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 __rcu *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_protected(q->vlan,
145 lockdep_is_held(&macvtap_lock));
147 int index = get_slot(vlan, q);
149 rcu_assign_pointer(vlan->taps[index], NULL);
150 rcu_assign_pointer(q->vlan, NULL);
155 spin_unlock(&macvtap_lock);
162 * Select a queue based on the rxq of the device on which this packet
163 * arrived. If the incoming device is not mq, calculate a flow hash
164 * to select a queue. If all fails, find the first available queue.
165 * Cache vlan->numvtaps since it can become zero during the execution
168 static struct macvtap_queue *macvtap_get_queue(struct net_device *dev,
171 struct macvlan_dev *vlan = netdev_priv(dev);
172 struct macvtap_queue *tap = NULL;
173 int numvtaps = vlan->numvtaps;
179 if (likely(skb_rx_queue_recorded(skb))) {
180 rxq = skb_get_rx_queue(skb);
182 while (unlikely(rxq >= numvtaps))
185 tap = rcu_dereference(vlan->taps[rxq]);
190 /* Check if we can use flow to select a queue */
191 rxq = skb_get_rxhash(skb);
193 tap = rcu_dereference(vlan->taps[rxq % numvtaps]);
198 /* Everything failed - find first available queue */
199 for (rxq = 0; rxq < MAX_MACVTAP_QUEUES; rxq++) {
200 tap = rcu_dereference(vlan->taps[rxq]);
210 * The net_device is going away, give up the reference
211 * that it holds on all queues and safely set the pointer
212 * from the queues to NULL.
214 static void macvtap_del_queues(struct net_device *dev)
216 struct macvlan_dev *vlan = netdev_priv(dev);
217 struct macvtap_queue *q, *qlist[MAX_MACVTAP_QUEUES];
220 /* macvtap_put_queue can free some slots, so go through all slots */
221 spin_lock(&macvtap_lock);
222 for (i = 0; i < MAX_MACVTAP_QUEUES && vlan->numvtaps; i++) {
223 q = rcu_dereference_protected(vlan->taps[i],
224 lockdep_is_held(&macvtap_lock));
227 rcu_assign_pointer(vlan->taps[i], NULL);
228 rcu_assign_pointer(q->vlan, NULL);
232 BUG_ON(vlan->numvtaps != 0);
233 spin_unlock(&macvtap_lock);
237 for (--j; j >= 0; j--)
238 sock_put(&qlist[j]->sk);
242 * Forward happens for data that gets sent from one macvlan
243 * endpoint to another one in bridge mode. We just take
244 * the skb and put it into the receive queue.
246 static int macvtap_forward(struct net_device *dev, struct sk_buff *skb)
248 struct macvtap_queue *q = macvtap_get_queue(dev, skb);
252 if (skb_queue_len(&q->sk.sk_receive_queue) >= dev->tx_queue_len)
255 skb_queue_tail(&q->sk.sk_receive_queue, skb);
256 wake_up_interruptible_poll(sk_sleep(&q->sk), POLLIN | POLLRDNORM | POLLRDBAND);
257 return NET_RX_SUCCESS;
265 * Receive is for data from the external interface (lowerdev),
266 * in case of macvtap, we can treat that the same way as
267 * forward, which macvlan cannot.
269 static int macvtap_receive(struct sk_buff *skb)
271 skb_push(skb, ETH_HLEN);
272 return macvtap_forward(skb->dev, skb);
275 static int macvtap_newlink(struct net *src_net,
276 struct net_device *dev,
278 struct nlattr *data[])
280 struct device *classdev;
284 err = macvlan_common_newlink(src_net, dev, tb, data,
285 macvtap_receive, macvtap_forward);
289 devt = MKDEV(MAJOR(macvtap_major), dev->ifindex);
291 classdev = device_create(macvtap_class, &dev->dev, devt,
292 dev, "tap%d", dev->ifindex);
293 if (IS_ERR(classdev)) {
294 err = PTR_ERR(classdev);
295 macvtap_del_queues(dev);
302 static void macvtap_dellink(struct net_device *dev,
303 struct list_head *head)
305 device_destroy(macvtap_class,
306 MKDEV(MAJOR(macvtap_major), dev->ifindex));
308 macvtap_del_queues(dev);
309 macvlan_dellink(dev, head);
312 static void macvtap_setup(struct net_device *dev)
314 macvlan_common_setup(dev);
315 dev->tx_queue_len = TUN_READQ_SIZE;
318 static struct rtnl_link_ops macvtap_link_ops __read_mostly = {
320 .setup = macvtap_setup,
321 .newlink = macvtap_newlink,
322 .dellink = macvtap_dellink,
326 static void macvtap_sock_write_space(struct sock *sk)
328 wait_queue_head_t *wqueue;
330 if (!sock_writeable(sk) ||
331 !test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags))
334 wqueue = sk_sleep(sk);
335 if (wqueue && waitqueue_active(wqueue))
336 wake_up_interruptible_poll(wqueue, POLLOUT | POLLWRNORM | POLLWRBAND);
339 static int macvtap_open(struct inode *inode, struct file *file)
341 struct net *net = current->nsproxy->net_ns;
342 struct net_device *dev = dev_get_by_index(net, iminor(inode));
343 struct macvtap_queue *q;
350 /* check if this is a macvtap device */
352 if (dev->rtnl_link_ops != &macvtap_link_ops)
356 q = (struct macvtap_queue *)sk_alloc(net, AF_UNSPEC, GFP_KERNEL,
362 init_waitqueue_head(&q->wq.wait);
363 q->sock.type = SOCK_RAW;
364 q->sock.state = SS_CONNECTED;
366 q->sock.ops = &macvtap_socket_ops;
367 sock_init_data(&q->sock, &q->sk);
368 q->sk.sk_write_space = macvtap_sock_write_space;
369 q->flags = IFF_VNET_HDR | IFF_NO_PI | IFF_TAP;
370 q->vnet_hdr_sz = sizeof(struct virtio_net_hdr);
372 err = macvtap_set_queue(dev, file, q);
383 static int macvtap_release(struct inode *inode, struct file *file)
385 struct macvtap_queue *q = file->private_data;
386 macvtap_put_queue(q);
390 static unsigned int macvtap_poll(struct file *file, poll_table * wait)
392 struct macvtap_queue *q = file->private_data;
393 unsigned int mask = POLLERR;
399 poll_wait(file, &q->wq.wait, wait);
401 if (!skb_queue_empty(&q->sk.sk_receive_queue))
402 mask |= POLLIN | POLLRDNORM;
404 if (sock_writeable(&q->sk) ||
405 (!test_and_set_bit(SOCK_ASYNC_NOSPACE, &q->sock.flags) &&
406 sock_writeable(&q->sk)))
407 mask |= POLLOUT | POLLWRNORM;
413 static inline struct sk_buff *macvtap_alloc_skb(struct sock *sk, size_t prepad,
414 size_t len, size_t linear,
415 int noblock, int *err)
419 /* Under a page? Don't bother with paged skb. */
420 if (prepad + len < PAGE_SIZE || !linear)
423 skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
428 skb_reserve(skb, prepad);
429 skb_put(skb, linear);
430 skb->data_len = len - linear;
431 skb->len += len - linear;
437 * macvtap_skb_from_vnet_hdr and macvtap_skb_to_vnet_hdr should
438 * be shared with the tun/tap driver.
440 static int macvtap_skb_from_vnet_hdr(struct sk_buff *skb,
441 struct virtio_net_hdr *vnet_hdr)
443 unsigned short gso_type = 0;
444 if (vnet_hdr->gso_type != VIRTIO_NET_HDR_GSO_NONE) {
445 switch (vnet_hdr->gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
446 case VIRTIO_NET_HDR_GSO_TCPV4:
447 gso_type = SKB_GSO_TCPV4;
449 case VIRTIO_NET_HDR_GSO_TCPV6:
450 gso_type = SKB_GSO_TCPV6;
452 case VIRTIO_NET_HDR_GSO_UDP:
453 gso_type = SKB_GSO_UDP;
459 if (vnet_hdr->gso_type & VIRTIO_NET_HDR_GSO_ECN)
460 gso_type |= SKB_GSO_TCP_ECN;
462 if (vnet_hdr->gso_size == 0)
466 if (vnet_hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
467 if (!skb_partial_csum_set(skb, vnet_hdr->csum_start,
468 vnet_hdr->csum_offset))
472 if (vnet_hdr->gso_type != VIRTIO_NET_HDR_GSO_NONE) {
473 skb_shinfo(skb)->gso_size = vnet_hdr->gso_size;
474 skb_shinfo(skb)->gso_type = gso_type;
476 /* Header must be checked, and gso_segs computed. */
477 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
478 skb_shinfo(skb)->gso_segs = 0;
483 static int macvtap_skb_to_vnet_hdr(const struct sk_buff *skb,
484 struct virtio_net_hdr *vnet_hdr)
486 memset(vnet_hdr, 0, sizeof(*vnet_hdr));
488 if (skb_is_gso(skb)) {
489 struct skb_shared_info *sinfo = skb_shinfo(skb);
491 /* This is a hint as to how much should be linear. */
492 vnet_hdr->hdr_len = skb_headlen(skb);
493 vnet_hdr->gso_size = sinfo->gso_size;
494 if (sinfo->gso_type & SKB_GSO_TCPV4)
495 vnet_hdr->gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
496 else if (sinfo->gso_type & SKB_GSO_TCPV6)
497 vnet_hdr->gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
498 else if (sinfo->gso_type & SKB_GSO_UDP)
499 vnet_hdr->gso_type = VIRTIO_NET_HDR_GSO_UDP;
502 if (sinfo->gso_type & SKB_GSO_TCP_ECN)
503 vnet_hdr->gso_type |= VIRTIO_NET_HDR_GSO_ECN;
505 vnet_hdr->gso_type = VIRTIO_NET_HDR_GSO_NONE;
507 if (skb->ip_summed == CHECKSUM_PARTIAL) {
508 vnet_hdr->flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
509 vnet_hdr->csum_start = skb_checksum_start_offset(skb);
510 vnet_hdr->csum_offset = skb->csum_offset;
511 } /* else everything is zero */
517 /* Get packet from user space buffer */
518 static ssize_t macvtap_get_user(struct macvtap_queue *q,
519 const struct iovec *iv, size_t count,
523 struct macvlan_dev *vlan;
526 struct virtio_net_hdr vnet_hdr = { 0 };
527 int vnet_hdr_len = 0;
529 if (q->flags & IFF_VNET_HDR) {
530 vnet_hdr_len = q->vnet_hdr_sz;
533 if (len < vnet_hdr_len)
537 err = memcpy_fromiovecend((void *)&vnet_hdr, iv, 0,
541 if ((vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
542 vnet_hdr.csum_start + vnet_hdr.csum_offset + 2 >
544 vnet_hdr.hdr_len = vnet_hdr.csum_start +
545 vnet_hdr.csum_offset + 2;
547 if (vnet_hdr.hdr_len > len)
552 if (unlikely(len < ETH_HLEN))
555 skb = macvtap_alloc_skb(&q->sk, NET_IP_ALIGN, len, vnet_hdr.hdr_len,
560 err = skb_copy_datagram_from_iovec(skb, 0, iv, vnet_hdr_len, len);
564 skb_set_network_header(skb, ETH_HLEN);
565 skb_reset_mac_header(skb);
566 skb->protocol = eth_hdr(skb)->h_proto;
569 err = macvtap_skb_from_vnet_hdr(skb, &vnet_hdr);
575 vlan = rcu_dereference_bh(q->vlan);
577 macvlan_start_xmit(skb, vlan->dev);
580 rcu_read_unlock_bh();
589 vlan = rcu_dereference_bh(q->vlan);
591 vlan->dev->stats.tx_dropped++;
592 rcu_read_unlock_bh();
597 static ssize_t macvtap_aio_write(struct kiocb *iocb, const struct iovec *iv,
598 unsigned long count, loff_t pos)
600 struct file *file = iocb->ki_filp;
601 ssize_t result = -ENOLINK;
602 struct macvtap_queue *q = file->private_data;
604 result = macvtap_get_user(q, iv, iov_length(iv, count),
605 file->f_flags & O_NONBLOCK);
609 /* Put packet to the user space buffer */
610 static ssize_t macvtap_put_user(struct macvtap_queue *q,
611 const struct sk_buff *skb,
612 const struct iovec *iv, int len)
614 struct macvlan_dev *vlan;
616 int vnet_hdr_len = 0;
618 if (q->flags & IFF_VNET_HDR) {
619 struct virtio_net_hdr vnet_hdr;
620 vnet_hdr_len = q->vnet_hdr_sz;
621 if ((len -= vnet_hdr_len) < 0)
624 ret = macvtap_skb_to_vnet_hdr(skb, &vnet_hdr);
628 if (memcpy_toiovecend(iv, (void *)&vnet_hdr, 0, sizeof(vnet_hdr)))
632 len = min_t(int, skb->len, len);
634 ret = skb_copy_datagram_const_iovec(skb, 0, iv, vnet_hdr_len, len);
637 vlan = rcu_dereference_bh(q->vlan);
639 macvlan_count_rx(vlan, len, ret == 0, 0);
640 rcu_read_unlock_bh();
642 return ret ? ret : (len + vnet_hdr_len);
645 static ssize_t macvtap_do_read(struct macvtap_queue *q, struct kiocb *iocb,
646 const struct iovec *iv, unsigned long len,
649 DECLARE_WAITQUEUE(wait, current);
653 add_wait_queue(sk_sleep(&q->sk), &wait);
655 current->state = TASK_INTERRUPTIBLE;
657 /* Read frames from the queue */
658 skb = skb_dequeue(&q->sk.sk_receive_queue);
664 if (signal_pending(current)) {
668 /* Nothing to read, let's sleep */
672 ret = macvtap_put_user(q, skb, iv, len);
677 current->state = TASK_RUNNING;
678 remove_wait_queue(sk_sleep(&q->sk), &wait);
682 static ssize_t macvtap_aio_read(struct kiocb *iocb, const struct iovec *iv,
683 unsigned long count, loff_t pos)
685 struct file *file = iocb->ki_filp;
686 struct macvtap_queue *q = file->private_data;
687 ssize_t len, ret = 0;
689 len = iov_length(iv, count);
695 ret = macvtap_do_read(q, iocb, iv, len, file->f_flags & O_NONBLOCK);
696 ret = min_t(ssize_t, ret, len); /* XXX copied from tun.c. Why? */
702 * provide compatibility with generic tun/tap interface
704 static long macvtap_ioctl(struct file *file, unsigned int cmd,
707 struct macvtap_queue *q = file->private_data;
708 struct macvlan_dev *vlan;
709 void __user *argp = (void __user *)arg;
710 struct ifreq __user *ifr = argp;
711 unsigned int __user *up = argp;
713 int __user *sp = argp;
719 /* ignore the name, just look at flags */
720 if (get_user(u, &ifr->ifr_flags))
724 if ((u & ~IFF_VNET_HDR) != (IFF_NO_PI | IFF_TAP))
733 vlan = rcu_dereference_bh(q->vlan);
736 rcu_read_unlock_bh();
742 if (copy_to_user(&ifr->ifr_name, vlan->dev->name, IFNAMSIZ) ||
743 put_user(q->flags, &ifr->ifr_flags))
749 if (put_user(IFF_TAP | IFF_NO_PI | IFF_VNET_HDR, up))
760 case TUNGETVNETHDRSZ:
766 case TUNSETVNETHDRSZ:
769 if (s < (int)sizeof(struct virtio_net_hdr))
776 /* let the user check for future flags */
777 if (arg & ~(TUN_F_CSUM | TUN_F_TSO4 | TUN_F_TSO6 |
778 TUN_F_TSO_ECN | TUN_F_UFO))
781 /* TODO: only accept frames with the features that
782 got enabled for forwarded frames */
783 if (!(q->flags & IFF_VNET_HDR))
793 static long macvtap_compat_ioctl(struct file *file, unsigned int cmd,
796 return macvtap_ioctl(file, cmd, (unsigned long)compat_ptr(arg));
800 static const struct file_operations macvtap_fops = {
801 .owner = THIS_MODULE,
802 .open = macvtap_open,
803 .release = macvtap_release,
804 .aio_read = macvtap_aio_read,
805 .aio_write = macvtap_aio_write,
806 .poll = macvtap_poll,
808 .unlocked_ioctl = macvtap_ioctl,
810 .compat_ioctl = macvtap_compat_ioctl,
814 static int macvtap_sendmsg(struct kiocb *iocb, struct socket *sock,
815 struct msghdr *m, size_t total_len)
817 struct macvtap_queue *q = container_of(sock, struct macvtap_queue, sock);
818 return macvtap_get_user(q, m->msg_iov, total_len,
819 m->msg_flags & MSG_DONTWAIT);
822 static int macvtap_recvmsg(struct kiocb *iocb, struct socket *sock,
823 struct msghdr *m, size_t total_len,
826 struct macvtap_queue *q = container_of(sock, struct macvtap_queue, sock);
828 if (flags & ~(MSG_DONTWAIT|MSG_TRUNC))
830 ret = macvtap_do_read(q, iocb, m->msg_iov, total_len,
831 flags & MSG_DONTWAIT);
832 if (ret > total_len) {
833 m->msg_flags |= MSG_TRUNC;
834 ret = flags & MSG_TRUNC ? ret : total_len;
839 /* Ops structure to mimic raw sockets with tun */
840 static const struct proto_ops macvtap_socket_ops = {
841 .sendmsg = macvtap_sendmsg,
842 .recvmsg = macvtap_recvmsg,
845 /* Get an underlying socket object from tun file. Returns error unless file is
846 * attached to a device. The returned object works like a packet socket, it
847 * can be used for sock_sendmsg/sock_recvmsg. The caller is responsible for
848 * holding a reference to the file for as long as the socket is in use. */
849 struct socket *macvtap_get_socket(struct file *file)
851 struct macvtap_queue *q;
852 if (file->f_op != &macvtap_fops)
853 return ERR_PTR(-EINVAL);
854 q = file->private_data;
856 return ERR_PTR(-EBADFD);
859 EXPORT_SYMBOL_GPL(macvtap_get_socket);
861 static int macvtap_init(void)
865 err = alloc_chrdev_region(&macvtap_major, 0,
866 MACVTAP_NUM_DEVS, "macvtap");
870 cdev_init(&macvtap_cdev, &macvtap_fops);
871 err = cdev_add(&macvtap_cdev, macvtap_major, MACVTAP_NUM_DEVS);
875 macvtap_class = class_create(THIS_MODULE, "macvtap");
876 if (IS_ERR(macvtap_class)) {
877 err = PTR_ERR(macvtap_class);
881 err = macvlan_link_register(&macvtap_link_ops);
888 class_unregister(macvtap_class);
890 cdev_del(&macvtap_cdev);
892 unregister_chrdev_region(macvtap_major, MACVTAP_NUM_DEVS);
896 module_init(macvtap_init);
898 static void macvtap_exit(void)
900 rtnl_link_unregister(&macvtap_link_ops);
901 class_unregister(macvtap_class);
902 cdev_del(&macvtap_cdev);
903 unregister_chrdev_region(macvtap_major, MACVTAP_NUM_DEVS);
905 module_exit(macvtap_exit);
907 MODULE_ALIAS_RTNL_LINK("macvtap");
908 MODULE_AUTHOR("Arnd Bergmann <arnd@arndb.de>");
909 MODULE_LICENSE("GPL");