static int tun_attach(struct tun_struct *tun, struct file *file)
156 struct tun_file *tfile = file->private_data;
161 netif_tx_lock_bh(tun->dev);
174 tun->socket.file = file;
175 netif_carrier_on(tun->dev);
177 sock_hold(tun->socket.sk);
178 atomic_inc(&tfile->count);
181 netif_tx_unlock_bh(tun->dev);
185 static void __tun_detach(struct tun_struct *tun)
187 /* Detach from net device */
188 netif_tx_lock_bh(tun->dev);
189 netif_carrier_off(tun->dev);
191 netif_tx_unlock_bh(tun->dev);
193 /* Drop read queue */
194 skb_queue_purge(&tun->socket.sk->sk_receive_queue);
196 /* Drop the extra count on the net device */
200 static void tun_detach(struct tun_struct *tun)
207 static struct tun_struct *__tun_get(struct tun_file *tfile)
209 struct tun_struct *tun = NULL;
211 if (atomic_inc_not_zero(&tfile->count))
217 static struct tun_struct *tun_get(struct file *file)
219 return __tun_get(file->private_data);
222 static void tun_put(struct tun_struct *tun)
224 struct tun_file *tfile = tun->tfile;
226 if (atomic_dec_and_test(&tfile->count))
227 tun_detach(tfile->tun);
231 static void addr_hash_set(u32 *mask, const u8 *addr)
233 int n = ether_crc(ETH_ALEN, addr) >> 26;
234 mask[n >> 5] |= (1 << (n & 31));
237 static unsigned int addr_hash_test(const u32 *mask, const u8 *addr)
239 int n = ether_crc(ETH_ALEN, addr) >> 26;
240 return mask[n >> 5] & (1 << (n & 31));
243 static int update_filter(struct tap_filter *filter, void __user *arg)
245 struct { u8 u[ETH_ALEN]; } *addr;
246 struct tun_filter uf;
247 int err, alen, n, nexact;
249 if (copy_from_user(&uf, arg, sizeof(uf)))
258 alen = ETH_ALEN * uf.count;
259 addr = kmalloc(alen, GFP_KERNEL);
263 if (copy_from_user(addr, arg + sizeof(uf), alen)) {
268 /* The filter is updated without holding any locks. Which is
269 * perfectly safe. We disable it first and in the worst
270 * case we'll accept a few undesired packets. */
274 /* Use first set of addresses as an exact filter */
275 for (n = 0; n < uf.count && n < FLT_EXACT_COUNT; n++)
276 memcpy(filter->addr[n], addr[n].u, ETH_ALEN);
280 /* Remaining multicast addresses are hashed,
281 * unicast will leave the filter disabled. */
282 memset(filter->mask, 0, sizeof(filter->mask));
283 for (; n < uf.count; n++) {
284 if (!is_multicast_ether_addr(addr[n].u)) {
285 err = 0; /* no filter */
288 addr_hash_set(filter->mask, addr[n].u);
291 /* For ALLMULTI just set the mask to all ones.
292 * This overrides the mask populated above. */
293 if ((uf.flags & TUN_FLT_ALLMULTI))
294 memset(filter->mask, ~0, sizeof(filter->mask));
296 /* Now enable the filter */
298 filter->count = nexact;
300 /* Return the number of exact filters */
308 /* Returns: 0 - drop, !=0 - accept */
309 static int run_filter(struct tap_filter *filter, const struct sk_buff *skb)
311 /* Cannot use eth_hdr(skb) here because skb_mac_hdr() is incorrect
313 struct ethhdr *eh = (struct ethhdr *) skb->data;
317 for (i = 0; i < filter->count; i++)
318 if (ether_addr_equal(eh->h_dest, filter->addr[i]))
321 /* Inexact match (multicast only) */
322 if (is_multicast_ether_addr(eh->h_dest))
323 return addr_hash_test(filter->mask, eh->h_dest);
329 * Checks whether the packet is accepted or not.
330 * Returns: 0 - drop, !=0 - accept
332 static int check_filter(struct tap_filter *filter, const struct sk_buff *skb)
337 return run_filter(filter, skb);
340 /* Network device part of the driver */
342 static const struct ethtool_ops tun_ethtool_ops;
344 /* Net device detach from fd. */
345 static void tun_net_uninit(struct net_device *dev)
347 struct tun_struct *tun = netdev_priv(dev);
348 struct tun_file *tfile = tun->tfile;
350 /* Inform the methods they need to stop using the dev.
353 wake_up_all(&tun->wq.wait);
354 if (atomic_dec_and_test(&tfile->count))
359 static void tun_free_netdev(struct net_device *dev)
361 struct tun_struct *tun = netdev_priv(dev);
363 BUG_ON(!test_bit(SOCK_EXTERNALLY_ALLOCATED, &tun->socket.flags));
365 sk_release_kernel(tun->socket.sk);
368 /* Net device open. */
369 static int tun_net_open(struct net_device *dev)
371 netif_start_queue(dev);
375 /* Net device close. */
376 static int tun_net_close(struct net_device *dev)
378 netif_stop_queue(dev);
382 /* Net device start xmit */
383 static netdev_tx_t tun_net_xmit(struct sk_buff *skb, struct net_device *dev)
385 struct tun_struct *tun = netdev_priv(dev);
387 tun_debug(KERN_INFO, tun, "tun_net_xmit %d\n", skb->len);
389 /* Drop packet if interface is not attached */
393 /* Drop if the filter does not like it.
394 * This is a noop if the filter is disabled.
395 * Filter can be enabled only for the TAP devices. */
396 if (!check_filter(&tun->txflt, skb))
399 if (tun->socket.sk->sk_filter &&
400 sk_filter(tun->socket.sk, skb))
403 if (skb_queue_len(&tun->socket.sk->sk_receive_queue) >= dev->tx_queue_len) {
404 if (!(tun->flags & TUN_ONE_QUEUE)) {
405 /* Normal queueing mode. */
406 /* Packet scheduler handles dropping of further packets. */
407 netif_stop_queue(dev);
409 /* We won't see all dropped packets individually, so overrun
410 * error is more appropriate. */
411 dev->stats.tx_fifo_errors++;
413 /* Single queue mode.
414 * Driver handles dropping of all packets itself. */
419 /* Orphan the skb - required as we might hang on to it
420 * for indefinite time. */
421 if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC)))
426 skb_queue_tail(&tun->socket.sk->sk_receive_queue, skb);
428 /* Notify and wake up reader process */
429 if (tun->flags & TUN_FASYNC)
430 kill_fasync(&tun->fasync, SIGIO, POLL_IN);
431 wake_up_interruptible_poll(&tun->wq.wait, POLLIN |
432 POLLRDNORM | POLLRDBAND);
436 dev->stats.tx_dropped++;
441 static void tun_net_mclist(struct net_device *dev)
444 * This callback is supposed to deal with mc filter in
445 * _rx_ path and has nothing to do with the _tx_ path.
446 * In rx path we always accept everything userspace gives us.
451 #define MAX_MTU 65535
454 tun_net_change_mtu(struct net_device *dev, int new_mtu)
456 if (new_mtu < MIN_MTU || new_mtu + dev->hard_header_len > MAX_MTU)
462 static netdev_features_t tun_net_fix_features(struct net_device *dev,
463 netdev_features_t features)
465 struct tun_struct *tun = netdev_priv(dev);
467 return (features & tun->set_features) | (features & ~TUN_USER_FEATURES);
469 #ifdef CONFIG_NET_POLL_CONTROLLER
470 static void tun_poll_controller(struct net_device *dev)
473 * Tun only receives frames when:
474 * 1) the char device endpoint gets data from user space
475 * 2) the tun socket gets a sendmsg call from user space
476 * Since both of those are syncronous operations, we are guaranteed
477 * never to have pending data when we poll for it
478 * so theres nothing to do here but return.
479 * We need this though so netpoll recognizes us as an interface that
480 * supports polling, which enables bridge devices in virt setups to
481 * still use netconsole
486 static const struct net_device_ops tun_netdev_ops = {
487 .ndo_uninit = tun_net_uninit,
488 .ndo_open = tun_net_open,
489 .ndo_stop = tun_net_close,
490 .ndo_start_xmit = tun_net_xmit,
491 .ndo_change_mtu = tun_net_change_mtu,
492 .ndo_fix_features = tun_net_fix_features,
493 #ifdef CONFIG_NET_POLL_CONTROLLER
494 .ndo_poll_controller = tun_poll_controller,
498 static const struct net_device_ops tap_netdev_ops = {
499 .ndo_uninit = tun_net_uninit,
500 .ndo_open = tun_net_open,
501 .ndo_stop = tun_net_close,
502 .ndo_start_xmit = tun_net_xmit,
503 .ndo_change_mtu = tun_net_change_mtu,
504 .ndo_fix_features = tun_net_fix_features,
505 .ndo_set_rx_mode = tun_net_mclist,
506 .ndo_set_mac_address = eth_mac_addr,
507 .ndo_validate_addr = eth_validate_addr,
508 #ifdef CONFIG_NET_POLL_CONTROLLER
509 .ndo_poll_controller = tun_poll_controller,
513 /* Initialize net device. */
514 static void tun_net_init(struct net_device *dev)
516 struct tun_struct *tun = netdev_priv(dev);
518 switch (tun->flags & TUN_TYPE_MASK) {
520 dev->netdev_ops = &tun_netdev_ops;
522 /* Point-to-Point TUN Device */
523 dev->hard_header_len = 0;
527 /* Zero header length */
528 dev->type = ARPHRD_NONE;
529 dev->flags = IFF_POINTOPOINT | IFF_NOARP | IFF_MULTICAST;
530 dev->tx_queue_len = TUN_READQ_SIZE; /* We prefer our own queue length */
534 dev->netdev_ops = &tap_netdev_ops;
535 /* Ethernet TAP Device */
537 dev->priv_flags &= ~IFF_TX_SKB_SHARING;
539 eth_hw_addr_random(dev);
541 dev->tx_queue_len = TUN_READQ_SIZE; /* We prefer our own queue length */
546 /* Character device part */
549 static unsigned int tun_chr_poll(struct file *file, poll_table * wait)
551 struct tun_file *tfile = file->private_data;
552 struct tun_struct *tun = __tun_get(tfile);
554 unsigned int mask = 0;
561 tun_debug(KERN_INFO, tun, "tun_chr_poll\n");
563 poll_wait(file, &tun->wq.wait, wait);
565 if (!skb_queue_empty(&sk->sk_receive_queue))
566 mask |= POLLIN | POLLRDNORM;
568 if (sock_writeable(sk) ||
569 (!test_and_set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags) &&
571 mask |= POLLOUT | POLLWRNORM;
573 if (tun->dev->reg_state != NETREG_REGISTERED)
580 /* prepad is the amount to reserve at front. len is length after that.
581 * linear is a hint as to how much to copy (usually headers). */
582 static struct sk_buff *tun_alloc_skb(struct tun_struct *tun,
583 size_t prepad, size_t len,
584 size_t linear, int noblock)
586 struct sock *sk = tun->socket.sk;
590 sock_update_classid(sk);
592 /* Under a page? Don't bother with paged skb. */
593 if (prepad + len < PAGE_SIZE || !linear)
596 skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
601 skb_reserve(skb, prepad);
602 skb_put(skb, linear);
603 skb->data_len = len - linear;
604 skb->len += len - linear;
609 /* set skb frags from iovec, this can move to core network code for reuse */
610 static int zerocopy_sg_from_iovec(struct sk_buff *skb, const struct iovec *from,
611 int offset, size_t count)
613 int len = iov_length(from, count) - offset;
614 int copy = skb_headlen(skb);
615 int size, offset1 = 0;
618 /* Skip over from offset */
619 while (count && (offset >= from->iov_len)) {
620 offset -= from->iov_len;
625 /* copy up to skb headlen */
626 while (count && (copy > 0)) {
627 size = min_t(unsigned int, copy, from->iov_len - offset);
628 if (copy_from_user(skb->data + offset1, from->iov_base + offset,
645 struct page *page[MAX_SKB_FRAGS];
648 unsigned long truesize;
650 len = from->iov_len - offset;
656 base = (unsigned long)from->iov_base + offset;
657 size = ((base & ~PAGE_MASK) + len + ~PAGE_MASK) >> PAGE_SHIFT;
658 if (i + size > MAX_SKB_FRAGS)
660 num_pages = get_user_pages_fast(base, size, 0, &page[i]);
661 if (num_pages != size) {
662 for (i = 0; i < num_pages; i++)
666 truesize = size * PAGE_SIZE;
667 skb->data_len += len;
669 skb->truesize += truesize;
670 atomic_add(truesize, &skb->sk->sk_wmem_alloc);
672 int off = base & ~PAGE_MASK;
673 int size = min_t(int, len, PAGE_SIZE - off);
674 __skb_fill_page_desc(skb, i, page[i], off, size);
675 skb_shinfo(skb)->nr_frags++;
676 /* increase sk_wmem_alloc */
687 /* Get packet from user space buffer */
688 static ssize_t tun_get_user(struct tun_struct *tun, void *msg_control,
689 const struct iovec *iv, size_t total_len,
690 size_t count, int noblock)
692 struct tun_pi pi = { 0, cpu_to_be16(ETH_P_IP) };
694 size_t len = total_len, align = NET_SKB_PAD;
695 struct virtio_net_hdr gso = { 0 };
698 bool zerocopy = false;
701 if (!(tun->flags & TUN_NO_PI)) {
702 if ((len -= sizeof(pi)) > total_len)
705 if (memcpy_fromiovecend((void *)&pi, iv, 0, sizeof(pi)))
707 offset += sizeof(pi);
710 if (tun->flags & TUN_VNET_HDR) {
711 if ((len -= tun->vnet_hdr_sz) > total_len)
714 if (memcpy_fromiovecend((void *)&gso, iv, offset, sizeof(gso)))
717 if ((gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
718 gso.csum_start + gso.csum_offset + 2 > gso.hdr_len)
719 gso.hdr_len = gso.csum_start + gso.csum_offset + 2;
721 if (gso.hdr_len > len)
723 offset += tun->vnet_hdr_sz;
726 if ((tun->flags & TUN_TYPE_MASK) == TUN_TAP_DEV) {
727 align += NET_IP_ALIGN;
728 if (unlikely(len < ETH_HLEN ||
729 (gso.hdr_len && gso.hdr_len < ETH_HLEN)))
737 /* Userspace may produce vectors with count greater than
738 * MAX_SKB_FRAGS, so we need to linearize parts of the skb
739 * to let the rest of data to be fit in the frags.
741 if (count > MAX_SKB_FRAGS) {
742 copylen = iov_length(iv, count - MAX_SKB_FRAGS);
743 if (copylen < offset)
749 /* There are 256 bytes to be copied in skb, so there is enough
750 * room for skb expand head in case it is used.
751 * The rest of the buffer is mapped from userspace.
753 if (copylen < gso.hdr_len)
754 copylen = gso.hdr_len;
756 copylen = GOODCOPY_LEN;
760 skb = tun_alloc_skb(tun, align, copylen, gso.hdr_len, noblock);
762 if (PTR_ERR(skb) != -EAGAIN)
763 tun->dev->stats.rx_dropped++;
768 err = zerocopy_sg_from_iovec(skb, iv, offset, count);
770 err = skb_copy_datagram_from_iovec(skb, 0, iv, offset, len);
773 tun->dev->stats.rx_dropped++;
778 if (gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
779 if (!skb_partial_csum_set(skb, gso.csum_start,
781 tun->dev->stats.rx_frame_errors++;
787 switch (tun->flags & TUN_TYPE_MASK) {
789 if (tun->flags & TUN_NO_PI) {
790 switch (skb->data[0] & 0xf0) {
792 pi.proto = htons(ETH_P_IP);
795 pi.proto = htons(ETH_P_IPV6);
798 tun->dev->stats.rx_dropped++;
804 skb_reset_mac_header(skb);
805 skb->protocol = pi.proto;
809 skb->protocol = eth_type_trans(skb, tun->dev);
813 if (gso.gso_type != VIRTIO_NET_HDR_GSO_NONE) {
815 switch (gso.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
816 case VIRTIO_NET_HDR_GSO_TCPV4:
817 skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
819 case VIRTIO_NET_HDR_GSO_TCPV6:
820 skb_shinfo(skb)->gso_type = SKB_GSO_TCPV6;
822 case VIRTIO_NET_HDR_GSO_UDP:
823 skb_shinfo(skb)->gso_type = SKB_GSO_UDP;
826 tun->dev->stats.rx_frame_errors++;
831 if (gso.gso_type & VIRTIO_NET_HDR_GSO_ECN)
832 skb_shinfo(skb)->gso_type |= SKB_GSO_TCP_ECN;
834 skb_shinfo(skb)->gso_size = gso.gso_size;
835 if (skb_shinfo(skb)->gso_size == 0) {
836 tun->dev->stats.rx_frame_errors++;
841 /* Header must be checked, and gso_segs computed. */
842 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
843 skb_shinfo(skb)->gso_segs = 0;
846 /* copy skb_ubuf_info for callback when skb has no error */
848 skb_shinfo(skb)->destructor_arg = msg_control;
849 skb_shinfo(skb)->tx_flags |= SKBTX_DEV_ZEROCOPY;
854 tun->dev->stats.rx_packets++;
855 tun->dev->stats.rx_bytes += len;
860 static ssize_t tun_chr_aio_write(struct kiocb *iocb, const struct iovec *iv,
861 unsigned long count, loff_t pos)
863 struct file *file = iocb->ki_filp;
864 struct tun_struct *tun = tun_get(file);
870 tun_debug(KERN_INFO, tun, "tun_chr_write %ld\n", count);
872 result = tun_get_user(tun, NULL, iv, iov_length(iv, count), count,
873 file->f_flags & O_NONBLOCK);
879 /* Put packet to the user space buffer */
880 static ssize_t tun_put_user(struct tun_struct *tun,
882 const struct iovec *iv, int len)
884 struct tun_pi pi = { 0, skb->protocol };
887 if (!(tun->flags & TUN_NO_PI)) {
888 if ((len -= sizeof(pi)) < 0)
891 if (len < skb->len) {
892 /* Packet will be striped */
893 pi.flags |= TUN_PKT_STRIP;
896 if (memcpy_toiovecend(iv, (void *) &pi, 0, sizeof(pi)))
901 if (tun->flags & TUN_VNET_HDR) {
902 struct virtio_net_hdr gso = { 0 }; /* no info leak */
903 if ((len -= tun->vnet_hdr_sz) < 0)
906 if (skb_is_gso(skb)) {
907 struct skb_shared_info *sinfo = skb_shinfo(skb);
909 /* This is a hint as to how much should be linear. */
910 gso.hdr_len = skb_headlen(skb);
911 gso.gso_size = sinfo->gso_size;
912 if (sinfo->gso_type & SKB_GSO_TCPV4)
913 gso.gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
914 else if (sinfo->gso_type & SKB_GSO_TCPV6)
915 gso.gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
916 else if (sinfo->gso_type & SKB_GSO_UDP)
917 gso.gso_type = VIRTIO_NET_HDR_GSO_UDP;
919 pr_err("unexpected GSO type: "
920 "0x%x, gso_size %d, hdr_len %d\n",
921 sinfo->gso_type, gso.gso_size,
923 print_hex_dump(KERN_ERR, "tun: ",
926 min((int)gso.hdr_len, 64), true);
930 if (sinfo->gso_type & SKB_GSO_TCP_ECN)
931 gso.gso_type |= VIRTIO_NET_HDR_GSO_ECN;
933 gso.gso_type = VIRTIO_NET_HDR_GSO_NONE;
935 if (skb->ip_summed == CHECKSUM_PARTIAL) {
936 gso.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
937 gso.csum_start = skb_checksum_start_offset(skb);
938 gso.csum_offset = skb->csum_offset;
939 } else if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
940 gso.flags = VIRTIO_NET_HDR_F_DATA_VALID;
941 } /* else everything is zero */
943 if (unlikely(memcpy_toiovecend(iv, (void *)&gso, total,
946 total += tun->vnet_hdr_sz;
949 len = min_t(int, skb->len, len);
951 skb_copy_datagram_const_iovec(skb, 0, iv, total, len);
954 tun->dev->stats.tx_packets++;
955 tun->dev->stats.tx_bytes += len;
960 static ssize_t tun_do_read(struct tun_struct *tun,
961 struct kiocb *iocb, const struct iovec *iv,
962 ssize_t len, int noblock)
964 DECLARE_WAITQUEUE(wait, current);
968 tun_debug(KERN_INFO, tun, "tun_chr_read\n");
970 if (unlikely(!noblock))
971 add_wait_queue(&tun->wq.wait, &wait);
973 current->state = TASK_INTERRUPTIBLE;
975 /* Read frames from the queue */
976 if (!(skb=skb_dequeue(&tun->socket.sk->sk_receive_queue))) {
981 if (signal_pending(current)) {
985 if (tun->dev->reg_state != NETREG_REGISTERED) {
990 /* Nothing to read, let's sleep */
994 netif_wake_queue(tun->dev);
996 ret = tun_put_user(tun, skb, iv, len);
1001 current->state = TASK_RUNNING;
1002 if (unlikely(!noblock))
1003 remove_wait_queue(&tun->wq.wait, &wait);
1008 static ssize_t tun_chr_aio_read(struct kiocb *iocb, const struct iovec *iv,
1009 unsigned long count, loff_t pos)
1011 struct file *file = iocb->ki_filp;
1012 struct tun_file *tfile = file->private_data;
1013 struct tun_struct *tun = __tun_get(tfile);
1018 len = iov_length(iv, count);
1024 ret = tun_do_read(tun, iocb, iv, len, file->f_flags & O_NONBLOCK);
1025 ret = min_t(ssize_t, ret, len);
1031 static void tun_setup(struct net_device *dev)
1033 struct tun_struct *tun = netdev_priv(dev);
1035 tun->owner = INVALID_UID;
1036 tun->group = INVALID_GID;
1038 dev->ethtool_ops = &tun_ethtool_ops;
1039 dev->destructor = tun_free_netdev;
1042 /* Trivial set of netlink ops to allow deleting tun or tap
1043 * device with netlink.
1045 static int tun_validate(struct nlattr *tb[], struct nlattr *data[])
1050 static struct rtnl_link_ops tun_link_ops __read_mostly = {
1052 .priv_size = sizeof(struct tun_struct),
1054 .validate = tun_validate,
1057 static void tun_sock_write_space(struct sock *sk)
1059 struct tun_struct *tun;
1060 wait_queue_head_t *wqueue;
1062 if (!sock_writeable(sk))
1065 if (!test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags))
1068 wqueue = sk_sleep(sk);
1069 if (wqueue && waitqueue_active(wqueue))
1070 wake_up_interruptible_sync_poll(wqueue, POLLOUT |
1071 POLLWRNORM | POLLWRBAND);
1073 tun = tun_sk(sk)->tun;
1074 kill_fasync(&tun->fasync, SIGIO, POLL_OUT);
1077 static void tun_sock_destruct(struct sock *sk)
1079 free_netdev(tun_sk(sk)->tun->dev);
1082 static int tun_sendmsg(struct kiocb *iocb, struct socket *sock,
1083 struct msghdr *m, size_t total_len)
1085 struct tun_struct *tun = container_of(sock, struct tun_struct, socket);
1086 return tun_get_user(tun, m->msg_control, m->msg_iov, total_len,
1087 m->msg_iovlen, m->msg_flags & MSG_DONTWAIT);
1090 static int tun_recvmsg(struct kiocb *iocb, struct socket *sock,
1091 struct msghdr *m, size_t total_len,
1094 struct tun_struct *tun = container_of(sock, struct tun_struct, socket);
1096 if (flags & ~(MSG_DONTWAIT|MSG_TRUNC))
1098 ret = tun_do_read(tun, iocb, m->msg_iov, total_len,
1099 flags & MSG_DONTWAIT);
1100 if (ret > total_len) {
1101 m->msg_flags |= MSG_TRUNC;
1102 ret = flags & MSG_TRUNC ? ret : total_len;
1107 static int tun_release(struct socket *sock)
1114 /* Ops structure to mimic raw sockets with tun */
1115 static const struct proto_ops tun_socket_ops = {
1116 .sendmsg = tun_sendmsg,
1117 .recvmsg = tun_recvmsg,
1118 .release = tun_release,
1121 static struct proto tun_proto = {
1123 .owner = THIS_MODULE,
1124 .obj_size = sizeof(struct tun_sock),
1127 static int tun_flags(struct tun_struct *tun)
1131 if (tun->flags & TUN_TUN_DEV)
1136 if (tun->flags & TUN_NO_PI)
1139 if (tun->flags & TUN_ONE_QUEUE)
1140 flags |= IFF_ONE_QUEUE;
1142 if (tun->flags & TUN_VNET_HDR)
1143 flags |= IFF_VNET_HDR;
1148 static ssize_t tun_show_flags(struct device *dev, struct device_attribute *attr,
1151 struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1152 return sprintf(buf, "0x%x\n", tun_flags(tun));
1155 static ssize_t tun_show_owner(struct device *dev, struct device_attribute *attr,
1158 struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1159 return uid_valid(tun->owner)?
1160 sprintf(buf, "%u\n",
1161 from_kuid_munged(current_user_ns(), tun->owner)):
1162 sprintf(buf, "-1\n");
1165 static ssize_t tun_show_group(struct device *dev, struct device_attribute *attr,
1168 struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1169 return gid_valid(tun->group) ?
1170 sprintf(buf, "%u\n",
1171 from_kgid_munged(current_user_ns(), tun->group)):
1172 sprintf(buf, "-1\n");
1175 static DEVICE_ATTR(tun_flags, 0444, tun_show_flags, NULL);
1176 static DEVICE_ATTR(owner, 0444, tun_show_owner, NULL);
1177 static DEVICE_ATTR(group, 0444, tun_show_group, NULL);
1179 static int tun_set_iff(struct net *net, struct file *file, struct ifreq *ifr)
1182 struct tun_struct *tun;
1183 struct net_device *dev;
1186 dev = __dev_get_by_name(net, ifr->ifr_name);
1188 const struct cred *cred = current_cred();
1190 if (ifr->ifr_flags & IFF_TUN_EXCL)
1192 if ((ifr->ifr_flags & IFF_TUN) && dev->netdev_ops == &tun_netdev_ops)
1193 tun = netdev_priv(dev);
1194 else if ((ifr->ifr_flags & IFF_TAP) && dev->netdev_ops == &tap_netdev_ops)
1195 tun = netdev_priv(dev);
1199 if (((uid_valid(tun->owner) && !uid_eq(cred->euid, tun->owner)) ||
1200 (gid_valid(tun->group) && !in_egroup_p(tun->group))) &&
1201 !capable(CAP_NET_ADMIN))
1203 err = security_tun_dev_attach(tun->socket.sk);
1207 err = tun_attach(tun, file);
1213 unsigned long flags = 0;
1215 if (!capable(CAP_NET_ADMIN))
1217 err = security_tun_dev_create();
1222 if (ifr->ifr_flags & IFF_TUN) {
1224 flags |= TUN_TUN_DEV;
1226 } else if (ifr->ifr_flags & IFF_TAP) {
1228 flags |= TUN_TAP_DEV;
1234 name = ifr->ifr_name;
1236 dev = alloc_netdev(sizeof(struct tun_struct), name,
1241 dev_net_set(dev, net);
1242 dev->rtnl_link_ops = &tun_link_ops;
1244 tun = netdev_priv(dev);
1247 tun->txflt.count = 0;
1248 tun->vnet_hdr_sz = sizeof(struct virtio_net_hdr);
1249 set_bit(SOCK_EXTERNALLY_ALLOCATED, &tun->socket.flags);
1252 sk = sk_alloc(&init_net, AF_UNSPEC, GFP_KERNEL, &tun_proto);
1256 sk_change_net(sk, net);
1257 tun->socket.wq = &tun->wq;
1258 init_waitqueue_head(&tun->wq.wait);
1259 tun->socket.ops = &tun_socket_ops;
1260 sock_init_data(&tun->socket, sk);
1261 sk->sk_write_space = tun_sock_write_space;
1262 sk->sk_sndbuf = INT_MAX;
1263 sock_set_flag(sk, SOCK_ZEROCOPY);
1265 tun_sk(sk)->tun = tun;
1267 security_tun_dev_post_create(sk);
1271 dev->hw_features = NETIF_F_SG | NETIF_F_FRAGLIST |
1273 dev->features = dev->hw_features;
1275 err = register_netdevice(tun->dev);
1279 if (device_create_file(&tun->dev->dev, &dev_attr_tun_flags) ||
1280 device_create_file(&tun->dev->dev, &dev_attr_owner) ||
1281 device_create_file(&tun->dev->dev, &dev_attr_group))
1282 pr_err("Failed to create tun sysfs files\n");
1284 sk->sk_destruct = tun_sock_destruct;
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);
1318 tun_free_netdev(dev);
1325 static int tun_get_iff(struct net *net, struct tun_struct *tun,
1328 tun_debug(KERN_INFO, tun, "tun_get_iff\n");
1330 strcpy(ifr->ifr_name, tun->dev->name);
1332 ifr->ifr_flags = tun_flags(tun);
1337 /* This is like a cut-down ethtool ops, except done via tun fd so no
1338 * privs required. */
1339 static int set_offload(struct tun_struct *tun, unsigned long arg)
1341 netdev_features_t features = 0;
1343 if (arg & TUN_F_CSUM) {
1344 features |= NETIF_F_HW_CSUM;
1347 if (arg & (TUN_F_TSO4|TUN_F_TSO6)) {
1348 if (arg & TUN_F_TSO_ECN) {
1349 features |= NETIF_F_TSO_ECN;
1350 arg &= ~TUN_F_TSO_ECN;
1352 if (arg & TUN_F_TSO4)
1353 features |= NETIF_F_TSO;
1354 if (arg & TUN_F_TSO6)
1355 features |= NETIF_F_TSO6;
1356 arg &= ~(TUN_F_TSO4|TUN_F_TSO6);
1359 if (arg & TUN_F_UFO) {
1360 features |= NETIF_F_UFO;
1365 /* This gives the user a way to test for new features in future by
1366 * trying to set them. */
1370 tun->set_features = features;
1371 netdev_update_features(tun->dev);
1376 static long __tun_chr_ioctl(struct file *file, unsigned int cmd,
1377 unsigned long arg, int ifreq_len)
1379 struct tun_file *tfile = file->private_data;
1380 struct tun_struct *tun;
1381 void __user* argp = (void __user*)arg;
1382 struct sock_fprog fprog;
1390 if (cmd == TUNSETIFF || _IOC_TYPE(cmd) == 0x89) {
1391 if (copy_from_user(&ifr, argp, ifreq_len))
1394 memset(&ifr, 0, sizeof(ifr));
1396 if (cmd == TUNGETFEATURES) {
1397 /* Currently this just means: "what IFF flags are valid?".
1398 * This is needed because we never checked for invalid flags on
1400 return put_user(IFF_TUN | IFF_TAP | IFF_NO_PI | IFF_ONE_QUEUE |
1402 (unsigned int __user*)argp);
1407 tun = __tun_get(tfile);
1408 if (cmd == TUNSETIFF && !tun) {
1409 ifr.ifr_name[IFNAMSIZ-1] = '\0';
1411 ret = tun_set_iff(tfile->net, file, &ifr);
1416 if (copy_to_user(argp, &ifr, ifreq_len))
1425 tun_debug(KERN_INFO, tun, "tun_chr_ioctl cmd %d\n", cmd);
1430 ret = tun_get_iff(current->nsproxy->net_ns, tun, &ifr);
1434 if (copy_to_user(argp, &ifr, ifreq_len))
1439 /* Disable/Enable checksum */
1441 /* [unimplemented] */
1442 tun_debug(KERN_INFO, tun, "ignored: set checksum %s\n",
1443 arg ? "disabled" : "enabled");
1447 /* Disable/Enable persist mode */
1449 tun->flags |= TUN_PERSIST;
1451 tun->flags &= ~TUN_PERSIST;
1453 tun_debug(KERN_INFO, tun, "persist %s\n",
1454 arg ? "enabled" : "disabled");
1458 /* Set owner of the device */
1459 owner = make_kuid(current_user_ns(), arg);
1460 if (!uid_valid(owner)) {
1465 tun_debug(KERN_INFO, tun, "owner set to %d\n",
1466 from_kuid(&init_user_ns, tun->owner));
1470 /* Set group of the device */
1471 group = make_kgid(current_user_ns(), arg);
1472 if (!gid_valid(group)) {
1477 tun_debug(KERN_INFO, tun, "group set to %d\n",
1478 from_kgid(&init_user_ns, tun->group));
1482 /* Only allow setting the type when the interface is down */
1483 if (tun->dev->flags & IFF_UP) {
1484 tun_debug(KERN_INFO, tun,
1485 "Linktype set failed because interface is up\n");
1488 tun->dev->type = (int) arg;
1489 tun_debug(KERN_INFO, tun, "linktype set to %d\n",
1501 ret = set_offload(tun, arg);
1504 case TUNSETTXFILTER:
1505 /* Can be set only for TAPs */
1507 if ((tun->flags & TUN_TYPE_MASK) != TUN_TAP_DEV)
1509 ret = update_filter(&tun->txflt, (void __user *)arg);
1513 /* Get hw address */
1514 memcpy(ifr.ifr_hwaddr.sa_data, tun->dev->dev_addr, ETH_ALEN);
1515 ifr.ifr_hwaddr.sa_family = tun->dev->type;
1516 if (copy_to_user(argp, &ifr, ifreq_len))
1521 /* Set hw address */
1522 tun_debug(KERN_DEBUG, tun, "set hw address: %pM\n",
1523 ifr.ifr_hwaddr.sa_data);
1525 ret = dev_set_mac_address(tun->dev, &ifr.ifr_hwaddr);
1529 sndbuf = tun->socket.sk->sk_sndbuf;
1530 if (copy_to_user(argp, &sndbuf, sizeof(sndbuf)))
1535 if (copy_from_user(&sndbuf, argp, sizeof(sndbuf))) {
1540 tun->socket.sk->sk_sndbuf = sndbuf;
1543 case TUNGETVNETHDRSZ:
1544 vnet_hdr_sz = tun->vnet_hdr_sz;
1545 if (copy_to_user(argp, &vnet_hdr_sz, sizeof(vnet_hdr_sz)))
1549 case TUNSETVNETHDRSZ:
1550 if (copy_from_user(&vnet_hdr_sz, argp, sizeof(vnet_hdr_sz))) {
1554 if (vnet_hdr_sz < (int)sizeof(struct virtio_net_hdr)) {
1559 tun->vnet_hdr_sz = vnet_hdr_sz;
1562 case TUNATTACHFILTER:
1563 /* Can be set only for TAPs */
1565 if ((tun->flags & TUN_TYPE_MASK) != TUN_TAP_DEV)
1568 if (copy_from_user(&fprog, argp, sizeof(fprog)))
1571 ret = sk_attach_filter(&fprog, tun->socket.sk);
1574 case TUNDETACHFILTER:
1575 /* Can be set only for TAPs */
1577 if ((tun->flags & TUN_TYPE_MASK) != TUN_TAP_DEV)
1579 ret = sk_detach_filter(tun->socket.sk);
1594 static long tun_chr_ioctl(struct file *file,
1595 unsigned int cmd, unsigned long arg)
1597 return __tun_chr_ioctl(file, cmd, arg, sizeof (struct ifreq));
1600 #ifdef CONFIG_COMPAT
1601 static long tun_chr_compat_ioctl(struct file *file,
1602 unsigned int cmd, unsigned long arg)
1607 case TUNSETTXFILTER:
1612 arg = (unsigned long)compat_ptr(arg);
1615 arg = (compat_ulong_t)arg;
1620 * compat_ifreq is shorter than ifreq, so we must not access beyond
1621 * the end of that structure. All fields that are used in this
1622 * driver are compatible though, we don't need to convert the
1625 return __tun_chr_ioctl(file, cmd, arg, sizeof(struct compat_ifreq));
1627 #endif /* CONFIG_COMPAT */
1629 static int tun_chr_fasync(int fd, struct file *file, int on)
1631 struct tun_struct *tun = tun_get(file);
1637 tun_debug(KERN_INFO, tun, "tun_chr_fasync %d\n", on);
1639 if ((ret = fasync_helper(fd, file, on, &tun->fasync)) < 0)
1643 ret = __f_setown(file, task_pid(current), PIDTYPE_PID, 0);
1646 tun->flags |= TUN_FASYNC;
1648 tun->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 = kmalloc(sizeof(*tfile), GFP_KERNEL);
1664 atomic_set(&tfile->count, 0);
1666 tfile->net = get_net(current->nsproxy->net_ns);
1667 file->private_data = tfile;
1671 static int tun_chr_close(struct inode *inode, struct file *file)
1673 struct tun_file *tfile = file->private_data;
1674 struct tun_struct *tun;
1676 tun = __tun_get(tfile);
1678 struct net_device *dev = tun->dev;
1680 tun_debug(KERN_INFO, tun, "tun_chr_close\n");
1684 /* If desirable, unregister the netdevice. */
1685 if (!(tun->flags & TUN_PERSIST)) {
1687 if (dev->reg_state == NETREG_REGISTERED)
1688 unregister_netdevice(dev);
1695 sock_put(tun->socket.sk);
1697 put_net(tfile->net);
1703 static const struct file_operations tun_fops = {
1704 .owner = THIS_MODULE,
1705 .llseek = no_llseek,
1706 .read = do_sync_read,
1707 .aio_read = tun_chr_aio_read,
1708 .write = do_sync_write,
1709 .aio_write = tun_chr_aio_write,
1710 .poll = tun_chr_poll,
1711 .unlocked_ioctl = tun_chr_ioctl,
1712 #ifdef CONFIG_COMPAT
1713 .compat_ioctl = tun_chr_compat_ioctl,
1715 .open = tun_chr_open,
1716 .release = tun_chr_close,
1717 .fasync = tun_chr_fasync
1720 static struct miscdevice tun_miscdev = {
1723 .nodename = "net/tun",
1727 /* ethtool interface */
1729 static int tun_get_settings(struct net_device *dev, struct ethtool_cmd *cmd)
1732 cmd->advertising = 0;
1733 ethtool_cmd_speed_set(cmd, SPEED_10);
1734 cmd->duplex = DUPLEX_FULL;
1735 cmd->port = PORT_TP;
1736 cmd->phy_address = 0;
1737 cmd->transceiver = XCVR_INTERNAL;
1738 cmd->autoneg = AUTONEG_DISABLE;
1744 static void tun_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
1746 struct tun_struct *tun = netdev_priv(dev);
1748 strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
1749 strlcpy(info->version, DRV_VERSION, sizeof(info->version));
1751 switch (tun->flags & TUN_TYPE_MASK) {
1753 strlcpy(info->bus_info, "tun", sizeof(info->bus_info));
1756 strlcpy(info->bus_info, "tap", sizeof(info->bus_info));
1761 static u32 tun_get_msglevel(struct net_device *dev)
1764 struct tun_struct *tun = netdev_priv(dev);
1771 static void tun_set_msglevel(struct net_device *dev, u32 value)
1774 struct tun_struct *tun = netdev_priv(dev);
1779 static const struct ethtool_ops tun_ethtool_ops = {
1780 .get_settings = tun_get_settings,
1781 .get_drvinfo = tun_get_drvinfo,
1782 .get_msglevel = tun_get_msglevel,
1783 .set_msglevel = tun_set_msglevel,
1784 .get_link = ethtool_op_get_link,
1788 static int __init tun_init(void)
1792 pr_info("%s, %s\n", DRV_DESCRIPTION, DRV_VERSION);
1793 pr_info("%s\n", DRV_COPYRIGHT);
1795 ret = rtnl_link_register(&tun_link_ops);
1797 pr_err("Can't register link_ops\n");
1801 ret = misc_register(&tun_miscdev);
1803 pr_err("Can't register misc device %d\n", TUN_MINOR);
1808 rtnl_link_unregister(&tun_link_ops);
1813 static void tun_cleanup(void)
1815 misc_deregister(&tun_miscdev);
1816 rtnl_link_unregister(&tun_link_ops);
1819 /* Get an underlying socket object from tun file. Returns error unless file is
1820 * attached to a device. The returned object works like a packet socket, it
1821 * can be used for sock_sendmsg/sock_recvmsg. The caller is responsible for
1822 * holding a reference to the file for as long as the socket is in use. */
1823 struct socket *tun_get_socket(struct file *file)
1825 struct tun_struct *tun;
1826 if (file->f_op != &tun_fops)
1827 return ERR_PTR(-EINVAL);
1828 tun = tun_get(file);
1830 return ERR_PTR(-EBADFD);
1832 return &tun->socket;
1834 EXPORT_SYMBOL_GPL(tun_get_socket);
1836 module_init(tun_init);
1837 module_exit(tun_cleanup);
1838 MODULE_DESCRIPTION(DRV_DESCRIPTION);
1839 MODULE_AUTHOR(DRV_COPYRIGHT);
1840 MODULE_LICENSE("GPL");
1841 MODULE_ALIAS_MISCDEV(TUN_MINOR);
1842 MODULE_ALIAS("devname:net/tun");