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 * get_slot: return a [unused/occupied] slot in vlan->taps[]:
88 * - if 'q' is NULL, return the first empty slot;
89 * - otherwise, return the slot this pointer occupies.
91 static int get_slot(struct macvlan_dev *vlan, struct macvtap_queue *q)
95 for (i = 0; i < MAX_MACVTAP_QUEUES; i++) {
96 if (rcu_dereference(vlan->taps[i]) == q)
100 /* Should never happen */
104 static int macvtap_set_queue(struct net_device *dev, struct file *file,
105 struct macvtap_queue *q)
107 struct macvlan_dev *vlan = netdev_priv(dev);
111 spin_lock(&macvtap_lock);
112 if (vlan->numvtaps == MAX_MACVTAP_QUEUES)
116 index = get_slot(vlan, NULL);
117 rcu_assign_pointer(q->vlan, vlan);
118 rcu_assign_pointer(vlan->taps[index], q);
122 file->private_data = q;
127 spin_unlock(&macvtap_lock);
132 * The file owning the queue got closed, give up both
133 * the reference that the files holds as well as the
134 * one from the macvlan_dev if that still exists.
136 * Using the spinlock makes sure that we don't get
137 * to the queue again after destroying it.
139 static void macvtap_put_queue(struct macvtap_queue *q)
141 struct macvlan_dev *vlan;
143 spin_lock(&macvtap_lock);
144 vlan = rcu_dereference(q->vlan);
146 int index = get_slot(vlan, q);
148 rcu_assign_pointer(vlan->taps[index], NULL);
149 rcu_assign_pointer(q->vlan, NULL);
154 spin_unlock(&macvtap_lock);
161 * Select a queue based on the rxq of the device on which this packet
162 * arrived. If the incoming device is not mq, calculate a flow hash
163 * to select a queue. If all fails, find the first available queue.
164 * Cache vlan->numvtaps since it can become zero during the execution
167 static struct macvtap_queue *macvtap_get_queue(struct net_device *dev,
170 struct macvlan_dev *vlan = netdev_priv(dev);
171 struct macvtap_queue *tap = NULL;
172 int numvtaps = vlan->numvtaps;
178 if (likely(skb_rx_queue_recorded(skb))) {
179 rxq = skb_get_rx_queue(skb);
181 while (unlikely(rxq >= numvtaps))
184 tap = rcu_dereference(vlan->taps[rxq]);
189 /* Check if we can use flow to select a queue */
190 rxq = skb_get_rxhash(skb);
192 tap = rcu_dereference(vlan->taps[rxq % numvtaps]);
197 /* Everything failed - find first available queue */
198 for (rxq = 0; rxq < MAX_MACVTAP_QUEUES; rxq++) {
199 tap = rcu_dereference(vlan->taps[rxq]);
209 * The net_device is going away, give up the reference
210 * that it holds on all queues and safely set the pointer
211 * from the queues to NULL.
213 static void macvtap_del_queues(struct net_device *dev)
215 struct macvlan_dev *vlan = netdev_priv(dev);
216 struct macvtap_queue *q, *qlist[MAX_MACVTAP_QUEUES];
219 /* macvtap_put_queue can free some slots, so go through all slots */
220 spin_lock(&macvtap_lock);
221 for (i = 0; i < MAX_MACVTAP_QUEUES && vlan->numvtaps; i++) {
222 q = rcu_dereference(vlan->taps[i]);
225 rcu_assign_pointer(vlan->taps[i], NULL);
226 rcu_assign_pointer(q->vlan, NULL);
230 BUG_ON(vlan->numvtaps != 0);
231 spin_unlock(&macvtap_lock);
235 for (--j; j >= 0; j--)
236 sock_put(&qlist[j]->sk);
240 * Forward happens for data that gets sent from one macvlan
241 * endpoint to another one in bridge mode. We just take
242 * the skb and put it into the receive queue.
244 static int macvtap_forward(struct net_device *dev, struct sk_buff *skb)
246 struct macvtap_queue *q = macvtap_get_queue(dev, skb);
250 if (skb_queue_len(&q->sk.sk_receive_queue) >= dev->tx_queue_len)
253 skb_queue_tail(&q->sk.sk_receive_queue, skb);
254 wake_up_interruptible_poll(sk_sleep(&q->sk), POLLIN | POLLRDNORM | POLLRDBAND);
255 return NET_RX_SUCCESS;
263 * Receive is for data from the external interface (lowerdev),
264 * in case of macvtap, we can treat that the same way as
265 * forward, which macvlan cannot.
267 static int macvtap_receive(struct sk_buff *skb)
269 skb_push(skb, ETH_HLEN);
270 return macvtap_forward(skb->dev, skb);
273 static int macvtap_newlink(struct net *src_net,
274 struct net_device *dev,
276 struct nlattr *data[])
278 struct device *classdev;
282 err = macvlan_common_newlink(src_net, dev, tb, data,
283 macvtap_receive, macvtap_forward);
287 devt = MKDEV(MAJOR(macvtap_major), dev->ifindex);
289 classdev = device_create(macvtap_class, &dev->dev, devt,
290 dev, "tap%d", dev->ifindex);
291 if (IS_ERR(classdev)) {
292 err = PTR_ERR(classdev);
293 macvtap_del_queues(dev);
300 static void macvtap_dellink(struct net_device *dev,
301 struct list_head *head)
303 device_destroy(macvtap_class,
304 MKDEV(MAJOR(macvtap_major), dev->ifindex));
306 macvtap_del_queues(dev);
307 macvlan_dellink(dev, head);
310 static void macvtap_setup(struct net_device *dev)
312 macvlan_common_setup(dev);
313 dev->tx_queue_len = TUN_READQ_SIZE;
316 static struct rtnl_link_ops macvtap_link_ops __read_mostly = {
318 .setup = macvtap_setup,
319 .newlink = macvtap_newlink,
320 .dellink = macvtap_dellink,
324 static void macvtap_sock_write_space(struct sock *sk)
326 wait_queue_head_t *wqueue;
328 if (!sock_writeable(sk) ||
329 !test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags))
332 wqueue = sk_sleep(sk);
333 if (wqueue && waitqueue_active(wqueue))
334 wake_up_interruptible_poll(wqueue, POLLOUT | POLLWRNORM | POLLWRBAND);
337 static int macvtap_open(struct inode *inode, struct file *file)
339 struct net *net = current->nsproxy->net_ns;
340 struct net_device *dev = dev_get_by_index(net, iminor(inode));
341 struct macvtap_queue *q;
348 /* check if this is a macvtap device */
350 if (dev->rtnl_link_ops != &macvtap_link_ops)
354 q = (struct macvtap_queue *)sk_alloc(net, AF_UNSPEC, GFP_KERNEL,
360 init_waitqueue_head(&q->wq.wait);
361 q->sock.type = SOCK_RAW;
362 q->sock.state = SS_CONNECTED;
364 q->sock.ops = &macvtap_socket_ops;
365 sock_init_data(&q->sock, &q->sk);
366 q->sk.sk_write_space = macvtap_sock_write_space;
367 q->flags = IFF_VNET_HDR | IFF_NO_PI | IFF_TAP;
368 q->vnet_hdr_sz = sizeof(struct virtio_net_hdr);
370 err = macvtap_set_queue(dev, file, q);
381 static int macvtap_release(struct inode *inode, struct file *file)
383 struct macvtap_queue *q = file->private_data;
384 macvtap_put_queue(q);
388 static unsigned int macvtap_poll(struct file *file, poll_table * wait)
390 struct macvtap_queue *q = file->private_data;
391 unsigned int mask = POLLERR;
397 poll_wait(file, &q->wq.wait, wait);
399 if (!skb_queue_empty(&q->sk.sk_receive_queue))
400 mask |= POLLIN | POLLRDNORM;
402 if (sock_writeable(&q->sk) ||
403 (!test_and_set_bit(SOCK_ASYNC_NOSPACE, &q->sock.flags) &&
404 sock_writeable(&q->sk)))
405 mask |= POLLOUT | POLLWRNORM;
411 static inline struct sk_buff *macvtap_alloc_skb(struct sock *sk, size_t prepad,
412 size_t len, size_t linear,
413 int noblock, int *err)
417 /* Under a page? Don't bother with paged skb. */
418 if (prepad + len < PAGE_SIZE || !linear)
421 skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
426 skb_reserve(skb, prepad);
427 skb_put(skb, linear);
428 skb->data_len = len - linear;
429 skb->len += len - linear;
435 * macvtap_skb_from_vnet_hdr and macvtap_skb_to_vnet_hdr should
436 * be shared with the tun/tap driver.
438 static int macvtap_skb_from_vnet_hdr(struct sk_buff *skb,
439 struct virtio_net_hdr *vnet_hdr)
441 unsigned short gso_type = 0;
442 if (vnet_hdr->gso_type != VIRTIO_NET_HDR_GSO_NONE) {
443 switch (vnet_hdr->gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
444 case VIRTIO_NET_HDR_GSO_TCPV4:
445 gso_type = SKB_GSO_TCPV4;
447 case VIRTIO_NET_HDR_GSO_TCPV6:
448 gso_type = SKB_GSO_TCPV6;
450 case VIRTIO_NET_HDR_GSO_UDP:
451 gso_type = SKB_GSO_UDP;
457 if (vnet_hdr->gso_type & VIRTIO_NET_HDR_GSO_ECN)
458 gso_type |= SKB_GSO_TCP_ECN;
460 if (vnet_hdr->gso_size == 0)
464 if (vnet_hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
465 if (!skb_partial_csum_set(skb, vnet_hdr->csum_start,
466 vnet_hdr->csum_offset))
470 if (vnet_hdr->gso_type != VIRTIO_NET_HDR_GSO_NONE) {
471 skb_shinfo(skb)->gso_size = vnet_hdr->gso_size;
472 skb_shinfo(skb)->gso_type = gso_type;
474 /* Header must be checked, and gso_segs computed. */
475 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
476 skb_shinfo(skb)->gso_segs = 0;
481 static int macvtap_skb_to_vnet_hdr(const struct sk_buff *skb,
482 struct virtio_net_hdr *vnet_hdr)
484 memset(vnet_hdr, 0, sizeof(*vnet_hdr));
486 if (skb_is_gso(skb)) {
487 struct skb_shared_info *sinfo = skb_shinfo(skb);
489 /* This is a hint as to how much should be linear. */
490 vnet_hdr->hdr_len = skb_headlen(skb);
491 vnet_hdr->gso_size = sinfo->gso_size;
492 if (sinfo->gso_type & SKB_GSO_TCPV4)
493 vnet_hdr->gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
494 else if (sinfo->gso_type & SKB_GSO_TCPV6)
495 vnet_hdr->gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
496 else if (sinfo->gso_type & SKB_GSO_UDP)
497 vnet_hdr->gso_type = VIRTIO_NET_HDR_GSO_UDP;
500 if (sinfo->gso_type & SKB_GSO_TCP_ECN)
501 vnet_hdr->gso_type |= VIRTIO_NET_HDR_GSO_ECN;
503 vnet_hdr->gso_type = VIRTIO_NET_HDR_GSO_NONE;
505 if (skb->ip_summed == CHECKSUM_PARTIAL) {
506 vnet_hdr->flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
507 vnet_hdr->csum_start = skb->csum_start -
509 vnet_hdr->csum_offset = skb->csum_offset;
510 } /* else everything is zero */
516 /* Get packet from user space buffer */
517 static ssize_t macvtap_get_user(struct macvtap_queue *q,
518 const struct iovec *iv, size_t count,
522 struct macvlan_dev *vlan;
525 struct virtio_net_hdr vnet_hdr = { 0 };
526 int vnet_hdr_len = 0;
528 if (q->flags & IFF_VNET_HDR) {
529 vnet_hdr_len = q->vnet_hdr_sz;
532 if ((len -= vnet_hdr_len) < 0)
535 err = memcpy_fromiovecend((void *)&vnet_hdr, iv, 0,
539 if ((vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
540 vnet_hdr.csum_start + vnet_hdr.csum_offset + 2 >
542 vnet_hdr.hdr_len = vnet_hdr.csum_start +
543 vnet_hdr.csum_offset + 2;
545 if (vnet_hdr.hdr_len > len)
550 if (unlikely(len < ETH_HLEN))
553 skb = macvtap_alloc_skb(&q->sk, NET_IP_ALIGN, len, vnet_hdr.hdr_len,
558 err = skb_copy_datagram_from_iovec(skb, 0, iv, vnet_hdr_len, len);
562 skb_set_network_header(skb, ETH_HLEN);
563 skb_reset_mac_header(skb);
564 skb->protocol = eth_hdr(skb)->h_proto;
567 err = macvtap_skb_from_vnet_hdr(skb, &vnet_hdr);
573 vlan = rcu_dereference(q->vlan);
575 macvlan_start_xmit(skb, vlan->dev);
578 rcu_read_unlock_bh();
587 vlan = rcu_dereference(q->vlan);
589 netdev_get_tx_queue(vlan->dev, 0)->tx_dropped++;
590 rcu_read_unlock_bh();
595 static ssize_t macvtap_aio_write(struct kiocb *iocb, const struct iovec *iv,
596 unsigned long count, loff_t pos)
598 struct file *file = iocb->ki_filp;
599 ssize_t result = -ENOLINK;
600 struct macvtap_queue *q = file->private_data;
602 result = macvtap_get_user(q, iv, iov_length(iv, count),
603 file->f_flags & O_NONBLOCK);
607 /* Put packet to the user space buffer */
608 static ssize_t macvtap_put_user(struct macvtap_queue *q,
609 const struct sk_buff *skb,
610 const struct iovec *iv, int len)
612 struct macvlan_dev *vlan;
614 int vnet_hdr_len = 0;
616 if (q->flags & IFF_VNET_HDR) {
617 struct virtio_net_hdr vnet_hdr;
618 vnet_hdr_len = q->vnet_hdr_sz;
619 if ((len -= vnet_hdr_len) < 0)
622 ret = macvtap_skb_to_vnet_hdr(skb, &vnet_hdr);
626 if (memcpy_toiovecend(iv, (void *)&vnet_hdr, 0, sizeof(vnet_hdr)))
630 len = min_t(int, skb->len, len);
632 ret = skb_copy_datagram_const_iovec(skb, 0, iv, vnet_hdr_len, len);
635 vlan = rcu_dereference(q->vlan);
637 macvlan_count_rx(vlan, len, ret == 0, 0);
638 rcu_read_unlock_bh();
640 return ret ? ret : (len + vnet_hdr_len);
643 static ssize_t macvtap_do_read(struct macvtap_queue *q, struct kiocb *iocb,
644 const struct iovec *iv, unsigned long len,
647 DECLARE_WAITQUEUE(wait, current);
651 add_wait_queue(sk_sleep(&q->sk), &wait);
653 current->state = TASK_INTERRUPTIBLE;
655 /* Read frames from the queue */
656 skb = skb_dequeue(&q->sk.sk_receive_queue);
662 if (signal_pending(current)) {
666 /* Nothing to read, let's sleep */
670 ret = macvtap_put_user(q, skb, iv, len);
675 current->state = TASK_RUNNING;
676 remove_wait_queue(sk_sleep(&q->sk), &wait);
680 static ssize_t macvtap_aio_read(struct kiocb *iocb, const struct iovec *iv,
681 unsigned long count, loff_t pos)
683 struct file *file = iocb->ki_filp;
684 struct macvtap_queue *q = file->private_data;
685 ssize_t len, ret = 0;
687 len = iov_length(iv, count);
693 ret = macvtap_do_read(q, iocb, iv, len, file->f_flags & O_NONBLOCK);
694 ret = min_t(ssize_t, ret, len); /* XXX copied from tun.c. Why? */
700 * provide compatibility with generic tun/tap interface
702 static long macvtap_ioctl(struct file *file, unsigned int cmd,
705 struct macvtap_queue *q = file->private_data;
706 struct macvlan_dev *vlan;
707 void __user *argp = (void __user *)arg;
708 struct ifreq __user *ifr = argp;
709 unsigned int __user *up = argp;
711 int __user *sp = argp;
717 /* ignore the name, just look at flags */
718 if (get_user(u, &ifr->ifr_flags))
722 if ((u & ~IFF_VNET_HDR) != (IFF_NO_PI | IFF_TAP))
731 vlan = rcu_dereference(q->vlan);
734 rcu_read_unlock_bh();
740 if (copy_to_user(&ifr->ifr_name, q->vlan->dev->name, IFNAMSIZ) ||
741 put_user(q->flags, &ifr->ifr_flags))
747 if (put_user(IFF_TAP | IFF_NO_PI | IFF_VNET_HDR, up))
758 case TUNGETVNETHDRSZ:
764 case TUNSETVNETHDRSZ:
767 if (s < (int)sizeof(struct virtio_net_hdr))
774 /* let the user check for future flags */
775 if (arg & ~(TUN_F_CSUM | TUN_F_TSO4 | TUN_F_TSO6 |
776 TUN_F_TSO_ECN | TUN_F_UFO))
779 /* TODO: only accept frames with the features that
780 got enabled for forwarded frames */
781 if (!(q->flags & IFF_VNET_HDR))
791 static long macvtap_compat_ioctl(struct file *file, unsigned int cmd,
794 return macvtap_ioctl(file, cmd, (unsigned long)compat_ptr(arg));
798 static const struct file_operations macvtap_fops = {
799 .owner = THIS_MODULE,
800 .open = macvtap_open,
801 .release = macvtap_release,
802 .aio_read = macvtap_aio_read,
803 .aio_write = macvtap_aio_write,
804 .poll = macvtap_poll,
806 .unlocked_ioctl = macvtap_ioctl,
808 .compat_ioctl = macvtap_compat_ioctl,
812 static int macvtap_sendmsg(struct kiocb *iocb, struct socket *sock,
813 struct msghdr *m, size_t total_len)
815 struct macvtap_queue *q = container_of(sock, struct macvtap_queue, sock);
816 return macvtap_get_user(q, m->msg_iov, total_len,
817 m->msg_flags & MSG_DONTWAIT);
820 static int macvtap_recvmsg(struct kiocb *iocb, struct socket *sock,
821 struct msghdr *m, size_t total_len,
824 struct macvtap_queue *q = container_of(sock, struct macvtap_queue, sock);
826 if (flags & ~(MSG_DONTWAIT|MSG_TRUNC))
828 ret = macvtap_do_read(q, iocb, m->msg_iov, total_len,
829 flags & MSG_DONTWAIT);
830 if (ret > total_len) {
831 m->msg_flags |= MSG_TRUNC;
832 ret = flags & MSG_TRUNC ? ret : total_len;
837 /* Ops structure to mimic raw sockets with tun */
838 static const struct proto_ops macvtap_socket_ops = {
839 .sendmsg = macvtap_sendmsg,
840 .recvmsg = macvtap_recvmsg,
843 /* Get an underlying socket object from tun file. Returns error unless file is
844 * attached to a device. The returned object works like a packet socket, it
845 * can be used for sock_sendmsg/sock_recvmsg. The caller is responsible for
846 * holding a reference to the file for as long as the socket is in use. */
847 struct socket *macvtap_get_socket(struct file *file)
849 struct macvtap_queue *q;
850 if (file->f_op != &macvtap_fops)
851 return ERR_PTR(-EINVAL);
852 q = file->private_data;
854 return ERR_PTR(-EBADFD);
857 EXPORT_SYMBOL_GPL(macvtap_get_socket);
859 static int macvtap_init(void)
863 err = alloc_chrdev_region(&macvtap_major, 0,
864 MACVTAP_NUM_DEVS, "macvtap");
868 cdev_init(&macvtap_cdev, &macvtap_fops);
869 err = cdev_add(&macvtap_cdev, macvtap_major, MACVTAP_NUM_DEVS);
873 macvtap_class = class_create(THIS_MODULE, "macvtap");
874 if (IS_ERR(macvtap_class)) {
875 err = PTR_ERR(macvtap_class);
879 err = macvlan_link_register(&macvtap_link_ops);
886 class_unregister(macvtap_class);
888 cdev_del(&macvtap_cdev);
890 unregister_chrdev_region(macvtap_major, MACVTAP_NUM_DEVS);
894 module_init(macvtap_init);
896 static void macvtap_exit(void)
898 rtnl_link_unregister(&macvtap_link_ops);
899 class_unregister(macvtap_class);
900 cdev_del(&macvtap_cdev);
901 unregister_chrdev_region(macvtap_major, MACVTAP_NUM_DEVS);
903 module_exit(macvtap_exit);
905 MODULE_ALIAS_RTNL_LINK("macvtap");
906 MODULE_AUTHOR("Arnd Bergmann <arnd@arndb.de>");
907 MODULE_LICENSE("GPL");