Merge branch 'master' of master.kernel.org:/pub/scm/linux/kernel/git/davem/net-2.6
[pandora-kernel.git] / drivers / net / tun.c
index 09fea31..4825c52 100644 (file)
@@ -63,6 +63,8 @@
 #include <linux/virtio_net.h>
 #include <net/net_namespace.h>
 #include <net/netns/generic.h>
+#include <net/rtnetlink.h>
+#include <net/sock.h>
 
 #include <asm/system.h>
 #include <asm/uaccess.h>
@@ -87,26 +89,127 @@ struct tap_filter {
        unsigned char   addr[FLT_EXACT_COUNT][ETH_ALEN];
 };
 
+struct tun_file {
+       atomic_t count;
+       struct tun_struct *tun;
+       struct net *net;
+       wait_queue_head_t       read_wait;
+};
+
+struct tun_sock;
+
 struct tun_struct {
-       struct list_head        list;
+       struct tun_file         *tfile;
        unsigned int            flags;
-       int                     attached;
        uid_t                   owner;
        gid_t                   group;
 
-       wait_queue_head_t       read_wait;
        struct sk_buff_head     readq;
 
        struct net_device       *dev;
        struct fasync_struct    *fasync;
 
        struct tap_filter       txflt;
+       struct sock             *sk;
+       struct socket           socket;
 
 #ifdef TUN_DEBUG
        int debug;
 #endif
 };
 
+struct tun_sock {
+       struct sock             sk;
+       struct tun_struct       *tun;
+};
+
+static inline struct tun_sock *tun_sk(struct sock *sk)
+{
+       return container_of(sk, struct tun_sock, sk);
+}
+
+static int tun_attach(struct tun_struct *tun, struct file *file)
+{
+       struct tun_file *tfile = file->private_data;
+       const struct cred *cred = current_cred();
+       int err;
+
+       ASSERT_RTNL();
+
+       /* Check permissions */
+       if (((tun->owner != -1 && cred->euid != tun->owner) ||
+            (tun->group != -1 && !in_egroup_p(tun->group))) &&
+               !capable(CAP_NET_ADMIN))
+               return -EPERM;
+
+       netif_tx_lock_bh(tun->dev);
+
+       err = -EINVAL;
+       if (tfile->tun)
+               goto out;
+
+       err = -EBUSY;
+       if (tun->tfile)
+               goto out;
+
+       err = 0;
+       tfile->tun = tun;
+       tun->tfile = tfile;
+       dev_hold(tun->dev);
+       atomic_inc(&tfile->count);
+
+out:
+       netif_tx_unlock_bh(tun->dev);
+       return err;
+}
+
+static void __tun_detach(struct tun_struct *tun)
+{
+       struct tun_file *tfile = tun->tfile;
+
+       /* Detach from net device */
+       netif_tx_lock_bh(tun->dev);
+       tfile->tun = NULL;
+       tun->tfile = NULL;
+       netif_tx_unlock_bh(tun->dev);
+
+       /* Drop read queue */
+       skb_queue_purge(&tun->readq);
+
+       /* Drop the extra count on the net device */
+       dev_put(tun->dev);
+}
+
+static void tun_detach(struct tun_struct *tun)
+{
+       rtnl_lock();
+       __tun_detach(tun);
+       rtnl_unlock();
+}
+
+static struct tun_struct *__tun_get(struct tun_file *tfile)
+{
+       struct tun_struct *tun = NULL;
+
+       if (atomic_inc_not_zero(&tfile->count))
+               tun = tfile->tun;
+
+       return tun;
+}
+
+static struct tun_struct *tun_get(struct file *file)
+{
+       return __tun_get(file->private_data);
+}
+
+static void tun_put(struct tun_struct *tun)
+{
+       struct tun_file *tfile = tun->tfile;
+
+       if (atomic_dec_and_test(&tfile->count))
+               tun_detach(tfile->tun);
+}
+
 /* TAP filterting */
 static void addr_hash_set(u32 *mask, const u8 *addr)
 {
@@ -219,13 +322,23 @@ static int check_filter(struct tap_filter *filter, const struct sk_buff *skb)
 
 /* Network device part of the driver */
 
-static int tun_net_id;
-struct tun_net {
-       struct list_head dev_list;
-};
-
 static const struct ethtool_ops tun_ethtool_ops;
 
+/* Net device detach from fd. */
+static void tun_net_uninit(struct net_device *dev)
+{
+       struct tun_struct *tun = netdev_priv(dev);
+       struct tun_file *tfile = tun->tfile;
+
+       /* Inform the methods they need to stop using the dev.
+        */
+       if (tfile) {
+               wake_up_all(&tfile->read_wait);
+               if (atomic_dec_and_test(&tfile->count))
+                       __tun_detach(tun);
+       }
+}
+
 /* Net device open. */
 static int tun_net_open(struct net_device *dev)
 {
@@ -248,7 +361,7 @@ static int tun_net_xmit(struct sk_buff *skb, struct net_device *dev)
        DBG(KERN_INFO "%s: tun_net_xmit %d\n", tun->dev->name, skb->len);
 
        /* Drop packet if interface is not attached */
-       if (!tun->attached)
+       if (!tun->tfile)
                goto drop;
 
        /* Drop if the filter does not like it.
@@ -280,7 +393,7 @@ static int tun_net_xmit(struct sk_buff *skb, struct net_device *dev)
        /* Notify and wake up reader process */
        if (tun->flags & TUN_FASYNC)
                kill_fasync(&tun->fasync, SIGIO, POLL_IN);
-       wake_up_interruptible(&tun->read_wait);
+       wake_up_interruptible(&tun->tfile->read_wait);
        return 0;
 
 drop:
@@ -312,6 +425,7 @@ tun_net_change_mtu(struct net_device *dev, int new_mtu)
 }
 
 static const struct net_device_ops tun_netdev_ops = {
+       .ndo_uninit             = tun_net_uninit,
        .ndo_open               = tun_net_open,
        .ndo_stop               = tun_net_close,
        .ndo_start_xmit         = tun_net_xmit,
@@ -319,6 +433,7 @@ static const struct net_device_ops tun_netdev_ops = {
 };
 
 static const struct net_device_ops tap_netdev_ops = {
+       .ndo_uninit             = tun_net_uninit,
        .ndo_open               = tun_net_open,
        .ndo_stop               = tun_net_close,
        .ndo_start_xmit         = tun_net_xmit,
@@ -365,86 +480,66 @@ static void tun_net_init(struct net_device *dev)
 /* Poll */
 static unsigned int tun_chr_poll(struct file *file, poll_table * wait)
 {
-       struct tun_struct *tun = file->private_data;
-       unsigned int mask = POLLOUT | POLLWRNORM;
+       struct tun_file *tfile = file->private_data;
+       struct tun_struct *tun = __tun_get(tfile);
+       struct sock *sk = tun->sk;
+       unsigned int mask = 0;
 
        if (!tun)
-               return -EBADFD;
+               return POLLERR;
 
        DBG(KERN_INFO "%s: tun_chr_poll\n", tun->dev->name);
 
-       poll_wait(file, &tun->read_wait, wait);
+       poll_wait(file, &tfile->read_wait, wait);
 
        if (!skb_queue_empty(&tun->readq))
                mask |= POLLIN | POLLRDNORM;
 
+       if (sock_writeable(sk) ||
+           (!test_and_set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags) &&
+            sock_writeable(sk)))
+               mask |= POLLOUT | POLLWRNORM;
+
+       if (tun->dev->reg_state != NETREG_REGISTERED)
+               mask = POLLERR;
+
+       tun_put(tun);
        return mask;
 }
 
 /* prepad is the amount to reserve at front.  len is length after that.
  * linear is a hint as to how much to copy (usually headers). */
-static struct sk_buff *tun_alloc_skb(size_t prepad, size_t len, size_t linear,
-                                    gfp_t gfp)
+static inline struct sk_buff *tun_alloc_skb(struct tun_struct *tun,
+                                           size_t prepad, size_t len,
+                                           size_t linear, int noblock)
 {
+       struct sock *sk = tun->sk;
        struct sk_buff *skb;
-       unsigned int i;
-
-       skb = alloc_skb(prepad + len, gfp|__GFP_NOWARN);
-       if (skb) {
-               skb_reserve(skb, prepad);
-               skb_put(skb, len);
-               return skb;
-       }
+       int err;
 
        /* Under a page?  Don't bother with paged skb. */
        if (prepad + len < PAGE_SIZE)
-               return NULL;
+               linear = len;
 
-       /* Start with a normal skb, and add pages. */
-       skb = alloc_skb(prepad + linear, gfp);
+       skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
+                                  &err);
        if (!skb)
-               return NULL;
+               return ERR_PTR(err);
 
        skb_reserve(skb, prepad);
        skb_put(skb, linear);
-
-       len -= linear;
-
-       for (i = 0; i < MAX_SKB_FRAGS; i++) {
-               skb_frag_t *f = &skb_shinfo(skb)->frags[i];
-
-               f->page = alloc_page(gfp|__GFP_ZERO);
-               if (!f->page)
-                       break;
-
-               f->page_offset = 0;
-               f->size = PAGE_SIZE;
-
-               skb->data_len += PAGE_SIZE;
-               skb->len += PAGE_SIZE;
-               skb->truesize += PAGE_SIZE;
-               skb_shinfo(skb)->nr_frags++;
-
-               if (len < PAGE_SIZE) {
-                       len = 0;
-                       break;
-               }
-               len -= PAGE_SIZE;
-       }
-
-       /* Too large, or alloc fail? */
-       if (unlikely(len)) {
-               kfree_skb(skb);
-               skb = NULL;
-       }
+       skb->data_len = len - linear;
+       skb->len += len - linear;
 
        return skb;
 }
 
 /* Get packet from user space buffer */
-static __inline__ ssize_t tun_get_user(struct tun_struct *tun, struct iovec *iv, size_t count)
+static __inline__ ssize_t tun_get_user(struct tun_struct *tun,
+                                      struct iovec *iv, size_t count,
+                                      int noblock)
 {
-       struct tun_pi pi = { 0, __constant_htons(ETH_P_IP) };
+       struct tun_pi pi = { 0, cpu_to_be16(ETH_P_IP) };
        struct sk_buff *skb;
        size_t len = count, align = 0;
        struct virtio_net_hdr gso = { 0 };
@@ -474,9 +569,11 @@ static __inline__ ssize_t tun_get_user(struct tun_struct *tun, struct iovec *iv,
                        return -EINVAL;
        }
 
-       if (!(skb = tun_alloc_skb(align, len, gso.hdr_len, GFP_KERNEL))) {
-               tun->dev->stats.rx_dropped++;
-               return -ENOMEM;
+       skb = tun_alloc_skb(tun, align, len, gso.hdr_len, noblock);
+       if (IS_ERR(skb)) {
+               if (PTR_ERR(skb) != -EAGAIN)
+                       tun->dev->stats.rx_dropped++;
+               return PTR_ERR(skb);
        }
 
        if (skb_copy_datagram_from_iovec(skb, 0, iv, len)) {
@@ -562,14 +659,20 @@ static __inline__ ssize_t tun_get_user(struct tun_struct *tun, struct iovec *iv,
 static ssize_t tun_chr_aio_write(struct kiocb *iocb, const struct iovec *iv,
                              unsigned long count, loff_t pos)
 {
-       struct tun_struct *tun = iocb->ki_filp->private_data;
+       struct file *file = iocb->ki_filp;
+       struct tun_struct *tun = file->private_data;
+       ssize_t result;
 
        if (!tun)
                return -EBADFD;
 
        DBG(KERN_INFO "%s: tun_chr_write %ld\n", tun->dev->name, count);
 
-       return tun_get_user(tun, (struct iovec *) iv, iov_length(iv, count));
+       result = tun_get_user(tun, (struct iovec *)iv, iov_length(iv, count),
+                             file->f_flags & O_NONBLOCK);
+
+       tun_put(tun);
+       return result;
 }
 
 /* Put packet to the user space buffer */
@@ -642,7 +745,8 @@ static ssize_t tun_chr_aio_read(struct kiocb *iocb, const struct iovec *iv,
                            unsigned long count, loff_t pos)
 {
        struct file *file = iocb->ki_filp;
-       struct tun_struct *tun = file->private_data;
+       struct tun_file *tfile = file->private_data;
+       struct tun_struct *tun = __tun_get(tfile);
        DECLARE_WAITQUEUE(wait, current);
        struct sk_buff *skb;
        ssize_t len, ret = 0;
@@ -653,10 +757,12 @@ static ssize_t tun_chr_aio_read(struct kiocb *iocb, const struct iovec *iv,
        DBG(KERN_INFO "%s: tun_chr_read\n", tun->dev->name);
 
        len = iov_length(iv, count);
-       if (len < 0)
-               return -EINVAL;
+       if (len < 0) {
+               ret = -EINVAL;
+               goto out;
+       }
 
-       add_wait_queue(&tun->read_wait, &wait);
+       add_wait_queue(&tfile->read_wait, &wait);
        while (len) {
                current->state = TASK_INTERRUPTIBLE;
 
@@ -670,6 +776,10 @@ static ssize_t tun_chr_aio_read(struct kiocb *iocb, const struct iovec *iv,
                                ret = -ERESTARTSYS;
                                break;
                        }
+                       if (tun->dev->reg_state != NETREG_REGISTERED) {
+                               ret = -EIO;
+                               break;
+                       }
 
                        /* Nothing to read, let's sleep */
                        schedule();
@@ -683,8 +793,10 @@ static ssize_t tun_chr_aio_read(struct kiocb *iocb, const struct iovec *iv,
        }
 
        current->state = TASK_RUNNING;
-       remove_wait_queue(&tun->read_wait, &wait);
+       remove_wait_queue(&tfile->read_wait, &wait);
 
+out:
+       tun_put(tun);
        return ret;
 }
 
@@ -693,54 +805,78 @@ static void tun_setup(struct net_device *dev)
        struct tun_struct *tun = netdev_priv(dev);
 
        skb_queue_head_init(&tun->readq);
-       init_waitqueue_head(&tun->read_wait);
 
        tun->owner = -1;
        tun->group = -1;
 
        dev->ethtool_ops = &tun_ethtool_ops;
        dev->destructor = free_netdev;
-       dev->features |= NETIF_F_NETNS_LOCAL;
 }
 
-static struct tun_struct *tun_get_by_name(struct tun_net *tn, const char *name)
+/* Trivial set of netlink ops to allow deleting tun or tap
+ * device with netlink.
+ */
+static int tun_validate(struct nlattr *tb[], struct nlattr *data[])
+{
+       return -EINVAL;
+}
+
+static struct rtnl_link_ops tun_link_ops __read_mostly = {
+       .kind           = DRV_NAME,
+       .priv_size      = sizeof(struct tun_struct),
+       .setup          = tun_setup,
+       .validate       = tun_validate,
+};
+
+static void tun_sock_write_space(struct sock *sk)
 {
        struct tun_struct *tun;
 
-       ASSERT_RTNL();
-       list_for_each_entry(tun, &tn->dev_list, list) {
-               if (!strncmp(tun->dev->name, name, IFNAMSIZ))
-                   return tun;
-       }
+       if (!sock_writeable(sk))
+               return;
+
+       if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
+               wake_up_interruptible_sync(sk->sk_sleep);
+
+       if (!test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags))
+               return;
 
-       return NULL;
+       tun = container_of(sk, struct tun_sock, sk)->tun;
+       kill_fasync(&tun->fasync, SIGIO, POLL_OUT);
 }
 
+static void tun_sock_destruct(struct sock *sk)
+{
+       dev_put(container_of(sk, struct tun_sock, sk)->tun->dev);
+}
+
+static struct proto tun_proto = {
+       .name           = "tun",
+       .owner          = THIS_MODULE,
+       .obj_size       = sizeof(struct tun_sock),
+};
+
 static int tun_set_iff(struct net *net, struct file *file, struct ifreq *ifr)
 {
-       struct tun_net *tn;
+       struct sock *sk;
        struct tun_struct *tun;
        struct net_device *dev;
-       const struct cred *cred = current_cred();
+       struct tun_file *tfile = file->private_data;
        int err;
 
-       tn = net_generic(net, tun_net_id);
-       tun = tun_get_by_name(tn, ifr->ifr_name);
-       if (tun) {
-               if (tun->attached)
-                       return -EBUSY;
-
-               /* Check permissions */
-               if (((tun->owner != -1 &&
-                     cred->euid != tun->owner) ||
-                    (tun->group != -1 &&
-                     cred->egid != tun->group)) &&
-                   !capable(CAP_NET_ADMIN)) {
-                       return -EPERM;
-               }
+       dev = __dev_get_by_name(net, ifr->ifr_name);
+       if (dev) {
+               if ((ifr->ifr_flags & IFF_TUN) && dev->netdev_ops == &tun_netdev_ops)
+                       tun = netdev_priv(dev);
+               else if ((ifr->ifr_flags & IFF_TAP) && dev->netdev_ops == &tap_netdev_ops)
+                       tun = netdev_priv(dev);
+               else
+                       return -EINVAL;
+
+               err = tun_attach(tun, file);
+               if (err < 0)
+                       return err;
        }
-       else if (__dev_get_by_name(net, ifr->ifr_name))
-               return -EINVAL;
        else {
                char *name;
                unsigned long flags = 0;
@@ -771,25 +907,45 @@ static int tun_set_iff(struct net *net, struct file *file, struct ifreq *ifr)
                        return -ENOMEM;
 
                dev_net_set(dev, net);
+               dev->rtnl_link_ops = &tun_link_ops;
 
                tun = netdev_priv(dev);
                tun->dev = dev;
                tun->flags = flags;
                tun->txflt.count = 0;
 
+               err = -ENOMEM;
+               sk = sk_alloc(net, AF_UNSPEC, GFP_KERNEL, &tun_proto);
+               if (!sk)
+                       goto err_free_dev;
+
+               /* This ref count is for tun->sk. */
+               dev_hold(dev);
+               sock_init_data(&tun->socket, sk);
+               sk->sk_write_space = tun_sock_write_space;
+               sk->sk_destruct = tun_sock_destruct;
+               sk->sk_sndbuf = INT_MAX;
+               sk->sk_sleep = &tfile->read_wait;
+
+               tun->sk = sk;
+               container_of(sk, struct tun_sock, sk)->tun = tun;
+
                tun_net_init(dev);
 
                if (strchr(dev->name, '%')) {
                        err = dev_alloc_name(dev, dev->name);
                        if (err < 0)
-                               goto err_free_dev;
+                               goto err_free_sk;
                }
 
+               err = -EINVAL;
                err = register_netdevice(tun->dev);
                if (err < 0)
                        goto err_free_dev;
 
-               list_add(&tun->list, &tn->dev_list);
+               err = tun_attach(tun, file);
+               if (err < 0)
+                       goto err_free_dev;
        }
 
        DBG(KERN_INFO "%s: tun_set_iff\n", tun->dev->name);
@@ -809,10 +965,6 @@ static int tun_set_iff(struct net *net, struct file *file, struct ifreq *ifr)
        else
                tun->flags &= ~TUN_VNET_HDR;
 
-       file->private_data = tun;
-       tun->attached = 1;
-       get_net(dev_net(tun->dev));
-
        /* Make sure persistent devices do not get stuck in
         * xoff state.
         */
@@ -822,6 +974,8 @@ static int tun_set_iff(struct net *net, struct file *file, struct ifreq *ifr)
        strcpy(ifr->ifr_name, tun->dev->name);
        return 0;
 
+ err_free_sk:
+       sock_put(sk);
  err_free_dev:
        free_netdev(dev);
  failed:
@@ -830,7 +984,7 @@ static int tun_set_iff(struct net *net, struct file *file, struct ifreq *ifr)
 
 static int tun_get_iff(struct net *net, struct file *file, struct ifreq *ifr)
 {
-       struct tun_struct *tun = file->private_data;
+       struct tun_struct *tun = tun_get(file);
 
        if (!tun)
                return -EBADFD;
@@ -855,6 +1009,7 @@ static int tun_get_iff(struct net *net, struct file *file, struct ifreq *ifr)
        if (tun->flags & TUN_VNET_HDR)
                ifr->ifr_flags |= IFF_VNET_HDR;
 
+       tun_put(tun);
        return 0;
 }
 
@@ -901,22 +1056,34 @@ static int set_offload(struct net_device *dev, unsigned long arg)
 static int tun_chr_ioctl(struct inode *inode, struct file *file,
                         unsigned int cmd, unsigned long arg)
 {
-       struct tun_struct *tun = file->private_data;
+       struct tun_file *tfile = file->private_data;
+       struct tun_struct *tun;
        void __user* argp = (void __user*)arg;
        struct ifreq ifr;
+       int sndbuf;
        int ret;
 
        if (cmd == TUNSETIFF || _IOC_TYPE(cmd) == 0x89)
                if (copy_from_user(&ifr, argp, sizeof ifr))
                        return -EFAULT;
 
+       if (cmd == TUNGETFEATURES) {
+               /* Currently this just means: "what IFF flags are valid?".
+                * This is needed because we never checked for invalid flags on
+                * TUNSETIFF. */
+               return put_user(IFF_TUN | IFF_TAP | IFF_NO_PI | IFF_ONE_QUEUE |
+                               IFF_VNET_HDR,
+                               (unsigned int __user*)argp);
+       }
+
+       tun = __tun_get(tfile);
        if (cmd == TUNSETIFF && !tun) {
                int err;
 
                ifr.ifr_name[IFNAMSIZ-1] = '\0';
 
                rtnl_lock();
-               err = tun_set_iff(current->nsproxy->net_ns, file, &ifr);
+               err = tun_set_iff(tfile->net, file, &ifr);
                rtnl_unlock();
 
                if (err)
@@ -927,28 +1094,21 @@ static int tun_chr_ioctl(struct inode *inode, struct file *file,
                return 0;
        }
 
-       if (cmd == TUNGETFEATURES) {
-               /* Currently this just means: "what IFF flags are valid?".
-                * This is needed because we never checked for invalid flags on
-                * TUNSETIFF. */
-               return put_user(IFF_TUN | IFF_TAP | IFF_NO_PI | IFF_ONE_QUEUE |
-                               IFF_VNET_HDR,
-                               (unsigned int __user*)argp);
-       }
 
        if (!tun)
                return -EBADFD;
 
        DBG(KERN_INFO "%s: tun_chr_ioctl cmd %d\n", tun->dev->name, cmd);
 
+       ret = 0;
        switch (cmd) {
        case TUNGETIFF:
                ret = tun_get_iff(current->nsproxy->net_ns, file, &ifr);
                if (ret)
-                       return ret;
+                       break;
 
                if (copy_to_user(argp, &ifr, sizeof(ifr)))
-                       return -EFAULT;
+                       ret = -EFAULT;
                break;
 
        case TUNSETNOCSUM:
@@ -1000,7 +1160,7 @@ static int tun_chr_ioctl(struct inode *inode, struct file *file,
                        ret = 0;
                }
                rtnl_unlock();
-               return ret;
+               break;
 
 #ifdef TUN_DEBUG
        case TUNSETDEBUG:
@@ -1011,24 +1171,25 @@ static int tun_chr_ioctl(struct inode *inode, struct file *file,
                rtnl_lock();
                ret = set_offload(tun->dev, arg);
                rtnl_unlock();
-               return ret;
+               break;
 
        case TUNSETTXFILTER:
                /* Can be set only for TAPs */
+               ret = -EINVAL;
                if ((tun->flags & TUN_TYPE_MASK) != TUN_TAP_DEV)
-                       return -EINVAL;
+                       break;
                rtnl_lock();
                ret = update_filter(&tun->txflt, (void __user *)arg);
                rtnl_unlock();
-               return ret;
+               break;
 
        case SIOCGIFHWADDR:
                /* Get hw addres */
                memcpy(ifr.ifr_hwaddr.sa_data, tun->dev->dev_addr, ETH_ALEN);
                ifr.ifr_hwaddr.sa_family = tun->dev->type;
                if (copy_to_user(argp, &ifr, sizeof ifr))
-                       return -EFAULT;
-               return 0;
+                       ret = -EFAULT;
+               break;
 
        case SIOCSIFHWADDR:
                /* Set hw address */
@@ -1038,18 +1199,35 @@ static int tun_chr_ioctl(struct inode *inode, struct file *file,
                rtnl_lock();
                ret = dev_set_mac_address(tun->dev, &ifr.ifr_hwaddr);
                rtnl_unlock();
-               return ret;
+               break;
+
+       case TUNGETSNDBUF:
+               sndbuf = tun->sk->sk_sndbuf;
+               if (copy_to_user(argp, &sndbuf, sizeof(sndbuf)))
+                       ret = -EFAULT;
+               break;
+
+       case TUNSETSNDBUF:
+               if (copy_from_user(&sndbuf, argp, sizeof(sndbuf))) {
+                       ret = -EFAULT;
+                       break;
+               }
+
+               tun->sk->sk_sndbuf = sndbuf;
+               break;
 
        default:
-               return -EINVAL;
+               ret = -EINVAL;
+               break;
        };
 
-       return 0;
+       tun_put(tun);
+       return ret;
 }
 
 static int tun_chr_fasync(int fd, struct file *file, int on)
 {
-       struct tun_struct *tun = file->private_data;
+       struct tun_struct *tun = tun_get(file);
        int ret;
 
        if (!tun)
@@ -1071,42 +1249,50 @@ static int tun_chr_fasync(int fd, struct file *file, int on)
        ret = 0;
 out:
        unlock_kernel();
+       tun_put(tun);
        return ret;
 }
 
 static int tun_chr_open(struct inode *inode, struct file * file)
 {
+       struct tun_file *tfile;
        cycle_kernel_lock();
        DBG1(KERN_INFO "tunX: tun_chr_open\n");
-       file->private_data = NULL;
+
+       tfile = kmalloc(sizeof(*tfile), GFP_KERNEL);
+       if (!tfile)
+               return -ENOMEM;
+       atomic_set(&tfile->count, 0);
+       tfile->tun = NULL;
+       tfile->net = get_net(current->nsproxy->net_ns);
+       init_waitqueue_head(&tfile->read_wait);
+       file->private_data = tfile;
        return 0;
 }
 
 static int tun_chr_close(struct inode *inode, struct file *file)
 {
-       struct tun_struct *tun = file->private_data;
-
-       if (!tun)
-               return 0;
+       struct tun_file *tfile = file->private_data;
+       struct tun_struct *tun = __tun_get(tfile);
 
-       DBG(KERN_INFO "%s: tun_chr_close\n", tun->dev->name);
 
-       rtnl_lock();
+       if (tun) {
+               DBG(KERN_INFO "%s: tun_chr_close\n", tun->dev->name);
 
-       /* Detach from net device */
-       file->private_data = NULL;
-       tun->attached = 0;
-       put_net(dev_net(tun->dev));
+               rtnl_lock();
+               __tun_detach(tun);
 
-       /* Drop read queue */
-       skb_queue_purge(&tun->readq);
+               /* If desireable, unregister the netdevice. */
+               if (!(tun->flags & TUN_PERSIST)) {
+                       sock_put(tun->sk);
+                       unregister_netdevice(tun->dev);
+               }
 
-       if (!(tun->flags & TUN_PERSIST)) {
-               list_del(&tun->list);
-               unregister_netdevice(tun->dev);
+               rtnl_unlock();
        }
 
-       rtnl_unlock();
+       put_net(tfile->net);
+       kfree(tfile);
 
        return 0;
 }
@@ -1187,7 +1373,7 @@ static void tun_set_msglevel(struct net_device *dev, u32 value)
 static u32 tun_get_link(struct net_device *dev)
 {
        struct tun_struct *tun = netdev_priv(dev);
-       return tun->attached;
+       return !!tun->tfile;
 }
 
 static u32 tun_get_rx_csum(struct net_device *dev)
@@ -1216,45 +1402,6 @@ static const struct ethtool_ops tun_ethtool_ops = {
        .set_rx_csum    = tun_set_rx_csum
 };
 
-static int tun_init_net(struct net *net)
-{
-       struct tun_net *tn;
-
-       tn = kmalloc(sizeof(*tn), GFP_KERNEL);
-       if (tn == NULL)
-               return -ENOMEM;
-
-       INIT_LIST_HEAD(&tn->dev_list);
-
-       if (net_assign_generic(net, tun_net_id, tn)) {
-               kfree(tn);
-               return -ENOMEM;
-       }
-
-       return 0;
-}
-
-static void tun_exit_net(struct net *net)
-{
-       struct tun_net *tn;
-       struct tun_struct *tun, *nxt;
-
-       tn = net_generic(net, tun_net_id);
-
-       rtnl_lock();
-       list_for_each_entry_safe(tun, nxt, &tn->dev_list, list) {
-               DBG(KERN_INFO "%s cleaned up\n", tun->dev->name);
-               unregister_netdevice(tun->dev);
-       }
-       rtnl_unlock();
-
-       kfree(tn);
-}
-
-static struct pernet_operations tun_net_ops = {
-       .init = tun_init_net,
-       .exit = tun_exit_net,
-};
 
 static int __init tun_init(void)
 {
@@ -1263,10 +1410,10 @@ static int __init tun_init(void)
        printk(KERN_INFO "tun: %s, %s\n", DRV_DESCRIPTION, DRV_VERSION);
        printk(KERN_INFO "tun: %s\n", DRV_COPYRIGHT);
 
-       ret = register_pernet_gen_device(&tun_net_id, &tun_net_ops);
+       ret = rtnl_link_register(&tun_link_ops);
        if (ret) {
-               printk(KERN_ERR "tun: Can't register pernet ops\n");
-               goto err_pernet;
+               printk(KERN_ERR "tun: Can't register link_ops\n");
+               goto err_linkops;
        }
 
        ret = misc_register(&tun_miscdev);
@@ -1274,18 +1421,17 @@ static int __init tun_init(void)
                printk(KERN_ERR "tun: Can't register misc device %d\n", TUN_MINOR);
                goto err_misc;
        }
-       return 0;
-
+       return  0;
 err_misc:
-       unregister_pernet_gen_device(tun_net_id, &tun_net_ops);
-err_pernet:
+       rtnl_link_unregister(&tun_link_ops);
+err_linkops:
        return ret;
 }
 
 static void tun_cleanup(void)
 {
        misc_deregister(&tun_miscdev);
-       unregister_pernet_gen_device(tun_net_id, &tun_net_ops);
+       rtnl_link_unregister(&tun_link_ops);
 }
 
 module_init(tun_init);