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 {
41 struct macvlan_dev *vlan;
46 static struct proto macvtap_proto = {
49 .obj_size = sizeof (struct macvtap_queue),
53 * Minor number matches netdev->ifindex, so need a potentially
54 * large value. This also makes it possible to split the
55 * tap functionality out again in the future by offering it
56 * from other drivers besides macvtap. As long as every device
57 * only has one tap, the interface numbers assure that the
58 * device nodes are unique.
60 static unsigned int macvtap_major;
61 #define MACVTAP_NUM_DEVS 65536
62 static struct class *macvtap_class;
63 static struct cdev macvtap_cdev;
65 static const struct proto_ops macvtap_socket_ops;
69 * The macvtap_queue and the macvlan_dev are loosely coupled, the
70 * pointers from one to the other can only be read while rcu_read_lock
71 * or macvtap_lock is held.
73 * Both the file and the macvlan_dev hold a reference on the macvtap_queue
74 * through sock_hold(&q->sk). When the macvlan_dev goes away first,
75 * q->vlan becomes inaccessible. When the files gets closed,
76 * macvtap_get_queue() fails.
78 * There may still be references to the struct sock inside of the
79 * queue from outbound SKBs, but these never reference back to the
80 * file or the dev. The data structure is freed through __sk_free
81 * when both our references and any pending SKBs are gone.
83 static DEFINE_SPINLOCK(macvtap_lock);
86 * Choose the next free queue, for now there is only one
88 static int macvtap_set_queue(struct net_device *dev, struct file *file,
89 struct macvtap_queue *q)
91 struct macvlan_dev *vlan = netdev_priv(dev);
94 spin_lock(&macvtap_lock);
95 if (rcu_dereference(vlan->tap))
99 rcu_assign_pointer(q->vlan, vlan);
100 rcu_assign_pointer(vlan->tap, q);
104 file->private_data = q;
107 spin_unlock(&macvtap_lock);
112 * The file owning the queue got closed, give up both
113 * the reference that the files holds as well as the
114 * one from the macvlan_dev if that still exists.
116 * Using the spinlock makes sure that we don't get
117 * to the queue again after destroying it.
119 static void macvtap_put_queue(struct macvtap_queue *q)
121 struct macvlan_dev *vlan;
123 spin_lock(&macvtap_lock);
124 vlan = rcu_dereference(q->vlan);
126 rcu_assign_pointer(vlan->tap, NULL);
127 rcu_assign_pointer(q->vlan, NULL);
131 spin_unlock(&macvtap_lock);
138 * Since we only support one queue, just dereference the pointer.
140 static struct macvtap_queue *macvtap_get_queue(struct net_device *dev,
143 struct macvlan_dev *vlan = netdev_priv(dev);
145 return rcu_dereference(vlan->tap);
149 * The net_device is going away, give up the reference
150 * that it holds on the queue (all the queues one day)
151 * and safely set the pointer from the queues to NULL.
153 static void macvtap_del_queues(struct net_device *dev)
155 struct macvlan_dev *vlan = netdev_priv(dev);
156 struct macvtap_queue *q;
158 spin_lock(&macvtap_lock);
159 q = rcu_dereference(vlan->tap);
161 spin_unlock(&macvtap_lock);
165 rcu_assign_pointer(vlan->tap, NULL);
166 rcu_assign_pointer(q->vlan, NULL);
167 spin_unlock(&macvtap_lock);
174 * Forward happens for data that gets sent from one macvlan
175 * endpoint to another one in bridge mode. We just take
176 * the skb and put it into the receive queue.
178 static int macvtap_forward(struct net_device *dev, struct sk_buff *skb)
180 struct macvtap_queue *q = macvtap_get_queue(dev, skb);
184 skb_queue_tail(&q->sk.sk_receive_queue, skb);
185 wake_up_interruptible_poll(sk_sleep(&q->sk), POLLIN | POLLRDNORM | POLLRDBAND);
190 * Receive is for data from the external interface (lowerdev),
191 * in case of macvtap, we can treat that the same way as
192 * forward, which macvlan cannot.
194 static int macvtap_receive(struct sk_buff *skb)
196 skb_push(skb, ETH_HLEN);
197 return macvtap_forward(skb->dev, skb);
200 static int macvtap_newlink(struct net *src_net,
201 struct net_device *dev,
203 struct nlattr *data[])
205 struct device *classdev;
209 err = macvlan_common_newlink(src_net, dev, tb, data,
210 macvtap_receive, macvtap_forward);
214 devt = MKDEV(MAJOR(macvtap_major), dev->ifindex);
216 classdev = device_create(macvtap_class, &dev->dev, devt,
217 dev, "tap%d", dev->ifindex);
218 if (IS_ERR(classdev)) {
219 err = PTR_ERR(classdev);
220 macvtap_del_queues(dev);
227 static void macvtap_dellink(struct net_device *dev,
228 struct list_head *head)
230 device_destroy(macvtap_class,
231 MKDEV(MAJOR(macvtap_major), dev->ifindex));
233 macvtap_del_queues(dev);
234 macvlan_dellink(dev, head);
237 static struct rtnl_link_ops macvtap_link_ops __read_mostly = {
239 .newlink = macvtap_newlink,
240 .dellink = macvtap_dellink,
244 static void macvtap_sock_write_space(struct sock *sk)
246 wait_queue_head_t *wqueue;
248 if (!sock_writeable(sk) ||
249 !test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags))
252 wqueue = sk_sleep(sk);
253 if (wqueue && waitqueue_active(wqueue))
254 wake_up_interruptible_poll(wqueue, POLLOUT | POLLWRNORM | POLLWRBAND);
257 static int macvtap_open(struct inode *inode, struct file *file)
259 struct net *net = current->nsproxy->net_ns;
260 struct net_device *dev = dev_get_by_index(net, iminor(inode));
261 struct macvtap_queue *q;
268 /* check if this is a macvtap device */
270 if (dev->rtnl_link_ops != &macvtap_link_ops)
274 q = (struct macvtap_queue *)sk_alloc(net, AF_UNSPEC, GFP_KERNEL,
280 init_waitqueue_head(&q->wq.wait);
281 q->sock.type = SOCK_RAW;
282 q->sock.state = SS_CONNECTED;
284 q->sock.ops = &macvtap_socket_ops;
285 sock_init_data(&q->sock, &q->sk);
286 q->sk.sk_write_space = macvtap_sock_write_space;
287 q->flags = IFF_VNET_HDR | IFF_NO_PI | IFF_TAP;
289 err = macvtap_set_queue(dev, file, q);
300 static int macvtap_release(struct inode *inode, struct file *file)
302 struct macvtap_queue *q = file->private_data;
303 macvtap_put_queue(q);
307 static unsigned int macvtap_poll(struct file *file, poll_table * wait)
309 struct macvtap_queue *q = file->private_data;
310 unsigned int mask = POLLERR;
316 poll_wait(file, &q->wq.wait, wait);
318 if (!skb_queue_empty(&q->sk.sk_receive_queue))
319 mask |= POLLIN | POLLRDNORM;
321 if (sock_writeable(&q->sk) ||
322 (!test_and_set_bit(SOCK_ASYNC_NOSPACE, &q->sock.flags) &&
323 sock_writeable(&q->sk)))
324 mask |= POLLOUT | POLLWRNORM;
330 static inline struct sk_buff *macvtap_alloc_skb(struct sock *sk, size_t prepad,
331 size_t len, size_t linear,
332 int noblock, int *err)
336 /* Under a page? Don't bother with paged skb. */
337 if (prepad + len < PAGE_SIZE || !linear)
340 skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
345 skb_reserve(skb, prepad);
346 skb_put(skb, linear);
347 skb->data_len = len - linear;
348 skb->len += len - linear;
354 * macvtap_skb_from_vnet_hdr and macvtap_skb_to_vnet_hdr should
355 * be shared with the tun/tap driver.
357 static int macvtap_skb_from_vnet_hdr(struct sk_buff *skb,
358 struct virtio_net_hdr *vnet_hdr)
360 unsigned short gso_type = 0;
361 if (vnet_hdr->gso_type != VIRTIO_NET_HDR_GSO_NONE) {
362 switch (vnet_hdr->gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
363 case VIRTIO_NET_HDR_GSO_TCPV4:
364 gso_type = SKB_GSO_TCPV4;
366 case VIRTIO_NET_HDR_GSO_TCPV6:
367 gso_type = SKB_GSO_TCPV6;
369 case VIRTIO_NET_HDR_GSO_UDP:
370 gso_type = SKB_GSO_UDP;
376 if (vnet_hdr->gso_type & VIRTIO_NET_HDR_GSO_ECN)
377 gso_type |= SKB_GSO_TCP_ECN;
379 if (vnet_hdr->gso_size == 0)
383 if (vnet_hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
384 if (!skb_partial_csum_set(skb, vnet_hdr->csum_start,
385 vnet_hdr->csum_offset))
389 if (vnet_hdr->gso_type != VIRTIO_NET_HDR_GSO_NONE) {
390 skb_shinfo(skb)->gso_size = vnet_hdr->gso_size;
391 skb_shinfo(skb)->gso_type = gso_type;
393 /* Header must be checked, and gso_segs computed. */
394 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
395 skb_shinfo(skb)->gso_segs = 0;
400 static int macvtap_skb_to_vnet_hdr(const struct sk_buff *skb,
401 struct virtio_net_hdr *vnet_hdr)
403 memset(vnet_hdr, 0, sizeof(*vnet_hdr));
405 if (skb_is_gso(skb)) {
406 struct skb_shared_info *sinfo = skb_shinfo(skb);
408 /* This is a hint as to how much should be linear. */
409 vnet_hdr->hdr_len = skb_headlen(skb);
410 vnet_hdr->gso_size = sinfo->gso_size;
411 if (sinfo->gso_type & SKB_GSO_TCPV4)
412 vnet_hdr->gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
413 else if (sinfo->gso_type & SKB_GSO_TCPV6)
414 vnet_hdr->gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
415 else if (sinfo->gso_type & SKB_GSO_UDP)
416 vnet_hdr->gso_type = VIRTIO_NET_HDR_GSO_UDP;
419 if (sinfo->gso_type & SKB_GSO_TCP_ECN)
420 vnet_hdr->gso_type |= VIRTIO_NET_HDR_GSO_ECN;
422 vnet_hdr->gso_type = VIRTIO_NET_HDR_GSO_NONE;
424 if (skb->ip_summed == CHECKSUM_PARTIAL) {
425 vnet_hdr->flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
426 vnet_hdr->csum_start = skb->csum_start -
428 vnet_hdr->csum_offset = skb->csum_offset;
429 } /* else everything is zero */
435 /* Get packet from user space buffer */
436 static ssize_t macvtap_get_user(struct macvtap_queue *q,
437 const struct iovec *iv, size_t count,
441 struct macvlan_dev *vlan;
444 struct virtio_net_hdr vnet_hdr = { 0 };
445 int vnet_hdr_len = 0;
447 if (q->flags & IFF_VNET_HDR) {
448 vnet_hdr_len = sizeof(vnet_hdr);
451 if ((len -= vnet_hdr_len) < 0)
454 err = memcpy_fromiovecend((void *)&vnet_hdr, iv, 0,
458 if ((vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
459 vnet_hdr.csum_start + vnet_hdr.csum_offset + 2 >
461 vnet_hdr.hdr_len = vnet_hdr.csum_start +
462 vnet_hdr.csum_offset + 2;
464 if (vnet_hdr.hdr_len > len)
469 if (unlikely(len < ETH_HLEN))
472 skb = macvtap_alloc_skb(&q->sk, NET_IP_ALIGN, len, vnet_hdr.hdr_len,
477 err = skb_copy_datagram_from_iovec(skb, 0, iv, vnet_hdr_len, len);
481 skb_set_network_header(skb, ETH_HLEN);
482 skb_reset_mac_header(skb);
483 skb->protocol = eth_hdr(skb)->h_proto;
486 err = macvtap_skb_from_vnet_hdr(skb, &vnet_hdr);
492 vlan = rcu_dereference(q->vlan);
494 macvlan_start_xmit(skb, vlan->dev);
497 rcu_read_unlock_bh();
506 vlan = rcu_dereference(q->vlan);
508 netdev_get_tx_queue(vlan->dev, 0)->tx_dropped++;
509 rcu_read_unlock_bh();
514 static ssize_t macvtap_aio_write(struct kiocb *iocb, const struct iovec *iv,
515 unsigned long count, loff_t pos)
517 struct file *file = iocb->ki_filp;
518 ssize_t result = -ENOLINK;
519 struct macvtap_queue *q = file->private_data;
521 result = macvtap_get_user(q, iv, iov_length(iv, count),
522 file->f_flags & O_NONBLOCK);
526 /* Put packet to the user space buffer */
527 static ssize_t macvtap_put_user(struct macvtap_queue *q,
528 const struct sk_buff *skb,
529 const struct iovec *iv, int len)
531 struct macvlan_dev *vlan;
533 int vnet_hdr_len = 0;
535 if (q->flags & IFF_VNET_HDR) {
536 struct virtio_net_hdr vnet_hdr;
537 vnet_hdr_len = sizeof (vnet_hdr);
538 if ((len -= vnet_hdr_len) < 0)
541 ret = macvtap_skb_to_vnet_hdr(skb, &vnet_hdr);
545 if (memcpy_toiovecend(iv, (void *)&vnet_hdr, 0, vnet_hdr_len))
549 len = min_t(int, skb->len, len);
551 ret = skb_copy_datagram_const_iovec(skb, 0, iv, vnet_hdr_len, len);
554 vlan = rcu_dereference(q->vlan);
556 macvlan_count_rx(vlan, len, ret == 0, 0);
557 rcu_read_unlock_bh();
559 return ret ? ret : (len + vnet_hdr_len);
562 static ssize_t macvtap_do_read(struct macvtap_queue *q, struct kiocb *iocb,
563 const struct iovec *iv, unsigned long len,
566 DECLARE_WAITQUEUE(wait, current);
570 add_wait_queue(sk_sleep(&q->sk), &wait);
572 current->state = TASK_INTERRUPTIBLE;
574 /* Read frames from the queue */
575 skb = skb_dequeue(&q->sk.sk_receive_queue);
581 if (signal_pending(current)) {
585 /* Nothing to read, let's sleep */
589 ret = macvtap_put_user(q, skb, iv, len);
594 current->state = TASK_RUNNING;
595 remove_wait_queue(sk_sleep(&q->sk), &wait);
599 static ssize_t macvtap_aio_read(struct kiocb *iocb, const struct iovec *iv,
600 unsigned long count, loff_t pos)
602 struct file *file = iocb->ki_filp;
603 struct macvtap_queue *q = file->private_data;
604 ssize_t len, ret = 0;
606 len = iov_length(iv, count);
612 ret = macvtap_do_read(q, iocb, iv, len, file->f_flags & O_NONBLOCK);
613 ret = min_t(ssize_t, ret, len); /* XXX copied from tun.c. Why? */
619 * provide compatibility with generic tun/tap interface
621 static long macvtap_ioctl(struct file *file, unsigned int cmd,
624 struct macvtap_queue *q = file->private_data;
625 struct macvlan_dev *vlan;
626 void __user *argp = (void __user *)arg;
627 struct ifreq __user *ifr = argp;
628 unsigned int __user *up = argp;
634 /* ignore the name, just look at flags */
635 if (get_user(u, &ifr->ifr_flags))
639 if ((u & ~IFF_VNET_HDR) != (IFF_NO_PI | IFF_TAP))
648 vlan = rcu_dereference(q->vlan);
651 rcu_read_unlock_bh();
657 if (copy_to_user(&ifr->ifr_name, q->vlan->dev->name, IFNAMSIZ) ||
658 put_user(q->flags, &ifr->ifr_flags))
664 if (put_user(IFF_TAP | IFF_NO_PI | IFF_VNET_HDR, up))
676 /* let the user check for future flags */
677 if (arg & ~(TUN_F_CSUM | TUN_F_TSO4 | TUN_F_TSO6 |
678 TUN_F_TSO_ECN | TUN_F_UFO))
681 /* TODO: only accept frames with the features that
682 got enabled for forwarded frames */
683 if (!(q->flags & IFF_VNET_HDR))
693 static long macvtap_compat_ioctl(struct file *file, unsigned int cmd,
696 return macvtap_ioctl(file, cmd, (unsigned long)compat_ptr(arg));
700 static const struct file_operations macvtap_fops = {
701 .owner = THIS_MODULE,
702 .open = macvtap_open,
703 .release = macvtap_release,
704 .aio_read = macvtap_aio_read,
705 .aio_write = macvtap_aio_write,
706 .poll = macvtap_poll,
708 .unlocked_ioctl = macvtap_ioctl,
710 .compat_ioctl = macvtap_compat_ioctl,
714 static int macvtap_sendmsg(struct kiocb *iocb, struct socket *sock,
715 struct msghdr *m, size_t total_len)
717 struct macvtap_queue *q = container_of(sock, struct macvtap_queue, sock);
718 return macvtap_get_user(q, m->msg_iov, total_len,
719 m->msg_flags & MSG_DONTWAIT);
722 static int macvtap_recvmsg(struct kiocb *iocb, struct socket *sock,
723 struct msghdr *m, size_t total_len,
726 struct macvtap_queue *q = container_of(sock, struct macvtap_queue, sock);
728 if (flags & ~(MSG_DONTWAIT|MSG_TRUNC))
730 ret = macvtap_do_read(q, iocb, m->msg_iov, total_len,
731 flags & MSG_DONTWAIT);
732 if (ret > total_len) {
733 m->msg_flags |= MSG_TRUNC;
734 ret = flags & MSG_TRUNC ? ret : total_len;
739 /* Ops structure to mimic raw sockets with tun */
740 static const struct proto_ops macvtap_socket_ops = {
741 .sendmsg = macvtap_sendmsg,
742 .recvmsg = macvtap_recvmsg,
745 /* Get an underlying socket object from tun file. Returns error unless file is
746 * attached to a device. The returned object works like a packet socket, it
747 * can be used for sock_sendmsg/sock_recvmsg. The caller is responsible for
748 * holding a reference to the file for as long as the socket is in use. */
749 struct socket *macvtap_get_socket(struct file *file)
751 struct macvtap_queue *q;
752 if (file->f_op != &macvtap_fops)
753 return ERR_PTR(-EINVAL);
754 q = file->private_data;
756 return ERR_PTR(-EBADFD);
759 EXPORT_SYMBOL_GPL(macvtap_get_socket);
761 static int macvtap_init(void)
765 err = alloc_chrdev_region(&macvtap_major, 0,
766 MACVTAP_NUM_DEVS, "macvtap");
770 cdev_init(&macvtap_cdev, &macvtap_fops);
771 err = cdev_add(&macvtap_cdev, macvtap_major, MACVTAP_NUM_DEVS);
775 macvtap_class = class_create(THIS_MODULE, "macvtap");
776 if (IS_ERR(macvtap_class)) {
777 err = PTR_ERR(macvtap_class);
781 err = macvlan_link_register(&macvtap_link_ops);
788 class_unregister(macvtap_class);
790 cdev_del(&macvtap_cdev);
792 unregister_chrdev_region(macvtap_major, MACVTAP_NUM_DEVS);
796 module_init(macvtap_init);
798 static void macvtap_exit(void)
800 rtnl_link_unregister(&macvtap_link_ops);
801 class_unregister(macvtap_class);
802 cdev_del(&macvtap_cdev);
803 unregister_chrdev_region(macvtap_major, MACVTAP_NUM_DEVS);
805 module_exit(macvtap_exit);
807 MODULE_ALIAS_RTNL_LINK("macvtap");
808 MODULE_AUTHOR("Arnd Bergmann <arnd@arndb.de>");
809 MODULE_LICENSE("GPL");