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)
71 #define TAP_FEATURES (NETIF_F_GSO | NETIF_F_SG)
75 * The macvtap_queue and the macvlan_dev are loosely coupled, the
76 * pointers from one to the other can only be read while rcu_read_lock
79 * Both the file and the macvlan_dev hold a reference on the macvtap_queue
80 * through sock_hold(&q->sk). When the macvlan_dev goes away first,
81 * q->vlan becomes inaccessible. When the files gets closed,
82 * macvtap_get_queue() fails.
84 * There may still be references to the struct sock inside of the
85 * queue from outbound SKBs, but these never reference back to the
86 * file or the dev. The data structure is freed through __sk_free
87 * when both our references and any pending SKBs are gone.
90 static int macvtap_enable_queue(struct net_device *dev, struct file *file,
91 struct macvtap_queue *q)
93 struct macvlan_dev *vlan = netdev_priv(dev);
102 rcu_assign_pointer(vlan->taps[vlan->numvtaps], q);
103 q->queue_index = vlan->numvtaps;
111 static int macvtap_set_queue(struct net_device *dev, struct file *file,
112 struct macvtap_queue *q)
114 struct macvlan_dev *vlan = netdev_priv(dev);
118 if (vlan->numqueues == MAX_MACVTAP_QUEUES)
122 rcu_assign_pointer(q->vlan, vlan);
123 rcu_assign_pointer(vlan->taps[vlan->numvtaps], q);
127 q->queue_index = vlan->numvtaps;
129 file->private_data = q;
130 list_add_tail(&q->next, &vlan->queue_list);
140 static int macvtap_disable_queue(struct macvtap_queue *q)
142 struct macvlan_dev *vlan;
143 struct macvtap_queue *nq;
149 vlan = rtnl_dereference(q->vlan);
152 int index = q->queue_index;
153 BUG_ON(index >= vlan->numvtaps);
154 nq = rtnl_dereference(vlan->taps[vlan->numvtaps - 1]);
155 nq->queue_index = index;
157 rcu_assign_pointer(vlan->taps[index], nq);
158 RCU_INIT_POINTER(vlan->taps[vlan->numvtaps - 1], NULL);
168 * The file owning the queue got closed, give up both
169 * the reference that the files holds as well as the
170 * one from the macvlan_dev if that still exists.
172 * Using the spinlock makes sure that we don't get
173 * to the queue again after destroying it.
175 static void macvtap_put_queue(struct macvtap_queue *q)
177 struct macvlan_dev *vlan;
180 vlan = rtnl_dereference(q->vlan);
184 BUG_ON(macvtap_disable_queue(q));
187 RCU_INIT_POINTER(q->vlan, NULL);
189 list_del_init(&q->next);
199 * Select a queue based on the rxq of the device on which this packet
200 * arrived. If the incoming device is not mq, calculate a flow hash
201 * to select a queue. If all fails, find the first available queue.
202 * Cache vlan->numvtaps since it can become zero during the execution
205 static struct macvtap_queue *macvtap_get_queue(struct net_device *dev,
208 struct macvlan_dev *vlan = netdev_priv(dev);
209 struct macvtap_queue *tap = NULL;
210 /* Access to taps array is protected by rcu, but access to numvtaps
211 * isn't. Below we use it to lookup a queue, but treat it as a hint
212 * and validate that the result isn't NULL - in case we are
213 * racing against queue removal.
215 int numvtaps = ACCESS_ONCE(vlan->numvtaps);
221 /* Check if we can use flow to select a queue */
222 rxq = skb_get_rxhash(skb);
224 tap = rcu_dereference(vlan->taps[rxq % numvtaps]);
228 if (likely(skb_rx_queue_recorded(skb))) {
229 rxq = skb_get_rx_queue(skb);
231 while (unlikely(rxq >= numvtaps))
234 tap = rcu_dereference(vlan->taps[rxq]);
238 tap = rcu_dereference(vlan->taps[0]);
244 * The net_device is going away, give up the reference
245 * that it holds on all queues and safely set the pointer
246 * from the queues to NULL.
248 static void macvtap_del_queues(struct net_device *dev)
250 struct macvlan_dev *vlan = netdev_priv(dev);
251 struct macvtap_queue *q, *tmp, *qlist[MAX_MACVTAP_QUEUES];
255 list_for_each_entry_safe(q, tmp, &vlan->queue_list, next) {
256 list_del_init(&q->next);
258 RCU_INIT_POINTER(q->vlan, NULL);
263 for (i = 0; i < vlan->numvtaps; i++)
264 RCU_INIT_POINTER(vlan->taps[i], NULL);
265 BUG_ON(vlan->numvtaps);
266 BUG_ON(vlan->numqueues);
267 /* guarantee that any future macvtap_set_queue will fail */
268 vlan->numvtaps = MAX_MACVTAP_QUEUES;
270 for (--j; j >= 0; j--)
271 sock_put(&qlist[j]->sk);
275 * Forward happens for data that gets sent from one macvlan
276 * endpoint to another one in bridge mode. We just take
277 * the skb and put it into the receive queue.
279 static int macvtap_forward(struct net_device *dev, struct sk_buff *skb)
281 struct macvlan_dev *vlan = netdev_priv(dev);
282 struct macvtap_queue *q = macvtap_get_queue(dev, skb);
283 netdev_features_t features = TAP_FEATURES;
288 if (skb_queue_len(&q->sk.sk_receive_queue) >= dev->tx_queue_len)
292 /* Apply the forward feature mask so that we perform segmentation
293 * according to users wishes.
295 features |= vlan->tap_features;
296 if (netif_needs_gso(skb, features)) {
297 struct sk_buff *segs = __skb_gso_segment(skb, features, false);
303 skb_queue_tail(&q->sk.sk_receive_queue, skb);
309 struct sk_buff *nskb = segs->next;
312 skb_queue_tail(&q->sk.sk_receive_queue, segs);
316 skb_queue_tail(&q->sk.sk_receive_queue, skb);
320 wake_up_interruptible_poll(sk_sleep(&q->sk), POLLIN | POLLRDNORM | POLLRDBAND);
321 return NET_RX_SUCCESS;
329 * Receive is for data from the external interface (lowerdev),
330 * in case of macvtap, we can treat that the same way as
331 * forward, which macvlan cannot.
333 static int macvtap_receive(struct sk_buff *skb)
335 skb_push(skb, ETH_HLEN);
336 return macvtap_forward(skb->dev, skb);
339 static int macvtap_get_minor(struct macvlan_dev *vlan)
341 int retval = -ENOMEM;
343 mutex_lock(&minor_lock);
344 retval = idr_alloc(&minor_idr, vlan, 1, MACVTAP_NUM_DEVS, GFP_KERNEL);
346 vlan->minor = retval;
347 } else if (retval == -ENOSPC) {
348 printk(KERN_ERR "too many macvtap devices\n");
351 mutex_unlock(&minor_lock);
352 return retval < 0 ? retval : 0;
355 static void macvtap_free_minor(struct macvlan_dev *vlan)
357 mutex_lock(&minor_lock);
359 idr_remove(&minor_idr, vlan->minor);
362 mutex_unlock(&minor_lock);
365 static struct net_device *dev_get_by_macvtap_minor(int minor)
367 struct net_device *dev = NULL;
368 struct macvlan_dev *vlan;
370 mutex_lock(&minor_lock);
371 vlan = idr_find(&minor_idr, minor);
376 mutex_unlock(&minor_lock);
380 static int macvtap_newlink(struct net *src_net,
381 struct net_device *dev,
383 struct nlattr *data[])
385 struct macvlan_dev *vlan = netdev_priv(dev);
386 INIT_LIST_HEAD(&vlan->queue_list);
388 /* Since macvlan supports all offloads by default, make
389 * tap support all offloads also.
391 vlan->tap_features = TUN_OFFLOADS;
393 /* Don't put anything that may fail after macvlan_common_newlink
394 * because we can't undo what it does.
396 return macvlan_common_newlink(src_net, dev, tb, data,
397 macvtap_receive, macvtap_forward);
400 static void macvtap_dellink(struct net_device *dev,
401 struct list_head *head)
403 macvtap_del_queues(dev);
404 macvlan_dellink(dev, head);
407 static void macvtap_setup(struct net_device *dev)
409 macvlan_common_setup(dev);
410 dev->tx_queue_len = TUN_READQ_SIZE;
413 static struct rtnl_link_ops macvtap_link_ops __read_mostly = {
415 .setup = macvtap_setup,
416 .newlink = macvtap_newlink,
417 .dellink = macvtap_dellink,
421 static void macvtap_sock_write_space(struct sock *sk)
423 wait_queue_head_t *wqueue;
425 if (!sock_writeable(sk) ||
426 !test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags))
429 wqueue = sk_sleep(sk);
430 if (wqueue && waitqueue_active(wqueue))
431 wake_up_interruptible_poll(wqueue, POLLOUT | POLLWRNORM | POLLWRBAND);
434 static void macvtap_sock_destruct(struct sock *sk)
436 skb_queue_purge(&sk->sk_receive_queue);
439 static int macvtap_open(struct inode *inode, struct file *file)
441 struct net *net = current->nsproxy->net_ns;
442 struct net_device *dev = dev_get_by_macvtap_minor(iminor(inode));
443 struct macvtap_queue *q;
451 q = (struct macvtap_queue *)sk_alloc(net, AF_UNSPEC, GFP_KERNEL,
456 RCU_INIT_POINTER(q->sock.wq, &q->wq);
457 init_waitqueue_head(&q->wq.wait);
458 q->sock.type = SOCK_RAW;
459 q->sock.state = SS_CONNECTED;
461 q->sock.ops = &macvtap_socket_ops;
462 sock_init_data(&q->sock, &q->sk);
463 q->sk.sk_write_space = macvtap_sock_write_space;
464 q->sk.sk_destruct = macvtap_sock_destruct;
465 q->flags = IFF_VNET_HDR | IFF_NO_PI | IFF_TAP;
466 q->vnet_hdr_sz = sizeof(struct virtio_net_hdr);
469 * so far only KVM virtio_net uses macvtap, enable zero copy between
470 * guest kernel and host kernel when lower device supports zerocopy
472 * The macvlan supports zerocopy iff the lower device supports zero
473 * copy so we don't have to look at the lower device directly.
475 if ((dev->features & NETIF_F_HIGHDMA) && (dev->features & NETIF_F_SG))
476 sock_set_flag(&q->sk, SOCK_ZEROCOPY);
478 err = macvtap_set_queue(dev, file, q);
489 static int macvtap_release(struct inode *inode, struct file *file)
491 struct macvtap_queue *q = file->private_data;
492 macvtap_put_queue(q);
496 static unsigned int macvtap_poll(struct file *file, poll_table * wait)
498 struct macvtap_queue *q = file->private_data;
499 unsigned int mask = POLLERR;
505 poll_wait(file, &q->wq.wait, wait);
507 if (!skb_queue_empty(&q->sk.sk_receive_queue))
508 mask |= POLLIN | POLLRDNORM;
510 if (sock_writeable(&q->sk) ||
511 (!test_and_set_bit(SOCK_ASYNC_NOSPACE, &q->sock.flags) &&
512 sock_writeable(&q->sk)))
513 mask |= POLLOUT | POLLWRNORM;
519 static inline struct sk_buff *macvtap_alloc_skb(struct sock *sk, size_t prepad,
520 size_t len, size_t linear,
521 int noblock, int *err)
525 /* Under a page? Don't bother with paged skb. */
526 if (prepad + len < PAGE_SIZE || !linear)
529 skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
534 skb_reserve(skb, prepad);
535 skb_put(skb, linear);
536 skb->data_len = len - linear;
537 skb->len += len - linear;
542 /* set skb frags from iovec, this can move to core network code for reuse */
543 static int zerocopy_sg_from_iovec(struct sk_buff *skb, const struct iovec *from,
544 int offset, size_t count)
546 int len = iov_length(from, count) - offset;
547 int copy = skb_headlen(skb);
548 int size, offset1 = 0;
551 /* Skip over from offset */
552 while (count && (offset >= from->iov_len)) {
553 offset -= from->iov_len;
558 /* copy up to skb headlen */
559 while (count && (copy > 0)) {
560 size = min_t(unsigned int, copy, from->iov_len - offset);
561 if (copy_from_user(skb->data + offset1, from->iov_base + offset,
578 struct page *page[MAX_SKB_FRAGS];
581 unsigned long truesize;
583 len = from->iov_len - offset;
589 base = (unsigned long)from->iov_base + offset;
590 size = ((base & ~PAGE_MASK) + len + ~PAGE_MASK) >> PAGE_SHIFT;
591 if (i + size > MAX_SKB_FRAGS)
593 num_pages = get_user_pages_fast(base, size, 0, &page[i]);
594 if (num_pages != size) {
597 for (j = 0; j < num_pages; j++)
598 put_page(page[i + j]);
601 truesize = size * PAGE_SIZE;
602 skb->data_len += len;
604 skb->truesize += truesize;
605 atomic_add(truesize, &skb->sk->sk_wmem_alloc);
607 int off = base & ~PAGE_MASK;
608 int size = min_t(int, len, PAGE_SIZE - off);
609 __skb_fill_page_desc(skb, i, page[i], off, size);
610 skb_shinfo(skb)->nr_frags++;
611 /* increase sk_wmem_alloc */
623 * macvtap_skb_from_vnet_hdr and macvtap_skb_to_vnet_hdr should
624 * be shared with the tun/tap driver.
626 static int macvtap_skb_from_vnet_hdr(struct sk_buff *skb,
627 struct virtio_net_hdr *vnet_hdr)
629 unsigned short gso_type = 0;
630 if (vnet_hdr->gso_type != VIRTIO_NET_HDR_GSO_NONE) {
631 switch (vnet_hdr->gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
632 case VIRTIO_NET_HDR_GSO_TCPV4:
633 gso_type = SKB_GSO_TCPV4;
635 case VIRTIO_NET_HDR_GSO_TCPV6:
636 gso_type = SKB_GSO_TCPV6;
638 case VIRTIO_NET_HDR_GSO_UDP:
639 gso_type = SKB_GSO_UDP;
645 if (vnet_hdr->gso_type & VIRTIO_NET_HDR_GSO_ECN)
646 gso_type |= SKB_GSO_TCP_ECN;
648 if (vnet_hdr->gso_size == 0)
652 if (vnet_hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
653 if (!skb_partial_csum_set(skb, vnet_hdr->csum_start,
654 vnet_hdr->csum_offset))
658 if (vnet_hdr->gso_type != VIRTIO_NET_HDR_GSO_NONE) {
659 skb_shinfo(skb)->gso_size = vnet_hdr->gso_size;
660 skb_shinfo(skb)->gso_type = gso_type;
662 /* Header must be checked, and gso_segs computed. */
663 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
664 skb_shinfo(skb)->gso_segs = 0;
669 static int macvtap_skb_to_vnet_hdr(const struct sk_buff *skb,
670 struct virtio_net_hdr *vnet_hdr)
672 memset(vnet_hdr, 0, sizeof(*vnet_hdr));
674 if (skb_is_gso(skb)) {
675 struct skb_shared_info *sinfo = skb_shinfo(skb);
677 /* This is a hint as to how much should be linear. */
678 vnet_hdr->hdr_len = skb_headlen(skb);
679 vnet_hdr->gso_size = sinfo->gso_size;
680 if (sinfo->gso_type & SKB_GSO_TCPV4)
681 vnet_hdr->gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
682 else if (sinfo->gso_type & SKB_GSO_TCPV6)
683 vnet_hdr->gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
684 else if (sinfo->gso_type & SKB_GSO_UDP)
685 vnet_hdr->gso_type = VIRTIO_NET_HDR_GSO_UDP;
688 if (sinfo->gso_type & SKB_GSO_TCP_ECN)
689 vnet_hdr->gso_type |= VIRTIO_NET_HDR_GSO_ECN;
691 vnet_hdr->gso_type = VIRTIO_NET_HDR_GSO_NONE;
693 if (skb->ip_summed == CHECKSUM_PARTIAL) {
694 vnet_hdr->flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
695 vnet_hdr->csum_start = skb_checksum_start_offset(skb);
696 vnet_hdr->csum_offset = skb->csum_offset;
697 } else if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
698 vnet_hdr->flags = VIRTIO_NET_HDR_F_DATA_VALID;
699 } /* else everything is zero */
704 static unsigned long iov_pages(const struct iovec *iv, int offset,
705 unsigned long nr_segs)
707 unsigned long seg, base;
708 int pages = 0, len, size;
710 while (nr_segs && (offset >= iv->iov_len)) {
711 offset -= iv->iov_len;
716 for (seg = 0; seg < nr_segs; seg++) {
717 base = (unsigned long)iv[seg].iov_base + offset;
718 len = iv[seg].iov_len - offset;
719 size = ((base & ~PAGE_MASK) + len + ~PAGE_MASK) >> PAGE_SHIFT;
727 /* Get packet from user space buffer */
728 static ssize_t macvtap_get_user(struct macvtap_queue *q, struct msghdr *m,
729 const struct iovec *iv, unsigned long total_len,
730 size_t count, int noblock)
733 struct macvlan_dev *vlan;
734 unsigned long len = total_len;
736 struct virtio_net_hdr vnet_hdr = { 0 };
737 int vnet_hdr_len = 0;
739 bool zerocopy = false;
742 if (q->flags & IFF_VNET_HDR) {
743 vnet_hdr_len = q->vnet_hdr_sz;
746 if (len < vnet_hdr_len)
750 err = memcpy_fromiovecend((void *)&vnet_hdr, iv, 0,
754 if ((vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
755 vnet_hdr.csum_start + vnet_hdr.csum_offset + 2 >
757 vnet_hdr.hdr_len = vnet_hdr.csum_start +
758 vnet_hdr.csum_offset + 2;
760 if (vnet_hdr.hdr_len > len)
765 if (unlikely(len < ETH_HLEN))
769 if (unlikely(count > UIO_MAXIOV))
772 if (m && m->msg_control && sock_flag(&q->sk, SOCK_ZEROCOPY)) {
773 copylen = vnet_hdr.hdr_len ? vnet_hdr.hdr_len : GOODCOPY_LEN;
775 if (iov_pages(iv, vnet_hdr_len + copylen, count)
782 linear = vnet_hdr.hdr_len;
785 skb = macvtap_alloc_skb(&q->sk, NET_IP_ALIGN, copylen,
786 linear, noblock, &err);
791 err = zerocopy_sg_from_iovec(skb, iv, vnet_hdr_len, count);
793 err = skb_copy_datagram_from_iovec(skb, 0, iv, vnet_hdr_len,
795 if (!err && m && m->msg_control) {
796 struct ubuf_info *uarg = m->msg_control;
797 uarg->callback(uarg, false);
804 skb_set_network_header(skb, ETH_HLEN);
805 skb_reset_mac_header(skb);
806 skb->protocol = eth_hdr(skb)->h_proto;
809 err = macvtap_skb_from_vnet_hdr(skb, &vnet_hdr);
814 skb_probe_transport_header(skb, ETH_HLEN);
817 vlan = rcu_dereference(q->vlan);
818 /* copy skb_ubuf_info for callback when skb has no error */
820 skb_shinfo(skb)->destructor_arg = m->msg_control;
821 skb_shinfo(skb)->tx_flags |= SKBTX_DEV_ZEROCOPY;
822 skb_shinfo(skb)->tx_flags |= SKBTX_SHARED_FRAG;
826 macvlan_start_xmit(skb, vlan->dev);
840 vlan = rcu_dereference(q->vlan);
842 vlan->dev->stats.tx_dropped++;
848 static ssize_t macvtap_aio_write(struct kiocb *iocb, const struct iovec *iv,
849 unsigned long count, loff_t pos)
851 struct file *file = iocb->ki_filp;
852 ssize_t result = -ENOLINK;
853 struct macvtap_queue *q = file->private_data;
855 result = macvtap_get_user(q, NULL, iv, iov_length(iv, count), count,
856 file->f_flags & O_NONBLOCK);
860 /* Put packet to the user space buffer */
861 static ssize_t macvtap_put_user(struct macvtap_queue *q,
862 const struct sk_buff *skb,
863 const struct iovec *iv, int len)
865 struct macvlan_dev *vlan;
867 int vnet_hdr_len = 0;
871 if (q->flags & IFF_VNET_HDR) {
872 struct virtio_net_hdr vnet_hdr;
873 vnet_hdr_len = q->vnet_hdr_sz;
874 if ((len -= vnet_hdr_len) < 0)
877 ret = macvtap_skb_to_vnet_hdr(skb, &vnet_hdr);
881 if (memcpy_toiovecend(iv, (void *)&vnet_hdr, 0, sizeof(vnet_hdr)))
884 copied = vnet_hdr_len;
886 if (!vlan_tx_tag_present(skb))
887 len = min_t(int, skb->len, len);
894 veth.h_vlan_proto = skb->vlan_proto;
895 veth.h_vlan_TCI = htons(vlan_tx_tag_get(skb));
897 vlan_offset = offsetof(struct vlan_ethhdr, h_vlan_proto);
898 len = min_t(int, skb->len + VLAN_HLEN, len);
900 copy = min_t(int, vlan_offset, len);
901 ret = skb_copy_datagram_const_iovec(skb, 0, iv, copied, copy);
907 copy = min_t(int, sizeof(veth), len);
908 ret = memcpy_toiovecend(iv, (void *)&veth, copied, copy);
915 ret = skb_copy_datagram_const_iovec(skb, vlan_offset, iv, copied, len);
920 vlan = rcu_dereference(q->vlan);
923 macvlan_count_rx(vlan, copied - vnet_hdr_len, ret == 0, 0);
928 return ret ? ret : copied;
931 static ssize_t macvtap_do_read(struct macvtap_queue *q, struct kiocb *iocb,
932 const struct iovec *iv, unsigned long len,
941 prepare_to_wait(sk_sleep(&q->sk), &wait,
944 /* Read frames from the queue */
945 skb = skb_dequeue(&q->sk.sk_receive_queue);
951 if (signal_pending(current)) {
955 /* Nothing to read, let's sleep */
959 ret = macvtap_put_user(q, skb, iv, len);
965 finish_wait(sk_sleep(&q->sk), &wait);
969 static ssize_t macvtap_aio_read(struct kiocb *iocb, const struct iovec *iv,
970 unsigned long count, loff_t pos)
972 struct file *file = iocb->ki_filp;
973 struct macvtap_queue *q = file->private_data;
974 ssize_t len, ret = 0;
976 len = iov_length(iv, count);
982 ret = macvtap_do_read(q, iocb, iv, len, file->f_flags & O_NONBLOCK);
983 ret = min_t(ssize_t, ret, len); /* XXX copied from tun.c. Why? */
988 static struct macvlan_dev *macvtap_get_vlan(struct macvtap_queue *q)
990 struct macvlan_dev *vlan;
993 vlan = rtnl_dereference(q->vlan);
1000 static void macvtap_put_vlan(struct macvlan_dev *vlan)
1005 static int macvtap_ioctl_set_queue(struct file *file, unsigned int flags)
1007 struct macvtap_queue *q = file->private_data;
1008 struct macvlan_dev *vlan;
1011 vlan = macvtap_get_vlan(q);
1015 if (flags & IFF_ATTACH_QUEUE)
1016 ret = macvtap_enable_queue(vlan->dev, file, q);
1017 else if (flags & IFF_DETACH_QUEUE)
1018 ret = macvtap_disable_queue(q);
1022 macvtap_put_vlan(vlan);
1026 static int set_offload(struct macvtap_queue *q, unsigned long arg)
1028 struct macvlan_dev *vlan;
1029 netdev_features_t features;
1030 netdev_features_t feature_mask = 0;
1032 vlan = rtnl_dereference(q->vlan);
1036 features = vlan->dev->features;
1038 if (arg & TUN_F_CSUM) {
1039 feature_mask = NETIF_F_HW_CSUM;
1041 if (arg & (TUN_F_TSO4 | TUN_F_TSO6)) {
1042 if (arg & TUN_F_TSO_ECN)
1043 feature_mask |= NETIF_F_TSO_ECN;
1044 if (arg & TUN_F_TSO4)
1045 feature_mask |= NETIF_F_TSO;
1046 if (arg & TUN_F_TSO6)
1047 feature_mask |= NETIF_F_TSO6;
1050 if (arg & TUN_F_UFO)
1051 feature_mask |= NETIF_F_UFO;
1054 /* tun/tap driver inverts the usage for TSO offloads, where
1055 * setting the TSO bit means that the userspace wants to
1056 * accept TSO frames and turning it off means that user space
1057 * does not support TSO.
1058 * For macvtap, we have to invert it to mean the same thing.
1059 * When user space turns off TSO, we turn off GSO/LRO so that
1060 * user-space will not receive TSO frames.
1062 if (feature_mask & (NETIF_F_TSO | NETIF_F_TSO6 | NETIF_F_UFO))
1063 features |= RX_OFFLOADS;
1065 features &= ~RX_OFFLOADS;
1067 /* tap_features are the same as features on tun/tap and
1068 * reflect user expectations.
1070 vlan->tap_features = feature_mask;
1071 vlan->set_features = features;
1072 netdev_update_features(vlan->dev);
1078 * provide compatibility with generic tun/tap interface
1080 static long macvtap_ioctl(struct file *file, unsigned int cmd,
1083 struct macvtap_queue *q = file->private_data;
1084 struct macvlan_dev *vlan;
1085 void __user *argp = (void __user *)arg;
1086 struct ifreq __user *ifr = argp;
1087 unsigned int __user *up = argp;
1089 int __user *sp = argp;
1095 /* ignore the name, just look at flags */
1096 if (get_user(u, &ifr->ifr_flags))
1100 if ((u & ~(IFF_VNET_HDR | IFF_MULTI_QUEUE)) !=
1101 (IFF_NO_PI | IFF_TAP))
1110 vlan = macvtap_get_vlan(q);
1117 if (copy_to_user(&ifr->ifr_name, vlan->dev->name, IFNAMSIZ) ||
1118 put_user(q->flags, &ifr->ifr_flags))
1120 macvtap_put_vlan(vlan);
1125 if (get_user(u, &ifr->ifr_flags))
1128 ret = macvtap_ioctl_set_queue(file, u);
1132 case TUNGETFEATURES:
1133 if (put_user(IFF_TAP | IFF_NO_PI | IFF_VNET_HDR |
1134 IFF_MULTI_QUEUE, up))
1139 if (get_user(u, up))
1142 q->sk.sk_sndbuf = u;
1145 case TUNGETVNETHDRSZ:
1147 if (put_user(s, sp))
1151 case TUNSETVNETHDRSZ:
1152 if (get_user(s, sp))
1154 if (s < (int)sizeof(struct virtio_net_hdr))
1161 /* let the user check for future flags */
1162 if (arg & ~(TUN_F_CSUM | TUN_F_TSO4 | TUN_F_TSO6 |
1163 TUN_F_TSO_ECN | TUN_F_UFO))
1166 /* TODO: only accept frames with the features that
1167 got enabled for forwarded frames */
1168 if (!(q->flags & IFF_VNET_HDR))
1171 ret = set_offload(q, arg);
1180 #ifdef CONFIG_COMPAT
1181 static long macvtap_compat_ioctl(struct file *file, unsigned int cmd,
1184 return macvtap_ioctl(file, cmd, (unsigned long)compat_ptr(arg));
1188 static const struct file_operations macvtap_fops = {
1189 .owner = THIS_MODULE,
1190 .open = macvtap_open,
1191 .release = macvtap_release,
1192 .aio_read = macvtap_aio_read,
1193 .aio_write = macvtap_aio_write,
1194 .poll = macvtap_poll,
1195 .llseek = no_llseek,
1196 .unlocked_ioctl = macvtap_ioctl,
1197 #ifdef CONFIG_COMPAT
1198 .compat_ioctl = macvtap_compat_ioctl,
1202 static int macvtap_sendmsg(struct kiocb *iocb, struct socket *sock,
1203 struct msghdr *m, size_t total_len)
1205 struct macvtap_queue *q = container_of(sock, struct macvtap_queue, sock);
1206 return macvtap_get_user(q, m, m->msg_iov, total_len, m->msg_iovlen,
1207 m->msg_flags & MSG_DONTWAIT);
1210 static int macvtap_recvmsg(struct kiocb *iocb, struct socket *sock,
1211 struct msghdr *m, size_t total_len,
1214 struct macvtap_queue *q = container_of(sock, struct macvtap_queue, sock);
1216 if (flags & ~(MSG_DONTWAIT|MSG_TRUNC))
1218 ret = macvtap_do_read(q, iocb, m->msg_iov, total_len,
1219 flags & MSG_DONTWAIT);
1220 if (ret > total_len) {
1221 m->msg_flags |= MSG_TRUNC;
1222 ret = flags & MSG_TRUNC ? ret : total_len;
1227 /* Ops structure to mimic raw sockets with tun */
1228 static const struct proto_ops macvtap_socket_ops = {
1229 .sendmsg = macvtap_sendmsg,
1230 .recvmsg = macvtap_recvmsg,
1233 /* Get an underlying socket object from tun file. Returns error unless file is
1234 * attached to a device. The returned object works like a packet socket, it
1235 * can be used for sock_sendmsg/sock_recvmsg. The caller is responsible for
1236 * holding a reference to the file for as long as the socket is in use. */
1237 struct socket *macvtap_get_socket(struct file *file)
1239 struct macvtap_queue *q;
1240 if (file->f_op != &macvtap_fops)
1241 return ERR_PTR(-EINVAL);
1242 q = file->private_data;
1244 return ERR_PTR(-EBADFD);
1247 EXPORT_SYMBOL_GPL(macvtap_get_socket);
1249 static int macvtap_device_event(struct notifier_block *unused,
1250 unsigned long event, void *ptr)
1252 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1253 struct macvlan_dev *vlan;
1254 struct device *classdev;
1258 if (dev->rtnl_link_ops != &macvtap_link_ops)
1261 vlan = netdev_priv(dev);
1264 case NETDEV_REGISTER:
1265 /* Create the device node here after the network device has
1266 * been registered but before register_netdevice has
1269 err = macvtap_get_minor(vlan);
1271 return notifier_from_errno(err);
1273 devt = MKDEV(MAJOR(macvtap_major), vlan->minor);
1274 classdev = device_create(macvtap_class, &dev->dev, devt,
1275 dev, "tap%d", dev->ifindex);
1276 if (IS_ERR(classdev)) {
1277 macvtap_free_minor(vlan);
1278 return notifier_from_errno(PTR_ERR(classdev));
1281 case NETDEV_UNREGISTER:
1282 devt = MKDEV(MAJOR(macvtap_major), vlan->minor);
1283 device_destroy(macvtap_class, devt);
1284 macvtap_free_minor(vlan);
1291 static struct notifier_block macvtap_notifier_block __read_mostly = {
1292 .notifier_call = macvtap_device_event,
1295 static int macvtap_init(void)
1299 err = alloc_chrdev_region(&macvtap_major, 0,
1300 MACVTAP_NUM_DEVS, "macvtap");
1304 cdev_init(&macvtap_cdev, &macvtap_fops);
1305 err = cdev_add(&macvtap_cdev, macvtap_major, MACVTAP_NUM_DEVS);
1309 macvtap_class = class_create(THIS_MODULE, "macvtap");
1310 if (IS_ERR(macvtap_class)) {
1311 err = PTR_ERR(macvtap_class);
1315 err = register_netdevice_notifier(&macvtap_notifier_block);
1319 err = macvlan_link_register(&macvtap_link_ops);
1326 unregister_netdevice_notifier(&macvtap_notifier_block);
1328 class_unregister(macvtap_class);
1330 cdev_del(&macvtap_cdev);
1332 unregister_chrdev_region(macvtap_major, MACVTAP_NUM_DEVS);
1336 module_init(macvtap_init);
1338 static void macvtap_exit(void)
1340 rtnl_link_unregister(&macvtap_link_ops);
1341 unregister_netdevice_notifier(&macvtap_notifier_block);
1342 class_unregister(macvtap_class);
1343 cdev_del(&macvtap_cdev);
1344 unregister_chrdev_region(macvtap_major, MACVTAP_NUM_DEVS);
1346 module_exit(macvtap_exit);
1348 MODULE_ALIAS_RTNL_LINK("macvtap");
1349 MODULE_AUTHOR("Arnd Bergmann <arnd@arndb.de>");
1350 MODULE_LICENSE("GPL");