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;
68 #define TUN_OFFLOADS (NETIF_F_HW_CSUM | NETIF_F_TSO_ECN | NETIF_F_TSO | \
69 NETIF_F_TSO6 | NETIF_F_UFO)
70 #define RX_OFFLOADS (NETIF_F_GRO | NETIF_F_LRO)
73 * The macvtap_queue and the macvlan_dev are loosely coupled, the
74 * pointers from one to the other can only be read while rcu_read_lock
77 * Both the file and the macvlan_dev hold a reference on the macvtap_queue
78 * through sock_hold(&q->sk). When the macvlan_dev goes away first,
79 * q->vlan becomes inaccessible. When the files gets closed,
80 * macvtap_get_queue() fails.
82 * There may still be references to the struct sock inside of the
83 * queue from outbound SKBs, but these never reference back to the
84 * file or the dev. The data structure is freed through __sk_free
85 * when both our references and any pending SKBs are gone.
88 static int macvtap_enable_queue(struct net_device *dev, struct file *file,
89 struct macvtap_queue *q)
91 struct macvlan_dev *vlan = netdev_priv(dev);
100 rcu_assign_pointer(vlan->taps[vlan->numvtaps], q);
101 q->queue_index = vlan->numvtaps;
109 static int macvtap_set_queue(struct net_device *dev, struct file *file,
110 struct macvtap_queue *q)
112 struct macvlan_dev *vlan = netdev_priv(dev);
116 if (vlan->numqueues == MAX_MACVTAP_QUEUES)
120 rcu_assign_pointer(q->vlan, vlan);
121 rcu_assign_pointer(vlan->taps[vlan->numvtaps], q);
125 q->queue_index = vlan->numvtaps;
127 file->private_data = q;
128 list_add_tail(&q->next, &vlan->queue_list);
138 static int macvtap_disable_queue(struct macvtap_queue *q)
140 struct macvlan_dev *vlan;
141 struct macvtap_queue *nq;
147 vlan = rtnl_dereference(q->vlan);
150 int index = q->queue_index;
151 BUG_ON(index >= vlan->numvtaps);
152 nq = rtnl_dereference(vlan->taps[vlan->numvtaps - 1]);
153 nq->queue_index = index;
155 rcu_assign_pointer(vlan->taps[index], nq);
156 RCU_INIT_POINTER(vlan->taps[vlan->numvtaps - 1], NULL);
166 * The file owning the queue got closed, give up both
167 * the reference that the files holds as well as the
168 * one from the macvlan_dev if that still exists.
170 * Using the spinlock makes sure that we don't get
171 * to the queue again after destroying it.
173 static void macvtap_put_queue(struct macvtap_queue *q)
175 struct macvlan_dev *vlan;
178 vlan = rtnl_dereference(q->vlan);
182 BUG_ON(macvtap_disable_queue(q));
185 RCU_INIT_POINTER(q->vlan, NULL);
187 list_del_init(&q->next);
197 * Select a queue based on the rxq of the device on which this packet
198 * arrived. If the incoming device is not mq, calculate a flow hash
199 * to select a queue. If all fails, find the first available queue.
200 * Cache vlan->numvtaps since it can become zero during the execution
203 static struct macvtap_queue *macvtap_get_queue(struct net_device *dev,
206 struct macvlan_dev *vlan = netdev_priv(dev);
207 struct macvtap_queue *tap = NULL;
208 /* Access to taps array is protected by rcu, but access to numvtaps
209 * isn't. Below we use it to lookup a queue, but treat it as a hint
210 * and validate that the result isn't NULL - in case we are
211 * racing against queue removal.
213 int numvtaps = ACCESS_ONCE(vlan->numvtaps);
219 /* Check if we can use flow to select a queue */
220 rxq = skb_get_rxhash(skb);
222 tap = rcu_dereference(vlan->taps[rxq % numvtaps]);
226 if (likely(skb_rx_queue_recorded(skb))) {
227 rxq = skb_get_rx_queue(skb);
229 while (unlikely(rxq >= numvtaps))
232 tap = rcu_dereference(vlan->taps[rxq]);
236 tap = rcu_dereference(vlan->taps[0]);
242 * The net_device is going away, give up the reference
243 * that it holds on all queues and safely set the pointer
244 * from the queues to NULL.
246 static void macvtap_del_queues(struct net_device *dev)
248 struct macvlan_dev *vlan = netdev_priv(dev);
249 struct macvtap_queue *q, *tmp, *qlist[MAX_MACVTAP_QUEUES];
253 list_for_each_entry_safe(q, tmp, &vlan->queue_list, next) {
254 list_del_init(&q->next);
256 RCU_INIT_POINTER(q->vlan, NULL);
261 for (i = 0; i < vlan->numvtaps; i++)
262 RCU_INIT_POINTER(vlan->taps[i], NULL);
263 BUG_ON(vlan->numvtaps);
264 BUG_ON(vlan->numqueues);
265 /* guarantee that any future macvtap_set_queue will fail */
266 vlan->numvtaps = MAX_MACVTAP_QUEUES;
268 for (--j; j >= 0; j--)
269 sock_put(&qlist[j]->sk);
273 * Forward happens for data that gets sent from one macvlan
274 * endpoint to another one in bridge mode. We just take
275 * the skb and put it into the receive queue.
277 static int macvtap_forward(struct net_device *dev, struct sk_buff *skb)
279 struct macvtap_queue *q = macvtap_get_queue(dev, skb);
283 if (skb_queue_len(&q->sk.sk_receive_queue) >= dev->tx_queue_len)
286 skb_queue_tail(&q->sk.sk_receive_queue, skb);
287 wake_up_interruptible_poll(sk_sleep(&q->sk), POLLIN | POLLRDNORM | POLLRDBAND);
288 return NET_RX_SUCCESS;
296 * Receive is for data from the external interface (lowerdev),
297 * in case of macvtap, we can treat that the same way as
298 * forward, which macvlan cannot.
300 static int macvtap_receive(struct sk_buff *skb)
302 skb_push(skb, ETH_HLEN);
303 return macvtap_forward(skb->dev, skb);
306 static int macvtap_get_minor(struct macvlan_dev *vlan)
308 int retval = -ENOMEM;
310 mutex_lock(&minor_lock);
311 retval = idr_alloc(&minor_idr, vlan, 1, MACVTAP_NUM_DEVS, GFP_KERNEL);
313 vlan->minor = retval;
314 } else if (retval == -ENOSPC) {
315 printk(KERN_ERR "too many macvtap devices\n");
318 mutex_unlock(&minor_lock);
319 return retval < 0 ? retval : 0;
322 static void macvtap_free_minor(struct macvlan_dev *vlan)
324 mutex_lock(&minor_lock);
326 idr_remove(&minor_idr, vlan->minor);
329 mutex_unlock(&minor_lock);
332 static struct net_device *dev_get_by_macvtap_minor(int minor)
334 struct net_device *dev = NULL;
335 struct macvlan_dev *vlan;
337 mutex_lock(&minor_lock);
338 vlan = idr_find(&minor_idr, minor);
343 mutex_unlock(&minor_lock);
347 static int macvtap_newlink(struct net *src_net,
348 struct net_device *dev,
350 struct nlattr *data[])
352 struct macvlan_dev *vlan = netdev_priv(dev);
353 INIT_LIST_HEAD(&vlan->queue_list);
355 /* Since macvlan supports all offloads by default, make
356 * tap support all offloads also.
358 vlan->tap_features = TUN_OFFLOADS;
360 /* Don't put anything that may fail after macvlan_common_newlink
361 * because we can't undo what it does.
363 return macvlan_common_newlink(src_net, dev, tb, data,
364 macvtap_receive, macvtap_forward);
367 static void macvtap_dellink(struct net_device *dev,
368 struct list_head *head)
370 macvtap_del_queues(dev);
371 macvlan_dellink(dev, head);
374 static void macvtap_setup(struct net_device *dev)
376 macvlan_common_setup(dev);
377 dev->tx_queue_len = TUN_READQ_SIZE;
380 static struct rtnl_link_ops macvtap_link_ops __read_mostly = {
382 .setup = macvtap_setup,
383 .newlink = macvtap_newlink,
384 .dellink = macvtap_dellink,
388 static void macvtap_sock_write_space(struct sock *sk)
390 wait_queue_head_t *wqueue;
392 if (!sock_writeable(sk) ||
393 !test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags))
396 wqueue = sk_sleep(sk);
397 if (wqueue && waitqueue_active(wqueue))
398 wake_up_interruptible_poll(wqueue, POLLOUT | POLLWRNORM | POLLWRBAND);
401 static void macvtap_sock_destruct(struct sock *sk)
403 skb_queue_purge(&sk->sk_receive_queue);
406 static int macvtap_open(struct inode *inode, struct file *file)
408 struct net *net = current->nsproxy->net_ns;
409 struct net_device *dev = dev_get_by_macvtap_minor(iminor(inode));
410 struct macvtap_queue *q;
418 q = (struct macvtap_queue *)sk_alloc(net, AF_UNSPEC, GFP_KERNEL,
423 RCU_INIT_POINTER(q->sock.wq, &q->wq);
424 init_waitqueue_head(&q->wq.wait);
425 q->sock.type = SOCK_RAW;
426 q->sock.state = SS_CONNECTED;
428 q->sock.ops = &macvtap_socket_ops;
429 sock_init_data(&q->sock, &q->sk);
430 q->sk.sk_write_space = macvtap_sock_write_space;
431 q->sk.sk_destruct = macvtap_sock_destruct;
432 q->flags = IFF_VNET_HDR | IFF_NO_PI | IFF_TAP;
433 q->vnet_hdr_sz = sizeof(struct virtio_net_hdr);
436 * so far only KVM virtio_net uses macvtap, enable zero copy between
437 * guest kernel and host kernel when lower device supports zerocopy
439 * The macvlan supports zerocopy iff the lower device supports zero
440 * copy so we don't have to look at the lower device directly.
442 if ((dev->features & NETIF_F_HIGHDMA) && (dev->features & NETIF_F_SG))
443 sock_set_flag(&q->sk, SOCK_ZEROCOPY);
445 err = macvtap_set_queue(dev, file, q);
456 static int macvtap_release(struct inode *inode, struct file *file)
458 struct macvtap_queue *q = file->private_data;
459 macvtap_put_queue(q);
463 static unsigned int macvtap_poll(struct file *file, poll_table * wait)
465 struct macvtap_queue *q = file->private_data;
466 unsigned int mask = POLLERR;
472 poll_wait(file, &q->wq.wait, wait);
474 if (!skb_queue_empty(&q->sk.sk_receive_queue))
475 mask |= POLLIN | POLLRDNORM;
477 if (sock_writeable(&q->sk) ||
478 (!test_and_set_bit(SOCK_ASYNC_NOSPACE, &q->sock.flags) &&
479 sock_writeable(&q->sk)))
480 mask |= POLLOUT | POLLWRNORM;
486 static inline struct sk_buff *macvtap_alloc_skb(struct sock *sk, size_t prepad,
487 size_t len, size_t linear,
488 int noblock, int *err)
492 /* Under a page? Don't bother with paged skb. */
493 if (prepad + len < PAGE_SIZE || !linear)
496 skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
501 skb_reserve(skb, prepad);
502 skb_put(skb, linear);
503 skb->data_len = len - linear;
504 skb->len += len - linear;
509 /* set skb frags from iovec, this can move to core network code for reuse */
510 static int zerocopy_sg_from_iovec(struct sk_buff *skb, const struct iovec *from,
511 int offset, size_t count)
513 int len = iov_length(from, count) - offset;
514 int copy = skb_headlen(skb);
515 int size, offset1 = 0;
518 /* Skip over from offset */
519 while (count && (offset >= from->iov_len)) {
520 offset -= from->iov_len;
525 /* copy up to skb headlen */
526 while (count && (copy > 0)) {
527 size = min_t(unsigned int, copy, from->iov_len - offset);
528 if (copy_from_user(skb->data + offset1, from->iov_base + offset,
545 struct page *page[MAX_SKB_FRAGS];
548 unsigned long truesize;
550 len = from->iov_len - offset;
556 base = (unsigned long)from->iov_base + offset;
557 size = ((base & ~PAGE_MASK) + len + ~PAGE_MASK) >> PAGE_SHIFT;
558 if (i + size > MAX_SKB_FRAGS)
560 num_pages = get_user_pages_fast(base, size, 0, &page[i]);
561 if (num_pages != size) {
562 for (i = 0; i < num_pages; i++)
566 truesize = size * PAGE_SIZE;
567 skb->data_len += len;
569 skb->truesize += truesize;
570 atomic_add(truesize, &skb->sk->sk_wmem_alloc);
572 int off = base & ~PAGE_MASK;
573 int size = min_t(int, len, PAGE_SIZE - off);
574 __skb_fill_page_desc(skb, i, page[i], off, size);
575 skb_shinfo(skb)->nr_frags++;
576 /* increase sk_wmem_alloc */
588 * macvtap_skb_from_vnet_hdr and macvtap_skb_to_vnet_hdr should
589 * be shared with the tun/tap driver.
591 static int macvtap_skb_from_vnet_hdr(struct sk_buff *skb,
592 struct virtio_net_hdr *vnet_hdr)
594 unsigned short gso_type = 0;
595 if (vnet_hdr->gso_type != VIRTIO_NET_HDR_GSO_NONE) {
596 switch (vnet_hdr->gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
597 case VIRTIO_NET_HDR_GSO_TCPV4:
598 gso_type = SKB_GSO_TCPV4;
600 case VIRTIO_NET_HDR_GSO_TCPV6:
601 gso_type = SKB_GSO_TCPV6;
603 case VIRTIO_NET_HDR_GSO_UDP:
604 gso_type = SKB_GSO_UDP;
610 if (vnet_hdr->gso_type & VIRTIO_NET_HDR_GSO_ECN)
611 gso_type |= SKB_GSO_TCP_ECN;
613 if (vnet_hdr->gso_size == 0)
617 if (vnet_hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
618 if (!skb_partial_csum_set(skb, vnet_hdr->csum_start,
619 vnet_hdr->csum_offset))
623 if (vnet_hdr->gso_type != VIRTIO_NET_HDR_GSO_NONE) {
624 skb_shinfo(skb)->gso_size = vnet_hdr->gso_size;
625 skb_shinfo(skb)->gso_type = gso_type;
627 /* Header must be checked, and gso_segs computed. */
628 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
629 skb_shinfo(skb)->gso_segs = 0;
634 static int macvtap_skb_to_vnet_hdr(const struct sk_buff *skb,
635 struct virtio_net_hdr *vnet_hdr)
637 memset(vnet_hdr, 0, sizeof(*vnet_hdr));
639 if (skb_is_gso(skb)) {
640 struct skb_shared_info *sinfo = skb_shinfo(skb);
642 /* This is a hint as to how much should be linear. */
643 vnet_hdr->hdr_len = skb_headlen(skb);
644 vnet_hdr->gso_size = sinfo->gso_size;
645 if (sinfo->gso_type & SKB_GSO_TCPV4)
646 vnet_hdr->gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
647 else if (sinfo->gso_type & SKB_GSO_TCPV6)
648 vnet_hdr->gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
649 else if (sinfo->gso_type & SKB_GSO_UDP)
650 vnet_hdr->gso_type = VIRTIO_NET_HDR_GSO_UDP;
653 if (sinfo->gso_type & SKB_GSO_TCP_ECN)
654 vnet_hdr->gso_type |= VIRTIO_NET_HDR_GSO_ECN;
656 vnet_hdr->gso_type = VIRTIO_NET_HDR_GSO_NONE;
658 if (skb->ip_summed == CHECKSUM_PARTIAL) {
659 vnet_hdr->flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
660 vnet_hdr->csum_start = skb_checksum_start_offset(skb);
661 vnet_hdr->csum_offset = skb->csum_offset;
662 } else if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
663 vnet_hdr->flags = VIRTIO_NET_HDR_F_DATA_VALID;
664 } /* else everything is zero */
670 /* Get packet from user space buffer */
671 static ssize_t macvtap_get_user(struct macvtap_queue *q, struct msghdr *m,
672 const struct iovec *iv, unsigned long total_len,
673 size_t count, int noblock)
676 struct macvlan_dev *vlan;
677 unsigned long len = total_len;
679 struct virtio_net_hdr vnet_hdr = { 0 };
680 int vnet_hdr_len = 0;
682 bool zerocopy = false;
684 if (q->flags & IFF_VNET_HDR) {
685 vnet_hdr_len = q->vnet_hdr_sz;
688 if (len < vnet_hdr_len)
692 err = memcpy_fromiovecend((void *)&vnet_hdr, iv, 0,
696 if ((vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
697 vnet_hdr.csum_start + vnet_hdr.csum_offset + 2 >
699 vnet_hdr.hdr_len = vnet_hdr.csum_start +
700 vnet_hdr.csum_offset + 2;
702 if (vnet_hdr.hdr_len > len)
707 if (unlikely(len < ETH_HLEN))
711 if (unlikely(count > UIO_MAXIOV))
714 if (m && m->msg_control && sock_flag(&q->sk, SOCK_ZEROCOPY))
718 /* Userspace may produce vectors with count greater than
719 * MAX_SKB_FRAGS, so we need to linearize parts of the skb
720 * to let the rest of data to be fit in the frags.
722 if (count > MAX_SKB_FRAGS) {
723 copylen = iov_length(iv, count - MAX_SKB_FRAGS);
724 if (copylen < vnet_hdr_len)
727 copylen -= vnet_hdr_len;
729 /* There are 256 bytes to be copied in skb, so there is enough
730 * room for skb expand head in case it is used.
731 * The rest buffer is mapped from userspace.
733 if (copylen < vnet_hdr.hdr_len)
734 copylen = vnet_hdr.hdr_len;
736 copylen = GOODCOPY_LEN;
740 skb = macvtap_alloc_skb(&q->sk, NET_IP_ALIGN, copylen,
741 vnet_hdr.hdr_len, noblock, &err);
746 err = zerocopy_sg_from_iovec(skb, iv, vnet_hdr_len, count);
748 err = skb_copy_datagram_from_iovec(skb, 0, iv, vnet_hdr_len,
753 skb_set_network_header(skb, ETH_HLEN);
754 skb_reset_mac_header(skb);
755 skb->protocol = eth_hdr(skb)->h_proto;
758 err = macvtap_skb_from_vnet_hdr(skb, &vnet_hdr);
763 skb_probe_transport_header(skb, ETH_HLEN);
766 vlan = rcu_dereference(q->vlan);
767 /* copy skb_ubuf_info for callback when skb has no error */
769 skb_shinfo(skb)->destructor_arg = m->msg_control;
770 skb_shinfo(skb)->tx_flags |= SKBTX_DEV_ZEROCOPY;
771 skb_shinfo(skb)->tx_flags |= SKBTX_SHARED_FRAG;
774 macvlan_start_xmit(skb, vlan->dev);
786 vlan = rcu_dereference(q->vlan);
788 vlan->dev->stats.tx_dropped++;
794 static ssize_t macvtap_aio_write(struct kiocb *iocb, const struct iovec *iv,
795 unsigned long count, loff_t pos)
797 struct file *file = iocb->ki_filp;
798 ssize_t result = -ENOLINK;
799 struct macvtap_queue *q = file->private_data;
801 result = macvtap_get_user(q, NULL, iv, iov_length(iv, count), count,
802 file->f_flags & O_NONBLOCK);
806 /* Put packet to the user space buffer */
807 static ssize_t macvtap_put_user(struct macvtap_queue *q,
808 const struct sk_buff *skb,
809 const struct iovec *iv, int len)
811 struct macvlan_dev *vlan;
813 int vnet_hdr_len = 0;
817 if (q->flags & IFF_VNET_HDR) {
818 struct virtio_net_hdr vnet_hdr;
819 vnet_hdr_len = q->vnet_hdr_sz;
820 if ((len -= vnet_hdr_len) < 0)
823 ret = macvtap_skb_to_vnet_hdr(skb, &vnet_hdr);
827 if (memcpy_toiovecend(iv, (void *)&vnet_hdr, 0, sizeof(vnet_hdr)))
830 copied = vnet_hdr_len;
832 if (!vlan_tx_tag_present(skb))
833 len = min_t(int, skb->len, len);
840 veth.h_vlan_proto = htons(ETH_P_8021Q);
841 veth.h_vlan_TCI = htons(vlan_tx_tag_get(skb));
843 vlan_offset = offsetof(struct vlan_ethhdr, h_vlan_proto);
844 len = min_t(int, skb->len + VLAN_HLEN, len);
846 copy = min_t(int, vlan_offset, len);
847 ret = skb_copy_datagram_const_iovec(skb, 0, iv, copied, copy);
853 copy = min_t(int, sizeof(veth), len);
854 ret = memcpy_toiovecend(iv, (void *)&veth, copied, copy);
861 ret = skb_copy_datagram_const_iovec(skb, vlan_offset, iv, copied, len);
866 vlan = rcu_dereference(q->vlan);
868 macvlan_count_rx(vlan, copied - vnet_hdr_len, ret == 0, 0);
871 return ret ? ret : copied;
874 static ssize_t macvtap_do_read(struct macvtap_queue *q, struct kiocb *iocb,
875 const struct iovec *iv, unsigned long len,
884 prepare_to_wait(sk_sleep(&q->sk), &wait,
887 /* Read frames from the queue */
888 skb = skb_dequeue(&q->sk.sk_receive_queue);
894 if (signal_pending(current)) {
898 /* Nothing to read, let's sleep */
902 ret = macvtap_put_user(q, skb, iv, len);
908 finish_wait(sk_sleep(&q->sk), &wait);
912 static ssize_t macvtap_aio_read(struct kiocb *iocb, const struct iovec *iv,
913 unsigned long count, loff_t pos)
915 struct file *file = iocb->ki_filp;
916 struct macvtap_queue *q = file->private_data;
917 ssize_t len, ret = 0;
919 len = iov_length(iv, count);
925 ret = macvtap_do_read(q, iocb, iv, len, file->f_flags & O_NONBLOCK);
926 ret = min_t(ssize_t, ret, len); /* XXX copied from tun.c. Why? */
931 static struct macvlan_dev *macvtap_get_vlan(struct macvtap_queue *q)
933 struct macvlan_dev *vlan;
936 vlan = rtnl_dereference(q->vlan);
943 static void macvtap_put_vlan(struct macvlan_dev *vlan)
948 static int macvtap_ioctl_set_queue(struct file *file, unsigned int flags)
950 struct macvtap_queue *q = file->private_data;
951 struct macvlan_dev *vlan;
954 vlan = macvtap_get_vlan(q);
958 if (flags & IFF_ATTACH_QUEUE)
959 ret = macvtap_enable_queue(vlan->dev, file, q);
960 else if (flags & IFF_DETACH_QUEUE)
961 ret = macvtap_disable_queue(q);
965 macvtap_put_vlan(vlan);
969 static int set_offload(struct macvtap_queue *q, unsigned long arg)
971 struct macvlan_dev *vlan;
972 netdev_features_t features;
973 netdev_features_t feature_mask = 0;
975 vlan = rtnl_dereference(q->vlan);
979 features = vlan->dev->features;
981 if (arg & TUN_F_CSUM) {
982 feature_mask = NETIF_F_HW_CSUM;
984 if (arg & (TUN_F_TSO4 | TUN_F_TSO6)) {
985 if (arg & TUN_F_TSO_ECN)
986 feature_mask |= NETIF_F_TSO_ECN;
987 if (arg & TUN_F_TSO4)
988 feature_mask |= NETIF_F_TSO;
989 if (arg & TUN_F_TSO6)
990 feature_mask |= NETIF_F_TSO6;
994 feature_mask |= NETIF_F_UFO;
997 /* tun/tap driver inverts the usage for TSO offloads, where
998 * setting the TSO bit means that the userspace wants to
999 * accept TSO frames and turning it off means that user space
1000 * does not support TSO.
1001 * For macvtap, we have to invert it to mean the same thing.
1002 * When user space turns off TSO, we turn off GSO/LRO so that
1003 * user-space will not receive TSO frames.
1005 if (feature_mask & (NETIF_F_TSO | NETIF_F_TSO6 | NETIF_F_UFO))
1006 features |= RX_OFFLOADS;
1008 features &= ~RX_OFFLOADS;
1010 /* tap_features are the same as features on tun/tap and
1011 * reflect user expectations.
1013 vlan->tap_features = vlan->dev->features &
1014 (feature_mask | ~TUN_OFFLOADS);
1015 vlan->set_features = features;
1016 netdev_update_features(vlan->dev);
1022 * provide compatibility with generic tun/tap interface
1024 static long macvtap_ioctl(struct file *file, unsigned int cmd,
1027 struct macvtap_queue *q = file->private_data;
1028 struct macvlan_dev *vlan;
1029 void __user *argp = (void __user *)arg;
1030 struct ifreq __user *ifr = argp;
1031 unsigned int __user *up = argp;
1033 int __user *sp = argp;
1039 /* ignore the name, just look at flags */
1040 if (get_user(u, &ifr->ifr_flags))
1044 if ((u & ~(IFF_VNET_HDR | IFF_MULTI_QUEUE)) !=
1045 (IFF_NO_PI | IFF_TAP))
1054 vlan = macvtap_get_vlan(q);
1061 if (copy_to_user(&ifr->ifr_name, vlan->dev->name, IFNAMSIZ) ||
1062 put_user(q->flags, &ifr->ifr_flags))
1064 macvtap_put_vlan(vlan);
1069 if (get_user(u, &ifr->ifr_flags))
1072 ret = macvtap_ioctl_set_queue(file, u);
1075 case TUNGETFEATURES:
1076 if (put_user(IFF_TAP | IFF_NO_PI | IFF_VNET_HDR |
1077 IFF_MULTI_QUEUE, up))
1082 if (get_user(u, up))
1085 q->sk.sk_sndbuf = u;
1088 case TUNGETVNETHDRSZ:
1090 if (put_user(s, sp))
1094 case TUNSETVNETHDRSZ:
1095 if (get_user(s, sp))
1097 if (s < (int)sizeof(struct virtio_net_hdr))
1104 /* let the user check for future flags */
1105 if (arg & ~(TUN_F_CSUM | TUN_F_TSO4 | TUN_F_TSO6 |
1106 TUN_F_TSO_ECN | TUN_F_UFO))
1109 /* TODO: only accept frames with the features that
1110 got enabled for forwarded frames */
1111 if (!(q->flags & IFF_VNET_HDR))
1114 ret = set_offload(q, arg);
1123 #ifdef CONFIG_COMPAT
1124 static long macvtap_compat_ioctl(struct file *file, unsigned int cmd,
1127 return macvtap_ioctl(file, cmd, (unsigned long)compat_ptr(arg));
1131 static const struct file_operations macvtap_fops = {
1132 .owner = THIS_MODULE,
1133 .open = macvtap_open,
1134 .release = macvtap_release,
1135 .aio_read = macvtap_aio_read,
1136 .aio_write = macvtap_aio_write,
1137 .poll = macvtap_poll,
1138 .llseek = no_llseek,
1139 .unlocked_ioctl = macvtap_ioctl,
1140 #ifdef CONFIG_COMPAT
1141 .compat_ioctl = macvtap_compat_ioctl,
1145 static int macvtap_sendmsg(struct kiocb *iocb, struct socket *sock,
1146 struct msghdr *m, size_t total_len)
1148 struct macvtap_queue *q = container_of(sock, struct macvtap_queue, sock);
1149 return macvtap_get_user(q, m, m->msg_iov, total_len, m->msg_iovlen,
1150 m->msg_flags & MSG_DONTWAIT);
1153 static int macvtap_recvmsg(struct kiocb *iocb, struct socket *sock,
1154 struct msghdr *m, size_t total_len,
1157 struct macvtap_queue *q = container_of(sock, struct macvtap_queue, sock);
1159 if (flags & ~(MSG_DONTWAIT|MSG_TRUNC))
1161 ret = macvtap_do_read(q, iocb, m->msg_iov, total_len,
1162 flags & MSG_DONTWAIT);
1163 if (ret > total_len) {
1164 m->msg_flags |= MSG_TRUNC;
1165 ret = flags & MSG_TRUNC ? ret : total_len;
1170 /* Ops structure to mimic raw sockets with tun */
1171 static const struct proto_ops macvtap_socket_ops = {
1172 .sendmsg = macvtap_sendmsg,
1173 .recvmsg = macvtap_recvmsg,
1176 /* Get an underlying socket object from tun file. Returns error unless file is
1177 * attached to a device. The returned object works like a packet socket, it
1178 * can be used for sock_sendmsg/sock_recvmsg. The caller is responsible for
1179 * holding a reference to the file for as long as the socket is in use. */
1180 struct socket *macvtap_get_socket(struct file *file)
1182 struct macvtap_queue *q;
1183 if (file->f_op != &macvtap_fops)
1184 return ERR_PTR(-EINVAL);
1185 q = file->private_data;
1187 return ERR_PTR(-EBADFD);
1190 EXPORT_SYMBOL_GPL(macvtap_get_socket);
1192 static int macvtap_device_event(struct notifier_block *unused,
1193 unsigned long event, void *ptr)
1195 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1196 struct macvlan_dev *vlan;
1197 struct device *classdev;
1201 if (dev->rtnl_link_ops != &macvtap_link_ops)
1204 vlan = netdev_priv(dev);
1207 case NETDEV_REGISTER:
1208 /* Create the device node here after the network device has
1209 * been registered but before register_netdevice has
1212 err = macvtap_get_minor(vlan);
1214 return notifier_from_errno(err);
1216 devt = MKDEV(MAJOR(macvtap_major), vlan->minor);
1217 classdev = device_create(macvtap_class, &dev->dev, devt,
1218 dev, "tap%d", dev->ifindex);
1219 if (IS_ERR(classdev)) {
1220 macvtap_free_minor(vlan);
1221 return notifier_from_errno(PTR_ERR(classdev));
1224 case NETDEV_UNREGISTER:
1225 devt = MKDEV(MAJOR(macvtap_major), vlan->minor);
1226 device_destroy(macvtap_class, devt);
1227 macvtap_free_minor(vlan);
1234 static struct notifier_block macvtap_notifier_block __read_mostly = {
1235 .notifier_call = macvtap_device_event,
1238 static int macvtap_init(void)
1242 err = alloc_chrdev_region(&macvtap_major, 0,
1243 MACVTAP_NUM_DEVS, "macvtap");
1247 cdev_init(&macvtap_cdev, &macvtap_fops);
1248 err = cdev_add(&macvtap_cdev, macvtap_major, MACVTAP_NUM_DEVS);
1252 macvtap_class = class_create(THIS_MODULE, "macvtap");
1253 if (IS_ERR(macvtap_class)) {
1254 err = PTR_ERR(macvtap_class);
1258 err = register_netdevice_notifier(&macvtap_notifier_block);
1262 err = macvlan_link_register(&macvtap_link_ops);
1269 unregister_netdevice_notifier(&macvtap_notifier_block);
1271 class_unregister(macvtap_class);
1273 cdev_del(&macvtap_cdev);
1275 unregister_chrdev_region(macvtap_major, MACVTAP_NUM_DEVS);
1279 module_init(macvtap_init);
1281 static void macvtap_exit(void)
1283 rtnl_link_unregister(&macvtap_link_ops);
1284 unregister_netdevice_notifier(&macvtap_notifier_block);
1285 class_unregister(macvtap_class);
1286 cdev_del(&macvtap_cdev);
1287 unregister_chrdev_region(macvtap_major, MACVTAP_NUM_DEVS);
1289 module_exit(macvtap_exit);
1291 MODULE_ALIAS_RTNL_LINK("macvtap");
1292 MODULE_AUTHOR("Arnd Bergmann <arnd@arndb.de>");
1293 MODULE_LICENSE("GPL");