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/init.h>
13 #include <linux/wait.h>
14 #include <linux/cdev.h>
17 #include <net/net_namespace.h>
18 #include <net/rtnetlink.h>
20 #include <linux/virtio_net.h>
23 * A macvtap queue is the central object of this driver, it connects
24 * an open character device to a macvlan interface. There can be
25 * multiple queues on one interface, which map back to queues
26 * implemented in hardware on the underlying device.
28 * macvtap_proto is used to allocate queues through the sock allocation
31 * TODO: multiqueue support is currently not implemented, even though
32 * macvtap is basically prepared for that. We will need to add this
33 * here as well as in virtio-net and qemu to get line rate on 10gbit
34 * adapters from a guest.
36 struct macvtap_queue {
39 struct macvlan_dev *vlan;
44 static struct proto macvtap_proto = {
47 .obj_size = sizeof (struct macvtap_queue),
51 * Minor number matches netdev->ifindex, so need a potentially
52 * large value. This also makes it possible to split the
53 * tap functionality out again in the future by offering it
54 * from other drivers besides macvtap. As long as every device
55 * only has one tap, the interface numbers assure that the
56 * device nodes are unique.
58 static unsigned int macvtap_major;
59 #define MACVTAP_NUM_DEVS 65536
60 static struct class *macvtap_class;
61 static struct cdev macvtap_cdev;
63 static const struct proto_ops macvtap_socket_ops;
67 * The macvtap_queue and the macvlan_dev are loosely coupled, the
68 * pointers from one to the other can only be read while rcu_read_lock
69 * or macvtap_lock is held.
71 * Both the file and the macvlan_dev hold a reference on the macvtap_queue
72 * through sock_hold(&q->sk). When the macvlan_dev goes away first,
73 * q->vlan becomes inaccessible. When the files gets closed,
74 * macvtap_get_queue() fails.
76 * There may still be references to the struct sock inside of the
77 * queue from outbound SKBs, but these never reference back to the
78 * file or the dev. The data structure is freed through __sk_free
79 * when both our references and any pending SKBs are gone.
81 static DEFINE_SPINLOCK(macvtap_lock);
84 * Choose the next free queue, for now there is only one
86 static int macvtap_set_queue(struct net_device *dev, struct file *file,
87 struct macvtap_queue *q)
89 struct macvlan_dev *vlan = netdev_priv(dev);
92 spin_lock(&macvtap_lock);
93 if (rcu_dereference(vlan->tap))
97 rcu_assign_pointer(q->vlan, vlan);
98 rcu_assign_pointer(vlan->tap, q);
102 file->private_data = q;
105 spin_unlock(&macvtap_lock);
110 * The file owning the queue got closed, give up both
111 * the reference that the files holds as well as the
112 * one from the macvlan_dev if that still exists.
114 * Using the spinlock makes sure that we don't get
115 * to the queue again after destroying it.
117 static void macvtap_put_queue(struct macvtap_queue *q)
119 struct macvlan_dev *vlan;
121 spin_lock(&macvtap_lock);
122 vlan = rcu_dereference(q->vlan);
124 rcu_assign_pointer(vlan->tap, NULL);
125 rcu_assign_pointer(q->vlan, NULL);
129 spin_unlock(&macvtap_lock);
136 * Since we only support one queue, just dereference the pointer.
138 static struct macvtap_queue *macvtap_get_queue(struct net_device *dev,
141 struct macvlan_dev *vlan = netdev_priv(dev);
143 return rcu_dereference(vlan->tap);
147 * The net_device is going away, give up the reference
148 * that it holds on the queue (all the queues one day)
149 * and safely set the pointer from the queues to NULL.
151 static void macvtap_del_queues(struct net_device *dev)
153 struct macvlan_dev *vlan = netdev_priv(dev);
154 struct macvtap_queue *q;
156 spin_lock(&macvtap_lock);
157 q = rcu_dereference(vlan->tap);
159 spin_unlock(&macvtap_lock);
163 rcu_assign_pointer(vlan->tap, NULL);
164 rcu_assign_pointer(q->vlan, NULL);
165 spin_unlock(&macvtap_lock);
172 * Forward happens for data that gets sent from one macvlan
173 * endpoint to another one in bridge mode. We just take
174 * the skb and put it into the receive queue.
176 static int macvtap_forward(struct net_device *dev, struct sk_buff *skb)
178 struct macvtap_queue *q = macvtap_get_queue(dev, skb);
182 skb_queue_tail(&q->sk.sk_receive_queue, skb);
183 wake_up_interruptible_poll(q->sk.sk_sleep, POLLIN | POLLRDNORM | POLLRDBAND);
188 * Receive is for data from the external interface (lowerdev),
189 * in case of macvtap, we can treat that the same way as
190 * forward, which macvlan cannot.
192 static int macvtap_receive(struct sk_buff *skb)
194 skb_push(skb, ETH_HLEN);
195 return macvtap_forward(skb->dev, skb);
198 static int macvtap_newlink(struct net *src_net,
199 struct net_device *dev,
201 struct nlattr *data[])
203 struct device *classdev;
207 err = macvlan_common_newlink(src_net, dev, tb, data,
208 macvtap_receive, macvtap_forward);
212 devt = MKDEV(MAJOR(macvtap_major), dev->ifindex);
214 classdev = device_create(macvtap_class, &dev->dev, devt,
215 dev, "tap%d", dev->ifindex);
216 if (IS_ERR(classdev)) {
217 err = PTR_ERR(classdev);
218 macvtap_del_queues(dev);
225 static void macvtap_dellink(struct net_device *dev,
226 struct list_head *head)
228 device_destroy(macvtap_class,
229 MKDEV(MAJOR(macvtap_major), dev->ifindex));
231 macvtap_del_queues(dev);
232 macvlan_dellink(dev, head);
235 static struct rtnl_link_ops macvtap_link_ops __read_mostly = {
237 .newlink = macvtap_newlink,
238 .dellink = macvtap_dellink,
242 static void macvtap_sock_write_space(struct sock *sk)
244 if (!sock_writeable(sk) ||
245 !test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags))
248 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
249 wake_up_interruptible_poll(sk->sk_sleep, POLLOUT | POLLWRNORM | POLLWRBAND);
252 static int macvtap_open(struct inode *inode, struct file *file)
254 struct net *net = current->nsproxy->net_ns;
255 struct net_device *dev = dev_get_by_index(net, iminor(inode));
256 struct macvtap_queue *q;
263 /* check if this is a macvtap device */
265 if (dev->rtnl_link_ops != &macvtap_link_ops)
269 q = (struct macvtap_queue *)sk_alloc(net, AF_UNSPEC, GFP_KERNEL,
274 init_waitqueue_head(&q->sock.wait);
275 q->sock.type = SOCK_RAW;
276 q->sock.state = SS_CONNECTED;
278 q->sock.ops = &macvtap_socket_ops;
279 sock_init_data(&q->sock, &q->sk);
280 q->sk.sk_write_space = macvtap_sock_write_space;
281 q->flags = IFF_VNET_HDR | IFF_NO_PI | IFF_TAP;
283 err = macvtap_set_queue(dev, file, q);
294 static int macvtap_release(struct inode *inode, struct file *file)
296 struct macvtap_queue *q = file->private_data;
297 macvtap_put_queue(q);
301 static unsigned int macvtap_poll(struct file *file, poll_table * wait)
303 struct macvtap_queue *q = file->private_data;
304 unsigned int mask = POLLERR;
310 poll_wait(file, &q->sock.wait, wait);
312 if (!skb_queue_empty(&q->sk.sk_receive_queue))
313 mask |= POLLIN | POLLRDNORM;
315 if (sock_writeable(&q->sk) ||
316 (!test_and_set_bit(SOCK_ASYNC_NOSPACE, &q->sock.flags) &&
317 sock_writeable(&q->sk)))
318 mask |= POLLOUT | POLLWRNORM;
324 static inline struct sk_buff *macvtap_alloc_skb(struct sock *sk, size_t prepad,
325 size_t len, size_t linear,
326 int noblock, int *err)
330 /* Under a page? Don't bother with paged skb. */
331 if (prepad + len < PAGE_SIZE || !linear)
334 skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
339 skb_reserve(skb, prepad);
340 skb_put(skb, linear);
341 skb->data_len = len - linear;
342 skb->len += len - linear;
348 * macvtap_skb_from_vnet_hdr and macvtap_skb_to_vnet_hdr should
349 * be shared with the tun/tap driver.
351 static int macvtap_skb_from_vnet_hdr(struct sk_buff *skb,
352 struct virtio_net_hdr *vnet_hdr)
354 unsigned short gso_type = 0;
355 if (vnet_hdr->gso_type != VIRTIO_NET_HDR_GSO_NONE) {
356 switch (vnet_hdr->gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
357 case VIRTIO_NET_HDR_GSO_TCPV4:
358 gso_type = SKB_GSO_TCPV4;
360 case VIRTIO_NET_HDR_GSO_TCPV6:
361 gso_type = SKB_GSO_TCPV6;
363 case VIRTIO_NET_HDR_GSO_UDP:
364 gso_type = SKB_GSO_UDP;
370 if (vnet_hdr->gso_type & VIRTIO_NET_HDR_GSO_ECN)
371 gso_type |= SKB_GSO_TCP_ECN;
373 if (vnet_hdr->gso_size == 0)
377 if (vnet_hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
378 if (!skb_partial_csum_set(skb, vnet_hdr->csum_start,
379 vnet_hdr->csum_offset))
383 if (vnet_hdr->gso_type != VIRTIO_NET_HDR_GSO_NONE) {
384 skb_shinfo(skb)->gso_size = vnet_hdr->gso_size;
385 skb_shinfo(skb)->gso_type = gso_type;
387 /* Header must be checked, and gso_segs computed. */
388 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
389 skb_shinfo(skb)->gso_segs = 0;
394 static int macvtap_skb_to_vnet_hdr(const struct sk_buff *skb,
395 struct virtio_net_hdr *vnet_hdr)
397 memset(vnet_hdr, 0, sizeof(*vnet_hdr));
399 if (skb_is_gso(skb)) {
400 struct skb_shared_info *sinfo = skb_shinfo(skb);
402 /* This is a hint as to how much should be linear. */
403 vnet_hdr->hdr_len = skb_headlen(skb);
404 vnet_hdr->gso_size = sinfo->gso_size;
405 if (sinfo->gso_type & SKB_GSO_TCPV4)
406 vnet_hdr->gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
407 else if (sinfo->gso_type & SKB_GSO_TCPV6)
408 vnet_hdr->gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
409 else if (sinfo->gso_type & SKB_GSO_UDP)
410 vnet_hdr->gso_type = VIRTIO_NET_HDR_GSO_UDP;
413 if (sinfo->gso_type & SKB_GSO_TCP_ECN)
414 vnet_hdr->gso_type |= VIRTIO_NET_HDR_GSO_ECN;
416 vnet_hdr->gso_type = VIRTIO_NET_HDR_GSO_NONE;
418 if (skb->ip_summed == CHECKSUM_PARTIAL) {
419 vnet_hdr->flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
420 vnet_hdr->csum_start = skb->csum_start -
422 vnet_hdr->csum_offset = skb->csum_offset;
423 } /* else everything is zero */
429 /* Get packet from user space buffer */
430 static ssize_t macvtap_get_user(struct macvtap_queue *q,
431 const struct iovec *iv, size_t count,
435 struct macvlan_dev *vlan;
438 struct virtio_net_hdr vnet_hdr = { 0 };
439 int vnet_hdr_len = 0;
441 if (q->flags & IFF_VNET_HDR) {
442 vnet_hdr_len = sizeof(vnet_hdr);
445 if ((len -= vnet_hdr_len) < 0)
448 err = memcpy_fromiovecend((void *)&vnet_hdr, iv, 0,
452 if ((vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
453 vnet_hdr.csum_start + vnet_hdr.csum_offset + 2 >
455 vnet_hdr.hdr_len = vnet_hdr.csum_start +
456 vnet_hdr.csum_offset + 2;
458 if (vnet_hdr.hdr_len > len)
463 if (unlikely(len < ETH_HLEN))
466 skb = macvtap_alloc_skb(&q->sk, NET_IP_ALIGN, len, vnet_hdr.hdr_len,
471 err = skb_copy_datagram_from_iovec(skb, 0, iv, vnet_hdr_len, len);
475 skb_set_network_header(skb, ETH_HLEN);
476 skb_reset_mac_header(skb);
477 skb->protocol = eth_hdr(skb)->h_proto;
480 err = macvtap_skb_from_vnet_hdr(skb, &vnet_hdr);
486 vlan = rcu_dereference(q->vlan);
488 macvlan_start_xmit(skb, vlan->dev);
491 rcu_read_unlock_bh();
500 vlan = rcu_dereference(q->vlan);
502 netdev_get_tx_queue(vlan->dev, 0)->tx_dropped++;
503 rcu_read_unlock_bh();
508 static ssize_t macvtap_aio_write(struct kiocb *iocb, const struct iovec *iv,
509 unsigned long count, loff_t pos)
511 struct file *file = iocb->ki_filp;
512 ssize_t result = -ENOLINK;
513 struct macvtap_queue *q = file->private_data;
515 result = macvtap_get_user(q, iv, iov_length(iv, count),
516 file->f_flags & O_NONBLOCK);
520 /* Put packet to the user space buffer */
521 static ssize_t macvtap_put_user(struct macvtap_queue *q,
522 const struct sk_buff *skb,
523 const struct iovec *iv, int len)
525 struct macvlan_dev *vlan;
527 int vnet_hdr_len = 0;
529 if (q->flags & IFF_VNET_HDR) {
530 struct virtio_net_hdr vnet_hdr;
531 vnet_hdr_len = sizeof (vnet_hdr);
532 if ((len -= vnet_hdr_len) < 0)
535 ret = macvtap_skb_to_vnet_hdr(skb, &vnet_hdr);
539 if (memcpy_toiovecend(iv, (void *)&vnet_hdr, 0, vnet_hdr_len))
543 len = min_t(int, skb->len, len);
545 ret = skb_copy_datagram_const_iovec(skb, 0, iv, vnet_hdr_len, len);
548 vlan = rcu_dereference(q->vlan);
550 macvlan_count_rx(vlan, len, ret == 0, 0);
551 rcu_read_unlock_bh();
553 return ret ? ret : (len + vnet_hdr_len);
556 static ssize_t macvtap_do_read(struct macvtap_queue *q, struct kiocb *iocb,
557 const struct iovec *iv, unsigned long len,
560 DECLARE_WAITQUEUE(wait, current);
564 add_wait_queue(q->sk.sk_sleep, &wait);
566 current->state = TASK_INTERRUPTIBLE;
568 /* Read frames from the queue */
569 skb = skb_dequeue(&q->sk.sk_receive_queue);
575 if (signal_pending(current)) {
579 /* Nothing to read, let's sleep */
583 ret = macvtap_put_user(q, skb, iv, len);
588 current->state = TASK_RUNNING;
589 remove_wait_queue(q->sk.sk_sleep, &wait);
593 static ssize_t macvtap_aio_read(struct kiocb *iocb, const struct iovec *iv,
594 unsigned long count, loff_t pos)
596 struct file *file = iocb->ki_filp;
597 struct macvtap_queue *q = file->private_data;
598 ssize_t len, ret = 0;
600 len = iov_length(iv, count);
606 ret = macvtap_do_read(q, iocb, iv, len, file->f_flags & O_NONBLOCK);
607 ret = min_t(ssize_t, ret, len); /* XXX copied from tun.c. Why? */
613 * provide compatibility with generic tun/tap interface
615 static long macvtap_ioctl(struct file *file, unsigned int cmd,
618 struct macvtap_queue *q = file->private_data;
619 struct macvlan_dev *vlan;
620 void __user *argp = (void __user *)arg;
621 struct ifreq __user *ifr = argp;
622 unsigned int __user *up = argp;
628 /* ignore the name, just look at flags */
629 if (get_user(u, &ifr->ifr_flags))
633 if ((u & ~IFF_VNET_HDR) != (IFF_NO_PI | IFF_TAP))
642 vlan = rcu_dereference(q->vlan);
645 rcu_read_unlock_bh();
651 if (copy_to_user(&ifr->ifr_name, q->vlan->dev->name, IFNAMSIZ) ||
652 put_user(q->flags, &ifr->ifr_flags))
658 if (put_user(IFF_TAP | IFF_NO_PI | IFF_VNET_HDR, up))
670 /* let the user check for future flags */
671 if (arg & ~(TUN_F_CSUM | TUN_F_TSO4 | TUN_F_TSO6 |
672 TUN_F_TSO_ECN | TUN_F_UFO))
675 /* TODO: only accept frames with the features that
676 got enabled for forwarded frames */
677 if (!(q->flags & IFF_VNET_HDR))
687 static long macvtap_compat_ioctl(struct file *file, unsigned int cmd,
690 return macvtap_ioctl(file, cmd, (unsigned long)compat_ptr(arg));
694 static const struct file_operations macvtap_fops = {
695 .owner = THIS_MODULE,
696 .open = macvtap_open,
697 .release = macvtap_release,
698 .aio_read = macvtap_aio_read,
699 .aio_write = macvtap_aio_write,
700 .poll = macvtap_poll,
702 .unlocked_ioctl = macvtap_ioctl,
704 .compat_ioctl = macvtap_compat_ioctl,
708 static int macvtap_sendmsg(struct kiocb *iocb, struct socket *sock,
709 struct msghdr *m, size_t total_len)
711 struct macvtap_queue *q = container_of(sock, struct macvtap_queue, sock);
712 return macvtap_get_user(q, m->msg_iov, total_len,
713 m->msg_flags & MSG_DONTWAIT);
716 static int macvtap_recvmsg(struct kiocb *iocb, struct socket *sock,
717 struct msghdr *m, size_t total_len,
720 struct macvtap_queue *q = container_of(sock, struct macvtap_queue, sock);
722 if (flags & ~(MSG_DONTWAIT|MSG_TRUNC))
724 ret = macvtap_do_read(q, iocb, m->msg_iov, total_len,
725 flags & MSG_DONTWAIT);
726 if (ret > total_len) {
727 m->msg_flags |= MSG_TRUNC;
728 ret = flags & MSG_TRUNC ? ret : total_len;
733 /* Ops structure to mimic raw sockets with tun */
734 static const struct proto_ops macvtap_socket_ops = {
735 .sendmsg = macvtap_sendmsg,
736 .recvmsg = macvtap_recvmsg,
739 /* Get an underlying socket object from tun file. Returns error unless file is
740 * attached to a device. The returned object works like a packet socket, it
741 * can be used for sock_sendmsg/sock_recvmsg. The caller is responsible for
742 * holding a reference to the file for as long as the socket is in use. */
743 struct socket *macvtap_get_socket(struct file *file)
745 struct macvtap_queue *q;
746 if (file->f_op != &macvtap_fops)
747 return ERR_PTR(-EINVAL);
748 q = file->private_data;
750 return ERR_PTR(-EBADFD);
753 EXPORT_SYMBOL_GPL(macvtap_get_socket);
755 static int macvtap_init(void)
759 err = alloc_chrdev_region(&macvtap_major, 0,
760 MACVTAP_NUM_DEVS, "macvtap");
764 cdev_init(&macvtap_cdev, &macvtap_fops);
765 err = cdev_add(&macvtap_cdev, macvtap_major, MACVTAP_NUM_DEVS);
769 macvtap_class = class_create(THIS_MODULE, "macvtap");
770 if (IS_ERR(macvtap_class)) {
771 err = PTR_ERR(macvtap_class);
775 err = macvlan_link_register(&macvtap_link_ops);
782 class_unregister(macvtap_class);
784 cdev_del(&macvtap_cdev);
786 unregister_chrdev_region(macvtap_major, MACVTAP_NUM_DEVS);
790 module_init(macvtap_init);
792 static void macvtap_exit(void)
794 rtnl_link_unregister(&macvtap_link_ops);
795 class_unregister(macvtap_class);
796 cdev_del(&macvtap_cdev);
797 unregister_chrdev_region(macvtap_major, MACVTAP_NUM_DEVS);
799 module_exit(macvtap_exit);
801 MODULE_ALIAS_RTNL_LINK("macvtap");
802 MODULE_AUTHOR("Arnd Bergmann <arnd@arndb.de>");
803 MODULE_LICENSE("GPL");