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
35 struct macvtap_queue {
40 struct macvlan_dev __rcu *vlan;
45 struct list_head next;
48 static struct proto macvtap_proto = {
51 .obj_size = sizeof (struct macvtap_queue),
55 * Variables for dealing with macvtaps device numbers.
57 static dev_t macvtap_major;
58 #define MACVTAP_NUM_DEVS (1U << MINORBITS)
59 static DEFINE_MUTEX(minor_lock);
60 static DEFINE_IDR(minor_idr);
62 #define GOODCOPY_LEN 128
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);
86 static int macvtap_enable_queue(struct net_device *dev, struct file *file,
87 struct macvtap_queue *q)
89 struct macvlan_dev *vlan = netdev_priv(dev);
92 spin_lock(&macvtap_lock);
98 rcu_assign_pointer(vlan->taps[vlan->numvtaps], q);
99 q->queue_index = vlan->numvtaps;
104 spin_unlock(&macvtap_lock);
108 static int macvtap_set_queue(struct net_device *dev, struct file *file,
109 struct macvtap_queue *q)
111 struct macvlan_dev *vlan = netdev_priv(dev);
114 spin_lock(&macvtap_lock);
115 if (vlan->numqueues == MAX_MACVTAP_QUEUES)
119 rcu_assign_pointer(q->vlan, vlan);
120 rcu_assign_pointer(vlan->taps[vlan->numvtaps], q);
124 q->queue_index = vlan->numvtaps;
126 file->private_data = q;
127 list_add_tail(&q->next, &vlan->queue_list);
133 spin_unlock(&macvtap_lock);
137 static int __macvtap_disable_queue(struct macvtap_queue *q)
139 struct macvlan_dev *vlan;
140 struct macvtap_queue *nq;
142 vlan = rcu_dereference_protected(q->vlan,
143 lockdep_is_held(&macvtap_lock));
149 int index = q->queue_index;
150 BUG_ON(index >= vlan->numvtaps);
151 nq = rcu_dereference_protected(vlan->taps[vlan->numvtaps - 1],
152 lockdep_is_held(&macvtap_lock));
153 nq->queue_index = index;
155 rcu_assign_pointer(vlan->taps[index], nq);
156 RCU_INIT_POINTER(vlan->taps[vlan->numvtaps - 1], NULL);
165 static int macvtap_disable_queue(struct macvtap_queue *q)
169 spin_lock(&macvtap_lock);
170 err = __macvtap_disable_queue(q);
171 spin_unlock(&macvtap_lock);
177 * The file owning the queue got closed, give up both
178 * the reference that the files holds as well as the
179 * one from the macvlan_dev if that still exists.
181 * Using the spinlock makes sure that we don't get
182 * to the queue again after destroying it.
184 static void macvtap_put_queue(struct macvtap_queue *q)
186 struct macvlan_dev *vlan;
188 spin_lock(&macvtap_lock);
189 vlan = rcu_dereference_protected(q->vlan,
190 lockdep_is_held(&macvtap_lock));
193 BUG_ON(__macvtap_disable_queue(q));
196 RCU_INIT_POINTER(q->vlan, NULL);
198 list_del_init(&q->next);
201 spin_unlock(&macvtap_lock);
208 * Select a queue based on the rxq of the device on which this packet
209 * arrived. If the incoming device is not mq, calculate a flow hash
210 * to select a queue. If all fails, find the first available queue.
211 * Cache vlan->numvtaps since it can become zero during the execution
214 static struct macvtap_queue *macvtap_get_queue(struct net_device *dev,
217 struct macvlan_dev *vlan = netdev_priv(dev);
218 struct macvtap_queue *tap = NULL;
219 /* Access to taps array is protected by rcu, but access to numvtaps
220 * isn't. Below we use it to lookup a queue, but treat it as a hint
221 * and validate that the result isn't NULL - in case we are
222 * racing against queue removal.
224 int numvtaps = ACCESS_ONCE(vlan->numvtaps);
230 /* Check if we can use flow to select a queue */
231 rxq = skb_get_rxhash(skb);
233 tap = rcu_dereference(vlan->taps[rxq % numvtaps]);
237 if (likely(skb_rx_queue_recorded(skb))) {
238 rxq = skb_get_rx_queue(skb);
240 while (unlikely(rxq >= numvtaps))
243 tap = rcu_dereference(vlan->taps[rxq]);
247 tap = rcu_dereference(vlan->taps[0]);
253 * The net_device is going away, give up the reference
254 * that it holds on all queues and safely set the pointer
255 * from the queues to NULL.
257 static void macvtap_del_queues(struct net_device *dev)
259 struct macvlan_dev *vlan = netdev_priv(dev);
260 struct macvtap_queue *q, *tmp, *qlist[MAX_MACVTAP_QUEUES];
263 spin_lock(&macvtap_lock);
264 list_for_each_entry_safe(q, tmp, &vlan->queue_list, next) {
265 list_del_init(&q->next);
267 RCU_INIT_POINTER(q->vlan, NULL);
272 for (i = 0; i < vlan->numvtaps; i++)
273 RCU_INIT_POINTER(vlan->taps[i], NULL);
274 BUG_ON(vlan->numvtaps);
275 BUG_ON(vlan->numqueues);
276 /* guarantee that any future macvtap_set_queue will fail */
277 vlan->numvtaps = MAX_MACVTAP_QUEUES;
278 spin_unlock(&macvtap_lock);
282 for (--j; j >= 0; j--)
283 sock_put(&qlist[j]->sk);
287 * Forward happens for data that gets sent from one macvlan
288 * endpoint to another one in bridge mode. We just take
289 * the skb and put it into the receive queue.
291 static int macvtap_forward(struct net_device *dev, struct sk_buff *skb)
293 struct macvtap_queue *q = macvtap_get_queue(dev, skb);
297 if (skb_queue_len(&q->sk.sk_receive_queue) >= dev->tx_queue_len)
300 skb_queue_tail(&q->sk.sk_receive_queue, skb);
301 wake_up_interruptible_poll(sk_sleep(&q->sk), POLLIN | POLLRDNORM | POLLRDBAND);
302 return NET_RX_SUCCESS;
310 * Receive is for data from the external interface (lowerdev),
311 * in case of macvtap, we can treat that the same way as
312 * forward, which macvlan cannot.
314 static int macvtap_receive(struct sk_buff *skb)
316 skb_push(skb, ETH_HLEN);
317 return macvtap_forward(skb->dev, skb);
320 static int macvtap_get_minor(struct macvlan_dev *vlan)
322 int retval = -ENOMEM;
324 mutex_lock(&minor_lock);
325 retval = idr_alloc(&minor_idr, vlan, 1, MACVTAP_NUM_DEVS, GFP_KERNEL);
327 vlan->minor = retval;
328 } else if (retval == -ENOSPC) {
329 printk(KERN_ERR "too many macvtap devices\n");
332 mutex_unlock(&minor_lock);
333 return retval < 0 ? retval : 0;
336 static void macvtap_free_minor(struct macvlan_dev *vlan)
338 mutex_lock(&minor_lock);
340 idr_remove(&minor_idr, vlan->minor);
343 mutex_unlock(&minor_lock);
346 static struct net_device *dev_get_by_macvtap_minor(int minor)
348 struct net_device *dev = NULL;
349 struct macvlan_dev *vlan;
351 mutex_lock(&minor_lock);
352 vlan = idr_find(&minor_idr, minor);
357 mutex_unlock(&minor_lock);
361 static int macvtap_newlink(struct net *src_net,
362 struct net_device *dev,
364 struct nlattr *data[])
366 struct macvlan_dev *vlan = netdev_priv(dev);
367 INIT_LIST_HEAD(&vlan->queue_list);
369 /* Don't put anything that may fail after macvlan_common_newlink
370 * because we can't undo what it does.
372 return macvlan_common_newlink(src_net, dev, tb, data,
373 macvtap_receive, macvtap_forward);
376 static void macvtap_dellink(struct net_device *dev,
377 struct list_head *head)
379 macvtap_del_queues(dev);
380 macvlan_dellink(dev, head);
383 static void macvtap_setup(struct net_device *dev)
385 macvlan_common_setup(dev);
386 dev->tx_queue_len = TUN_READQ_SIZE;
389 static struct rtnl_link_ops macvtap_link_ops __read_mostly = {
391 .setup = macvtap_setup,
392 .newlink = macvtap_newlink,
393 .dellink = macvtap_dellink,
397 static void macvtap_sock_write_space(struct sock *sk)
399 wait_queue_head_t *wqueue;
401 if (!sock_writeable(sk) ||
402 !test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags))
405 wqueue = sk_sleep(sk);
406 if (wqueue && waitqueue_active(wqueue))
407 wake_up_interruptible_poll(wqueue, POLLOUT | POLLWRNORM | POLLWRBAND);
410 static void macvtap_sock_destruct(struct sock *sk)
412 skb_queue_purge(&sk->sk_receive_queue);
415 static int macvtap_open(struct inode *inode, struct file *file)
417 struct net *net = current->nsproxy->net_ns;
418 struct net_device *dev = dev_get_by_macvtap_minor(iminor(inode));
419 struct macvtap_queue *q;
427 q = (struct macvtap_queue *)sk_alloc(net, AF_UNSPEC, GFP_KERNEL,
433 init_waitqueue_head(&q->wq.wait);
434 q->sock.type = SOCK_RAW;
435 q->sock.state = SS_CONNECTED;
437 q->sock.ops = &macvtap_socket_ops;
438 sock_init_data(&q->sock, &q->sk);
439 q->sk.sk_write_space = macvtap_sock_write_space;
440 q->sk.sk_destruct = macvtap_sock_destruct;
441 q->flags = IFF_VNET_HDR | IFF_NO_PI | IFF_TAP;
442 q->vnet_hdr_sz = sizeof(struct virtio_net_hdr);
445 * so far only KVM virtio_net uses macvtap, enable zero copy between
446 * guest kernel and host kernel when lower device supports zerocopy
448 * The macvlan supports zerocopy iff the lower device supports zero
449 * copy so we don't have to look at the lower device directly.
451 if ((dev->features & NETIF_F_HIGHDMA) && (dev->features & NETIF_F_SG))
452 sock_set_flag(&q->sk, SOCK_ZEROCOPY);
454 err = macvtap_set_queue(dev, file, q);
465 static int macvtap_release(struct inode *inode, struct file *file)
467 struct macvtap_queue *q = file->private_data;
468 macvtap_put_queue(q);
472 static unsigned int macvtap_poll(struct file *file, poll_table * wait)
474 struct macvtap_queue *q = file->private_data;
475 unsigned int mask = POLLERR;
481 poll_wait(file, &q->wq.wait, wait);
483 if (!skb_queue_empty(&q->sk.sk_receive_queue))
484 mask |= POLLIN | POLLRDNORM;
486 if (sock_writeable(&q->sk) ||
487 (!test_and_set_bit(SOCK_ASYNC_NOSPACE, &q->sock.flags) &&
488 sock_writeable(&q->sk)))
489 mask |= POLLOUT | POLLWRNORM;
495 static inline struct sk_buff *macvtap_alloc_skb(struct sock *sk, size_t prepad,
496 size_t len, size_t linear,
497 int noblock, int *err)
501 /* Under a page? Don't bother with paged skb. */
502 if (prepad + len < PAGE_SIZE || !linear)
505 skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
510 skb_reserve(skb, prepad);
511 skb_put(skb, linear);
512 skb->data_len = len - linear;
513 skb->len += len - linear;
518 /* set skb frags from iovec, this can move to core network code for reuse */
519 static int zerocopy_sg_from_iovec(struct sk_buff *skb, const struct iovec *from,
520 int offset, size_t count)
522 int len = iov_length(from, count) - offset;
523 int copy = skb_headlen(skb);
524 int size, offset1 = 0;
527 /* Skip over from offset */
528 while (count && (offset >= from->iov_len)) {
529 offset -= from->iov_len;
534 /* copy up to skb headlen */
535 while (count && (copy > 0)) {
536 size = min_t(unsigned int, copy, from->iov_len - offset);
537 if (copy_from_user(skb->data + offset1, from->iov_base + offset,
554 struct page *page[MAX_SKB_FRAGS];
557 unsigned long truesize;
559 len = from->iov_len - offset;
565 base = (unsigned long)from->iov_base + offset;
566 size = ((base & ~PAGE_MASK) + len + ~PAGE_MASK) >> PAGE_SHIFT;
567 if (i + size > MAX_SKB_FRAGS)
569 num_pages = get_user_pages_fast(base, size, 0, &page[i]);
570 if (num_pages != size) {
571 for (i = 0; i < num_pages; i++)
575 truesize = size * PAGE_SIZE;
576 skb->data_len += len;
578 skb->truesize += truesize;
579 atomic_add(truesize, &skb->sk->sk_wmem_alloc);
581 int off = base & ~PAGE_MASK;
582 int size = min_t(int, len, PAGE_SIZE - off);
583 __skb_fill_page_desc(skb, i, page[i], off, size);
584 skb_shinfo(skb)->nr_frags++;
585 /* increase sk_wmem_alloc */
597 * macvtap_skb_from_vnet_hdr and macvtap_skb_to_vnet_hdr should
598 * be shared with the tun/tap driver.
600 static int macvtap_skb_from_vnet_hdr(struct sk_buff *skb,
601 struct virtio_net_hdr *vnet_hdr)
603 unsigned short gso_type = 0;
604 if (vnet_hdr->gso_type != VIRTIO_NET_HDR_GSO_NONE) {
605 switch (vnet_hdr->gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
606 case VIRTIO_NET_HDR_GSO_TCPV4:
607 gso_type = SKB_GSO_TCPV4;
609 case VIRTIO_NET_HDR_GSO_TCPV6:
610 gso_type = SKB_GSO_TCPV6;
612 case VIRTIO_NET_HDR_GSO_UDP:
613 gso_type = SKB_GSO_UDP;
619 if (vnet_hdr->gso_type & VIRTIO_NET_HDR_GSO_ECN)
620 gso_type |= SKB_GSO_TCP_ECN;
622 if (vnet_hdr->gso_size == 0)
626 if (vnet_hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
627 if (!skb_partial_csum_set(skb, vnet_hdr->csum_start,
628 vnet_hdr->csum_offset))
632 if (vnet_hdr->gso_type != VIRTIO_NET_HDR_GSO_NONE) {
633 skb_shinfo(skb)->gso_size = vnet_hdr->gso_size;
634 skb_shinfo(skb)->gso_type = gso_type;
636 /* Header must be checked, and gso_segs computed. */
637 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
638 skb_shinfo(skb)->gso_segs = 0;
643 static int macvtap_skb_to_vnet_hdr(const struct sk_buff *skb,
644 struct virtio_net_hdr *vnet_hdr)
646 memset(vnet_hdr, 0, sizeof(*vnet_hdr));
648 if (skb_is_gso(skb)) {
649 struct skb_shared_info *sinfo = skb_shinfo(skb);
651 /* This is a hint as to how much should be linear. */
652 vnet_hdr->hdr_len = skb_headlen(skb);
653 vnet_hdr->gso_size = sinfo->gso_size;
654 if (sinfo->gso_type & SKB_GSO_TCPV4)
655 vnet_hdr->gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
656 else if (sinfo->gso_type & SKB_GSO_TCPV6)
657 vnet_hdr->gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
658 else if (sinfo->gso_type & SKB_GSO_UDP)
659 vnet_hdr->gso_type = VIRTIO_NET_HDR_GSO_UDP;
662 if (sinfo->gso_type & SKB_GSO_TCP_ECN)
663 vnet_hdr->gso_type |= VIRTIO_NET_HDR_GSO_ECN;
665 vnet_hdr->gso_type = VIRTIO_NET_HDR_GSO_NONE;
667 if (skb->ip_summed == CHECKSUM_PARTIAL) {
668 vnet_hdr->flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
669 vnet_hdr->csum_start = skb_checksum_start_offset(skb);
670 vnet_hdr->csum_offset = skb->csum_offset;
671 } else if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
672 vnet_hdr->flags = VIRTIO_NET_HDR_F_DATA_VALID;
673 } /* else everything is zero */
679 /* Get packet from user space buffer */
680 static ssize_t macvtap_get_user(struct macvtap_queue *q, struct msghdr *m,
681 const struct iovec *iv, unsigned long total_len,
682 size_t count, int noblock)
685 struct macvlan_dev *vlan;
686 unsigned long len = total_len;
688 struct virtio_net_hdr vnet_hdr = { 0 };
689 int vnet_hdr_len = 0;
691 bool zerocopy = false;
693 if (q->flags & IFF_VNET_HDR) {
694 vnet_hdr_len = q->vnet_hdr_sz;
697 if (len < vnet_hdr_len)
701 err = memcpy_fromiovecend((void *)&vnet_hdr, iv, 0,
705 if ((vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
706 vnet_hdr.csum_start + vnet_hdr.csum_offset + 2 >
708 vnet_hdr.hdr_len = vnet_hdr.csum_start +
709 vnet_hdr.csum_offset + 2;
711 if (vnet_hdr.hdr_len > len)
716 if (unlikely(len < ETH_HLEN))
720 if (unlikely(count > UIO_MAXIOV))
723 if (m && m->msg_control && sock_flag(&q->sk, SOCK_ZEROCOPY))
727 /* Userspace may produce vectors with count greater than
728 * MAX_SKB_FRAGS, so we need to linearize parts of the skb
729 * to let the rest of data to be fit in the frags.
731 if (count > MAX_SKB_FRAGS) {
732 copylen = iov_length(iv, count - MAX_SKB_FRAGS);
733 if (copylen < vnet_hdr_len)
736 copylen -= vnet_hdr_len;
738 /* There are 256 bytes to be copied in skb, so there is enough
739 * room for skb expand head in case it is used.
740 * The rest buffer is mapped from userspace.
742 if (copylen < vnet_hdr.hdr_len)
743 copylen = vnet_hdr.hdr_len;
745 copylen = GOODCOPY_LEN;
749 skb = macvtap_alloc_skb(&q->sk, NET_IP_ALIGN, copylen,
750 vnet_hdr.hdr_len, noblock, &err);
755 err = zerocopy_sg_from_iovec(skb, iv, vnet_hdr_len, count);
757 err = skb_copy_datagram_from_iovec(skb, 0, iv, vnet_hdr_len,
762 skb_set_network_header(skb, ETH_HLEN);
763 skb_reset_mac_header(skb);
764 skb->protocol = eth_hdr(skb)->h_proto;
767 err = macvtap_skb_from_vnet_hdr(skb, &vnet_hdr);
772 skb_probe_transport_header(skb, ETH_HLEN);
775 vlan = rcu_dereference_bh(q->vlan);
776 /* copy skb_ubuf_info for callback when skb has no error */
778 skb_shinfo(skb)->destructor_arg = m->msg_control;
779 skb_shinfo(skb)->tx_flags |= SKBTX_DEV_ZEROCOPY;
780 skb_shinfo(skb)->tx_flags |= SKBTX_SHARED_FRAG;
783 macvlan_start_xmit(skb, vlan->dev);
786 rcu_read_unlock_bh();
795 vlan = rcu_dereference_bh(q->vlan);
797 vlan->dev->stats.tx_dropped++;
798 rcu_read_unlock_bh();
803 static ssize_t macvtap_aio_write(struct kiocb *iocb, const struct iovec *iv,
804 unsigned long count, loff_t pos)
806 struct file *file = iocb->ki_filp;
807 ssize_t result = -ENOLINK;
808 struct macvtap_queue *q = file->private_data;
810 result = macvtap_get_user(q, NULL, iv, iov_length(iv, count), count,
811 file->f_flags & O_NONBLOCK);
815 /* Put packet to the user space buffer */
816 static ssize_t macvtap_put_user(struct macvtap_queue *q,
817 const struct sk_buff *skb,
818 const struct iovec *iv, int len)
820 struct macvlan_dev *vlan;
822 int vnet_hdr_len = 0;
826 if (q->flags & IFF_VNET_HDR) {
827 struct virtio_net_hdr vnet_hdr;
828 vnet_hdr_len = q->vnet_hdr_sz;
829 if ((len -= vnet_hdr_len) < 0)
832 ret = macvtap_skb_to_vnet_hdr(skb, &vnet_hdr);
836 if (memcpy_toiovecend(iv, (void *)&vnet_hdr, 0, sizeof(vnet_hdr)))
839 copied = vnet_hdr_len;
841 if (!vlan_tx_tag_present(skb))
842 len = min_t(int, skb->len, len);
849 veth.h_vlan_proto = htons(ETH_P_8021Q);
850 veth.h_vlan_TCI = htons(vlan_tx_tag_get(skb));
852 vlan_offset = offsetof(struct vlan_ethhdr, h_vlan_proto);
853 len = min_t(int, skb->len + VLAN_HLEN, len);
855 copy = min_t(int, vlan_offset, len);
856 ret = skb_copy_datagram_const_iovec(skb, 0, iv, copied, copy);
862 copy = min_t(int, sizeof(veth), len);
863 ret = memcpy_toiovecend(iv, (void *)&veth, copied, copy);
870 ret = skb_copy_datagram_const_iovec(skb, vlan_offset, iv, copied, len);
875 vlan = rcu_dereference_bh(q->vlan);
877 macvlan_count_rx(vlan, copied - vnet_hdr_len, ret == 0, 0);
878 rcu_read_unlock_bh();
880 return ret ? ret : copied;
883 static ssize_t macvtap_do_read(struct macvtap_queue *q, struct kiocb *iocb,
884 const struct iovec *iv, unsigned long len,
893 prepare_to_wait(sk_sleep(&q->sk), &wait,
896 /* Read frames from the queue */
897 skb = skb_dequeue(&q->sk.sk_receive_queue);
903 if (signal_pending(current)) {
907 /* Nothing to read, let's sleep */
911 ret = macvtap_put_user(q, skb, iv, len);
917 finish_wait(sk_sleep(&q->sk), &wait);
921 static ssize_t macvtap_aio_read(struct kiocb *iocb, const struct iovec *iv,
922 unsigned long count, loff_t pos)
924 struct file *file = iocb->ki_filp;
925 struct macvtap_queue *q = file->private_data;
926 ssize_t len, ret = 0;
928 len = iov_length(iv, count);
934 ret = macvtap_do_read(q, iocb, iv, len, file->f_flags & O_NONBLOCK);
935 ret = min_t(ssize_t, ret, len); /* XXX copied from tun.c. Why? */
940 static struct macvlan_dev *macvtap_get_vlan(struct macvtap_queue *q)
942 struct macvlan_dev *vlan;
945 vlan = rcu_dereference_bh(q->vlan);
948 rcu_read_unlock_bh();
953 static void macvtap_put_vlan(struct macvlan_dev *vlan)
958 static int macvtap_ioctl_set_queue(struct file *file, unsigned int flags)
960 struct macvtap_queue *q = file->private_data;
961 struct macvlan_dev *vlan;
964 vlan = macvtap_get_vlan(q);
968 if (flags & IFF_ATTACH_QUEUE)
969 ret = macvtap_enable_queue(vlan->dev, file, q);
970 else if (flags & IFF_DETACH_QUEUE)
971 ret = macvtap_disable_queue(q);
973 macvtap_put_vlan(vlan);
978 * provide compatibility with generic tun/tap interface
980 static long macvtap_ioctl(struct file *file, unsigned int cmd,
983 struct macvtap_queue *q = file->private_data;
984 struct macvlan_dev *vlan;
985 void __user *argp = (void __user *)arg;
986 struct ifreq __user *ifr = argp;
987 unsigned int __user *up = argp;
989 int __user *sp = argp;
995 /* ignore the name, just look at flags */
996 if (get_user(u, &ifr->ifr_flags))
1000 if ((u & ~(IFF_VNET_HDR | IFF_MULTI_QUEUE)) !=
1001 (IFF_NO_PI | IFF_TAP))
1009 vlan = macvtap_get_vlan(q);
1014 if (copy_to_user(&ifr->ifr_name, vlan->dev->name, IFNAMSIZ) ||
1015 put_user(q->flags, &ifr->ifr_flags))
1017 macvtap_put_vlan(vlan);
1021 if (get_user(u, &ifr->ifr_flags))
1023 return macvtap_ioctl_set_queue(file, u);
1025 case TUNGETFEATURES:
1026 if (put_user(IFF_TAP | IFF_NO_PI | IFF_VNET_HDR |
1027 IFF_MULTI_QUEUE, up))
1032 if (get_user(u, up))
1035 q->sk.sk_sndbuf = u;
1038 case TUNGETVNETHDRSZ:
1040 if (put_user(s, sp))
1044 case TUNSETVNETHDRSZ:
1045 if (get_user(s, sp))
1047 if (s < (int)sizeof(struct virtio_net_hdr))
1054 /* let the user check for future flags */
1055 if (arg & ~(TUN_F_CSUM | TUN_F_TSO4 | TUN_F_TSO6 |
1056 TUN_F_TSO_ECN | TUN_F_UFO))
1059 /* TODO: only accept frames with the features that
1060 got enabled for forwarded frames */
1061 if (!(q->flags & IFF_VNET_HDR))
1070 #ifdef CONFIG_COMPAT
1071 static long macvtap_compat_ioctl(struct file *file, unsigned int cmd,
1074 return macvtap_ioctl(file, cmd, (unsigned long)compat_ptr(arg));
1078 static const struct file_operations macvtap_fops = {
1079 .owner = THIS_MODULE,
1080 .open = macvtap_open,
1081 .release = macvtap_release,
1082 .aio_read = macvtap_aio_read,
1083 .aio_write = macvtap_aio_write,
1084 .poll = macvtap_poll,
1085 .llseek = no_llseek,
1086 .unlocked_ioctl = macvtap_ioctl,
1087 #ifdef CONFIG_COMPAT
1088 .compat_ioctl = macvtap_compat_ioctl,
1092 static int macvtap_sendmsg(struct kiocb *iocb, struct socket *sock,
1093 struct msghdr *m, size_t total_len)
1095 struct macvtap_queue *q = container_of(sock, struct macvtap_queue, sock);
1096 return macvtap_get_user(q, m, m->msg_iov, total_len, m->msg_iovlen,
1097 m->msg_flags & MSG_DONTWAIT);
1100 static int macvtap_recvmsg(struct kiocb *iocb, struct socket *sock,
1101 struct msghdr *m, size_t total_len,
1104 struct macvtap_queue *q = container_of(sock, struct macvtap_queue, sock);
1106 if (flags & ~(MSG_DONTWAIT|MSG_TRUNC))
1108 ret = macvtap_do_read(q, iocb, m->msg_iov, total_len,
1109 flags & MSG_DONTWAIT);
1110 if (ret > total_len) {
1111 m->msg_flags |= MSG_TRUNC;
1112 ret = flags & MSG_TRUNC ? ret : total_len;
1117 /* Ops structure to mimic raw sockets with tun */
1118 static const struct proto_ops macvtap_socket_ops = {
1119 .sendmsg = macvtap_sendmsg,
1120 .recvmsg = macvtap_recvmsg,
1123 /* Get an underlying socket object from tun file. Returns error unless file is
1124 * attached to a device. The returned object works like a packet socket, it
1125 * can be used for sock_sendmsg/sock_recvmsg. The caller is responsible for
1126 * holding a reference to the file for as long as the socket is in use. */
1127 struct socket *macvtap_get_socket(struct file *file)
1129 struct macvtap_queue *q;
1130 if (file->f_op != &macvtap_fops)
1131 return ERR_PTR(-EINVAL);
1132 q = file->private_data;
1134 return ERR_PTR(-EBADFD);
1137 EXPORT_SYMBOL_GPL(macvtap_get_socket);
1139 static int macvtap_device_event(struct notifier_block *unused,
1140 unsigned long event, void *ptr)
1142 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1143 struct macvlan_dev *vlan;
1144 struct device *classdev;
1148 if (dev->rtnl_link_ops != &macvtap_link_ops)
1151 vlan = netdev_priv(dev);
1154 case NETDEV_REGISTER:
1155 /* Create the device node here after the network device has
1156 * been registered but before register_netdevice has
1159 err = macvtap_get_minor(vlan);
1161 return notifier_from_errno(err);
1163 devt = MKDEV(MAJOR(macvtap_major), vlan->minor);
1164 classdev = device_create(macvtap_class, &dev->dev, devt,
1165 dev, "tap%d", dev->ifindex);
1166 if (IS_ERR(classdev)) {
1167 macvtap_free_minor(vlan);
1168 return notifier_from_errno(PTR_ERR(classdev));
1171 case NETDEV_UNREGISTER:
1172 devt = MKDEV(MAJOR(macvtap_major), vlan->minor);
1173 device_destroy(macvtap_class, devt);
1174 macvtap_free_minor(vlan);
1181 static struct notifier_block macvtap_notifier_block __read_mostly = {
1182 .notifier_call = macvtap_device_event,
1185 static int macvtap_init(void)
1189 err = alloc_chrdev_region(&macvtap_major, 0,
1190 MACVTAP_NUM_DEVS, "macvtap");
1194 cdev_init(&macvtap_cdev, &macvtap_fops);
1195 err = cdev_add(&macvtap_cdev, macvtap_major, MACVTAP_NUM_DEVS);
1199 macvtap_class = class_create(THIS_MODULE, "macvtap");
1200 if (IS_ERR(macvtap_class)) {
1201 err = PTR_ERR(macvtap_class);
1205 err = register_netdevice_notifier(&macvtap_notifier_block);
1209 err = macvlan_link_register(&macvtap_link_ops);
1216 unregister_netdevice_notifier(&macvtap_notifier_block);
1218 class_unregister(macvtap_class);
1220 cdev_del(&macvtap_cdev);
1222 unregister_chrdev_region(macvtap_major, MACVTAP_NUM_DEVS);
1226 module_init(macvtap_init);
1228 static void macvtap_exit(void)
1230 rtnl_link_unregister(&macvtap_link_ops);
1231 unregister_netdevice_notifier(&macvtap_notifier_block);
1232 class_unregister(macvtap_class);
1233 cdev_del(&macvtap_cdev);
1234 unregister_chrdev_region(macvtap_major, MACVTAP_NUM_DEVS);
1236 module_exit(macvtap_exit);
1238 MODULE_ALIAS_RTNL_LINK("macvtap");
1239 MODULE_AUTHOR("Arnd Bergmann <arnd@arndb.de>");
1240 MODULE_LICENSE("GPL");