1 // SPDX-License-Identifier: GPL-2.0-or-later
3 * TUN - Universal TUN/TAP device driver.
4 * Copyright (C) 1999-2002 Maxim Krasnyansky <maxk@qualcomm.com>
6 * $Id: tun.c,v 1.15 2002/03/01 02:44:24 maxk Exp $
12 * Mike Kershaw <dragorn@kismetwireless.net> 2005/08/14
13 * Add TUNSETLINK ioctl to set the link encapsulation
15 * Mark Smith <markzzzsmith@yahoo.com.au>
16 * Use eth_random_addr() for tap MAC address.
18 * Harald Roelle <harald.roelle@ifi.lmu.de> 2004/04/20
19 * Fixes in packet dropping, queue length setting and queue wakeup.
20 * Increased default tx queue length.
24 * Daniel Podlejski <underley@underley.eu.org>
25 * Modifications for 2.3.99-pre5 kernel.
28 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
30 #define DRV_NAME "tun"
31 #define DRV_VERSION "1.6"
32 #define DRV_DESCRIPTION "Universal TUN/TAP device driver"
33 #define DRV_COPYRIGHT "(C) 1999-2004 Max Krasnyansky <maxk@qualcomm.com>"
35 #include <linux/module.h>
36 #include <linux/errno.h>
37 #include <linux/kernel.h>
38 #include <linux/sched/signal.h>
39 #include <linux/major.h>
40 #include <linux/slab.h>
41 #include <linux/poll.h>
42 #include <linux/fcntl.h>
43 #include <linux/init.h>
44 #include <linux/skbuff.h>
45 #include <linux/netdevice.h>
46 #include <linux/etherdevice.h>
47 #include <linux/miscdevice.h>
48 #include <linux/ethtool.h>
49 #include <linux/rtnetlink.h>
50 #include <linux/compat.h>
52 #include <linux/if_arp.h>
53 #include <linux/if_ether.h>
54 #include <linux/if_tun.h>
55 #include <linux/if_vlan.h>
56 #include <linux/crc32.h>
57 #include <linux/nsproxy.h>
58 #include <linux/virtio_net.h>
59 #include <linux/rcupdate.h>
60 #include <net/net_namespace.h>
61 #include <net/netns/generic.h>
62 #include <net/rtnetlink.h>
65 #include <net/ip_tunnels.h>
66 #include <linux/seq_file.h>
67 #include <linux/uio.h>
68 #include <linux/skb_array.h>
69 #include <linux/bpf.h>
70 #include <linux/bpf_trace.h>
71 #include <linux/mutex.h>
72 #include <linux/ieee802154.h>
73 #include <linux/if_ltalk.h>
74 #include <uapi/linux/if_fddi.h>
75 #include <uapi/linux/if_hippi.h>
76 #include <uapi/linux/if_fc.h>
79 #include <net/6lowpan.h>
81 #include <linux/uaccess.h>
82 #include <linux/proc_fs.h>
84 static void tun_default_link_ksettings(struct net_device *dev,
85 struct ethtool_link_ksettings *cmd);
87 #define TUN_RX_PAD (NET_IP_ALIGN + NET_SKB_PAD)
89 /* TUN device flags */
91 /* IFF_ATTACH_QUEUE is never stored in device flags,
92 * overload it to mean fasync when stored there.
94 #define TUN_FASYNC IFF_ATTACH_QUEUE
95 /* High bits in flags field are unused. */
96 #define TUN_VNET_LE 0x80000000
97 #define TUN_VNET_BE 0x40000000
99 #define TUN_FEATURES (IFF_NO_PI | IFF_ONE_QUEUE | IFF_VNET_HDR | \
100 IFF_MULTI_QUEUE | IFF_NAPI | IFF_NAPI_FRAGS)
102 #define GOODCOPY_LEN 128
104 #define FLT_EXACT_COUNT 8
106 unsigned int count; /* Number of addrs. Zero means disabled */
107 u32 mask[2]; /* Mask of the hashed addrs */
108 unsigned char addr[FLT_EXACT_COUNT][ETH_ALEN];
111 /* MAX_TAP_QUEUES 256 is chosen to allow rx/tx queues to be equal
112 * to max number of VCPUs in guest. */
113 #define MAX_TAP_QUEUES 256
114 #define MAX_TAP_FLOWS 4096
116 #define TUN_FLOW_EXPIRE (3 * HZ)
118 /* A tun_file connects an open character device to a tuntap netdevice. It
119 * also contains all socket related structures (except sock_fprog and tap_filter)
120 * to serve as one transmit queue for tuntap device. The sock_fprog and
121 * tap_filter were kept in tun_struct since they were used for filtering for the
122 * netdevice not for a specific queue (at least I didn't see the requirement for
126 * The tun_file and tun_struct are loosely coupled, the pointer from one to the
127 * other can only be read while rcu_read_lock or rtnl_lock is held.
131 struct socket socket;
132 struct tun_struct __rcu *tun;
133 struct fasync_struct *fasync;
134 /* only used for fasnyc */
138 unsigned int ifindex;
140 struct napi_struct napi;
142 bool napi_frags_enabled;
143 struct mutex napi_mutex; /* Protects access to the above napi */
144 struct list_head next;
145 struct tun_struct *detached;
146 struct ptr_ring tx_ring;
147 struct xdp_rxq_info xdp_rxq;
155 struct tun_flow_entry {
156 struct hlist_node hash_link;
158 struct tun_struct *tun;
163 unsigned long updated ____cacheline_aligned_in_smp;
166 #define TUN_NUM_FLOW_ENTRIES 1024
167 #define TUN_MASK_FLOW_ENTRIES (TUN_NUM_FLOW_ENTRIES - 1)
171 struct bpf_prog *prog;
174 /* Since the socket were moved to tun_file, to preserve the behavior of persist
175 * device, socket filter, sndbuf and vnet header size were restore when the
176 * file were attached to a persist device.
179 struct tun_file __rcu *tfiles[MAX_TAP_QUEUES];
180 unsigned int numqueues;
185 struct net_device *dev;
186 netdev_features_t set_features;
187 #define TUN_USER_FEATURES (NETIF_F_HW_CSUM|NETIF_F_TSO_ECN|NETIF_F_TSO| \
188 NETIF_F_TSO6 | NETIF_F_GSO_UDP_L4)
193 struct tap_filter txflt;
194 struct sock_fprog fprog;
195 /* protected by rtnl lock */
196 bool filter_attached;
199 struct hlist_head flows[TUN_NUM_FLOW_ENTRIES];
200 struct timer_list flow_gc_timer;
201 unsigned long ageing_time;
202 unsigned int numdisabled;
203 struct list_head disabled;
207 atomic_long_t rx_frame_errors;
208 struct bpf_prog __rcu *xdp_prog;
209 struct tun_prog __rcu *steering_prog;
210 struct tun_prog __rcu *filter_prog;
211 struct ethtool_link_ksettings link_ksettings;
222 static void tun_flow_init(struct tun_struct *tun);
223 static void tun_flow_uninit(struct tun_struct *tun);
225 static int tun_napi_receive(struct napi_struct *napi, int budget)
227 struct tun_file *tfile = container_of(napi, struct tun_file, napi);
228 struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
229 struct sk_buff_head process_queue;
233 __skb_queue_head_init(&process_queue);
235 spin_lock(&queue->lock);
236 skb_queue_splice_tail_init(queue, &process_queue);
237 spin_unlock(&queue->lock);
239 while (received < budget && (skb = __skb_dequeue(&process_queue))) {
240 napi_gro_receive(napi, skb);
244 if (!skb_queue_empty(&process_queue)) {
245 spin_lock(&queue->lock);
246 skb_queue_splice(&process_queue, queue);
247 spin_unlock(&queue->lock);
253 static int tun_napi_poll(struct napi_struct *napi, int budget)
255 unsigned int received;
257 received = tun_napi_receive(napi, budget);
259 if (received < budget)
260 napi_complete_done(napi, received);
265 static void tun_napi_init(struct tun_struct *tun, struct tun_file *tfile,
266 bool napi_en, bool napi_frags)
268 tfile->napi_enabled = napi_en;
269 tfile->napi_frags_enabled = napi_en && napi_frags;
271 netif_napi_add_tx(tun->dev, &tfile->napi, tun_napi_poll);
272 napi_enable(&tfile->napi);
276 static void tun_napi_enable(struct tun_file *tfile)
278 if (tfile->napi_enabled)
279 napi_enable(&tfile->napi);
282 static void tun_napi_disable(struct tun_file *tfile)
284 if (tfile->napi_enabled)
285 napi_disable(&tfile->napi);
288 static void tun_napi_del(struct tun_file *tfile)
290 if (tfile->napi_enabled)
291 netif_napi_del(&tfile->napi);
294 static bool tun_napi_frags_enabled(const struct tun_file *tfile)
296 return tfile->napi_frags_enabled;
299 #ifdef CONFIG_TUN_VNET_CROSS_LE
300 static inline bool tun_legacy_is_little_endian(struct tun_struct *tun)
302 return tun->flags & TUN_VNET_BE ? false :
303 virtio_legacy_is_little_endian();
306 static long tun_get_vnet_be(struct tun_struct *tun, int __user *argp)
308 int be = !!(tun->flags & TUN_VNET_BE);
310 if (put_user(be, argp))
316 static long tun_set_vnet_be(struct tun_struct *tun, int __user *argp)
320 if (get_user(be, argp))
324 tun->flags |= TUN_VNET_BE;
326 tun->flags &= ~TUN_VNET_BE;
331 static inline bool tun_legacy_is_little_endian(struct tun_struct *tun)
333 return virtio_legacy_is_little_endian();
336 static long tun_get_vnet_be(struct tun_struct *tun, int __user *argp)
341 static long tun_set_vnet_be(struct tun_struct *tun, int __user *argp)
345 #endif /* CONFIG_TUN_VNET_CROSS_LE */
347 static inline bool tun_is_little_endian(struct tun_struct *tun)
349 return tun->flags & TUN_VNET_LE ||
350 tun_legacy_is_little_endian(tun);
353 static inline u16 tun16_to_cpu(struct tun_struct *tun, __virtio16 val)
355 return __virtio16_to_cpu(tun_is_little_endian(tun), val);
358 static inline __virtio16 cpu_to_tun16(struct tun_struct *tun, u16 val)
360 return __cpu_to_virtio16(tun_is_little_endian(tun), val);
363 static inline u32 tun_hashfn(u32 rxhash)
365 return rxhash & TUN_MASK_FLOW_ENTRIES;
368 static struct tun_flow_entry *tun_flow_find(struct hlist_head *head, u32 rxhash)
370 struct tun_flow_entry *e;
372 hlist_for_each_entry_rcu(e, head, hash_link) {
373 if (e->rxhash == rxhash)
379 static struct tun_flow_entry *tun_flow_create(struct tun_struct *tun,
380 struct hlist_head *head,
381 u32 rxhash, u16 queue_index)
383 struct tun_flow_entry *e = kmalloc(sizeof(*e), GFP_ATOMIC);
386 netif_info(tun, tx_queued, tun->dev,
387 "create flow: hash %u index %u\n",
388 rxhash, queue_index);
389 e->updated = jiffies;
392 e->queue_index = queue_index;
394 hlist_add_head_rcu(&e->hash_link, head);
400 static void tun_flow_delete(struct tun_struct *tun, struct tun_flow_entry *e)
402 netif_info(tun, tx_queued, tun->dev, "delete flow: hash %u index %u\n",
403 e->rxhash, e->queue_index);
404 hlist_del_rcu(&e->hash_link);
409 static void tun_flow_flush(struct tun_struct *tun)
413 spin_lock_bh(&tun->lock);
414 for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
415 struct tun_flow_entry *e;
416 struct hlist_node *n;
418 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link)
419 tun_flow_delete(tun, e);
421 spin_unlock_bh(&tun->lock);
424 static void tun_flow_delete_by_queue(struct tun_struct *tun, u16 queue_index)
428 spin_lock_bh(&tun->lock);
429 for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
430 struct tun_flow_entry *e;
431 struct hlist_node *n;
433 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
434 if (e->queue_index == queue_index)
435 tun_flow_delete(tun, e);
438 spin_unlock_bh(&tun->lock);
441 static void tun_flow_cleanup(struct timer_list *t)
443 struct tun_struct *tun = from_timer(tun, t, flow_gc_timer);
444 unsigned long delay = tun->ageing_time;
445 unsigned long next_timer = jiffies + delay;
446 unsigned long count = 0;
449 spin_lock(&tun->lock);
450 for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
451 struct tun_flow_entry *e;
452 struct hlist_node *n;
454 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
455 unsigned long this_timer;
457 this_timer = e->updated + delay;
458 if (time_before_eq(this_timer, jiffies)) {
459 tun_flow_delete(tun, e);
463 if (time_before(this_timer, next_timer))
464 next_timer = this_timer;
469 mod_timer(&tun->flow_gc_timer, round_jiffies_up(next_timer));
470 spin_unlock(&tun->lock);
473 static void tun_flow_update(struct tun_struct *tun, u32 rxhash,
474 struct tun_file *tfile)
476 struct hlist_head *head;
477 struct tun_flow_entry *e;
478 unsigned long delay = tun->ageing_time;
479 u16 queue_index = tfile->queue_index;
481 head = &tun->flows[tun_hashfn(rxhash)];
485 e = tun_flow_find(head, rxhash);
487 /* TODO: keep queueing to old queue until it's empty? */
488 if (READ_ONCE(e->queue_index) != queue_index)
489 WRITE_ONCE(e->queue_index, queue_index);
490 if (e->updated != jiffies)
491 e->updated = jiffies;
492 sock_rps_record_flow_hash(e->rps_rxhash);
494 spin_lock_bh(&tun->lock);
495 if (!tun_flow_find(head, rxhash) &&
496 tun->flow_count < MAX_TAP_FLOWS)
497 tun_flow_create(tun, head, rxhash, queue_index);
499 if (!timer_pending(&tun->flow_gc_timer))
500 mod_timer(&tun->flow_gc_timer,
501 round_jiffies_up(jiffies + delay));
502 spin_unlock_bh(&tun->lock);
508 /* Save the hash received in the stack receive path and update the
509 * flow_hash table accordingly.
511 static inline void tun_flow_save_rps_rxhash(struct tun_flow_entry *e, u32 hash)
513 if (unlikely(e->rps_rxhash != hash))
514 e->rps_rxhash = hash;
517 /* We try to identify a flow through its rxhash. The reason that
518 * we do not check rxq no. is because some cards(e.g 82599), chooses
519 * the rxq based on the txq where the last packet of the flow comes. As
520 * the userspace application move between processors, we may get a
521 * different rxq no. here.
523 static u16 tun_automq_select_queue(struct tun_struct *tun, struct sk_buff *skb)
525 struct tun_flow_entry *e;
529 numqueues = READ_ONCE(tun->numqueues);
531 txq = __skb_get_hash_symmetric(skb);
532 e = tun_flow_find(&tun->flows[tun_hashfn(txq)], txq);
534 tun_flow_save_rps_rxhash(e, txq);
535 txq = e->queue_index;
537 /* use multiply and shift instead of expensive divide */
538 txq = ((u64)txq * numqueues) >> 32;
544 static u16 tun_ebpf_select_queue(struct tun_struct *tun, struct sk_buff *skb)
546 struct tun_prog *prog;
550 numqueues = READ_ONCE(tun->numqueues);
554 prog = rcu_dereference(tun->steering_prog);
556 ret = bpf_prog_run_clear_cb(prog->prog, skb);
558 return ret % numqueues;
561 static u16 tun_select_queue(struct net_device *dev, struct sk_buff *skb,
562 struct net_device *sb_dev)
564 struct tun_struct *tun = netdev_priv(dev);
568 if (rcu_dereference(tun->steering_prog))
569 ret = tun_ebpf_select_queue(tun, skb);
571 ret = tun_automq_select_queue(tun, skb);
577 static inline bool tun_not_capable(struct tun_struct *tun)
579 const struct cred *cred = current_cred();
580 struct net *net = dev_net(tun->dev);
582 return ((uid_valid(tun->owner) && !uid_eq(cred->euid, tun->owner)) ||
583 (gid_valid(tun->group) && !in_egroup_p(tun->group))) &&
584 !ns_capable(net->user_ns, CAP_NET_ADMIN);
587 static void tun_set_real_num_queues(struct tun_struct *tun)
589 netif_set_real_num_tx_queues(tun->dev, tun->numqueues);
590 netif_set_real_num_rx_queues(tun->dev, tun->numqueues);
593 static void tun_disable_queue(struct tun_struct *tun, struct tun_file *tfile)
595 tfile->detached = tun;
596 list_add_tail(&tfile->next, &tun->disabled);
600 static struct tun_struct *tun_enable_queue(struct tun_file *tfile)
602 struct tun_struct *tun = tfile->detached;
604 tfile->detached = NULL;
605 list_del_init(&tfile->next);
610 void tun_ptr_free(void *ptr)
614 if (tun_is_xdp_frame(ptr)) {
615 struct xdp_frame *xdpf = tun_ptr_to_xdp(ptr);
617 xdp_return_frame(xdpf);
619 __skb_array_destroy_skb(ptr);
622 EXPORT_SYMBOL_GPL(tun_ptr_free);
624 static void tun_queue_purge(struct tun_file *tfile)
628 while ((ptr = ptr_ring_consume(&tfile->tx_ring)) != NULL)
631 skb_queue_purge(&tfile->sk.sk_write_queue);
632 skb_queue_purge(&tfile->sk.sk_error_queue);
635 static void __tun_detach(struct tun_file *tfile, bool clean)
637 struct tun_file *ntfile;
638 struct tun_struct *tun;
640 tun = rtnl_dereference(tfile->tun);
643 if (!tfile->detached)
644 tun_napi_disable(tfile);
648 if (tun && !tfile->detached) {
649 u16 index = tfile->queue_index;
650 BUG_ON(index >= tun->numqueues);
652 rcu_assign_pointer(tun->tfiles[index],
653 tun->tfiles[tun->numqueues - 1]);
654 ntfile = rtnl_dereference(tun->tfiles[index]);
655 ntfile->queue_index = index;
656 ntfile->xdp_rxq.queue_index = index;
657 rcu_assign_pointer(tun->tfiles[tun->numqueues - 1],
662 RCU_INIT_POINTER(tfile->tun, NULL);
663 sock_put(&tfile->sk);
665 tun_disable_queue(tun, tfile);
666 tun_napi_disable(tfile);
670 tun_flow_delete_by_queue(tun, tun->numqueues + 1);
671 /* Drop read queue */
672 tun_queue_purge(tfile);
673 tun_set_real_num_queues(tun);
674 } else if (tfile->detached && clean) {
675 tun = tun_enable_queue(tfile);
676 sock_put(&tfile->sk);
680 if (tun && tun->numqueues == 0 && tun->numdisabled == 0) {
681 netif_carrier_off(tun->dev);
683 if (!(tun->flags & IFF_PERSIST) &&
684 tun->dev->reg_state == NETREG_REGISTERED)
685 unregister_netdevice(tun->dev);
688 xdp_rxq_info_unreg(&tfile->xdp_rxq);
689 ptr_ring_cleanup(&tfile->tx_ring, tun_ptr_free);
693 static void tun_detach(struct tun_file *tfile, bool clean)
695 struct tun_struct *tun;
696 struct net_device *dev;
699 tun = rtnl_dereference(tfile->tun);
700 dev = tun ? tun->dev : NULL;
701 __tun_detach(tfile, clean);
703 netdev_state_change(dev);
707 sock_put(&tfile->sk);
710 static void tun_detach_all(struct net_device *dev)
712 struct tun_struct *tun = netdev_priv(dev);
713 struct tun_file *tfile, *tmp;
714 int i, n = tun->numqueues;
716 for (i = 0; i < n; i++) {
717 tfile = rtnl_dereference(tun->tfiles[i]);
719 tun_napi_disable(tfile);
720 tfile->socket.sk->sk_shutdown = RCV_SHUTDOWN;
721 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
722 RCU_INIT_POINTER(tfile->tun, NULL);
725 list_for_each_entry(tfile, &tun->disabled, next) {
726 tfile->socket.sk->sk_shutdown = RCV_SHUTDOWN;
727 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
728 RCU_INIT_POINTER(tfile->tun, NULL);
730 BUG_ON(tun->numqueues != 0);
733 for (i = 0; i < n; i++) {
734 tfile = rtnl_dereference(tun->tfiles[i]);
736 /* Drop read queue */
737 tun_queue_purge(tfile);
738 xdp_rxq_info_unreg(&tfile->xdp_rxq);
739 sock_put(&tfile->sk);
741 list_for_each_entry_safe(tfile, tmp, &tun->disabled, next) {
743 tun_enable_queue(tfile);
744 tun_queue_purge(tfile);
745 xdp_rxq_info_unreg(&tfile->xdp_rxq);
746 sock_put(&tfile->sk);
748 BUG_ON(tun->numdisabled != 0);
750 if (tun->flags & IFF_PERSIST)
751 module_put(THIS_MODULE);
754 static int tun_attach(struct tun_struct *tun, struct file *file,
755 bool skip_filter, bool napi, bool napi_frags,
758 struct tun_file *tfile = file->private_data;
759 struct net_device *dev = tun->dev;
762 err = security_tun_dev_attach(tfile->socket.sk, tun->security);
767 if (rtnl_dereference(tfile->tun) && !tfile->detached)
771 if (!(tun->flags & IFF_MULTI_QUEUE) && tun->numqueues == 1)
775 if (!tfile->detached &&
776 tun->numqueues + tun->numdisabled == MAX_TAP_QUEUES)
781 /* Re-attach the filter to persist device */
782 if (!skip_filter && (tun->filter_attached == true)) {
783 lock_sock(tfile->socket.sk);
784 err = sk_attach_filter(&tun->fprog, tfile->socket.sk);
785 release_sock(tfile->socket.sk);
790 if (!tfile->detached &&
791 ptr_ring_resize(&tfile->tx_ring, dev->tx_queue_len,
792 GFP_KERNEL, tun_ptr_free)) {
797 tfile->queue_index = tun->numqueues;
798 tfile->socket.sk->sk_shutdown &= ~RCV_SHUTDOWN;
800 if (tfile->detached) {
801 /* Re-attach detached tfile, updating XDP queue_index */
802 WARN_ON(!xdp_rxq_info_is_reg(&tfile->xdp_rxq));
804 if (tfile->xdp_rxq.queue_index != tfile->queue_index)
805 tfile->xdp_rxq.queue_index = tfile->queue_index;
807 /* Setup XDP RX-queue info, for new tfile getting attached */
808 err = xdp_rxq_info_reg(&tfile->xdp_rxq,
809 tun->dev, tfile->queue_index, 0);
812 err = xdp_rxq_info_reg_mem_model(&tfile->xdp_rxq,
813 MEM_TYPE_PAGE_SHARED, NULL);
815 xdp_rxq_info_unreg(&tfile->xdp_rxq);
821 if (tfile->detached) {
822 tun_enable_queue(tfile);
823 tun_napi_enable(tfile);
825 sock_hold(&tfile->sk);
826 tun_napi_init(tun, tfile, napi, napi_frags);
829 if (rtnl_dereference(tun->xdp_prog))
830 sock_set_flag(&tfile->sk, SOCK_XDP);
832 /* device is allowed to go away first, so no need to hold extra
836 /* Publish tfile->tun and tun->tfiles only after we've fully
837 * initialized tfile; otherwise we risk using half-initialized
841 rcu_assign_pointer(tfile->tun, tun);
842 rcu_assign_pointer(tun->tfiles[tun->numqueues], tfile);
844 tun_set_real_num_queues(tun);
849 static struct tun_struct *tun_get(struct tun_file *tfile)
851 struct tun_struct *tun;
854 tun = rcu_dereference(tfile->tun);
862 static void tun_put(struct tun_struct *tun)
868 static void addr_hash_set(u32 *mask, const u8 *addr)
870 int n = ether_crc(ETH_ALEN, addr) >> 26;
871 mask[n >> 5] |= (1 << (n & 31));
874 static unsigned int addr_hash_test(const u32 *mask, const u8 *addr)
876 int n = ether_crc(ETH_ALEN, addr) >> 26;
877 return mask[n >> 5] & (1 << (n & 31));
880 static int update_filter(struct tap_filter *filter, void __user *arg)
882 struct { u8 u[ETH_ALEN]; } *addr;
883 struct tun_filter uf;
884 int err, alen, n, nexact;
886 if (copy_from_user(&uf, arg, sizeof(uf)))
895 alen = ETH_ALEN * uf.count;
896 addr = memdup_user(arg + sizeof(uf), alen);
898 return PTR_ERR(addr);
900 /* The filter is updated without holding any locks. Which is
901 * perfectly safe. We disable it first and in the worst
902 * case we'll accept a few undesired packets. */
906 /* Use first set of addresses as an exact filter */
907 for (n = 0; n < uf.count && n < FLT_EXACT_COUNT; n++)
908 memcpy(filter->addr[n], addr[n].u, ETH_ALEN);
912 /* Remaining multicast addresses are hashed,
913 * unicast will leave the filter disabled. */
914 memset(filter->mask, 0, sizeof(filter->mask));
915 for (; n < uf.count; n++) {
916 if (!is_multicast_ether_addr(addr[n].u)) {
917 err = 0; /* no filter */
920 addr_hash_set(filter->mask, addr[n].u);
923 /* For ALLMULTI just set the mask to all ones.
924 * This overrides the mask populated above. */
925 if ((uf.flags & TUN_FLT_ALLMULTI))
926 memset(filter->mask, ~0, sizeof(filter->mask));
928 /* Now enable the filter */
930 filter->count = nexact;
932 /* Return the number of exact filters */
939 /* Returns: 0 - drop, !=0 - accept */
940 static int run_filter(struct tap_filter *filter, const struct sk_buff *skb)
942 /* Cannot use eth_hdr(skb) here because skb_mac_hdr() is incorrect
944 struct ethhdr *eh = (struct ethhdr *) skb->data;
948 for (i = 0; i < filter->count; i++)
949 if (ether_addr_equal(eh->h_dest, filter->addr[i]))
952 /* Inexact match (multicast only) */
953 if (is_multicast_ether_addr(eh->h_dest))
954 return addr_hash_test(filter->mask, eh->h_dest);
960 * Checks whether the packet is accepted or not.
961 * Returns: 0 - drop, !=0 - accept
963 static int check_filter(struct tap_filter *filter, const struct sk_buff *skb)
968 return run_filter(filter, skb);
971 /* Network device part of the driver */
973 static const struct ethtool_ops tun_ethtool_ops;
975 static int tun_net_init(struct net_device *dev)
977 struct tun_struct *tun = netdev_priv(dev);
978 struct ifreq *ifr = tun->ifr;
981 dev->tstats = netdev_alloc_pcpu_stats(struct pcpu_sw_netstats);
985 spin_lock_init(&tun->lock);
987 err = security_tun_dev_alloc_security(&tun->security);
989 free_percpu(dev->tstats);
995 dev->hw_features = NETIF_F_SG | NETIF_F_FRAGLIST |
996 TUN_USER_FEATURES | NETIF_F_HW_VLAN_CTAG_TX |
997 NETIF_F_HW_VLAN_STAG_TX;
998 dev->features = dev->hw_features | NETIF_F_LLTX;
999 dev->vlan_features = dev->features &
1000 ~(NETIF_F_HW_VLAN_CTAG_TX |
1001 NETIF_F_HW_VLAN_STAG_TX);
1003 tun->flags = (tun->flags & ~TUN_FEATURES) |
1004 (ifr->ifr_flags & TUN_FEATURES);
1006 INIT_LIST_HEAD(&tun->disabled);
1007 err = tun_attach(tun, tun->file, false, ifr->ifr_flags & IFF_NAPI,
1008 ifr->ifr_flags & IFF_NAPI_FRAGS, false);
1010 tun_flow_uninit(tun);
1011 security_tun_dev_free_security(tun->security);
1012 free_percpu(dev->tstats);
1018 /* Net device detach from fd. */
1019 static void tun_net_uninit(struct net_device *dev)
1021 tun_detach_all(dev);
1024 /* Net device open. */
1025 static int tun_net_open(struct net_device *dev)
1027 netif_tx_start_all_queues(dev);
1032 /* Net device close. */
1033 static int tun_net_close(struct net_device *dev)
1035 netif_tx_stop_all_queues(dev);
1039 /* Net device start xmit */
1040 static void tun_automq_xmit(struct tun_struct *tun, struct sk_buff *skb)
1043 if (tun->numqueues == 1 && static_branch_unlikely(&rps_needed)) {
1044 /* Select queue was not called for the skbuff, so we extract the
1045 * RPS hash and save it into the flow_table here.
1047 struct tun_flow_entry *e;
1050 rxhash = __skb_get_hash_symmetric(skb);
1051 e = tun_flow_find(&tun->flows[tun_hashfn(rxhash)], rxhash);
1053 tun_flow_save_rps_rxhash(e, rxhash);
1058 static unsigned int run_ebpf_filter(struct tun_struct *tun,
1059 struct sk_buff *skb,
1062 struct tun_prog *prog = rcu_dereference(tun->filter_prog);
1065 len = bpf_prog_run_clear_cb(prog->prog, skb);
1070 /* Net device start xmit */
1071 static netdev_tx_t tun_net_xmit(struct sk_buff *skb, struct net_device *dev)
1073 struct tun_struct *tun = netdev_priv(dev);
1074 enum skb_drop_reason drop_reason;
1075 int txq = skb->queue_mapping;
1076 struct netdev_queue *queue;
1077 struct tun_file *tfile;
1081 tfile = rcu_dereference(tun->tfiles[txq]);
1083 /* Drop packet if interface is not attached */
1085 drop_reason = SKB_DROP_REASON_DEV_READY;
1089 if (!rcu_dereference(tun->steering_prog))
1090 tun_automq_xmit(tun, skb);
1092 netif_info(tun, tx_queued, tun->dev, "%s %d\n", __func__, skb->len);
1094 /* Drop if the filter does not like it.
1095 * This is a noop if the filter is disabled.
1096 * Filter can be enabled only for the TAP devices. */
1097 if (!check_filter(&tun->txflt, skb)) {
1098 drop_reason = SKB_DROP_REASON_TAP_TXFILTER;
1102 if (tfile->socket.sk->sk_filter &&
1103 sk_filter(tfile->socket.sk, skb)) {
1104 drop_reason = SKB_DROP_REASON_SOCKET_FILTER;
1108 len = run_ebpf_filter(tun, skb, len);
1110 drop_reason = SKB_DROP_REASON_TAP_FILTER;
1114 if (pskb_trim(skb, len)) {
1115 drop_reason = SKB_DROP_REASON_NOMEM;
1119 if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC))) {
1120 drop_reason = SKB_DROP_REASON_SKB_UCOPY_FAULT;
1124 skb_tx_timestamp(skb);
1126 /* Orphan the skb - required as we might hang on to it
1127 * for indefinite time.
1133 if (ptr_ring_produce(&tfile->tx_ring, skb)) {
1134 drop_reason = SKB_DROP_REASON_FULL_RING;
1138 /* NETIF_F_LLTX requires to do our own update of trans_start */
1139 queue = netdev_get_tx_queue(dev, txq);
1140 txq_trans_cond_update(queue);
1142 /* Notify and wake up reader process */
1143 if (tfile->flags & TUN_FASYNC)
1144 kill_fasync(&tfile->fasync, SIGIO, POLL_IN);
1145 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
1148 return NETDEV_TX_OK;
1151 dev_core_stats_tx_dropped_inc(dev);
1153 kfree_skb_reason(skb, drop_reason);
1155 return NET_XMIT_DROP;
1158 static void tun_net_mclist(struct net_device *dev)
1161 * This callback is supposed to deal with mc filter in
1162 * _rx_ path and has nothing to do with the _tx_ path.
1163 * In rx path we always accept everything userspace gives us.
1167 static netdev_features_t tun_net_fix_features(struct net_device *dev,
1168 netdev_features_t features)
1170 struct tun_struct *tun = netdev_priv(dev);
1172 return (features & tun->set_features) | (features & ~TUN_USER_FEATURES);
1175 static void tun_set_headroom(struct net_device *dev, int new_hr)
1177 struct tun_struct *tun = netdev_priv(dev);
1179 if (new_hr < NET_SKB_PAD)
1180 new_hr = NET_SKB_PAD;
1182 tun->align = new_hr;
1186 tun_net_get_stats64(struct net_device *dev, struct rtnl_link_stats64 *stats)
1188 struct tun_struct *tun = netdev_priv(dev);
1190 dev_get_tstats64(dev, stats);
1192 stats->rx_frame_errors +=
1193 (unsigned long)atomic_long_read(&tun->rx_frame_errors);
1196 static int tun_xdp_set(struct net_device *dev, struct bpf_prog *prog,
1197 struct netlink_ext_ack *extack)
1199 struct tun_struct *tun = netdev_priv(dev);
1200 struct tun_file *tfile;
1201 struct bpf_prog *old_prog;
1204 old_prog = rtnl_dereference(tun->xdp_prog);
1205 rcu_assign_pointer(tun->xdp_prog, prog);
1207 bpf_prog_put(old_prog);
1209 for (i = 0; i < tun->numqueues; i++) {
1210 tfile = rtnl_dereference(tun->tfiles[i]);
1212 sock_set_flag(&tfile->sk, SOCK_XDP);
1214 sock_reset_flag(&tfile->sk, SOCK_XDP);
1216 list_for_each_entry(tfile, &tun->disabled, next) {
1218 sock_set_flag(&tfile->sk, SOCK_XDP);
1220 sock_reset_flag(&tfile->sk, SOCK_XDP);
1226 static int tun_xdp(struct net_device *dev, struct netdev_bpf *xdp)
1228 switch (xdp->command) {
1229 case XDP_SETUP_PROG:
1230 return tun_xdp_set(dev, xdp->prog, xdp->extack);
1236 static int tun_net_change_carrier(struct net_device *dev, bool new_carrier)
1239 struct tun_struct *tun = netdev_priv(dev);
1241 if (!tun->numqueues)
1244 netif_carrier_on(dev);
1246 netif_carrier_off(dev);
1251 static const struct net_device_ops tun_netdev_ops = {
1252 .ndo_init = tun_net_init,
1253 .ndo_uninit = tun_net_uninit,
1254 .ndo_open = tun_net_open,
1255 .ndo_stop = tun_net_close,
1256 .ndo_start_xmit = tun_net_xmit,
1257 .ndo_fix_features = tun_net_fix_features,
1258 .ndo_select_queue = tun_select_queue,
1259 .ndo_set_rx_headroom = tun_set_headroom,
1260 .ndo_get_stats64 = tun_net_get_stats64,
1261 .ndo_change_carrier = tun_net_change_carrier,
1264 static void __tun_xdp_flush_tfile(struct tun_file *tfile)
1266 /* Notify and wake up reader process */
1267 if (tfile->flags & TUN_FASYNC)
1268 kill_fasync(&tfile->fasync, SIGIO, POLL_IN);
1269 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
1272 static int tun_xdp_xmit(struct net_device *dev, int n,
1273 struct xdp_frame **frames, u32 flags)
1275 struct tun_struct *tun = netdev_priv(dev);
1276 struct tun_file *tfile;
1281 if (unlikely(flags & ~XDP_XMIT_FLAGS_MASK))
1287 numqueues = READ_ONCE(tun->numqueues);
1290 return -ENXIO; /* Caller will free/return all frames */
1293 tfile = rcu_dereference(tun->tfiles[smp_processor_id() %
1295 if (unlikely(!tfile))
1298 spin_lock(&tfile->tx_ring.producer_lock);
1299 for (i = 0; i < n; i++) {
1300 struct xdp_frame *xdp = frames[i];
1301 /* Encode the XDP flag into lowest bit for consumer to differ
1302 * XDP buffer from sk_buff.
1304 void *frame = tun_xdp_to_ptr(xdp);
1306 if (__ptr_ring_produce(&tfile->tx_ring, frame)) {
1307 dev_core_stats_tx_dropped_inc(dev);
1312 spin_unlock(&tfile->tx_ring.producer_lock);
1314 if (flags & XDP_XMIT_FLUSH)
1315 __tun_xdp_flush_tfile(tfile);
1321 static int tun_xdp_tx(struct net_device *dev, struct xdp_buff *xdp)
1323 struct xdp_frame *frame = xdp_convert_buff_to_frame(xdp);
1326 if (unlikely(!frame))
1329 nxmit = tun_xdp_xmit(dev, 1, &frame, XDP_XMIT_FLUSH);
1331 xdp_return_frame_rx_napi(frame);
1335 static const struct net_device_ops tap_netdev_ops = {
1336 .ndo_init = tun_net_init,
1337 .ndo_uninit = tun_net_uninit,
1338 .ndo_open = tun_net_open,
1339 .ndo_stop = tun_net_close,
1340 .ndo_start_xmit = tun_net_xmit,
1341 .ndo_fix_features = tun_net_fix_features,
1342 .ndo_set_rx_mode = tun_net_mclist,
1343 .ndo_set_mac_address = eth_mac_addr,
1344 .ndo_validate_addr = eth_validate_addr,
1345 .ndo_select_queue = tun_select_queue,
1346 .ndo_features_check = passthru_features_check,
1347 .ndo_set_rx_headroom = tun_set_headroom,
1348 .ndo_get_stats64 = dev_get_tstats64,
1350 .ndo_xdp_xmit = tun_xdp_xmit,
1351 .ndo_change_carrier = tun_net_change_carrier,
1354 static void tun_flow_init(struct tun_struct *tun)
1358 for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++)
1359 INIT_HLIST_HEAD(&tun->flows[i]);
1361 tun->ageing_time = TUN_FLOW_EXPIRE;
1362 timer_setup(&tun->flow_gc_timer, tun_flow_cleanup, 0);
1363 mod_timer(&tun->flow_gc_timer,
1364 round_jiffies_up(jiffies + tun->ageing_time));
1367 static void tun_flow_uninit(struct tun_struct *tun)
1369 del_timer_sync(&tun->flow_gc_timer);
1370 tun_flow_flush(tun);
1374 #define MAX_MTU 65535
1376 /* Initialize net device. */
1377 static void tun_net_initialize(struct net_device *dev)
1379 struct tun_struct *tun = netdev_priv(dev);
1381 switch (tun->flags & TUN_TYPE_MASK) {
1383 dev->netdev_ops = &tun_netdev_ops;
1384 dev->header_ops = &ip_tunnel_header_ops;
1386 /* Point-to-Point TUN Device */
1387 dev->hard_header_len = 0;
1391 /* Zero header length */
1392 dev->type = ARPHRD_NONE;
1393 dev->flags = IFF_POINTOPOINT | IFF_NOARP | IFF_MULTICAST;
1397 dev->netdev_ops = &tap_netdev_ops;
1398 /* Ethernet TAP Device */
1400 dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1401 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
1403 eth_hw_addr_random(dev);
1405 /* Currently tun does not support XDP, only tap does. */
1406 dev->xdp_features = NETDEV_XDP_ACT_BASIC |
1407 NETDEV_XDP_ACT_REDIRECT |
1408 NETDEV_XDP_ACT_NDO_XMIT;
1413 dev->min_mtu = MIN_MTU;
1414 dev->max_mtu = MAX_MTU - dev->hard_header_len;
1417 static bool tun_sock_writeable(struct tun_struct *tun, struct tun_file *tfile)
1419 struct sock *sk = tfile->socket.sk;
1421 return (tun->dev->flags & IFF_UP) && sock_writeable(sk);
1424 /* Character device part */
1427 static __poll_t tun_chr_poll(struct file *file, poll_table *wait)
1429 struct tun_file *tfile = file->private_data;
1430 struct tun_struct *tun = tun_get(tfile);
1437 sk = tfile->socket.sk;
1439 poll_wait(file, sk_sleep(sk), wait);
1441 if (!ptr_ring_empty(&tfile->tx_ring))
1442 mask |= EPOLLIN | EPOLLRDNORM;
1444 /* Make sure SOCKWQ_ASYNC_NOSPACE is set if not writable to
1445 * guarantee EPOLLOUT to be raised by either here or
1446 * tun_sock_write_space(). Then process could get notification
1447 * after it writes to a down device and meets -EIO.
1449 if (tun_sock_writeable(tun, tfile) ||
1450 (!test_and_set_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags) &&
1451 tun_sock_writeable(tun, tfile)))
1452 mask |= EPOLLOUT | EPOLLWRNORM;
1454 if (tun->dev->reg_state != NETREG_REGISTERED)
1461 static struct sk_buff *tun_napi_alloc_frags(struct tun_file *tfile,
1463 const struct iov_iter *it)
1465 struct sk_buff *skb;
1470 if (it->nr_segs > MAX_SKB_FRAGS + 1 ||
1471 len > (ETH_MAX_MTU - NET_SKB_PAD - NET_IP_ALIGN))
1472 return ERR_PTR(-EMSGSIZE);
1475 skb = napi_get_frags(&tfile->napi);
1478 return ERR_PTR(-ENOMEM);
1480 linear = iov_iter_single_seg_count(it);
1481 err = __skb_grow(skb, linear);
1486 skb->data_len = len - linear;
1487 skb->truesize += skb->data_len;
1489 for (i = 1; i < it->nr_segs; i++) {
1490 const struct iovec *iov = iter_iov(it);
1491 size_t fragsz = iov->iov_len;
1495 if (fragsz == 0 || fragsz > PAGE_SIZE) {
1499 frag = netdev_alloc_frag(fragsz);
1504 page = virt_to_head_page(frag);
1505 skb_fill_page_desc(skb, i - 1, page,
1506 frag - page_address(page), fragsz);
1511 /* frees skb and all frags allocated with napi_alloc_frag() */
1512 napi_free_frags(&tfile->napi);
1513 return ERR_PTR(err);
1516 /* prepad is the amount to reserve at front. len is length after that.
1517 * linear is a hint as to how much to copy (usually headers). */
1518 static struct sk_buff *tun_alloc_skb(struct tun_file *tfile,
1519 size_t prepad, size_t len,
1520 size_t linear, int noblock)
1522 struct sock *sk = tfile->socket.sk;
1523 struct sk_buff *skb;
1526 /* Under a page? Don't bother with paged skb. */
1527 if (prepad + len < PAGE_SIZE)
1530 if (len - linear > MAX_SKB_FRAGS * (PAGE_SIZE << PAGE_ALLOC_COSTLY_ORDER))
1531 linear = len - MAX_SKB_FRAGS * (PAGE_SIZE << PAGE_ALLOC_COSTLY_ORDER);
1532 skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
1533 &err, PAGE_ALLOC_COSTLY_ORDER);
1535 return ERR_PTR(err);
1537 skb_reserve(skb, prepad);
1538 skb_put(skb, linear);
1539 skb->data_len = len - linear;
1540 skb->len += len - linear;
1545 static void tun_rx_batched(struct tun_struct *tun, struct tun_file *tfile,
1546 struct sk_buff *skb, int more)
1548 struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
1549 struct sk_buff_head process_queue;
1550 u32 rx_batched = tun->rx_batched;
1553 if (!rx_batched || (!more && skb_queue_empty(queue))) {
1555 skb_record_rx_queue(skb, tfile->queue_index);
1556 netif_receive_skb(skb);
1561 spin_lock(&queue->lock);
1562 if (!more || skb_queue_len(queue) == rx_batched) {
1563 __skb_queue_head_init(&process_queue);
1564 skb_queue_splice_tail_init(queue, &process_queue);
1567 __skb_queue_tail(queue, skb);
1569 spin_unlock(&queue->lock);
1572 struct sk_buff *nskb;
1575 while ((nskb = __skb_dequeue(&process_queue))) {
1576 skb_record_rx_queue(nskb, tfile->queue_index);
1577 netif_receive_skb(nskb);
1579 skb_record_rx_queue(skb, tfile->queue_index);
1580 netif_receive_skb(skb);
1585 static bool tun_can_build_skb(struct tun_struct *tun, struct tun_file *tfile,
1586 int len, int noblock, bool zerocopy)
1588 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
1591 if (tfile->socket.sk->sk_sndbuf != INT_MAX)
1600 if (SKB_DATA_ALIGN(len + TUN_RX_PAD + XDP_PACKET_HEADROOM) +
1601 SKB_DATA_ALIGN(sizeof(struct skb_shared_info)) > PAGE_SIZE)
1607 static struct sk_buff *__tun_build_skb(struct tun_file *tfile,
1608 struct page_frag *alloc_frag, char *buf,
1609 int buflen, int len, int pad)
1611 struct sk_buff *skb = build_skb(buf, buflen);
1614 return ERR_PTR(-ENOMEM);
1616 skb_reserve(skb, pad);
1618 skb_set_owner_w(skb, tfile->socket.sk);
1620 get_page(alloc_frag->page);
1621 alloc_frag->offset += buflen;
1626 static int tun_xdp_act(struct tun_struct *tun, struct bpf_prog *xdp_prog,
1627 struct xdp_buff *xdp, u32 act)
1633 err = xdp_do_redirect(tun->dev, xdp, xdp_prog);
1635 dev_core_stats_rx_dropped_inc(tun->dev);
1638 dev_sw_netstats_rx_add(tun->dev, xdp->data_end - xdp->data);
1641 err = tun_xdp_tx(tun->dev, xdp);
1643 dev_core_stats_rx_dropped_inc(tun->dev);
1646 dev_sw_netstats_rx_add(tun->dev, xdp->data_end - xdp->data);
1651 bpf_warn_invalid_xdp_action(tun->dev, xdp_prog, act);
1654 trace_xdp_exception(tun->dev, xdp_prog, act);
1657 dev_core_stats_rx_dropped_inc(tun->dev);
1664 static struct sk_buff *tun_build_skb(struct tun_struct *tun,
1665 struct tun_file *tfile,
1666 struct iov_iter *from,
1667 struct virtio_net_hdr *hdr,
1668 int len, int *skb_xdp)
1670 struct page_frag *alloc_frag = ¤t->task_frag;
1671 struct bpf_prog *xdp_prog;
1672 int buflen = SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
1675 int pad = TUN_RX_PAD;
1679 xdp_prog = rcu_dereference(tun->xdp_prog);
1681 pad += XDP_PACKET_HEADROOM;
1682 buflen += SKB_DATA_ALIGN(len + pad);
1685 alloc_frag->offset = ALIGN((u64)alloc_frag->offset, SMP_CACHE_BYTES);
1686 if (unlikely(!skb_page_frag_refill(buflen, alloc_frag, GFP_KERNEL)))
1687 return ERR_PTR(-ENOMEM);
1689 buf = (char *)page_address(alloc_frag->page) + alloc_frag->offset;
1690 copied = copy_page_from_iter(alloc_frag->page,
1691 alloc_frag->offset + pad,
1694 return ERR_PTR(-EFAULT);
1696 /* There's a small window that XDP may be set after the check
1697 * of xdp_prog above, this should be rare and for simplicity
1698 * we do XDP on skb in case the headroom is not enough.
1700 if (hdr->gso_type || !xdp_prog) {
1702 return __tun_build_skb(tfile, alloc_frag, buf, buflen, len,
1710 xdp_prog = rcu_dereference(tun->xdp_prog);
1712 struct xdp_buff xdp;
1715 xdp_init_buff(&xdp, buflen, &tfile->xdp_rxq);
1716 xdp_prepare_buff(&xdp, buf, pad, len, false);
1718 act = bpf_prog_run_xdp(xdp_prog, &xdp);
1719 if (act == XDP_REDIRECT || act == XDP_TX) {
1720 get_page(alloc_frag->page);
1721 alloc_frag->offset += buflen;
1723 err = tun_xdp_act(tun, xdp_prog, &xdp, act);
1725 if (act == XDP_REDIRECT || act == XDP_TX)
1726 put_page(alloc_frag->page);
1730 if (err == XDP_REDIRECT)
1732 if (err != XDP_PASS)
1735 pad = xdp.data - xdp.data_hard_start;
1736 len = xdp.data_end - xdp.data;
1741 return __tun_build_skb(tfile, alloc_frag, buf, buflen, len, pad);
1749 /* Get packet from user space buffer */
1750 static ssize_t tun_get_user(struct tun_struct *tun, struct tun_file *tfile,
1751 void *msg_control, struct iov_iter *from,
1752 int noblock, bool more)
1754 struct tun_pi pi = { 0, cpu_to_be16(ETH_P_IP) };
1755 struct sk_buff *skb;
1756 size_t total_len = iov_iter_count(from);
1757 size_t len = total_len, align = tun->align, linear;
1758 struct virtio_net_hdr gso = { 0 };
1761 bool zerocopy = false;
1765 bool frags = tun_napi_frags_enabled(tfile);
1766 enum skb_drop_reason drop_reason = SKB_DROP_REASON_NOT_SPECIFIED;
1768 if (!(tun->flags & IFF_NO_PI)) {
1769 if (len < sizeof(pi))
1773 if (!copy_from_iter_full(&pi, sizeof(pi), from))
1777 if (tun->flags & IFF_VNET_HDR) {
1778 int vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
1780 if (len < vnet_hdr_sz)
1784 if (!copy_from_iter_full(&gso, sizeof(gso), from))
1787 if ((gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
1788 tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2 > tun16_to_cpu(tun, gso.hdr_len))
1789 gso.hdr_len = cpu_to_tun16(tun, tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2);
1791 if (tun16_to_cpu(tun, gso.hdr_len) > len)
1793 iov_iter_advance(from, vnet_hdr_sz - sizeof(gso));
1796 if ((tun->flags & TUN_TYPE_MASK) == IFF_TAP) {
1797 align += NET_IP_ALIGN;
1798 if (unlikely(len < ETH_HLEN ||
1799 (gso.hdr_len && tun16_to_cpu(tun, gso.hdr_len) < ETH_HLEN)))
1803 good_linear = SKB_MAX_HEAD(align);
1806 struct iov_iter i = *from;
1808 /* There are 256 bytes to be copied in skb, so there is
1809 * enough room for skb expand head in case it is used.
1810 * The rest of the buffer is mapped from userspace.
1812 copylen = gso.hdr_len ? tun16_to_cpu(tun, gso.hdr_len) : GOODCOPY_LEN;
1813 if (copylen > good_linear)
1814 copylen = good_linear;
1816 iov_iter_advance(&i, copylen);
1817 if (iov_iter_npages(&i, INT_MAX) <= MAX_SKB_FRAGS)
1821 if (!frags && tun_can_build_skb(tun, tfile, len, noblock, zerocopy)) {
1822 /* For the packet that is not easy to be processed
1823 * (e.g gso or jumbo packet), we will do it at after
1824 * skb was created with generic XDP routine.
1826 skb = tun_build_skb(tun, tfile, from, &gso, len, &skb_xdp);
1827 err = PTR_ERR_OR_ZERO(skb);
1835 if (tun16_to_cpu(tun, gso.hdr_len) > good_linear)
1836 linear = good_linear;
1838 linear = tun16_to_cpu(tun, gso.hdr_len);
1842 mutex_lock(&tfile->napi_mutex);
1843 skb = tun_napi_alloc_frags(tfile, copylen, from);
1844 /* tun_napi_alloc_frags() enforces a layout for the skb.
1845 * If zerocopy is enabled, then this layout will be
1846 * overwritten by zerocopy_sg_from_iter().
1851 linear = min_t(size_t, good_linear, copylen);
1853 skb = tun_alloc_skb(tfile, align, copylen, linear,
1857 err = PTR_ERR_OR_ZERO(skb);
1862 err = zerocopy_sg_from_iter(skb, from);
1864 err = skb_copy_datagram_from_iter(skb, 0, from, len);
1868 drop_reason = SKB_DROP_REASON_SKB_UCOPY_FAULT;
1873 if (virtio_net_hdr_to_skb(skb, &gso, tun_is_little_endian(tun))) {
1874 atomic_long_inc(&tun->rx_frame_errors);
1879 switch (tun->flags & TUN_TYPE_MASK) {
1881 if (tun->flags & IFF_NO_PI) {
1882 u8 ip_version = skb->len ? (skb->data[0] >> 4) : 0;
1884 switch (ip_version) {
1886 pi.proto = htons(ETH_P_IP);
1889 pi.proto = htons(ETH_P_IPV6);
1897 skb_reset_mac_header(skb);
1898 skb->protocol = pi.proto;
1899 skb->dev = tun->dev;
1902 if (frags && !pskb_may_pull(skb, ETH_HLEN)) {
1904 drop_reason = SKB_DROP_REASON_HDR_TRUNC;
1907 skb->protocol = eth_type_trans(skb, tun->dev);
1911 /* copy skb_ubuf_info for callback when skb has no error */
1913 skb_zcopy_init(skb, msg_control);
1914 } else if (msg_control) {
1915 struct ubuf_info *uarg = msg_control;
1916 uarg->callback(NULL, uarg, false);
1919 skb_reset_network_header(skb);
1920 skb_probe_transport_header(skb);
1921 skb_record_rx_queue(skb, tfile->queue_index);
1924 struct bpf_prog *xdp_prog;
1929 xdp_prog = rcu_dereference(tun->xdp_prog);
1931 ret = do_xdp_generic(xdp_prog, skb);
1932 if (ret != XDP_PASS) {
1942 /* Compute the costly rx hash only if needed for flow updates.
1943 * We may get a very small possibility of OOO during switching, not
1944 * worth to optimize.
1946 if (!rcu_access_pointer(tun->steering_prog) && tun->numqueues > 1 &&
1948 rxhash = __skb_get_hash_symmetric(skb);
1951 if (unlikely(!(tun->dev->flags & IFF_UP))) {
1954 drop_reason = SKB_DROP_REASON_DEV_READY;
1961 /* Exercise flow dissector code path. */
1962 skb_push(skb, ETH_HLEN);
1963 headlen = eth_get_headlen(tun->dev, skb->data,
1966 if (unlikely(headlen > skb_headlen(skb))) {
1969 dev_core_stats_rx_dropped_inc(tun->dev);
1971 napi_free_frags(&tfile->napi);
1973 mutex_unlock(&tfile->napi_mutex);
1977 if (likely(napi_schedule_prep(&tfile->napi))) {
1979 napi_gro_frags(&tfile->napi);
1980 napi_complete(&tfile->napi);
1986 mutex_unlock(&tfile->napi_mutex);
1987 } else if (tfile->napi_enabled) {
1988 struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
1991 spin_lock_bh(&queue->lock);
1993 if (unlikely(tfile->detached)) {
1994 spin_unlock_bh(&queue->lock);
2000 __skb_queue_tail(queue, skb);
2001 queue_len = skb_queue_len(queue);
2002 spin_unlock(&queue->lock);
2004 if (!more || queue_len > NAPI_POLL_WEIGHT)
2005 napi_schedule(&tfile->napi);
2008 } else if (!IS_ENABLED(CONFIG_4KSTACKS)) {
2009 tun_rx_batched(tun, tfile, skb, more);
2016 dev_sw_netstats_rx_add(tun->dev, len);
2020 tun_flow_update(tun, rxhash, tfile);
2026 dev_core_stats_rx_dropped_inc(tun->dev);
2029 if (!IS_ERR_OR_NULL(skb))
2030 kfree_skb_reason(skb, drop_reason);
2034 tfile->napi.skb = NULL;
2035 mutex_unlock(&tfile->napi_mutex);
2038 return err ?: total_len;
2041 static ssize_t tun_chr_write_iter(struct kiocb *iocb, struct iov_iter *from)
2043 struct file *file = iocb->ki_filp;
2044 struct tun_file *tfile = file->private_data;
2045 struct tun_struct *tun = tun_get(tfile);
2052 if ((file->f_flags & O_NONBLOCK) || (iocb->ki_flags & IOCB_NOWAIT))
2055 result = tun_get_user(tun, tfile, NULL, from, noblock, false);
2061 static ssize_t tun_put_user_xdp(struct tun_struct *tun,
2062 struct tun_file *tfile,
2063 struct xdp_frame *xdp_frame,
2064 struct iov_iter *iter)
2066 int vnet_hdr_sz = 0;
2067 size_t size = xdp_frame->len;
2070 if (tun->flags & IFF_VNET_HDR) {
2071 struct virtio_net_hdr gso = { 0 };
2073 vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
2074 if (unlikely(iov_iter_count(iter) < vnet_hdr_sz))
2076 if (unlikely(copy_to_iter(&gso, sizeof(gso), iter) !=
2079 iov_iter_advance(iter, vnet_hdr_sz - sizeof(gso));
2082 ret = copy_to_iter(xdp_frame->data, size, iter) + vnet_hdr_sz;
2085 dev_sw_netstats_tx_add(tun->dev, 1, ret);
2091 /* Put packet to the user space buffer */
2092 static ssize_t tun_put_user(struct tun_struct *tun,
2093 struct tun_file *tfile,
2094 struct sk_buff *skb,
2095 struct iov_iter *iter)
2097 struct tun_pi pi = { 0, skb->protocol };
2099 int vlan_offset = 0;
2101 int vnet_hdr_sz = 0;
2103 if (skb_vlan_tag_present(skb))
2104 vlan_hlen = VLAN_HLEN;
2106 if (tun->flags & IFF_VNET_HDR)
2107 vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
2109 total = skb->len + vlan_hlen + vnet_hdr_sz;
2111 if (!(tun->flags & IFF_NO_PI)) {
2112 if (iov_iter_count(iter) < sizeof(pi))
2115 total += sizeof(pi);
2116 if (iov_iter_count(iter) < total) {
2117 /* Packet will be striped */
2118 pi.flags |= TUN_PKT_STRIP;
2121 if (copy_to_iter(&pi, sizeof(pi), iter) != sizeof(pi))
2126 struct virtio_net_hdr gso;
2128 if (iov_iter_count(iter) < vnet_hdr_sz)
2131 if (virtio_net_hdr_from_skb(skb, &gso,
2132 tun_is_little_endian(tun), true,
2134 struct skb_shared_info *sinfo = skb_shinfo(skb);
2135 pr_err("unexpected GSO type: "
2136 "0x%x, gso_size %d, hdr_len %d\n",
2137 sinfo->gso_type, tun16_to_cpu(tun, gso.gso_size),
2138 tun16_to_cpu(tun, gso.hdr_len));
2139 print_hex_dump(KERN_ERR, "tun: ",
2142 min((int)tun16_to_cpu(tun, gso.hdr_len), 64), true);
2147 if (copy_to_iter(&gso, sizeof(gso), iter) != sizeof(gso))
2150 iov_iter_advance(iter, vnet_hdr_sz - sizeof(gso));
2157 veth.h_vlan_proto = skb->vlan_proto;
2158 veth.h_vlan_TCI = htons(skb_vlan_tag_get(skb));
2160 vlan_offset = offsetof(struct vlan_ethhdr, h_vlan_proto);
2162 ret = skb_copy_datagram_iter(skb, 0, iter, vlan_offset);
2163 if (ret || !iov_iter_count(iter))
2166 ret = copy_to_iter(&veth, sizeof(veth), iter);
2167 if (ret != sizeof(veth) || !iov_iter_count(iter))
2171 skb_copy_datagram_iter(skb, vlan_offset, iter, skb->len - vlan_offset);
2174 /* caller is in process context, */
2176 dev_sw_netstats_tx_add(tun->dev, 1, skb->len + vlan_hlen);
2182 static void *tun_ring_recv(struct tun_file *tfile, int noblock, int *err)
2184 DECLARE_WAITQUEUE(wait, current);
2188 ptr = ptr_ring_consume(&tfile->tx_ring);
2196 add_wait_queue(&tfile->socket.wq.wait, &wait);
2199 set_current_state(TASK_INTERRUPTIBLE);
2200 ptr = ptr_ring_consume(&tfile->tx_ring);
2203 if (signal_pending(current)) {
2204 error = -ERESTARTSYS;
2207 if (tfile->socket.sk->sk_shutdown & RCV_SHUTDOWN) {
2215 __set_current_state(TASK_RUNNING);
2216 remove_wait_queue(&tfile->socket.wq.wait, &wait);
2223 static ssize_t tun_do_read(struct tun_struct *tun, struct tun_file *tfile,
2224 struct iov_iter *to,
2225 int noblock, void *ptr)
2230 if (!iov_iter_count(to)) {
2236 /* Read frames from ring */
2237 ptr = tun_ring_recv(tfile, noblock, &err);
2242 if (tun_is_xdp_frame(ptr)) {
2243 struct xdp_frame *xdpf = tun_ptr_to_xdp(ptr);
2245 ret = tun_put_user_xdp(tun, tfile, xdpf, to);
2246 xdp_return_frame(xdpf);
2248 struct sk_buff *skb = ptr;
2250 ret = tun_put_user(tun, tfile, skb, to);
2251 if (unlikely(ret < 0))
2260 static ssize_t tun_chr_read_iter(struct kiocb *iocb, struct iov_iter *to)
2262 struct file *file = iocb->ki_filp;
2263 struct tun_file *tfile = file->private_data;
2264 struct tun_struct *tun = tun_get(tfile);
2265 ssize_t len = iov_iter_count(to), ret;
2271 if ((file->f_flags & O_NONBLOCK) || (iocb->ki_flags & IOCB_NOWAIT))
2274 ret = tun_do_read(tun, tfile, to, noblock, NULL);
2275 ret = min_t(ssize_t, ret, len);
2282 static void tun_prog_free(struct rcu_head *rcu)
2284 struct tun_prog *prog = container_of(rcu, struct tun_prog, rcu);
2286 bpf_prog_destroy(prog->prog);
2290 static int __tun_set_ebpf(struct tun_struct *tun,
2291 struct tun_prog __rcu **prog_p,
2292 struct bpf_prog *prog)
2294 struct tun_prog *old, *new = NULL;
2297 new = kmalloc(sizeof(*new), GFP_KERNEL);
2303 spin_lock_bh(&tun->lock);
2304 old = rcu_dereference_protected(*prog_p,
2305 lockdep_is_held(&tun->lock));
2306 rcu_assign_pointer(*prog_p, new);
2307 spin_unlock_bh(&tun->lock);
2310 call_rcu(&old->rcu, tun_prog_free);
2315 static void tun_free_netdev(struct net_device *dev)
2317 struct tun_struct *tun = netdev_priv(dev);
2319 BUG_ON(!(list_empty(&tun->disabled)));
2321 free_percpu(dev->tstats);
2322 tun_flow_uninit(tun);
2323 security_tun_dev_free_security(tun->security);
2324 __tun_set_ebpf(tun, &tun->steering_prog, NULL);
2325 __tun_set_ebpf(tun, &tun->filter_prog, NULL);
2328 static void tun_setup(struct net_device *dev)
2330 struct tun_struct *tun = netdev_priv(dev);
2332 tun->owner = INVALID_UID;
2333 tun->group = INVALID_GID;
2334 tun_default_link_ksettings(dev, &tun->link_ksettings);
2336 dev->ethtool_ops = &tun_ethtool_ops;
2337 dev->needs_free_netdev = true;
2338 dev->priv_destructor = tun_free_netdev;
2339 /* We prefer our own queue length */
2340 dev->tx_queue_len = TUN_READQ_SIZE;
2343 /* Trivial set of netlink ops to allow deleting tun or tap
2344 * device with netlink.
2346 static int tun_validate(struct nlattr *tb[], struct nlattr *data[],
2347 struct netlink_ext_ack *extack)
2349 NL_SET_ERR_MSG(extack,
2350 "tun/tap creation via rtnetlink is not supported.");
2354 static size_t tun_get_size(const struct net_device *dev)
2356 BUILD_BUG_ON(sizeof(u32) != sizeof(uid_t));
2357 BUILD_BUG_ON(sizeof(u32) != sizeof(gid_t));
2359 return nla_total_size(sizeof(uid_t)) + /* OWNER */
2360 nla_total_size(sizeof(gid_t)) + /* GROUP */
2361 nla_total_size(sizeof(u8)) + /* TYPE */
2362 nla_total_size(sizeof(u8)) + /* PI */
2363 nla_total_size(sizeof(u8)) + /* VNET_HDR */
2364 nla_total_size(sizeof(u8)) + /* PERSIST */
2365 nla_total_size(sizeof(u8)) + /* MULTI_QUEUE */
2366 nla_total_size(sizeof(u32)) + /* NUM_QUEUES */
2367 nla_total_size(sizeof(u32)) + /* NUM_DISABLED_QUEUES */
2371 static int tun_fill_info(struct sk_buff *skb, const struct net_device *dev)
2373 struct tun_struct *tun = netdev_priv(dev);
2375 if (nla_put_u8(skb, IFLA_TUN_TYPE, tun->flags & TUN_TYPE_MASK))
2376 goto nla_put_failure;
2377 if (uid_valid(tun->owner) &&
2378 nla_put_u32(skb, IFLA_TUN_OWNER,
2379 from_kuid_munged(current_user_ns(), tun->owner)))
2380 goto nla_put_failure;
2381 if (gid_valid(tun->group) &&
2382 nla_put_u32(skb, IFLA_TUN_GROUP,
2383 from_kgid_munged(current_user_ns(), tun->group)))
2384 goto nla_put_failure;
2385 if (nla_put_u8(skb, IFLA_TUN_PI, !(tun->flags & IFF_NO_PI)))
2386 goto nla_put_failure;
2387 if (nla_put_u8(skb, IFLA_TUN_VNET_HDR, !!(tun->flags & IFF_VNET_HDR)))
2388 goto nla_put_failure;
2389 if (nla_put_u8(skb, IFLA_TUN_PERSIST, !!(tun->flags & IFF_PERSIST)))
2390 goto nla_put_failure;
2391 if (nla_put_u8(skb, IFLA_TUN_MULTI_QUEUE,
2392 !!(tun->flags & IFF_MULTI_QUEUE)))
2393 goto nla_put_failure;
2394 if (tun->flags & IFF_MULTI_QUEUE) {
2395 if (nla_put_u32(skb, IFLA_TUN_NUM_QUEUES, tun->numqueues))
2396 goto nla_put_failure;
2397 if (nla_put_u32(skb, IFLA_TUN_NUM_DISABLED_QUEUES,
2399 goto nla_put_failure;
2408 static struct rtnl_link_ops tun_link_ops __read_mostly = {
2410 .priv_size = sizeof(struct tun_struct),
2412 .validate = tun_validate,
2413 .get_size = tun_get_size,
2414 .fill_info = tun_fill_info,
2417 static void tun_sock_write_space(struct sock *sk)
2419 struct tun_file *tfile;
2420 wait_queue_head_t *wqueue;
2422 if (!sock_writeable(sk))
2425 if (!test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags))
2428 wqueue = sk_sleep(sk);
2429 if (wqueue && waitqueue_active(wqueue))
2430 wake_up_interruptible_sync_poll(wqueue, EPOLLOUT |
2431 EPOLLWRNORM | EPOLLWRBAND);
2433 tfile = container_of(sk, struct tun_file, sk);
2434 kill_fasync(&tfile->fasync, SIGIO, POLL_OUT);
2437 static void tun_put_page(struct tun_page *tpage)
2440 __page_frag_cache_drain(tpage->page, tpage->count);
2443 static int tun_xdp_one(struct tun_struct *tun,
2444 struct tun_file *tfile,
2445 struct xdp_buff *xdp, int *flush,
2446 struct tun_page *tpage)
2448 unsigned int datasize = xdp->data_end - xdp->data;
2449 struct tun_xdp_hdr *hdr = xdp->data_hard_start;
2450 struct virtio_net_hdr *gso = &hdr->gso;
2451 struct bpf_prog *xdp_prog;
2452 struct sk_buff *skb = NULL;
2453 struct sk_buff_head *queue;
2454 u32 rxhash = 0, act;
2455 int buflen = hdr->buflen;
2457 bool skb_xdp = false;
2460 xdp_prog = rcu_dereference(tun->xdp_prog);
2462 if (gso->gso_type) {
2467 xdp_init_buff(xdp, buflen, &tfile->xdp_rxq);
2468 xdp_set_data_meta_invalid(xdp);
2470 act = bpf_prog_run_xdp(xdp_prog, xdp);
2471 ret = tun_xdp_act(tun, xdp_prog, xdp, act);
2473 put_page(virt_to_head_page(xdp->data));
2486 page = virt_to_head_page(xdp->data);
2487 if (tpage->page == page) {
2490 tun_put_page(tpage);
2499 skb = build_skb(xdp->data_hard_start, buflen);
2505 skb_reserve(skb, xdp->data - xdp->data_hard_start);
2506 skb_put(skb, xdp->data_end - xdp->data);
2508 if (virtio_net_hdr_to_skb(skb, gso, tun_is_little_endian(tun))) {
2509 atomic_long_inc(&tun->rx_frame_errors);
2515 skb->protocol = eth_type_trans(skb, tun->dev);
2516 skb_reset_network_header(skb);
2517 skb_probe_transport_header(skb);
2518 skb_record_rx_queue(skb, tfile->queue_index);
2521 ret = do_xdp_generic(xdp_prog, skb);
2522 if (ret != XDP_PASS) {
2528 if (!rcu_dereference(tun->steering_prog) && tun->numqueues > 1 &&
2530 rxhash = __skb_get_hash_symmetric(skb);
2532 if (tfile->napi_enabled) {
2533 queue = &tfile->sk.sk_write_queue;
2534 spin_lock(&queue->lock);
2536 if (unlikely(tfile->detached)) {
2537 spin_unlock(&queue->lock);
2542 __skb_queue_tail(queue, skb);
2543 spin_unlock(&queue->lock);
2546 netif_receive_skb(skb);
2550 /* No need to disable preemption here since this function is
2551 * always called with bh disabled
2553 dev_sw_netstats_rx_add(tun->dev, datasize);
2556 tun_flow_update(tun, rxhash, tfile);
2562 static int tun_sendmsg(struct socket *sock, struct msghdr *m, size_t total_len)
2565 struct tun_file *tfile = container_of(sock, struct tun_file, socket);
2566 struct tun_struct *tun = tun_get(tfile);
2567 struct tun_msg_ctl *ctl = m->msg_control;
2568 struct xdp_buff *xdp;
2573 if (m->msg_controllen == sizeof(struct tun_msg_ctl) &&
2574 ctl && ctl->type == TUN_MSG_PTR) {
2575 struct tun_page tpage;
2577 int flush = 0, queued = 0;
2579 memset(&tpage, 0, sizeof(tpage));
2584 for (i = 0; i < n; i++) {
2585 xdp = &((struct xdp_buff *)ctl->ptr)[i];
2586 ret = tun_xdp_one(tun, tfile, xdp, &flush, &tpage);
2594 if (tfile->napi_enabled && queued > 0)
2595 napi_schedule(&tfile->napi);
2600 tun_put_page(&tpage);
2606 ret = tun_get_user(tun, tfile, ctl ? ctl->ptr : NULL, &m->msg_iter,
2607 m->msg_flags & MSG_DONTWAIT,
2608 m->msg_flags & MSG_MORE);
2614 static int tun_recvmsg(struct socket *sock, struct msghdr *m, size_t total_len,
2617 struct tun_file *tfile = container_of(sock, struct tun_file, socket);
2618 struct tun_struct *tun = tun_get(tfile);
2619 void *ptr = m->msg_control;
2627 if (flags & ~(MSG_DONTWAIT|MSG_TRUNC|MSG_ERRQUEUE)) {
2631 if (flags & MSG_ERRQUEUE) {
2632 ret = sock_recv_errqueue(sock->sk, m, total_len,
2633 SOL_PACKET, TUN_TX_TIMESTAMP);
2636 ret = tun_do_read(tun, tfile, &m->msg_iter, flags & MSG_DONTWAIT, ptr);
2637 if (ret > (ssize_t)total_len) {
2638 m->msg_flags |= MSG_TRUNC;
2639 ret = flags & MSG_TRUNC ? ret : total_len;
2652 static int tun_ptr_peek_len(void *ptr)
2655 if (tun_is_xdp_frame(ptr)) {
2656 struct xdp_frame *xdpf = tun_ptr_to_xdp(ptr);
2660 return __skb_array_len_with_tag(ptr);
2666 static int tun_peek_len(struct socket *sock)
2668 struct tun_file *tfile = container_of(sock, struct tun_file, socket);
2669 struct tun_struct *tun;
2672 tun = tun_get(tfile);
2676 ret = PTR_RING_PEEK_CALL(&tfile->tx_ring, tun_ptr_peek_len);
2682 /* Ops structure to mimic raw sockets with tun */
2683 static const struct proto_ops tun_socket_ops = {
2684 .peek_len = tun_peek_len,
2685 .sendmsg = tun_sendmsg,
2686 .recvmsg = tun_recvmsg,
2689 static struct proto tun_proto = {
2691 .owner = THIS_MODULE,
2692 .obj_size = sizeof(struct tun_file),
2695 static int tun_flags(struct tun_struct *tun)
2697 return tun->flags & (TUN_FEATURES | IFF_PERSIST | IFF_TUN | IFF_TAP);
2700 static ssize_t tun_flags_show(struct device *dev, struct device_attribute *attr,
2703 struct tun_struct *tun = netdev_priv(to_net_dev(dev));
2704 return sysfs_emit(buf, "0x%x\n", tun_flags(tun));
2707 static ssize_t owner_show(struct device *dev, struct device_attribute *attr,
2710 struct tun_struct *tun = netdev_priv(to_net_dev(dev));
2711 return uid_valid(tun->owner)?
2712 sysfs_emit(buf, "%u\n",
2713 from_kuid_munged(current_user_ns(), tun->owner)) :
2714 sysfs_emit(buf, "-1\n");
2717 static ssize_t group_show(struct device *dev, struct device_attribute *attr,
2720 struct tun_struct *tun = netdev_priv(to_net_dev(dev));
2721 return gid_valid(tun->group) ?
2722 sysfs_emit(buf, "%u\n",
2723 from_kgid_munged(current_user_ns(), tun->group)) :
2724 sysfs_emit(buf, "-1\n");
2727 static DEVICE_ATTR_RO(tun_flags);
2728 static DEVICE_ATTR_RO(owner);
2729 static DEVICE_ATTR_RO(group);
2731 static struct attribute *tun_dev_attrs[] = {
2732 &dev_attr_tun_flags.attr,
2733 &dev_attr_owner.attr,
2734 &dev_attr_group.attr,
2738 static const struct attribute_group tun_attr_group = {
2739 .attrs = tun_dev_attrs
2742 static int tun_set_iff(struct net *net, struct file *file, struct ifreq *ifr)
2744 struct tun_struct *tun;
2745 struct tun_file *tfile = file->private_data;
2746 struct net_device *dev;
2749 if (tfile->detached)
2752 if ((ifr->ifr_flags & IFF_NAPI_FRAGS)) {
2753 if (!capable(CAP_NET_ADMIN))
2756 if (!(ifr->ifr_flags & IFF_NAPI) ||
2757 (ifr->ifr_flags & TUN_TYPE_MASK) != IFF_TAP)
2761 dev = __dev_get_by_name(net, ifr->ifr_name);
2763 if (ifr->ifr_flags & IFF_TUN_EXCL)
2765 if ((ifr->ifr_flags & IFF_TUN) && dev->netdev_ops == &tun_netdev_ops)
2766 tun = netdev_priv(dev);
2767 else if ((ifr->ifr_flags & IFF_TAP) && dev->netdev_ops == &tap_netdev_ops)
2768 tun = netdev_priv(dev);
2772 if (!!(ifr->ifr_flags & IFF_MULTI_QUEUE) !=
2773 !!(tun->flags & IFF_MULTI_QUEUE))
2776 if (tun_not_capable(tun))
2778 err = security_tun_dev_open(tun->security);
2782 err = tun_attach(tun, file, ifr->ifr_flags & IFF_NOFILTER,
2783 ifr->ifr_flags & IFF_NAPI,
2784 ifr->ifr_flags & IFF_NAPI_FRAGS, true);
2788 if (tun->flags & IFF_MULTI_QUEUE &&
2789 (tun->numqueues + tun->numdisabled > 1)) {
2790 /* One or more queue has already been attached, no need
2791 * to initialize the device again.
2793 netdev_state_change(dev);
2797 tun->flags = (tun->flags & ~TUN_FEATURES) |
2798 (ifr->ifr_flags & TUN_FEATURES);
2800 netdev_state_change(dev);
2803 unsigned long flags = 0;
2804 int queues = ifr->ifr_flags & IFF_MULTI_QUEUE ?
2807 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
2809 err = security_tun_dev_create();
2814 if (ifr->ifr_flags & IFF_TUN) {
2818 } else if (ifr->ifr_flags & IFF_TAP) {
2826 name = ifr->ifr_name;
2828 dev = alloc_netdev_mqs(sizeof(struct tun_struct), name,
2829 NET_NAME_UNKNOWN, tun_setup, queues,
2835 dev_net_set(dev, net);
2836 dev->rtnl_link_ops = &tun_link_ops;
2837 dev->ifindex = tfile->ifindex;
2838 dev->sysfs_groups[0] = &tun_attr_group;
2840 tun = netdev_priv(dev);
2843 tun->txflt.count = 0;
2844 tun->vnet_hdr_sz = sizeof(struct virtio_net_hdr);
2846 tun->align = NET_SKB_PAD;
2847 tun->filter_attached = false;
2848 tun->sndbuf = tfile->socket.sk->sk_sndbuf;
2849 tun->rx_batched = 0;
2850 RCU_INIT_POINTER(tun->steering_prog, NULL);
2855 tun_net_initialize(dev);
2857 err = register_netdevice(tun->dev);
2862 /* free_netdev() won't check refcnt, to avoid race
2863 * with dev_put() we need publish tun after registration.
2865 rcu_assign_pointer(tfile->tun, tun);
2868 if (ifr->ifr_flags & IFF_NO_CARRIER)
2869 netif_carrier_off(tun->dev);
2871 netif_carrier_on(tun->dev);
2873 /* Make sure persistent devices do not get stuck in
2876 if (netif_running(tun->dev))
2877 netif_tx_wake_all_queues(tun->dev);
2879 strcpy(ifr->ifr_name, tun->dev->name);
2883 static void tun_get_iff(struct tun_struct *tun, struct ifreq *ifr)
2885 strcpy(ifr->ifr_name, tun->dev->name);
2887 ifr->ifr_flags = tun_flags(tun);
2891 /* This is like a cut-down ethtool ops, except done via tun fd so no
2892 * privs required. */
2893 static int set_offload(struct tun_struct *tun, unsigned long arg)
2895 netdev_features_t features = 0;
2897 if (arg & TUN_F_CSUM) {
2898 features |= NETIF_F_HW_CSUM;
2901 if (arg & (TUN_F_TSO4|TUN_F_TSO6)) {
2902 if (arg & TUN_F_TSO_ECN) {
2903 features |= NETIF_F_TSO_ECN;
2904 arg &= ~TUN_F_TSO_ECN;
2906 if (arg & TUN_F_TSO4)
2907 features |= NETIF_F_TSO;
2908 if (arg & TUN_F_TSO6)
2909 features |= NETIF_F_TSO6;
2910 arg &= ~(TUN_F_TSO4|TUN_F_TSO6);
2915 /* TODO: for now USO4 and USO6 should work simultaneously */
2916 if (arg & TUN_F_USO4 && arg & TUN_F_USO6) {
2917 features |= NETIF_F_GSO_UDP_L4;
2918 arg &= ~(TUN_F_USO4 | TUN_F_USO6);
2922 /* This gives the user a way to test for new features in future by
2923 * trying to set them. */
2927 tun->set_features = features;
2928 tun->dev->wanted_features &= ~TUN_USER_FEATURES;
2929 tun->dev->wanted_features |= features;
2930 netdev_update_features(tun->dev);
2935 static void tun_detach_filter(struct tun_struct *tun, int n)
2938 struct tun_file *tfile;
2940 for (i = 0; i < n; i++) {
2941 tfile = rtnl_dereference(tun->tfiles[i]);
2942 lock_sock(tfile->socket.sk);
2943 sk_detach_filter(tfile->socket.sk);
2944 release_sock(tfile->socket.sk);
2947 tun->filter_attached = false;
2950 static int tun_attach_filter(struct tun_struct *tun)
2953 struct tun_file *tfile;
2955 for (i = 0; i < tun->numqueues; i++) {
2956 tfile = rtnl_dereference(tun->tfiles[i]);
2957 lock_sock(tfile->socket.sk);
2958 ret = sk_attach_filter(&tun->fprog, tfile->socket.sk);
2959 release_sock(tfile->socket.sk);
2961 tun_detach_filter(tun, i);
2966 tun->filter_attached = true;
2970 static void tun_set_sndbuf(struct tun_struct *tun)
2972 struct tun_file *tfile;
2975 for (i = 0; i < tun->numqueues; i++) {
2976 tfile = rtnl_dereference(tun->tfiles[i]);
2977 tfile->socket.sk->sk_sndbuf = tun->sndbuf;
2981 static int tun_set_queue(struct file *file, struct ifreq *ifr)
2983 struct tun_file *tfile = file->private_data;
2984 struct tun_struct *tun;
2989 if (ifr->ifr_flags & IFF_ATTACH_QUEUE) {
2990 tun = tfile->detached;
2995 ret = security_tun_dev_attach_queue(tun->security);
2998 ret = tun_attach(tun, file, false, tun->flags & IFF_NAPI,
2999 tun->flags & IFF_NAPI_FRAGS, true);
3000 } else if (ifr->ifr_flags & IFF_DETACH_QUEUE) {
3001 tun = rtnl_dereference(tfile->tun);
3002 if (!tun || !(tun->flags & IFF_MULTI_QUEUE) || tfile->detached)
3005 __tun_detach(tfile, false);
3010 netdev_state_change(tun->dev);
3017 static int tun_set_ebpf(struct tun_struct *tun, struct tun_prog __rcu **prog_p,
3020 struct bpf_prog *prog;
3023 if (copy_from_user(&fd, data, sizeof(fd)))
3029 prog = bpf_prog_get_type(fd, BPF_PROG_TYPE_SOCKET_FILTER);
3031 return PTR_ERR(prog);
3034 return __tun_set_ebpf(tun, prog_p, prog);
3037 /* Return correct value for tun->dev->addr_len based on tun->dev->type. */
3038 static unsigned char tun_get_addr_len(unsigned short type)
3042 case ARPHRD_TUNNEL6:
3043 return sizeof(struct in6_addr);
3050 case ARPHRD_IEEE802154:
3051 case ARPHRD_IEEE802154_MONITOR:
3052 return IEEE802154_EXTENDED_ADDR_LEN;
3053 case ARPHRD_PHONET_PIPE:
3057 case ARPHRD_6LOWPAN:
3058 return EUI64_ADDR_LEN;
3063 case ARPHRD_IEEE802:
3066 return ROSE_ADDR_LEN;
3068 return AX25_ADDR_LEN;
3069 case ARPHRD_LOCALTLK:
3076 static long __tun_chr_ioctl(struct file *file, unsigned int cmd,
3077 unsigned long arg, int ifreq_len)
3079 struct tun_file *tfile = file->private_data;
3080 struct net *net = sock_net(&tfile->sk);
3081 struct tun_struct *tun;
3082 void __user* argp = (void __user*)arg;
3083 unsigned int carrier;
3092 bool do_notify = false;
3094 if (cmd == TUNSETIFF || cmd == TUNSETQUEUE ||
3095 (_IOC_TYPE(cmd) == SOCK_IOC_TYPE && cmd != SIOCGSKNS)) {
3096 if (copy_from_user(&ifr, argp, ifreq_len))
3099 memset(&ifr, 0, sizeof(ifr));
3101 if (cmd == TUNGETFEATURES) {
3102 /* Currently this just means: "what IFF flags are valid?".
3103 * This is needed because we never checked for invalid flags on
3106 return put_user(IFF_TUN | IFF_TAP | IFF_NO_CARRIER |
3107 TUN_FEATURES, (unsigned int __user*)argp);
3108 } else if (cmd == TUNSETQUEUE) {
3109 return tun_set_queue(file, &ifr);
3110 } else if (cmd == SIOCGSKNS) {
3111 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
3113 return open_related_ns(&net->ns, get_net_ns);
3118 tun = tun_get(tfile);
3119 if (cmd == TUNSETIFF) {
3124 ifr.ifr_name[IFNAMSIZ-1] = '\0';
3126 ret = tun_set_iff(net, file, &ifr);
3131 if (copy_to_user(argp, &ifr, ifreq_len))
3135 if (cmd == TUNSETIFINDEX) {
3141 if (copy_from_user(&ifindex, argp, sizeof(ifindex)))
3147 tfile->ifindex = ifindex;
3155 netif_info(tun, drv, tun->dev, "tun_chr_ioctl cmd %u\n", cmd);
3157 net = dev_net(tun->dev);
3161 tun_get_iff(tun, &ifr);
3163 if (tfile->detached)
3164 ifr.ifr_flags |= IFF_DETACH_QUEUE;
3165 if (!tfile->socket.sk->sk_filter)
3166 ifr.ifr_flags |= IFF_NOFILTER;
3168 if (copy_to_user(argp, &ifr, ifreq_len))
3173 /* Disable/Enable checksum */
3175 /* [unimplemented] */
3176 netif_info(tun, drv, tun->dev, "ignored: set checksum %s\n",
3177 arg ? "disabled" : "enabled");
3181 /* Disable/Enable persist mode. Keep an extra reference to the
3182 * module to prevent the module being unprobed.
3184 if (arg && !(tun->flags & IFF_PERSIST)) {
3185 tun->flags |= IFF_PERSIST;
3186 __module_get(THIS_MODULE);
3189 if (!arg && (tun->flags & IFF_PERSIST)) {
3190 tun->flags &= ~IFF_PERSIST;
3191 module_put(THIS_MODULE);
3195 netif_info(tun, drv, tun->dev, "persist %s\n",
3196 arg ? "enabled" : "disabled");
3200 /* Set owner of the device */
3201 owner = make_kuid(current_user_ns(), arg);
3202 if (!uid_valid(owner)) {
3208 netif_info(tun, drv, tun->dev, "owner set to %u\n",
3209 from_kuid(&init_user_ns, tun->owner));
3213 /* Set group of the device */
3214 group = make_kgid(current_user_ns(), arg);
3215 if (!gid_valid(group)) {
3221 netif_info(tun, drv, tun->dev, "group set to %u\n",
3222 from_kgid(&init_user_ns, tun->group));
3226 /* Only allow setting the type when the interface is down */
3227 if (tun->dev->flags & IFF_UP) {
3228 netif_info(tun, drv, tun->dev,
3229 "Linktype set failed because interface is up\n");
3232 ret = call_netdevice_notifiers(NETDEV_PRE_TYPE_CHANGE,
3234 ret = notifier_to_errno(ret);
3236 netif_info(tun, drv, tun->dev,
3237 "Refused to change device type\n");
3240 tun->dev->type = (int) arg;
3241 tun->dev->addr_len = tun_get_addr_len(tun->dev->type);
3242 netif_info(tun, drv, tun->dev, "linktype set to %d\n",
3244 call_netdevice_notifiers(NETDEV_POST_TYPE_CHANGE,
3250 tun->msg_enable = (u32)arg;
3254 ret = set_offload(tun, arg);
3257 case TUNSETTXFILTER:
3258 /* Can be set only for TAPs */
3260 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3262 ret = update_filter(&tun->txflt, (void __user *)arg);
3266 /* Get hw address */
3267 dev_get_mac_address(&ifr.ifr_hwaddr, net, tun->dev->name);
3268 if (copy_to_user(argp, &ifr, ifreq_len))
3273 /* Set hw address */
3274 ret = dev_set_mac_address_user(tun->dev, &ifr.ifr_hwaddr, NULL);
3278 sndbuf = tfile->socket.sk->sk_sndbuf;
3279 if (copy_to_user(argp, &sndbuf, sizeof(sndbuf)))
3284 if (copy_from_user(&sndbuf, argp, sizeof(sndbuf))) {
3293 tun->sndbuf = sndbuf;
3294 tun_set_sndbuf(tun);
3297 case TUNGETVNETHDRSZ:
3298 vnet_hdr_sz = tun->vnet_hdr_sz;
3299 if (copy_to_user(argp, &vnet_hdr_sz, sizeof(vnet_hdr_sz)))
3303 case TUNSETVNETHDRSZ:
3304 if (copy_from_user(&vnet_hdr_sz, argp, sizeof(vnet_hdr_sz))) {
3308 if (vnet_hdr_sz < (int)sizeof(struct virtio_net_hdr)) {
3313 tun->vnet_hdr_sz = vnet_hdr_sz;
3317 le = !!(tun->flags & TUN_VNET_LE);
3318 if (put_user(le, (int __user *)argp))
3323 if (get_user(le, (int __user *)argp)) {
3328 tun->flags |= TUN_VNET_LE;
3330 tun->flags &= ~TUN_VNET_LE;
3334 ret = tun_get_vnet_be(tun, argp);
3338 ret = tun_set_vnet_be(tun, argp);
3341 case TUNATTACHFILTER:
3342 /* Can be set only for TAPs */
3344 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3347 if (copy_from_user(&tun->fprog, argp, sizeof(tun->fprog)))
3350 ret = tun_attach_filter(tun);
3353 case TUNDETACHFILTER:
3354 /* Can be set only for TAPs */
3356 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3359 tun_detach_filter(tun, tun->numqueues);
3364 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3367 if (copy_to_user(argp, &tun->fprog, sizeof(tun->fprog)))
3372 case TUNSETSTEERINGEBPF:
3373 ret = tun_set_ebpf(tun, &tun->steering_prog, argp);
3376 case TUNSETFILTEREBPF:
3377 ret = tun_set_ebpf(tun, &tun->filter_prog, argp);
3382 if (copy_from_user(&carrier, argp, sizeof(carrier)))
3385 ret = tun_net_change_carrier(tun->dev, (bool)carrier);
3388 case TUNGETDEVNETNS:
3390 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
3392 ret = open_related_ns(&net->ns, get_net_ns);
3401 netdev_state_change(tun->dev);
3410 static long tun_chr_ioctl(struct file *file,
3411 unsigned int cmd, unsigned long arg)
3413 return __tun_chr_ioctl(file, cmd, arg, sizeof (struct ifreq));
3416 #ifdef CONFIG_COMPAT
3417 static long tun_chr_compat_ioctl(struct file *file,
3418 unsigned int cmd, unsigned long arg)
3423 case TUNSETTXFILTER:
3428 arg = (unsigned long)compat_ptr(arg);
3431 arg = (compat_ulong_t)arg;
3436 * compat_ifreq is shorter than ifreq, so we must not access beyond
3437 * the end of that structure. All fields that are used in this
3438 * driver are compatible though, we don't need to convert the
3441 return __tun_chr_ioctl(file, cmd, arg, sizeof(struct compat_ifreq));
3443 #endif /* CONFIG_COMPAT */
3445 static int tun_chr_fasync(int fd, struct file *file, int on)
3447 struct tun_file *tfile = file->private_data;
3450 if ((ret = fasync_helper(fd, file, on, &tfile->fasync)) < 0)
3454 __f_setown(file, task_pid(current), PIDTYPE_TGID, 0);
3455 tfile->flags |= TUN_FASYNC;
3457 tfile->flags &= ~TUN_FASYNC;
3463 static int tun_chr_open(struct inode *inode, struct file * file)
3465 struct net *net = current->nsproxy->net_ns;
3466 struct tun_file *tfile;
3468 tfile = (struct tun_file *)sk_alloc(net, AF_UNSPEC, GFP_KERNEL,
3472 if (ptr_ring_init(&tfile->tx_ring, 0, GFP_KERNEL)) {
3473 sk_free(&tfile->sk);
3477 mutex_init(&tfile->napi_mutex);
3478 RCU_INIT_POINTER(tfile->tun, NULL);
3482 init_waitqueue_head(&tfile->socket.wq.wait);
3484 tfile->socket.file = file;
3485 tfile->socket.ops = &tun_socket_ops;
3487 sock_init_data_uid(&tfile->socket, &tfile->sk, current_fsuid());
3489 tfile->sk.sk_write_space = tun_sock_write_space;
3490 tfile->sk.sk_sndbuf = INT_MAX;
3492 file->private_data = tfile;
3493 INIT_LIST_HEAD(&tfile->next);
3495 sock_set_flag(&tfile->sk, SOCK_ZEROCOPY);
3497 /* tun groks IOCB_NOWAIT just fine, mark it as such */
3498 file->f_mode |= FMODE_NOWAIT;
3502 static int tun_chr_close(struct inode *inode, struct file *file)
3504 struct tun_file *tfile = file->private_data;
3506 tun_detach(tfile, true);
3511 #ifdef CONFIG_PROC_FS
3512 static void tun_chr_show_fdinfo(struct seq_file *m, struct file *file)
3514 struct tun_file *tfile = file->private_data;
3515 struct tun_struct *tun;
3518 memset(&ifr, 0, sizeof(ifr));
3521 tun = tun_get(tfile);
3523 tun_get_iff(tun, &ifr);
3529 seq_printf(m, "iff:\t%s\n", ifr.ifr_name);
3533 static const struct file_operations tun_fops = {
3534 .owner = THIS_MODULE,
3535 .llseek = no_llseek,
3536 .read_iter = tun_chr_read_iter,
3537 .write_iter = tun_chr_write_iter,
3538 .poll = tun_chr_poll,
3539 .unlocked_ioctl = tun_chr_ioctl,
3540 #ifdef CONFIG_COMPAT
3541 .compat_ioctl = tun_chr_compat_ioctl,
3543 .open = tun_chr_open,
3544 .release = tun_chr_close,
3545 .fasync = tun_chr_fasync,
3546 #ifdef CONFIG_PROC_FS
3547 .show_fdinfo = tun_chr_show_fdinfo,
3551 static struct miscdevice tun_miscdev = {
3554 .nodename = "net/tun",
3558 /* ethtool interface */
3560 static void tun_default_link_ksettings(struct net_device *dev,
3561 struct ethtool_link_ksettings *cmd)
3563 ethtool_link_ksettings_zero_link_mode(cmd, supported);
3564 ethtool_link_ksettings_zero_link_mode(cmd, advertising);
3565 cmd->base.speed = SPEED_10000;
3566 cmd->base.duplex = DUPLEX_FULL;
3567 cmd->base.port = PORT_TP;
3568 cmd->base.phy_address = 0;
3569 cmd->base.autoneg = AUTONEG_DISABLE;
3572 static int tun_get_link_ksettings(struct net_device *dev,
3573 struct ethtool_link_ksettings *cmd)
3575 struct tun_struct *tun = netdev_priv(dev);
3577 memcpy(cmd, &tun->link_ksettings, sizeof(*cmd));
3581 static int tun_set_link_ksettings(struct net_device *dev,
3582 const struct ethtool_link_ksettings *cmd)
3584 struct tun_struct *tun = netdev_priv(dev);
3586 memcpy(&tun->link_ksettings, cmd, sizeof(*cmd));
3590 static void tun_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
3592 struct tun_struct *tun = netdev_priv(dev);
3594 strscpy(info->driver, DRV_NAME, sizeof(info->driver));
3595 strscpy(info->version, DRV_VERSION, sizeof(info->version));
3597 switch (tun->flags & TUN_TYPE_MASK) {
3599 strscpy(info->bus_info, "tun", sizeof(info->bus_info));
3602 strscpy(info->bus_info, "tap", sizeof(info->bus_info));
3607 static u32 tun_get_msglevel(struct net_device *dev)
3609 struct tun_struct *tun = netdev_priv(dev);
3611 return tun->msg_enable;
3614 static void tun_set_msglevel(struct net_device *dev, u32 value)
3616 struct tun_struct *tun = netdev_priv(dev);
3618 tun->msg_enable = value;
3621 static int tun_get_coalesce(struct net_device *dev,
3622 struct ethtool_coalesce *ec,
3623 struct kernel_ethtool_coalesce *kernel_coal,
3624 struct netlink_ext_ack *extack)
3626 struct tun_struct *tun = netdev_priv(dev);
3628 ec->rx_max_coalesced_frames = tun->rx_batched;
3633 static int tun_set_coalesce(struct net_device *dev,
3634 struct ethtool_coalesce *ec,
3635 struct kernel_ethtool_coalesce *kernel_coal,
3636 struct netlink_ext_ack *extack)
3638 struct tun_struct *tun = netdev_priv(dev);
3640 if (ec->rx_max_coalesced_frames > NAPI_POLL_WEIGHT)
3641 tun->rx_batched = NAPI_POLL_WEIGHT;
3643 tun->rx_batched = ec->rx_max_coalesced_frames;
3648 static const struct ethtool_ops tun_ethtool_ops = {
3649 .supported_coalesce_params = ETHTOOL_COALESCE_RX_MAX_FRAMES,
3650 .get_drvinfo = tun_get_drvinfo,
3651 .get_msglevel = tun_get_msglevel,
3652 .set_msglevel = tun_set_msglevel,
3653 .get_link = ethtool_op_get_link,
3654 .get_ts_info = ethtool_op_get_ts_info,
3655 .get_coalesce = tun_get_coalesce,
3656 .set_coalesce = tun_set_coalesce,
3657 .get_link_ksettings = tun_get_link_ksettings,
3658 .set_link_ksettings = tun_set_link_ksettings,
3661 static int tun_queue_resize(struct tun_struct *tun)
3663 struct net_device *dev = tun->dev;
3664 struct tun_file *tfile;
3665 struct ptr_ring **rings;
3666 int n = tun->numqueues + tun->numdisabled;
3669 rings = kmalloc_array(n, sizeof(*rings), GFP_KERNEL);
3673 for (i = 0; i < tun->numqueues; i++) {
3674 tfile = rtnl_dereference(tun->tfiles[i]);
3675 rings[i] = &tfile->tx_ring;
3677 list_for_each_entry(tfile, &tun->disabled, next)
3678 rings[i++] = &tfile->tx_ring;
3680 ret = ptr_ring_resize_multiple(rings, n,
3681 dev->tx_queue_len, GFP_KERNEL,
3688 static int tun_device_event(struct notifier_block *unused,
3689 unsigned long event, void *ptr)
3691 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
3692 struct tun_struct *tun = netdev_priv(dev);
3695 if (dev->rtnl_link_ops != &tun_link_ops)
3699 case NETDEV_CHANGE_TX_QUEUE_LEN:
3700 if (tun_queue_resize(tun))
3704 for (i = 0; i < tun->numqueues; i++) {
3705 struct tun_file *tfile;
3707 tfile = rtnl_dereference(tun->tfiles[i]);
3708 tfile->socket.sk->sk_write_space(tfile->socket.sk);
3718 static struct notifier_block tun_notifier_block __read_mostly = {
3719 .notifier_call = tun_device_event,
3722 static int __init tun_init(void)
3726 pr_info("%s, %s\n", DRV_DESCRIPTION, DRV_VERSION);
3728 ret = rtnl_link_register(&tun_link_ops);
3730 pr_err("Can't register link_ops\n");
3734 ret = misc_register(&tun_miscdev);
3736 pr_err("Can't register misc device %d\n", TUN_MINOR);
3740 ret = register_netdevice_notifier(&tun_notifier_block);
3742 pr_err("Can't register netdevice notifier\n");
3749 misc_deregister(&tun_miscdev);
3751 rtnl_link_unregister(&tun_link_ops);
3756 static void __exit tun_cleanup(void)
3758 misc_deregister(&tun_miscdev);
3759 rtnl_link_unregister(&tun_link_ops);
3760 unregister_netdevice_notifier(&tun_notifier_block);
3763 /* Get an underlying socket object from tun file. Returns error unless file is
3764 * attached to a device. The returned object works like a packet socket, it
3765 * can be used for sock_sendmsg/sock_recvmsg. The caller is responsible for
3766 * holding a reference to the file for as long as the socket is in use. */
3767 struct socket *tun_get_socket(struct file *file)
3769 struct tun_file *tfile;
3770 if (file->f_op != &tun_fops)
3771 return ERR_PTR(-EINVAL);
3772 tfile = file->private_data;
3774 return ERR_PTR(-EBADFD);
3775 return &tfile->socket;
3777 EXPORT_SYMBOL_GPL(tun_get_socket);
3779 struct ptr_ring *tun_get_tx_ring(struct file *file)
3781 struct tun_file *tfile;
3783 if (file->f_op != &tun_fops)
3784 return ERR_PTR(-EINVAL);
3785 tfile = file->private_data;
3787 return ERR_PTR(-EBADFD);
3788 return &tfile->tx_ring;
3790 EXPORT_SYMBOL_GPL(tun_get_tx_ring);
3792 module_init(tun_init);
3793 module_exit(tun_cleanup);
3794 MODULE_DESCRIPTION(DRV_DESCRIPTION);
3795 MODULE_AUTHOR(DRV_COPYRIGHT);
3796 MODULE_LICENSE("GPL");
3797 MODULE_ALIAS_MISCDEV(TUN_MINOR);
3798 MODULE_ALIAS("devname:net/tun");