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 struct tun_pcpu_stats {
119 u64_stats_t rx_packets;
120 u64_stats_t rx_bytes;
121 u64_stats_t tx_packets;
122 u64_stats_t tx_bytes;
123 struct u64_stats_sync syncp;
129 /* A tun_file connects an open character device to a tuntap netdevice. It
130 * also contains all socket related structures (except sock_fprog and tap_filter)
131 * to serve as one transmit queue for tuntap device. The sock_fprog and
132 * tap_filter were kept in tun_struct since they were used for filtering for the
133 * netdevice not for a specific queue (at least I didn't see the requirement for
137 * The tun_file and tun_struct are loosely coupled, the pointer from one to the
138 * other can only be read while rcu_read_lock or rtnl_lock is held.
142 struct socket socket;
143 struct tun_struct __rcu *tun;
144 struct fasync_struct *fasync;
145 /* only used for fasnyc */
149 unsigned int ifindex;
151 struct napi_struct napi;
153 bool napi_frags_enabled;
154 struct mutex napi_mutex; /* Protects access to the above napi */
155 struct list_head next;
156 struct tun_struct *detached;
157 struct ptr_ring tx_ring;
158 struct xdp_rxq_info xdp_rxq;
166 struct tun_flow_entry {
167 struct hlist_node hash_link;
169 struct tun_struct *tun;
174 unsigned long updated ____cacheline_aligned_in_smp;
177 #define TUN_NUM_FLOW_ENTRIES 1024
178 #define TUN_MASK_FLOW_ENTRIES (TUN_NUM_FLOW_ENTRIES - 1)
182 struct bpf_prog *prog;
185 /* Since the socket were moved to tun_file, to preserve the behavior of persist
186 * device, socket filter, sndbuf and vnet header size were restore when the
187 * file were attached to a persist device.
190 struct tun_file __rcu *tfiles[MAX_TAP_QUEUES];
191 unsigned int numqueues;
196 struct net_device *dev;
197 netdev_features_t set_features;
198 #define TUN_USER_FEATURES (NETIF_F_HW_CSUM|NETIF_F_TSO_ECN|NETIF_F_TSO| \
204 struct tap_filter txflt;
205 struct sock_fprog fprog;
206 /* protected by rtnl lock */
207 bool filter_attached;
210 struct hlist_head flows[TUN_NUM_FLOW_ENTRIES];
211 struct timer_list flow_gc_timer;
212 unsigned long ageing_time;
213 unsigned int numdisabled;
214 struct list_head disabled;
218 struct tun_pcpu_stats __percpu *pcpu_stats;
219 struct bpf_prog __rcu *xdp_prog;
220 struct tun_prog __rcu *steering_prog;
221 struct tun_prog __rcu *filter_prog;
222 struct ethtool_link_ksettings link_ksettings;
233 static void tun_flow_init(struct tun_struct *tun);
234 static void tun_flow_uninit(struct tun_struct *tun);
236 static int tun_napi_receive(struct napi_struct *napi, int budget)
238 struct tun_file *tfile = container_of(napi, struct tun_file, napi);
239 struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
240 struct sk_buff_head process_queue;
244 __skb_queue_head_init(&process_queue);
246 spin_lock(&queue->lock);
247 skb_queue_splice_tail_init(queue, &process_queue);
248 spin_unlock(&queue->lock);
250 while (received < budget && (skb = __skb_dequeue(&process_queue))) {
251 napi_gro_receive(napi, skb);
255 if (!skb_queue_empty(&process_queue)) {
256 spin_lock(&queue->lock);
257 skb_queue_splice(&process_queue, queue);
258 spin_unlock(&queue->lock);
264 static int tun_napi_poll(struct napi_struct *napi, int budget)
266 unsigned int received;
268 received = tun_napi_receive(napi, budget);
270 if (received < budget)
271 napi_complete_done(napi, received);
276 static void tun_napi_init(struct tun_struct *tun, struct tun_file *tfile,
277 bool napi_en, bool napi_frags)
279 tfile->napi_enabled = napi_en;
280 tfile->napi_frags_enabled = napi_en && napi_frags;
282 netif_tx_napi_add(tun->dev, &tfile->napi, tun_napi_poll,
284 napi_enable(&tfile->napi);
288 static void tun_napi_enable(struct tun_file *tfile)
290 if (tfile->napi_enabled)
291 napi_enable(&tfile->napi);
294 static void tun_napi_disable(struct tun_file *tfile)
296 if (tfile->napi_enabled)
297 napi_disable(&tfile->napi);
300 static void tun_napi_del(struct tun_file *tfile)
302 if (tfile->napi_enabled)
303 netif_napi_del(&tfile->napi);
306 static bool tun_napi_frags_enabled(const struct tun_file *tfile)
308 return tfile->napi_frags_enabled;
311 #ifdef CONFIG_TUN_VNET_CROSS_LE
312 static inline bool tun_legacy_is_little_endian(struct tun_struct *tun)
314 return tun->flags & TUN_VNET_BE ? false :
315 virtio_legacy_is_little_endian();
318 static long tun_get_vnet_be(struct tun_struct *tun, int __user *argp)
320 int be = !!(tun->flags & TUN_VNET_BE);
322 if (put_user(be, argp))
328 static long tun_set_vnet_be(struct tun_struct *tun, int __user *argp)
332 if (get_user(be, argp))
336 tun->flags |= TUN_VNET_BE;
338 tun->flags &= ~TUN_VNET_BE;
343 static inline bool tun_legacy_is_little_endian(struct tun_struct *tun)
345 return virtio_legacy_is_little_endian();
348 static long tun_get_vnet_be(struct tun_struct *tun, int __user *argp)
353 static long tun_set_vnet_be(struct tun_struct *tun, int __user *argp)
357 #endif /* CONFIG_TUN_VNET_CROSS_LE */
359 static inline bool tun_is_little_endian(struct tun_struct *tun)
361 return tun->flags & TUN_VNET_LE ||
362 tun_legacy_is_little_endian(tun);
365 static inline u16 tun16_to_cpu(struct tun_struct *tun, __virtio16 val)
367 return __virtio16_to_cpu(tun_is_little_endian(tun), val);
370 static inline __virtio16 cpu_to_tun16(struct tun_struct *tun, u16 val)
372 return __cpu_to_virtio16(tun_is_little_endian(tun), val);
375 static inline u32 tun_hashfn(u32 rxhash)
377 return rxhash & TUN_MASK_FLOW_ENTRIES;
380 static struct tun_flow_entry *tun_flow_find(struct hlist_head *head, u32 rxhash)
382 struct tun_flow_entry *e;
384 hlist_for_each_entry_rcu(e, head, hash_link) {
385 if (e->rxhash == rxhash)
391 static struct tun_flow_entry *tun_flow_create(struct tun_struct *tun,
392 struct hlist_head *head,
393 u32 rxhash, u16 queue_index)
395 struct tun_flow_entry *e = kmalloc(sizeof(*e), GFP_ATOMIC);
398 netif_info(tun, tx_queued, tun->dev,
399 "create flow: hash %u index %u\n",
400 rxhash, queue_index);
401 e->updated = jiffies;
404 e->queue_index = queue_index;
406 hlist_add_head_rcu(&e->hash_link, head);
412 static void tun_flow_delete(struct tun_struct *tun, struct tun_flow_entry *e)
414 netif_info(tun, tx_queued, tun->dev, "delete flow: hash %u index %u\n",
415 e->rxhash, e->queue_index);
416 hlist_del_rcu(&e->hash_link);
421 static void tun_flow_flush(struct tun_struct *tun)
425 spin_lock_bh(&tun->lock);
426 for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
427 struct tun_flow_entry *e;
428 struct hlist_node *n;
430 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link)
431 tun_flow_delete(tun, e);
433 spin_unlock_bh(&tun->lock);
436 static void tun_flow_delete_by_queue(struct tun_struct *tun, u16 queue_index)
440 spin_lock_bh(&tun->lock);
441 for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
442 struct tun_flow_entry *e;
443 struct hlist_node *n;
445 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
446 if (e->queue_index == queue_index)
447 tun_flow_delete(tun, e);
450 spin_unlock_bh(&tun->lock);
453 static void tun_flow_cleanup(struct timer_list *t)
455 struct tun_struct *tun = from_timer(tun, t, flow_gc_timer);
456 unsigned long delay = tun->ageing_time;
457 unsigned long next_timer = jiffies + delay;
458 unsigned long count = 0;
461 spin_lock(&tun->lock);
462 for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
463 struct tun_flow_entry *e;
464 struct hlist_node *n;
466 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
467 unsigned long this_timer;
469 this_timer = e->updated + delay;
470 if (time_before_eq(this_timer, jiffies)) {
471 tun_flow_delete(tun, e);
475 if (time_before(this_timer, next_timer))
476 next_timer = this_timer;
481 mod_timer(&tun->flow_gc_timer, round_jiffies_up(next_timer));
482 spin_unlock(&tun->lock);
485 static void tun_flow_update(struct tun_struct *tun, u32 rxhash,
486 struct tun_file *tfile)
488 struct hlist_head *head;
489 struct tun_flow_entry *e;
490 unsigned long delay = tun->ageing_time;
491 u16 queue_index = tfile->queue_index;
493 head = &tun->flows[tun_hashfn(rxhash)];
497 e = tun_flow_find(head, rxhash);
499 /* TODO: keep queueing to old queue until it's empty? */
500 if (READ_ONCE(e->queue_index) != queue_index)
501 WRITE_ONCE(e->queue_index, queue_index);
502 if (e->updated != jiffies)
503 e->updated = jiffies;
504 sock_rps_record_flow_hash(e->rps_rxhash);
506 spin_lock_bh(&tun->lock);
507 if (!tun_flow_find(head, rxhash) &&
508 tun->flow_count < MAX_TAP_FLOWS)
509 tun_flow_create(tun, head, rxhash, queue_index);
511 if (!timer_pending(&tun->flow_gc_timer))
512 mod_timer(&tun->flow_gc_timer,
513 round_jiffies_up(jiffies + delay));
514 spin_unlock_bh(&tun->lock);
520 /* Save the hash received in the stack receive path and update the
521 * flow_hash table accordingly.
523 static inline void tun_flow_save_rps_rxhash(struct tun_flow_entry *e, u32 hash)
525 if (unlikely(e->rps_rxhash != hash))
526 e->rps_rxhash = hash;
529 /* We try to identify a flow through its rxhash. The reason that
530 * we do not check rxq no. is because some cards(e.g 82599), chooses
531 * the rxq based on the txq where the last packet of the flow comes. As
532 * the userspace application move between processors, we may get a
533 * different rxq no. here.
535 static u16 tun_automq_select_queue(struct tun_struct *tun, struct sk_buff *skb)
537 struct tun_flow_entry *e;
541 numqueues = READ_ONCE(tun->numqueues);
543 txq = __skb_get_hash_symmetric(skb);
544 e = tun_flow_find(&tun->flows[tun_hashfn(txq)], txq);
546 tun_flow_save_rps_rxhash(e, txq);
547 txq = e->queue_index;
549 /* use multiply and shift instead of expensive divide */
550 txq = ((u64)txq * numqueues) >> 32;
556 static u16 tun_ebpf_select_queue(struct tun_struct *tun, struct sk_buff *skb)
558 struct tun_prog *prog;
562 numqueues = READ_ONCE(tun->numqueues);
566 prog = rcu_dereference(tun->steering_prog);
568 ret = bpf_prog_run_clear_cb(prog->prog, skb);
570 return ret % numqueues;
573 static u16 tun_select_queue(struct net_device *dev, struct sk_buff *skb,
574 struct net_device *sb_dev)
576 struct tun_struct *tun = netdev_priv(dev);
580 if (rcu_dereference(tun->steering_prog))
581 ret = tun_ebpf_select_queue(tun, skb);
583 ret = tun_automq_select_queue(tun, skb);
589 static inline bool tun_not_capable(struct tun_struct *tun)
591 const struct cred *cred = current_cred();
592 struct net *net = dev_net(tun->dev);
594 return ((uid_valid(tun->owner) && !uid_eq(cred->euid, tun->owner)) ||
595 (gid_valid(tun->group) && !in_egroup_p(tun->group))) &&
596 !ns_capable(net->user_ns, CAP_NET_ADMIN);
599 static void tun_set_real_num_queues(struct tun_struct *tun)
601 netif_set_real_num_tx_queues(tun->dev, tun->numqueues);
602 netif_set_real_num_rx_queues(tun->dev, tun->numqueues);
605 static void tun_disable_queue(struct tun_struct *tun, struct tun_file *tfile)
607 tfile->detached = tun;
608 list_add_tail(&tfile->next, &tun->disabled);
612 static struct tun_struct *tun_enable_queue(struct tun_file *tfile)
614 struct tun_struct *tun = tfile->detached;
616 tfile->detached = NULL;
617 list_del_init(&tfile->next);
622 void tun_ptr_free(void *ptr)
626 if (tun_is_xdp_frame(ptr)) {
627 struct xdp_frame *xdpf = tun_ptr_to_xdp(ptr);
629 xdp_return_frame(xdpf);
631 __skb_array_destroy_skb(ptr);
634 EXPORT_SYMBOL_GPL(tun_ptr_free);
636 static void tun_queue_purge(struct tun_file *tfile)
640 while ((ptr = ptr_ring_consume(&tfile->tx_ring)) != NULL)
643 skb_queue_purge(&tfile->sk.sk_write_queue);
644 skb_queue_purge(&tfile->sk.sk_error_queue);
647 static void __tun_detach(struct tun_file *tfile, bool clean)
649 struct tun_file *ntfile;
650 struct tun_struct *tun;
652 tun = rtnl_dereference(tfile->tun);
655 if (!tfile->detached)
656 tun_napi_disable(tfile);
660 if (tun && !tfile->detached) {
661 u16 index = tfile->queue_index;
662 BUG_ON(index >= tun->numqueues);
664 rcu_assign_pointer(tun->tfiles[index],
665 tun->tfiles[tun->numqueues - 1]);
666 ntfile = rtnl_dereference(tun->tfiles[index]);
667 ntfile->queue_index = index;
668 ntfile->xdp_rxq.queue_index = index;
669 rcu_assign_pointer(tun->tfiles[tun->numqueues - 1],
674 RCU_INIT_POINTER(tfile->tun, NULL);
675 sock_put(&tfile->sk);
677 tun_disable_queue(tun, tfile);
678 tun_napi_disable(tfile);
682 tun_flow_delete_by_queue(tun, tun->numqueues + 1);
683 /* Drop read queue */
684 tun_queue_purge(tfile);
685 tun_set_real_num_queues(tun);
686 } else if (tfile->detached && clean) {
687 tun = tun_enable_queue(tfile);
688 sock_put(&tfile->sk);
692 if (tun && tun->numqueues == 0 && tun->numdisabled == 0) {
693 netif_carrier_off(tun->dev);
695 if (!(tun->flags & IFF_PERSIST) &&
696 tun->dev->reg_state == NETREG_REGISTERED)
697 unregister_netdevice(tun->dev);
700 xdp_rxq_info_unreg(&tfile->xdp_rxq);
701 ptr_ring_cleanup(&tfile->tx_ring, tun_ptr_free);
705 static void tun_detach(struct tun_file *tfile, bool clean)
707 struct tun_struct *tun;
708 struct net_device *dev;
711 tun = rtnl_dereference(tfile->tun);
712 dev = tun ? tun->dev : NULL;
713 __tun_detach(tfile, clean);
715 netdev_state_change(dev);
719 sock_put(&tfile->sk);
722 static void tun_detach_all(struct net_device *dev)
724 struct tun_struct *tun = netdev_priv(dev);
725 struct tun_file *tfile, *tmp;
726 int i, n = tun->numqueues;
728 for (i = 0; i < n; i++) {
729 tfile = rtnl_dereference(tun->tfiles[i]);
731 tun_napi_disable(tfile);
732 tfile->socket.sk->sk_shutdown = RCV_SHUTDOWN;
733 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
734 RCU_INIT_POINTER(tfile->tun, NULL);
737 list_for_each_entry(tfile, &tun->disabled, next) {
738 tfile->socket.sk->sk_shutdown = RCV_SHUTDOWN;
739 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
740 RCU_INIT_POINTER(tfile->tun, NULL);
742 BUG_ON(tun->numqueues != 0);
745 for (i = 0; i < n; i++) {
746 tfile = rtnl_dereference(tun->tfiles[i]);
748 /* Drop read queue */
749 tun_queue_purge(tfile);
750 xdp_rxq_info_unreg(&tfile->xdp_rxq);
751 sock_put(&tfile->sk);
753 list_for_each_entry_safe(tfile, tmp, &tun->disabled, next) {
755 tun_enable_queue(tfile);
756 tun_queue_purge(tfile);
757 xdp_rxq_info_unreg(&tfile->xdp_rxq);
758 sock_put(&tfile->sk);
760 BUG_ON(tun->numdisabled != 0);
762 if (tun->flags & IFF_PERSIST)
763 module_put(THIS_MODULE);
766 static int tun_attach(struct tun_struct *tun, struct file *file,
767 bool skip_filter, bool napi, bool napi_frags,
770 struct tun_file *tfile = file->private_data;
771 struct net_device *dev = tun->dev;
774 err = security_tun_dev_attach(tfile->socket.sk, tun->security);
779 if (rtnl_dereference(tfile->tun) && !tfile->detached)
783 if (!(tun->flags & IFF_MULTI_QUEUE) && tun->numqueues == 1)
787 if (!tfile->detached &&
788 tun->numqueues + tun->numdisabled == MAX_TAP_QUEUES)
793 /* Re-attach the filter to persist device */
794 if (!skip_filter && (tun->filter_attached == true)) {
795 lock_sock(tfile->socket.sk);
796 err = sk_attach_filter(&tun->fprog, tfile->socket.sk);
797 release_sock(tfile->socket.sk);
802 if (!tfile->detached &&
803 ptr_ring_resize(&tfile->tx_ring, dev->tx_queue_len,
804 GFP_KERNEL, tun_ptr_free)) {
809 tfile->queue_index = tun->numqueues;
810 tfile->socket.sk->sk_shutdown &= ~RCV_SHUTDOWN;
812 if (tfile->detached) {
813 /* Re-attach detached tfile, updating XDP queue_index */
814 WARN_ON(!xdp_rxq_info_is_reg(&tfile->xdp_rxq));
816 if (tfile->xdp_rxq.queue_index != tfile->queue_index)
817 tfile->xdp_rxq.queue_index = tfile->queue_index;
819 /* Setup XDP RX-queue info, for new tfile getting attached */
820 err = xdp_rxq_info_reg(&tfile->xdp_rxq,
821 tun->dev, tfile->queue_index);
824 err = xdp_rxq_info_reg_mem_model(&tfile->xdp_rxq,
825 MEM_TYPE_PAGE_SHARED, NULL);
827 xdp_rxq_info_unreg(&tfile->xdp_rxq);
833 if (tfile->detached) {
834 tun_enable_queue(tfile);
835 tun_napi_enable(tfile);
837 sock_hold(&tfile->sk);
838 tun_napi_init(tun, tfile, napi, napi_frags);
841 if (rtnl_dereference(tun->xdp_prog))
842 sock_set_flag(&tfile->sk, SOCK_XDP);
844 /* device is allowed to go away first, so no need to hold extra
848 /* Publish tfile->tun and tun->tfiles only after we've fully
849 * initialized tfile; otherwise we risk using half-initialized
853 rcu_assign_pointer(tfile->tun, tun);
854 rcu_assign_pointer(tun->tfiles[tun->numqueues], tfile);
856 tun_set_real_num_queues(tun);
861 static struct tun_struct *tun_get(struct tun_file *tfile)
863 struct tun_struct *tun;
866 tun = rcu_dereference(tfile->tun);
874 static void tun_put(struct tun_struct *tun)
880 static void addr_hash_set(u32 *mask, const u8 *addr)
882 int n = ether_crc(ETH_ALEN, addr) >> 26;
883 mask[n >> 5] |= (1 << (n & 31));
886 static unsigned int addr_hash_test(const u32 *mask, const u8 *addr)
888 int n = ether_crc(ETH_ALEN, addr) >> 26;
889 return mask[n >> 5] & (1 << (n & 31));
892 static int update_filter(struct tap_filter *filter, void __user *arg)
894 struct { u8 u[ETH_ALEN]; } *addr;
895 struct tun_filter uf;
896 int err, alen, n, nexact;
898 if (copy_from_user(&uf, arg, sizeof(uf)))
907 alen = ETH_ALEN * uf.count;
908 addr = memdup_user(arg + sizeof(uf), alen);
910 return PTR_ERR(addr);
912 /* The filter is updated without holding any locks. Which is
913 * perfectly safe. We disable it first and in the worst
914 * case we'll accept a few undesired packets. */
918 /* Use first set of addresses as an exact filter */
919 for (n = 0; n < uf.count && n < FLT_EXACT_COUNT; n++)
920 memcpy(filter->addr[n], addr[n].u, ETH_ALEN);
924 /* Remaining multicast addresses are hashed,
925 * unicast will leave the filter disabled. */
926 memset(filter->mask, 0, sizeof(filter->mask));
927 for (; n < uf.count; n++) {
928 if (!is_multicast_ether_addr(addr[n].u)) {
929 err = 0; /* no filter */
932 addr_hash_set(filter->mask, addr[n].u);
935 /* For ALLMULTI just set the mask to all ones.
936 * This overrides the mask populated above. */
937 if ((uf.flags & TUN_FLT_ALLMULTI))
938 memset(filter->mask, ~0, sizeof(filter->mask));
940 /* Now enable the filter */
942 filter->count = nexact;
944 /* Return the number of exact filters */
951 /* Returns: 0 - drop, !=0 - accept */
952 static int run_filter(struct tap_filter *filter, const struct sk_buff *skb)
954 /* Cannot use eth_hdr(skb) here because skb_mac_hdr() is incorrect
956 struct ethhdr *eh = (struct ethhdr *) skb->data;
960 for (i = 0; i < filter->count; i++)
961 if (ether_addr_equal(eh->h_dest, filter->addr[i]))
964 /* Inexact match (multicast only) */
965 if (is_multicast_ether_addr(eh->h_dest))
966 return addr_hash_test(filter->mask, eh->h_dest);
972 * Checks whether the packet is accepted or not.
973 * Returns: 0 - drop, !=0 - accept
975 static int check_filter(struct tap_filter *filter, const struct sk_buff *skb)
980 return run_filter(filter, skb);
983 /* Network device part of the driver */
985 static const struct ethtool_ops tun_ethtool_ops;
987 static int tun_net_init(struct net_device *dev)
989 struct tun_struct *tun = netdev_priv(dev);
990 struct ifreq *ifr = tun->ifr;
993 tun->pcpu_stats = netdev_alloc_pcpu_stats(struct tun_pcpu_stats);
994 if (!tun->pcpu_stats)
997 spin_lock_init(&tun->lock);
999 err = security_tun_dev_alloc_security(&tun->security);
1001 free_percpu(tun->pcpu_stats);
1007 dev->hw_features = NETIF_F_SG | NETIF_F_FRAGLIST |
1008 TUN_USER_FEATURES | NETIF_F_HW_VLAN_CTAG_TX |
1009 NETIF_F_HW_VLAN_STAG_TX;
1010 dev->features = dev->hw_features | NETIF_F_LLTX;
1011 dev->vlan_features = dev->features &
1012 ~(NETIF_F_HW_VLAN_CTAG_TX |
1013 NETIF_F_HW_VLAN_STAG_TX);
1015 tun->flags = (tun->flags & ~TUN_FEATURES) |
1016 (ifr->ifr_flags & TUN_FEATURES);
1018 INIT_LIST_HEAD(&tun->disabled);
1019 err = tun_attach(tun, tun->file, false, ifr->ifr_flags & IFF_NAPI,
1020 ifr->ifr_flags & IFF_NAPI_FRAGS, false);
1022 tun_flow_uninit(tun);
1023 security_tun_dev_free_security(tun->security);
1024 free_percpu(tun->pcpu_stats);
1030 /* Net device detach from fd. */
1031 static void tun_net_uninit(struct net_device *dev)
1033 tun_detach_all(dev);
1036 /* Net device open. */
1037 static int tun_net_open(struct net_device *dev)
1039 netif_tx_start_all_queues(dev);
1044 /* Net device close. */
1045 static int tun_net_close(struct net_device *dev)
1047 netif_tx_stop_all_queues(dev);
1051 /* Net device start xmit */
1052 static void tun_automq_xmit(struct tun_struct *tun, struct sk_buff *skb)
1055 if (tun->numqueues == 1 && static_branch_unlikely(&rps_needed)) {
1056 /* Select queue was not called for the skbuff, so we extract the
1057 * RPS hash and save it into the flow_table here.
1059 struct tun_flow_entry *e;
1062 rxhash = __skb_get_hash_symmetric(skb);
1063 e = tun_flow_find(&tun->flows[tun_hashfn(rxhash)], rxhash);
1065 tun_flow_save_rps_rxhash(e, rxhash);
1070 static unsigned int run_ebpf_filter(struct tun_struct *tun,
1071 struct sk_buff *skb,
1074 struct tun_prog *prog = rcu_dereference(tun->filter_prog);
1077 len = bpf_prog_run_clear_cb(prog->prog, skb);
1082 /* Net device start xmit */
1083 static netdev_tx_t tun_net_xmit(struct sk_buff *skb, struct net_device *dev)
1085 struct tun_struct *tun = netdev_priv(dev);
1086 int txq = skb->queue_mapping;
1087 struct netdev_queue *queue;
1088 struct tun_file *tfile;
1092 tfile = rcu_dereference(tun->tfiles[txq]);
1094 /* Drop packet if interface is not attached */
1098 if (!rcu_dereference(tun->steering_prog))
1099 tun_automq_xmit(tun, skb);
1101 netif_info(tun, tx_queued, tun->dev, "%s %d\n", __func__, skb->len);
1103 /* Drop if the filter does not like it.
1104 * This is a noop if the filter is disabled.
1105 * Filter can be enabled only for the TAP devices. */
1106 if (!check_filter(&tun->txflt, skb))
1109 if (tfile->socket.sk->sk_filter &&
1110 sk_filter(tfile->socket.sk, skb))
1113 len = run_ebpf_filter(tun, skb, len);
1114 if (len == 0 || pskb_trim(skb, len))
1117 if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC)))
1120 skb_tx_timestamp(skb);
1122 /* Orphan the skb - required as we might hang on to it
1123 * for indefinite time.
1129 if (ptr_ring_produce(&tfile->tx_ring, skb))
1132 /* NETIF_F_LLTX requires to do our own update of trans_start */
1133 queue = netdev_get_tx_queue(dev, txq);
1134 queue->trans_start = jiffies;
1136 /* Notify and wake up reader process */
1137 if (tfile->flags & TUN_FASYNC)
1138 kill_fasync(&tfile->fasync, SIGIO, POLL_IN);
1139 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
1142 return NETDEV_TX_OK;
1145 this_cpu_inc(tun->pcpu_stats->tx_dropped);
1149 return NET_XMIT_DROP;
1152 static void tun_net_mclist(struct net_device *dev)
1155 * This callback is supposed to deal with mc filter in
1156 * _rx_ path and has nothing to do with the _tx_ path.
1157 * In rx path we always accept everything userspace gives us.
1161 static netdev_features_t tun_net_fix_features(struct net_device *dev,
1162 netdev_features_t features)
1164 struct tun_struct *tun = netdev_priv(dev);
1166 return (features & tun->set_features) | (features & ~TUN_USER_FEATURES);
1169 static void tun_set_headroom(struct net_device *dev, int new_hr)
1171 struct tun_struct *tun = netdev_priv(dev);
1173 if (new_hr < NET_SKB_PAD)
1174 new_hr = NET_SKB_PAD;
1176 tun->align = new_hr;
1180 tun_net_get_stats64(struct net_device *dev, struct rtnl_link_stats64 *stats)
1182 u32 rx_dropped = 0, tx_dropped = 0, rx_frame_errors = 0;
1183 struct tun_struct *tun = netdev_priv(dev);
1184 struct tun_pcpu_stats *p;
1187 for_each_possible_cpu(i) {
1188 u64 rxpackets, rxbytes, txpackets, txbytes;
1191 p = per_cpu_ptr(tun->pcpu_stats, i);
1193 start = u64_stats_fetch_begin(&p->syncp);
1194 rxpackets = u64_stats_read(&p->rx_packets);
1195 rxbytes = u64_stats_read(&p->rx_bytes);
1196 txpackets = u64_stats_read(&p->tx_packets);
1197 txbytes = u64_stats_read(&p->tx_bytes);
1198 } while (u64_stats_fetch_retry(&p->syncp, start));
1200 stats->rx_packets += rxpackets;
1201 stats->rx_bytes += rxbytes;
1202 stats->tx_packets += txpackets;
1203 stats->tx_bytes += txbytes;
1206 rx_dropped += p->rx_dropped;
1207 rx_frame_errors += p->rx_frame_errors;
1208 tx_dropped += p->tx_dropped;
1210 stats->rx_dropped = rx_dropped;
1211 stats->rx_frame_errors = rx_frame_errors;
1212 stats->tx_dropped = tx_dropped;
1215 static int tun_xdp_set(struct net_device *dev, struct bpf_prog *prog,
1216 struct netlink_ext_ack *extack)
1218 struct tun_struct *tun = netdev_priv(dev);
1219 struct tun_file *tfile;
1220 struct bpf_prog *old_prog;
1223 old_prog = rtnl_dereference(tun->xdp_prog);
1224 rcu_assign_pointer(tun->xdp_prog, prog);
1226 bpf_prog_put(old_prog);
1228 for (i = 0; i < tun->numqueues; i++) {
1229 tfile = rtnl_dereference(tun->tfiles[i]);
1231 sock_set_flag(&tfile->sk, SOCK_XDP);
1233 sock_reset_flag(&tfile->sk, SOCK_XDP);
1235 list_for_each_entry(tfile, &tun->disabled, next) {
1237 sock_set_flag(&tfile->sk, SOCK_XDP);
1239 sock_reset_flag(&tfile->sk, SOCK_XDP);
1245 static int tun_xdp(struct net_device *dev, struct netdev_bpf *xdp)
1247 switch (xdp->command) {
1248 case XDP_SETUP_PROG:
1249 return tun_xdp_set(dev, xdp->prog, xdp->extack);
1255 static int tun_net_change_carrier(struct net_device *dev, bool new_carrier)
1258 struct tun_struct *tun = netdev_priv(dev);
1260 if (!tun->numqueues)
1263 netif_carrier_on(dev);
1265 netif_carrier_off(dev);
1270 static const struct net_device_ops tun_netdev_ops = {
1271 .ndo_init = tun_net_init,
1272 .ndo_uninit = tun_net_uninit,
1273 .ndo_open = tun_net_open,
1274 .ndo_stop = tun_net_close,
1275 .ndo_start_xmit = tun_net_xmit,
1276 .ndo_fix_features = tun_net_fix_features,
1277 .ndo_select_queue = tun_select_queue,
1278 .ndo_set_rx_headroom = tun_set_headroom,
1279 .ndo_get_stats64 = tun_net_get_stats64,
1280 .ndo_change_carrier = tun_net_change_carrier,
1283 static void __tun_xdp_flush_tfile(struct tun_file *tfile)
1285 /* Notify and wake up reader process */
1286 if (tfile->flags & TUN_FASYNC)
1287 kill_fasync(&tfile->fasync, SIGIO, POLL_IN);
1288 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
1291 static int tun_xdp_xmit(struct net_device *dev, int n,
1292 struct xdp_frame **frames, u32 flags)
1294 struct tun_struct *tun = netdev_priv(dev);
1295 struct tun_file *tfile;
1301 if (unlikely(flags & ~XDP_XMIT_FLAGS_MASK))
1307 numqueues = READ_ONCE(tun->numqueues);
1310 return -ENXIO; /* Caller will free/return all frames */
1313 tfile = rcu_dereference(tun->tfiles[smp_processor_id() %
1315 if (unlikely(!tfile))
1318 spin_lock(&tfile->tx_ring.producer_lock);
1319 for (i = 0; i < n; i++) {
1320 struct xdp_frame *xdp = frames[i];
1321 /* Encode the XDP flag into lowest bit for consumer to differ
1322 * XDP buffer from sk_buff.
1324 void *frame = tun_xdp_to_ptr(xdp);
1326 if (__ptr_ring_produce(&tfile->tx_ring, frame)) {
1327 this_cpu_inc(tun->pcpu_stats->tx_dropped);
1328 xdp_return_frame_rx_napi(xdp);
1332 spin_unlock(&tfile->tx_ring.producer_lock);
1334 if (flags & XDP_XMIT_FLUSH)
1335 __tun_xdp_flush_tfile(tfile);
1341 static int tun_xdp_tx(struct net_device *dev, struct xdp_buff *xdp)
1343 struct xdp_frame *frame = xdp_convert_buff_to_frame(xdp);
1345 if (unlikely(!frame))
1348 return tun_xdp_xmit(dev, 1, &frame, XDP_XMIT_FLUSH);
1351 static const struct net_device_ops tap_netdev_ops = {
1352 .ndo_init = tun_net_init,
1353 .ndo_uninit = tun_net_uninit,
1354 .ndo_open = tun_net_open,
1355 .ndo_stop = tun_net_close,
1356 .ndo_start_xmit = tun_net_xmit,
1357 .ndo_fix_features = tun_net_fix_features,
1358 .ndo_set_rx_mode = tun_net_mclist,
1359 .ndo_set_mac_address = eth_mac_addr,
1360 .ndo_validate_addr = eth_validate_addr,
1361 .ndo_select_queue = tun_select_queue,
1362 .ndo_features_check = passthru_features_check,
1363 .ndo_set_rx_headroom = tun_set_headroom,
1364 .ndo_get_stats64 = tun_net_get_stats64,
1366 .ndo_xdp_xmit = tun_xdp_xmit,
1367 .ndo_change_carrier = tun_net_change_carrier,
1370 static void tun_flow_init(struct tun_struct *tun)
1374 for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++)
1375 INIT_HLIST_HEAD(&tun->flows[i]);
1377 tun->ageing_time = TUN_FLOW_EXPIRE;
1378 timer_setup(&tun->flow_gc_timer, tun_flow_cleanup, 0);
1379 mod_timer(&tun->flow_gc_timer,
1380 round_jiffies_up(jiffies + tun->ageing_time));
1383 static void tun_flow_uninit(struct tun_struct *tun)
1385 del_timer_sync(&tun->flow_gc_timer);
1386 tun_flow_flush(tun);
1390 #define MAX_MTU 65535
1392 /* Initialize net device. */
1393 static void tun_net_initialize(struct net_device *dev)
1395 struct tun_struct *tun = netdev_priv(dev);
1397 switch (tun->flags & TUN_TYPE_MASK) {
1399 dev->netdev_ops = &tun_netdev_ops;
1400 dev->header_ops = &ip_tunnel_header_ops;
1402 /* Point-to-Point TUN Device */
1403 dev->hard_header_len = 0;
1407 /* Zero header length */
1408 dev->type = ARPHRD_NONE;
1409 dev->flags = IFF_POINTOPOINT | IFF_NOARP | IFF_MULTICAST;
1413 dev->netdev_ops = &tap_netdev_ops;
1414 /* Ethernet TAP Device */
1416 dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1417 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
1419 eth_hw_addr_random(dev);
1424 dev->min_mtu = MIN_MTU;
1425 dev->max_mtu = MAX_MTU - dev->hard_header_len;
1428 static bool tun_sock_writeable(struct tun_struct *tun, struct tun_file *tfile)
1430 struct sock *sk = tfile->socket.sk;
1432 return (tun->dev->flags & IFF_UP) && sock_writeable(sk);
1435 /* Character device part */
1438 static __poll_t tun_chr_poll(struct file *file, poll_table *wait)
1440 struct tun_file *tfile = file->private_data;
1441 struct tun_struct *tun = tun_get(tfile);
1448 sk = tfile->socket.sk;
1450 poll_wait(file, sk_sleep(sk), wait);
1452 if (!ptr_ring_empty(&tfile->tx_ring))
1453 mask |= EPOLLIN | EPOLLRDNORM;
1455 /* Make sure SOCKWQ_ASYNC_NOSPACE is set if not writable to
1456 * guarantee EPOLLOUT to be raised by either here or
1457 * tun_sock_write_space(). Then process could get notification
1458 * after it writes to a down device and meets -EIO.
1460 if (tun_sock_writeable(tun, tfile) ||
1461 (!test_and_set_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags) &&
1462 tun_sock_writeable(tun, tfile)))
1463 mask |= EPOLLOUT | EPOLLWRNORM;
1465 if (tun->dev->reg_state != NETREG_REGISTERED)
1472 static struct sk_buff *tun_napi_alloc_frags(struct tun_file *tfile,
1474 const struct iov_iter *it)
1476 struct sk_buff *skb;
1481 if (it->nr_segs > MAX_SKB_FRAGS + 1 ||
1482 len > (ETH_MAX_MTU - NET_SKB_PAD - NET_IP_ALIGN))
1483 return ERR_PTR(-EMSGSIZE);
1486 skb = napi_get_frags(&tfile->napi);
1489 return ERR_PTR(-ENOMEM);
1491 linear = iov_iter_single_seg_count(it);
1492 err = __skb_grow(skb, linear);
1497 skb->data_len = len - linear;
1498 skb->truesize += skb->data_len;
1500 for (i = 1; i < it->nr_segs; i++) {
1501 size_t fragsz = it->iov[i].iov_len;
1505 if (fragsz == 0 || fragsz > PAGE_SIZE) {
1509 frag = netdev_alloc_frag(fragsz);
1514 page = virt_to_head_page(frag);
1515 skb_fill_page_desc(skb, i - 1, page,
1516 frag - page_address(page), fragsz);
1521 /* frees skb and all frags allocated with napi_alloc_frag() */
1522 napi_free_frags(&tfile->napi);
1523 return ERR_PTR(err);
1526 /* prepad is the amount to reserve at front. len is length after that.
1527 * linear is a hint as to how much to copy (usually headers). */
1528 static struct sk_buff *tun_alloc_skb(struct tun_file *tfile,
1529 size_t prepad, size_t len,
1530 size_t linear, int noblock)
1532 struct sock *sk = tfile->socket.sk;
1533 struct sk_buff *skb;
1536 /* Under a page? Don't bother with paged skb. */
1537 if (prepad + len < PAGE_SIZE || !linear)
1540 skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
1543 return ERR_PTR(err);
1545 skb_reserve(skb, prepad);
1546 skb_put(skb, linear);
1547 skb->data_len = len - linear;
1548 skb->len += len - linear;
1553 static void tun_rx_batched(struct tun_struct *tun, struct tun_file *tfile,
1554 struct sk_buff *skb, int more)
1556 struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
1557 struct sk_buff_head process_queue;
1558 u32 rx_batched = tun->rx_batched;
1561 if (!rx_batched || (!more && skb_queue_empty(queue))) {
1563 skb_record_rx_queue(skb, tfile->queue_index);
1564 netif_receive_skb(skb);
1569 spin_lock(&queue->lock);
1570 if (!more || skb_queue_len(queue) == rx_batched) {
1571 __skb_queue_head_init(&process_queue);
1572 skb_queue_splice_tail_init(queue, &process_queue);
1575 __skb_queue_tail(queue, skb);
1577 spin_unlock(&queue->lock);
1580 struct sk_buff *nskb;
1583 while ((nskb = __skb_dequeue(&process_queue))) {
1584 skb_record_rx_queue(nskb, tfile->queue_index);
1585 netif_receive_skb(nskb);
1587 skb_record_rx_queue(skb, tfile->queue_index);
1588 netif_receive_skb(skb);
1593 static bool tun_can_build_skb(struct tun_struct *tun, struct tun_file *tfile,
1594 int len, int noblock, bool zerocopy)
1596 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
1599 if (tfile->socket.sk->sk_sndbuf != INT_MAX)
1608 if (SKB_DATA_ALIGN(len + TUN_RX_PAD + XDP_PACKET_HEADROOM) +
1609 SKB_DATA_ALIGN(sizeof(struct skb_shared_info)) > PAGE_SIZE)
1615 static struct sk_buff *__tun_build_skb(struct tun_file *tfile,
1616 struct page_frag *alloc_frag, char *buf,
1617 int buflen, int len, int pad)
1619 struct sk_buff *skb = build_skb(buf, buflen);
1622 return ERR_PTR(-ENOMEM);
1624 skb_reserve(skb, pad);
1626 skb_set_owner_w(skb, tfile->socket.sk);
1628 get_page(alloc_frag->page);
1629 alloc_frag->offset += buflen;
1634 static int tun_xdp_act(struct tun_struct *tun, struct bpf_prog *xdp_prog,
1635 struct xdp_buff *xdp, u32 act)
1641 err = xdp_do_redirect(tun->dev, xdp, xdp_prog);
1646 err = tun_xdp_tx(tun->dev, xdp);
1653 bpf_warn_invalid_xdp_action(act);
1656 trace_xdp_exception(tun->dev, xdp_prog, act);
1659 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1666 static struct sk_buff *tun_build_skb(struct tun_struct *tun,
1667 struct tun_file *tfile,
1668 struct iov_iter *from,
1669 struct virtio_net_hdr *hdr,
1670 int len, int *skb_xdp)
1672 struct page_frag *alloc_frag = ¤t->task_frag;
1673 struct bpf_prog *xdp_prog;
1674 int buflen = SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
1677 int pad = TUN_RX_PAD;
1681 xdp_prog = rcu_dereference(tun->xdp_prog);
1683 pad += XDP_PACKET_HEADROOM;
1684 buflen += SKB_DATA_ALIGN(len + pad);
1687 alloc_frag->offset = ALIGN((u64)alloc_frag->offset, SMP_CACHE_BYTES);
1688 if (unlikely(!skb_page_frag_refill(buflen, alloc_frag, GFP_KERNEL)))
1689 return ERR_PTR(-ENOMEM);
1691 buf = (char *)page_address(alloc_frag->page) + alloc_frag->offset;
1692 copied = copy_page_from_iter(alloc_frag->page,
1693 alloc_frag->offset + pad,
1696 return ERR_PTR(-EFAULT);
1698 /* There's a small window that XDP may be set after the check
1699 * of xdp_prog above, this should be rare and for simplicity
1700 * we do XDP on skb in case the headroom is not enough.
1702 if (hdr->gso_type || !xdp_prog) {
1704 return __tun_build_skb(tfile, alloc_frag, buf, buflen, len,
1712 xdp_prog = rcu_dereference(tun->xdp_prog);
1714 struct xdp_buff xdp;
1717 xdp.data_hard_start = buf;
1718 xdp.data = buf + pad;
1719 xdp_set_data_meta_invalid(&xdp);
1720 xdp.data_end = xdp.data + len;
1721 xdp.rxq = &tfile->xdp_rxq;
1722 xdp.frame_sz = buflen;
1724 act = bpf_prog_run_xdp(xdp_prog, &xdp);
1725 if (act == XDP_REDIRECT || act == XDP_TX) {
1726 get_page(alloc_frag->page);
1727 alloc_frag->offset += buflen;
1729 err = tun_xdp_act(tun, xdp_prog, &xdp, act);
1731 if (act == XDP_REDIRECT || act == XDP_TX)
1732 put_page(alloc_frag->page);
1736 if (err == XDP_REDIRECT)
1738 if (err != XDP_PASS)
1741 pad = xdp.data - xdp.data_hard_start;
1742 len = xdp.data_end - xdp.data;
1747 return __tun_build_skb(tfile, alloc_frag, buf, buflen, len, pad);
1755 /* Get packet from user space buffer */
1756 static ssize_t tun_get_user(struct tun_struct *tun, struct tun_file *tfile,
1757 void *msg_control, struct iov_iter *from,
1758 int noblock, bool more)
1760 struct tun_pi pi = { 0, cpu_to_be16(ETH_P_IP) };
1761 struct sk_buff *skb;
1762 size_t total_len = iov_iter_count(from);
1763 size_t len = total_len, align = tun->align, linear;
1764 struct virtio_net_hdr gso = { 0 };
1765 struct tun_pcpu_stats *stats;
1768 bool zerocopy = false;
1772 bool frags = tun_napi_frags_enabled(tfile);
1774 if (!(tun->flags & IFF_NO_PI)) {
1775 if (len < sizeof(pi))
1779 if (!copy_from_iter_full(&pi, sizeof(pi), from))
1783 if (tun->flags & IFF_VNET_HDR) {
1784 int vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
1786 if (len < vnet_hdr_sz)
1790 if (!copy_from_iter_full(&gso, sizeof(gso), from))
1793 if ((gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
1794 tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2 > tun16_to_cpu(tun, gso.hdr_len))
1795 gso.hdr_len = cpu_to_tun16(tun, tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2);
1797 if (tun16_to_cpu(tun, gso.hdr_len) > len)
1799 iov_iter_advance(from, vnet_hdr_sz - sizeof(gso));
1802 if ((tun->flags & TUN_TYPE_MASK) == IFF_TAP) {
1803 align += NET_IP_ALIGN;
1804 if (unlikely(len < ETH_HLEN ||
1805 (gso.hdr_len && tun16_to_cpu(tun, gso.hdr_len) < ETH_HLEN)))
1809 good_linear = SKB_MAX_HEAD(align);
1812 struct iov_iter i = *from;
1814 /* There are 256 bytes to be copied in skb, so there is
1815 * enough room for skb expand head in case it is used.
1816 * The rest of the buffer is mapped from userspace.
1818 copylen = gso.hdr_len ? tun16_to_cpu(tun, gso.hdr_len) : GOODCOPY_LEN;
1819 if (copylen > good_linear)
1820 copylen = good_linear;
1822 iov_iter_advance(&i, copylen);
1823 if (iov_iter_npages(&i, INT_MAX) <= MAX_SKB_FRAGS)
1827 if (!frags && tun_can_build_skb(tun, tfile, len, noblock, zerocopy)) {
1828 /* For the packet that is not easy to be processed
1829 * (e.g gso or jumbo packet), we will do it at after
1830 * skb was created with generic XDP routine.
1832 skb = tun_build_skb(tun, tfile, from, &gso, len, &skb_xdp);
1834 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1835 return PTR_ERR(skb);
1842 if (tun16_to_cpu(tun, gso.hdr_len) > good_linear)
1843 linear = good_linear;
1845 linear = tun16_to_cpu(tun, gso.hdr_len);
1849 mutex_lock(&tfile->napi_mutex);
1850 skb = tun_napi_alloc_frags(tfile, copylen, from);
1851 /* tun_napi_alloc_frags() enforces a layout for the skb.
1852 * If zerocopy is enabled, then this layout will be
1853 * overwritten by zerocopy_sg_from_iter().
1857 skb = tun_alloc_skb(tfile, align, copylen, linear,
1862 if (PTR_ERR(skb) != -EAGAIN)
1863 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1865 mutex_unlock(&tfile->napi_mutex);
1866 return PTR_ERR(skb);
1870 err = zerocopy_sg_from_iter(skb, from);
1872 err = skb_copy_datagram_from_iter(skb, 0, from, len);
1877 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1880 tfile->napi.skb = NULL;
1881 mutex_unlock(&tfile->napi_mutex);
1888 if (virtio_net_hdr_to_skb(skb, &gso, tun_is_little_endian(tun))) {
1889 this_cpu_inc(tun->pcpu_stats->rx_frame_errors);
1892 tfile->napi.skb = NULL;
1893 mutex_unlock(&tfile->napi_mutex);
1899 switch (tun->flags & TUN_TYPE_MASK) {
1901 if (tun->flags & IFF_NO_PI) {
1902 u8 ip_version = skb->len ? (skb->data[0] >> 4) : 0;
1904 switch (ip_version) {
1906 pi.proto = htons(ETH_P_IP);
1909 pi.proto = htons(ETH_P_IPV6);
1912 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1918 skb_reset_mac_header(skb);
1919 skb->protocol = pi.proto;
1920 skb->dev = tun->dev;
1923 if (frags && !pskb_may_pull(skb, ETH_HLEN)) {
1927 skb->protocol = eth_type_trans(skb, tun->dev);
1931 /* copy skb_ubuf_info for callback when skb has no error */
1933 skb_shinfo(skb)->destructor_arg = msg_control;
1934 skb_shinfo(skb)->tx_flags |= SKBTX_DEV_ZEROCOPY;
1935 skb_shinfo(skb)->tx_flags |= SKBTX_SHARED_FRAG;
1936 } else if (msg_control) {
1937 struct ubuf_info *uarg = msg_control;
1938 uarg->callback(uarg, false);
1941 skb_reset_network_header(skb);
1942 skb_probe_transport_header(skb);
1943 skb_record_rx_queue(skb, tfile->queue_index);
1946 struct bpf_prog *xdp_prog;
1951 xdp_prog = rcu_dereference(tun->xdp_prog);
1953 ret = do_xdp_generic(xdp_prog, skb);
1954 if (ret != XDP_PASS) {
1958 tfile->napi.skb = NULL;
1959 mutex_unlock(&tfile->napi_mutex);
1968 /* Compute the costly rx hash only if needed for flow updates.
1969 * We may get a very small possibility of OOO during switching, not
1970 * worth to optimize.
1972 if (!rcu_access_pointer(tun->steering_prog) && tun->numqueues > 1 &&
1974 rxhash = __skb_get_hash_symmetric(skb);
1977 if (unlikely(!(tun->dev->flags & IFF_UP))) {
1986 /* Exercise flow dissector code path. */
1987 skb_push(skb, ETH_HLEN);
1988 headlen = eth_get_headlen(tun->dev, skb->data,
1991 if (unlikely(headlen > skb_headlen(skb))) {
1994 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1996 napi_free_frags(&tfile->napi);
1998 mutex_unlock(&tfile->napi_mutex);
2002 if (likely(napi_schedule_prep(&tfile->napi))) {
2004 napi_gro_frags(&tfile->napi);
2005 napi_complete(&tfile->napi);
2011 mutex_unlock(&tfile->napi_mutex);
2012 } else if (tfile->napi_enabled) {
2013 struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
2016 spin_lock_bh(&queue->lock);
2017 __skb_queue_tail(queue, skb);
2018 queue_len = skb_queue_len(queue);
2019 spin_unlock(&queue->lock);
2021 if (!more || queue_len > NAPI_POLL_WEIGHT)
2022 napi_schedule(&tfile->napi);
2025 } else if (!IS_ENABLED(CONFIG_4KSTACKS)) {
2026 tun_rx_batched(tun, tfile, skb, more);
2032 stats = get_cpu_ptr(tun->pcpu_stats);
2033 u64_stats_update_begin(&stats->syncp);
2034 u64_stats_inc(&stats->rx_packets);
2035 u64_stats_add(&stats->rx_bytes, len);
2036 u64_stats_update_end(&stats->syncp);
2040 tun_flow_update(tun, rxhash, tfile);
2045 static ssize_t tun_chr_write_iter(struct kiocb *iocb, struct iov_iter *from)
2047 struct file *file = iocb->ki_filp;
2048 struct tun_file *tfile = file->private_data;
2049 struct tun_struct *tun = tun_get(tfile);
2056 if ((file->f_flags & O_NONBLOCK) || (iocb->ki_flags & IOCB_NOWAIT))
2059 result = tun_get_user(tun, tfile, NULL, from, noblock, false);
2065 static ssize_t tun_put_user_xdp(struct tun_struct *tun,
2066 struct tun_file *tfile,
2067 struct xdp_frame *xdp_frame,
2068 struct iov_iter *iter)
2070 int vnet_hdr_sz = 0;
2071 size_t size = xdp_frame->len;
2072 struct tun_pcpu_stats *stats;
2075 if (tun->flags & IFF_VNET_HDR) {
2076 struct virtio_net_hdr gso = { 0 };
2078 vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
2079 if (unlikely(iov_iter_count(iter) < vnet_hdr_sz))
2081 if (unlikely(copy_to_iter(&gso, sizeof(gso), iter) !=
2084 iov_iter_advance(iter, vnet_hdr_sz - sizeof(gso));
2087 ret = copy_to_iter(xdp_frame->data, size, iter) + vnet_hdr_sz;
2089 stats = get_cpu_ptr(tun->pcpu_stats);
2090 u64_stats_update_begin(&stats->syncp);
2091 u64_stats_inc(&stats->tx_packets);
2092 u64_stats_add(&stats->tx_bytes, ret);
2093 u64_stats_update_end(&stats->syncp);
2094 put_cpu_ptr(tun->pcpu_stats);
2099 /* Put packet to the user space buffer */
2100 static ssize_t tun_put_user(struct tun_struct *tun,
2101 struct tun_file *tfile,
2102 struct sk_buff *skb,
2103 struct iov_iter *iter)
2105 struct tun_pi pi = { 0, skb->protocol };
2106 struct tun_pcpu_stats *stats;
2108 int vlan_offset = 0;
2110 int vnet_hdr_sz = 0;
2112 if (skb_vlan_tag_present(skb))
2113 vlan_hlen = VLAN_HLEN;
2115 if (tun->flags & IFF_VNET_HDR)
2116 vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
2118 total = skb->len + vlan_hlen + vnet_hdr_sz;
2120 if (!(tun->flags & IFF_NO_PI)) {
2121 if (iov_iter_count(iter) < sizeof(pi))
2124 total += sizeof(pi);
2125 if (iov_iter_count(iter) < total) {
2126 /* Packet will be striped */
2127 pi.flags |= TUN_PKT_STRIP;
2130 if (copy_to_iter(&pi, sizeof(pi), iter) != sizeof(pi))
2135 struct virtio_net_hdr gso;
2137 if (iov_iter_count(iter) < vnet_hdr_sz)
2140 if (virtio_net_hdr_from_skb(skb, &gso,
2141 tun_is_little_endian(tun), true,
2143 struct skb_shared_info *sinfo = skb_shinfo(skb);
2144 pr_err("unexpected GSO type: "
2145 "0x%x, gso_size %d, hdr_len %d\n",
2146 sinfo->gso_type, tun16_to_cpu(tun, gso.gso_size),
2147 tun16_to_cpu(tun, gso.hdr_len));
2148 print_hex_dump(KERN_ERR, "tun: ",
2151 min((int)tun16_to_cpu(tun, gso.hdr_len), 64), true);
2156 if (copy_to_iter(&gso, sizeof(gso), iter) != sizeof(gso))
2159 iov_iter_advance(iter, vnet_hdr_sz - sizeof(gso));
2166 veth.h_vlan_proto = skb->vlan_proto;
2167 veth.h_vlan_TCI = htons(skb_vlan_tag_get(skb));
2169 vlan_offset = offsetof(struct vlan_ethhdr, h_vlan_proto);
2171 ret = skb_copy_datagram_iter(skb, 0, iter, vlan_offset);
2172 if (ret || !iov_iter_count(iter))
2175 ret = copy_to_iter(&veth, sizeof(veth), iter);
2176 if (ret != sizeof(veth) || !iov_iter_count(iter))
2180 skb_copy_datagram_iter(skb, vlan_offset, iter, skb->len - vlan_offset);
2183 /* caller is in process context, */
2184 stats = get_cpu_ptr(tun->pcpu_stats);
2185 u64_stats_update_begin(&stats->syncp);
2186 u64_stats_inc(&stats->tx_packets);
2187 u64_stats_add(&stats->tx_bytes, skb->len + vlan_hlen);
2188 u64_stats_update_end(&stats->syncp);
2189 put_cpu_ptr(tun->pcpu_stats);
2194 static void *tun_ring_recv(struct tun_file *tfile, int noblock, int *err)
2196 DECLARE_WAITQUEUE(wait, current);
2200 ptr = ptr_ring_consume(&tfile->tx_ring);
2208 add_wait_queue(&tfile->socket.wq.wait, &wait);
2211 set_current_state(TASK_INTERRUPTIBLE);
2212 ptr = ptr_ring_consume(&tfile->tx_ring);
2215 if (signal_pending(current)) {
2216 error = -ERESTARTSYS;
2219 if (tfile->socket.sk->sk_shutdown & RCV_SHUTDOWN) {
2227 __set_current_state(TASK_RUNNING);
2228 remove_wait_queue(&tfile->socket.wq.wait, &wait);
2235 static ssize_t tun_do_read(struct tun_struct *tun, struct tun_file *tfile,
2236 struct iov_iter *to,
2237 int noblock, void *ptr)
2242 if (!iov_iter_count(to)) {
2248 /* Read frames from ring */
2249 ptr = tun_ring_recv(tfile, noblock, &err);
2254 if (tun_is_xdp_frame(ptr)) {
2255 struct xdp_frame *xdpf = tun_ptr_to_xdp(ptr);
2257 ret = tun_put_user_xdp(tun, tfile, xdpf, to);
2258 xdp_return_frame(xdpf);
2260 struct sk_buff *skb = ptr;
2262 ret = tun_put_user(tun, tfile, skb, to);
2263 if (unlikely(ret < 0))
2272 static ssize_t tun_chr_read_iter(struct kiocb *iocb, struct iov_iter *to)
2274 struct file *file = iocb->ki_filp;
2275 struct tun_file *tfile = file->private_data;
2276 struct tun_struct *tun = tun_get(tfile);
2277 ssize_t len = iov_iter_count(to), ret;
2283 if ((file->f_flags & O_NONBLOCK) || (iocb->ki_flags & IOCB_NOWAIT))
2286 ret = tun_do_read(tun, tfile, to, noblock, NULL);
2287 ret = min_t(ssize_t, ret, len);
2294 static void tun_prog_free(struct rcu_head *rcu)
2296 struct tun_prog *prog = container_of(rcu, struct tun_prog, rcu);
2298 bpf_prog_destroy(prog->prog);
2302 static int __tun_set_ebpf(struct tun_struct *tun,
2303 struct tun_prog __rcu **prog_p,
2304 struct bpf_prog *prog)
2306 struct tun_prog *old, *new = NULL;
2309 new = kmalloc(sizeof(*new), GFP_KERNEL);
2315 spin_lock_bh(&tun->lock);
2316 old = rcu_dereference_protected(*prog_p,
2317 lockdep_is_held(&tun->lock));
2318 rcu_assign_pointer(*prog_p, new);
2319 spin_unlock_bh(&tun->lock);
2322 call_rcu(&old->rcu, tun_prog_free);
2327 static void tun_free_netdev(struct net_device *dev)
2329 struct tun_struct *tun = netdev_priv(dev);
2331 BUG_ON(!(list_empty(&tun->disabled)));
2333 free_percpu(tun->pcpu_stats);
2335 tun_flow_uninit(tun);
2336 security_tun_dev_free_security(tun->security);
2337 __tun_set_ebpf(tun, &tun->steering_prog, NULL);
2338 __tun_set_ebpf(tun, &tun->filter_prog, NULL);
2341 static void tun_setup(struct net_device *dev)
2343 struct tun_struct *tun = netdev_priv(dev);
2345 tun->owner = INVALID_UID;
2346 tun->group = INVALID_GID;
2347 tun_default_link_ksettings(dev, &tun->link_ksettings);
2349 dev->ethtool_ops = &tun_ethtool_ops;
2350 dev->needs_free_netdev = true;
2351 dev->priv_destructor = tun_free_netdev;
2352 /* We prefer our own queue length */
2353 dev->tx_queue_len = TUN_READQ_SIZE;
2356 /* Trivial set of netlink ops to allow deleting tun or tap
2357 * device with netlink.
2359 static int tun_validate(struct nlattr *tb[], struct nlattr *data[],
2360 struct netlink_ext_ack *extack)
2362 NL_SET_ERR_MSG(extack,
2363 "tun/tap creation via rtnetlink is not supported.");
2367 static size_t tun_get_size(const struct net_device *dev)
2369 BUILD_BUG_ON(sizeof(u32) != sizeof(uid_t));
2370 BUILD_BUG_ON(sizeof(u32) != sizeof(gid_t));
2372 return nla_total_size(sizeof(uid_t)) + /* OWNER */
2373 nla_total_size(sizeof(gid_t)) + /* GROUP */
2374 nla_total_size(sizeof(u8)) + /* TYPE */
2375 nla_total_size(sizeof(u8)) + /* PI */
2376 nla_total_size(sizeof(u8)) + /* VNET_HDR */
2377 nla_total_size(sizeof(u8)) + /* PERSIST */
2378 nla_total_size(sizeof(u8)) + /* MULTI_QUEUE */
2379 nla_total_size(sizeof(u32)) + /* NUM_QUEUES */
2380 nla_total_size(sizeof(u32)) + /* NUM_DISABLED_QUEUES */
2384 static int tun_fill_info(struct sk_buff *skb, const struct net_device *dev)
2386 struct tun_struct *tun = netdev_priv(dev);
2388 if (nla_put_u8(skb, IFLA_TUN_TYPE, tun->flags & TUN_TYPE_MASK))
2389 goto nla_put_failure;
2390 if (uid_valid(tun->owner) &&
2391 nla_put_u32(skb, IFLA_TUN_OWNER,
2392 from_kuid_munged(current_user_ns(), tun->owner)))
2393 goto nla_put_failure;
2394 if (gid_valid(tun->group) &&
2395 nla_put_u32(skb, IFLA_TUN_GROUP,
2396 from_kgid_munged(current_user_ns(), tun->group)))
2397 goto nla_put_failure;
2398 if (nla_put_u8(skb, IFLA_TUN_PI, !(tun->flags & IFF_NO_PI)))
2399 goto nla_put_failure;
2400 if (nla_put_u8(skb, IFLA_TUN_VNET_HDR, !!(tun->flags & IFF_VNET_HDR)))
2401 goto nla_put_failure;
2402 if (nla_put_u8(skb, IFLA_TUN_PERSIST, !!(tun->flags & IFF_PERSIST)))
2403 goto nla_put_failure;
2404 if (nla_put_u8(skb, IFLA_TUN_MULTI_QUEUE,
2405 !!(tun->flags & IFF_MULTI_QUEUE)))
2406 goto nla_put_failure;
2407 if (tun->flags & IFF_MULTI_QUEUE) {
2408 if (nla_put_u32(skb, IFLA_TUN_NUM_QUEUES, tun->numqueues))
2409 goto nla_put_failure;
2410 if (nla_put_u32(skb, IFLA_TUN_NUM_DISABLED_QUEUES,
2412 goto nla_put_failure;
2421 static struct rtnl_link_ops tun_link_ops __read_mostly = {
2423 .priv_size = sizeof(struct tun_struct),
2425 .validate = tun_validate,
2426 .get_size = tun_get_size,
2427 .fill_info = tun_fill_info,
2430 static void tun_sock_write_space(struct sock *sk)
2432 struct tun_file *tfile;
2433 wait_queue_head_t *wqueue;
2435 if (!sock_writeable(sk))
2438 if (!test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags))
2441 wqueue = sk_sleep(sk);
2442 if (wqueue && waitqueue_active(wqueue))
2443 wake_up_interruptible_sync_poll(wqueue, EPOLLOUT |
2444 EPOLLWRNORM | EPOLLWRBAND);
2446 tfile = container_of(sk, struct tun_file, sk);
2447 kill_fasync(&tfile->fasync, SIGIO, POLL_OUT);
2450 static void tun_put_page(struct tun_page *tpage)
2453 __page_frag_cache_drain(tpage->page, tpage->count);
2456 static int tun_xdp_one(struct tun_struct *tun,
2457 struct tun_file *tfile,
2458 struct xdp_buff *xdp, int *flush,
2459 struct tun_page *tpage)
2461 unsigned int datasize = xdp->data_end - xdp->data;
2462 struct tun_xdp_hdr *hdr = xdp->data_hard_start;
2463 struct virtio_net_hdr *gso = &hdr->gso;
2464 struct tun_pcpu_stats *stats;
2465 struct bpf_prog *xdp_prog;
2466 struct sk_buff *skb = NULL;
2467 u32 rxhash = 0, act;
2468 int buflen = hdr->buflen;
2470 bool skb_xdp = false;
2473 xdp_prog = rcu_dereference(tun->xdp_prog);
2475 if (gso->gso_type) {
2479 xdp_set_data_meta_invalid(xdp);
2480 xdp->rxq = &tfile->xdp_rxq;
2481 xdp->frame_sz = buflen;
2483 act = bpf_prog_run_xdp(xdp_prog, xdp);
2484 err = tun_xdp_act(tun, xdp_prog, xdp, act);
2486 put_page(virt_to_head_page(xdp->data));
2499 page = virt_to_head_page(xdp->data);
2500 if (tpage->page == page) {
2503 tun_put_page(tpage);
2512 skb = build_skb(xdp->data_hard_start, buflen);
2518 skb_reserve(skb, xdp->data - xdp->data_hard_start);
2519 skb_put(skb, xdp->data_end - xdp->data);
2521 if (virtio_net_hdr_to_skb(skb, gso, tun_is_little_endian(tun))) {
2522 this_cpu_inc(tun->pcpu_stats->rx_frame_errors);
2528 skb->protocol = eth_type_trans(skb, tun->dev);
2529 skb_reset_network_header(skb);
2530 skb_probe_transport_header(skb);
2531 skb_record_rx_queue(skb, tfile->queue_index);
2534 err = do_xdp_generic(xdp_prog, skb);
2535 if (err != XDP_PASS)
2539 if (!rcu_dereference(tun->steering_prog) && tun->numqueues > 1 &&
2541 rxhash = __skb_get_hash_symmetric(skb);
2543 netif_receive_skb(skb);
2545 /* No need for get_cpu_ptr() here since this function is
2546 * always called with bh disabled
2548 stats = this_cpu_ptr(tun->pcpu_stats);
2549 u64_stats_update_begin(&stats->syncp);
2550 u64_stats_inc(&stats->rx_packets);
2551 u64_stats_add(&stats->rx_bytes, datasize);
2552 u64_stats_update_end(&stats->syncp);
2555 tun_flow_update(tun, rxhash, tfile);
2561 static int tun_sendmsg(struct socket *sock, struct msghdr *m, size_t total_len)
2564 struct tun_file *tfile = container_of(sock, struct tun_file, socket);
2565 struct tun_struct *tun = tun_get(tfile);
2566 struct tun_msg_ctl *ctl = m->msg_control;
2567 struct xdp_buff *xdp;
2572 if (m->msg_controllen == sizeof(struct tun_msg_ctl) &&
2573 ctl && ctl->type == TUN_MSG_PTR) {
2574 struct tun_page tpage;
2578 memset(&tpage, 0, sizeof(tpage));
2583 for (i = 0; i < n; i++) {
2584 xdp = &((struct xdp_buff *)ctl->ptr)[i];
2585 tun_xdp_one(tun, tfile, xdp, &flush, &tpage);
2594 tun_put_page(&tpage);
2600 ret = tun_get_user(tun, tfile, ctl ? ctl->ptr : NULL, &m->msg_iter,
2601 m->msg_flags & MSG_DONTWAIT,
2602 m->msg_flags & MSG_MORE);
2608 static int tun_recvmsg(struct socket *sock, struct msghdr *m, size_t total_len,
2611 struct tun_file *tfile = container_of(sock, struct tun_file, socket);
2612 struct tun_struct *tun = tun_get(tfile);
2613 void *ptr = m->msg_control;
2621 if (flags & ~(MSG_DONTWAIT|MSG_TRUNC|MSG_ERRQUEUE)) {
2625 if (flags & MSG_ERRQUEUE) {
2626 ret = sock_recv_errqueue(sock->sk, m, total_len,
2627 SOL_PACKET, TUN_TX_TIMESTAMP);
2630 ret = tun_do_read(tun, tfile, &m->msg_iter, flags & MSG_DONTWAIT, ptr);
2631 if (ret > (ssize_t)total_len) {
2632 m->msg_flags |= MSG_TRUNC;
2633 ret = flags & MSG_TRUNC ? ret : total_len;
2646 static int tun_ptr_peek_len(void *ptr)
2649 if (tun_is_xdp_frame(ptr)) {
2650 struct xdp_frame *xdpf = tun_ptr_to_xdp(ptr);
2654 return __skb_array_len_with_tag(ptr);
2660 static int tun_peek_len(struct socket *sock)
2662 struct tun_file *tfile = container_of(sock, struct tun_file, socket);
2663 struct tun_struct *tun;
2666 tun = tun_get(tfile);
2670 ret = PTR_RING_PEEK_CALL(&tfile->tx_ring, tun_ptr_peek_len);
2676 /* Ops structure to mimic raw sockets with tun */
2677 static const struct proto_ops tun_socket_ops = {
2678 .peek_len = tun_peek_len,
2679 .sendmsg = tun_sendmsg,
2680 .recvmsg = tun_recvmsg,
2683 static struct proto tun_proto = {
2685 .owner = THIS_MODULE,
2686 .obj_size = sizeof(struct tun_file),
2689 static int tun_flags(struct tun_struct *tun)
2691 return tun->flags & (TUN_FEATURES | IFF_PERSIST | IFF_TUN | IFF_TAP);
2694 static ssize_t tun_show_flags(struct device *dev, struct device_attribute *attr,
2697 struct tun_struct *tun = netdev_priv(to_net_dev(dev));
2698 return sprintf(buf, "0x%x\n", tun_flags(tun));
2701 static ssize_t tun_show_owner(struct device *dev, struct device_attribute *attr,
2704 struct tun_struct *tun = netdev_priv(to_net_dev(dev));
2705 return uid_valid(tun->owner)?
2706 sprintf(buf, "%u\n",
2707 from_kuid_munged(current_user_ns(), tun->owner)):
2708 sprintf(buf, "-1\n");
2711 static ssize_t tun_show_group(struct device *dev, struct device_attribute *attr,
2714 struct tun_struct *tun = netdev_priv(to_net_dev(dev));
2715 return gid_valid(tun->group) ?
2716 sprintf(buf, "%u\n",
2717 from_kgid_munged(current_user_ns(), tun->group)):
2718 sprintf(buf, "-1\n");
2721 static DEVICE_ATTR(tun_flags, 0444, tun_show_flags, NULL);
2722 static DEVICE_ATTR(owner, 0444, tun_show_owner, NULL);
2723 static DEVICE_ATTR(group, 0444, tun_show_group, NULL);
2725 static struct attribute *tun_dev_attrs[] = {
2726 &dev_attr_tun_flags.attr,
2727 &dev_attr_owner.attr,
2728 &dev_attr_group.attr,
2732 static const struct attribute_group tun_attr_group = {
2733 .attrs = tun_dev_attrs
2736 static int tun_set_iff(struct net *net, struct file *file, struct ifreq *ifr)
2738 struct tun_struct *tun;
2739 struct tun_file *tfile = file->private_data;
2740 struct net_device *dev;
2743 if (tfile->detached)
2746 if ((ifr->ifr_flags & IFF_NAPI_FRAGS)) {
2747 if (!capable(CAP_NET_ADMIN))
2750 if (!(ifr->ifr_flags & IFF_NAPI) ||
2751 (ifr->ifr_flags & TUN_TYPE_MASK) != IFF_TAP)
2755 dev = __dev_get_by_name(net, ifr->ifr_name);
2757 if (ifr->ifr_flags & IFF_TUN_EXCL)
2759 if ((ifr->ifr_flags & IFF_TUN) && dev->netdev_ops == &tun_netdev_ops)
2760 tun = netdev_priv(dev);
2761 else if ((ifr->ifr_flags & IFF_TAP) && dev->netdev_ops == &tap_netdev_ops)
2762 tun = netdev_priv(dev);
2766 if (!!(ifr->ifr_flags & IFF_MULTI_QUEUE) !=
2767 !!(tun->flags & IFF_MULTI_QUEUE))
2770 if (tun_not_capable(tun))
2772 err = security_tun_dev_open(tun->security);
2776 err = tun_attach(tun, file, ifr->ifr_flags & IFF_NOFILTER,
2777 ifr->ifr_flags & IFF_NAPI,
2778 ifr->ifr_flags & IFF_NAPI_FRAGS, true);
2782 if (tun->flags & IFF_MULTI_QUEUE &&
2783 (tun->numqueues + tun->numdisabled > 1)) {
2784 /* One or more queue has already been attached, no need
2785 * to initialize the device again.
2787 netdev_state_change(dev);
2791 tun->flags = (tun->flags & ~TUN_FEATURES) |
2792 (ifr->ifr_flags & TUN_FEATURES);
2794 netdev_state_change(dev);
2797 unsigned long flags = 0;
2798 int queues = ifr->ifr_flags & IFF_MULTI_QUEUE ?
2801 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
2803 err = security_tun_dev_create();
2808 if (ifr->ifr_flags & IFF_TUN) {
2812 } else if (ifr->ifr_flags & IFF_TAP) {
2820 name = ifr->ifr_name;
2822 dev = alloc_netdev_mqs(sizeof(struct tun_struct), name,
2823 NET_NAME_UNKNOWN, tun_setup, queues,
2829 dev_net_set(dev, net);
2830 dev->rtnl_link_ops = &tun_link_ops;
2831 dev->ifindex = tfile->ifindex;
2832 dev->sysfs_groups[0] = &tun_attr_group;
2834 tun = netdev_priv(dev);
2837 tun->txflt.count = 0;
2838 tun->vnet_hdr_sz = sizeof(struct virtio_net_hdr);
2840 tun->align = NET_SKB_PAD;
2841 tun->filter_attached = false;
2842 tun->sndbuf = tfile->socket.sk->sk_sndbuf;
2843 tun->rx_batched = 0;
2844 RCU_INIT_POINTER(tun->steering_prog, NULL);
2849 tun_net_initialize(dev);
2851 err = register_netdevice(tun->dev);
2856 /* free_netdev() won't check refcnt, to aovid race
2857 * with dev_put() we need publish tun after registration.
2859 rcu_assign_pointer(tfile->tun, tun);
2862 netif_carrier_on(tun->dev);
2864 /* Make sure persistent devices do not get stuck in
2867 if (netif_running(tun->dev))
2868 netif_tx_wake_all_queues(tun->dev);
2870 strcpy(ifr->ifr_name, tun->dev->name);
2874 static void tun_get_iff(struct tun_struct *tun, struct ifreq *ifr)
2876 strcpy(ifr->ifr_name, tun->dev->name);
2878 ifr->ifr_flags = tun_flags(tun);
2882 /* This is like a cut-down ethtool ops, except done via tun fd so no
2883 * privs required. */
2884 static int set_offload(struct tun_struct *tun, unsigned long arg)
2886 netdev_features_t features = 0;
2888 if (arg & TUN_F_CSUM) {
2889 features |= NETIF_F_HW_CSUM;
2892 if (arg & (TUN_F_TSO4|TUN_F_TSO6)) {
2893 if (arg & TUN_F_TSO_ECN) {
2894 features |= NETIF_F_TSO_ECN;
2895 arg &= ~TUN_F_TSO_ECN;
2897 if (arg & TUN_F_TSO4)
2898 features |= NETIF_F_TSO;
2899 if (arg & TUN_F_TSO6)
2900 features |= NETIF_F_TSO6;
2901 arg &= ~(TUN_F_TSO4|TUN_F_TSO6);
2907 /* This gives the user a way to test for new features in future by
2908 * trying to set them. */
2912 tun->set_features = features;
2913 tun->dev->wanted_features &= ~TUN_USER_FEATURES;
2914 tun->dev->wanted_features |= features;
2915 netdev_update_features(tun->dev);
2920 static void tun_detach_filter(struct tun_struct *tun, int n)
2923 struct tun_file *tfile;
2925 for (i = 0; i < n; i++) {
2926 tfile = rtnl_dereference(tun->tfiles[i]);
2927 lock_sock(tfile->socket.sk);
2928 sk_detach_filter(tfile->socket.sk);
2929 release_sock(tfile->socket.sk);
2932 tun->filter_attached = false;
2935 static int tun_attach_filter(struct tun_struct *tun)
2938 struct tun_file *tfile;
2940 for (i = 0; i < tun->numqueues; i++) {
2941 tfile = rtnl_dereference(tun->tfiles[i]);
2942 lock_sock(tfile->socket.sk);
2943 ret = sk_attach_filter(&tun->fprog, tfile->socket.sk);
2944 release_sock(tfile->socket.sk);
2946 tun_detach_filter(tun, i);
2951 tun->filter_attached = true;
2955 static void tun_set_sndbuf(struct tun_struct *tun)
2957 struct tun_file *tfile;
2960 for (i = 0; i < tun->numqueues; i++) {
2961 tfile = rtnl_dereference(tun->tfiles[i]);
2962 tfile->socket.sk->sk_sndbuf = tun->sndbuf;
2966 static int tun_set_queue(struct file *file, struct ifreq *ifr)
2968 struct tun_file *tfile = file->private_data;
2969 struct tun_struct *tun;
2974 if (ifr->ifr_flags & IFF_ATTACH_QUEUE) {
2975 tun = tfile->detached;
2980 ret = security_tun_dev_attach_queue(tun->security);
2983 ret = tun_attach(tun, file, false, tun->flags & IFF_NAPI,
2984 tun->flags & IFF_NAPI_FRAGS, true);
2985 } else if (ifr->ifr_flags & IFF_DETACH_QUEUE) {
2986 tun = rtnl_dereference(tfile->tun);
2987 if (!tun || !(tun->flags & IFF_MULTI_QUEUE) || tfile->detached)
2990 __tun_detach(tfile, false);
2995 netdev_state_change(tun->dev);
3002 static int tun_set_ebpf(struct tun_struct *tun, struct tun_prog __rcu **prog_p,
3005 struct bpf_prog *prog;
3008 if (copy_from_user(&fd, data, sizeof(fd)))
3014 prog = bpf_prog_get_type(fd, BPF_PROG_TYPE_SOCKET_FILTER);
3016 return PTR_ERR(prog);
3019 return __tun_set_ebpf(tun, prog_p, prog);
3022 /* Return correct value for tun->dev->addr_len based on tun->dev->type. */
3023 static unsigned char tun_get_addr_len(unsigned short type)
3027 case ARPHRD_TUNNEL6:
3028 return sizeof(struct in6_addr);
3035 case ARPHRD_IEEE802154:
3036 case ARPHRD_IEEE802154_MONITOR:
3037 return IEEE802154_EXTENDED_ADDR_LEN;
3038 case ARPHRD_PHONET_PIPE:
3042 case ARPHRD_6LOWPAN:
3043 return EUI64_ADDR_LEN;
3048 case ARPHRD_IEEE802:
3051 return ROSE_ADDR_LEN;
3053 return AX25_ADDR_LEN;
3054 case ARPHRD_LOCALTLK:
3061 static long __tun_chr_ioctl(struct file *file, unsigned int cmd,
3062 unsigned long arg, int ifreq_len)
3064 struct tun_file *tfile = file->private_data;
3065 struct net *net = sock_net(&tfile->sk);
3066 struct tun_struct *tun;
3067 void __user* argp = (void __user*)arg;
3068 unsigned int carrier;
3077 bool do_notify = false;
3079 if (cmd == TUNSETIFF || cmd == TUNSETQUEUE ||
3080 (_IOC_TYPE(cmd) == SOCK_IOC_TYPE && cmd != SIOCGSKNS)) {
3081 if (copy_from_user(&ifr, argp, ifreq_len))
3084 memset(&ifr, 0, sizeof(ifr));
3086 if (cmd == TUNGETFEATURES) {
3087 /* Currently this just means: "what IFF flags are valid?".
3088 * This is needed because we never checked for invalid flags on
3091 return put_user(IFF_TUN | IFF_TAP | TUN_FEATURES,
3092 (unsigned int __user*)argp);
3093 } else if (cmd == TUNSETQUEUE) {
3094 return tun_set_queue(file, &ifr);
3095 } else if (cmd == SIOCGSKNS) {
3096 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
3098 return open_related_ns(&net->ns, get_net_ns);
3104 tun = tun_get(tfile);
3105 if (cmd == TUNSETIFF) {
3110 ifr.ifr_name[IFNAMSIZ-1] = '\0';
3112 ret = tun_set_iff(net, file, &ifr);
3117 if (copy_to_user(argp, &ifr, ifreq_len))
3121 if (cmd == TUNSETIFINDEX) {
3127 if (copy_from_user(&ifindex, argp, sizeof(ifindex)))
3133 tfile->ifindex = ifindex;
3141 netif_info(tun, drv, tun->dev, "tun_chr_ioctl cmd %u\n", cmd);
3143 net = dev_net(tun->dev);
3147 tun_get_iff(tun, &ifr);
3149 if (tfile->detached)
3150 ifr.ifr_flags |= IFF_DETACH_QUEUE;
3151 if (!tfile->socket.sk->sk_filter)
3152 ifr.ifr_flags |= IFF_NOFILTER;
3154 if (copy_to_user(argp, &ifr, ifreq_len))
3159 /* Disable/Enable checksum */
3161 /* [unimplemented] */
3162 netif_info(tun, drv, tun->dev, "ignored: set checksum %s\n",
3163 arg ? "disabled" : "enabled");
3167 /* Disable/Enable persist mode. Keep an extra reference to the
3168 * module to prevent the module being unprobed.
3170 if (arg && !(tun->flags & IFF_PERSIST)) {
3171 tun->flags |= IFF_PERSIST;
3172 __module_get(THIS_MODULE);
3175 if (!arg && (tun->flags & IFF_PERSIST)) {
3176 tun->flags &= ~IFF_PERSIST;
3177 module_put(THIS_MODULE);
3181 netif_info(tun, drv, tun->dev, "persist %s\n",
3182 arg ? "enabled" : "disabled");
3186 /* Set owner of the device */
3187 owner = make_kuid(current_user_ns(), arg);
3188 if (!uid_valid(owner)) {
3194 netif_info(tun, drv, tun->dev, "owner set to %u\n",
3195 from_kuid(&init_user_ns, tun->owner));
3199 /* Set group of the device */
3200 group = make_kgid(current_user_ns(), arg);
3201 if (!gid_valid(group)) {
3207 netif_info(tun, drv, tun->dev, "group set to %u\n",
3208 from_kgid(&init_user_ns, tun->group));
3212 /* Only allow setting the type when the interface is down */
3213 if (tun->dev->flags & IFF_UP) {
3214 netif_info(tun, drv, tun->dev,
3215 "Linktype set failed because interface is up\n");
3218 tun->dev->type = (int) arg;
3219 tun->dev->addr_len = tun_get_addr_len(tun->dev->type);
3220 netif_info(tun, drv, tun->dev, "linktype set to %d\n",
3227 tun->msg_enable = (u32)arg;
3231 ret = set_offload(tun, arg);
3234 case TUNSETTXFILTER:
3235 /* Can be set only for TAPs */
3237 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3239 ret = update_filter(&tun->txflt, (void __user *)arg);
3243 /* Get hw address */
3244 dev_get_mac_address(&ifr.ifr_hwaddr, net, tun->dev->name);
3245 if (copy_to_user(argp, &ifr, ifreq_len))
3250 /* Set hw address */
3251 ret = dev_set_mac_address_user(tun->dev, &ifr.ifr_hwaddr, NULL);
3255 sndbuf = tfile->socket.sk->sk_sndbuf;
3256 if (copy_to_user(argp, &sndbuf, sizeof(sndbuf)))
3261 if (copy_from_user(&sndbuf, argp, sizeof(sndbuf))) {
3270 tun->sndbuf = sndbuf;
3271 tun_set_sndbuf(tun);
3274 case TUNGETVNETHDRSZ:
3275 vnet_hdr_sz = tun->vnet_hdr_sz;
3276 if (copy_to_user(argp, &vnet_hdr_sz, sizeof(vnet_hdr_sz)))
3280 case TUNSETVNETHDRSZ:
3281 if (copy_from_user(&vnet_hdr_sz, argp, sizeof(vnet_hdr_sz))) {
3285 if (vnet_hdr_sz < (int)sizeof(struct virtio_net_hdr)) {
3290 tun->vnet_hdr_sz = vnet_hdr_sz;
3294 le = !!(tun->flags & TUN_VNET_LE);
3295 if (put_user(le, (int __user *)argp))
3300 if (get_user(le, (int __user *)argp)) {
3305 tun->flags |= TUN_VNET_LE;
3307 tun->flags &= ~TUN_VNET_LE;
3311 ret = tun_get_vnet_be(tun, argp);
3315 ret = tun_set_vnet_be(tun, argp);
3318 case TUNATTACHFILTER:
3319 /* Can be set only for TAPs */
3321 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3324 if (copy_from_user(&tun->fprog, argp, sizeof(tun->fprog)))
3327 ret = tun_attach_filter(tun);
3330 case TUNDETACHFILTER:
3331 /* Can be set only for TAPs */
3333 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3336 tun_detach_filter(tun, tun->numqueues);
3341 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3344 if (copy_to_user(argp, &tun->fprog, sizeof(tun->fprog)))
3349 case TUNSETSTEERINGEBPF:
3350 ret = tun_set_ebpf(tun, &tun->steering_prog, argp);
3353 case TUNSETFILTEREBPF:
3354 ret = tun_set_ebpf(tun, &tun->filter_prog, argp);
3359 if (copy_from_user(&carrier, argp, sizeof(carrier)))
3362 ret = tun_net_change_carrier(tun->dev, (bool)carrier);
3365 case TUNGETDEVNETNS:
3367 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
3369 ret = open_related_ns(&net->ns, get_net_ns);
3378 netdev_state_change(tun->dev);
3387 static long tun_chr_ioctl(struct file *file,
3388 unsigned int cmd, unsigned long arg)
3390 return __tun_chr_ioctl(file, cmd, arg, sizeof (struct ifreq));
3393 #ifdef CONFIG_COMPAT
3394 static long tun_chr_compat_ioctl(struct file *file,
3395 unsigned int cmd, unsigned long arg)
3400 case TUNSETTXFILTER:
3405 arg = (unsigned long)compat_ptr(arg);
3408 arg = (compat_ulong_t)arg;
3413 * compat_ifreq is shorter than ifreq, so we must not access beyond
3414 * the end of that structure. All fields that are used in this
3415 * driver are compatible though, we don't need to convert the
3418 return __tun_chr_ioctl(file, cmd, arg, sizeof(struct compat_ifreq));
3420 #endif /* CONFIG_COMPAT */
3422 static int tun_chr_fasync(int fd, struct file *file, int on)
3424 struct tun_file *tfile = file->private_data;
3427 if ((ret = fasync_helper(fd, file, on, &tfile->fasync)) < 0)
3431 __f_setown(file, task_pid(current), PIDTYPE_TGID, 0);
3432 tfile->flags |= TUN_FASYNC;
3434 tfile->flags &= ~TUN_FASYNC;
3440 static int tun_chr_open(struct inode *inode, struct file * file)
3442 struct net *net = current->nsproxy->net_ns;
3443 struct tun_file *tfile;
3445 tfile = (struct tun_file *)sk_alloc(net, AF_UNSPEC, GFP_KERNEL,
3449 if (ptr_ring_init(&tfile->tx_ring, 0, GFP_KERNEL)) {
3450 sk_free(&tfile->sk);
3454 mutex_init(&tfile->napi_mutex);
3455 RCU_INIT_POINTER(tfile->tun, NULL);
3459 init_waitqueue_head(&tfile->socket.wq.wait);
3461 tfile->socket.file = file;
3462 tfile->socket.ops = &tun_socket_ops;
3464 sock_init_data_uid(&tfile->socket, &tfile->sk, current_fsuid());
3466 tfile->sk.sk_write_space = tun_sock_write_space;
3467 tfile->sk.sk_sndbuf = INT_MAX;
3469 file->private_data = tfile;
3470 INIT_LIST_HEAD(&tfile->next);
3472 sock_set_flag(&tfile->sk, SOCK_ZEROCOPY);
3477 static int tun_chr_close(struct inode *inode, struct file *file)
3479 struct tun_file *tfile = file->private_data;
3481 tun_detach(tfile, true);
3486 #ifdef CONFIG_PROC_FS
3487 static void tun_chr_show_fdinfo(struct seq_file *m, struct file *file)
3489 struct tun_file *tfile = file->private_data;
3490 struct tun_struct *tun;
3493 memset(&ifr, 0, sizeof(ifr));
3496 tun = tun_get(tfile);
3498 tun_get_iff(tun, &ifr);
3504 seq_printf(m, "iff:\t%s\n", ifr.ifr_name);
3508 static const struct file_operations tun_fops = {
3509 .owner = THIS_MODULE,
3510 .llseek = no_llseek,
3511 .read_iter = tun_chr_read_iter,
3512 .write_iter = tun_chr_write_iter,
3513 .poll = tun_chr_poll,
3514 .unlocked_ioctl = tun_chr_ioctl,
3515 #ifdef CONFIG_COMPAT
3516 .compat_ioctl = tun_chr_compat_ioctl,
3518 .open = tun_chr_open,
3519 .release = tun_chr_close,
3520 .fasync = tun_chr_fasync,
3521 #ifdef CONFIG_PROC_FS
3522 .show_fdinfo = tun_chr_show_fdinfo,
3526 static struct miscdevice tun_miscdev = {
3529 .nodename = "net/tun",
3533 /* ethtool interface */
3535 static void tun_default_link_ksettings(struct net_device *dev,
3536 struct ethtool_link_ksettings *cmd)
3538 ethtool_link_ksettings_zero_link_mode(cmd, supported);
3539 ethtool_link_ksettings_zero_link_mode(cmd, advertising);
3540 cmd->base.speed = SPEED_10;
3541 cmd->base.duplex = DUPLEX_FULL;
3542 cmd->base.port = PORT_TP;
3543 cmd->base.phy_address = 0;
3544 cmd->base.autoneg = AUTONEG_DISABLE;
3547 static int tun_get_link_ksettings(struct net_device *dev,
3548 struct ethtool_link_ksettings *cmd)
3550 struct tun_struct *tun = netdev_priv(dev);
3552 memcpy(cmd, &tun->link_ksettings, sizeof(*cmd));
3556 static int tun_set_link_ksettings(struct net_device *dev,
3557 const struct ethtool_link_ksettings *cmd)
3559 struct tun_struct *tun = netdev_priv(dev);
3561 memcpy(&tun->link_ksettings, cmd, sizeof(*cmd));
3565 static void tun_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
3567 struct tun_struct *tun = netdev_priv(dev);
3569 strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
3570 strlcpy(info->version, DRV_VERSION, sizeof(info->version));
3572 switch (tun->flags & TUN_TYPE_MASK) {
3574 strlcpy(info->bus_info, "tun", sizeof(info->bus_info));
3577 strlcpy(info->bus_info, "tap", sizeof(info->bus_info));
3582 static u32 tun_get_msglevel(struct net_device *dev)
3584 struct tun_struct *tun = netdev_priv(dev);
3586 return tun->msg_enable;
3589 static void tun_set_msglevel(struct net_device *dev, u32 value)
3591 struct tun_struct *tun = netdev_priv(dev);
3593 tun->msg_enable = value;
3596 static int tun_get_coalesce(struct net_device *dev,
3597 struct ethtool_coalesce *ec)
3599 struct tun_struct *tun = netdev_priv(dev);
3601 ec->rx_max_coalesced_frames = tun->rx_batched;
3606 static int tun_set_coalesce(struct net_device *dev,
3607 struct ethtool_coalesce *ec)
3609 struct tun_struct *tun = netdev_priv(dev);
3611 if (ec->rx_max_coalesced_frames > NAPI_POLL_WEIGHT)
3612 tun->rx_batched = NAPI_POLL_WEIGHT;
3614 tun->rx_batched = ec->rx_max_coalesced_frames;
3619 static const struct ethtool_ops tun_ethtool_ops = {
3620 .supported_coalesce_params = ETHTOOL_COALESCE_RX_MAX_FRAMES,
3621 .get_drvinfo = tun_get_drvinfo,
3622 .get_msglevel = tun_get_msglevel,
3623 .set_msglevel = tun_set_msglevel,
3624 .get_link = ethtool_op_get_link,
3625 .get_ts_info = ethtool_op_get_ts_info,
3626 .get_coalesce = tun_get_coalesce,
3627 .set_coalesce = tun_set_coalesce,
3628 .get_link_ksettings = tun_get_link_ksettings,
3629 .set_link_ksettings = tun_set_link_ksettings,
3632 static int tun_queue_resize(struct tun_struct *tun)
3634 struct net_device *dev = tun->dev;
3635 struct tun_file *tfile;
3636 struct ptr_ring **rings;
3637 int n = tun->numqueues + tun->numdisabled;
3640 rings = kmalloc_array(n, sizeof(*rings), GFP_KERNEL);
3644 for (i = 0; i < tun->numqueues; i++) {
3645 tfile = rtnl_dereference(tun->tfiles[i]);
3646 rings[i] = &tfile->tx_ring;
3648 list_for_each_entry(tfile, &tun->disabled, next)
3649 rings[i++] = &tfile->tx_ring;
3651 ret = ptr_ring_resize_multiple(rings, n,
3652 dev->tx_queue_len, GFP_KERNEL,
3659 static int tun_device_event(struct notifier_block *unused,
3660 unsigned long event, void *ptr)
3662 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
3663 struct tun_struct *tun = netdev_priv(dev);
3666 if (dev->rtnl_link_ops != &tun_link_ops)
3670 case NETDEV_CHANGE_TX_QUEUE_LEN:
3671 if (tun_queue_resize(tun))
3675 for (i = 0; i < tun->numqueues; i++) {
3676 struct tun_file *tfile;
3678 tfile = rtnl_dereference(tun->tfiles[i]);
3679 tfile->socket.sk->sk_write_space(tfile->socket.sk);
3689 static struct notifier_block tun_notifier_block __read_mostly = {
3690 .notifier_call = tun_device_event,
3693 static int __init tun_init(void)
3697 pr_info("%s, %s\n", DRV_DESCRIPTION, DRV_VERSION);
3699 ret = rtnl_link_register(&tun_link_ops);
3701 pr_err("Can't register link_ops\n");
3705 ret = misc_register(&tun_miscdev);
3707 pr_err("Can't register misc device %d\n", TUN_MINOR);
3711 ret = register_netdevice_notifier(&tun_notifier_block);
3713 pr_err("Can't register netdevice notifier\n");
3720 misc_deregister(&tun_miscdev);
3722 rtnl_link_unregister(&tun_link_ops);
3727 static void tun_cleanup(void)
3729 misc_deregister(&tun_miscdev);
3730 rtnl_link_unregister(&tun_link_ops);
3731 unregister_netdevice_notifier(&tun_notifier_block);
3734 /* Get an underlying socket object from tun file. Returns error unless file is
3735 * attached to a device. The returned object works like a packet socket, it
3736 * can be used for sock_sendmsg/sock_recvmsg. The caller is responsible for
3737 * holding a reference to the file for as long as the socket is in use. */
3738 struct socket *tun_get_socket(struct file *file)
3740 struct tun_file *tfile;
3741 if (file->f_op != &tun_fops)
3742 return ERR_PTR(-EINVAL);
3743 tfile = file->private_data;
3745 return ERR_PTR(-EBADFD);
3746 return &tfile->socket;
3748 EXPORT_SYMBOL_GPL(tun_get_socket);
3750 struct ptr_ring *tun_get_tx_ring(struct file *file)
3752 struct tun_file *tfile;
3754 if (file->f_op != &tun_fops)
3755 return ERR_PTR(-EINVAL);
3756 tfile = file->private_data;
3758 return ERR_PTR(-EBADFD);
3759 return &tfile->tx_ring;
3761 EXPORT_SYMBOL_GPL(tun_get_tx_ring);
3763 module_init(tun_init);
3764 module_exit(tun_cleanup);
3765 MODULE_DESCRIPTION(DRV_DESCRIPTION);
3766 MODULE_AUTHOR(DRV_COPYRIGHT);
3767 MODULE_LICENSE("GPL");
3768 MODULE_ALIAS_MISCDEV(TUN_MINOR);
3769 MODULE_ALIAS("devname:net/tun");