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 macvlan_dev *vlan = netdev_priv(dev);
280 struct macvtap_queue *q = macvtap_get_queue(dev, skb);
281 netdev_features_t features;
285 if (skb_queue_len(&q->sk.sk_receive_queue) >= dev->tx_queue_len)
289 /* Apply the forward feature mask so that we perform segmentation
290 * according to users wishes.
292 features = netif_skb_features(skb) & vlan->tap_features;
293 if (netif_needs_gso(skb, features)) {
294 struct sk_buff *segs = __skb_gso_segment(skb, features, false);
300 skb_queue_tail(&q->sk.sk_receive_queue, skb);
306 struct sk_buff *nskb = segs->next;
309 skb_queue_tail(&q->sk.sk_receive_queue, segs);
313 skb_queue_tail(&q->sk.sk_receive_queue, skb);
317 wake_up_interruptible_poll(sk_sleep(&q->sk), POLLIN | POLLRDNORM | POLLRDBAND);
318 return NET_RX_SUCCESS;
326 * Receive is for data from the external interface (lowerdev),
327 * in case of macvtap, we can treat that the same way as
328 * forward, which macvlan cannot.
330 static int macvtap_receive(struct sk_buff *skb)
332 skb_push(skb, ETH_HLEN);
333 return macvtap_forward(skb->dev, skb);
336 static int macvtap_get_minor(struct macvlan_dev *vlan)
338 int retval = -ENOMEM;
340 mutex_lock(&minor_lock);
341 retval = idr_alloc(&minor_idr, vlan, 1, MACVTAP_NUM_DEVS, GFP_KERNEL);
343 vlan->minor = retval;
344 } else if (retval == -ENOSPC) {
345 printk(KERN_ERR "too many macvtap devices\n");
348 mutex_unlock(&minor_lock);
349 return retval < 0 ? retval : 0;
352 static void macvtap_free_minor(struct macvlan_dev *vlan)
354 mutex_lock(&minor_lock);
356 idr_remove(&minor_idr, vlan->minor);
359 mutex_unlock(&minor_lock);
362 static struct net_device *dev_get_by_macvtap_minor(int minor)
364 struct net_device *dev = NULL;
365 struct macvlan_dev *vlan;
367 mutex_lock(&minor_lock);
368 vlan = idr_find(&minor_idr, minor);
373 mutex_unlock(&minor_lock);
377 static int macvtap_newlink(struct net *src_net,
378 struct net_device *dev,
380 struct nlattr *data[])
382 struct macvlan_dev *vlan = netdev_priv(dev);
383 INIT_LIST_HEAD(&vlan->queue_list);
385 /* Since macvlan supports all offloads by default, make
386 * tap support all offloads also.
388 vlan->tap_features = TUN_OFFLOADS;
390 /* Don't put anything that may fail after macvlan_common_newlink
391 * because we can't undo what it does.
393 return macvlan_common_newlink(src_net, dev, tb, data,
394 macvtap_receive, macvtap_forward);
397 static void macvtap_dellink(struct net_device *dev,
398 struct list_head *head)
400 macvtap_del_queues(dev);
401 macvlan_dellink(dev, head);
404 static void macvtap_setup(struct net_device *dev)
406 macvlan_common_setup(dev);
407 dev->tx_queue_len = TUN_READQ_SIZE;
410 static struct rtnl_link_ops macvtap_link_ops __read_mostly = {
412 .setup = macvtap_setup,
413 .newlink = macvtap_newlink,
414 .dellink = macvtap_dellink,
418 static void macvtap_sock_write_space(struct sock *sk)
420 wait_queue_head_t *wqueue;
422 if (!sock_writeable(sk) ||
423 !test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags))
426 wqueue = sk_sleep(sk);
427 if (wqueue && waitqueue_active(wqueue))
428 wake_up_interruptible_poll(wqueue, POLLOUT | POLLWRNORM | POLLWRBAND);
431 static void macvtap_sock_destruct(struct sock *sk)
433 skb_queue_purge(&sk->sk_receive_queue);
436 static int macvtap_open(struct inode *inode, struct file *file)
438 struct net *net = current->nsproxy->net_ns;
439 struct net_device *dev = dev_get_by_macvtap_minor(iminor(inode));
440 struct macvtap_queue *q;
448 q = (struct macvtap_queue *)sk_alloc(net, AF_UNSPEC, GFP_KERNEL,
453 RCU_INIT_POINTER(q->sock.wq, &q->wq);
454 init_waitqueue_head(&q->wq.wait);
455 q->sock.type = SOCK_RAW;
456 q->sock.state = SS_CONNECTED;
458 q->sock.ops = &macvtap_socket_ops;
459 sock_init_data(&q->sock, &q->sk);
460 q->sk.sk_write_space = macvtap_sock_write_space;
461 q->sk.sk_destruct = macvtap_sock_destruct;
462 q->flags = IFF_VNET_HDR | IFF_NO_PI | IFF_TAP;
463 q->vnet_hdr_sz = sizeof(struct virtio_net_hdr);
466 * so far only KVM virtio_net uses macvtap, enable zero copy between
467 * guest kernel and host kernel when lower device supports zerocopy
469 * The macvlan supports zerocopy iff the lower device supports zero
470 * copy so we don't have to look at the lower device directly.
472 if ((dev->features & NETIF_F_HIGHDMA) && (dev->features & NETIF_F_SG))
473 sock_set_flag(&q->sk, SOCK_ZEROCOPY);
475 err = macvtap_set_queue(dev, file, q);
486 static int macvtap_release(struct inode *inode, struct file *file)
488 struct macvtap_queue *q = file->private_data;
489 macvtap_put_queue(q);
493 static unsigned int macvtap_poll(struct file *file, poll_table * wait)
495 struct macvtap_queue *q = file->private_data;
496 unsigned int mask = POLLERR;
502 poll_wait(file, &q->wq.wait, wait);
504 if (!skb_queue_empty(&q->sk.sk_receive_queue))
505 mask |= POLLIN | POLLRDNORM;
507 if (sock_writeable(&q->sk) ||
508 (!test_and_set_bit(SOCK_ASYNC_NOSPACE, &q->sock.flags) &&
509 sock_writeable(&q->sk)))
510 mask |= POLLOUT | POLLWRNORM;
516 static inline struct sk_buff *macvtap_alloc_skb(struct sock *sk, size_t prepad,
517 size_t len, size_t linear,
518 int noblock, int *err)
522 /* Under a page? Don't bother with paged skb. */
523 if (prepad + len < PAGE_SIZE || !linear)
526 skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
531 skb_reserve(skb, prepad);
532 skb_put(skb, linear);
533 skb->data_len = len - linear;
534 skb->len += len - linear;
539 /* set skb frags from iovec, this can move to core network code for reuse */
540 static int zerocopy_sg_from_iovec(struct sk_buff *skb, const struct iovec *from,
541 int offset, size_t count)
543 int len = iov_length(from, count) - offset;
544 int copy = skb_headlen(skb);
545 int size, offset1 = 0;
548 /* Skip over from offset */
549 while (count && (offset >= from->iov_len)) {
550 offset -= from->iov_len;
555 /* copy up to skb headlen */
556 while (count && (copy > 0)) {
557 size = min_t(unsigned int, copy, from->iov_len - offset);
558 if (copy_from_user(skb->data + offset1, from->iov_base + offset,
575 struct page *page[MAX_SKB_FRAGS];
578 unsigned long truesize;
580 len = from->iov_len - offset;
586 base = (unsigned long)from->iov_base + offset;
587 size = ((base & ~PAGE_MASK) + len + ~PAGE_MASK) >> PAGE_SHIFT;
588 if (i + size > MAX_SKB_FRAGS)
590 num_pages = get_user_pages_fast(base, size, 0, &page[i]);
591 if (num_pages != size) {
592 for (i = 0; i < num_pages; i++)
596 truesize = size * PAGE_SIZE;
597 skb->data_len += len;
599 skb->truesize += truesize;
600 atomic_add(truesize, &skb->sk->sk_wmem_alloc);
602 int off = base & ~PAGE_MASK;
603 int size = min_t(int, len, PAGE_SIZE - off);
604 __skb_fill_page_desc(skb, i, page[i], off, size);
605 skb_shinfo(skb)->nr_frags++;
606 /* increase sk_wmem_alloc */
618 * macvtap_skb_from_vnet_hdr and macvtap_skb_to_vnet_hdr should
619 * be shared with the tun/tap driver.
621 static int macvtap_skb_from_vnet_hdr(struct sk_buff *skb,
622 struct virtio_net_hdr *vnet_hdr)
624 unsigned short gso_type = 0;
625 if (vnet_hdr->gso_type != VIRTIO_NET_HDR_GSO_NONE) {
626 switch (vnet_hdr->gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
627 case VIRTIO_NET_HDR_GSO_TCPV4:
628 gso_type = SKB_GSO_TCPV4;
630 case VIRTIO_NET_HDR_GSO_TCPV6:
631 gso_type = SKB_GSO_TCPV6;
633 case VIRTIO_NET_HDR_GSO_UDP:
634 gso_type = SKB_GSO_UDP;
640 if (vnet_hdr->gso_type & VIRTIO_NET_HDR_GSO_ECN)
641 gso_type |= SKB_GSO_TCP_ECN;
643 if (vnet_hdr->gso_size == 0)
647 if (vnet_hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
648 if (!skb_partial_csum_set(skb, vnet_hdr->csum_start,
649 vnet_hdr->csum_offset))
653 if (vnet_hdr->gso_type != VIRTIO_NET_HDR_GSO_NONE) {
654 skb_shinfo(skb)->gso_size = vnet_hdr->gso_size;
655 skb_shinfo(skb)->gso_type = gso_type;
657 /* Header must be checked, and gso_segs computed. */
658 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
659 skb_shinfo(skb)->gso_segs = 0;
664 static int macvtap_skb_to_vnet_hdr(const struct sk_buff *skb,
665 struct virtio_net_hdr *vnet_hdr)
667 memset(vnet_hdr, 0, sizeof(*vnet_hdr));
669 if (skb_is_gso(skb)) {
670 struct skb_shared_info *sinfo = skb_shinfo(skb);
672 /* This is a hint as to how much should be linear. */
673 vnet_hdr->hdr_len = skb_headlen(skb);
674 vnet_hdr->gso_size = sinfo->gso_size;
675 if (sinfo->gso_type & SKB_GSO_TCPV4)
676 vnet_hdr->gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
677 else if (sinfo->gso_type & SKB_GSO_TCPV6)
678 vnet_hdr->gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
679 else if (sinfo->gso_type & SKB_GSO_UDP)
680 vnet_hdr->gso_type = VIRTIO_NET_HDR_GSO_UDP;
683 if (sinfo->gso_type & SKB_GSO_TCP_ECN)
684 vnet_hdr->gso_type |= VIRTIO_NET_HDR_GSO_ECN;
686 vnet_hdr->gso_type = VIRTIO_NET_HDR_GSO_NONE;
688 if (skb->ip_summed == CHECKSUM_PARTIAL) {
689 vnet_hdr->flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
690 vnet_hdr->csum_start = skb_checksum_start_offset(skb);
691 vnet_hdr->csum_offset = skb->csum_offset;
692 } else if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
693 vnet_hdr->flags = VIRTIO_NET_HDR_F_DATA_VALID;
694 } /* else everything is zero */
700 /* Get packet from user space buffer */
701 static ssize_t macvtap_get_user(struct macvtap_queue *q, struct msghdr *m,
702 const struct iovec *iv, unsigned long total_len,
703 size_t count, int noblock)
706 struct macvlan_dev *vlan;
707 unsigned long len = total_len;
709 struct virtio_net_hdr vnet_hdr = { 0 };
710 int vnet_hdr_len = 0;
712 bool zerocopy = false;
714 if (q->flags & IFF_VNET_HDR) {
715 vnet_hdr_len = q->vnet_hdr_sz;
718 if (len < vnet_hdr_len)
722 err = memcpy_fromiovecend((void *)&vnet_hdr, iv, 0,
726 if ((vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
727 vnet_hdr.csum_start + vnet_hdr.csum_offset + 2 >
729 vnet_hdr.hdr_len = vnet_hdr.csum_start +
730 vnet_hdr.csum_offset + 2;
732 if (vnet_hdr.hdr_len > len)
737 if (unlikely(len < ETH_HLEN))
741 if (unlikely(count > UIO_MAXIOV))
744 if (m && m->msg_control && sock_flag(&q->sk, SOCK_ZEROCOPY))
748 /* Userspace may produce vectors with count greater than
749 * MAX_SKB_FRAGS, so we need to linearize parts of the skb
750 * to let the rest of data to be fit in the frags.
752 if (count > MAX_SKB_FRAGS) {
753 copylen = iov_length(iv, count - MAX_SKB_FRAGS);
754 if (copylen < vnet_hdr_len)
757 copylen -= vnet_hdr_len;
759 /* There are 256 bytes to be copied in skb, so there is enough
760 * room for skb expand head in case it is used.
761 * The rest buffer is mapped from userspace.
763 if (copylen < vnet_hdr.hdr_len)
764 copylen = vnet_hdr.hdr_len;
766 copylen = GOODCOPY_LEN;
770 skb = macvtap_alloc_skb(&q->sk, NET_IP_ALIGN, copylen,
771 vnet_hdr.hdr_len, noblock, &err);
776 err = zerocopy_sg_from_iovec(skb, iv, vnet_hdr_len, count);
778 err = skb_copy_datagram_from_iovec(skb, 0, iv, vnet_hdr_len,
783 skb_set_network_header(skb, ETH_HLEN);
784 skb_reset_mac_header(skb);
785 skb->protocol = eth_hdr(skb)->h_proto;
788 err = macvtap_skb_from_vnet_hdr(skb, &vnet_hdr);
793 skb_probe_transport_header(skb, ETH_HLEN);
796 vlan = rcu_dereference(q->vlan);
797 /* copy skb_ubuf_info for callback when skb has no error */
799 skb_shinfo(skb)->destructor_arg = m->msg_control;
800 skb_shinfo(skb)->tx_flags |= SKBTX_DEV_ZEROCOPY;
801 skb_shinfo(skb)->tx_flags |= SKBTX_SHARED_FRAG;
804 macvlan_start_xmit(skb, vlan->dev);
816 vlan = rcu_dereference(q->vlan);
818 vlan->dev->stats.tx_dropped++;
824 static ssize_t macvtap_aio_write(struct kiocb *iocb, const struct iovec *iv,
825 unsigned long count, loff_t pos)
827 struct file *file = iocb->ki_filp;
828 ssize_t result = -ENOLINK;
829 struct macvtap_queue *q = file->private_data;
831 result = macvtap_get_user(q, NULL, iv, iov_length(iv, count), count,
832 file->f_flags & O_NONBLOCK);
836 /* Put packet to the user space buffer */
837 static ssize_t macvtap_put_user(struct macvtap_queue *q,
838 const struct sk_buff *skb,
839 const struct iovec *iv, int len)
841 struct macvlan_dev *vlan;
843 int vnet_hdr_len = 0;
847 if (q->flags & IFF_VNET_HDR) {
848 struct virtio_net_hdr vnet_hdr;
849 vnet_hdr_len = q->vnet_hdr_sz;
850 if ((len -= vnet_hdr_len) < 0)
853 ret = macvtap_skb_to_vnet_hdr(skb, &vnet_hdr);
857 if (memcpy_toiovecend(iv, (void *)&vnet_hdr, 0, sizeof(vnet_hdr)))
860 copied = vnet_hdr_len;
862 if (!vlan_tx_tag_present(skb))
863 len = min_t(int, skb->len, len);
870 veth.h_vlan_proto = htons(ETH_P_8021Q);
871 veth.h_vlan_TCI = htons(vlan_tx_tag_get(skb));
873 vlan_offset = offsetof(struct vlan_ethhdr, h_vlan_proto);
874 len = min_t(int, skb->len + VLAN_HLEN, len);
876 copy = min_t(int, vlan_offset, len);
877 ret = skb_copy_datagram_const_iovec(skb, 0, iv, copied, copy);
883 copy = min_t(int, sizeof(veth), len);
884 ret = memcpy_toiovecend(iv, (void *)&veth, copied, copy);
891 ret = skb_copy_datagram_const_iovec(skb, vlan_offset, iv, copied, len);
896 vlan = rcu_dereference(q->vlan);
898 macvlan_count_rx(vlan, copied - vnet_hdr_len, ret == 0, 0);
901 return ret ? ret : copied;
904 static ssize_t macvtap_do_read(struct macvtap_queue *q, struct kiocb *iocb,
905 const struct iovec *iv, unsigned long len,
914 prepare_to_wait(sk_sleep(&q->sk), &wait,
917 /* Read frames from the queue */
918 skb = skb_dequeue(&q->sk.sk_receive_queue);
924 if (signal_pending(current)) {
928 /* Nothing to read, let's sleep */
932 ret = macvtap_put_user(q, skb, iv, len);
938 finish_wait(sk_sleep(&q->sk), &wait);
942 static ssize_t macvtap_aio_read(struct kiocb *iocb, const struct iovec *iv,
943 unsigned long count, loff_t pos)
945 struct file *file = iocb->ki_filp;
946 struct macvtap_queue *q = file->private_data;
947 ssize_t len, ret = 0;
949 len = iov_length(iv, count);
955 ret = macvtap_do_read(q, iocb, iv, len, file->f_flags & O_NONBLOCK);
956 ret = min_t(ssize_t, ret, len); /* XXX copied from tun.c. Why? */
961 static struct macvlan_dev *macvtap_get_vlan(struct macvtap_queue *q)
963 struct macvlan_dev *vlan;
966 vlan = rtnl_dereference(q->vlan);
973 static void macvtap_put_vlan(struct macvlan_dev *vlan)
978 static int macvtap_ioctl_set_queue(struct file *file, unsigned int flags)
980 struct macvtap_queue *q = file->private_data;
981 struct macvlan_dev *vlan;
984 vlan = macvtap_get_vlan(q);
988 if (flags & IFF_ATTACH_QUEUE)
989 ret = macvtap_enable_queue(vlan->dev, file, q);
990 else if (flags & IFF_DETACH_QUEUE)
991 ret = macvtap_disable_queue(q);
995 macvtap_put_vlan(vlan);
999 static int set_offload(struct macvtap_queue *q, unsigned long arg)
1001 struct macvlan_dev *vlan;
1002 netdev_features_t features;
1003 netdev_features_t feature_mask = 0;
1005 vlan = rtnl_dereference(q->vlan);
1009 features = vlan->dev->features;
1011 if (arg & TUN_F_CSUM) {
1012 feature_mask = NETIF_F_HW_CSUM;
1014 if (arg & (TUN_F_TSO4 | TUN_F_TSO6)) {
1015 if (arg & TUN_F_TSO_ECN)
1016 feature_mask |= NETIF_F_TSO_ECN;
1017 if (arg & TUN_F_TSO4)
1018 feature_mask |= NETIF_F_TSO;
1019 if (arg & TUN_F_TSO6)
1020 feature_mask |= NETIF_F_TSO6;
1023 if (arg & TUN_F_UFO)
1024 feature_mask |= NETIF_F_UFO;
1027 /* tun/tap driver inverts the usage for TSO offloads, where
1028 * setting the TSO bit means that the userspace wants to
1029 * accept TSO frames and turning it off means that user space
1030 * does not support TSO.
1031 * For macvtap, we have to invert it to mean the same thing.
1032 * When user space turns off TSO, we turn off GSO/LRO so that
1033 * user-space will not receive TSO frames.
1035 if (feature_mask & (NETIF_F_TSO | NETIF_F_TSO6 | NETIF_F_UFO))
1036 features |= RX_OFFLOADS;
1038 features &= ~RX_OFFLOADS;
1040 /* tap_features are the same as features on tun/tap and
1041 * reflect user expectations.
1043 vlan->tap_features = vlan->dev->features &
1044 (feature_mask | ~TUN_OFFLOADS);
1045 vlan->set_features = features;
1046 netdev_update_features(vlan->dev);
1052 * provide compatibility with generic tun/tap interface
1054 static long macvtap_ioctl(struct file *file, unsigned int cmd,
1057 struct macvtap_queue *q = file->private_data;
1058 struct macvlan_dev *vlan;
1059 void __user *argp = (void __user *)arg;
1060 struct ifreq __user *ifr = argp;
1061 unsigned int __user *up = argp;
1063 int __user *sp = argp;
1069 /* ignore the name, just look at flags */
1070 if (get_user(u, &ifr->ifr_flags))
1074 if ((u & ~(IFF_VNET_HDR | IFF_MULTI_QUEUE)) !=
1075 (IFF_NO_PI | IFF_TAP))
1084 vlan = macvtap_get_vlan(q);
1091 if (copy_to_user(&ifr->ifr_name, vlan->dev->name, IFNAMSIZ) ||
1092 put_user(q->flags, &ifr->ifr_flags))
1094 macvtap_put_vlan(vlan);
1099 if (get_user(u, &ifr->ifr_flags))
1102 ret = macvtap_ioctl_set_queue(file, u);
1105 case TUNGETFEATURES:
1106 if (put_user(IFF_TAP | IFF_NO_PI | IFF_VNET_HDR |
1107 IFF_MULTI_QUEUE, up))
1112 if (get_user(u, up))
1115 q->sk.sk_sndbuf = u;
1118 case TUNGETVNETHDRSZ:
1120 if (put_user(s, sp))
1124 case TUNSETVNETHDRSZ:
1125 if (get_user(s, sp))
1127 if (s < (int)sizeof(struct virtio_net_hdr))
1134 /* let the user check for future flags */
1135 if (arg & ~(TUN_F_CSUM | TUN_F_TSO4 | TUN_F_TSO6 |
1136 TUN_F_TSO_ECN | TUN_F_UFO))
1139 /* TODO: only accept frames with the features that
1140 got enabled for forwarded frames */
1141 if (!(q->flags & IFF_VNET_HDR))
1144 ret = set_offload(q, arg);
1153 #ifdef CONFIG_COMPAT
1154 static long macvtap_compat_ioctl(struct file *file, unsigned int cmd,
1157 return macvtap_ioctl(file, cmd, (unsigned long)compat_ptr(arg));
1161 static const struct file_operations macvtap_fops = {
1162 .owner = THIS_MODULE,
1163 .open = macvtap_open,
1164 .release = macvtap_release,
1165 .aio_read = macvtap_aio_read,
1166 .aio_write = macvtap_aio_write,
1167 .poll = macvtap_poll,
1168 .llseek = no_llseek,
1169 .unlocked_ioctl = macvtap_ioctl,
1170 #ifdef CONFIG_COMPAT
1171 .compat_ioctl = macvtap_compat_ioctl,
1175 static int macvtap_sendmsg(struct kiocb *iocb, struct socket *sock,
1176 struct msghdr *m, size_t total_len)
1178 struct macvtap_queue *q = container_of(sock, struct macvtap_queue, sock);
1179 return macvtap_get_user(q, m, m->msg_iov, total_len, m->msg_iovlen,
1180 m->msg_flags & MSG_DONTWAIT);
1183 static int macvtap_recvmsg(struct kiocb *iocb, struct socket *sock,
1184 struct msghdr *m, size_t total_len,
1187 struct macvtap_queue *q = container_of(sock, struct macvtap_queue, sock);
1189 if (flags & ~(MSG_DONTWAIT|MSG_TRUNC))
1191 ret = macvtap_do_read(q, iocb, m->msg_iov, total_len,
1192 flags & MSG_DONTWAIT);
1193 if (ret > total_len) {
1194 m->msg_flags |= MSG_TRUNC;
1195 ret = flags & MSG_TRUNC ? ret : total_len;
1200 /* Ops structure to mimic raw sockets with tun */
1201 static const struct proto_ops macvtap_socket_ops = {
1202 .sendmsg = macvtap_sendmsg,
1203 .recvmsg = macvtap_recvmsg,
1206 /* Get an underlying socket object from tun file. Returns error unless file is
1207 * attached to a device. The returned object works like a packet socket, it
1208 * can be used for sock_sendmsg/sock_recvmsg. The caller is responsible for
1209 * holding a reference to the file for as long as the socket is in use. */
1210 struct socket *macvtap_get_socket(struct file *file)
1212 struct macvtap_queue *q;
1213 if (file->f_op != &macvtap_fops)
1214 return ERR_PTR(-EINVAL);
1215 q = file->private_data;
1217 return ERR_PTR(-EBADFD);
1220 EXPORT_SYMBOL_GPL(macvtap_get_socket);
1222 static int macvtap_device_event(struct notifier_block *unused,
1223 unsigned long event, void *ptr)
1225 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1226 struct macvlan_dev *vlan;
1227 struct device *classdev;
1231 if (dev->rtnl_link_ops != &macvtap_link_ops)
1234 vlan = netdev_priv(dev);
1237 case NETDEV_REGISTER:
1238 /* Create the device node here after the network device has
1239 * been registered but before register_netdevice has
1242 err = macvtap_get_minor(vlan);
1244 return notifier_from_errno(err);
1246 devt = MKDEV(MAJOR(macvtap_major), vlan->minor);
1247 classdev = device_create(macvtap_class, &dev->dev, devt,
1248 dev, "tap%d", dev->ifindex);
1249 if (IS_ERR(classdev)) {
1250 macvtap_free_minor(vlan);
1251 return notifier_from_errno(PTR_ERR(classdev));
1254 case NETDEV_UNREGISTER:
1255 devt = MKDEV(MAJOR(macvtap_major), vlan->minor);
1256 device_destroy(macvtap_class, devt);
1257 macvtap_free_minor(vlan);
1264 static struct notifier_block macvtap_notifier_block __read_mostly = {
1265 .notifier_call = macvtap_device_event,
1268 static int macvtap_init(void)
1272 err = alloc_chrdev_region(&macvtap_major, 0,
1273 MACVTAP_NUM_DEVS, "macvtap");
1277 cdev_init(&macvtap_cdev, &macvtap_fops);
1278 err = cdev_add(&macvtap_cdev, macvtap_major, MACVTAP_NUM_DEVS);
1282 macvtap_class = class_create(THIS_MODULE, "macvtap");
1283 if (IS_ERR(macvtap_class)) {
1284 err = PTR_ERR(macvtap_class);
1288 err = register_netdevice_notifier(&macvtap_notifier_block);
1292 err = macvlan_link_register(&macvtap_link_ops);
1299 unregister_netdevice_notifier(&macvtap_notifier_block);
1301 class_unregister(macvtap_class);
1303 cdev_del(&macvtap_cdev);
1305 unregister_chrdev_region(macvtap_major, MACVTAP_NUM_DEVS);
1309 module_init(macvtap_init);
1311 static void macvtap_exit(void)
1313 rtnl_link_unregister(&macvtap_link_ops);
1314 unregister_netdevice_notifier(&macvtap_notifier_block);
1315 class_unregister(macvtap_class);
1316 cdev_del(&macvtap_cdev);
1317 unregister_chrdev_region(macvtap_major, MACVTAP_NUM_DEVS);
1319 module_exit(macvtap_exit);
1321 MODULE_ALIAS_RTNL_LINK("macvtap");
1322 MODULE_AUTHOR("Arnd Bergmann <arnd@arndb.de>");
1323 MODULE_LICENSE("GPL");