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 struct class *macvtap_class;
133 static struct cdev macvtap_cdev;
135 static const struct proto_ops macvtap_socket_ops;
137 #define TUN_OFFLOADS (NETIF_F_HW_CSUM | NETIF_F_TSO_ECN | NETIF_F_TSO | \
138 NETIF_F_TSO6 | NETIF_F_UFO)
139 #define RX_OFFLOADS (NETIF_F_GRO | NETIF_F_LRO)
140 #define TAP_FEATURES (NETIF_F_GSO | NETIF_F_SG)
142 static struct macvlan_dev *macvtap_get_vlan_rcu(const struct net_device *dev)
144 return rcu_dereference(dev->rx_handler_data);
149 * The macvtap_queue and the macvlan_dev are loosely coupled, the
150 * pointers from one to the other can only be read while rcu_read_lock
153 * Both the file and the macvlan_dev hold a reference on the macvtap_queue
154 * through sock_hold(&q->sk). When the macvlan_dev goes away first,
155 * q->vlan becomes inaccessible. When the files gets closed,
156 * macvtap_get_queue() fails.
158 * There may still be references to the struct sock inside of the
159 * queue from outbound SKBs, but these never reference back to the
160 * file or the dev. The data structure is freed through __sk_free
161 * when both our references and any pending SKBs are gone.
164 static int macvtap_enable_queue(struct net_device *dev, struct file *file,
165 struct macvtap_queue *q)
167 struct macvlan_dev *vlan = netdev_priv(dev);
176 rcu_assign_pointer(vlan->taps[vlan->numvtaps], q);
177 q->queue_index = vlan->numvtaps;
186 static int macvtap_set_queue(struct net_device *dev, struct file *file,
187 struct macvtap_queue *q)
189 struct macvlan_dev *vlan = netdev_priv(dev);
191 if (vlan->numqueues == MAX_MACVTAP_QUEUES)
194 rcu_assign_pointer(q->vlan, vlan);
195 rcu_assign_pointer(vlan->taps[vlan->numvtaps], q);
199 q->queue_index = vlan->numvtaps;
201 file->private_data = q;
202 list_add_tail(&q->next, &vlan->queue_list);
210 static int macvtap_disable_queue(struct macvtap_queue *q)
212 struct macvlan_dev *vlan;
213 struct macvtap_queue *nq;
219 vlan = rtnl_dereference(q->vlan);
222 int index = q->queue_index;
223 BUG_ON(index >= vlan->numvtaps);
224 nq = rtnl_dereference(vlan->taps[vlan->numvtaps - 1]);
225 nq->queue_index = index;
227 rcu_assign_pointer(vlan->taps[index], nq);
228 RCU_INIT_POINTER(vlan->taps[vlan->numvtaps - 1], NULL);
238 * The file owning the queue got closed, give up both
239 * the reference that the files holds as well as the
240 * one from the macvlan_dev if that still exists.
242 * Using the spinlock makes sure that we don't get
243 * to the queue again after destroying it.
245 static void macvtap_put_queue(struct macvtap_queue *q)
247 struct macvlan_dev *vlan;
250 vlan = rtnl_dereference(q->vlan);
254 BUG_ON(macvtap_disable_queue(q));
257 RCU_INIT_POINTER(q->vlan, NULL);
259 list_del_init(&q->next);
269 * Select a queue based on the rxq of the device on which this packet
270 * arrived. If the incoming device is not mq, calculate a flow hash
271 * to select a queue. If all fails, find the first available queue.
272 * Cache vlan->numvtaps since it can become zero during the execution
275 static struct macvtap_queue *macvtap_get_queue(struct net_device *dev,
278 struct macvlan_dev *vlan = netdev_priv(dev);
279 struct macvtap_queue *tap = NULL;
280 /* Access to taps array is protected by rcu, but access to numvtaps
281 * isn't. Below we use it to lookup a queue, but treat it as a hint
282 * and validate that the result isn't NULL - in case we are
283 * racing against queue removal.
285 int numvtaps = ACCESS_ONCE(vlan->numvtaps);
291 /* Check if we can use flow to select a queue */
292 rxq = skb_get_hash(skb);
294 tap = rcu_dereference(vlan->taps[rxq % numvtaps]);
298 if (likely(skb_rx_queue_recorded(skb))) {
299 rxq = skb_get_rx_queue(skb);
301 while (unlikely(rxq >= numvtaps))
304 tap = rcu_dereference(vlan->taps[rxq]);
308 tap = rcu_dereference(vlan->taps[0]);
314 * The net_device is going away, give up the reference
315 * that it holds on all queues and safely set the pointer
316 * from the queues to NULL.
318 static void macvtap_del_queues(struct net_device *dev)
320 struct macvlan_dev *vlan = netdev_priv(dev);
321 struct macvtap_queue *q, *tmp, *qlist[MAX_MACVTAP_QUEUES];
325 list_for_each_entry_safe(q, tmp, &vlan->queue_list, next) {
326 list_del_init(&q->next);
328 RCU_INIT_POINTER(q->vlan, NULL);
333 for (i = 0; i < vlan->numvtaps; i++)
334 RCU_INIT_POINTER(vlan->taps[i], NULL);
335 BUG_ON(vlan->numvtaps);
336 BUG_ON(vlan->numqueues);
337 /* guarantee that any future macvtap_set_queue will fail */
338 vlan->numvtaps = MAX_MACVTAP_QUEUES;
340 for (--j; j >= 0; j--)
341 sock_put(&qlist[j]->sk);
344 static rx_handler_result_t macvtap_handle_frame(struct sk_buff **pskb)
346 struct sk_buff *skb = *pskb;
347 struct net_device *dev = skb->dev;
348 struct macvlan_dev *vlan;
349 struct macvtap_queue *q;
350 netdev_features_t features = TAP_FEATURES;
352 vlan = macvtap_get_vlan_rcu(dev);
354 return RX_HANDLER_PASS;
356 q = macvtap_get_queue(dev, skb);
358 return RX_HANDLER_PASS;
360 if (skb_queue_len(&q->sk.sk_receive_queue) >= dev->tx_queue_len)
363 skb_push(skb, ETH_HLEN);
365 /* Apply the forward feature mask so that we perform segmentation
366 * according to users wishes. This only works if VNET_HDR is
369 if (q->flags & IFF_VNET_HDR)
370 features |= vlan->tap_features;
371 if (netif_needs_gso(skb, features)) {
372 struct sk_buff *segs = __skb_gso_segment(skb, features, false);
378 skb_queue_tail(&q->sk.sk_receive_queue, skb);
384 struct sk_buff *nskb = segs->next;
387 skb_queue_tail(&q->sk.sk_receive_queue, segs);
391 /* If we receive a partial checksum and the tap side
392 * doesn't support checksum offload, compute the checksum.
393 * Note: it doesn't matter which checksum feature to
394 * check, we either support them all or none.
396 if (skb->ip_summed == CHECKSUM_PARTIAL &&
397 !(features & NETIF_F_ALL_CSUM) &&
398 skb_checksum_help(skb))
400 skb_queue_tail(&q->sk.sk_receive_queue, skb);
404 wake_up_interruptible_poll(sk_sleep(&q->sk), POLLIN | POLLRDNORM | POLLRDBAND);
405 return RX_HANDLER_CONSUMED;
408 /* Count errors/drops only here, thus don't care about args. */
409 macvlan_count_rx(vlan, 0, 0, 0);
411 return RX_HANDLER_CONSUMED;
414 static int macvtap_get_minor(struct macvlan_dev *vlan)
416 int retval = -ENOMEM;
418 mutex_lock(&minor_lock);
419 retval = idr_alloc(&minor_idr, vlan, 1, MACVTAP_NUM_DEVS, GFP_KERNEL);
421 vlan->minor = retval;
422 } else if (retval == -ENOSPC) {
423 printk(KERN_ERR "too many macvtap devices\n");
426 mutex_unlock(&minor_lock);
427 return retval < 0 ? retval : 0;
430 static void macvtap_free_minor(struct macvlan_dev *vlan)
432 mutex_lock(&minor_lock);
434 idr_remove(&minor_idr, vlan->minor);
437 mutex_unlock(&minor_lock);
440 static struct net_device *dev_get_by_macvtap_minor(int minor)
442 struct net_device *dev = NULL;
443 struct macvlan_dev *vlan;
445 mutex_lock(&minor_lock);
446 vlan = idr_find(&minor_idr, minor);
451 mutex_unlock(&minor_lock);
455 static int macvtap_newlink(struct net *src_net,
456 struct net_device *dev,
458 struct nlattr *data[])
460 struct macvlan_dev *vlan = netdev_priv(dev);
463 INIT_LIST_HEAD(&vlan->queue_list);
465 /* Since macvlan supports all offloads by default, make
466 * tap support all offloads also.
468 vlan->tap_features = TUN_OFFLOADS;
470 err = netdev_rx_handler_register(dev, macvtap_handle_frame, vlan);
474 /* Don't put anything that may fail after macvlan_common_newlink
475 * because we can't undo what it does.
477 return macvlan_common_newlink(src_net, dev, tb, data);
480 static void macvtap_dellink(struct net_device *dev,
481 struct list_head *head)
483 netdev_rx_handler_unregister(dev);
484 macvtap_del_queues(dev);
485 macvlan_dellink(dev, head);
488 static void macvtap_setup(struct net_device *dev)
490 macvlan_common_setup(dev);
491 dev->tx_queue_len = TUN_READQ_SIZE;
494 static struct rtnl_link_ops macvtap_link_ops __read_mostly = {
496 .setup = macvtap_setup,
497 .newlink = macvtap_newlink,
498 .dellink = macvtap_dellink,
502 static void macvtap_sock_write_space(struct sock *sk)
504 wait_queue_head_t *wqueue;
506 if (!sock_writeable(sk) ||
507 !test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags))
510 wqueue = sk_sleep(sk);
511 if (wqueue && waitqueue_active(wqueue))
512 wake_up_interruptible_poll(wqueue, POLLOUT | POLLWRNORM | POLLWRBAND);
515 static void macvtap_sock_destruct(struct sock *sk)
517 skb_queue_purge(&sk->sk_receive_queue);
520 static int macvtap_open(struct inode *inode, struct file *file)
522 struct net *net = current->nsproxy->net_ns;
523 struct net_device *dev;
524 struct macvtap_queue *q;
528 dev = dev_get_by_macvtap_minor(iminor(inode));
533 q = (struct macvtap_queue *)sk_alloc(net, AF_UNSPEC, GFP_KERNEL,
538 RCU_INIT_POINTER(q->sock.wq, &q->wq);
539 init_waitqueue_head(&q->wq.wait);
540 q->sock.type = SOCK_RAW;
541 q->sock.state = SS_CONNECTED;
543 q->sock.ops = &macvtap_socket_ops;
544 sock_init_data(&q->sock, &q->sk);
545 q->sk.sk_write_space = macvtap_sock_write_space;
546 q->sk.sk_destruct = macvtap_sock_destruct;
547 q->flags = IFF_VNET_HDR | IFF_NO_PI | IFF_TAP;
548 q->vnet_hdr_sz = sizeof(struct virtio_net_hdr);
551 * so far only KVM virtio_net uses macvtap, enable zero copy between
552 * guest kernel and host kernel when lower device supports zerocopy
554 * The macvlan supports zerocopy iff the lower device supports zero
555 * copy so we don't have to look at the lower device directly.
557 if ((dev->features & NETIF_F_HIGHDMA) && (dev->features & NETIF_F_SG))
558 sock_set_flag(&q->sk, SOCK_ZEROCOPY);
560 err = macvtap_set_queue(dev, file, q);
572 static int macvtap_release(struct inode *inode, struct file *file)
574 struct macvtap_queue *q = file->private_data;
575 macvtap_put_queue(q);
579 static unsigned int macvtap_poll(struct file *file, poll_table * wait)
581 struct macvtap_queue *q = file->private_data;
582 unsigned int mask = POLLERR;
588 poll_wait(file, &q->wq.wait, wait);
590 if (!skb_queue_empty(&q->sk.sk_receive_queue))
591 mask |= POLLIN | POLLRDNORM;
593 if (sock_writeable(&q->sk) ||
594 (!test_and_set_bit(SOCK_ASYNC_NOSPACE, &q->sock.flags) &&
595 sock_writeable(&q->sk)))
596 mask |= POLLOUT | POLLWRNORM;
602 static inline struct sk_buff *macvtap_alloc_skb(struct sock *sk, size_t prepad,
603 size_t len, size_t linear,
604 int noblock, int *err)
608 /* Under a page? Don't bother with paged skb. */
609 if (prepad + len < PAGE_SIZE || !linear)
612 skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
617 skb_reserve(skb, prepad);
618 skb_put(skb, linear);
619 skb->data_len = len - linear;
620 skb->len += len - linear;
626 * macvtap_skb_from_vnet_hdr and macvtap_skb_to_vnet_hdr should
627 * be shared with the tun/tap driver.
629 static int macvtap_skb_from_vnet_hdr(struct macvtap_queue *q,
631 struct virtio_net_hdr *vnet_hdr)
633 unsigned short gso_type = 0;
634 if (vnet_hdr->gso_type != VIRTIO_NET_HDR_GSO_NONE) {
635 switch (vnet_hdr->gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
636 case VIRTIO_NET_HDR_GSO_TCPV4:
637 gso_type = SKB_GSO_TCPV4;
639 case VIRTIO_NET_HDR_GSO_TCPV6:
640 gso_type = SKB_GSO_TCPV6;
642 case VIRTIO_NET_HDR_GSO_UDP:
643 gso_type = SKB_GSO_UDP;
649 if (vnet_hdr->gso_type & VIRTIO_NET_HDR_GSO_ECN)
650 gso_type |= SKB_GSO_TCP_ECN;
652 if (vnet_hdr->gso_size == 0)
656 if (vnet_hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
657 if (!skb_partial_csum_set(skb, macvtap16_to_cpu(q, vnet_hdr->csum_start),
658 macvtap16_to_cpu(q, vnet_hdr->csum_offset)))
662 if (vnet_hdr->gso_type != VIRTIO_NET_HDR_GSO_NONE) {
663 skb_shinfo(skb)->gso_size = macvtap16_to_cpu(q, vnet_hdr->gso_size);
664 skb_shinfo(skb)->gso_type = gso_type;
666 /* Header must be checked, and gso_segs computed. */
667 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
668 skb_shinfo(skb)->gso_segs = 0;
673 static void macvtap_skb_to_vnet_hdr(struct macvtap_queue *q,
674 const struct sk_buff *skb,
675 struct virtio_net_hdr *vnet_hdr)
677 memset(vnet_hdr, 0, sizeof(*vnet_hdr));
679 if (skb_is_gso(skb)) {
680 struct skb_shared_info *sinfo = skb_shinfo(skb);
682 /* This is a hint as to how much should be linear. */
683 vnet_hdr->hdr_len = cpu_to_macvtap16(q, skb_headlen(skb));
684 vnet_hdr->gso_size = cpu_to_macvtap16(q, sinfo->gso_size);
685 if (sinfo->gso_type & SKB_GSO_TCPV4)
686 vnet_hdr->gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
687 else if (sinfo->gso_type & SKB_GSO_TCPV6)
688 vnet_hdr->gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
689 else if (sinfo->gso_type & SKB_GSO_UDP)
690 vnet_hdr->gso_type = VIRTIO_NET_HDR_GSO_UDP;
693 if (sinfo->gso_type & SKB_GSO_TCP_ECN)
694 vnet_hdr->gso_type |= VIRTIO_NET_HDR_GSO_ECN;
696 vnet_hdr->gso_type = VIRTIO_NET_HDR_GSO_NONE;
698 if (skb->ip_summed == CHECKSUM_PARTIAL) {
699 vnet_hdr->flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
700 if (skb_vlan_tag_present(skb))
701 vnet_hdr->csum_start = cpu_to_macvtap16(q,
702 skb_checksum_start_offset(skb) + VLAN_HLEN);
704 vnet_hdr->csum_start = cpu_to_macvtap16(q,
705 skb_checksum_start_offset(skb));
706 vnet_hdr->csum_offset = cpu_to_macvtap16(q, skb->csum_offset);
707 } else if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
708 vnet_hdr->flags = VIRTIO_NET_HDR_F_DATA_VALID;
709 } /* else everything is zero */
712 /* Neighbour code has some assumptions on HH_DATA_MOD alignment */
713 #define MACVTAP_RESERVE HH_DATA_OFF(ETH_HLEN)
715 /* Get packet from user space buffer */
716 static ssize_t macvtap_get_user(struct macvtap_queue *q, struct msghdr *m,
717 struct iov_iter *from, int noblock)
719 int good_linear = SKB_MAX_HEAD(MACVTAP_RESERVE);
721 struct macvlan_dev *vlan;
722 unsigned long total_len = iov_iter_count(from);
723 unsigned long len = total_len;
725 struct virtio_net_hdr vnet_hdr = { 0 };
726 int vnet_hdr_len = 0;
728 bool zerocopy = false;
732 if (q->flags & IFF_VNET_HDR) {
733 vnet_hdr_len = q->vnet_hdr_sz;
736 if (len < vnet_hdr_len)
741 n = copy_from_iter(&vnet_hdr, sizeof(vnet_hdr), from);
742 if (n != sizeof(vnet_hdr))
744 iov_iter_advance(from, vnet_hdr_len - sizeof(vnet_hdr));
745 if ((vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
746 macvtap16_to_cpu(q, vnet_hdr.csum_start) +
747 macvtap16_to_cpu(q, vnet_hdr.csum_offset) + 2 >
748 macvtap16_to_cpu(q, vnet_hdr.hdr_len))
749 vnet_hdr.hdr_len = cpu_to_macvtap16(q,
750 macvtap16_to_cpu(q, vnet_hdr.csum_start) +
751 macvtap16_to_cpu(q, vnet_hdr.csum_offset) + 2);
753 if (macvtap16_to_cpu(q, vnet_hdr.hdr_len) > len)
758 if (unlikely(len < ETH_HLEN))
761 if (m && m->msg_control && sock_flag(&q->sk, SOCK_ZEROCOPY)) {
764 copylen = vnet_hdr.hdr_len ?
765 macvtap16_to_cpu(q, vnet_hdr.hdr_len) : GOODCOPY_LEN;
766 if (copylen > good_linear)
767 copylen = good_linear;
770 iov_iter_advance(&i, copylen);
771 if (iov_iter_npages(&i, INT_MAX) <= MAX_SKB_FRAGS)
777 if (macvtap16_to_cpu(q, vnet_hdr.hdr_len) > good_linear)
778 linear = good_linear;
780 linear = macvtap16_to_cpu(q, vnet_hdr.hdr_len);
783 skb = macvtap_alloc_skb(&q->sk, MACVTAP_RESERVE, copylen,
784 linear, noblock, &err);
789 err = zerocopy_sg_from_iter(skb, from);
791 err = skb_copy_datagram_from_iter(skb, 0, from, len);
792 if (!err && m && m->msg_control) {
793 struct ubuf_info *uarg = m->msg_control;
794 uarg->callback(uarg, false);
801 skb_set_network_header(skb, ETH_HLEN);
802 skb_reset_mac_header(skb);
803 skb->protocol = eth_hdr(skb)->h_proto;
806 err = macvtap_skb_from_vnet_hdr(q, skb, &vnet_hdr);
811 skb_probe_transport_header(skb, ETH_HLEN);
814 vlan = rcu_dereference(q->vlan);
815 /* copy skb_ubuf_info for callback when skb has no error */
817 skb_shinfo(skb)->destructor_arg = m->msg_control;
818 skb_shinfo(skb)->tx_flags |= SKBTX_DEV_ZEROCOPY;
819 skb_shinfo(skb)->tx_flags |= SKBTX_SHARED_FRAG;
822 skb->dev = vlan->dev;
836 vlan = rcu_dereference(q->vlan);
838 this_cpu_inc(vlan->pcpu_stats->tx_dropped);
844 static ssize_t macvtap_write_iter(struct kiocb *iocb, struct iov_iter *from)
846 struct file *file = iocb->ki_filp;
847 struct macvtap_queue *q = file->private_data;
849 return macvtap_get_user(q, NULL, from, file->f_flags & O_NONBLOCK);
852 /* Put packet to the user space buffer */
853 static ssize_t macvtap_put_user(struct macvtap_queue *q,
854 const struct sk_buff *skb,
855 struct iov_iter *iter)
858 int vnet_hdr_len = 0;
862 if (q->flags & IFF_VNET_HDR) {
863 struct virtio_net_hdr vnet_hdr;
864 vnet_hdr_len = q->vnet_hdr_sz;
865 if (iov_iter_count(iter) < vnet_hdr_len)
868 macvtap_skb_to_vnet_hdr(q, skb, &vnet_hdr);
870 if (copy_to_iter(&vnet_hdr, sizeof(vnet_hdr), iter) !=
874 iov_iter_advance(iter, vnet_hdr_len - sizeof(vnet_hdr));
876 total = vnet_hdr_len;
879 if (skb_vlan_tag_present(skb)) {
884 veth.h_vlan_proto = skb->vlan_proto;
885 veth.h_vlan_TCI = htons(skb_vlan_tag_get(skb));
887 vlan_offset = offsetof(struct vlan_ethhdr, h_vlan_proto);
890 ret = skb_copy_datagram_iter(skb, 0, iter, vlan_offset);
891 if (ret || !iov_iter_count(iter))
894 ret = copy_to_iter(&veth, sizeof(veth), iter);
895 if (ret != sizeof(veth) || !iov_iter_count(iter))
899 ret = skb_copy_datagram_iter(skb, vlan_offset, iter,
900 skb->len - vlan_offset);
903 return ret ? ret : total;
906 static ssize_t macvtap_do_read(struct macvtap_queue *q,
914 if (!iov_iter_count(to))
919 prepare_to_wait(sk_sleep(&q->sk), &wait,
922 /* Read frames from the queue */
923 skb = skb_dequeue(&q->sk.sk_receive_queue);
930 if (signal_pending(current)) {
934 /* Nothing to read, let's sleep */
938 ret = macvtap_put_user(q, skb, to);
939 if (unlikely(ret < 0))
945 finish_wait(sk_sleep(&q->sk), &wait);
949 static ssize_t macvtap_read_iter(struct kiocb *iocb, struct iov_iter *to)
951 struct file *file = iocb->ki_filp;
952 struct macvtap_queue *q = file->private_data;
953 ssize_t len = iov_iter_count(to), ret;
955 ret = macvtap_do_read(q, to, file->f_flags & O_NONBLOCK);
956 ret = min_t(ssize_t, ret, len);
962 static struct macvlan_dev *macvtap_get_vlan(struct macvtap_queue *q)
964 struct macvlan_dev *vlan;
967 vlan = rtnl_dereference(q->vlan);
974 static void macvtap_put_vlan(struct macvlan_dev *vlan)
979 static int macvtap_ioctl_set_queue(struct file *file, unsigned int flags)
981 struct macvtap_queue *q = file->private_data;
982 struct macvlan_dev *vlan;
985 vlan = macvtap_get_vlan(q);
989 if (flags & IFF_ATTACH_QUEUE)
990 ret = macvtap_enable_queue(vlan->dev, file, q);
991 else if (flags & IFF_DETACH_QUEUE)
992 ret = macvtap_disable_queue(q);
996 macvtap_put_vlan(vlan);
1000 static int set_offload(struct macvtap_queue *q, unsigned long arg)
1002 struct macvlan_dev *vlan;
1003 netdev_features_t features;
1004 netdev_features_t feature_mask = 0;
1006 vlan = rtnl_dereference(q->vlan);
1010 features = vlan->dev->features;
1012 if (arg & TUN_F_CSUM) {
1013 feature_mask = NETIF_F_HW_CSUM;
1015 if (arg & (TUN_F_TSO4 | TUN_F_TSO6)) {
1016 if (arg & TUN_F_TSO_ECN)
1017 feature_mask |= NETIF_F_TSO_ECN;
1018 if (arg & TUN_F_TSO4)
1019 feature_mask |= NETIF_F_TSO;
1020 if (arg & TUN_F_TSO6)
1021 feature_mask |= NETIF_F_TSO6;
1024 if (arg & TUN_F_UFO)
1025 feature_mask |= NETIF_F_UFO;
1028 /* tun/tap driver inverts the usage for TSO offloads, where
1029 * setting the TSO bit means that the userspace wants to
1030 * accept TSO frames and turning it off means that user space
1031 * does not support TSO.
1032 * For macvtap, we have to invert it to mean the same thing.
1033 * When user space turns off TSO, we turn off GSO/LRO so that
1034 * user-space will not receive TSO frames.
1036 if (feature_mask & (NETIF_F_TSO | NETIF_F_TSO6 | NETIF_F_UFO))
1037 features |= RX_OFFLOADS;
1039 features &= ~RX_OFFLOADS;
1041 /* tap_features are the same as features on tun/tap and
1042 * reflect user expectations.
1044 vlan->tap_features = feature_mask;
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 & ~MACVTAP_FEATURES) != (IFF_NO_PI | IFF_TAP))
1077 q->flags = (q->flags & ~MACVTAP_FEATURES) | u;
1083 vlan = macvtap_get_vlan(q);
1091 if (copy_to_user(&ifr->ifr_name, vlan->dev->name, IFNAMSIZ) ||
1092 put_user(u, &ifr->ifr_flags))
1094 macvtap_put_vlan(vlan);
1099 if (get_user(u, &ifr->ifr_flags))
1102 ret = macvtap_ioctl_set_queue(file, u);
1106 case TUNGETFEATURES:
1107 if (put_user(IFF_TAP | IFF_NO_PI | MACVTAP_FEATURES, 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 s = !!(q->flags & MACVTAP_VNET_LE);
1135 if (put_user(s, sp))
1140 if (get_user(s, sp))
1143 q->flags |= MACVTAP_VNET_LE;
1145 q->flags &= ~MACVTAP_VNET_LE;
1149 return macvtap_get_vnet_be(q, sp);
1152 return macvtap_set_vnet_be(q, sp);
1155 /* let the user check for future flags */
1156 if (arg & ~(TUN_F_CSUM | TUN_F_TSO4 | TUN_F_TSO6 |
1157 TUN_F_TSO_ECN | TUN_F_UFO))
1161 ret = set_offload(q, arg);
1170 #ifdef CONFIG_COMPAT
1171 static long macvtap_compat_ioctl(struct file *file, unsigned int cmd,
1174 return macvtap_ioctl(file, cmd, (unsigned long)compat_ptr(arg));
1178 static const struct file_operations macvtap_fops = {
1179 .owner = THIS_MODULE,
1180 .open = macvtap_open,
1181 .release = macvtap_release,
1182 .read_iter = macvtap_read_iter,
1183 .write_iter = macvtap_write_iter,
1184 .poll = macvtap_poll,
1185 .llseek = no_llseek,
1186 .unlocked_ioctl = macvtap_ioctl,
1187 #ifdef CONFIG_COMPAT
1188 .compat_ioctl = macvtap_compat_ioctl,
1192 static int macvtap_sendmsg(struct socket *sock, struct msghdr *m,
1195 struct macvtap_queue *q = container_of(sock, struct macvtap_queue, sock);
1196 return macvtap_get_user(q, m, &m->msg_iter, m->msg_flags & MSG_DONTWAIT);
1199 static int macvtap_recvmsg(struct socket *sock, struct msghdr *m,
1200 size_t total_len, int flags)
1202 struct macvtap_queue *q = container_of(sock, struct macvtap_queue, sock);
1204 if (flags & ~(MSG_DONTWAIT|MSG_TRUNC))
1206 ret = macvtap_do_read(q, &m->msg_iter, flags & MSG_DONTWAIT);
1207 if (ret > total_len) {
1208 m->msg_flags |= MSG_TRUNC;
1209 ret = flags & MSG_TRUNC ? ret : total_len;
1214 /* Ops structure to mimic raw sockets with tun */
1215 static const struct proto_ops macvtap_socket_ops = {
1216 .sendmsg = macvtap_sendmsg,
1217 .recvmsg = macvtap_recvmsg,
1220 /* Get an underlying socket object from tun file. Returns error unless file is
1221 * attached to a device. The returned object works like a packet socket, it
1222 * can be used for sock_sendmsg/sock_recvmsg. The caller is responsible for
1223 * holding a reference to the file for as long as the socket is in use. */
1224 struct socket *macvtap_get_socket(struct file *file)
1226 struct macvtap_queue *q;
1227 if (file->f_op != &macvtap_fops)
1228 return ERR_PTR(-EINVAL);
1229 q = file->private_data;
1231 return ERR_PTR(-EBADFD);
1234 EXPORT_SYMBOL_GPL(macvtap_get_socket);
1236 static int macvtap_device_event(struct notifier_block *unused,
1237 unsigned long event, void *ptr)
1239 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1240 struct macvlan_dev *vlan;
1241 struct device *classdev;
1245 if (dev->rtnl_link_ops != &macvtap_link_ops)
1248 vlan = netdev_priv(dev);
1251 case NETDEV_REGISTER:
1252 /* Create the device node here after the network device has
1253 * been registered but before register_netdevice has
1256 err = macvtap_get_minor(vlan);
1258 return notifier_from_errno(err);
1260 devt = MKDEV(MAJOR(macvtap_major), vlan->minor);
1261 classdev = device_create(macvtap_class, &dev->dev, devt,
1262 dev, "tap%d", dev->ifindex);
1263 if (IS_ERR(classdev)) {
1264 macvtap_free_minor(vlan);
1265 return notifier_from_errno(PTR_ERR(classdev));
1268 case NETDEV_UNREGISTER:
1269 devt = MKDEV(MAJOR(macvtap_major), vlan->minor);
1270 device_destroy(macvtap_class, devt);
1271 macvtap_free_minor(vlan);
1278 static struct notifier_block macvtap_notifier_block __read_mostly = {
1279 .notifier_call = macvtap_device_event,
1282 static int macvtap_init(void)
1286 err = alloc_chrdev_region(&macvtap_major, 0,
1287 MACVTAP_NUM_DEVS, "macvtap");
1291 cdev_init(&macvtap_cdev, &macvtap_fops);
1292 err = cdev_add(&macvtap_cdev, macvtap_major, MACVTAP_NUM_DEVS);
1296 macvtap_class = class_create(THIS_MODULE, "macvtap");
1297 if (IS_ERR(macvtap_class)) {
1298 err = PTR_ERR(macvtap_class);
1302 err = register_netdevice_notifier(&macvtap_notifier_block);
1306 err = macvlan_link_register(&macvtap_link_ops);
1313 unregister_netdevice_notifier(&macvtap_notifier_block);
1315 class_unregister(macvtap_class);
1317 cdev_del(&macvtap_cdev);
1319 unregister_chrdev_region(macvtap_major, MACVTAP_NUM_DEVS);
1323 module_init(macvtap_init);
1325 static void macvtap_exit(void)
1327 rtnl_link_unregister(&macvtap_link_ops);
1328 unregister_netdevice_notifier(&macvtap_notifier_block);
1329 class_unregister(macvtap_class);
1330 cdev_del(&macvtap_cdev);
1331 unregister_chrdev_region(macvtap_major, MACVTAP_NUM_DEVS);
1333 module_exit(macvtap_exit);
1335 MODULE_ALIAS_RTNL_LINK("macvtap");
1336 MODULE_AUTHOR("Arnd Bergmann <arnd@arndb.de>");
1337 MODULE_LICENSE("GPL");