1 #include <linux/etherdevice.h>
2 #include <linux/if_macvlan.h>
3 #include <linux/interrupt.h>
4 #include <linux/nsproxy.h>
5 #include <linux/compat.h>
6 #include <linux/if_tun.h>
7 #include <linux/module.h>
8 #include <linux/skbuff.h>
9 #include <linux/cache.h>
10 #include <linux/sched.h>
11 #include <linux/types.h>
12 #include <linux/slab.h>
13 #include <linux/init.h>
14 #include <linux/wait.h>
15 #include <linux/cdev.h>
18 #include <net/net_namespace.h>
19 #include <net/rtnetlink.h>
21 #include <linux/virtio_net.h>
24 * A macvtap queue is the central object of this driver, it connects
25 * an open character device to a macvlan interface. There can be
26 * multiple queues on one interface, which map back to queues
27 * implemented in hardware on the underlying device.
29 * macvtap_proto is used to allocate queues through the sock allocation
32 * TODO: multiqueue support is currently not implemented, even though
33 * macvtap is basically prepared for that. We will need to add this
34 * here as well as in virtio-net and qemu to get line rate on 10gbit
35 * adapters from a guest.
37 struct macvtap_queue {
42 struct macvlan_dev *vlan;
47 static struct proto macvtap_proto = {
50 .obj_size = sizeof (struct macvtap_queue),
54 * Minor number matches netdev->ifindex, so need a potentially
55 * large value. This also makes it possible to split the
56 * tap functionality out again in the future by offering it
57 * from other drivers besides macvtap. As long as every device
58 * only has one tap, the interface numbers assure that the
59 * device nodes are unique.
61 static dev_t macvtap_major;
62 #define MACVTAP_NUM_DEVS 65536
63 static struct class *macvtap_class;
64 static struct cdev macvtap_cdev;
66 static const struct proto_ops macvtap_socket_ops;
70 * The macvtap_queue and the macvlan_dev are loosely coupled, the
71 * pointers from one to the other can only be read while rcu_read_lock
72 * or macvtap_lock is held.
74 * Both the file and the macvlan_dev hold a reference on the macvtap_queue
75 * through sock_hold(&q->sk). When the macvlan_dev goes away first,
76 * q->vlan becomes inaccessible. When the files gets closed,
77 * macvtap_get_queue() fails.
79 * There may still be references to the struct sock inside of the
80 * queue from outbound SKBs, but these never reference back to the
81 * file or the dev. The data structure is freed through __sk_free
82 * when both our references and any pending SKBs are gone.
84 static DEFINE_SPINLOCK(macvtap_lock);
87 * Choose the next free queue, for now there is only one
89 static int macvtap_set_queue(struct net_device *dev, struct file *file,
90 struct macvtap_queue *q)
92 struct macvlan_dev *vlan = netdev_priv(dev);
95 spin_lock(&macvtap_lock);
96 if (rcu_dereference(vlan->tap))
100 rcu_assign_pointer(q->vlan, vlan);
101 rcu_assign_pointer(vlan->tap, q);
105 file->private_data = q;
108 spin_unlock(&macvtap_lock);
113 * The file owning the queue got closed, give up both
114 * the reference that the files holds as well as the
115 * one from the macvlan_dev if that still exists.
117 * Using the spinlock makes sure that we don't get
118 * to the queue again after destroying it.
120 static void macvtap_put_queue(struct macvtap_queue *q)
122 struct macvlan_dev *vlan;
124 spin_lock(&macvtap_lock);
125 vlan = rcu_dereference(q->vlan);
127 rcu_assign_pointer(vlan->tap, NULL);
128 rcu_assign_pointer(q->vlan, NULL);
132 spin_unlock(&macvtap_lock);
139 * Since we only support one queue, just dereference the pointer.
141 static struct macvtap_queue *macvtap_get_queue(struct net_device *dev,
144 struct macvlan_dev *vlan = netdev_priv(dev);
146 return rcu_dereference(vlan->tap);
150 * The net_device is going away, give up the reference
151 * that it holds on the queue (all the queues one day)
152 * and safely set the pointer from the queues to NULL.
154 static void macvtap_del_queues(struct net_device *dev)
156 struct macvlan_dev *vlan = netdev_priv(dev);
157 struct macvtap_queue *q;
159 spin_lock(&macvtap_lock);
160 q = rcu_dereference(vlan->tap);
162 spin_unlock(&macvtap_lock);
166 rcu_assign_pointer(vlan->tap, NULL);
167 rcu_assign_pointer(q->vlan, NULL);
168 spin_unlock(&macvtap_lock);
175 * Forward happens for data that gets sent from one macvlan
176 * endpoint to another one in bridge mode. We just take
177 * the skb and put it into the receive queue.
179 static int macvtap_forward(struct net_device *dev, struct sk_buff *skb)
181 struct macvtap_queue *q = macvtap_get_queue(dev, skb);
185 if (skb_queue_len(&q->sk.sk_receive_queue) >= dev->tx_queue_len)
188 skb_queue_tail(&q->sk.sk_receive_queue, skb);
189 wake_up_interruptible_poll(sk_sleep(&q->sk), POLLIN | POLLRDNORM | POLLRDBAND);
190 return NET_RX_SUCCESS;
198 * Receive is for data from the external interface (lowerdev),
199 * in case of macvtap, we can treat that the same way as
200 * forward, which macvlan cannot.
202 static int macvtap_receive(struct sk_buff *skb)
204 skb_push(skb, ETH_HLEN);
205 return macvtap_forward(skb->dev, skb);
208 static int macvtap_newlink(struct net *src_net,
209 struct net_device *dev,
211 struct nlattr *data[])
213 struct device *classdev;
217 err = macvlan_common_newlink(src_net, dev, tb, data,
218 macvtap_receive, macvtap_forward);
222 devt = MKDEV(MAJOR(macvtap_major), dev->ifindex);
224 classdev = device_create(macvtap_class, &dev->dev, devt,
225 dev, "tap%d", dev->ifindex);
226 if (IS_ERR(classdev)) {
227 err = PTR_ERR(classdev);
228 macvtap_del_queues(dev);
235 static void macvtap_dellink(struct net_device *dev,
236 struct list_head *head)
238 device_destroy(macvtap_class,
239 MKDEV(MAJOR(macvtap_major), dev->ifindex));
241 macvtap_del_queues(dev);
242 macvlan_dellink(dev, head);
245 static void macvtap_setup(struct net_device *dev)
247 macvlan_common_setup(dev);
248 dev->tx_queue_len = TUN_READQ_SIZE;
251 static struct rtnl_link_ops macvtap_link_ops __read_mostly = {
253 .setup = macvtap_setup,
254 .newlink = macvtap_newlink,
255 .dellink = macvtap_dellink,
259 static void macvtap_sock_write_space(struct sock *sk)
261 wait_queue_head_t *wqueue;
263 if (!sock_writeable(sk) ||
264 !test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags))
267 wqueue = sk_sleep(sk);
268 if (wqueue && waitqueue_active(wqueue))
269 wake_up_interruptible_poll(wqueue, POLLOUT | POLLWRNORM | POLLWRBAND);
272 static int macvtap_open(struct inode *inode, struct file *file)
274 struct net *net = current->nsproxy->net_ns;
275 struct net_device *dev = dev_get_by_index(net, iminor(inode));
276 struct macvtap_queue *q;
283 /* check if this is a macvtap device */
285 if (dev->rtnl_link_ops != &macvtap_link_ops)
289 q = (struct macvtap_queue *)sk_alloc(net, AF_UNSPEC, GFP_KERNEL,
295 init_waitqueue_head(&q->wq.wait);
296 q->sock.type = SOCK_RAW;
297 q->sock.state = SS_CONNECTED;
299 q->sock.ops = &macvtap_socket_ops;
300 sock_init_data(&q->sock, &q->sk);
301 q->sk.sk_write_space = macvtap_sock_write_space;
302 q->flags = IFF_VNET_HDR | IFF_NO_PI | IFF_TAP;
303 q->vnet_hdr_sz = sizeof(struct virtio_net_hdr);
305 err = macvtap_set_queue(dev, file, q);
316 static int macvtap_release(struct inode *inode, struct file *file)
318 struct macvtap_queue *q = file->private_data;
319 macvtap_put_queue(q);
323 static unsigned int macvtap_poll(struct file *file, poll_table * wait)
325 struct macvtap_queue *q = file->private_data;
326 unsigned int mask = POLLERR;
332 poll_wait(file, &q->wq.wait, wait);
334 if (!skb_queue_empty(&q->sk.sk_receive_queue))
335 mask |= POLLIN | POLLRDNORM;
337 if (sock_writeable(&q->sk) ||
338 (!test_and_set_bit(SOCK_ASYNC_NOSPACE, &q->sock.flags) &&
339 sock_writeable(&q->sk)))
340 mask |= POLLOUT | POLLWRNORM;
346 static inline struct sk_buff *macvtap_alloc_skb(struct sock *sk, size_t prepad,
347 size_t len, size_t linear,
348 int noblock, int *err)
352 /* Under a page? Don't bother with paged skb. */
353 if (prepad + len < PAGE_SIZE || !linear)
356 skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
361 skb_reserve(skb, prepad);
362 skb_put(skb, linear);
363 skb->data_len = len - linear;
364 skb->len += len - linear;
370 * macvtap_skb_from_vnet_hdr and macvtap_skb_to_vnet_hdr should
371 * be shared with the tun/tap driver.
373 static int macvtap_skb_from_vnet_hdr(struct sk_buff *skb,
374 struct virtio_net_hdr *vnet_hdr)
376 unsigned short gso_type = 0;
377 if (vnet_hdr->gso_type != VIRTIO_NET_HDR_GSO_NONE) {
378 switch (vnet_hdr->gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
379 case VIRTIO_NET_HDR_GSO_TCPV4:
380 gso_type = SKB_GSO_TCPV4;
382 case VIRTIO_NET_HDR_GSO_TCPV6:
383 gso_type = SKB_GSO_TCPV6;
385 case VIRTIO_NET_HDR_GSO_UDP:
386 gso_type = SKB_GSO_UDP;
392 if (vnet_hdr->gso_type & VIRTIO_NET_HDR_GSO_ECN)
393 gso_type |= SKB_GSO_TCP_ECN;
395 if (vnet_hdr->gso_size == 0)
399 if (vnet_hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
400 if (!skb_partial_csum_set(skb, vnet_hdr->csum_start,
401 vnet_hdr->csum_offset))
405 if (vnet_hdr->gso_type != VIRTIO_NET_HDR_GSO_NONE) {
406 skb_shinfo(skb)->gso_size = vnet_hdr->gso_size;
407 skb_shinfo(skb)->gso_type = gso_type;
409 /* Header must be checked, and gso_segs computed. */
410 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
411 skb_shinfo(skb)->gso_segs = 0;
416 static int macvtap_skb_to_vnet_hdr(const struct sk_buff *skb,
417 struct virtio_net_hdr *vnet_hdr)
419 memset(vnet_hdr, 0, sizeof(*vnet_hdr));
421 if (skb_is_gso(skb)) {
422 struct skb_shared_info *sinfo = skb_shinfo(skb);
424 /* This is a hint as to how much should be linear. */
425 vnet_hdr->hdr_len = skb_headlen(skb);
426 vnet_hdr->gso_size = sinfo->gso_size;
427 if (sinfo->gso_type & SKB_GSO_TCPV4)
428 vnet_hdr->gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
429 else if (sinfo->gso_type & SKB_GSO_TCPV6)
430 vnet_hdr->gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
431 else if (sinfo->gso_type & SKB_GSO_UDP)
432 vnet_hdr->gso_type = VIRTIO_NET_HDR_GSO_UDP;
435 if (sinfo->gso_type & SKB_GSO_TCP_ECN)
436 vnet_hdr->gso_type |= VIRTIO_NET_HDR_GSO_ECN;
438 vnet_hdr->gso_type = VIRTIO_NET_HDR_GSO_NONE;
440 if (skb->ip_summed == CHECKSUM_PARTIAL) {
441 vnet_hdr->flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
442 vnet_hdr->csum_start = skb->csum_start -
444 vnet_hdr->csum_offset = skb->csum_offset;
445 } /* else everything is zero */
451 /* Get packet from user space buffer */
452 static ssize_t macvtap_get_user(struct macvtap_queue *q,
453 const struct iovec *iv, size_t count,
457 struct macvlan_dev *vlan;
460 struct virtio_net_hdr vnet_hdr = { 0 };
461 int vnet_hdr_len = 0;
463 if (q->flags & IFF_VNET_HDR) {
464 vnet_hdr_len = q->vnet_hdr_sz;
467 if ((len -= vnet_hdr_len) < 0)
470 err = memcpy_fromiovecend((void *)&vnet_hdr, iv, 0,
474 if ((vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
475 vnet_hdr.csum_start + vnet_hdr.csum_offset + 2 >
477 vnet_hdr.hdr_len = vnet_hdr.csum_start +
478 vnet_hdr.csum_offset + 2;
480 if (vnet_hdr.hdr_len > len)
485 if (unlikely(len < ETH_HLEN))
488 skb = macvtap_alloc_skb(&q->sk, NET_IP_ALIGN, len, vnet_hdr.hdr_len,
493 err = skb_copy_datagram_from_iovec(skb, 0, iv, vnet_hdr_len, len);
497 skb_set_network_header(skb, ETH_HLEN);
498 skb_reset_mac_header(skb);
499 skb->protocol = eth_hdr(skb)->h_proto;
502 err = macvtap_skb_from_vnet_hdr(skb, &vnet_hdr);
508 vlan = rcu_dereference(q->vlan);
510 macvlan_start_xmit(skb, vlan->dev);
513 rcu_read_unlock_bh();
522 vlan = rcu_dereference(q->vlan);
524 netdev_get_tx_queue(vlan->dev, 0)->tx_dropped++;
525 rcu_read_unlock_bh();
530 static ssize_t macvtap_aio_write(struct kiocb *iocb, const struct iovec *iv,
531 unsigned long count, loff_t pos)
533 struct file *file = iocb->ki_filp;
534 ssize_t result = -ENOLINK;
535 struct macvtap_queue *q = file->private_data;
537 result = macvtap_get_user(q, iv, iov_length(iv, count),
538 file->f_flags & O_NONBLOCK);
542 /* Put packet to the user space buffer */
543 static ssize_t macvtap_put_user(struct macvtap_queue *q,
544 const struct sk_buff *skb,
545 const struct iovec *iv, int len)
547 struct macvlan_dev *vlan;
549 int vnet_hdr_len = 0;
551 if (q->flags & IFF_VNET_HDR) {
552 struct virtio_net_hdr vnet_hdr;
553 vnet_hdr_len = q->vnet_hdr_sz;
554 if ((len -= vnet_hdr_len) < 0)
557 ret = macvtap_skb_to_vnet_hdr(skb, &vnet_hdr);
561 if (memcpy_toiovecend(iv, (void *)&vnet_hdr, 0, sizeof(vnet_hdr)))
565 len = min_t(int, skb->len, len);
567 ret = skb_copy_datagram_const_iovec(skb, 0, iv, vnet_hdr_len, len);
570 vlan = rcu_dereference(q->vlan);
572 macvlan_count_rx(vlan, len, ret == 0, 0);
573 rcu_read_unlock_bh();
575 return ret ? ret : (len + vnet_hdr_len);
578 static ssize_t macvtap_do_read(struct macvtap_queue *q, struct kiocb *iocb,
579 const struct iovec *iv, unsigned long len,
582 DECLARE_WAITQUEUE(wait, current);
586 add_wait_queue(sk_sleep(&q->sk), &wait);
588 current->state = TASK_INTERRUPTIBLE;
590 /* Read frames from the queue */
591 skb = skb_dequeue(&q->sk.sk_receive_queue);
597 if (signal_pending(current)) {
601 /* Nothing to read, let's sleep */
605 ret = macvtap_put_user(q, skb, iv, len);
610 current->state = TASK_RUNNING;
611 remove_wait_queue(sk_sleep(&q->sk), &wait);
615 static ssize_t macvtap_aio_read(struct kiocb *iocb, const struct iovec *iv,
616 unsigned long count, loff_t pos)
618 struct file *file = iocb->ki_filp;
619 struct macvtap_queue *q = file->private_data;
620 ssize_t len, ret = 0;
622 len = iov_length(iv, count);
628 ret = macvtap_do_read(q, iocb, iv, len, file->f_flags & O_NONBLOCK);
629 ret = min_t(ssize_t, ret, len); /* XXX copied from tun.c. Why? */
635 * provide compatibility with generic tun/tap interface
637 static long macvtap_ioctl(struct file *file, unsigned int cmd,
640 struct macvtap_queue *q = file->private_data;
641 struct macvlan_dev *vlan;
642 void __user *argp = (void __user *)arg;
643 struct ifreq __user *ifr = argp;
644 unsigned int __user *up = argp;
646 int __user *sp = argp;
652 /* ignore the name, just look at flags */
653 if (get_user(u, &ifr->ifr_flags))
657 if ((u & ~IFF_VNET_HDR) != (IFF_NO_PI | IFF_TAP))
666 vlan = rcu_dereference(q->vlan);
669 rcu_read_unlock_bh();
675 if (copy_to_user(&ifr->ifr_name, q->vlan->dev->name, IFNAMSIZ) ||
676 put_user(q->flags, &ifr->ifr_flags))
682 if (put_user(IFF_TAP | IFF_NO_PI | IFF_VNET_HDR, up))
693 case TUNGETVNETHDRSZ:
699 case TUNSETVNETHDRSZ:
702 if (s < (int)sizeof(struct virtio_net_hdr))
709 /* let the user check for future flags */
710 if (arg & ~(TUN_F_CSUM | TUN_F_TSO4 | TUN_F_TSO6 |
711 TUN_F_TSO_ECN | TUN_F_UFO))
714 /* TODO: only accept frames with the features that
715 got enabled for forwarded frames */
716 if (!(q->flags & IFF_VNET_HDR))
726 static long macvtap_compat_ioctl(struct file *file, unsigned int cmd,
729 return macvtap_ioctl(file, cmd, (unsigned long)compat_ptr(arg));
733 static const struct file_operations macvtap_fops = {
734 .owner = THIS_MODULE,
735 .open = macvtap_open,
736 .release = macvtap_release,
737 .aio_read = macvtap_aio_read,
738 .aio_write = macvtap_aio_write,
739 .poll = macvtap_poll,
741 .unlocked_ioctl = macvtap_ioctl,
743 .compat_ioctl = macvtap_compat_ioctl,
747 static int macvtap_sendmsg(struct kiocb *iocb, struct socket *sock,
748 struct msghdr *m, size_t total_len)
750 struct macvtap_queue *q = container_of(sock, struct macvtap_queue, sock);
751 return macvtap_get_user(q, m->msg_iov, total_len,
752 m->msg_flags & MSG_DONTWAIT);
755 static int macvtap_recvmsg(struct kiocb *iocb, struct socket *sock,
756 struct msghdr *m, size_t total_len,
759 struct macvtap_queue *q = container_of(sock, struct macvtap_queue, sock);
761 if (flags & ~(MSG_DONTWAIT|MSG_TRUNC))
763 ret = macvtap_do_read(q, iocb, m->msg_iov, total_len,
764 flags & MSG_DONTWAIT);
765 if (ret > total_len) {
766 m->msg_flags |= MSG_TRUNC;
767 ret = flags & MSG_TRUNC ? ret : total_len;
772 /* Ops structure to mimic raw sockets with tun */
773 static const struct proto_ops macvtap_socket_ops = {
774 .sendmsg = macvtap_sendmsg,
775 .recvmsg = macvtap_recvmsg,
778 /* Get an underlying socket object from tun file. Returns error unless file is
779 * attached to a device. The returned object works like a packet socket, it
780 * can be used for sock_sendmsg/sock_recvmsg. The caller is responsible for
781 * holding a reference to the file for as long as the socket is in use. */
782 struct socket *macvtap_get_socket(struct file *file)
784 struct macvtap_queue *q;
785 if (file->f_op != &macvtap_fops)
786 return ERR_PTR(-EINVAL);
787 q = file->private_data;
789 return ERR_PTR(-EBADFD);
792 EXPORT_SYMBOL_GPL(macvtap_get_socket);
794 static int macvtap_init(void)
798 err = alloc_chrdev_region(&macvtap_major, 0,
799 MACVTAP_NUM_DEVS, "macvtap");
803 cdev_init(&macvtap_cdev, &macvtap_fops);
804 err = cdev_add(&macvtap_cdev, macvtap_major, MACVTAP_NUM_DEVS);
808 macvtap_class = class_create(THIS_MODULE, "macvtap");
809 if (IS_ERR(macvtap_class)) {
810 err = PTR_ERR(macvtap_class);
814 err = macvlan_link_register(&macvtap_link_ops);
821 class_unregister(macvtap_class);
823 cdev_del(&macvtap_cdev);
825 unregister_chrdev_region(macvtap_major, MACVTAP_NUM_DEVS);
829 module_init(macvtap_init);
831 static void macvtap_exit(void)
833 rtnl_link_unregister(&macvtap_link_ops);
834 class_unregister(macvtap_class);
835 cdev_del(&macvtap_cdev);
836 unregister_chrdev_region(macvtap_major, MACVTAP_NUM_DEVS);
838 module_exit(macvtap_exit);
840 MODULE_ALIAS_RTNL_LINK("macvtap");
841 MODULE_AUTHOR("Arnd Bergmann <arnd@arndb.de>");
842 MODULE_LICENSE("GPL");