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 * Variables for dealing with macvtaps device numbers.
56 static dev_t macvtap_major;
57 #define MACVTAP_NUM_DEVS (1U << MINORBITS)
58 static DEFINE_MUTEX(minor_lock);
59 static DEFINE_IDR(minor_idr);
61 #define GOODCOPY_LEN 128
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 * get_slot: return a [unused/occupied] slot in vlan->taps[]:
87 * - if 'q' is NULL, return the first empty slot;
88 * - otherwise, return the slot this pointer occupies.
90 static int get_slot(struct macvlan_dev *vlan, struct macvtap_queue *q)
94 for (i = 0; i < MAX_MACVTAP_QUEUES; i++) {
95 if (rcu_dereference(vlan->taps[i]) == q)
99 /* Should never happen */
103 static int macvtap_set_queue(struct net_device *dev, struct file *file,
104 struct macvtap_queue *q)
106 struct macvlan_dev *vlan = netdev_priv(dev);
110 spin_lock(&macvtap_lock);
111 if (vlan->numvtaps == MAX_MACVTAP_QUEUES)
115 index = get_slot(vlan, NULL);
116 rcu_assign_pointer(q->vlan, vlan);
117 rcu_assign_pointer(vlan->taps[index], q);
121 file->private_data = q;
126 spin_unlock(&macvtap_lock);
131 * The file owning the queue got closed, give up both
132 * the reference that the files holds as well as the
133 * one from the macvlan_dev if that still exists.
135 * Using the spinlock makes sure that we don't get
136 * to the queue again after destroying it.
138 static void macvtap_put_queue(struct macvtap_queue *q)
140 struct macvlan_dev *vlan;
142 spin_lock(&macvtap_lock);
143 vlan = rcu_dereference_protected(q->vlan,
144 lockdep_is_held(&macvtap_lock));
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_protected(vlan->taps[i],
223 lockdep_is_held(&macvtap_lock));
226 rcu_assign_pointer(vlan->taps[i], NULL);
227 rcu_assign_pointer(q->vlan, NULL);
231 BUG_ON(vlan->numvtaps != 0);
232 /* guarantee that any future macvtap_set_queue will fail */
233 vlan->numvtaps = MAX_MACVTAP_QUEUES;
234 spin_unlock(&macvtap_lock);
238 for (--j; j >= 0; j--)
239 sock_put(&qlist[j]->sk);
243 * Forward happens for data that gets sent from one macvlan
244 * endpoint to another one in bridge mode. We just take
245 * the skb and put it into the receive queue.
247 static int macvtap_forward(struct net_device *dev, struct sk_buff *skb)
249 struct macvtap_queue *q = macvtap_get_queue(dev, skb);
253 if (skb_queue_len(&q->sk.sk_receive_queue) >= dev->tx_queue_len)
256 skb_queue_tail(&q->sk.sk_receive_queue, skb);
257 wake_up_interruptible_poll(sk_sleep(&q->sk), POLLIN | POLLRDNORM | POLLRDBAND);
258 return NET_RX_SUCCESS;
266 * Receive is for data from the external interface (lowerdev),
267 * in case of macvtap, we can treat that the same way as
268 * forward, which macvlan cannot.
270 static int macvtap_receive(struct sk_buff *skb)
272 skb_push(skb, ETH_HLEN);
273 return macvtap_forward(skb->dev, skb);
276 static int macvtap_get_minor(struct macvlan_dev *vlan)
278 int retval = -ENOMEM;
281 mutex_lock(&minor_lock);
282 if (idr_pre_get(&minor_idr, GFP_KERNEL) == 0)
285 retval = idr_get_new_above(&minor_idr, vlan, 1, &id);
287 if (retval == -EAGAIN)
291 if (id < MACVTAP_NUM_DEVS) {
294 printk(KERN_ERR "too many macvtap devices\n");
296 idr_remove(&minor_idr, id);
299 mutex_unlock(&minor_lock);
303 static void macvtap_free_minor(struct macvlan_dev *vlan)
305 mutex_lock(&minor_lock);
307 idr_remove(&minor_idr, vlan->minor);
310 mutex_unlock(&minor_lock);
313 static struct net_device *dev_get_by_macvtap_minor(int minor)
315 struct net_device *dev = NULL;
316 struct macvlan_dev *vlan;
318 mutex_lock(&minor_lock);
319 vlan = idr_find(&minor_idr, minor);
324 mutex_unlock(&minor_lock);
328 static int macvtap_newlink(struct net *src_net,
329 struct net_device *dev,
331 struct nlattr *data[])
333 /* Don't put anything that may fail after macvlan_common_newlink
334 * because we can't undo what it does.
336 return macvlan_common_newlink(src_net, dev, tb, data,
337 macvtap_receive, macvtap_forward);
340 static void macvtap_dellink(struct net_device *dev,
341 struct list_head *head)
343 macvtap_del_queues(dev);
344 macvlan_dellink(dev, head);
347 static void macvtap_setup(struct net_device *dev)
349 macvlan_common_setup(dev);
350 dev->tx_queue_len = TUN_READQ_SIZE;
353 static struct rtnl_link_ops macvtap_link_ops __read_mostly = {
355 .setup = macvtap_setup,
356 .newlink = macvtap_newlink,
357 .dellink = macvtap_dellink,
361 static void macvtap_sock_write_space(struct sock *sk)
363 wait_queue_head_t *wqueue;
365 if (!sock_writeable(sk) ||
366 !test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags))
369 wqueue = sk_sleep(sk);
370 if (wqueue && waitqueue_active(wqueue))
371 wake_up_interruptible_poll(wqueue, POLLOUT | POLLWRNORM | POLLWRBAND);
374 static void macvtap_sock_destruct(struct sock *sk)
376 skb_queue_purge(&sk->sk_receive_queue);
379 static int macvtap_open(struct inode *inode, struct file *file)
381 struct net *net = current->nsproxy->net_ns;
382 struct net_device *dev = dev_get_by_macvtap_minor(iminor(inode));
383 struct macvtap_queue *q;
391 q = (struct macvtap_queue *)sk_alloc(net, AF_UNSPEC, GFP_KERNEL,
397 init_waitqueue_head(&q->wq.wait);
398 q->sock.type = SOCK_RAW;
399 q->sock.state = SS_CONNECTED;
401 q->sock.ops = &macvtap_socket_ops;
402 sock_init_data(&q->sock, &q->sk);
403 q->sk.sk_write_space = macvtap_sock_write_space;
404 q->sk.sk_destruct = macvtap_sock_destruct;
405 q->flags = IFF_VNET_HDR | IFF_NO_PI | IFF_TAP;
406 q->vnet_hdr_sz = sizeof(struct virtio_net_hdr);
409 * so far only KVM virtio_net uses macvtap, enable zero copy between
410 * guest kernel and host kernel when lower device supports zerocopy
412 * The macvlan supports zerocopy iff the lower device supports zero
413 * copy so we don't have to look at the lower device directly.
415 if ((dev->features & NETIF_F_HIGHDMA) && (dev->features & NETIF_F_SG))
416 sock_set_flag(&q->sk, SOCK_ZEROCOPY);
418 err = macvtap_set_queue(dev, file, q);
429 static int macvtap_release(struct inode *inode, struct file *file)
431 struct macvtap_queue *q = file->private_data;
432 macvtap_put_queue(q);
436 static unsigned int macvtap_poll(struct file *file, poll_table * wait)
438 struct macvtap_queue *q = file->private_data;
439 unsigned int mask = POLLERR;
445 poll_wait(file, &q->wq.wait, wait);
447 if (!skb_queue_empty(&q->sk.sk_receive_queue))
448 mask |= POLLIN | POLLRDNORM;
450 if (sock_writeable(&q->sk) ||
451 (!test_and_set_bit(SOCK_ASYNC_NOSPACE, &q->sock.flags) &&
452 sock_writeable(&q->sk)))
453 mask |= POLLOUT | POLLWRNORM;
459 static inline struct sk_buff *macvtap_alloc_skb(struct sock *sk, size_t prepad,
460 size_t len, size_t linear,
461 int noblock, int *err)
465 /* Under a page? Don't bother with paged skb. */
466 if (prepad + len < PAGE_SIZE || !linear)
469 skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
474 skb_reserve(skb, prepad);
475 skb_put(skb, linear);
476 skb->data_len = len - linear;
477 skb->len += len - linear;
482 /* set skb frags from iovec, this can move to core network code for reuse */
483 static int zerocopy_sg_from_iovec(struct sk_buff *skb, const struct iovec *from,
484 int offset, size_t count)
486 int len = iov_length(from, count) - offset;
487 int copy = skb_headlen(skb);
488 int size, offset1 = 0;
491 /* Skip over from offset */
492 while (count && (offset >= from->iov_len)) {
493 offset -= from->iov_len;
498 /* copy up to skb headlen */
499 while (count && (copy > 0)) {
500 size = min_t(unsigned int, copy, from->iov_len - offset);
501 if (copy_from_user(skb->data + offset1, from->iov_base + offset,
518 struct page *page[MAX_SKB_FRAGS];
521 unsigned long truesize;
523 len = from->iov_len - offset;
529 base = (unsigned long)from->iov_base + offset;
530 size = ((base & ~PAGE_MASK) + len + ~PAGE_MASK) >> PAGE_SHIFT;
531 if (i + size > MAX_SKB_FRAGS)
533 num_pages = get_user_pages_fast(base, size, 0, &page[i]);
534 if (num_pages != size) {
535 for (i = 0; i < num_pages; i++)
539 truesize = size * PAGE_SIZE;
540 skb->data_len += len;
542 skb->truesize += truesize;
543 atomic_add(truesize, &skb->sk->sk_wmem_alloc);
545 int off = base & ~PAGE_MASK;
546 int size = min_t(int, len, PAGE_SIZE - off);
547 __skb_fill_page_desc(skb, i, page[i], off, size);
548 skb_shinfo(skb)->nr_frags++;
549 /* increase sk_wmem_alloc */
561 * macvtap_skb_from_vnet_hdr and macvtap_skb_to_vnet_hdr should
562 * be shared with the tun/tap driver.
564 static int macvtap_skb_from_vnet_hdr(struct sk_buff *skb,
565 struct virtio_net_hdr *vnet_hdr)
567 unsigned short gso_type = 0;
568 if (vnet_hdr->gso_type != VIRTIO_NET_HDR_GSO_NONE) {
569 switch (vnet_hdr->gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
570 case VIRTIO_NET_HDR_GSO_TCPV4:
571 gso_type = SKB_GSO_TCPV4;
573 case VIRTIO_NET_HDR_GSO_TCPV6:
574 gso_type = SKB_GSO_TCPV6;
576 case VIRTIO_NET_HDR_GSO_UDP:
577 gso_type = SKB_GSO_UDP;
583 if (vnet_hdr->gso_type & VIRTIO_NET_HDR_GSO_ECN)
584 gso_type |= SKB_GSO_TCP_ECN;
586 if (vnet_hdr->gso_size == 0)
590 if (vnet_hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
591 if (!skb_partial_csum_set(skb, vnet_hdr->csum_start,
592 vnet_hdr->csum_offset))
596 if (vnet_hdr->gso_type != VIRTIO_NET_HDR_GSO_NONE) {
597 skb_shinfo(skb)->gso_size = vnet_hdr->gso_size;
598 skb_shinfo(skb)->gso_type = gso_type;
600 /* Header must be checked, and gso_segs computed. */
601 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
602 skb_shinfo(skb)->gso_segs = 0;
607 static int macvtap_skb_to_vnet_hdr(const struct sk_buff *skb,
608 struct virtio_net_hdr *vnet_hdr)
610 memset(vnet_hdr, 0, sizeof(*vnet_hdr));
612 if (skb_is_gso(skb)) {
613 struct skb_shared_info *sinfo = skb_shinfo(skb);
615 /* This is a hint as to how much should be linear. */
616 vnet_hdr->hdr_len = skb_headlen(skb);
617 vnet_hdr->gso_size = sinfo->gso_size;
618 if (sinfo->gso_type & SKB_GSO_TCPV4)
619 vnet_hdr->gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
620 else if (sinfo->gso_type & SKB_GSO_TCPV6)
621 vnet_hdr->gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
622 else if (sinfo->gso_type & SKB_GSO_UDP)
623 vnet_hdr->gso_type = VIRTIO_NET_HDR_GSO_UDP;
626 if (sinfo->gso_type & SKB_GSO_TCP_ECN)
627 vnet_hdr->gso_type |= VIRTIO_NET_HDR_GSO_ECN;
629 vnet_hdr->gso_type = VIRTIO_NET_HDR_GSO_NONE;
631 if (skb->ip_summed == CHECKSUM_PARTIAL) {
632 vnet_hdr->flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
633 vnet_hdr->csum_start = skb_checksum_start_offset(skb);
634 vnet_hdr->csum_offset = skb->csum_offset;
635 } else if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
636 vnet_hdr->flags = VIRTIO_NET_HDR_F_DATA_VALID;
637 } /* else everything is zero */
643 /* Get packet from user space buffer */
644 static ssize_t macvtap_get_user(struct macvtap_queue *q, struct msghdr *m,
645 const struct iovec *iv, unsigned long total_len,
646 size_t count, int noblock)
649 struct macvlan_dev *vlan;
650 unsigned long len = total_len;
652 struct virtio_net_hdr vnet_hdr = { 0 };
653 int vnet_hdr_len = 0;
655 bool zerocopy = false;
657 if (q->flags & IFF_VNET_HDR) {
658 vnet_hdr_len = q->vnet_hdr_sz;
661 if (len < vnet_hdr_len)
665 err = memcpy_fromiovecend((void *)&vnet_hdr, iv, 0,
669 if ((vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
670 vnet_hdr.csum_start + vnet_hdr.csum_offset + 2 >
672 vnet_hdr.hdr_len = vnet_hdr.csum_start +
673 vnet_hdr.csum_offset + 2;
675 if (vnet_hdr.hdr_len > len)
680 if (unlikely(len < ETH_HLEN))
684 if (unlikely(count > UIO_MAXIOV))
687 if (m && m->msg_control && sock_flag(&q->sk, SOCK_ZEROCOPY))
691 /* Userspace may produce vectors with count greater than
692 * MAX_SKB_FRAGS, so we need to linearize parts of the skb
693 * to let the rest of data to be fit in the frags.
695 if (count > MAX_SKB_FRAGS) {
696 copylen = iov_length(iv, count - MAX_SKB_FRAGS);
697 if (copylen < vnet_hdr_len)
700 copylen -= vnet_hdr_len;
702 /* There are 256 bytes to be copied in skb, so there is enough
703 * room for skb expand head in case it is used.
704 * The rest buffer is mapped from userspace.
706 if (copylen < vnet_hdr.hdr_len)
707 copylen = vnet_hdr.hdr_len;
709 copylen = GOODCOPY_LEN;
713 skb = macvtap_alloc_skb(&q->sk, NET_IP_ALIGN, copylen,
714 vnet_hdr.hdr_len, noblock, &err);
719 err = zerocopy_sg_from_iovec(skb, iv, vnet_hdr_len, count);
721 err = skb_copy_datagram_from_iovec(skb, 0, iv, vnet_hdr_len,
726 skb_set_network_header(skb, ETH_HLEN);
727 skb_reset_mac_header(skb);
728 skb->protocol = eth_hdr(skb)->h_proto;
731 err = macvtap_skb_from_vnet_hdr(skb, &vnet_hdr);
737 vlan = rcu_dereference_bh(q->vlan);
738 /* copy skb_ubuf_info for callback when skb has no error */
740 skb_shinfo(skb)->destructor_arg = m->msg_control;
741 skb_shinfo(skb)->tx_flags |= SKBTX_DEV_ZEROCOPY;
744 macvlan_start_xmit(skb, vlan->dev);
747 rcu_read_unlock_bh();
756 vlan = rcu_dereference_bh(q->vlan);
758 vlan->dev->stats.tx_dropped++;
759 rcu_read_unlock_bh();
764 static ssize_t macvtap_aio_write(struct kiocb *iocb, const struct iovec *iv,
765 unsigned long count, loff_t pos)
767 struct file *file = iocb->ki_filp;
768 ssize_t result = -ENOLINK;
769 struct macvtap_queue *q = file->private_data;
771 result = macvtap_get_user(q, NULL, iv, iov_length(iv, count), count,
772 file->f_flags & O_NONBLOCK);
776 /* Put packet to the user space buffer */
777 static ssize_t macvtap_put_user(struct macvtap_queue *q,
778 const struct sk_buff *skb,
779 const struct iovec *iv, int len)
781 struct macvlan_dev *vlan;
783 int vnet_hdr_len = 0;
785 if (q->flags & IFF_VNET_HDR) {
786 struct virtio_net_hdr vnet_hdr;
787 vnet_hdr_len = q->vnet_hdr_sz;
788 if ((len -= vnet_hdr_len) < 0)
791 ret = macvtap_skb_to_vnet_hdr(skb, &vnet_hdr);
795 if (memcpy_toiovecend(iv, (void *)&vnet_hdr, 0, sizeof(vnet_hdr)))
799 len = min_t(int, skb->len, len);
801 ret = skb_copy_datagram_const_iovec(skb, 0, iv, vnet_hdr_len, len);
804 vlan = rcu_dereference_bh(q->vlan);
806 macvlan_count_rx(vlan, len, ret == 0, 0);
807 rcu_read_unlock_bh();
809 return ret ? ret : (len + vnet_hdr_len);
812 static ssize_t macvtap_do_read(struct macvtap_queue *q, struct kiocb *iocb,
813 const struct iovec *iv, unsigned long len,
816 DECLARE_WAITQUEUE(wait, current);
820 add_wait_queue(sk_sleep(&q->sk), &wait);
822 current->state = TASK_INTERRUPTIBLE;
824 /* Read frames from the queue */
825 skb = skb_dequeue(&q->sk.sk_receive_queue);
831 if (signal_pending(current)) {
835 /* Nothing to read, let's sleep */
839 ret = macvtap_put_user(q, skb, iv, len);
844 current->state = TASK_RUNNING;
845 remove_wait_queue(sk_sleep(&q->sk), &wait);
849 static ssize_t macvtap_aio_read(struct kiocb *iocb, const struct iovec *iv,
850 unsigned long count, loff_t pos)
852 struct file *file = iocb->ki_filp;
853 struct macvtap_queue *q = file->private_data;
854 ssize_t len, ret = 0;
856 len = iov_length(iv, count);
862 ret = macvtap_do_read(q, iocb, iv, len, file->f_flags & O_NONBLOCK);
863 ret = min_t(ssize_t, ret, len); /* XXX copied from tun.c. Why? */
869 * provide compatibility with generic tun/tap interface
871 static long macvtap_ioctl(struct file *file, unsigned int cmd,
874 struct macvtap_queue *q = file->private_data;
875 struct macvlan_dev *vlan;
876 void __user *argp = (void __user *)arg;
877 struct ifreq __user *ifr = argp;
878 unsigned int __user *up = argp;
880 int __user *sp = argp;
886 /* ignore the name, just look at flags */
887 if (get_user(u, &ifr->ifr_flags))
891 if ((u & ~IFF_VNET_HDR) != (IFF_NO_PI | IFF_TAP))
900 vlan = rcu_dereference_bh(q->vlan);
903 rcu_read_unlock_bh();
909 if (copy_to_user(&ifr->ifr_name, vlan->dev->name, IFNAMSIZ) ||
910 put_user(q->flags, &ifr->ifr_flags))
916 if (put_user(IFF_TAP | IFF_NO_PI | IFF_VNET_HDR, up))
927 case TUNGETVNETHDRSZ:
933 case TUNSETVNETHDRSZ:
936 if (s < (int)sizeof(struct virtio_net_hdr))
943 /* let the user check for future flags */
944 if (arg & ~(TUN_F_CSUM | TUN_F_TSO4 | TUN_F_TSO6 |
945 TUN_F_TSO_ECN | TUN_F_UFO))
948 /* TODO: only accept frames with the features that
949 got enabled for forwarded frames */
950 if (!(q->flags & IFF_VNET_HDR))
960 static long macvtap_compat_ioctl(struct file *file, unsigned int cmd,
963 return macvtap_ioctl(file, cmd, (unsigned long)compat_ptr(arg));
967 static const struct file_operations macvtap_fops = {
968 .owner = THIS_MODULE,
969 .open = macvtap_open,
970 .release = macvtap_release,
971 .aio_read = macvtap_aio_read,
972 .aio_write = macvtap_aio_write,
973 .poll = macvtap_poll,
975 .unlocked_ioctl = macvtap_ioctl,
977 .compat_ioctl = macvtap_compat_ioctl,
981 static int macvtap_sendmsg(struct kiocb *iocb, struct socket *sock,
982 struct msghdr *m, size_t total_len)
984 struct macvtap_queue *q = container_of(sock, struct macvtap_queue, sock);
985 return macvtap_get_user(q, m, m->msg_iov, total_len, m->msg_iovlen,
986 m->msg_flags & MSG_DONTWAIT);
989 static int macvtap_recvmsg(struct kiocb *iocb, struct socket *sock,
990 struct msghdr *m, size_t total_len,
993 struct macvtap_queue *q = container_of(sock, struct macvtap_queue, sock);
995 if (flags & ~(MSG_DONTWAIT|MSG_TRUNC))
997 ret = macvtap_do_read(q, iocb, m->msg_iov, total_len,
998 flags & MSG_DONTWAIT);
999 if (ret > total_len) {
1000 m->msg_flags |= MSG_TRUNC;
1001 ret = flags & MSG_TRUNC ? ret : total_len;
1006 /* Ops structure to mimic raw sockets with tun */
1007 static const struct proto_ops macvtap_socket_ops = {
1008 .sendmsg = macvtap_sendmsg,
1009 .recvmsg = macvtap_recvmsg,
1012 /* Get an underlying socket object from tun file. Returns error unless file is
1013 * attached to a device. The returned object works like a packet socket, it
1014 * can be used for sock_sendmsg/sock_recvmsg. The caller is responsible for
1015 * holding a reference to the file for as long as the socket is in use. */
1016 struct socket *macvtap_get_socket(struct file *file)
1018 struct macvtap_queue *q;
1019 if (file->f_op != &macvtap_fops)
1020 return ERR_PTR(-EINVAL);
1021 q = file->private_data;
1023 return ERR_PTR(-EBADFD);
1026 EXPORT_SYMBOL_GPL(macvtap_get_socket);
1028 static int macvtap_device_event(struct notifier_block *unused,
1029 unsigned long event, void *ptr)
1031 struct net_device *dev = ptr;
1032 struct macvlan_dev *vlan;
1033 struct device *classdev;
1037 if (dev->rtnl_link_ops != &macvtap_link_ops)
1040 vlan = netdev_priv(dev);
1043 case NETDEV_REGISTER:
1044 /* Create the device node here after the network device has
1045 * been registered but before register_netdevice has
1048 err = macvtap_get_minor(vlan);
1050 return notifier_from_errno(err);
1052 devt = MKDEV(MAJOR(macvtap_major), vlan->minor);
1053 classdev = device_create(macvtap_class, &dev->dev, devt,
1054 dev, "tap%d", dev->ifindex);
1055 if (IS_ERR(classdev)) {
1056 macvtap_free_minor(vlan);
1057 return notifier_from_errno(PTR_ERR(classdev));
1060 case NETDEV_UNREGISTER:
1061 devt = MKDEV(MAJOR(macvtap_major), vlan->minor);
1062 device_destroy(macvtap_class, devt);
1063 macvtap_free_minor(vlan);
1070 static struct notifier_block macvtap_notifier_block __read_mostly = {
1071 .notifier_call = macvtap_device_event,
1074 static int macvtap_init(void)
1078 err = alloc_chrdev_region(&macvtap_major, 0,
1079 MACVTAP_NUM_DEVS, "macvtap");
1083 cdev_init(&macvtap_cdev, &macvtap_fops);
1084 err = cdev_add(&macvtap_cdev, macvtap_major, MACVTAP_NUM_DEVS);
1088 macvtap_class = class_create(THIS_MODULE, "macvtap");
1089 if (IS_ERR(macvtap_class)) {
1090 err = PTR_ERR(macvtap_class);
1094 err = register_netdevice_notifier(&macvtap_notifier_block);
1098 err = macvlan_link_register(&macvtap_link_ops);
1105 unregister_netdevice_notifier(&macvtap_notifier_block);
1107 class_unregister(macvtap_class);
1109 cdev_del(&macvtap_cdev);
1111 unregister_chrdev_region(macvtap_major, MACVTAP_NUM_DEVS);
1115 module_init(macvtap_init);
1117 static void macvtap_exit(void)
1119 rtnl_link_unregister(&macvtap_link_ops);
1120 unregister_netdevice_notifier(&macvtap_notifier_block);
1121 class_unregister(macvtap_class);
1122 cdev_del(&macvtap_cdev);
1123 unregister_chrdev_region(macvtap_major, MACVTAP_NUM_DEVS);
1125 module_exit(macvtap_exit);
1127 MODULE_ALIAS_RTNL_LINK("macvtap");
1128 MODULE_AUTHOR("Arnd Bergmann <arnd@arndb.de>");
1129 MODULE_LICENSE("GPL");