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>
21 #include <net/net_namespace.h>
22 #include <net/rtnetlink.h>
24 #include <linux/virtio_net.h>
25 #include <linux/skb_array.h>
27 #define TAP_IFFEATURES (IFF_VNET_HDR | IFF_MULTI_QUEUE)
29 #define TAP_VNET_LE 0x80000000
30 #define TAP_VNET_BE 0x40000000
32 #ifdef CONFIG_TUN_VNET_CROSS_LE
33 static inline bool tap_legacy_is_little_endian(struct tap_queue *q)
35 return q->flags & TAP_VNET_BE ? false :
36 virtio_legacy_is_little_endian();
39 static long tap_get_vnet_be(struct tap_queue *q, int __user *sp)
41 int s = !!(q->flags & TAP_VNET_BE);
49 static long tap_set_vnet_be(struct tap_queue *q, int __user *sp)
57 q->flags |= TAP_VNET_BE;
59 q->flags &= ~TAP_VNET_BE;
64 static inline bool tap_legacy_is_little_endian(struct tap_queue *q)
66 return virtio_legacy_is_little_endian();
69 static long tap_get_vnet_be(struct tap_queue *q, int __user *argp)
74 static long tap_set_vnet_be(struct tap_queue *q, int __user *argp)
78 #endif /* CONFIG_TUN_VNET_CROSS_LE */
80 static inline bool tap_is_little_endian(struct tap_queue *q)
82 return q->flags & TAP_VNET_LE ||
83 tap_legacy_is_little_endian(q);
86 static inline u16 tap16_to_cpu(struct tap_queue *q, __virtio16 val)
88 return __virtio16_to_cpu(tap_is_little_endian(q), val);
91 static inline __virtio16 cpu_to_tap16(struct tap_queue *q, u16 val)
93 return __cpu_to_virtio16(tap_is_little_endian(q), val);
96 static struct proto tap_proto = {
99 .obj_size = sizeof(struct tap_queue),
102 #define TAP_NUM_DEVS (1U << MINORBITS)
104 static LIST_HEAD(major_list);
109 struct idr minor_idr;
110 spinlock_t minor_lock;
111 const char *device_name;
112 struct list_head next;
115 #define GOODCOPY_LEN 128
117 static const struct proto_ops tap_socket_ops;
119 #define RX_OFFLOADS (NETIF_F_GRO | NETIF_F_LRO)
120 #define TAP_FEATURES (NETIF_F_GSO | NETIF_F_SG | NETIF_F_FRAGLIST)
122 static struct tap_dev *tap_dev_get_rcu(const struct net_device *dev)
124 return rcu_dereference(dev->rx_handler_data);
129 * The tap_queue and the macvlan_dev are loosely coupled, the
130 * pointers from one to the other can only be read while rcu_read_lock
133 * Both the file and the macvlan_dev hold a reference on the tap_queue
134 * through sock_hold(&q->sk). When the macvlan_dev goes away first,
135 * q->vlan becomes inaccessible. When the files gets closed,
136 * tap_get_queue() fails.
138 * There may still be references to the struct sock inside of the
139 * queue from outbound SKBs, but these never reference back to the
140 * file or the dev. The data structure is freed through __sk_free
141 * when both our references and any pending SKBs are gone.
144 static int tap_enable_queue(struct tap_dev *tap, struct file *file,
155 rcu_assign_pointer(tap->taps[tap->numvtaps], q);
156 q->queue_index = tap->numvtaps;
165 static int tap_set_queue(struct tap_dev *tap, struct file *file,
168 if (tap->numqueues == MAX_TAP_QUEUES)
171 rcu_assign_pointer(q->tap, tap);
172 rcu_assign_pointer(tap->taps[tap->numvtaps], q);
176 q->queue_index = tap->numvtaps;
178 file->private_data = q;
179 list_add_tail(&q->next, &tap->queue_list);
187 static int tap_disable_queue(struct tap_queue *q)
190 struct tap_queue *nq;
196 tap = rtnl_dereference(q->tap);
199 int index = q->queue_index;
200 BUG_ON(index >= tap->numvtaps);
201 nq = rtnl_dereference(tap->taps[tap->numvtaps - 1]);
202 nq->queue_index = index;
204 rcu_assign_pointer(tap->taps[index], nq);
205 RCU_INIT_POINTER(tap->taps[tap->numvtaps - 1], NULL);
215 * The file owning the queue got closed, give up both
216 * the reference that the files holds as well as the
217 * one from the macvlan_dev if that still exists.
219 * Using the spinlock makes sure that we don't get
220 * to the queue again after destroying it.
222 static void tap_put_queue(struct tap_queue *q)
227 tap = rtnl_dereference(q->tap);
231 BUG_ON(tap_disable_queue(q));
234 RCU_INIT_POINTER(q->tap, NULL);
236 list_del_init(&q->next);
246 * Select a queue based on the rxq of the device on which this packet
247 * arrived. If the incoming device is not mq, calculate a flow hash
248 * to select a queue. If all fails, find the first available queue.
249 * Cache vlan->numvtaps since it can become zero during the execution
252 static struct tap_queue *tap_get_queue(struct tap_dev *tap,
255 struct tap_queue *queue = NULL;
256 /* Access to taps array is protected by rcu, but access to numvtaps
257 * isn't. Below we use it to lookup a queue, but treat it as a hint
258 * and validate that the result isn't NULL - in case we are
259 * racing against queue removal.
261 int numvtaps = READ_ONCE(tap->numvtaps);
270 /* Check if we can use flow to select a queue */
271 rxq = skb_get_hash(skb);
273 queue = rcu_dereference(tap->taps[rxq % numvtaps]);
277 if (likely(skb_rx_queue_recorded(skb))) {
278 rxq = skb_get_rx_queue(skb);
280 while (unlikely(rxq >= numvtaps))
283 queue = rcu_dereference(tap->taps[rxq]);
288 queue = rcu_dereference(tap->taps[0]);
294 * The net_device is going away, give up the reference
295 * that it holds on all queues and safely set the pointer
296 * from the queues to NULL.
298 void tap_del_queues(struct tap_dev *tap)
300 struct tap_queue *q, *tmp;
303 list_for_each_entry_safe(q, tmp, &tap->queue_list, next) {
304 list_del_init(&q->next);
305 RCU_INIT_POINTER(q->tap, NULL);
311 BUG_ON(tap->numvtaps);
312 BUG_ON(tap->numqueues);
313 /* guarantee that any future tap_set_queue will fail */
314 tap->numvtaps = MAX_TAP_QUEUES;
316 EXPORT_SYMBOL_GPL(tap_del_queues);
318 rx_handler_result_t tap_handle_frame(struct sk_buff **pskb)
320 struct sk_buff *skb = *pskb;
321 struct net_device *dev = skb->dev;
324 netdev_features_t features = TAP_FEATURES;
325 enum skb_drop_reason drop_reason;
327 tap = tap_dev_get_rcu(dev);
329 return RX_HANDLER_PASS;
331 q = tap_get_queue(tap, skb);
333 return RX_HANDLER_PASS;
335 skb_push(skb, ETH_HLEN);
337 /* Apply the forward feature mask so that we perform segmentation
338 * according to users wishes. This only works if VNET_HDR is
341 if (q->flags & IFF_VNET_HDR)
342 features |= tap->tap_features;
343 if (netif_needs_gso(skb, features)) {
344 struct sk_buff *segs = __skb_gso_segment(skb, features, false);
345 struct sk_buff *next;
348 drop_reason = SKB_DROP_REASON_SKB_GSO_SEG;
353 if (ptr_ring_produce(&q->ring, skb)) {
354 drop_reason = SKB_DROP_REASON_FULL_RING;
361 skb_list_walk_safe(segs, skb, next) {
362 skb_mark_not_on_list(skb);
363 if (ptr_ring_produce(&q->ring, skb)) {
364 drop_reason = SKB_DROP_REASON_FULL_RING;
365 kfree_skb_reason(skb, drop_reason);
366 kfree_skb_list_reason(next, drop_reason);
371 /* If we receive a partial checksum and the tap side
372 * doesn't support checksum offload, compute the checksum.
373 * Note: it doesn't matter which checksum feature to
374 * check, we either support them all or none.
376 if (skb->ip_summed == CHECKSUM_PARTIAL &&
377 !(features & NETIF_F_CSUM_MASK) &&
378 skb_checksum_help(skb)) {
379 drop_reason = SKB_DROP_REASON_SKB_CSUM;
382 if (ptr_ring_produce(&q->ring, skb)) {
383 drop_reason = SKB_DROP_REASON_FULL_RING;
389 wake_up_interruptible_poll(sk_sleep(&q->sk), EPOLLIN | EPOLLRDNORM | EPOLLRDBAND);
390 return RX_HANDLER_CONSUMED;
393 /* Count errors/drops only here, thus don't care about args. */
394 if (tap->count_rx_dropped)
395 tap->count_rx_dropped(tap);
396 kfree_skb_reason(skb, drop_reason);
397 return RX_HANDLER_CONSUMED;
399 EXPORT_SYMBOL_GPL(tap_handle_frame);
401 static struct major_info *tap_get_major(int major)
403 struct major_info *tap_major;
405 list_for_each_entry_rcu(tap_major, &major_list, next) {
406 if (tap_major->major == major)
413 int tap_get_minor(dev_t major, struct tap_dev *tap)
415 int retval = -ENOMEM;
416 struct major_info *tap_major;
419 tap_major = tap_get_major(MAJOR(major));
425 spin_lock(&tap_major->minor_lock);
426 retval = idr_alloc(&tap_major->minor_idr, tap, 1, TAP_NUM_DEVS, GFP_ATOMIC);
429 } else if (retval == -ENOSPC) {
430 netdev_err(tap->dev, "Too many tap devices\n");
433 spin_unlock(&tap_major->minor_lock);
437 return retval < 0 ? retval : 0;
439 EXPORT_SYMBOL_GPL(tap_get_minor);
441 void tap_free_minor(dev_t major, struct tap_dev *tap)
443 struct major_info *tap_major;
446 tap_major = tap_get_major(MAJOR(major));
451 spin_lock(&tap_major->minor_lock);
453 idr_remove(&tap_major->minor_idr, tap->minor);
456 spin_unlock(&tap_major->minor_lock);
461 EXPORT_SYMBOL_GPL(tap_free_minor);
463 static struct tap_dev *dev_get_by_tap_file(int major, int minor)
465 struct net_device *dev = NULL;
467 struct major_info *tap_major;
470 tap_major = tap_get_major(major);
476 spin_lock(&tap_major->minor_lock);
477 tap = idr_find(&tap_major->minor_idr, minor);
482 spin_unlock(&tap_major->minor_lock);
489 static void tap_sock_write_space(struct sock *sk)
491 wait_queue_head_t *wqueue;
493 if (!sock_writeable(sk) ||
494 !test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags))
497 wqueue = sk_sleep(sk);
498 if (wqueue && waitqueue_active(wqueue))
499 wake_up_interruptible_poll(wqueue, EPOLLOUT | EPOLLWRNORM | EPOLLWRBAND);
502 static void tap_sock_destruct(struct sock *sk)
504 struct tap_queue *q = container_of(sk, struct tap_queue, sk);
506 ptr_ring_cleanup(&q->ring, __skb_array_destroy_skb);
509 static int tap_open(struct inode *inode, struct file *file)
511 struct net *net = current->nsproxy->net_ns;
517 tap = dev_get_by_tap_file(imajor(inode), iminor(inode));
522 q = (struct tap_queue *)sk_alloc(net, AF_UNSPEC, GFP_KERNEL,
526 if (ptr_ring_init(&q->ring, tap->dev->tx_queue_len, GFP_KERNEL)) {
531 init_waitqueue_head(&q->sock.wq.wait);
532 q->sock.type = SOCK_RAW;
533 q->sock.state = SS_CONNECTED;
535 q->sock.ops = &tap_socket_ops;
536 sock_init_data(&q->sock, &q->sk);
537 q->sk.sk_write_space = tap_sock_write_space;
538 q->sk.sk_destruct = tap_sock_destruct;
539 q->flags = IFF_VNET_HDR | IFF_NO_PI | IFF_TAP;
540 q->vnet_hdr_sz = sizeof(struct virtio_net_hdr);
543 * so far only KVM virtio_net uses tap, enable zero copy between
544 * guest kernel and host kernel when lower device supports zerocopy
546 * The macvlan supports zerocopy iff the lower device supports zero
547 * copy so we don't have to look at the lower device directly.
549 if ((tap->dev->features & NETIF_F_HIGHDMA) && (tap->dev->features & NETIF_F_SG))
550 sock_set_flag(&q->sk, SOCK_ZEROCOPY);
552 err = tap_set_queue(tap, file, q);
554 /* tap_sock_destruct() will take care of freeing ptr_ring */
573 static int tap_release(struct inode *inode, struct file *file)
575 struct tap_queue *q = file->private_data;
580 static __poll_t tap_poll(struct file *file, poll_table *wait)
582 struct tap_queue *q = file->private_data;
583 __poll_t mask = EPOLLERR;
589 poll_wait(file, &q->sock.wq.wait, wait);
591 if (!ptr_ring_empty(&q->ring))
592 mask |= EPOLLIN | EPOLLRDNORM;
594 if (sock_writeable(&q->sk) ||
595 (!test_and_set_bit(SOCKWQ_ASYNC_NOSPACE, &q->sock.flags) &&
596 sock_writeable(&q->sk)))
597 mask |= EPOLLOUT | EPOLLWRNORM;
603 static inline struct sk_buff *tap_alloc_skb(struct sock *sk, size_t prepad,
604 size_t len, size_t linear,
605 int noblock, int *err)
609 /* Under a page? Don't bother with paged skb. */
610 if (prepad + len < PAGE_SIZE || !linear)
613 skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
618 skb_reserve(skb, prepad);
619 skb_put(skb, linear);
620 skb->data_len = len - linear;
621 skb->len += len - linear;
626 /* Neighbour code has some assumptions on HH_DATA_MOD alignment */
627 #define TAP_RESERVE HH_DATA_OFF(ETH_HLEN)
629 /* Get packet from user space buffer */
630 static ssize_t tap_get_user(struct tap_queue *q, void *msg_control,
631 struct iov_iter *from, int noblock)
633 int good_linear = SKB_MAX_HEAD(TAP_RESERVE);
636 unsigned long total_len = iov_iter_count(from);
637 unsigned long len = total_len;
639 struct virtio_net_hdr vnet_hdr = { 0 };
640 int vnet_hdr_len = 0;
643 bool zerocopy = false;
645 enum skb_drop_reason drop_reason;
647 if (q->flags & IFF_VNET_HDR) {
648 vnet_hdr_len = READ_ONCE(q->vnet_hdr_sz);
651 if (len < vnet_hdr_len)
656 if (!copy_from_iter_full(&vnet_hdr, sizeof(vnet_hdr), from))
658 iov_iter_advance(from, vnet_hdr_len - sizeof(vnet_hdr));
659 if ((vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
660 tap16_to_cpu(q, vnet_hdr.csum_start) +
661 tap16_to_cpu(q, vnet_hdr.csum_offset) + 2 >
662 tap16_to_cpu(q, vnet_hdr.hdr_len))
663 vnet_hdr.hdr_len = cpu_to_tap16(q,
664 tap16_to_cpu(q, vnet_hdr.csum_start) +
665 tap16_to_cpu(q, vnet_hdr.csum_offset) + 2);
667 if (tap16_to_cpu(q, vnet_hdr.hdr_len) > len)
672 if (unlikely(len < ETH_HLEN))
675 if (msg_control && sock_flag(&q->sk, SOCK_ZEROCOPY)) {
678 copylen = vnet_hdr.hdr_len ?
679 tap16_to_cpu(q, vnet_hdr.hdr_len) : GOODCOPY_LEN;
680 if (copylen > good_linear)
681 copylen = good_linear;
682 else if (copylen < ETH_HLEN)
686 iov_iter_advance(&i, copylen);
687 if (iov_iter_npages(&i, INT_MAX) <= MAX_SKB_FRAGS)
693 linear = tap16_to_cpu(q, vnet_hdr.hdr_len);
694 if (linear > good_linear)
695 linear = good_linear;
696 else if (linear < ETH_HLEN)
700 skb = tap_alloc_skb(&q->sk, TAP_RESERVE, copylen,
701 linear, noblock, &err);
706 err = zerocopy_sg_from_iter(skb, from);
708 err = skb_copy_datagram_from_iter(skb, 0, from, len);
711 drop_reason = SKB_DROP_REASON_SKB_UCOPY_FAULT;
715 skb_set_network_header(skb, ETH_HLEN);
716 skb_reset_mac_header(skb);
717 skb->protocol = eth_hdr(skb)->h_proto;
720 tap = rcu_dereference(q->tap);
729 err = virtio_net_hdr_to_skb(skb, &vnet_hdr,
730 tap_is_little_endian(q));
733 drop_reason = SKB_DROP_REASON_DEV_HDR;
738 skb_probe_transport_header(skb);
740 /* Move network header to the right position for VLAN tagged packets */
741 if (eth_type_vlan(skb->protocol) &&
742 __vlan_get_protocol(skb, skb->protocol, &depth) != 0)
743 skb_set_network_header(skb, depth);
745 /* copy skb_ubuf_info for callback when skb has no error */
747 skb_zcopy_init(skb, msg_control);
748 } else if (msg_control) {
749 struct ubuf_info *uarg = msg_control;
750 uarg->callback(NULL, uarg, false);
758 kfree_skb_reason(skb, drop_reason);
762 tap = rcu_dereference(q->tap);
763 if (tap && tap->count_tx_dropped)
764 tap->count_tx_dropped(tap);
770 static ssize_t tap_write_iter(struct kiocb *iocb, struct iov_iter *from)
772 struct file *file = iocb->ki_filp;
773 struct tap_queue *q = file->private_data;
775 return tap_get_user(q, NULL, from, file->f_flags & O_NONBLOCK);
778 /* Put packet to the user space buffer */
779 static ssize_t tap_put_user(struct tap_queue *q,
780 const struct sk_buff *skb,
781 struct iov_iter *iter)
784 int vnet_hdr_len = 0;
788 if (q->flags & IFF_VNET_HDR) {
789 int vlan_hlen = skb_vlan_tag_present(skb) ? VLAN_HLEN : 0;
790 struct virtio_net_hdr vnet_hdr;
792 vnet_hdr_len = READ_ONCE(q->vnet_hdr_sz);
793 if (iov_iter_count(iter) < vnet_hdr_len)
796 if (virtio_net_hdr_from_skb(skb, &vnet_hdr,
797 tap_is_little_endian(q), true,
801 if (copy_to_iter(&vnet_hdr, sizeof(vnet_hdr), iter) !=
805 iov_iter_advance(iter, vnet_hdr_len - sizeof(vnet_hdr));
807 total = vnet_hdr_len;
810 if (skb_vlan_tag_present(skb)) {
815 veth.h_vlan_proto = skb->vlan_proto;
816 veth.h_vlan_TCI = htons(skb_vlan_tag_get(skb));
818 vlan_offset = offsetof(struct vlan_ethhdr, h_vlan_proto);
821 ret = skb_copy_datagram_iter(skb, 0, iter, vlan_offset);
822 if (ret || !iov_iter_count(iter))
825 ret = copy_to_iter(&veth, sizeof(veth), iter);
826 if (ret != sizeof(veth) || !iov_iter_count(iter))
830 ret = skb_copy_datagram_iter(skb, vlan_offset, iter,
831 skb->len - vlan_offset);
834 return ret ? ret : total;
837 static ssize_t tap_do_read(struct tap_queue *q,
839 int noblock, struct sk_buff *skb)
844 if (!iov_iter_count(to)) {
854 prepare_to_wait(sk_sleep(&q->sk), &wait,
857 /* Read frames from the queue */
858 skb = ptr_ring_consume(&q->ring);
865 if (signal_pending(current)) {
869 /* Nothing to read, let's sleep */
873 finish_wait(sk_sleep(&q->sk), &wait);
877 ret = tap_put_user(q, skb, to);
878 if (unlikely(ret < 0))
886 static ssize_t tap_read_iter(struct kiocb *iocb, struct iov_iter *to)
888 struct file *file = iocb->ki_filp;
889 struct tap_queue *q = file->private_data;
890 ssize_t len = iov_iter_count(to), ret;
892 ret = tap_do_read(q, to, file->f_flags & O_NONBLOCK, NULL);
893 ret = min_t(ssize_t, ret, len);
899 static struct tap_dev *tap_get_tap_dev(struct tap_queue *q)
904 tap = rtnl_dereference(q->tap);
911 static void tap_put_tap_dev(struct tap_dev *tap)
916 static int tap_ioctl_set_queue(struct file *file, unsigned int flags)
918 struct tap_queue *q = file->private_data;
922 tap = tap_get_tap_dev(q);
926 if (flags & IFF_ATTACH_QUEUE)
927 ret = tap_enable_queue(tap, file, q);
928 else if (flags & IFF_DETACH_QUEUE)
929 ret = tap_disable_queue(q);
933 tap_put_tap_dev(tap);
937 static int set_offload(struct tap_queue *q, unsigned long arg)
940 netdev_features_t features;
941 netdev_features_t feature_mask = 0;
943 tap = rtnl_dereference(q->tap);
947 features = tap->dev->features;
949 if (arg & TUN_F_CSUM) {
950 feature_mask = NETIF_F_HW_CSUM;
952 if (arg & (TUN_F_TSO4 | TUN_F_TSO6)) {
953 if (arg & TUN_F_TSO_ECN)
954 feature_mask |= NETIF_F_TSO_ECN;
955 if (arg & TUN_F_TSO4)
956 feature_mask |= NETIF_F_TSO;
957 if (arg & TUN_F_TSO6)
958 feature_mask |= NETIF_F_TSO6;
962 /* tun/tap driver inverts the usage for TSO offloads, where
963 * setting the TSO bit means that the userspace wants to
964 * accept TSO frames and turning it off means that user space
965 * does not support TSO.
966 * For tap, we have to invert it to mean the same thing.
967 * When user space turns off TSO, we turn off GSO/LRO so that
968 * user-space will not receive TSO frames.
970 if (feature_mask & (NETIF_F_TSO | NETIF_F_TSO6))
971 features |= RX_OFFLOADS;
973 features &= ~RX_OFFLOADS;
975 /* tap_features are the same as features on tun/tap and
976 * reflect user expectations.
978 tap->tap_features = feature_mask;
979 if (tap->update_features)
980 tap->update_features(tap, features);
986 * provide compatibility with generic tun/tap interface
988 static long tap_ioctl(struct file *file, unsigned int cmd,
991 struct tap_queue *q = file->private_data;
993 void __user *argp = (void __user *)arg;
994 struct ifreq __user *ifr = argp;
995 unsigned int __user *up = argp;
997 int __user *sp = argp;
1004 /* ignore the name, just look at flags */
1005 if (get_user(u, &ifr->ifr_flags))
1009 if ((u & ~TAP_IFFEATURES) != (IFF_NO_PI | IFF_TAP))
1012 q->flags = (q->flags & ~TAP_IFFEATURES) | u;
1018 tap = tap_get_tap_dev(q);
1026 if (copy_to_user(&ifr->ifr_name, tap->dev->name, IFNAMSIZ) ||
1027 put_user(u, &ifr->ifr_flags))
1029 tap_put_tap_dev(tap);
1034 if (get_user(u, &ifr->ifr_flags))
1037 ret = tap_ioctl_set_queue(file, u);
1041 case TUNGETFEATURES:
1042 if (put_user(IFF_TAP | IFF_NO_PI | TAP_IFFEATURES, up))
1047 if (get_user(s, sp))
1052 q->sk.sk_sndbuf = s;
1055 case TUNGETVNETHDRSZ:
1057 if (put_user(s, sp))
1061 case TUNSETVNETHDRSZ:
1062 if (get_user(s, sp))
1064 if (s < (int)sizeof(struct virtio_net_hdr))
1071 s = !!(q->flags & TAP_VNET_LE);
1072 if (put_user(s, sp))
1077 if (get_user(s, sp))
1080 q->flags |= TAP_VNET_LE;
1082 q->flags &= ~TAP_VNET_LE;
1086 return tap_get_vnet_be(q, sp);
1089 return tap_set_vnet_be(q, sp);
1092 /* let the user check for future flags */
1093 if (arg & ~(TUN_F_CSUM | TUN_F_TSO4 | TUN_F_TSO6 |
1094 TUN_F_TSO_ECN | TUN_F_UFO))
1098 ret = set_offload(q, arg);
1104 tap = tap_get_tap_dev(q);
1110 dev_get_mac_address(&sa, dev_net(tap->dev), tap->dev->name);
1111 if (copy_to_user(&ifr->ifr_name, tap->dev->name, IFNAMSIZ) ||
1112 copy_to_user(&ifr->ifr_hwaddr, &sa, sizeof(sa)))
1114 tap_put_tap_dev(tap);
1119 if (copy_from_user(&sa, &ifr->ifr_hwaddr, sizeof(sa)))
1122 tap = tap_get_tap_dev(q);
1127 ret = dev_set_mac_address_user(tap->dev, &sa, NULL);
1128 tap_put_tap_dev(tap);
1137 static const struct file_operations tap_fops = {
1138 .owner = THIS_MODULE,
1140 .release = tap_release,
1141 .read_iter = tap_read_iter,
1142 .write_iter = tap_write_iter,
1144 .llseek = no_llseek,
1145 .unlocked_ioctl = tap_ioctl,
1146 .compat_ioctl = compat_ptr_ioctl,
1149 static int tap_get_user_xdp(struct tap_queue *q, struct xdp_buff *xdp)
1151 struct tun_xdp_hdr *hdr = xdp->data_hard_start;
1152 struct virtio_net_hdr *gso = &hdr->gso;
1153 int buflen = hdr->buflen;
1154 int vnet_hdr_len = 0;
1155 struct tap_dev *tap;
1156 struct sk_buff *skb;
1159 if (q->flags & IFF_VNET_HDR)
1160 vnet_hdr_len = READ_ONCE(q->vnet_hdr_sz);
1162 skb = build_skb(xdp->data_hard_start, buflen);
1168 skb_reserve(skb, xdp->data - xdp->data_hard_start);
1169 skb_put(skb, xdp->data_end - xdp->data);
1171 skb_set_network_header(skb, ETH_HLEN);
1172 skb_reset_mac_header(skb);
1173 skb->protocol = eth_hdr(skb)->h_proto;
1176 err = virtio_net_hdr_to_skb(skb, gso, tap_is_little_endian(q));
1181 /* Move network header to the right position for VLAN tagged packets */
1182 if (eth_type_vlan(skb->protocol) &&
1183 __vlan_get_protocol(skb, skb->protocol, &depth) != 0)
1184 skb_set_network_header(skb, depth);
1187 tap = rcu_dereference(q->tap);
1189 skb->dev = tap->dev;
1190 skb_probe_transport_header(skb);
1191 dev_queue_xmit(skb);
1203 tap = rcu_dereference(q->tap);
1204 if (tap && tap->count_tx_dropped)
1205 tap->count_tx_dropped(tap);
1210 static int tap_sendmsg(struct socket *sock, struct msghdr *m,
1213 struct tap_queue *q = container_of(sock, struct tap_queue, sock);
1214 struct tun_msg_ctl *ctl = m->msg_control;
1215 struct xdp_buff *xdp;
1218 if (m->msg_controllen == sizeof(struct tun_msg_ctl) &&
1219 ctl && ctl->type == TUN_MSG_PTR) {
1220 for (i = 0; i < ctl->num; i++) {
1221 xdp = &((struct xdp_buff *)ctl->ptr)[i];
1222 tap_get_user_xdp(q, xdp);
1227 return tap_get_user(q, ctl ? ctl->ptr : NULL, &m->msg_iter,
1228 m->msg_flags & MSG_DONTWAIT);
1231 static int tap_recvmsg(struct socket *sock, struct msghdr *m,
1232 size_t total_len, int flags)
1234 struct tap_queue *q = container_of(sock, struct tap_queue, sock);
1235 struct sk_buff *skb = m->msg_control;
1237 if (flags & ~(MSG_DONTWAIT|MSG_TRUNC)) {
1241 ret = tap_do_read(q, &m->msg_iter, flags & MSG_DONTWAIT, skb);
1242 if (ret > total_len) {
1243 m->msg_flags |= MSG_TRUNC;
1244 ret = flags & MSG_TRUNC ? ret : total_len;
1249 static int tap_peek_len(struct socket *sock)
1251 struct tap_queue *q = container_of(sock, struct tap_queue,
1253 return PTR_RING_PEEK_CALL(&q->ring, __skb_array_len_with_tag);
1256 /* Ops structure to mimic raw sockets with tun */
1257 static const struct proto_ops tap_socket_ops = {
1258 .sendmsg = tap_sendmsg,
1259 .recvmsg = tap_recvmsg,
1260 .peek_len = tap_peek_len,
1263 /* Get an underlying socket object from tun file. Returns error unless file is
1264 * attached to a device. The returned object works like a packet socket, it
1265 * can be used for sock_sendmsg/sock_recvmsg. The caller is responsible for
1266 * holding a reference to the file for as long as the socket is in use. */
1267 struct socket *tap_get_socket(struct file *file)
1269 struct tap_queue *q;
1270 if (file->f_op != &tap_fops)
1271 return ERR_PTR(-EINVAL);
1272 q = file->private_data;
1274 return ERR_PTR(-EBADFD);
1277 EXPORT_SYMBOL_GPL(tap_get_socket);
1279 struct ptr_ring *tap_get_ptr_ring(struct file *file)
1281 struct tap_queue *q;
1283 if (file->f_op != &tap_fops)
1284 return ERR_PTR(-EINVAL);
1285 q = file->private_data;
1287 return ERR_PTR(-EBADFD);
1290 EXPORT_SYMBOL_GPL(tap_get_ptr_ring);
1292 int tap_queue_resize(struct tap_dev *tap)
1294 struct net_device *dev = tap->dev;
1295 struct tap_queue *q;
1296 struct ptr_ring **rings;
1297 int n = tap->numqueues;
1300 rings = kmalloc_array(n, sizeof(*rings), GFP_KERNEL);
1304 list_for_each_entry(q, &tap->queue_list, next)
1305 rings[i++] = &q->ring;
1307 ret = ptr_ring_resize_multiple(rings, n,
1308 dev->tx_queue_len, GFP_KERNEL,
1309 __skb_array_destroy_skb);
1314 EXPORT_SYMBOL_GPL(tap_queue_resize);
1316 static int tap_list_add(dev_t major, const char *device_name)
1318 struct major_info *tap_major;
1320 tap_major = kzalloc(sizeof(*tap_major), GFP_ATOMIC);
1324 tap_major->major = MAJOR(major);
1326 idr_init(&tap_major->minor_idr);
1327 spin_lock_init(&tap_major->minor_lock);
1329 tap_major->device_name = device_name;
1331 list_add_tail_rcu(&tap_major->next, &major_list);
1335 int tap_create_cdev(struct cdev *tap_cdev, dev_t *tap_major,
1336 const char *device_name, struct module *module)
1340 err = alloc_chrdev_region(tap_major, 0, TAP_NUM_DEVS, device_name);
1344 cdev_init(tap_cdev, &tap_fops);
1345 tap_cdev->owner = module;
1346 err = cdev_add(tap_cdev, *tap_major, TAP_NUM_DEVS);
1350 err = tap_list_add(*tap_major, device_name);
1359 unregister_chrdev_region(*tap_major, TAP_NUM_DEVS);
1363 EXPORT_SYMBOL_GPL(tap_create_cdev);
1365 void tap_destroy_cdev(dev_t major, struct cdev *tap_cdev)
1367 struct major_info *tap_major, *tmp;
1370 unregister_chrdev_region(major, TAP_NUM_DEVS);
1371 list_for_each_entry_safe(tap_major, tmp, &major_list, next) {
1372 if (tap_major->major == MAJOR(major)) {
1373 idr_destroy(&tap_major->minor_idr);
1374 list_del_rcu(&tap_major->next);
1375 kfree_rcu(tap_major, rcu);
1379 EXPORT_SYMBOL_GPL(tap_destroy_cdev);
1381 MODULE_AUTHOR("Arnd Bergmann <arnd@arndb.de>");
1382 MODULE_AUTHOR("Sainath Grandhi <sainath.grandhi@intel.com>");
1383 MODULE_LICENSE("GPL");