1 #include <linux/etherdevice.h>
2 #include <linux/if_macvlan.h>
3 #include <linux/if_vlan.h>
4 #include <linux/interrupt.h>
5 #include <linux/nsproxy.h>
6 #include <linux/compat.h>
7 #include <linux/if_tun.h>
8 #include <linux/module.h>
9 #include <linux/skbuff.h>
10 #include <linux/cache.h>
11 #include <linux/sched.h>
12 #include <linux/types.h>
13 #include <linux/slab.h>
14 #include <linux/init.h>
15 #include <linux/wait.h>
16 #include <linux/cdev.h>
17 #include <linux/idr.h>
20 #include <net/net_namespace.h>
21 #include <net/rtnetlink.h>
23 #include <linux/virtio_net.h>
26 * A macvtap queue is the central object of this driver, it connects
27 * an open character device to a macvlan interface. There can be
28 * multiple queues on one interface, which map back to queues
29 * implemented in hardware on the underlying device.
31 * macvtap_proto is used to allocate queues through the sock allocation
34 * TODO: multiqueue support is currently not implemented, even though
35 * macvtap is basically prepared for that. We will need to add this
36 * here as well as in virtio-net and qemu to get line rate on 10gbit
37 * adapters from a guest.
39 struct macvtap_queue {
44 struct macvlan_dev __rcu *vlan;
49 static struct proto macvtap_proto = {
52 .obj_size = sizeof (struct macvtap_queue),
56 * Variables for dealing with macvtaps device numbers.
58 static dev_t macvtap_major;
59 #define MACVTAP_NUM_DEVS (1U << MINORBITS)
60 static DEFINE_MUTEX(minor_lock);
61 static DEFINE_IDR(minor_idr);
63 #define GOODCOPY_LEN 128
64 static struct class *macvtap_class;
65 static struct cdev macvtap_cdev;
67 static const struct proto_ops macvtap_socket_ops;
71 * The macvtap_queue and the macvlan_dev are loosely coupled, the
72 * pointers from one to the other can only be read while rcu_read_lock
73 * or macvtap_lock is held.
75 * Both the file and the macvlan_dev hold a reference on the macvtap_queue
76 * through sock_hold(&q->sk). When the macvlan_dev goes away first,
77 * q->vlan becomes inaccessible. When the files gets closed,
78 * macvtap_get_queue() fails.
80 * There may still be references to the struct sock inside of the
81 * queue from outbound SKBs, but these never reference back to the
82 * file or the dev. The data structure is freed through __sk_free
83 * when both our references and any pending SKBs are gone.
85 static DEFINE_SPINLOCK(macvtap_lock);
88 * get_slot: return a [unused/occupied] slot in vlan->taps[]:
89 * - if 'q' is NULL, return the first empty slot;
90 * - otherwise, return the slot this pointer occupies.
92 static int get_slot(struct macvlan_dev *vlan, struct macvtap_queue *q)
96 for (i = 0; i < MAX_MACVTAP_QUEUES; i++) {
97 if (rcu_dereference_protected(vlan->taps[i],
98 lockdep_is_held(&macvtap_lock)) == q)
102 /* Should never happen */
106 static int macvtap_set_queue(struct net_device *dev, struct file *file,
107 struct macvtap_queue *q)
109 struct macvlan_dev *vlan = netdev_priv(dev);
113 spin_lock(&macvtap_lock);
114 if (vlan->numvtaps == MAX_MACVTAP_QUEUES)
118 index = get_slot(vlan, NULL);
119 rcu_assign_pointer(q->vlan, vlan);
120 rcu_assign_pointer(vlan->taps[index], q);
124 file->private_data = q;
129 spin_unlock(&macvtap_lock);
134 * The file owning the queue got closed, give up both
135 * the reference that the files holds as well as the
136 * one from the macvlan_dev if that still exists.
138 * Using the spinlock makes sure that we don't get
139 * to the queue again after destroying it.
141 static void macvtap_put_queue(struct macvtap_queue *q)
143 struct macvlan_dev *vlan;
145 spin_lock(&macvtap_lock);
146 vlan = rcu_dereference_protected(q->vlan,
147 lockdep_is_held(&macvtap_lock));
149 int index = get_slot(vlan, q);
151 RCU_INIT_POINTER(vlan->taps[index], NULL);
152 RCU_INIT_POINTER(q->vlan, NULL);
157 spin_unlock(&macvtap_lock);
164 * Select a queue based on the rxq of the device on which this packet
165 * arrived. If the incoming device is not mq, calculate a flow hash
166 * to select a queue. If all fails, find the first available queue.
167 * Cache vlan->numvtaps since it can become zero during the execution
170 static struct macvtap_queue *macvtap_get_queue(struct net_device *dev,
173 struct macvlan_dev *vlan = netdev_priv(dev);
174 struct macvtap_queue *tap = NULL;
175 int numvtaps = vlan->numvtaps;
181 /* Check if we can use flow to select a queue */
182 rxq = skb_get_rxhash(skb);
184 tap = rcu_dereference(vlan->taps[rxq % numvtaps]);
189 if (likely(skb_rx_queue_recorded(skb))) {
190 rxq = skb_get_rx_queue(skb);
192 while (unlikely(rxq >= numvtaps))
195 tap = rcu_dereference(vlan->taps[rxq]);
200 /* Everything failed - find first available queue */
201 for (rxq = 0; rxq < MAX_MACVTAP_QUEUES; rxq++) {
202 tap = rcu_dereference(vlan->taps[rxq]);
212 * The net_device is going away, give up the reference
213 * that it holds on all queues and safely set the pointer
214 * from the queues to NULL.
216 static void macvtap_del_queues(struct net_device *dev)
218 struct macvlan_dev *vlan = netdev_priv(dev);
219 struct macvtap_queue *q, *qlist[MAX_MACVTAP_QUEUES];
222 /* macvtap_put_queue can free some slots, so go through all slots */
223 spin_lock(&macvtap_lock);
224 for (i = 0; i < MAX_MACVTAP_QUEUES && vlan->numvtaps; i++) {
225 q = rcu_dereference_protected(vlan->taps[i],
226 lockdep_is_held(&macvtap_lock));
229 RCU_INIT_POINTER(vlan->taps[i], NULL);
230 RCU_INIT_POINTER(q->vlan, NULL);
234 BUG_ON(vlan->numvtaps != 0);
235 /* guarantee that any future macvtap_set_queue will fail */
236 vlan->numvtaps = MAX_MACVTAP_QUEUES;
237 spin_unlock(&macvtap_lock);
241 for (--j; j >= 0; j--)
242 sock_put(&qlist[j]->sk);
246 * Forward happens for data that gets sent from one macvlan
247 * endpoint to another one in bridge mode. We just take
248 * the skb and put it into the receive queue.
250 static int macvtap_forward(struct net_device *dev, struct sk_buff *skb)
252 struct macvtap_queue *q = macvtap_get_queue(dev, skb);
256 if (skb_queue_len(&q->sk.sk_receive_queue) >= dev->tx_queue_len)
259 skb_queue_tail(&q->sk.sk_receive_queue, skb);
260 wake_up_interruptible_poll(sk_sleep(&q->sk), POLLIN | POLLRDNORM | POLLRDBAND);
261 return NET_RX_SUCCESS;
269 * Receive is for data from the external interface (lowerdev),
270 * in case of macvtap, we can treat that the same way as
271 * forward, which macvlan cannot.
273 static int macvtap_receive(struct sk_buff *skb)
275 skb_push(skb, ETH_HLEN);
276 return macvtap_forward(skb->dev, skb);
279 static int macvtap_get_minor(struct macvlan_dev *vlan)
281 int retval = -ENOMEM;
284 mutex_lock(&minor_lock);
285 if (idr_pre_get(&minor_idr, GFP_KERNEL) == 0)
288 retval = idr_get_new_above(&minor_idr, vlan, 1, &id);
290 if (retval == -EAGAIN)
294 if (id < MACVTAP_NUM_DEVS) {
297 printk(KERN_ERR "too many macvtap devices\n");
299 idr_remove(&minor_idr, id);
302 mutex_unlock(&minor_lock);
306 static void macvtap_free_minor(struct macvlan_dev *vlan)
308 mutex_lock(&minor_lock);
310 idr_remove(&minor_idr, vlan->minor);
313 mutex_unlock(&minor_lock);
316 static struct net_device *dev_get_by_macvtap_minor(int minor)
318 struct net_device *dev = NULL;
319 struct macvlan_dev *vlan;
321 mutex_lock(&minor_lock);
322 vlan = idr_find(&minor_idr, minor);
327 mutex_unlock(&minor_lock);
331 static int macvtap_newlink(struct net *src_net,
332 struct net_device *dev,
334 struct nlattr *data[])
336 /* Don't put anything that may fail after macvlan_common_newlink
337 * because we can't undo what it does.
339 return macvlan_common_newlink(src_net, dev, tb, data,
340 macvtap_receive, macvtap_forward);
343 static void macvtap_dellink(struct net_device *dev,
344 struct list_head *head)
346 macvtap_del_queues(dev);
347 macvlan_dellink(dev, head);
350 static void macvtap_setup(struct net_device *dev)
352 macvlan_common_setup(dev);
353 dev->tx_queue_len = TUN_READQ_SIZE;
356 static struct rtnl_link_ops macvtap_link_ops __read_mostly = {
358 .setup = macvtap_setup,
359 .newlink = macvtap_newlink,
360 .dellink = macvtap_dellink,
364 static void macvtap_sock_write_space(struct sock *sk)
366 wait_queue_head_t *wqueue;
368 if (!sock_writeable(sk) ||
369 !test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags))
372 wqueue = sk_sleep(sk);
373 if (wqueue && waitqueue_active(wqueue))
374 wake_up_interruptible_poll(wqueue, POLLOUT | POLLWRNORM | POLLWRBAND);
377 static void macvtap_sock_destruct(struct sock *sk)
379 skb_queue_purge(&sk->sk_receive_queue);
382 static int macvtap_open(struct inode *inode, struct file *file)
384 struct net *net = current->nsproxy->net_ns;
385 struct net_device *dev = dev_get_by_macvtap_minor(iminor(inode));
386 struct macvtap_queue *q;
394 q = (struct macvtap_queue *)sk_alloc(net, AF_UNSPEC, GFP_KERNEL,
400 init_waitqueue_head(&q->wq.wait);
401 q->sock.type = SOCK_RAW;
402 q->sock.state = SS_CONNECTED;
404 q->sock.ops = &macvtap_socket_ops;
405 sock_init_data(&q->sock, &q->sk);
406 q->sk.sk_write_space = macvtap_sock_write_space;
407 q->sk.sk_destruct = macvtap_sock_destruct;
408 q->flags = IFF_VNET_HDR | IFF_NO_PI | IFF_TAP;
409 q->vnet_hdr_sz = sizeof(struct virtio_net_hdr);
412 * so far only KVM virtio_net uses macvtap, enable zero copy between
413 * guest kernel and host kernel when lower device supports zerocopy
415 * The macvlan supports zerocopy iff the lower device supports zero
416 * copy so we don't have to look at the lower device directly.
418 if ((dev->features & NETIF_F_HIGHDMA) && (dev->features & NETIF_F_SG))
419 sock_set_flag(&q->sk, SOCK_ZEROCOPY);
421 err = macvtap_set_queue(dev, file, q);
432 static int macvtap_release(struct inode *inode, struct file *file)
434 struct macvtap_queue *q = file->private_data;
435 macvtap_put_queue(q);
439 static unsigned int macvtap_poll(struct file *file, poll_table * wait)
441 struct macvtap_queue *q = file->private_data;
442 unsigned int mask = POLLERR;
448 poll_wait(file, &q->wq.wait, wait);
450 if (!skb_queue_empty(&q->sk.sk_receive_queue))
451 mask |= POLLIN | POLLRDNORM;
453 if (sock_writeable(&q->sk) ||
454 (!test_and_set_bit(SOCK_ASYNC_NOSPACE, &q->sock.flags) &&
455 sock_writeable(&q->sk)))
456 mask |= POLLOUT | POLLWRNORM;
462 static inline struct sk_buff *macvtap_alloc_skb(struct sock *sk, size_t prepad,
463 size_t len, size_t linear,
464 int noblock, int *err)
468 /* Under a page? Don't bother with paged skb. */
469 if (prepad + len < PAGE_SIZE || !linear)
472 skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
477 skb_reserve(skb, prepad);
478 skb_put(skb, linear);
479 skb->data_len = len - linear;
480 skb->len += len - linear;
485 /* set skb frags from iovec, this can move to core network code for reuse */
486 static int zerocopy_sg_from_iovec(struct sk_buff *skb, const struct iovec *from,
487 int offset, size_t count)
489 int len = iov_length(from, count) - offset;
490 int copy = skb_headlen(skb);
491 int size, offset1 = 0;
494 /* Skip over from offset */
495 while (count && (offset >= from->iov_len)) {
496 offset -= from->iov_len;
501 /* copy up to skb headlen */
502 while (count && (copy > 0)) {
503 size = min_t(unsigned int, copy, from->iov_len - offset);
504 if (copy_from_user(skb->data + offset1, from->iov_base + offset,
521 struct page *page[MAX_SKB_FRAGS];
524 unsigned long truesize;
526 len = from->iov_len - offset;
532 base = (unsigned long)from->iov_base + offset;
533 size = ((base & ~PAGE_MASK) + len + ~PAGE_MASK) >> PAGE_SHIFT;
534 if (i + size > MAX_SKB_FRAGS)
536 num_pages = get_user_pages_fast(base, size, 0, &page[i]);
537 if (num_pages != size) {
538 for (i = 0; i < num_pages; i++)
542 truesize = size * PAGE_SIZE;
543 skb->data_len += len;
545 skb->truesize += truesize;
546 atomic_add(truesize, &skb->sk->sk_wmem_alloc);
548 int off = base & ~PAGE_MASK;
549 int size = min_t(int, len, PAGE_SIZE - off);
550 __skb_fill_page_desc(skb, i, page[i], off, size);
551 skb_shinfo(skb)->nr_frags++;
552 /* increase sk_wmem_alloc */
564 * macvtap_skb_from_vnet_hdr and macvtap_skb_to_vnet_hdr should
565 * be shared with the tun/tap driver.
567 static int macvtap_skb_from_vnet_hdr(struct sk_buff *skb,
568 struct virtio_net_hdr *vnet_hdr)
570 unsigned short gso_type = 0;
571 if (vnet_hdr->gso_type != VIRTIO_NET_HDR_GSO_NONE) {
572 switch (vnet_hdr->gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
573 case VIRTIO_NET_HDR_GSO_TCPV4:
574 gso_type = SKB_GSO_TCPV4;
576 case VIRTIO_NET_HDR_GSO_TCPV6:
577 gso_type = SKB_GSO_TCPV6;
579 case VIRTIO_NET_HDR_GSO_UDP:
580 gso_type = SKB_GSO_UDP;
586 if (vnet_hdr->gso_type & VIRTIO_NET_HDR_GSO_ECN)
587 gso_type |= SKB_GSO_TCP_ECN;
589 if (vnet_hdr->gso_size == 0)
593 if (vnet_hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
594 if (!skb_partial_csum_set(skb, vnet_hdr->csum_start,
595 vnet_hdr->csum_offset))
599 if (vnet_hdr->gso_type != VIRTIO_NET_HDR_GSO_NONE) {
600 skb_shinfo(skb)->gso_size = vnet_hdr->gso_size;
601 skb_shinfo(skb)->gso_type = gso_type;
603 /* Header must be checked, and gso_segs computed. */
604 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
605 skb_shinfo(skb)->gso_segs = 0;
610 static int macvtap_skb_to_vnet_hdr(const struct sk_buff *skb,
611 struct virtio_net_hdr *vnet_hdr)
613 memset(vnet_hdr, 0, sizeof(*vnet_hdr));
615 if (skb_is_gso(skb)) {
616 struct skb_shared_info *sinfo = skb_shinfo(skb);
618 /* This is a hint as to how much should be linear. */
619 vnet_hdr->hdr_len = skb_headlen(skb);
620 vnet_hdr->gso_size = sinfo->gso_size;
621 if (sinfo->gso_type & SKB_GSO_TCPV4)
622 vnet_hdr->gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
623 else if (sinfo->gso_type & SKB_GSO_TCPV6)
624 vnet_hdr->gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
625 else if (sinfo->gso_type & SKB_GSO_UDP)
626 vnet_hdr->gso_type = VIRTIO_NET_HDR_GSO_UDP;
629 if (sinfo->gso_type & SKB_GSO_TCP_ECN)
630 vnet_hdr->gso_type |= VIRTIO_NET_HDR_GSO_ECN;
632 vnet_hdr->gso_type = VIRTIO_NET_HDR_GSO_NONE;
634 if (skb->ip_summed == CHECKSUM_PARTIAL) {
635 vnet_hdr->flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
636 vnet_hdr->csum_start = skb_checksum_start_offset(skb);
637 vnet_hdr->csum_offset = skb->csum_offset;
638 } else if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
639 vnet_hdr->flags = VIRTIO_NET_HDR_F_DATA_VALID;
640 } /* else everything is zero */
646 /* Get packet from user space buffer */
647 static ssize_t macvtap_get_user(struct macvtap_queue *q, struct msghdr *m,
648 const struct iovec *iv, unsigned long total_len,
649 size_t count, int noblock)
652 struct macvlan_dev *vlan;
653 unsigned long len = total_len;
655 struct virtio_net_hdr vnet_hdr = { 0 };
656 int vnet_hdr_len = 0;
658 bool zerocopy = false;
660 if (q->flags & IFF_VNET_HDR) {
661 vnet_hdr_len = q->vnet_hdr_sz;
664 if (len < vnet_hdr_len)
668 err = memcpy_fromiovecend((void *)&vnet_hdr, iv, 0,
672 if ((vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
673 vnet_hdr.csum_start + vnet_hdr.csum_offset + 2 >
675 vnet_hdr.hdr_len = vnet_hdr.csum_start +
676 vnet_hdr.csum_offset + 2;
678 if (vnet_hdr.hdr_len > len)
683 if (unlikely(len < ETH_HLEN))
687 if (unlikely(count > UIO_MAXIOV))
690 if (m && m->msg_control && sock_flag(&q->sk, SOCK_ZEROCOPY))
694 /* Userspace may produce vectors with count greater than
695 * MAX_SKB_FRAGS, so we need to linearize parts of the skb
696 * to let the rest of data to be fit in the frags.
698 if (count > MAX_SKB_FRAGS) {
699 copylen = iov_length(iv, count - MAX_SKB_FRAGS);
700 if (copylen < vnet_hdr_len)
703 copylen -= vnet_hdr_len;
705 /* There are 256 bytes to be copied in skb, so there is enough
706 * room for skb expand head in case it is used.
707 * The rest buffer is mapped from userspace.
709 if (copylen < vnet_hdr.hdr_len)
710 copylen = vnet_hdr.hdr_len;
712 copylen = GOODCOPY_LEN;
716 skb = macvtap_alloc_skb(&q->sk, NET_IP_ALIGN, copylen,
717 vnet_hdr.hdr_len, noblock, &err);
722 err = zerocopy_sg_from_iovec(skb, iv, vnet_hdr_len, count);
724 err = skb_copy_datagram_from_iovec(skb, 0, iv, vnet_hdr_len,
729 skb_set_network_header(skb, ETH_HLEN);
730 skb_reset_mac_header(skb);
731 skb->protocol = eth_hdr(skb)->h_proto;
734 err = macvtap_skb_from_vnet_hdr(skb, &vnet_hdr);
740 vlan = rcu_dereference_bh(q->vlan);
741 /* copy skb_ubuf_info for callback when skb has no error */
743 skb_shinfo(skb)->destructor_arg = m->msg_control;
744 skb_shinfo(skb)->tx_flags |= SKBTX_DEV_ZEROCOPY;
745 skb_shinfo(skb)->tx_flags |= SKBTX_SHARED_FRAG;
748 macvlan_start_xmit(skb, vlan->dev);
751 rcu_read_unlock_bh();
760 vlan = rcu_dereference_bh(q->vlan);
762 vlan->dev->stats.tx_dropped++;
763 rcu_read_unlock_bh();
768 static ssize_t macvtap_aio_write(struct kiocb *iocb, const struct iovec *iv,
769 unsigned long count, loff_t pos)
771 struct file *file = iocb->ki_filp;
772 ssize_t result = -ENOLINK;
773 struct macvtap_queue *q = file->private_data;
775 result = macvtap_get_user(q, NULL, iv, iov_length(iv, count), count,
776 file->f_flags & O_NONBLOCK);
780 /* Put packet to the user space buffer */
781 static ssize_t macvtap_put_user(struct macvtap_queue *q,
782 const struct sk_buff *skb,
783 const struct iovec *iv, int len)
785 struct macvlan_dev *vlan;
787 int vnet_hdr_len = 0;
791 if (q->flags & IFF_VNET_HDR) {
792 struct virtio_net_hdr vnet_hdr;
793 vnet_hdr_len = q->vnet_hdr_sz;
794 if ((len -= vnet_hdr_len) < 0)
797 ret = macvtap_skb_to_vnet_hdr(skb, &vnet_hdr);
801 if (memcpy_toiovecend(iv, (void *)&vnet_hdr, 0, sizeof(vnet_hdr)))
804 copied = vnet_hdr_len;
806 if (!vlan_tx_tag_present(skb))
807 len = min_t(int, skb->len, len);
814 veth.h_vlan_proto = htons(ETH_P_8021Q);
815 veth.h_vlan_TCI = htons(vlan_tx_tag_get(skb));
817 vlan_offset = offsetof(struct vlan_ethhdr, h_vlan_proto);
818 len = min_t(int, skb->len + VLAN_HLEN, len);
820 copy = min_t(int, vlan_offset, len);
821 ret = skb_copy_datagram_const_iovec(skb, 0, iv, copied, copy);
827 copy = min_t(int, sizeof(veth), len);
828 ret = memcpy_toiovecend(iv, (void *)&veth, copied, copy);
835 ret = skb_copy_datagram_const_iovec(skb, vlan_offset, iv, copied, len);
840 vlan = rcu_dereference_bh(q->vlan);
842 macvlan_count_rx(vlan, copied - vnet_hdr_len, ret == 0, 0);
843 rcu_read_unlock_bh();
845 return ret ? ret : copied;
848 static ssize_t macvtap_do_read(struct macvtap_queue *q, struct kiocb *iocb,
849 const struct iovec *iv, unsigned long len,
857 prepare_to_wait(sk_sleep(&q->sk), &wait, TASK_INTERRUPTIBLE);
859 /* Read frames from the queue */
860 skb = skb_dequeue(&q->sk.sk_receive_queue);
866 if (signal_pending(current)) {
870 /* Nothing to read, let's sleep */
874 ret = macvtap_put_user(q, skb, iv, len);
879 finish_wait(sk_sleep(&q->sk), &wait);
883 static ssize_t macvtap_aio_read(struct kiocb *iocb, const struct iovec *iv,
884 unsigned long count, loff_t pos)
886 struct file *file = iocb->ki_filp;
887 struct macvtap_queue *q = file->private_data;
888 ssize_t len, ret = 0;
890 len = iov_length(iv, count);
896 ret = macvtap_do_read(q, iocb, iv, len, file->f_flags & O_NONBLOCK);
897 ret = min_t(ssize_t, ret, len); /* XXX copied from tun.c. Why? */
903 * provide compatibility with generic tun/tap interface
905 static long macvtap_ioctl(struct file *file, unsigned int cmd,
908 struct macvtap_queue *q = file->private_data;
909 struct macvlan_dev *vlan;
910 void __user *argp = (void __user *)arg;
911 struct ifreq __user *ifr = argp;
912 unsigned int __user *up = argp;
914 int __user *sp = argp;
920 /* ignore the name, just look at flags */
921 if (get_user(u, &ifr->ifr_flags))
925 if ((u & ~IFF_VNET_HDR) != (IFF_NO_PI | IFF_TAP))
934 vlan = rcu_dereference_bh(q->vlan);
937 rcu_read_unlock_bh();
943 if (copy_to_user(&ifr->ifr_name, vlan->dev->name, IFNAMSIZ) ||
944 put_user(q->flags, &ifr->ifr_flags))
950 if (put_user(IFF_TAP | IFF_NO_PI | IFF_VNET_HDR, up))
961 case TUNGETVNETHDRSZ:
967 case TUNSETVNETHDRSZ:
970 if (s < (int)sizeof(struct virtio_net_hdr))
977 /* let the user check for future flags */
978 if (arg & ~(TUN_F_CSUM | TUN_F_TSO4 | TUN_F_TSO6 |
979 TUN_F_TSO_ECN | TUN_F_UFO))
982 /* TODO: only accept frames with the features that
983 got enabled for forwarded frames */
984 if (!(q->flags & IFF_VNET_HDR))
994 static long macvtap_compat_ioctl(struct file *file, unsigned int cmd,
997 return macvtap_ioctl(file, cmd, (unsigned long)compat_ptr(arg));
1001 static const struct file_operations macvtap_fops = {
1002 .owner = THIS_MODULE,
1003 .open = macvtap_open,
1004 .release = macvtap_release,
1005 .aio_read = macvtap_aio_read,
1006 .aio_write = macvtap_aio_write,
1007 .poll = macvtap_poll,
1008 .llseek = no_llseek,
1009 .unlocked_ioctl = macvtap_ioctl,
1010 #ifdef CONFIG_COMPAT
1011 .compat_ioctl = macvtap_compat_ioctl,
1015 static int macvtap_sendmsg(struct kiocb *iocb, struct socket *sock,
1016 struct msghdr *m, size_t total_len)
1018 struct macvtap_queue *q = container_of(sock, struct macvtap_queue, sock);
1019 return macvtap_get_user(q, m, m->msg_iov, total_len, m->msg_iovlen,
1020 m->msg_flags & MSG_DONTWAIT);
1023 static int macvtap_recvmsg(struct kiocb *iocb, struct socket *sock,
1024 struct msghdr *m, size_t total_len,
1027 struct macvtap_queue *q = container_of(sock, struct macvtap_queue, sock);
1029 if (flags & ~(MSG_DONTWAIT|MSG_TRUNC))
1031 ret = macvtap_do_read(q, iocb, m->msg_iov, total_len,
1032 flags & MSG_DONTWAIT);
1033 if (ret > total_len) {
1034 m->msg_flags |= MSG_TRUNC;
1035 ret = flags & MSG_TRUNC ? ret : total_len;
1040 /* Ops structure to mimic raw sockets with tun */
1041 static const struct proto_ops macvtap_socket_ops = {
1042 .sendmsg = macvtap_sendmsg,
1043 .recvmsg = macvtap_recvmsg,
1046 /* Get an underlying socket object from tun file. Returns error unless file is
1047 * attached to a device. The returned object works like a packet socket, it
1048 * can be used for sock_sendmsg/sock_recvmsg. The caller is responsible for
1049 * holding a reference to the file for as long as the socket is in use. */
1050 struct socket *macvtap_get_socket(struct file *file)
1052 struct macvtap_queue *q;
1053 if (file->f_op != &macvtap_fops)
1054 return ERR_PTR(-EINVAL);
1055 q = file->private_data;
1057 return ERR_PTR(-EBADFD);
1060 EXPORT_SYMBOL_GPL(macvtap_get_socket);
1062 static int macvtap_device_event(struct notifier_block *unused,
1063 unsigned long event, void *ptr)
1065 struct net_device *dev = ptr;
1066 struct macvlan_dev *vlan;
1067 struct device *classdev;
1071 if (dev->rtnl_link_ops != &macvtap_link_ops)
1074 vlan = netdev_priv(dev);
1077 case NETDEV_REGISTER:
1078 /* Create the device node here after the network device has
1079 * been registered but before register_netdevice has
1082 err = macvtap_get_minor(vlan);
1084 return notifier_from_errno(err);
1086 devt = MKDEV(MAJOR(macvtap_major), vlan->minor);
1087 classdev = device_create(macvtap_class, &dev->dev, devt,
1088 dev, "tap%d", dev->ifindex);
1089 if (IS_ERR(classdev)) {
1090 macvtap_free_minor(vlan);
1091 return notifier_from_errno(PTR_ERR(classdev));
1094 case NETDEV_UNREGISTER:
1095 devt = MKDEV(MAJOR(macvtap_major), vlan->minor);
1096 device_destroy(macvtap_class, devt);
1097 macvtap_free_minor(vlan);
1104 static struct notifier_block macvtap_notifier_block __read_mostly = {
1105 .notifier_call = macvtap_device_event,
1108 static int macvtap_init(void)
1112 err = alloc_chrdev_region(&macvtap_major, 0,
1113 MACVTAP_NUM_DEVS, "macvtap");
1117 cdev_init(&macvtap_cdev, &macvtap_fops);
1118 err = cdev_add(&macvtap_cdev, macvtap_major, MACVTAP_NUM_DEVS);
1122 macvtap_class = class_create(THIS_MODULE, "macvtap");
1123 if (IS_ERR(macvtap_class)) {
1124 err = PTR_ERR(macvtap_class);
1128 err = register_netdevice_notifier(&macvtap_notifier_block);
1132 err = macvlan_link_register(&macvtap_link_ops);
1139 unregister_netdevice_notifier(&macvtap_notifier_block);
1141 class_unregister(macvtap_class);
1143 cdev_del(&macvtap_cdev);
1145 unregister_chrdev_region(macvtap_major, MACVTAP_NUM_DEVS);
1149 module_init(macvtap_init);
1151 static void macvtap_exit(void)
1153 rtnl_link_unregister(&macvtap_link_ops);
1154 unregister_netdevice_notifier(&macvtap_notifier_block);
1155 class_unregister(macvtap_class);
1156 cdev_del(&macvtap_cdev);
1157 unregister_chrdev_region(macvtap_major, MACVTAP_NUM_DEVS);
1159 module_exit(macvtap_exit);
1161 MODULE_ALIAS_RTNL_LINK("macvtap");
1162 MODULE_AUTHOR("Arnd Bergmann <arnd@arndb.de>");
1163 MODULE_LICENSE("GPL");