1 // SPDX-License-Identifier: GPL-2.0-only
2 #include <linux/etherdevice.h>
3 #include <linux/if_tap.h>
4 #include <linux/if_vlan.h>
5 #include <linux/interrupt.h>
6 #include <linux/nsproxy.h>
7 #include <linux/compat.h>
8 #include <linux/if_tun.h>
9 #include <linux/module.h>
10 #include <linux/skbuff.h>
11 #include <linux/cache.h>
12 #include <linux/sched/signal.h>
13 #include <linux/types.h>
14 #include <linux/slab.h>
15 #include <linux/wait.h>
16 #include <linux/cdev.h>
17 #include <linux/idr.h>
19 #include <linux/uio.h>
22 #include <net/net_namespace.h>
23 #include <net/rtnetlink.h>
25 #include <linux/virtio_net.h>
26 #include <linux/skb_array.h>
28 #define TAP_IFFEATURES (IFF_VNET_HDR | IFF_MULTI_QUEUE)
30 #define TAP_VNET_LE 0x80000000
31 #define TAP_VNET_BE 0x40000000
33 #ifdef CONFIG_TUN_VNET_CROSS_LE
34 static inline bool tap_legacy_is_little_endian(struct tap_queue *q)
36 return q->flags & TAP_VNET_BE ? false :
37 virtio_legacy_is_little_endian();
40 static long tap_get_vnet_be(struct tap_queue *q, int __user *sp)
42 int s = !!(q->flags & TAP_VNET_BE);
50 static long tap_set_vnet_be(struct tap_queue *q, int __user *sp)
58 q->flags |= TAP_VNET_BE;
60 q->flags &= ~TAP_VNET_BE;
65 static inline bool tap_legacy_is_little_endian(struct tap_queue *q)
67 return virtio_legacy_is_little_endian();
70 static long tap_get_vnet_be(struct tap_queue *q, int __user *argp)
75 static long tap_set_vnet_be(struct tap_queue *q, int __user *argp)
79 #endif /* CONFIG_TUN_VNET_CROSS_LE */
81 static inline bool tap_is_little_endian(struct tap_queue *q)
83 return q->flags & TAP_VNET_LE ||
84 tap_legacy_is_little_endian(q);
87 static inline u16 tap16_to_cpu(struct tap_queue *q, __virtio16 val)
89 return __virtio16_to_cpu(tap_is_little_endian(q), val);
92 static inline __virtio16 cpu_to_tap16(struct tap_queue *q, u16 val)
94 return __cpu_to_virtio16(tap_is_little_endian(q), val);
97 static struct proto tap_proto = {
100 .obj_size = sizeof(struct tap_queue),
103 #define TAP_NUM_DEVS (1U << MINORBITS)
105 static LIST_HEAD(major_list);
110 struct idr minor_idr;
111 spinlock_t minor_lock;
112 const char *device_name;
113 struct list_head next;
116 #define GOODCOPY_LEN 128
118 static const struct proto_ops tap_socket_ops;
120 #define RX_OFFLOADS (NETIF_F_GRO | NETIF_F_LRO)
121 #define TAP_FEATURES (NETIF_F_GSO | NETIF_F_SG | NETIF_F_FRAGLIST)
123 static struct tap_dev *tap_dev_get_rcu(const struct net_device *dev)
125 return rcu_dereference(dev->rx_handler_data);
130 * The tap_queue and the macvlan_dev are loosely coupled, the
131 * pointers from one to the other can only be read while rcu_read_lock
134 * Both the file and the macvlan_dev hold a reference on the tap_queue
135 * through sock_hold(&q->sk). When the macvlan_dev goes away first,
136 * q->vlan becomes inaccessible. When the files gets closed,
137 * tap_get_queue() fails.
139 * There may still be references to the struct sock inside of the
140 * queue from outbound SKBs, but these never reference back to the
141 * file or the dev. The data structure is freed through __sk_free
142 * when both our references and any pending SKBs are gone.
145 static int tap_enable_queue(struct tap_dev *tap, struct file *file,
156 rcu_assign_pointer(tap->taps[tap->numvtaps], q);
157 q->queue_index = tap->numvtaps;
166 static int tap_set_queue(struct tap_dev *tap, struct file *file,
169 if (tap->numqueues == MAX_TAP_QUEUES)
172 rcu_assign_pointer(q->tap, tap);
173 rcu_assign_pointer(tap->taps[tap->numvtaps], q);
177 q->queue_index = tap->numvtaps;
179 file->private_data = q;
180 list_add_tail(&q->next, &tap->queue_list);
188 static int tap_disable_queue(struct tap_queue *q)
191 struct tap_queue *nq;
197 tap = rtnl_dereference(q->tap);
200 int index = q->queue_index;
201 BUG_ON(index >= tap->numvtaps);
202 nq = rtnl_dereference(tap->taps[tap->numvtaps - 1]);
203 nq->queue_index = index;
205 rcu_assign_pointer(tap->taps[index], nq);
206 RCU_INIT_POINTER(tap->taps[tap->numvtaps - 1], NULL);
216 * The file owning the queue got closed, give up both
217 * the reference that the files holds as well as the
218 * one from the macvlan_dev if that still exists.
220 * Using the spinlock makes sure that we don't get
221 * to the queue again after destroying it.
223 static void tap_put_queue(struct tap_queue *q)
228 tap = rtnl_dereference(q->tap);
232 BUG_ON(tap_disable_queue(q));
235 RCU_INIT_POINTER(q->tap, NULL);
237 list_del_init(&q->next);
247 * Select a queue based on the rxq of the device on which this packet
248 * arrived. If the incoming device is not mq, calculate a flow hash
249 * to select a queue. If all fails, find the first available queue.
250 * Cache vlan->numvtaps since it can become zero during the execution
253 static struct tap_queue *tap_get_queue(struct tap_dev *tap,
256 struct tap_queue *queue = NULL;
257 /* Access to taps array is protected by rcu, but access to numvtaps
258 * isn't. Below we use it to lookup a queue, but treat it as a hint
259 * and validate that the result isn't NULL - in case we are
260 * racing against queue removal.
262 int numvtaps = READ_ONCE(tap->numvtaps);
271 /* Check if we can use flow to select a queue */
272 rxq = skb_get_hash(skb);
274 queue = rcu_dereference(tap->taps[rxq % numvtaps]);
278 if (likely(skb_rx_queue_recorded(skb))) {
279 rxq = skb_get_rx_queue(skb);
281 while (unlikely(rxq >= numvtaps))
284 queue = rcu_dereference(tap->taps[rxq]);
289 queue = rcu_dereference(tap->taps[0]);
295 * The net_device is going away, give up the reference
296 * that it holds on all queues and safely set the pointer
297 * from the queues to NULL.
299 void tap_del_queues(struct tap_dev *tap)
301 struct tap_queue *q, *tmp;
304 list_for_each_entry_safe(q, tmp, &tap->queue_list, next) {
305 list_del_init(&q->next);
306 RCU_INIT_POINTER(q->tap, NULL);
312 BUG_ON(tap->numvtaps);
313 BUG_ON(tap->numqueues);
314 /* guarantee that any future tap_set_queue will fail */
315 tap->numvtaps = MAX_TAP_QUEUES;
317 EXPORT_SYMBOL_GPL(tap_del_queues);
319 rx_handler_result_t tap_handle_frame(struct sk_buff **pskb)
321 struct sk_buff *skb = *pskb;
322 struct net_device *dev = skb->dev;
325 netdev_features_t features = TAP_FEATURES;
326 enum skb_drop_reason drop_reason;
328 tap = tap_dev_get_rcu(dev);
330 return RX_HANDLER_PASS;
332 q = tap_get_queue(tap, skb);
334 return RX_HANDLER_PASS;
336 skb_push(skb, ETH_HLEN);
338 /* Apply the forward feature mask so that we perform segmentation
339 * according to users wishes. This only works if VNET_HDR is
342 if (q->flags & IFF_VNET_HDR)
343 features |= tap->tap_features;
344 if (netif_needs_gso(skb, features)) {
345 struct sk_buff *segs = __skb_gso_segment(skb, features, false);
346 struct sk_buff *next;
349 drop_reason = SKB_DROP_REASON_SKB_GSO_SEG;
354 if (ptr_ring_produce(&q->ring, skb)) {
355 drop_reason = SKB_DROP_REASON_FULL_RING;
362 skb_list_walk_safe(segs, skb, next) {
363 skb_mark_not_on_list(skb);
364 if (ptr_ring_produce(&q->ring, skb)) {
365 drop_reason = SKB_DROP_REASON_FULL_RING;
366 kfree_skb_reason(skb, drop_reason);
367 kfree_skb_list_reason(next, drop_reason);
372 /* If we receive a partial checksum and the tap side
373 * doesn't support checksum offload, compute the checksum.
374 * Note: it doesn't matter which checksum feature to
375 * check, we either support them all or none.
377 if (skb->ip_summed == CHECKSUM_PARTIAL &&
378 !(features & NETIF_F_CSUM_MASK) &&
379 skb_checksum_help(skb)) {
380 drop_reason = SKB_DROP_REASON_SKB_CSUM;
383 if (ptr_ring_produce(&q->ring, skb)) {
384 drop_reason = SKB_DROP_REASON_FULL_RING;
390 wake_up_interruptible_poll(sk_sleep(&q->sk), EPOLLIN | EPOLLRDNORM | EPOLLRDBAND);
391 return RX_HANDLER_CONSUMED;
394 /* Count errors/drops only here, thus don't care about args. */
395 if (tap->count_rx_dropped)
396 tap->count_rx_dropped(tap);
397 kfree_skb_reason(skb, drop_reason);
398 return RX_HANDLER_CONSUMED;
400 EXPORT_SYMBOL_GPL(tap_handle_frame);
402 static struct major_info *tap_get_major(int major)
404 struct major_info *tap_major;
406 list_for_each_entry_rcu(tap_major, &major_list, next) {
407 if (tap_major->major == major)
414 int tap_get_minor(dev_t major, struct tap_dev *tap)
416 int retval = -ENOMEM;
417 struct major_info *tap_major;
420 tap_major = tap_get_major(MAJOR(major));
426 spin_lock(&tap_major->minor_lock);
427 retval = idr_alloc(&tap_major->minor_idr, tap, 1, TAP_NUM_DEVS, GFP_ATOMIC);
430 } else if (retval == -ENOSPC) {
431 netdev_err(tap->dev, "Too many tap devices\n");
434 spin_unlock(&tap_major->minor_lock);
438 return retval < 0 ? retval : 0;
440 EXPORT_SYMBOL_GPL(tap_get_minor);
442 void tap_free_minor(dev_t major, struct tap_dev *tap)
444 struct major_info *tap_major;
447 tap_major = tap_get_major(MAJOR(major));
452 spin_lock(&tap_major->minor_lock);
454 idr_remove(&tap_major->minor_idr, tap->minor);
457 spin_unlock(&tap_major->minor_lock);
462 EXPORT_SYMBOL_GPL(tap_free_minor);
464 static struct tap_dev *dev_get_by_tap_file(int major, int minor)
466 struct net_device *dev = NULL;
468 struct major_info *tap_major;
471 tap_major = tap_get_major(major);
477 spin_lock(&tap_major->minor_lock);
478 tap = idr_find(&tap_major->minor_idr, minor);
483 spin_unlock(&tap_major->minor_lock);
490 static void tap_sock_write_space(struct sock *sk)
492 wait_queue_head_t *wqueue;
494 if (!sock_writeable(sk) ||
495 !test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags))
498 wqueue = sk_sleep(sk);
499 if (wqueue && waitqueue_active(wqueue))
500 wake_up_interruptible_poll(wqueue, EPOLLOUT | EPOLLWRNORM | EPOLLWRBAND);
503 static void tap_sock_destruct(struct sock *sk)
505 struct tap_queue *q = container_of(sk, struct tap_queue, sk);
507 ptr_ring_cleanup(&q->ring, __skb_array_destroy_skb);
510 static int tap_open(struct inode *inode, struct file *file)
512 struct net *net = current->nsproxy->net_ns;
518 tap = dev_get_by_tap_file(imajor(inode), iminor(inode));
523 q = (struct tap_queue *)sk_alloc(net, AF_UNSPEC, GFP_KERNEL,
527 if (ptr_ring_init(&q->ring, tap->dev->tx_queue_len, GFP_KERNEL)) {
532 init_waitqueue_head(&q->sock.wq.wait);
533 q->sock.type = SOCK_RAW;
534 q->sock.state = SS_CONNECTED;
536 q->sock.ops = &tap_socket_ops;
537 sock_init_data_uid(&q->sock, &q->sk, inode->i_uid);
538 q->sk.sk_write_space = tap_sock_write_space;
539 q->sk.sk_destruct = tap_sock_destruct;
540 q->flags = IFF_VNET_HDR | IFF_NO_PI | IFF_TAP;
541 q->vnet_hdr_sz = sizeof(struct virtio_net_hdr);
544 * so far only KVM virtio_net uses tap, enable zero copy between
545 * guest kernel and host kernel when lower device supports zerocopy
547 * The macvlan supports zerocopy iff the lower device supports zero
548 * copy so we don't have to look at the lower device directly.
550 if ((tap->dev->features & NETIF_F_HIGHDMA) && (tap->dev->features & NETIF_F_SG))
551 sock_set_flag(&q->sk, SOCK_ZEROCOPY);
553 err = tap_set_queue(tap, file, q);
555 /* tap_sock_destruct() will take care of freeing ptr_ring */
559 /* tap groks IOCB_NOWAIT just fine, mark it as such */
560 file->f_mode |= FMODE_NOWAIT;
577 static int tap_release(struct inode *inode, struct file *file)
579 struct tap_queue *q = file->private_data;
584 static __poll_t tap_poll(struct file *file, poll_table *wait)
586 struct tap_queue *q = file->private_data;
587 __poll_t mask = EPOLLERR;
593 poll_wait(file, &q->sock.wq.wait, wait);
595 if (!ptr_ring_empty(&q->ring))
596 mask |= EPOLLIN | EPOLLRDNORM;
598 if (sock_writeable(&q->sk) ||
599 (!test_and_set_bit(SOCKWQ_ASYNC_NOSPACE, &q->sock.flags) &&
600 sock_writeable(&q->sk)))
601 mask |= EPOLLOUT | EPOLLWRNORM;
607 static inline struct sk_buff *tap_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 if (len - linear > MAX_SKB_FRAGS * (PAGE_SIZE << PAGE_ALLOC_COSTLY_ORDER))
618 linear = len - MAX_SKB_FRAGS * (PAGE_SIZE << PAGE_ALLOC_COSTLY_ORDER);
619 skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
620 err, PAGE_ALLOC_COSTLY_ORDER);
624 skb_reserve(skb, prepad);
625 skb_put(skb, linear);
626 skb->data_len = len - linear;
627 skb->len += len - linear;
632 /* Neighbour code has some assumptions on HH_DATA_MOD alignment */
633 #define TAP_RESERVE HH_DATA_OFF(ETH_HLEN)
635 /* Get packet from user space buffer */
636 static ssize_t tap_get_user(struct tap_queue *q, void *msg_control,
637 struct iov_iter *from, int noblock)
639 int good_linear = SKB_MAX_HEAD(TAP_RESERVE);
642 unsigned long total_len = iov_iter_count(from);
643 unsigned long len = total_len;
645 struct virtio_net_hdr vnet_hdr = { 0 };
646 int vnet_hdr_len = 0;
649 bool zerocopy = false;
651 enum skb_drop_reason drop_reason;
653 if (q->flags & IFF_VNET_HDR) {
654 vnet_hdr_len = READ_ONCE(q->vnet_hdr_sz);
657 if (len < vnet_hdr_len)
662 if (!copy_from_iter_full(&vnet_hdr, sizeof(vnet_hdr), from))
664 iov_iter_advance(from, vnet_hdr_len - sizeof(vnet_hdr));
665 if ((vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
666 tap16_to_cpu(q, vnet_hdr.csum_start) +
667 tap16_to_cpu(q, vnet_hdr.csum_offset) + 2 >
668 tap16_to_cpu(q, vnet_hdr.hdr_len))
669 vnet_hdr.hdr_len = cpu_to_tap16(q,
670 tap16_to_cpu(q, vnet_hdr.csum_start) +
671 tap16_to_cpu(q, vnet_hdr.csum_offset) + 2);
673 if (tap16_to_cpu(q, vnet_hdr.hdr_len) > len)
678 if (unlikely(len < ETH_HLEN))
681 if (msg_control && sock_flag(&q->sk, SOCK_ZEROCOPY)) {
684 copylen = vnet_hdr.hdr_len ?
685 tap16_to_cpu(q, vnet_hdr.hdr_len) : GOODCOPY_LEN;
686 if (copylen > good_linear)
687 copylen = good_linear;
688 else if (copylen < ETH_HLEN)
692 iov_iter_advance(&i, copylen);
693 if (iov_iter_npages(&i, INT_MAX) <= MAX_SKB_FRAGS)
699 linear = tap16_to_cpu(q, vnet_hdr.hdr_len);
700 if (linear > good_linear)
701 linear = good_linear;
702 else if (linear < ETH_HLEN)
706 skb = tap_alloc_skb(&q->sk, TAP_RESERVE, copylen,
707 linear, noblock, &err);
712 err = zerocopy_sg_from_iter(skb, from);
714 err = skb_copy_datagram_from_iter(skb, 0, from, len);
717 drop_reason = SKB_DROP_REASON_SKB_UCOPY_FAULT;
721 skb_set_network_header(skb, ETH_HLEN);
722 skb_reset_mac_header(skb);
723 skb->protocol = eth_hdr(skb)->h_proto;
726 tap = rcu_dereference(q->tap);
735 err = virtio_net_hdr_to_skb(skb, &vnet_hdr,
736 tap_is_little_endian(q));
739 drop_reason = SKB_DROP_REASON_DEV_HDR;
744 skb_probe_transport_header(skb);
746 /* Move network header to the right position for VLAN tagged packets */
747 if (eth_type_vlan(skb->protocol) &&
748 vlan_get_protocol_and_depth(skb, skb->protocol, &depth) != 0)
749 skb_set_network_header(skb, depth);
751 /* copy skb_ubuf_info for callback when skb has no error */
753 skb_zcopy_init(skb, msg_control);
754 } else if (msg_control) {
755 struct ubuf_info *uarg = msg_control;
756 uarg->callback(NULL, uarg, false);
764 kfree_skb_reason(skb, drop_reason);
768 tap = rcu_dereference(q->tap);
769 if (tap && tap->count_tx_dropped)
770 tap->count_tx_dropped(tap);
776 static ssize_t tap_write_iter(struct kiocb *iocb, struct iov_iter *from)
778 struct file *file = iocb->ki_filp;
779 struct tap_queue *q = file->private_data;
782 if ((file->f_flags & O_NONBLOCK) || (iocb->ki_flags & IOCB_NOWAIT))
785 return tap_get_user(q, NULL, from, noblock);
788 /* Put packet to the user space buffer */
789 static ssize_t tap_put_user(struct tap_queue *q,
790 const struct sk_buff *skb,
791 struct iov_iter *iter)
794 int vnet_hdr_len = 0;
798 if (q->flags & IFF_VNET_HDR) {
799 int vlan_hlen = skb_vlan_tag_present(skb) ? VLAN_HLEN : 0;
800 struct virtio_net_hdr vnet_hdr;
802 vnet_hdr_len = READ_ONCE(q->vnet_hdr_sz);
803 if (iov_iter_count(iter) < vnet_hdr_len)
806 if (virtio_net_hdr_from_skb(skb, &vnet_hdr,
807 tap_is_little_endian(q), true,
811 if (copy_to_iter(&vnet_hdr, sizeof(vnet_hdr), iter) !=
815 iov_iter_advance(iter, vnet_hdr_len - sizeof(vnet_hdr));
817 total = vnet_hdr_len;
820 if (skb_vlan_tag_present(skb)) {
825 veth.h_vlan_proto = skb->vlan_proto;
826 veth.h_vlan_TCI = htons(skb_vlan_tag_get(skb));
828 vlan_offset = offsetof(struct vlan_ethhdr, h_vlan_proto);
831 ret = skb_copy_datagram_iter(skb, 0, iter, vlan_offset);
832 if (ret || !iov_iter_count(iter))
835 ret = copy_to_iter(&veth, sizeof(veth), iter);
836 if (ret != sizeof(veth) || !iov_iter_count(iter))
840 ret = skb_copy_datagram_iter(skb, vlan_offset, iter,
841 skb->len - vlan_offset);
844 return ret ? ret : total;
847 static ssize_t tap_do_read(struct tap_queue *q,
849 int noblock, struct sk_buff *skb)
854 if (!iov_iter_count(to)) {
864 prepare_to_wait(sk_sleep(&q->sk), &wait,
867 /* Read frames from the queue */
868 skb = ptr_ring_consume(&q->ring);
875 if (signal_pending(current)) {
879 /* Nothing to read, let's sleep */
883 finish_wait(sk_sleep(&q->sk), &wait);
887 ret = tap_put_user(q, skb, to);
888 if (unlikely(ret < 0))
896 static ssize_t tap_read_iter(struct kiocb *iocb, struct iov_iter *to)
898 struct file *file = iocb->ki_filp;
899 struct tap_queue *q = file->private_data;
900 ssize_t len = iov_iter_count(to), ret;
903 if ((file->f_flags & O_NONBLOCK) || (iocb->ki_flags & IOCB_NOWAIT))
906 ret = tap_do_read(q, to, noblock, NULL);
907 ret = min_t(ssize_t, ret, len);
913 static struct tap_dev *tap_get_tap_dev(struct tap_queue *q)
918 tap = rtnl_dereference(q->tap);
925 static void tap_put_tap_dev(struct tap_dev *tap)
930 static int tap_ioctl_set_queue(struct file *file, unsigned int flags)
932 struct tap_queue *q = file->private_data;
936 tap = tap_get_tap_dev(q);
940 if (flags & IFF_ATTACH_QUEUE)
941 ret = tap_enable_queue(tap, file, q);
942 else if (flags & IFF_DETACH_QUEUE)
943 ret = tap_disable_queue(q);
947 tap_put_tap_dev(tap);
951 static int set_offload(struct tap_queue *q, unsigned long arg)
954 netdev_features_t features;
955 netdev_features_t feature_mask = 0;
957 tap = rtnl_dereference(q->tap);
961 features = tap->dev->features;
963 if (arg & TUN_F_CSUM) {
964 feature_mask = NETIF_F_HW_CSUM;
966 if (arg & (TUN_F_TSO4 | TUN_F_TSO6)) {
967 if (arg & TUN_F_TSO_ECN)
968 feature_mask |= NETIF_F_TSO_ECN;
969 if (arg & TUN_F_TSO4)
970 feature_mask |= NETIF_F_TSO;
971 if (arg & TUN_F_TSO6)
972 feature_mask |= NETIF_F_TSO6;
975 /* TODO: for now USO4 and USO6 should work simultaneously */
976 if ((arg & (TUN_F_USO4 | TUN_F_USO6)) == (TUN_F_USO4 | TUN_F_USO6))
977 features |= NETIF_F_GSO_UDP_L4;
980 /* tun/tap driver inverts the usage for TSO offloads, where
981 * setting the TSO bit means that the userspace wants to
982 * accept TSO frames and turning it off means that user space
983 * does not support TSO.
984 * For tap, we have to invert it to mean the same thing.
985 * When user space turns off TSO, we turn off GSO/LRO so that
986 * user-space will not receive TSO frames.
988 if (feature_mask & (NETIF_F_TSO | NETIF_F_TSO6) ||
989 (feature_mask & (TUN_F_USO4 | TUN_F_USO6)) == (TUN_F_USO4 | TUN_F_USO6))
990 features |= RX_OFFLOADS;
992 features &= ~RX_OFFLOADS;
994 /* tap_features are the same as features on tun/tap and
995 * reflect user expectations.
997 tap->tap_features = feature_mask;
998 if (tap->update_features)
999 tap->update_features(tap, features);
1005 * provide compatibility with generic tun/tap interface
1007 static long tap_ioctl(struct file *file, unsigned int cmd,
1010 struct tap_queue *q = file->private_data;
1011 struct tap_dev *tap;
1012 void __user *argp = (void __user *)arg;
1013 struct ifreq __user *ifr = argp;
1014 unsigned int __user *up = argp;
1016 int __user *sp = argp;
1023 /* ignore the name, just look at flags */
1024 if (get_user(u, &ifr->ifr_flags))
1028 if ((u & ~TAP_IFFEATURES) != (IFF_NO_PI | IFF_TAP))
1031 q->flags = (q->flags & ~TAP_IFFEATURES) | u;
1037 tap = tap_get_tap_dev(q);
1045 if (copy_to_user(&ifr->ifr_name, tap->dev->name, IFNAMSIZ) ||
1046 put_user(u, &ifr->ifr_flags))
1048 tap_put_tap_dev(tap);
1053 if (get_user(u, &ifr->ifr_flags))
1056 ret = tap_ioctl_set_queue(file, u);
1060 case TUNGETFEATURES:
1061 if (put_user(IFF_TAP | IFF_NO_PI | TAP_IFFEATURES, up))
1066 if (get_user(s, sp))
1071 q->sk.sk_sndbuf = s;
1074 case TUNGETVNETHDRSZ:
1076 if (put_user(s, sp))
1080 case TUNSETVNETHDRSZ:
1081 if (get_user(s, sp))
1083 if (s < (int)sizeof(struct virtio_net_hdr))
1090 s = !!(q->flags & TAP_VNET_LE);
1091 if (put_user(s, sp))
1096 if (get_user(s, sp))
1099 q->flags |= TAP_VNET_LE;
1101 q->flags &= ~TAP_VNET_LE;
1105 return tap_get_vnet_be(q, sp);
1108 return tap_set_vnet_be(q, sp);
1111 /* let the user check for future flags */
1112 if (arg & ~(TUN_F_CSUM | TUN_F_TSO4 | TUN_F_TSO6 |
1113 TUN_F_TSO_ECN | TUN_F_UFO |
1114 TUN_F_USO4 | TUN_F_USO6))
1118 ret = set_offload(q, arg);
1124 tap = tap_get_tap_dev(q);
1130 dev_get_mac_address(&sa, dev_net(tap->dev), tap->dev->name);
1131 if (copy_to_user(&ifr->ifr_name, tap->dev->name, IFNAMSIZ) ||
1132 copy_to_user(&ifr->ifr_hwaddr, &sa, sizeof(sa)))
1134 tap_put_tap_dev(tap);
1139 if (copy_from_user(&sa, &ifr->ifr_hwaddr, sizeof(sa)))
1142 tap = tap_get_tap_dev(q);
1147 ret = dev_set_mac_address_user(tap->dev, &sa, NULL);
1148 tap_put_tap_dev(tap);
1157 static const struct file_operations tap_fops = {
1158 .owner = THIS_MODULE,
1160 .release = tap_release,
1161 .read_iter = tap_read_iter,
1162 .write_iter = tap_write_iter,
1164 .llseek = no_llseek,
1165 .unlocked_ioctl = tap_ioctl,
1166 .compat_ioctl = compat_ptr_ioctl,
1169 static int tap_get_user_xdp(struct tap_queue *q, struct xdp_buff *xdp)
1171 struct tun_xdp_hdr *hdr = xdp->data_hard_start;
1172 struct virtio_net_hdr *gso = &hdr->gso;
1173 int buflen = hdr->buflen;
1174 int vnet_hdr_len = 0;
1175 struct tap_dev *tap;
1176 struct sk_buff *skb;
1179 if (q->flags & IFF_VNET_HDR)
1180 vnet_hdr_len = READ_ONCE(q->vnet_hdr_sz);
1182 skb = build_skb(xdp->data_hard_start, buflen);
1188 skb_reserve(skb, xdp->data - xdp->data_hard_start);
1189 skb_put(skb, xdp->data_end - xdp->data);
1191 skb_set_network_header(skb, ETH_HLEN);
1192 skb_reset_mac_header(skb);
1193 skb->protocol = eth_hdr(skb)->h_proto;
1196 err = virtio_net_hdr_to_skb(skb, gso, tap_is_little_endian(q));
1201 /* Move network header to the right position for VLAN tagged packets */
1202 if (eth_type_vlan(skb->protocol) &&
1203 vlan_get_protocol_and_depth(skb, skb->protocol, &depth) != 0)
1204 skb_set_network_header(skb, depth);
1207 tap = rcu_dereference(q->tap);
1209 skb->dev = tap->dev;
1210 skb_probe_transport_header(skb);
1211 dev_queue_xmit(skb);
1223 tap = rcu_dereference(q->tap);
1224 if (tap && tap->count_tx_dropped)
1225 tap->count_tx_dropped(tap);
1230 static int tap_sendmsg(struct socket *sock, struct msghdr *m,
1233 struct tap_queue *q = container_of(sock, struct tap_queue, sock);
1234 struct tun_msg_ctl *ctl = m->msg_control;
1235 struct xdp_buff *xdp;
1238 if (m->msg_controllen == sizeof(struct tun_msg_ctl) &&
1239 ctl && ctl->type == TUN_MSG_PTR) {
1240 for (i = 0; i < ctl->num; i++) {
1241 xdp = &((struct xdp_buff *)ctl->ptr)[i];
1242 tap_get_user_xdp(q, xdp);
1247 return tap_get_user(q, ctl ? ctl->ptr : NULL, &m->msg_iter,
1248 m->msg_flags & MSG_DONTWAIT);
1251 static int tap_recvmsg(struct socket *sock, struct msghdr *m,
1252 size_t total_len, int flags)
1254 struct tap_queue *q = container_of(sock, struct tap_queue, sock);
1255 struct sk_buff *skb = m->msg_control;
1257 if (flags & ~(MSG_DONTWAIT|MSG_TRUNC)) {
1261 ret = tap_do_read(q, &m->msg_iter, flags & MSG_DONTWAIT, skb);
1262 if (ret > total_len) {
1263 m->msg_flags |= MSG_TRUNC;
1264 ret = flags & MSG_TRUNC ? ret : total_len;
1269 static int tap_peek_len(struct socket *sock)
1271 struct tap_queue *q = container_of(sock, struct tap_queue,
1273 return PTR_RING_PEEK_CALL(&q->ring, __skb_array_len_with_tag);
1276 /* Ops structure to mimic raw sockets with tun */
1277 static const struct proto_ops tap_socket_ops = {
1278 .sendmsg = tap_sendmsg,
1279 .recvmsg = tap_recvmsg,
1280 .peek_len = tap_peek_len,
1283 /* Get an underlying socket object from tun file. Returns error unless file is
1284 * attached to a device. The returned object works like a packet socket, it
1285 * can be used for sock_sendmsg/sock_recvmsg. The caller is responsible for
1286 * holding a reference to the file for as long as the socket is in use. */
1287 struct socket *tap_get_socket(struct file *file)
1289 struct tap_queue *q;
1290 if (file->f_op != &tap_fops)
1291 return ERR_PTR(-EINVAL);
1292 q = file->private_data;
1294 return ERR_PTR(-EBADFD);
1297 EXPORT_SYMBOL_GPL(tap_get_socket);
1299 struct ptr_ring *tap_get_ptr_ring(struct file *file)
1301 struct tap_queue *q;
1303 if (file->f_op != &tap_fops)
1304 return ERR_PTR(-EINVAL);
1305 q = file->private_data;
1307 return ERR_PTR(-EBADFD);
1310 EXPORT_SYMBOL_GPL(tap_get_ptr_ring);
1312 int tap_queue_resize(struct tap_dev *tap)
1314 struct net_device *dev = tap->dev;
1315 struct tap_queue *q;
1316 struct ptr_ring **rings;
1317 int n = tap->numqueues;
1320 rings = kmalloc_array(n, sizeof(*rings), GFP_KERNEL);
1324 list_for_each_entry(q, &tap->queue_list, next)
1325 rings[i++] = &q->ring;
1327 ret = ptr_ring_resize_multiple(rings, n,
1328 dev->tx_queue_len, GFP_KERNEL,
1329 __skb_array_destroy_skb);
1334 EXPORT_SYMBOL_GPL(tap_queue_resize);
1336 static int tap_list_add(dev_t major, const char *device_name)
1338 struct major_info *tap_major;
1340 tap_major = kzalloc(sizeof(*tap_major), GFP_ATOMIC);
1344 tap_major->major = MAJOR(major);
1346 idr_init(&tap_major->minor_idr);
1347 spin_lock_init(&tap_major->minor_lock);
1349 tap_major->device_name = device_name;
1351 list_add_tail_rcu(&tap_major->next, &major_list);
1355 int tap_create_cdev(struct cdev *tap_cdev, dev_t *tap_major,
1356 const char *device_name, struct module *module)
1360 err = alloc_chrdev_region(tap_major, 0, TAP_NUM_DEVS, device_name);
1364 cdev_init(tap_cdev, &tap_fops);
1365 tap_cdev->owner = module;
1366 err = cdev_add(tap_cdev, *tap_major, TAP_NUM_DEVS);
1370 err = tap_list_add(*tap_major, device_name);
1379 unregister_chrdev_region(*tap_major, TAP_NUM_DEVS);
1383 EXPORT_SYMBOL_GPL(tap_create_cdev);
1385 void tap_destroy_cdev(dev_t major, struct cdev *tap_cdev)
1387 struct major_info *tap_major, *tmp;
1390 unregister_chrdev_region(major, TAP_NUM_DEVS);
1391 list_for_each_entry_safe(tap_major, tmp, &major_list, next) {
1392 if (tap_major->major == MAJOR(major)) {
1393 idr_destroy(&tap_major->minor_idr);
1394 list_del_rcu(&tap_major->next);
1395 kfree_rcu(tap_major, rcu);
1399 EXPORT_SYMBOL_GPL(tap_destroy_cdev);
1401 MODULE_AUTHOR("Arnd Bergmann <arnd@arndb.de>");
1402 MODULE_AUTHOR("Sainath Grandhi <sainath.grandhi@intel.com>");
1403 MODULE_LICENSE("GPL");