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/wait.h>
15 #include <linux/cdev.h>
16 #include <linux/idr.h>
18 #include <linux/uio.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 #define MACVTAP_FEATURES (IFF_VNET_HDR | IFF_MULTI_QUEUE)
50 #define MACVTAP_VNET_LE 0x80000000
51 #define MACVTAP_VNET_BE 0x40000000
53 #ifdef CONFIG_TUN_VNET_CROSS_LE
54 static inline bool macvtap_legacy_is_little_endian(struct macvtap_queue *q)
56 return q->flags & MACVTAP_VNET_BE ? false :
57 virtio_legacy_is_little_endian();
60 static long macvtap_get_vnet_be(struct macvtap_queue *q, int __user *sp)
62 int s = !!(q->flags & MACVTAP_VNET_BE);
70 static long macvtap_set_vnet_be(struct macvtap_queue *q, int __user *sp)
78 q->flags |= MACVTAP_VNET_BE;
80 q->flags &= ~MACVTAP_VNET_BE;
85 static inline bool macvtap_legacy_is_little_endian(struct macvtap_queue *q)
87 return virtio_legacy_is_little_endian();
90 static long macvtap_get_vnet_be(struct macvtap_queue *q, int __user *argp)
95 static long macvtap_set_vnet_be(struct macvtap_queue *q, int __user *argp)
99 #endif /* CONFIG_TUN_VNET_CROSS_LE */
101 static inline bool macvtap_is_little_endian(struct macvtap_queue *q)
103 return q->flags & MACVTAP_VNET_LE ||
104 macvtap_legacy_is_little_endian(q);
107 static inline u16 macvtap16_to_cpu(struct macvtap_queue *q, __virtio16 val)
109 return __virtio16_to_cpu(macvtap_is_little_endian(q), val);
112 static inline __virtio16 cpu_to_macvtap16(struct macvtap_queue *q, u16 val)
114 return __cpu_to_virtio16(macvtap_is_little_endian(q), val);
117 static struct proto macvtap_proto = {
119 .owner = THIS_MODULE,
120 .obj_size = sizeof (struct macvtap_queue),
124 * Variables for dealing with macvtaps device numbers.
126 static dev_t macvtap_major;
127 #define MACVTAP_NUM_DEVS (1U << MINORBITS)
128 static DEFINE_MUTEX(minor_lock);
129 static DEFINE_IDR(minor_idr);
131 #define GOODCOPY_LEN 128
132 static const void *macvtap_net_namespace(struct device *d)
134 struct net_device *dev = to_net_dev(d->parent);
138 static struct class macvtap_class = {
140 .owner = THIS_MODULE,
141 .ns_type = &net_ns_type_operations,
142 .namespace = macvtap_net_namespace,
144 static struct cdev macvtap_cdev;
146 static const struct proto_ops macvtap_socket_ops;
148 #define TUN_OFFLOADS (NETIF_F_HW_CSUM | NETIF_F_TSO_ECN | NETIF_F_TSO | \
149 NETIF_F_TSO6 | NETIF_F_UFO)
150 #define RX_OFFLOADS (NETIF_F_GRO | NETIF_F_LRO)
151 #define TAP_FEATURES (NETIF_F_GSO | NETIF_F_SG | NETIF_F_FRAGLIST)
153 static struct macvlan_dev *macvtap_get_vlan_rcu(const struct net_device *dev)
155 return rcu_dereference(dev->rx_handler_data);
160 * The macvtap_queue and the macvlan_dev are loosely coupled, the
161 * pointers from one to the other can only be read while rcu_read_lock
164 * Both the file and the macvlan_dev hold a reference on the macvtap_queue
165 * through sock_hold(&q->sk). When the macvlan_dev goes away first,
166 * q->vlan becomes inaccessible. When the files gets closed,
167 * macvtap_get_queue() fails.
169 * There may still be references to the struct sock inside of the
170 * queue from outbound SKBs, but these never reference back to the
171 * file or the dev. The data structure is freed through __sk_free
172 * when both our references and any pending SKBs are gone.
175 static int macvtap_enable_queue(struct net_device *dev, struct file *file,
176 struct macvtap_queue *q)
178 struct macvlan_dev *vlan = netdev_priv(dev);
187 rcu_assign_pointer(vlan->taps[vlan->numvtaps], q);
188 q->queue_index = vlan->numvtaps;
197 static int macvtap_set_queue(struct net_device *dev, struct file *file,
198 struct macvtap_queue *q)
200 struct macvlan_dev *vlan = netdev_priv(dev);
202 if (vlan->numqueues == MAX_MACVTAP_QUEUES)
205 rcu_assign_pointer(q->vlan, vlan);
206 rcu_assign_pointer(vlan->taps[vlan->numvtaps], q);
210 q->queue_index = vlan->numvtaps;
212 file->private_data = q;
213 list_add_tail(&q->next, &vlan->queue_list);
221 static int macvtap_disable_queue(struct macvtap_queue *q)
223 struct macvlan_dev *vlan;
224 struct macvtap_queue *nq;
230 vlan = rtnl_dereference(q->vlan);
233 int index = q->queue_index;
234 BUG_ON(index >= vlan->numvtaps);
235 nq = rtnl_dereference(vlan->taps[vlan->numvtaps - 1]);
236 nq->queue_index = index;
238 rcu_assign_pointer(vlan->taps[index], nq);
239 RCU_INIT_POINTER(vlan->taps[vlan->numvtaps - 1], NULL);
249 * The file owning the queue got closed, give up both
250 * the reference that the files holds as well as the
251 * one from the macvlan_dev if that still exists.
253 * Using the spinlock makes sure that we don't get
254 * to the queue again after destroying it.
256 static void macvtap_put_queue(struct macvtap_queue *q)
258 struct macvlan_dev *vlan;
261 vlan = rtnl_dereference(q->vlan);
265 BUG_ON(macvtap_disable_queue(q));
268 RCU_INIT_POINTER(q->vlan, NULL);
270 list_del_init(&q->next);
280 * Select a queue based on the rxq of the device on which this packet
281 * arrived. If the incoming device is not mq, calculate a flow hash
282 * to select a queue. If all fails, find the first available queue.
283 * Cache vlan->numvtaps since it can become zero during the execution
286 static struct macvtap_queue *macvtap_get_queue(struct net_device *dev,
289 struct macvlan_dev *vlan = netdev_priv(dev);
290 struct macvtap_queue *tap = NULL;
291 /* Access to taps array is protected by rcu, but access to numvtaps
292 * isn't. Below we use it to lookup a queue, but treat it as a hint
293 * and validate that the result isn't NULL - in case we are
294 * racing against queue removal.
296 int numvtaps = ACCESS_ONCE(vlan->numvtaps);
302 /* Check if we can use flow to select a queue */
303 rxq = skb_get_hash(skb);
305 tap = rcu_dereference(vlan->taps[rxq % numvtaps]);
309 if (likely(skb_rx_queue_recorded(skb))) {
310 rxq = skb_get_rx_queue(skb);
312 while (unlikely(rxq >= numvtaps))
315 tap = rcu_dereference(vlan->taps[rxq]);
319 tap = rcu_dereference(vlan->taps[0]);
325 * The net_device is going away, give up the reference
326 * that it holds on all queues and safely set the pointer
327 * from the queues to NULL.
329 static void macvtap_del_queues(struct net_device *dev)
331 struct macvlan_dev *vlan = netdev_priv(dev);
332 struct macvtap_queue *q, *tmp;
335 list_for_each_entry_safe(q, tmp, &vlan->queue_list, next) {
336 list_del_init(&q->next);
337 RCU_INIT_POINTER(q->vlan, NULL);
343 BUG_ON(vlan->numvtaps);
344 BUG_ON(vlan->numqueues);
345 /* guarantee that any future macvtap_set_queue will fail */
346 vlan->numvtaps = MAX_MACVTAP_QUEUES;
349 static rx_handler_result_t macvtap_handle_frame(struct sk_buff **pskb)
351 struct sk_buff *skb = *pskb;
352 struct net_device *dev = skb->dev;
353 struct macvlan_dev *vlan;
354 struct macvtap_queue *q;
355 netdev_features_t features = TAP_FEATURES;
357 vlan = macvtap_get_vlan_rcu(dev);
359 return RX_HANDLER_PASS;
361 q = macvtap_get_queue(dev, skb);
363 return RX_HANDLER_PASS;
365 if (skb_queue_len(&q->sk.sk_receive_queue) >= dev->tx_queue_len)
368 skb_push(skb, ETH_HLEN);
370 /* Apply the forward feature mask so that we perform segmentation
371 * according to users wishes. This only works if VNET_HDR is
374 if (q->flags & IFF_VNET_HDR)
375 features |= vlan->tap_features;
376 if (netif_needs_gso(skb, features)) {
377 struct sk_buff *segs = __skb_gso_segment(skb, features, false);
383 skb_queue_tail(&q->sk.sk_receive_queue, skb);
389 struct sk_buff *nskb = segs->next;
392 skb_queue_tail(&q->sk.sk_receive_queue, segs);
396 /* If we receive a partial checksum and the tap side
397 * doesn't support checksum offload, compute the checksum.
398 * Note: it doesn't matter which checksum feature to
399 * check, we either support them all or none.
401 if (skb->ip_summed == CHECKSUM_PARTIAL &&
402 !(features & NETIF_F_CSUM_MASK) &&
403 skb_checksum_help(skb))
405 skb_queue_tail(&q->sk.sk_receive_queue, skb);
409 wake_up_interruptible_poll(sk_sleep(&q->sk), POLLIN | POLLRDNORM | POLLRDBAND);
410 return RX_HANDLER_CONSUMED;
413 /* Count errors/drops only here, thus don't care about args. */
414 macvlan_count_rx(vlan, 0, 0, 0);
416 return RX_HANDLER_CONSUMED;
419 static int macvtap_get_minor(struct macvlan_dev *vlan)
421 int retval = -ENOMEM;
423 mutex_lock(&minor_lock);
424 retval = idr_alloc(&minor_idr, vlan, 1, MACVTAP_NUM_DEVS, GFP_KERNEL);
426 vlan->minor = retval;
427 } else if (retval == -ENOSPC) {
428 printk(KERN_ERR "too many macvtap devices\n");
431 mutex_unlock(&minor_lock);
432 return retval < 0 ? retval : 0;
435 static void macvtap_free_minor(struct macvlan_dev *vlan)
437 mutex_lock(&minor_lock);
439 idr_remove(&minor_idr, vlan->minor);
442 mutex_unlock(&minor_lock);
445 static struct net_device *dev_get_by_macvtap_minor(int minor)
447 struct net_device *dev = NULL;
448 struct macvlan_dev *vlan;
450 mutex_lock(&minor_lock);
451 vlan = idr_find(&minor_idr, minor);
456 mutex_unlock(&minor_lock);
460 static int macvtap_newlink(struct net *src_net,
461 struct net_device *dev,
463 struct nlattr *data[])
465 struct macvlan_dev *vlan = netdev_priv(dev);
468 INIT_LIST_HEAD(&vlan->queue_list);
470 /* Since macvlan supports all offloads by default, make
471 * tap support all offloads also.
473 vlan->tap_features = TUN_OFFLOADS;
475 err = netdev_rx_handler_register(dev, macvtap_handle_frame, vlan);
479 /* Don't put anything that may fail after macvlan_common_newlink
480 * because we can't undo what it does.
482 return macvlan_common_newlink(src_net, dev, tb, data);
485 static void macvtap_dellink(struct net_device *dev,
486 struct list_head *head)
488 netdev_rx_handler_unregister(dev);
489 macvtap_del_queues(dev);
490 macvlan_dellink(dev, head);
493 static void macvtap_setup(struct net_device *dev)
495 macvlan_common_setup(dev);
496 dev->tx_queue_len = TUN_READQ_SIZE;
499 static struct rtnl_link_ops macvtap_link_ops __read_mostly = {
501 .setup = macvtap_setup,
502 .newlink = macvtap_newlink,
503 .dellink = macvtap_dellink,
507 static void macvtap_sock_write_space(struct sock *sk)
509 wait_queue_head_t *wqueue;
511 if (!sock_writeable(sk) ||
512 !test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags))
515 wqueue = sk_sleep(sk);
516 if (wqueue && waitqueue_active(wqueue))
517 wake_up_interruptible_poll(wqueue, POLLOUT | POLLWRNORM | POLLWRBAND);
520 static void macvtap_sock_destruct(struct sock *sk)
522 skb_queue_purge(&sk->sk_receive_queue);
525 static int macvtap_open(struct inode *inode, struct file *file)
527 struct net *net = current->nsproxy->net_ns;
528 struct net_device *dev;
529 struct macvtap_queue *q;
533 dev = dev_get_by_macvtap_minor(iminor(inode));
538 q = (struct macvtap_queue *)sk_alloc(net, AF_UNSPEC, GFP_KERNEL,
543 RCU_INIT_POINTER(q->sock.wq, &q->wq);
544 init_waitqueue_head(&q->wq.wait);
545 q->sock.type = SOCK_RAW;
546 q->sock.state = SS_CONNECTED;
548 q->sock.ops = &macvtap_socket_ops;
549 sock_init_data(&q->sock, &q->sk);
550 q->sk.sk_write_space = macvtap_sock_write_space;
551 q->sk.sk_destruct = macvtap_sock_destruct;
552 q->flags = IFF_VNET_HDR | IFF_NO_PI | IFF_TAP;
553 q->vnet_hdr_sz = sizeof(struct virtio_net_hdr);
556 * so far only KVM virtio_net uses macvtap, enable zero copy between
557 * guest kernel and host kernel when lower device supports zerocopy
559 * The macvlan supports zerocopy iff the lower device supports zero
560 * copy so we don't have to look at the lower device directly.
562 if ((dev->features & NETIF_F_HIGHDMA) && (dev->features & NETIF_F_SG))
563 sock_set_flag(&q->sk, SOCK_ZEROCOPY);
565 err = macvtap_set_queue(dev, file, q);
577 static int macvtap_release(struct inode *inode, struct file *file)
579 struct macvtap_queue *q = file->private_data;
580 macvtap_put_queue(q);
584 static unsigned int macvtap_poll(struct file *file, poll_table * wait)
586 struct macvtap_queue *q = file->private_data;
587 unsigned int mask = POLLERR;
593 poll_wait(file, &q->wq.wait, wait);
595 if (!skb_queue_empty(&q->sk.sk_receive_queue))
596 mask |= POLLIN | POLLRDNORM;
598 if (sock_writeable(&q->sk) ||
599 (!test_and_set_bit(SOCKWQ_ASYNC_NOSPACE, &q->sock.flags) &&
600 sock_writeable(&q->sk)))
601 mask |= POLLOUT | POLLWRNORM;
607 static inline struct sk_buff *macvtap_alloc_skb(struct sock *sk, size_t prepad,
608 size_t len, size_t linear,
609 int noblock, int *err)
613 /* Under a page? Don't bother with paged skb. */
614 if (prepad + len < PAGE_SIZE || !linear)
617 skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
622 skb_reserve(skb, prepad);
623 skb_put(skb, linear);
624 skb->data_len = len - linear;
625 skb->len += len - linear;
631 * macvtap_skb_from_vnet_hdr and macvtap_skb_to_vnet_hdr should
632 * be shared with the tun/tap driver.
634 static int macvtap_skb_from_vnet_hdr(struct macvtap_queue *q,
636 struct virtio_net_hdr *vnet_hdr)
638 unsigned short gso_type = 0;
639 if (vnet_hdr->gso_type != VIRTIO_NET_HDR_GSO_NONE) {
640 switch (vnet_hdr->gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
641 case VIRTIO_NET_HDR_GSO_TCPV4:
642 gso_type = SKB_GSO_TCPV4;
644 case VIRTIO_NET_HDR_GSO_TCPV6:
645 gso_type = SKB_GSO_TCPV6;
647 case VIRTIO_NET_HDR_GSO_UDP:
648 gso_type = SKB_GSO_UDP;
654 if (vnet_hdr->gso_type & VIRTIO_NET_HDR_GSO_ECN)
655 gso_type |= SKB_GSO_TCP_ECN;
657 if (vnet_hdr->gso_size == 0)
661 if (vnet_hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
662 if (!skb_partial_csum_set(skb, macvtap16_to_cpu(q, vnet_hdr->csum_start),
663 macvtap16_to_cpu(q, vnet_hdr->csum_offset)))
667 if (vnet_hdr->gso_type != VIRTIO_NET_HDR_GSO_NONE) {
668 skb_shinfo(skb)->gso_size = macvtap16_to_cpu(q, vnet_hdr->gso_size);
669 skb_shinfo(skb)->gso_type = gso_type;
671 /* Header must be checked, and gso_segs computed. */
672 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
673 skb_shinfo(skb)->gso_segs = 0;
678 static void macvtap_skb_to_vnet_hdr(struct macvtap_queue *q,
679 const struct sk_buff *skb,
680 struct virtio_net_hdr *vnet_hdr)
682 memset(vnet_hdr, 0, sizeof(*vnet_hdr));
684 if (skb_is_gso(skb)) {
685 struct skb_shared_info *sinfo = skb_shinfo(skb);
687 /* This is a hint as to how much should be linear. */
688 vnet_hdr->hdr_len = cpu_to_macvtap16(q, skb_headlen(skb));
689 vnet_hdr->gso_size = cpu_to_macvtap16(q, sinfo->gso_size);
690 if (sinfo->gso_type & SKB_GSO_TCPV4)
691 vnet_hdr->gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
692 else if (sinfo->gso_type & SKB_GSO_TCPV6)
693 vnet_hdr->gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
694 else if (sinfo->gso_type & SKB_GSO_UDP)
695 vnet_hdr->gso_type = VIRTIO_NET_HDR_GSO_UDP;
698 if (sinfo->gso_type & SKB_GSO_TCP_ECN)
699 vnet_hdr->gso_type |= VIRTIO_NET_HDR_GSO_ECN;
701 vnet_hdr->gso_type = VIRTIO_NET_HDR_GSO_NONE;
703 if (skb->ip_summed == CHECKSUM_PARTIAL) {
704 vnet_hdr->flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
705 if (skb_vlan_tag_present(skb))
706 vnet_hdr->csum_start = cpu_to_macvtap16(q,
707 skb_checksum_start_offset(skb) + VLAN_HLEN);
709 vnet_hdr->csum_start = cpu_to_macvtap16(q,
710 skb_checksum_start_offset(skb));
711 vnet_hdr->csum_offset = cpu_to_macvtap16(q, skb->csum_offset);
712 } else if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
713 vnet_hdr->flags = VIRTIO_NET_HDR_F_DATA_VALID;
714 } /* else everything is zero */
717 /* Neighbour code has some assumptions on HH_DATA_MOD alignment */
718 #define MACVTAP_RESERVE HH_DATA_OFF(ETH_HLEN)
720 /* Get packet from user space buffer */
721 static ssize_t macvtap_get_user(struct macvtap_queue *q, struct msghdr *m,
722 struct iov_iter *from, int noblock)
724 int good_linear = SKB_MAX_HEAD(MACVTAP_RESERVE);
726 struct macvlan_dev *vlan;
727 unsigned long total_len = iov_iter_count(from);
728 unsigned long len = total_len;
730 struct virtio_net_hdr vnet_hdr = { 0 };
731 int vnet_hdr_len = 0;
734 bool zerocopy = false;
738 if (q->flags & IFF_VNET_HDR) {
739 vnet_hdr_len = q->vnet_hdr_sz;
742 if (len < vnet_hdr_len)
747 n = copy_from_iter(&vnet_hdr, sizeof(vnet_hdr), from);
748 if (n != sizeof(vnet_hdr))
750 iov_iter_advance(from, vnet_hdr_len - sizeof(vnet_hdr));
751 if ((vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
752 macvtap16_to_cpu(q, vnet_hdr.csum_start) +
753 macvtap16_to_cpu(q, vnet_hdr.csum_offset) + 2 >
754 macvtap16_to_cpu(q, vnet_hdr.hdr_len))
755 vnet_hdr.hdr_len = cpu_to_macvtap16(q,
756 macvtap16_to_cpu(q, vnet_hdr.csum_start) +
757 macvtap16_to_cpu(q, vnet_hdr.csum_offset) + 2);
759 if (macvtap16_to_cpu(q, vnet_hdr.hdr_len) > len)
764 if (unlikely(len < ETH_HLEN))
767 if (m && m->msg_control && sock_flag(&q->sk, SOCK_ZEROCOPY)) {
770 copylen = vnet_hdr.hdr_len ?
771 macvtap16_to_cpu(q, vnet_hdr.hdr_len) : GOODCOPY_LEN;
772 if (copylen > good_linear)
773 copylen = good_linear;
774 else if (copylen < ETH_HLEN)
778 iov_iter_advance(&i, copylen);
779 if (iov_iter_npages(&i, INT_MAX) <= MAX_SKB_FRAGS)
785 linear = macvtap16_to_cpu(q, vnet_hdr.hdr_len);
786 if (linear > good_linear)
787 linear = good_linear;
788 else if (linear < ETH_HLEN)
792 skb = macvtap_alloc_skb(&q->sk, MACVTAP_RESERVE, copylen,
793 linear, noblock, &err);
798 err = zerocopy_sg_from_iter(skb, from);
800 err = skb_copy_datagram_from_iter(skb, 0, from, len);
801 if (!err && m && m->msg_control) {
802 struct ubuf_info *uarg = m->msg_control;
803 uarg->callback(uarg, false);
810 skb_set_network_header(skb, ETH_HLEN);
811 skb_reset_mac_header(skb);
812 skb->protocol = eth_hdr(skb)->h_proto;
815 err = macvtap_skb_from_vnet_hdr(q, skb, &vnet_hdr);
820 skb_probe_transport_header(skb, ETH_HLEN);
822 /* Move network header to the right position for VLAN tagged packets */
823 if ((skb->protocol == htons(ETH_P_8021Q) ||
824 skb->protocol == htons(ETH_P_8021AD)) &&
825 __vlan_get_protocol(skb, skb->protocol, &depth) != 0)
826 skb_set_network_header(skb, depth);
829 vlan = rcu_dereference(q->vlan);
830 /* copy skb_ubuf_info for callback when skb has no error */
832 skb_shinfo(skb)->destructor_arg = m->msg_control;
833 skb_shinfo(skb)->tx_flags |= SKBTX_DEV_ZEROCOPY;
834 skb_shinfo(skb)->tx_flags |= SKBTX_SHARED_FRAG;
837 skb->dev = vlan->dev;
851 vlan = rcu_dereference(q->vlan);
853 this_cpu_inc(vlan->pcpu_stats->tx_dropped);
859 static ssize_t macvtap_write_iter(struct kiocb *iocb, struct iov_iter *from)
861 struct file *file = iocb->ki_filp;
862 struct macvtap_queue *q = file->private_data;
864 return macvtap_get_user(q, NULL, from, file->f_flags & O_NONBLOCK);
867 /* Put packet to the user space buffer */
868 static ssize_t macvtap_put_user(struct macvtap_queue *q,
869 const struct sk_buff *skb,
870 struct iov_iter *iter)
873 int vnet_hdr_len = 0;
877 if (q->flags & IFF_VNET_HDR) {
878 struct virtio_net_hdr vnet_hdr;
879 vnet_hdr_len = q->vnet_hdr_sz;
880 if (iov_iter_count(iter) < vnet_hdr_len)
883 macvtap_skb_to_vnet_hdr(q, skb, &vnet_hdr);
885 if (copy_to_iter(&vnet_hdr, sizeof(vnet_hdr), iter) !=
889 iov_iter_advance(iter, vnet_hdr_len - sizeof(vnet_hdr));
891 total = vnet_hdr_len;
894 if (skb_vlan_tag_present(skb)) {
899 veth.h_vlan_proto = skb->vlan_proto;
900 veth.h_vlan_TCI = htons(skb_vlan_tag_get(skb));
902 vlan_offset = offsetof(struct vlan_ethhdr, h_vlan_proto);
905 ret = skb_copy_datagram_iter(skb, 0, iter, vlan_offset);
906 if (ret || !iov_iter_count(iter))
909 ret = copy_to_iter(&veth, sizeof(veth), iter);
910 if (ret != sizeof(veth) || !iov_iter_count(iter))
914 ret = skb_copy_datagram_iter(skb, vlan_offset, iter,
915 skb->len - vlan_offset);
918 return ret ? ret : total;
921 static ssize_t macvtap_do_read(struct macvtap_queue *q,
929 if (!iov_iter_count(to))
934 prepare_to_wait(sk_sleep(&q->sk), &wait,
937 /* Read frames from the queue */
938 skb = skb_dequeue(&q->sk.sk_receive_queue);
945 if (signal_pending(current)) {
949 /* Nothing to read, let's sleep */
953 finish_wait(sk_sleep(&q->sk), &wait);
956 ret = macvtap_put_user(q, skb, to);
957 if (unlikely(ret < 0))
965 static ssize_t macvtap_read_iter(struct kiocb *iocb, struct iov_iter *to)
967 struct file *file = iocb->ki_filp;
968 struct macvtap_queue *q = file->private_data;
969 ssize_t len = iov_iter_count(to), ret;
971 ret = macvtap_do_read(q, to, file->f_flags & O_NONBLOCK);
972 ret = min_t(ssize_t, ret, len);
978 static struct macvlan_dev *macvtap_get_vlan(struct macvtap_queue *q)
980 struct macvlan_dev *vlan;
983 vlan = rtnl_dereference(q->vlan);
990 static void macvtap_put_vlan(struct macvlan_dev *vlan)
995 static int macvtap_ioctl_set_queue(struct file *file, unsigned int flags)
997 struct macvtap_queue *q = file->private_data;
998 struct macvlan_dev *vlan;
1001 vlan = macvtap_get_vlan(q);
1005 if (flags & IFF_ATTACH_QUEUE)
1006 ret = macvtap_enable_queue(vlan->dev, file, q);
1007 else if (flags & IFF_DETACH_QUEUE)
1008 ret = macvtap_disable_queue(q);
1012 macvtap_put_vlan(vlan);
1016 static int set_offload(struct macvtap_queue *q, unsigned long arg)
1018 struct macvlan_dev *vlan;
1019 netdev_features_t features;
1020 netdev_features_t feature_mask = 0;
1022 vlan = rtnl_dereference(q->vlan);
1026 features = vlan->dev->features;
1028 if (arg & TUN_F_CSUM) {
1029 feature_mask = NETIF_F_HW_CSUM;
1031 if (arg & (TUN_F_TSO4 | TUN_F_TSO6)) {
1032 if (arg & TUN_F_TSO_ECN)
1033 feature_mask |= NETIF_F_TSO_ECN;
1034 if (arg & TUN_F_TSO4)
1035 feature_mask |= NETIF_F_TSO;
1036 if (arg & TUN_F_TSO6)
1037 feature_mask |= NETIF_F_TSO6;
1040 if (arg & TUN_F_UFO)
1041 feature_mask |= NETIF_F_UFO;
1044 /* tun/tap driver inverts the usage for TSO offloads, where
1045 * setting the TSO bit means that the userspace wants to
1046 * accept TSO frames and turning it off means that user space
1047 * does not support TSO.
1048 * For macvtap, we have to invert it to mean the same thing.
1049 * When user space turns off TSO, we turn off GSO/LRO so that
1050 * user-space will not receive TSO frames.
1052 if (feature_mask & (NETIF_F_TSO | NETIF_F_TSO6 | NETIF_F_UFO))
1053 features |= RX_OFFLOADS;
1055 features &= ~RX_OFFLOADS;
1057 /* tap_features are the same as features on tun/tap and
1058 * reflect user expectations.
1060 vlan->tap_features = feature_mask;
1061 vlan->set_features = features;
1062 netdev_update_features(vlan->dev);
1068 * provide compatibility with generic tun/tap interface
1070 static long macvtap_ioctl(struct file *file, unsigned int cmd,
1073 struct macvtap_queue *q = file->private_data;
1074 struct macvlan_dev *vlan;
1075 void __user *argp = (void __user *)arg;
1076 struct ifreq __user *ifr = argp;
1077 unsigned int __user *up = argp;
1079 int __user *sp = argp;
1086 /* ignore the name, just look at flags */
1087 if (get_user(u, &ifr->ifr_flags))
1091 if ((u & ~MACVTAP_FEATURES) != (IFF_NO_PI | IFF_TAP))
1094 q->flags = (q->flags & ~MACVTAP_FEATURES) | u;
1100 vlan = macvtap_get_vlan(q);
1108 if (copy_to_user(&ifr->ifr_name, vlan->dev->name, IFNAMSIZ) ||
1109 put_user(u, &ifr->ifr_flags))
1111 macvtap_put_vlan(vlan);
1116 if (get_user(u, &ifr->ifr_flags))
1119 ret = macvtap_ioctl_set_queue(file, u);
1123 case TUNGETFEATURES:
1124 if (put_user(IFF_TAP | IFF_NO_PI | MACVTAP_FEATURES, up))
1129 if (get_user(s, sp))
1132 q->sk.sk_sndbuf = s;
1135 case TUNGETVNETHDRSZ:
1137 if (put_user(s, sp))
1141 case TUNSETVNETHDRSZ:
1142 if (get_user(s, sp))
1144 if (s < (int)sizeof(struct virtio_net_hdr))
1151 s = !!(q->flags & MACVTAP_VNET_LE);
1152 if (put_user(s, sp))
1157 if (get_user(s, sp))
1160 q->flags |= MACVTAP_VNET_LE;
1162 q->flags &= ~MACVTAP_VNET_LE;
1166 return macvtap_get_vnet_be(q, sp);
1169 return macvtap_set_vnet_be(q, sp);
1172 /* let the user check for future flags */
1173 if (arg & ~(TUN_F_CSUM | TUN_F_TSO4 | TUN_F_TSO6 |
1174 TUN_F_TSO_ECN | TUN_F_UFO))
1178 ret = set_offload(q, arg);
1184 vlan = macvtap_get_vlan(q);
1190 u = vlan->dev->type;
1191 if (copy_to_user(&ifr->ifr_name, vlan->dev->name, IFNAMSIZ) ||
1192 copy_to_user(&ifr->ifr_hwaddr.sa_data, vlan->dev->dev_addr, ETH_ALEN) ||
1193 put_user(u, &ifr->ifr_hwaddr.sa_family))
1195 macvtap_put_vlan(vlan);
1200 if (copy_from_user(&sa, &ifr->ifr_hwaddr, sizeof(sa)))
1203 vlan = macvtap_get_vlan(q);
1208 ret = dev_set_mac_address(vlan->dev, &sa);
1209 macvtap_put_vlan(vlan);
1218 #ifdef CONFIG_COMPAT
1219 static long macvtap_compat_ioctl(struct file *file, unsigned int cmd,
1222 return macvtap_ioctl(file, cmd, (unsigned long)compat_ptr(arg));
1226 static const struct file_operations macvtap_fops = {
1227 .owner = THIS_MODULE,
1228 .open = macvtap_open,
1229 .release = macvtap_release,
1230 .read_iter = macvtap_read_iter,
1231 .write_iter = macvtap_write_iter,
1232 .poll = macvtap_poll,
1233 .llseek = no_llseek,
1234 .unlocked_ioctl = macvtap_ioctl,
1235 #ifdef CONFIG_COMPAT
1236 .compat_ioctl = macvtap_compat_ioctl,
1240 static int macvtap_sendmsg(struct socket *sock, struct msghdr *m,
1243 struct macvtap_queue *q = container_of(sock, struct macvtap_queue, sock);
1244 return macvtap_get_user(q, m, &m->msg_iter, m->msg_flags & MSG_DONTWAIT);
1247 static int macvtap_recvmsg(struct socket *sock, struct msghdr *m,
1248 size_t total_len, int flags)
1250 struct macvtap_queue *q = container_of(sock, struct macvtap_queue, sock);
1252 if (flags & ~(MSG_DONTWAIT|MSG_TRUNC))
1254 ret = macvtap_do_read(q, &m->msg_iter, flags & MSG_DONTWAIT);
1255 if (ret > total_len) {
1256 m->msg_flags |= MSG_TRUNC;
1257 ret = flags & MSG_TRUNC ? ret : total_len;
1262 /* Ops structure to mimic raw sockets with tun */
1263 static const struct proto_ops macvtap_socket_ops = {
1264 .sendmsg = macvtap_sendmsg,
1265 .recvmsg = macvtap_recvmsg,
1268 /* Get an underlying socket object from tun file. Returns error unless file is
1269 * attached to a device. The returned object works like a packet socket, it
1270 * can be used for sock_sendmsg/sock_recvmsg. The caller is responsible for
1271 * holding a reference to the file for as long as the socket is in use. */
1272 struct socket *macvtap_get_socket(struct file *file)
1274 struct macvtap_queue *q;
1275 if (file->f_op != &macvtap_fops)
1276 return ERR_PTR(-EINVAL);
1277 q = file->private_data;
1279 return ERR_PTR(-EBADFD);
1282 EXPORT_SYMBOL_GPL(macvtap_get_socket);
1284 static int macvtap_device_event(struct notifier_block *unused,
1285 unsigned long event, void *ptr)
1287 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1288 struct macvlan_dev *vlan;
1289 struct device *classdev;
1292 char tap_name[IFNAMSIZ];
1294 if (dev->rtnl_link_ops != &macvtap_link_ops)
1297 snprintf(tap_name, IFNAMSIZ, "tap%d", dev->ifindex);
1298 vlan = netdev_priv(dev);
1301 case NETDEV_REGISTER:
1302 /* Create the device node here after the network device has
1303 * been registered but before register_netdevice has
1306 err = macvtap_get_minor(vlan);
1308 return notifier_from_errno(err);
1310 devt = MKDEV(MAJOR(macvtap_major), vlan->minor);
1311 classdev = device_create(&macvtap_class, &dev->dev, devt,
1313 if (IS_ERR(classdev)) {
1314 macvtap_free_minor(vlan);
1315 return notifier_from_errno(PTR_ERR(classdev));
1317 err = sysfs_create_link(&dev->dev.kobj, &classdev->kobj,
1320 return notifier_from_errno(err);
1322 case NETDEV_UNREGISTER:
1323 /* vlan->minor == 0 if NETDEV_REGISTER above failed */
1324 if (vlan->minor == 0)
1326 sysfs_remove_link(&dev->dev.kobj, tap_name);
1327 devt = MKDEV(MAJOR(macvtap_major), vlan->minor);
1328 device_destroy(&macvtap_class, devt);
1329 macvtap_free_minor(vlan);
1336 static struct notifier_block macvtap_notifier_block __read_mostly = {
1337 .notifier_call = macvtap_device_event,
1340 static int macvtap_init(void)
1344 err = alloc_chrdev_region(&macvtap_major, 0,
1345 MACVTAP_NUM_DEVS, "macvtap");
1349 cdev_init(&macvtap_cdev, &macvtap_fops);
1350 err = cdev_add(&macvtap_cdev, macvtap_major, MACVTAP_NUM_DEVS);
1354 err = class_register(&macvtap_class);
1358 err = register_netdevice_notifier(&macvtap_notifier_block);
1362 err = macvlan_link_register(&macvtap_link_ops);
1369 unregister_netdevice_notifier(&macvtap_notifier_block);
1371 class_unregister(&macvtap_class);
1373 cdev_del(&macvtap_cdev);
1375 unregister_chrdev_region(macvtap_major, MACVTAP_NUM_DEVS);
1379 module_init(macvtap_init);
1381 static void macvtap_exit(void)
1383 rtnl_link_unregister(&macvtap_link_ops);
1384 unregister_netdevice_notifier(&macvtap_notifier_block);
1385 class_unregister(&macvtap_class);
1386 cdev_del(&macvtap_cdev);
1387 unregister_chrdev_region(macvtap_major, MACVTAP_NUM_DEVS);
1388 idr_destroy(&minor_idr);
1390 module_exit(macvtap_exit);
1392 MODULE_ALIAS_RTNL_LINK("macvtap");
1394 MODULE_LICENSE("GPL");