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>
24 #include <net/flow_keys.h>
27 * A macvtap queue is the central object of this driver, it connects
28 * an open character device to a macvlan interface. There can be
29 * multiple queues on one interface, which map back to queues
30 * implemented in hardware on the underlying device.
32 * macvtap_proto is used to allocate queues through the sock allocation
35 * TODO: multiqueue support is currently not implemented, even though
36 * macvtap is basically prepared for that. We will need to add this
37 * here as well as in virtio-net and qemu to get line rate on 10gbit
38 * adapters from a guest.
40 struct macvtap_queue {
45 struct macvlan_dev __rcu *vlan;
50 static struct proto macvtap_proto = {
53 .obj_size = sizeof (struct macvtap_queue),
57 * Variables for dealing with macvtaps device numbers.
59 static dev_t macvtap_major;
60 #define MACVTAP_NUM_DEVS (1U << MINORBITS)
61 static DEFINE_MUTEX(minor_lock);
62 static DEFINE_IDR(minor_idr);
64 #define GOODCOPY_LEN 128
65 static struct class *macvtap_class;
66 static struct cdev macvtap_cdev;
68 static const struct proto_ops macvtap_socket_ops;
72 * The macvtap_queue and the macvlan_dev are loosely coupled, the
73 * pointers from one to the other can only be read while rcu_read_lock
74 * or macvtap_lock is held.
76 * Both the file and the macvlan_dev hold a reference on the macvtap_queue
77 * through sock_hold(&q->sk). When the macvlan_dev goes away first,
78 * q->vlan becomes inaccessible. When the files gets closed,
79 * macvtap_get_queue() fails.
81 * There may still be references to the struct sock inside of the
82 * queue from outbound SKBs, but these never reference back to the
83 * file or the dev. The data structure is freed through __sk_free
84 * when both our references and any pending SKBs are gone.
86 static DEFINE_SPINLOCK(macvtap_lock);
89 * get_slot: return a [unused/occupied] slot in vlan->taps[]:
90 * - if 'q' is NULL, return the first empty slot;
91 * - otherwise, return the slot this pointer occupies.
93 static int get_slot(struct macvlan_dev *vlan, struct macvtap_queue *q)
97 for (i = 0; i < MAX_MACVTAP_QUEUES; i++) {
98 if (rcu_dereference_protected(vlan->taps[i],
99 lockdep_is_held(&macvtap_lock)) == q)
103 /* Should never happen */
107 static int macvtap_set_queue(struct net_device *dev, struct file *file,
108 struct macvtap_queue *q)
110 struct macvlan_dev *vlan = netdev_priv(dev);
114 spin_lock(&macvtap_lock);
115 if (vlan->numvtaps == MAX_MACVTAP_QUEUES)
119 index = get_slot(vlan, NULL);
120 rcu_assign_pointer(q->vlan, vlan);
121 rcu_assign_pointer(vlan->taps[index], q);
125 file->private_data = q;
130 spin_unlock(&macvtap_lock);
135 * The file owning the queue got closed, give up both
136 * the reference that the files holds as well as the
137 * one from the macvlan_dev if that still exists.
139 * Using the spinlock makes sure that we don't get
140 * to the queue again after destroying it.
142 static void macvtap_put_queue(struct macvtap_queue *q)
144 struct macvlan_dev *vlan;
146 spin_lock(&macvtap_lock);
147 vlan = rcu_dereference_protected(q->vlan,
148 lockdep_is_held(&macvtap_lock));
150 int index = get_slot(vlan, q);
152 RCU_INIT_POINTER(vlan->taps[index], NULL);
153 RCU_INIT_POINTER(q->vlan, NULL);
158 spin_unlock(&macvtap_lock);
165 * Select a queue based on the rxq of the device on which this packet
166 * arrived. If the incoming device is not mq, calculate a flow hash
167 * to select a queue. If all fails, find the first available queue.
168 * Cache vlan->numvtaps since it can become zero during the execution
171 static struct macvtap_queue *macvtap_get_queue(struct net_device *dev,
174 struct macvlan_dev *vlan = netdev_priv(dev);
175 struct macvtap_queue *tap = NULL;
176 int numvtaps = vlan->numvtaps;
182 /* Check if we can use flow to select a queue */
183 rxq = skb_get_rxhash(skb);
185 tap = rcu_dereference(vlan->taps[rxq % numvtaps]);
190 if (likely(skb_rx_queue_recorded(skb))) {
191 rxq = skb_get_rx_queue(skb);
193 while (unlikely(rxq >= numvtaps))
196 tap = rcu_dereference(vlan->taps[rxq]);
201 /* Everything failed - find first available queue */
202 for (rxq = 0; rxq < MAX_MACVTAP_QUEUES; rxq++) {
203 tap = rcu_dereference(vlan->taps[rxq]);
213 * The net_device is going away, give up the reference
214 * that it holds on all queues and safely set the pointer
215 * from the queues to NULL.
217 static void macvtap_del_queues(struct net_device *dev)
219 struct macvlan_dev *vlan = netdev_priv(dev);
220 struct macvtap_queue *q, *qlist[MAX_MACVTAP_QUEUES];
223 /* macvtap_put_queue can free some slots, so go through all slots */
224 spin_lock(&macvtap_lock);
225 for (i = 0; i < MAX_MACVTAP_QUEUES && vlan->numvtaps; i++) {
226 q = rcu_dereference_protected(vlan->taps[i],
227 lockdep_is_held(&macvtap_lock));
230 RCU_INIT_POINTER(vlan->taps[i], NULL);
231 RCU_INIT_POINTER(q->vlan, NULL);
235 BUG_ON(vlan->numvtaps != 0);
236 /* guarantee that any future macvtap_set_queue will fail */
237 vlan->numvtaps = MAX_MACVTAP_QUEUES;
238 spin_unlock(&macvtap_lock);
242 for (--j; j >= 0; j--)
243 sock_put(&qlist[j]->sk);
247 * Forward happens for data that gets sent from one macvlan
248 * endpoint to another one in bridge mode. We just take
249 * the skb and put it into the receive queue.
251 static int macvtap_forward(struct net_device *dev, struct sk_buff *skb)
253 struct macvtap_queue *q = macvtap_get_queue(dev, skb);
257 if (skb_queue_len(&q->sk.sk_receive_queue) >= dev->tx_queue_len)
260 skb_queue_tail(&q->sk.sk_receive_queue, skb);
261 wake_up_interruptible_poll(sk_sleep(&q->sk), POLLIN | POLLRDNORM | POLLRDBAND);
262 return NET_RX_SUCCESS;
270 * Receive is for data from the external interface (lowerdev),
271 * in case of macvtap, we can treat that the same way as
272 * forward, which macvlan cannot.
274 static int macvtap_receive(struct sk_buff *skb)
276 skb_push(skb, ETH_HLEN);
277 return macvtap_forward(skb->dev, skb);
280 static int macvtap_get_minor(struct macvlan_dev *vlan)
282 int retval = -ENOMEM;
284 mutex_lock(&minor_lock);
285 retval = idr_alloc(&minor_idr, vlan, 1, MACVTAP_NUM_DEVS, GFP_KERNEL);
287 vlan->minor = retval;
288 } else if (retval == -ENOSPC) {
289 printk(KERN_ERR "too many macvtap devices\n");
292 mutex_unlock(&minor_lock);
293 return retval < 0 ? retval : 0;
296 static void macvtap_free_minor(struct macvlan_dev *vlan)
298 mutex_lock(&minor_lock);
300 idr_remove(&minor_idr, vlan->minor);
303 mutex_unlock(&minor_lock);
306 static struct net_device *dev_get_by_macvtap_minor(int minor)
308 struct net_device *dev = NULL;
309 struct macvlan_dev *vlan;
311 mutex_lock(&minor_lock);
312 vlan = idr_find(&minor_idr, minor);
317 mutex_unlock(&minor_lock);
321 static int macvtap_newlink(struct net *src_net,
322 struct net_device *dev,
324 struct nlattr *data[])
326 /* Don't put anything that may fail after macvlan_common_newlink
327 * because we can't undo what it does.
329 return macvlan_common_newlink(src_net, dev, tb, data,
330 macvtap_receive, macvtap_forward);
333 static void macvtap_dellink(struct net_device *dev,
334 struct list_head *head)
336 macvtap_del_queues(dev);
337 macvlan_dellink(dev, head);
340 static void macvtap_setup(struct net_device *dev)
342 macvlan_common_setup(dev);
343 dev->tx_queue_len = TUN_READQ_SIZE;
346 static struct rtnl_link_ops macvtap_link_ops __read_mostly = {
348 .setup = macvtap_setup,
349 .newlink = macvtap_newlink,
350 .dellink = macvtap_dellink,
354 static void macvtap_sock_write_space(struct sock *sk)
356 wait_queue_head_t *wqueue;
358 if (!sock_writeable(sk) ||
359 !test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags))
362 wqueue = sk_sleep(sk);
363 if (wqueue && waitqueue_active(wqueue))
364 wake_up_interruptible_poll(wqueue, POLLOUT | POLLWRNORM | POLLWRBAND);
367 static void macvtap_sock_destruct(struct sock *sk)
369 skb_queue_purge(&sk->sk_receive_queue);
372 static int macvtap_open(struct inode *inode, struct file *file)
374 struct net *net = current->nsproxy->net_ns;
375 struct net_device *dev = dev_get_by_macvtap_minor(iminor(inode));
376 struct macvtap_queue *q;
384 q = (struct macvtap_queue *)sk_alloc(net, AF_UNSPEC, GFP_KERNEL,
390 init_waitqueue_head(&q->wq.wait);
391 q->sock.type = SOCK_RAW;
392 q->sock.state = SS_CONNECTED;
394 q->sock.ops = &macvtap_socket_ops;
395 sock_init_data(&q->sock, &q->sk);
396 q->sk.sk_write_space = macvtap_sock_write_space;
397 q->sk.sk_destruct = macvtap_sock_destruct;
398 q->flags = IFF_VNET_HDR | IFF_NO_PI | IFF_TAP;
399 q->vnet_hdr_sz = sizeof(struct virtio_net_hdr);
402 * so far only KVM virtio_net uses macvtap, enable zero copy between
403 * guest kernel and host kernel when lower device supports zerocopy
405 * The macvlan supports zerocopy iff the lower device supports zero
406 * copy so we don't have to look at the lower device directly.
408 if ((dev->features & NETIF_F_HIGHDMA) && (dev->features & NETIF_F_SG))
409 sock_set_flag(&q->sk, SOCK_ZEROCOPY);
411 err = macvtap_set_queue(dev, file, q);
422 static int macvtap_release(struct inode *inode, struct file *file)
424 struct macvtap_queue *q = file->private_data;
425 macvtap_put_queue(q);
429 static unsigned int macvtap_poll(struct file *file, poll_table * wait)
431 struct macvtap_queue *q = file->private_data;
432 unsigned int mask = POLLERR;
438 poll_wait(file, &q->wq.wait, wait);
440 if (!skb_queue_empty(&q->sk.sk_receive_queue))
441 mask |= POLLIN | POLLRDNORM;
443 if (sock_writeable(&q->sk) ||
444 (!test_and_set_bit(SOCK_ASYNC_NOSPACE, &q->sock.flags) &&
445 sock_writeable(&q->sk)))
446 mask |= POLLOUT | POLLWRNORM;
452 static inline struct sk_buff *macvtap_alloc_skb(struct sock *sk, size_t prepad,
453 size_t len, size_t linear,
454 int noblock, int *err)
458 /* Under a page? Don't bother with paged skb. */
459 if (prepad + len < PAGE_SIZE || !linear)
462 skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
467 skb_reserve(skb, prepad);
468 skb_put(skb, linear);
469 skb->data_len = len - linear;
470 skb->len += len - linear;
475 /* set skb frags from iovec, this can move to core network code for reuse */
476 static int zerocopy_sg_from_iovec(struct sk_buff *skb, const struct iovec *from,
477 int offset, size_t count)
479 int len = iov_length(from, count) - offset;
480 int copy = skb_headlen(skb);
481 int size, offset1 = 0;
484 /* Skip over from offset */
485 while (count && (offset >= from->iov_len)) {
486 offset -= from->iov_len;
491 /* copy up to skb headlen */
492 while (count && (copy > 0)) {
493 size = min_t(unsigned int, copy, from->iov_len - offset);
494 if (copy_from_user(skb->data + offset1, from->iov_base + offset,
511 struct page *page[MAX_SKB_FRAGS];
514 unsigned long truesize;
516 len = from->iov_len - offset;
522 base = (unsigned long)from->iov_base + offset;
523 size = ((base & ~PAGE_MASK) + len + ~PAGE_MASK) >> PAGE_SHIFT;
524 if (i + size > MAX_SKB_FRAGS)
526 num_pages = get_user_pages_fast(base, size, 0, &page[i]);
527 if (num_pages != size) {
528 for (i = 0; i < num_pages; i++)
532 truesize = size * PAGE_SIZE;
533 skb->data_len += len;
535 skb->truesize += truesize;
536 atomic_add(truesize, &skb->sk->sk_wmem_alloc);
538 int off = base & ~PAGE_MASK;
539 int size = min_t(int, len, PAGE_SIZE - off);
540 __skb_fill_page_desc(skb, i, page[i], off, size);
541 skb_shinfo(skb)->nr_frags++;
542 /* increase sk_wmem_alloc */
554 * macvtap_skb_from_vnet_hdr and macvtap_skb_to_vnet_hdr should
555 * be shared with the tun/tap driver.
557 static int macvtap_skb_from_vnet_hdr(struct sk_buff *skb,
558 struct virtio_net_hdr *vnet_hdr)
560 unsigned short gso_type = 0;
561 if (vnet_hdr->gso_type != VIRTIO_NET_HDR_GSO_NONE) {
562 switch (vnet_hdr->gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
563 case VIRTIO_NET_HDR_GSO_TCPV4:
564 gso_type = SKB_GSO_TCPV4;
566 case VIRTIO_NET_HDR_GSO_TCPV6:
567 gso_type = SKB_GSO_TCPV6;
569 case VIRTIO_NET_HDR_GSO_UDP:
570 gso_type = SKB_GSO_UDP;
576 if (vnet_hdr->gso_type & VIRTIO_NET_HDR_GSO_ECN)
577 gso_type |= SKB_GSO_TCP_ECN;
579 if (vnet_hdr->gso_size == 0)
583 if (vnet_hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
584 if (!skb_partial_csum_set(skb, vnet_hdr->csum_start,
585 vnet_hdr->csum_offset))
589 if (vnet_hdr->gso_type != VIRTIO_NET_HDR_GSO_NONE) {
590 skb_shinfo(skb)->gso_size = vnet_hdr->gso_size;
591 skb_shinfo(skb)->gso_type = gso_type;
593 /* Header must be checked, and gso_segs computed. */
594 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
595 skb_shinfo(skb)->gso_segs = 0;
600 static int macvtap_skb_to_vnet_hdr(const struct sk_buff *skb,
601 struct virtio_net_hdr *vnet_hdr)
603 memset(vnet_hdr, 0, sizeof(*vnet_hdr));
605 if (skb_is_gso(skb)) {
606 struct skb_shared_info *sinfo = skb_shinfo(skb);
608 /* This is a hint as to how much should be linear. */
609 vnet_hdr->hdr_len = skb_headlen(skb);
610 vnet_hdr->gso_size = sinfo->gso_size;
611 if (sinfo->gso_type & SKB_GSO_TCPV4)
612 vnet_hdr->gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
613 else if (sinfo->gso_type & SKB_GSO_TCPV6)
614 vnet_hdr->gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
615 else if (sinfo->gso_type & SKB_GSO_UDP)
616 vnet_hdr->gso_type = VIRTIO_NET_HDR_GSO_UDP;
619 if (sinfo->gso_type & SKB_GSO_TCP_ECN)
620 vnet_hdr->gso_type |= VIRTIO_NET_HDR_GSO_ECN;
622 vnet_hdr->gso_type = VIRTIO_NET_HDR_GSO_NONE;
624 if (skb->ip_summed == CHECKSUM_PARTIAL) {
625 vnet_hdr->flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
626 vnet_hdr->csum_start = skb_checksum_start_offset(skb);
627 vnet_hdr->csum_offset = skb->csum_offset;
628 } else if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
629 vnet_hdr->flags = VIRTIO_NET_HDR_F_DATA_VALID;
630 } /* else everything is zero */
636 /* Get packet from user space buffer */
637 static ssize_t macvtap_get_user(struct macvtap_queue *q, struct msghdr *m,
638 const struct iovec *iv, unsigned long total_len,
639 size_t count, int noblock)
642 struct macvlan_dev *vlan;
643 unsigned long len = total_len;
645 struct virtio_net_hdr vnet_hdr = { 0 };
646 int vnet_hdr_len = 0;
648 bool zerocopy = false;
649 struct flow_keys keys;
651 if (q->flags & IFF_VNET_HDR) {
652 vnet_hdr_len = q->vnet_hdr_sz;
655 if (len < vnet_hdr_len)
659 err = memcpy_fromiovecend((void *)&vnet_hdr, iv, 0,
663 if ((vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
664 vnet_hdr.csum_start + vnet_hdr.csum_offset + 2 >
666 vnet_hdr.hdr_len = vnet_hdr.csum_start +
667 vnet_hdr.csum_offset + 2;
669 if (vnet_hdr.hdr_len > len)
674 if (unlikely(len < ETH_HLEN))
678 if (unlikely(count > UIO_MAXIOV))
681 if (m && m->msg_control && sock_flag(&q->sk, SOCK_ZEROCOPY))
685 /* Userspace may produce vectors with count greater than
686 * MAX_SKB_FRAGS, so we need to linearize parts of the skb
687 * to let the rest of data to be fit in the frags.
689 if (count > MAX_SKB_FRAGS) {
690 copylen = iov_length(iv, count - MAX_SKB_FRAGS);
691 if (copylen < vnet_hdr_len)
694 copylen -= vnet_hdr_len;
696 /* There are 256 bytes to be copied in skb, so there is enough
697 * room for skb expand head in case it is used.
698 * The rest buffer is mapped from userspace.
700 if (copylen < vnet_hdr.hdr_len)
701 copylen = vnet_hdr.hdr_len;
703 copylen = GOODCOPY_LEN;
707 skb = macvtap_alloc_skb(&q->sk, NET_IP_ALIGN, copylen,
708 vnet_hdr.hdr_len, noblock, &err);
713 err = zerocopy_sg_from_iovec(skb, iv, vnet_hdr_len, count);
715 err = skb_copy_datagram_from_iovec(skb, 0, iv, vnet_hdr_len,
720 skb_set_network_header(skb, ETH_HLEN);
721 skb_reset_mac_header(skb);
722 skb->protocol = eth_hdr(skb)->h_proto;
725 err = macvtap_skb_from_vnet_hdr(skb, &vnet_hdr);
730 if (skb->ip_summed == CHECKSUM_PARTIAL)
731 skb_set_transport_header(skb, skb_checksum_start_offset(skb));
732 else if (skb_flow_dissect(skb, &keys))
733 skb_set_transport_header(skb, keys.thoff);
735 skb_set_transport_header(skb, ETH_HLEN);
738 vlan = rcu_dereference_bh(q->vlan);
739 /* copy skb_ubuf_info for callback when skb has no error */
741 skb_shinfo(skb)->destructor_arg = m->msg_control;
742 skb_shinfo(skb)->tx_flags |= SKBTX_DEV_ZEROCOPY;
743 skb_shinfo(skb)->tx_flags |= SKBTX_SHARED_FRAG;
746 macvlan_start_xmit(skb, vlan->dev);
749 rcu_read_unlock_bh();
758 vlan = rcu_dereference_bh(q->vlan);
760 vlan->dev->stats.tx_dropped++;
761 rcu_read_unlock_bh();
766 static ssize_t macvtap_aio_write(struct kiocb *iocb, const struct iovec *iv,
767 unsigned long count, loff_t pos)
769 struct file *file = iocb->ki_filp;
770 ssize_t result = -ENOLINK;
771 struct macvtap_queue *q = file->private_data;
773 result = macvtap_get_user(q, NULL, iv, iov_length(iv, count), count,
774 file->f_flags & O_NONBLOCK);
778 /* Put packet to the user space buffer */
779 static ssize_t macvtap_put_user(struct macvtap_queue *q,
780 const struct sk_buff *skb,
781 const struct iovec *iv, int len)
783 struct macvlan_dev *vlan;
785 int vnet_hdr_len = 0;
789 if (q->flags & IFF_VNET_HDR) {
790 struct virtio_net_hdr vnet_hdr;
791 vnet_hdr_len = q->vnet_hdr_sz;
792 if ((len -= vnet_hdr_len) < 0)
795 ret = macvtap_skb_to_vnet_hdr(skb, &vnet_hdr);
799 if (memcpy_toiovecend(iv, (void *)&vnet_hdr, 0, sizeof(vnet_hdr)))
802 copied = vnet_hdr_len;
804 if (!vlan_tx_tag_present(skb))
805 len = min_t(int, skb->len, len);
812 veth.h_vlan_proto = htons(ETH_P_8021Q);
813 veth.h_vlan_TCI = htons(vlan_tx_tag_get(skb));
815 vlan_offset = offsetof(struct vlan_ethhdr, h_vlan_proto);
816 len = min_t(int, skb->len + VLAN_HLEN, len);
818 copy = min_t(int, vlan_offset, len);
819 ret = skb_copy_datagram_const_iovec(skb, 0, iv, copied, copy);
825 copy = min_t(int, sizeof(veth), len);
826 ret = memcpy_toiovecend(iv, (void *)&veth, copied, copy);
833 ret = skb_copy_datagram_const_iovec(skb, vlan_offset, iv, copied, len);
838 vlan = rcu_dereference_bh(q->vlan);
840 macvlan_count_rx(vlan, copied - vnet_hdr_len, ret == 0, 0);
841 rcu_read_unlock_bh();
843 return ret ? ret : copied;
846 static ssize_t macvtap_do_read(struct macvtap_queue *q, struct kiocb *iocb,
847 const struct iovec *iv, unsigned long len,
855 prepare_to_wait(sk_sleep(&q->sk), &wait, TASK_INTERRUPTIBLE);
857 /* Read frames from the queue */
858 skb = skb_dequeue(&q->sk.sk_receive_queue);
864 if (signal_pending(current)) {
868 /* Nothing to read, let's sleep */
872 ret = macvtap_put_user(q, skb, iv, len);
877 finish_wait(sk_sleep(&q->sk), &wait);
881 static ssize_t macvtap_aio_read(struct kiocb *iocb, const struct iovec *iv,
882 unsigned long count, loff_t pos)
884 struct file *file = iocb->ki_filp;
885 struct macvtap_queue *q = file->private_data;
886 ssize_t len, ret = 0;
888 len = iov_length(iv, count);
894 ret = macvtap_do_read(q, iocb, iv, len, file->f_flags & O_NONBLOCK);
895 ret = min_t(ssize_t, ret, len); /* XXX copied from tun.c. Why? */
901 * provide compatibility with generic tun/tap interface
903 static long macvtap_ioctl(struct file *file, unsigned int cmd,
906 struct macvtap_queue *q = file->private_data;
907 struct macvlan_dev *vlan;
908 void __user *argp = (void __user *)arg;
909 struct ifreq __user *ifr = argp;
910 unsigned int __user *up = argp;
912 int __user *sp = argp;
918 /* ignore the name, just look at flags */
919 if (get_user(u, &ifr->ifr_flags))
923 if ((u & ~IFF_VNET_HDR) != (IFF_NO_PI | IFF_TAP))
932 vlan = rcu_dereference_bh(q->vlan);
935 rcu_read_unlock_bh();
941 if (copy_to_user(&ifr->ifr_name, vlan->dev->name, IFNAMSIZ) ||
942 put_user(q->flags, &ifr->ifr_flags))
948 if (put_user(IFF_TAP | IFF_NO_PI | IFF_VNET_HDR, up))
959 case TUNGETVNETHDRSZ:
965 case TUNSETVNETHDRSZ:
968 if (s < (int)sizeof(struct virtio_net_hdr))
975 /* let the user check for future flags */
976 if (arg & ~(TUN_F_CSUM | TUN_F_TSO4 | TUN_F_TSO6 |
977 TUN_F_TSO_ECN | TUN_F_UFO))
980 /* TODO: only accept frames with the features that
981 got enabled for forwarded frames */
982 if (!(q->flags & IFF_VNET_HDR))
992 static long macvtap_compat_ioctl(struct file *file, unsigned int cmd,
995 return macvtap_ioctl(file, cmd, (unsigned long)compat_ptr(arg));
999 static const struct file_operations macvtap_fops = {
1000 .owner = THIS_MODULE,
1001 .open = macvtap_open,
1002 .release = macvtap_release,
1003 .aio_read = macvtap_aio_read,
1004 .aio_write = macvtap_aio_write,
1005 .poll = macvtap_poll,
1006 .llseek = no_llseek,
1007 .unlocked_ioctl = macvtap_ioctl,
1008 #ifdef CONFIG_COMPAT
1009 .compat_ioctl = macvtap_compat_ioctl,
1013 static int macvtap_sendmsg(struct kiocb *iocb, struct socket *sock,
1014 struct msghdr *m, size_t total_len)
1016 struct macvtap_queue *q = container_of(sock, struct macvtap_queue, sock);
1017 return macvtap_get_user(q, m, m->msg_iov, total_len, m->msg_iovlen,
1018 m->msg_flags & MSG_DONTWAIT);
1021 static int macvtap_recvmsg(struct kiocb *iocb, struct socket *sock,
1022 struct msghdr *m, size_t total_len,
1025 struct macvtap_queue *q = container_of(sock, struct macvtap_queue, sock);
1027 if (flags & ~(MSG_DONTWAIT|MSG_TRUNC))
1029 ret = macvtap_do_read(q, iocb, m->msg_iov, total_len,
1030 flags & MSG_DONTWAIT);
1031 if (ret > total_len) {
1032 m->msg_flags |= MSG_TRUNC;
1033 ret = flags & MSG_TRUNC ? ret : total_len;
1038 /* Ops structure to mimic raw sockets with tun */
1039 static const struct proto_ops macvtap_socket_ops = {
1040 .sendmsg = macvtap_sendmsg,
1041 .recvmsg = macvtap_recvmsg,
1044 /* Get an underlying socket object from tun file. Returns error unless file is
1045 * attached to a device. The returned object works like a packet socket, it
1046 * can be used for sock_sendmsg/sock_recvmsg. The caller is responsible for
1047 * holding a reference to the file for as long as the socket is in use. */
1048 struct socket *macvtap_get_socket(struct file *file)
1050 struct macvtap_queue *q;
1051 if (file->f_op != &macvtap_fops)
1052 return ERR_PTR(-EINVAL);
1053 q = file->private_data;
1055 return ERR_PTR(-EBADFD);
1058 EXPORT_SYMBOL_GPL(macvtap_get_socket);
1060 static int macvtap_device_event(struct notifier_block *unused,
1061 unsigned long event, void *ptr)
1063 struct net_device *dev = ptr;
1064 struct macvlan_dev *vlan;
1065 struct device *classdev;
1069 if (dev->rtnl_link_ops != &macvtap_link_ops)
1072 vlan = netdev_priv(dev);
1075 case NETDEV_REGISTER:
1076 /* Create the device node here after the network device has
1077 * been registered but before register_netdevice has
1080 err = macvtap_get_minor(vlan);
1082 return notifier_from_errno(err);
1084 devt = MKDEV(MAJOR(macvtap_major), vlan->minor);
1085 classdev = device_create(macvtap_class, &dev->dev, devt,
1086 dev, "tap%d", dev->ifindex);
1087 if (IS_ERR(classdev)) {
1088 macvtap_free_minor(vlan);
1089 return notifier_from_errno(PTR_ERR(classdev));
1092 case NETDEV_UNREGISTER:
1093 devt = MKDEV(MAJOR(macvtap_major), vlan->minor);
1094 device_destroy(macvtap_class, devt);
1095 macvtap_free_minor(vlan);
1102 static struct notifier_block macvtap_notifier_block __read_mostly = {
1103 .notifier_call = macvtap_device_event,
1106 static int macvtap_init(void)
1110 err = alloc_chrdev_region(&macvtap_major, 0,
1111 MACVTAP_NUM_DEVS, "macvtap");
1115 cdev_init(&macvtap_cdev, &macvtap_fops);
1116 err = cdev_add(&macvtap_cdev, macvtap_major, MACVTAP_NUM_DEVS);
1120 macvtap_class = class_create(THIS_MODULE, "macvtap");
1121 if (IS_ERR(macvtap_class)) {
1122 err = PTR_ERR(macvtap_class);
1126 err = register_netdevice_notifier(&macvtap_notifier_block);
1130 err = macvlan_link_register(&macvtap_link_ops);
1137 unregister_netdevice_notifier(&macvtap_notifier_block);
1139 class_unregister(macvtap_class);
1141 cdev_del(&macvtap_cdev);
1143 unregister_chrdev_region(macvtap_major, MACVTAP_NUM_DEVS);
1147 module_init(macvtap_init);
1149 static void macvtap_exit(void)
1151 rtnl_link_unregister(&macvtap_link_ops);
1152 unregister_netdevice_notifier(&macvtap_notifier_block);
1153 class_unregister(macvtap_class);
1154 cdev_del(&macvtap_cdev);
1155 unregister_chrdev_region(macvtap_major, MACVTAP_NUM_DEVS);
1157 module_exit(macvtap_exit);
1159 MODULE_ALIAS_RTNL_LINK("macvtap");
1160 MODULE_AUTHOR("Arnd Bergmann <arnd@arndb.de>");
1161 MODULE_LICENSE("GPL");