2 * TUN - Universal TUN/TAP device driver.
3 * Copyright (C) 1999-2002 Maxim Krasnyansky <maxk@qualcomm.com>
5 * This program is free software; you can redistribute it and/or modify
6 * it under the terms of the GNU General Public License as published by
7 * the Free Software Foundation; either version 2 of the License, or
8 * (at your option) any later version.
10 * This program is distributed in the hope that it will be useful,
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
13 * GNU General Public License for more details.
15 * $Id: tun.c,v 1.15 2002/03/01 02:44:24 maxk Exp $
21 * Mike Kershaw <dragorn@kismetwireless.net> 2005/08/14
22 * Add TUNSETLINK ioctl to set the link encapsulation
24 * Mark Smith <markzzzsmith@yahoo.com.au>
25 * Use eth_random_addr() for tap MAC address.
27 * Harald Roelle <harald.roelle@ifi.lmu.de> 2004/04/20
28 * Fixes in packet dropping, queue length setting and queue wakeup.
29 * Increased default tx queue length.
33 * Daniel Podlejski <underley@underley.eu.org>
34 * Modifications for 2.3.99-pre5 kernel.
37 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
39 #define DRV_NAME "tun"
40 #define DRV_VERSION "1.6"
41 #define DRV_DESCRIPTION "Universal TUN/TAP device driver"
42 #define DRV_COPYRIGHT "(C) 1999-2004 Max Krasnyansky <maxk@qualcomm.com>"
44 #include <linux/module.h>
45 #include <linux/errno.h>
46 #include <linux/kernel.h>
47 #include <linux/sched/signal.h>
48 #include <linux/major.h>
49 #include <linux/slab.h>
50 #include <linux/poll.h>
51 #include <linux/fcntl.h>
52 #include <linux/init.h>
53 #include <linux/skbuff.h>
54 #include <linux/netdevice.h>
55 #include <linux/etherdevice.h>
56 #include <linux/miscdevice.h>
57 #include <linux/ethtool.h>
58 #include <linux/rtnetlink.h>
59 #include <linux/compat.h>
61 #include <linux/if_arp.h>
62 #include <linux/if_ether.h>
63 #include <linux/if_tun.h>
64 #include <linux/if_vlan.h>
65 #include <linux/crc32.h>
66 #include <linux/nsproxy.h>
67 #include <linux/virtio_net.h>
68 #include <linux/rcupdate.h>
69 #include <net/net_namespace.h>
70 #include <net/netns/generic.h>
71 #include <net/rtnetlink.h>
73 #include <linux/seq_file.h>
74 #include <linux/uio.h>
75 #include <linux/skb_array.h>
76 #include <linux/bpf.h>
77 #include <linux/bpf_trace.h>
78 #include <linux/mutex.h>
80 #include <linux/uaccess.h>
81 #include <linux/proc_fs.h>
83 /* Uncomment to enable debugging */
84 /* #define TUN_DEBUG 1 */
89 #define tun_debug(level, tun, fmt, args...) \
92 netdev_printk(level, tun->dev, fmt, ##args); \
94 #define DBG1(level, fmt, args...) \
97 printk(level fmt, ##args); \
100 #define tun_debug(level, tun, fmt, args...) \
103 netdev_printk(level, tun->dev, fmt, ##args); \
105 #define DBG1(level, fmt, args...) \
108 printk(level fmt, ##args); \
112 #define TUN_HEADROOM 256
113 #define TUN_RX_PAD (NET_IP_ALIGN + NET_SKB_PAD)
115 /* TUN device flags */
117 /* IFF_ATTACH_QUEUE is never stored in device flags,
118 * overload it to mean fasync when stored there.
120 #define TUN_FASYNC IFF_ATTACH_QUEUE
121 /* High bits in flags field are unused. */
122 #define TUN_VNET_LE 0x80000000
123 #define TUN_VNET_BE 0x40000000
125 #define TUN_FEATURES (IFF_NO_PI | IFF_ONE_QUEUE | IFF_VNET_HDR | \
126 IFF_MULTI_QUEUE | IFF_NAPI | IFF_NAPI_FRAGS)
128 #define GOODCOPY_LEN 128
130 #define FLT_EXACT_COUNT 8
132 unsigned int count; /* Number of addrs. Zero means disabled */
133 u32 mask[2]; /* Mask of the hashed addrs */
134 unsigned char addr[FLT_EXACT_COUNT][ETH_ALEN];
137 /* MAX_TAP_QUEUES 256 is chosen to allow rx/tx queues to be equal
138 * to max number of VCPUs in guest. */
139 #define MAX_TAP_QUEUES 256
140 #define MAX_TAP_FLOWS 4096
142 #define TUN_FLOW_EXPIRE (3 * HZ)
144 struct tun_pcpu_stats {
149 struct u64_stats_sync syncp;
155 /* A tun_file connects an open character device to a tuntap netdevice. It
156 * also contains all socket related structures (except sock_fprog and tap_filter)
157 * to serve as one transmit queue for tuntap device. The sock_fprog and
158 * tap_filter were kept in tun_struct since they were used for filtering for the
159 * netdevice not for a specific queue (at least I didn't see the requirement for
163 * The tun_file and tun_struct are loosely coupled, the pointer from one to the
164 * other can only be read while rcu_read_lock or rtnl_lock is held.
168 struct socket socket;
170 struct tun_struct __rcu *tun;
171 struct fasync_struct *fasync;
172 /* only used for fasnyc */
176 unsigned int ifindex;
178 struct napi_struct napi;
180 struct mutex napi_mutex; /* Protects access to the above napi */
181 struct list_head next;
182 struct tun_struct *detached;
183 struct ptr_ring tx_ring;
184 struct xdp_rxq_info xdp_rxq;
187 struct tun_flow_entry {
188 struct hlist_node hash_link;
190 struct tun_struct *tun;
195 unsigned long updated;
198 #define TUN_NUM_FLOW_ENTRIES 1024
202 struct bpf_prog *prog;
205 /* Since the socket were moved to tun_file, to preserve the behavior of persist
206 * device, socket filter, sndbuf and vnet header size were restore when the
207 * file were attached to a persist device.
210 struct tun_file __rcu *tfiles[MAX_TAP_QUEUES];
211 unsigned int numqueues;
216 struct net_device *dev;
217 netdev_features_t set_features;
218 #define TUN_USER_FEATURES (NETIF_F_HW_CSUM|NETIF_F_TSO_ECN|NETIF_F_TSO| \
224 struct tap_filter txflt;
225 struct sock_fprog fprog;
226 /* protected by rtnl lock */
227 bool filter_attached;
232 struct hlist_head flows[TUN_NUM_FLOW_ENTRIES];
233 struct timer_list flow_gc_timer;
234 unsigned long ageing_time;
235 unsigned int numdisabled;
236 struct list_head disabled;
240 struct tun_pcpu_stats __percpu *pcpu_stats;
241 struct bpf_prog __rcu *xdp_prog;
242 struct tun_prog __rcu *steering_prog;
243 struct tun_prog __rcu *filter_prog;
251 bool tun_is_xdp_frame(void *ptr)
253 return (unsigned long)ptr & TUN_XDP_FLAG;
255 EXPORT_SYMBOL(tun_is_xdp_frame);
257 void *tun_xdp_to_ptr(void *ptr)
259 return (void *)((unsigned long)ptr | TUN_XDP_FLAG);
261 EXPORT_SYMBOL(tun_xdp_to_ptr);
263 void *tun_ptr_to_xdp(void *ptr)
265 return (void *)((unsigned long)ptr & ~TUN_XDP_FLAG);
267 EXPORT_SYMBOL(tun_ptr_to_xdp);
269 static int tun_napi_receive(struct napi_struct *napi, int budget)
271 struct tun_file *tfile = container_of(napi, struct tun_file, napi);
272 struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
273 struct sk_buff_head process_queue;
277 __skb_queue_head_init(&process_queue);
279 spin_lock(&queue->lock);
280 skb_queue_splice_tail_init(queue, &process_queue);
281 spin_unlock(&queue->lock);
283 while (received < budget && (skb = __skb_dequeue(&process_queue))) {
284 napi_gro_receive(napi, skb);
288 if (!skb_queue_empty(&process_queue)) {
289 spin_lock(&queue->lock);
290 skb_queue_splice(&process_queue, queue);
291 spin_unlock(&queue->lock);
297 static int tun_napi_poll(struct napi_struct *napi, int budget)
299 unsigned int received;
301 received = tun_napi_receive(napi, budget);
303 if (received < budget)
304 napi_complete_done(napi, received);
309 static void tun_napi_init(struct tun_struct *tun, struct tun_file *tfile,
312 tfile->napi_enabled = napi_en;
314 netif_napi_add(tun->dev, &tfile->napi, tun_napi_poll,
316 napi_enable(&tfile->napi);
317 mutex_init(&tfile->napi_mutex);
321 static void tun_napi_disable(struct tun_struct *tun, struct tun_file *tfile)
323 if (tfile->napi_enabled)
324 napi_disable(&tfile->napi);
327 static void tun_napi_del(struct tun_struct *tun, struct tun_file *tfile)
329 if (tfile->napi_enabled)
330 netif_napi_del(&tfile->napi);
333 static bool tun_napi_frags_enabled(const struct tun_struct *tun)
335 return READ_ONCE(tun->flags) & IFF_NAPI_FRAGS;
338 #ifdef CONFIG_TUN_VNET_CROSS_LE
339 static inline bool tun_legacy_is_little_endian(struct tun_struct *tun)
341 return tun->flags & TUN_VNET_BE ? false :
342 virtio_legacy_is_little_endian();
345 static long tun_get_vnet_be(struct tun_struct *tun, int __user *argp)
347 int be = !!(tun->flags & TUN_VNET_BE);
349 if (put_user(be, argp))
355 static long tun_set_vnet_be(struct tun_struct *tun, int __user *argp)
359 if (get_user(be, argp))
363 tun->flags |= TUN_VNET_BE;
365 tun->flags &= ~TUN_VNET_BE;
370 static inline bool tun_legacy_is_little_endian(struct tun_struct *tun)
372 return virtio_legacy_is_little_endian();
375 static long tun_get_vnet_be(struct tun_struct *tun, int __user *argp)
380 static long tun_set_vnet_be(struct tun_struct *tun, int __user *argp)
384 #endif /* CONFIG_TUN_VNET_CROSS_LE */
386 static inline bool tun_is_little_endian(struct tun_struct *tun)
388 return tun->flags & TUN_VNET_LE ||
389 tun_legacy_is_little_endian(tun);
392 static inline u16 tun16_to_cpu(struct tun_struct *tun, __virtio16 val)
394 return __virtio16_to_cpu(tun_is_little_endian(tun), val);
397 static inline __virtio16 cpu_to_tun16(struct tun_struct *tun, u16 val)
399 return __cpu_to_virtio16(tun_is_little_endian(tun), val);
402 static inline u32 tun_hashfn(u32 rxhash)
404 return rxhash & 0x3ff;
407 static struct tun_flow_entry *tun_flow_find(struct hlist_head *head, u32 rxhash)
409 struct tun_flow_entry *e;
411 hlist_for_each_entry_rcu(e, head, hash_link) {
412 if (e->rxhash == rxhash)
418 static struct tun_flow_entry *tun_flow_create(struct tun_struct *tun,
419 struct hlist_head *head,
420 u32 rxhash, u16 queue_index)
422 struct tun_flow_entry *e = kmalloc(sizeof(*e), GFP_ATOMIC);
425 tun_debug(KERN_INFO, tun, "create flow: hash %u index %u\n",
426 rxhash, queue_index);
427 e->updated = jiffies;
430 e->queue_index = queue_index;
432 hlist_add_head_rcu(&e->hash_link, head);
438 static void tun_flow_delete(struct tun_struct *tun, struct tun_flow_entry *e)
440 tun_debug(KERN_INFO, tun, "delete flow: hash %u index %u\n",
441 e->rxhash, e->queue_index);
442 hlist_del_rcu(&e->hash_link);
447 static void tun_flow_flush(struct tun_struct *tun)
451 spin_lock_bh(&tun->lock);
452 for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
453 struct tun_flow_entry *e;
454 struct hlist_node *n;
456 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link)
457 tun_flow_delete(tun, e);
459 spin_unlock_bh(&tun->lock);
462 static void tun_flow_delete_by_queue(struct tun_struct *tun, u16 queue_index)
466 spin_lock_bh(&tun->lock);
467 for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
468 struct tun_flow_entry *e;
469 struct hlist_node *n;
471 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
472 if (e->queue_index == queue_index)
473 tun_flow_delete(tun, e);
476 spin_unlock_bh(&tun->lock);
479 static void tun_flow_cleanup(struct timer_list *t)
481 struct tun_struct *tun = from_timer(tun, t, flow_gc_timer);
482 unsigned long delay = tun->ageing_time;
483 unsigned long next_timer = jiffies + delay;
484 unsigned long count = 0;
487 tun_debug(KERN_INFO, tun, "tun_flow_cleanup\n");
489 spin_lock(&tun->lock);
490 for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
491 struct tun_flow_entry *e;
492 struct hlist_node *n;
494 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
495 unsigned long this_timer;
497 this_timer = e->updated + delay;
498 if (time_before_eq(this_timer, jiffies)) {
499 tun_flow_delete(tun, e);
503 if (time_before(this_timer, next_timer))
504 next_timer = this_timer;
509 mod_timer(&tun->flow_gc_timer, round_jiffies_up(next_timer));
510 spin_unlock(&tun->lock);
513 static void tun_flow_update(struct tun_struct *tun, u32 rxhash,
514 struct tun_file *tfile)
516 struct hlist_head *head;
517 struct tun_flow_entry *e;
518 unsigned long delay = tun->ageing_time;
519 u16 queue_index = tfile->queue_index;
524 head = &tun->flows[tun_hashfn(rxhash)];
528 e = tun_flow_find(head, rxhash);
530 /* TODO: keep queueing to old queue until it's empty? */
531 e->queue_index = queue_index;
532 e->updated = jiffies;
533 sock_rps_record_flow_hash(e->rps_rxhash);
535 spin_lock_bh(&tun->lock);
536 if (!tun_flow_find(head, rxhash) &&
537 tun->flow_count < MAX_TAP_FLOWS)
538 tun_flow_create(tun, head, rxhash, queue_index);
540 if (!timer_pending(&tun->flow_gc_timer))
541 mod_timer(&tun->flow_gc_timer,
542 round_jiffies_up(jiffies + delay));
543 spin_unlock_bh(&tun->lock);
550 * Save the hash received in the stack receive path and update the
551 * flow_hash table accordingly.
553 static inline void tun_flow_save_rps_rxhash(struct tun_flow_entry *e, u32 hash)
555 if (unlikely(e->rps_rxhash != hash))
556 e->rps_rxhash = hash;
559 /* We try to identify a flow through its rxhash first. The reason that
560 * we do not check rxq no. is because some cards(e.g 82599), chooses
561 * the rxq based on the txq where the last packet of the flow comes. As
562 * the userspace application move between processors, we may get a
563 * different rxq no. here. If we could not get rxhash, then we would
564 * hope the rxq no. may help here.
566 static u16 tun_automq_select_queue(struct tun_struct *tun, struct sk_buff *skb)
568 struct tun_flow_entry *e;
572 numqueues = READ_ONCE(tun->numqueues);
574 txq = __skb_get_hash_symmetric(skb);
576 e = tun_flow_find(&tun->flows[tun_hashfn(txq)], txq);
578 tun_flow_save_rps_rxhash(e, txq);
579 txq = e->queue_index;
581 /* use multiply and shift instead of expensive divide */
582 txq = ((u64)txq * numqueues) >> 32;
583 } else if (likely(skb_rx_queue_recorded(skb))) {
584 txq = skb_get_rx_queue(skb);
585 while (unlikely(txq >= numqueues))
592 static u16 tun_ebpf_select_queue(struct tun_struct *tun, struct sk_buff *skb)
594 struct tun_prog *prog;
597 prog = rcu_dereference(tun->steering_prog);
599 ret = bpf_prog_run_clear_cb(prog->prog, skb);
601 return ret % tun->numqueues;
604 static u16 tun_select_queue(struct net_device *dev, struct sk_buff *skb,
605 void *accel_priv, select_queue_fallback_t fallback)
607 struct tun_struct *tun = netdev_priv(dev);
611 if (rcu_dereference(tun->steering_prog))
612 ret = tun_ebpf_select_queue(tun, skb);
614 ret = tun_automq_select_queue(tun, skb);
620 static inline bool tun_not_capable(struct tun_struct *tun)
622 const struct cred *cred = current_cred();
623 struct net *net = dev_net(tun->dev);
625 return ((uid_valid(tun->owner) && !uid_eq(cred->euid, tun->owner)) ||
626 (gid_valid(tun->group) && !in_egroup_p(tun->group))) &&
627 !ns_capable(net->user_ns, CAP_NET_ADMIN);
630 static void tun_set_real_num_queues(struct tun_struct *tun)
632 netif_set_real_num_tx_queues(tun->dev, tun->numqueues);
633 netif_set_real_num_rx_queues(tun->dev, tun->numqueues);
636 static void tun_disable_queue(struct tun_struct *tun, struct tun_file *tfile)
638 tfile->detached = tun;
639 list_add_tail(&tfile->next, &tun->disabled);
643 static struct tun_struct *tun_enable_queue(struct tun_file *tfile)
645 struct tun_struct *tun = tfile->detached;
647 tfile->detached = NULL;
648 list_del_init(&tfile->next);
653 void tun_ptr_free(void *ptr)
657 if (tun_is_xdp_frame(ptr)) {
658 struct xdp_frame *xdpf = tun_ptr_to_xdp(ptr);
660 xdp_return_frame(xdpf);
662 __skb_array_destroy_skb(ptr);
665 EXPORT_SYMBOL_GPL(tun_ptr_free);
667 static void tun_queue_purge(struct tun_file *tfile)
671 while ((ptr = ptr_ring_consume(&tfile->tx_ring)) != NULL)
674 skb_queue_purge(&tfile->sk.sk_write_queue);
675 skb_queue_purge(&tfile->sk.sk_error_queue);
678 static void tun_cleanup_tx_ring(struct tun_file *tfile)
680 if (tfile->tx_ring.queue) {
681 ptr_ring_cleanup(&tfile->tx_ring, tun_ptr_free);
682 xdp_rxq_info_unreg(&tfile->xdp_rxq);
683 memset(&tfile->tx_ring, 0, sizeof(tfile->tx_ring));
687 static void __tun_detach(struct tun_file *tfile, bool clean)
689 struct tun_file *ntfile;
690 struct tun_struct *tun;
692 tun = rtnl_dereference(tfile->tun);
695 tun_napi_disable(tun, tfile);
696 tun_napi_del(tun, tfile);
699 if (tun && !tfile->detached) {
700 u16 index = tfile->queue_index;
701 BUG_ON(index >= tun->numqueues);
703 rcu_assign_pointer(tun->tfiles[index],
704 tun->tfiles[tun->numqueues - 1]);
705 ntfile = rtnl_dereference(tun->tfiles[index]);
706 ntfile->queue_index = index;
710 RCU_INIT_POINTER(tfile->tun, NULL);
711 sock_put(&tfile->sk);
713 tun_disable_queue(tun, tfile);
716 tun_flow_delete_by_queue(tun, tun->numqueues + 1);
717 /* Drop read queue */
718 tun_queue_purge(tfile);
719 tun_set_real_num_queues(tun);
720 } else if (tfile->detached && clean) {
721 tun = tun_enable_queue(tfile);
722 sock_put(&tfile->sk);
726 if (tun && tun->numqueues == 0 && tun->numdisabled == 0) {
727 netif_carrier_off(tun->dev);
729 if (!(tun->flags & IFF_PERSIST) &&
730 tun->dev->reg_state == NETREG_REGISTERED)
731 unregister_netdevice(tun->dev);
733 tun_cleanup_tx_ring(tfile);
734 sock_put(&tfile->sk);
738 static void tun_detach(struct tun_file *tfile, bool clean)
740 struct tun_struct *tun;
741 struct net_device *dev;
744 tun = rtnl_dereference(tfile->tun);
745 dev = tun ? tun->dev : NULL;
746 __tun_detach(tfile, clean);
748 netdev_state_change(dev);
752 static void tun_detach_all(struct net_device *dev)
754 struct tun_struct *tun = netdev_priv(dev);
755 struct tun_file *tfile, *tmp;
756 int i, n = tun->numqueues;
758 for (i = 0; i < n; i++) {
759 tfile = rtnl_dereference(tun->tfiles[i]);
761 tun_napi_disable(tun, tfile);
762 tfile->socket.sk->sk_shutdown = RCV_SHUTDOWN;
763 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
764 RCU_INIT_POINTER(tfile->tun, NULL);
767 list_for_each_entry(tfile, &tun->disabled, next) {
768 tfile->socket.sk->sk_shutdown = RCV_SHUTDOWN;
769 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
770 RCU_INIT_POINTER(tfile->tun, NULL);
772 BUG_ON(tun->numqueues != 0);
775 for (i = 0; i < n; i++) {
776 tfile = rtnl_dereference(tun->tfiles[i]);
777 tun_napi_del(tun, tfile);
778 /* Drop read queue */
779 tun_queue_purge(tfile);
780 sock_put(&tfile->sk);
781 tun_cleanup_tx_ring(tfile);
783 list_for_each_entry_safe(tfile, tmp, &tun->disabled, next) {
784 tun_enable_queue(tfile);
785 tun_queue_purge(tfile);
786 sock_put(&tfile->sk);
787 tun_cleanup_tx_ring(tfile);
789 BUG_ON(tun->numdisabled != 0);
791 if (tun->flags & IFF_PERSIST)
792 module_put(THIS_MODULE);
795 static int tun_attach(struct tun_struct *tun, struct file *file,
796 bool skip_filter, bool napi)
798 struct tun_file *tfile = file->private_data;
799 struct net_device *dev = tun->dev;
802 err = security_tun_dev_attach(tfile->socket.sk, tun->security);
807 if (rtnl_dereference(tfile->tun) && !tfile->detached)
811 if (!(tun->flags & IFF_MULTI_QUEUE) && tun->numqueues == 1)
815 if (!tfile->detached &&
816 tun->numqueues + tun->numdisabled == MAX_TAP_QUEUES)
821 /* Re-attach the filter to persist device */
822 if (!skip_filter && (tun->filter_attached == true)) {
823 lock_sock(tfile->socket.sk);
824 err = sk_attach_filter(&tun->fprog, tfile->socket.sk);
825 release_sock(tfile->socket.sk);
830 if (!tfile->detached &&
831 ptr_ring_init(&tfile->tx_ring, dev->tx_queue_len, GFP_KERNEL)) {
836 tfile->queue_index = tun->numqueues;
837 tfile->socket.sk->sk_shutdown &= ~RCV_SHUTDOWN;
839 if (tfile->detached) {
840 /* Re-attach detached tfile, updating XDP queue_index */
841 WARN_ON(!xdp_rxq_info_is_reg(&tfile->xdp_rxq));
843 if (tfile->xdp_rxq.queue_index != tfile->queue_index)
844 tfile->xdp_rxq.queue_index = tfile->queue_index;
846 /* Setup XDP RX-queue info, for new tfile getting attached */
847 err = xdp_rxq_info_reg(&tfile->xdp_rxq,
848 tun->dev, tfile->queue_index);
851 err = xdp_rxq_info_reg_mem_model(&tfile->xdp_rxq,
852 MEM_TYPE_PAGE_SHARED, NULL);
854 xdp_rxq_info_unreg(&tfile->xdp_rxq);
860 rcu_assign_pointer(tfile->tun, tun);
861 rcu_assign_pointer(tun->tfiles[tun->numqueues], tfile);
864 if (tfile->detached) {
865 tun_enable_queue(tfile);
867 sock_hold(&tfile->sk);
868 tun_napi_init(tun, tfile, napi);
871 tun_set_real_num_queues(tun);
873 /* device is allowed to go away first, so no need to hold extra
881 static struct tun_struct *tun_get(struct tun_file *tfile)
883 struct tun_struct *tun;
886 tun = rcu_dereference(tfile->tun);
894 static void tun_put(struct tun_struct *tun)
900 static void addr_hash_set(u32 *mask, const u8 *addr)
902 int n = ether_crc(ETH_ALEN, addr) >> 26;
903 mask[n >> 5] |= (1 << (n & 31));
906 static unsigned int addr_hash_test(const u32 *mask, const u8 *addr)
908 int n = ether_crc(ETH_ALEN, addr) >> 26;
909 return mask[n >> 5] & (1 << (n & 31));
912 static int update_filter(struct tap_filter *filter, void __user *arg)
914 struct { u8 u[ETH_ALEN]; } *addr;
915 struct tun_filter uf;
916 int err, alen, n, nexact;
918 if (copy_from_user(&uf, arg, sizeof(uf)))
927 alen = ETH_ALEN * uf.count;
928 addr = memdup_user(arg + sizeof(uf), alen);
930 return PTR_ERR(addr);
932 /* The filter is updated without holding any locks. Which is
933 * perfectly safe. We disable it first and in the worst
934 * case we'll accept a few undesired packets. */
938 /* Use first set of addresses as an exact filter */
939 for (n = 0; n < uf.count && n < FLT_EXACT_COUNT; n++)
940 memcpy(filter->addr[n], addr[n].u, ETH_ALEN);
944 /* Remaining multicast addresses are hashed,
945 * unicast will leave the filter disabled. */
946 memset(filter->mask, 0, sizeof(filter->mask));
947 for (; n < uf.count; n++) {
948 if (!is_multicast_ether_addr(addr[n].u)) {
949 err = 0; /* no filter */
952 addr_hash_set(filter->mask, addr[n].u);
955 /* For ALLMULTI just set the mask to all ones.
956 * This overrides the mask populated above. */
957 if ((uf.flags & TUN_FLT_ALLMULTI))
958 memset(filter->mask, ~0, sizeof(filter->mask));
960 /* Now enable the filter */
962 filter->count = nexact;
964 /* Return the number of exact filters */
971 /* Returns: 0 - drop, !=0 - accept */
972 static int run_filter(struct tap_filter *filter, const struct sk_buff *skb)
974 /* Cannot use eth_hdr(skb) here because skb_mac_hdr() is incorrect
976 struct ethhdr *eh = (struct ethhdr *) skb->data;
980 for (i = 0; i < filter->count; i++)
981 if (ether_addr_equal(eh->h_dest, filter->addr[i]))
984 /* Inexact match (multicast only) */
985 if (is_multicast_ether_addr(eh->h_dest))
986 return addr_hash_test(filter->mask, eh->h_dest);
992 * Checks whether the packet is accepted or not.
993 * Returns: 0 - drop, !=0 - accept
995 static int check_filter(struct tap_filter *filter, const struct sk_buff *skb)
1000 return run_filter(filter, skb);
1003 /* Network device part of the driver */
1005 static const struct ethtool_ops tun_ethtool_ops;
1007 /* Net device detach from fd. */
1008 static void tun_net_uninit(struct net_device *dev)
1010 tun_detach_all(dev);
1013 /* Net device open. */
1014 static int tun_net_open(struct net_device *dev)
1016 struct tun_struct *tun = netdev_priv(dev);
1019 netif_tx_start_all_queues(dev);
1021 for (i = 0; i < tun->numqueues; i++) {
1022 struct tun_file *tfile;
1024 tfile = rtnl_dereference(tun->tfiles[i]);
1025 tfile->socket.sk->sk_write_space(tfile->socket.sk);
1031 /* Net device close. */
1032 static int tun_net_close(struct net_device *dev)
1034 netif_tx_stop_all_queues(dev);
1038 /* Net device start xmit */
1039 static void tun_automq_xmit(struct tun_struct *tun, struct sk_buff *skb)
1042 if (tun->numqueues == 1 && static_key_false(&rps_needed)) {
1043 /* Select queue was not called for the skbuff, so we extract the
1044 * RPS hash and save it into the flow_table here.
1048 rxhash = __skb_get_hash_symmetric(skb);
1050 struct tun_flow_entry *e;
1051 e = tun_flow_find(&tun->flows[tun_hashfn(rxhash)],
1054 tun_flow_save_rps_rxhash(e, rxhash);
1060 static unsigned int run_ebpf_filter(struct tun_struct *tun,
1061 struct sk_buff *skb,
1064 struct tun_prog *prog = rcu_dereference(tun->filter_prog);
1067 len = bpf_prog_run_clear_cb(prog->prog, skb);
1072 /* Net device start xmit */
1073 static netdev_tx_t tun_net_xmit(struct sk_buff *skb, struct net_device *dev)
1075 struct tun_struct *tun = netdev_priv(dev);
1076 int txq = skb->queue_mapping;
1077 struct tun_file *tfile;
1081 tfile = rcu_dereference(tun->tfiles[txq]);
1083 /* Drop packet if interface is not attached */
1084 if (txq >= tun->numqueues)
1087 if (!rcu_dereference(tun->steering_prog))
1088 tun_automq_xmit(tun, skb);
1090 tun_debug(KERN_INFO, tun, "tun_net_xmit %d\n", 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))
1100 if (tfile->socket.sk->sk_filter &&
1101 sk_filter(tfile->socket.sk, skb))
1104 len = run_ebpf_filter(tun, skb, len);
1105 if (len == 0 || pskb_trim(skb, len))
1108 if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC)))
1111 skb_tx_timestamp(skb);
1113 /* Orphan the skb - required as we might hang on to it
1114 * for indefinite time.
1120 if (ptr_ring_produce(&tfile->tx_ring, skb))
1123 /* Notify and wake up reader process */
1124 if (tfile->flags & TUN_FASYNC)
1125 kill_fasync(&tfile->fasync, SIGIO, POLL_IN);
1126 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
1129 return NETDEV_TX_OK;
1132 this_cpu_inc(tun->pcpu_stats->tx_dropped);
1136 return NET_XMIT_DROP;
1139 static void tun_net_mclist(struct net_device *dev)
1142 * This callback is supposed to deal with mc filter in
1143 * _rx_ path and has nothing to do with the _tx_ path.
1144 * In rx path we always accept everything userspace gives us.
1148 static netdev_features_t tun_net_fix_features(struct net_device *dev,
1149 netdev_features_t features)
1151 struct tun_struct *tun = netdev_priv(dev);
1153 return (features & tun->set_features) | (features & ~TUN_USER_FEATURES);
1155 #ifdef CONFIG_NET_POLL_CONTROLLER
1156 static void tun_poll_controller(struct net_device *dev)
1159 * Tun only receives frames when:
1160 * 1) the char device endpoint gets data from user space
1161 * 2) the tun socket gets a sendmsg call from user space
1162 * If NAPI is not enabled, since both of those are synchronous
1163 * operations, we are guaranteed never to have pending data when we poll
1164 * for it so there is nothing to do here but return.
1165 * We need this though so netpoll recognizes us as an interface that
1166 * supports polling, which enables bridge devices in virt setups to
1167 * still use netconsole
1168 * If NAPI is enabled, however, we need to schedule polling for all
1169 * queues unless we are using napi_gro_frags(), which we call in
1170 * process context and not in NAPI context.
1172 struct tun_struct *tun = netdev_priv(dev);
1174 if (tun->flags & IFF_NAPI) {
1175 struct tun_file *tfile;
1178 if (tun_napi_frags_enabled(tun))
1182 for (i = 0; i < tun->numqueues; i++) {
1183 tfile = rcu_dereference(tun->tfiles[i]);
1184 if (tfile->napi_enabled)
1185 napi_schedule(&tfile->napi);
1193 static void tun_set_headroom(struct net_device *dev, int new_hr)
1195 struct tun_struct *tun = netdev_priv(dev);
1197 if (new_hr < NET_SKB_PAD)
1198 new_hr = NET_SKB_PAD;
1200 tun->align = new_hr;
1204 tun_net_get_stats64(struct net_device *dev, struct rtnl_link_stats64 *stats)
1206 u32 rx_dropped = 0, tx_dropped = 0, rx_frame_errors = 0;
1207 struct tun_struct *tun = netdev_priv(dev);
1208 struct tun_pcpu_stats *p;
1211 for_each_possible_cpu(i) {
1212 u64 rxpackets, rxbytes, txpackets, txbytes;
1215 p = per_cpu_ptr(tun->pcpu_stats, i);
1217 start = u64_stats_fetch_begin(&p->syncp);
1218 rxpackets = p->rx_packets;
1219 rxbytes = p->rx_bytes;
1220 txpackets = p->tx_packets;
1221 txbytes = p->tx_bytes;
1222 } while (u64_stats_fetch_retry(&p->syncp, start));
1224 stats->rx_packets += rxpackets;
1225 stats->rx_bytes += rxbytes;
1226 stats->tx_packets += txpackets;
1227 stats->tx_bytes += txbytes;
1230 rx_dropped += p->rx_dropped;
1231 rx_frame_errors += p->rx_frame_errors;
1232 tx_dropped += p->tx_dropped;
1234 stats->rx_dropped = rx_dropped;
1235 stats->rx_frame_errors = rx_frame_errors;
1236 stats->tx_dropped = tx_dropped;
1239 static int tun_xdp_set(struct net_device *dev, struct bpf_prog *prog,
1240 struct netlink_ext_ack *extack)
1242 struct tun_struct *tun = netdev_priv(dev);
1243 struct bpf_prog *old_prog;
1245 old_prog = rtnl_dereference(tun->xdp_prog);
1246 rcu_assign_pointer(tun->xdp_prog, prog);
1248 bpf_prog_put(old_prog);
1253 static u32 tun_xdp_query(struct net_device *dev)
1255 struct tun_struct *tun = netdev_priv(dev);
1256 const struct bpf_prog *xdp_prog;
1258 xdp_prog = rtnl_dereference(tun->xdp_prog);
1260 return xdp_prog->aux->id;
1265 static int tun_xdp(struct net_device *dev, struct netdev_bpf *xdp)
1267 switch (xdp->command) {
1268 case XDP_SETUP_PROG:
1269 return tun_xdp_set(dev, xdp->prog, xdp->extack);
1270 case XDP_QUERY_PROG:
1271 xdp->prog_id = tun_xdp_query(dev);
1272 xdp->prog_attached = !!xdp->prog_id;
1279 static const struct net_device_ops tun_netdev_ops = {
1280 .ndo_uninit = tun_net_uninit,
1281 .ndo_open = tun_net_open,
1282 .ndo_stop = tun_net_close,
1283 .ndo_start_xmit = tun_net_xmit,
1284 .ndo_fix_features = tun_net_fix_features,
1285 .ndo_select_queue = tun_select_queue,
1286 #ifdef CONFIG_NET_POLL_CONTROLLER
1287 .ndo_poll_controller = tun_poll_controller,
1289 .ndo_set_rx_headroom = tun_set_headroom,
1290 .ndo_get_stats64 = tun_net_get_stats64,
1293 static int tun_xdp_xmit(struct net_device *dev, struct xdp_frame *frame)
1295 struct tun_struct *tun = netdev_priv(dev);
1296 struct tun_file *tfile;
1302 numqueues = READ_ONCE(tun->numqueues);
1308 tfile = rcu_dereference(tun->tfiles[smp_processor_id() %
1310 /* Encode the XDP flag into lowest bit for consumer to differ
1311 * XDP buffer from sk_buff.
1313 if (ptr_ring_produce(&tfile->tx_ring, tun_xdp_to_ptr(frame))) {
1314 this_cpu_inc(tun->pcpu_stats->tx_dropped);
1323 static int tun_xdp_tx(struct net_device *dev, struct xdp_buff *xdp)
1325 struct xdp_frame *frame = convert_to_xdp_frame(xdp);
1327 if (unlikely(!frame))
1330 return tun_xdp_xmit(dev, frame);
1333 static void tun_xdp_flush(struct net_device *dev)
1335 struct tun_struct *tun = netdev_priv(dev);
1336 struct tun_file *tfile;
1341 numqueues = READ_ONCE(tun->numqueues);
1345 tfile = rcu_dereference(tun->tfiles[smp_processor_id() %
1347 /* Notify and wake up reader process */
1348 if (tfile->flags & TUN_FASYNC)
1349 kill_fasync(&tfile->fasync, SIGIO, POLL_IN);
1350 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
1356 static const struct net_device_ops tap_netdev_ops = {
1357 .ndo_uninit = tun_net_uninit,
1358 .ndo_open = tun_net_open,
1359 .ndo_stop = tun_net_close,
1360 .ndo_start_xmit = tun_net_xmit,
1361 .ndo_fix_features = tun_net_fix_features,
1362 .ndo_set_rx_mode = tun_net_mclist,
1363 .ndo_set_mac_address = eth_mac_addr,
1364 .ndo_validate_addr = eth_validate_addr,
1365 .ndo_select_queue = tun_select_queue,
1366 #ifdef CONFIG_NET_POLL_CONTROLLER
1367 .ndo_poll_controller = tun_poll_controller,
1369 .ndo_features_check = passthru_features_check,
1370 .ndo_set_rx_headroom = tun_set_headroom,
1371 .ndo_get_stats64 = tun_net_get_stats64,
1373 .ndo_xdp_xmit = tun_xdp_xmit,
1374 .ndo_xdp_flush = tun_xdp_flush,
1377 static void tun_flow_init(struct tun_struct *tun)
1381 for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++)
1382 INIT_HLIST_HEAD(&tun->flows[i]);
1384 tun->ageing_time = TUN_FLOW_EXPIRE;
1385 timer_setup(&tun->flow_gc_timer, tun_flow_cleanup, 0);
1386 mod_timer(&tun->flow_gc_timer,
1387 round_jiffies_up(jiffies + tun->ageing_time));
1390 static void tun_flow_uninit(struct tun_struct *tun)
1392 del_timer_sync(&tun->flow_gc_timer);
1393 tun_flow_flush(tun);
1397 #define MAX_MTU 65535
1399 /* Initialize net device. */
1400 static void tun_net_init(struct net_device *dev)
1402 struct tun_struct *tun = netdev_priv(dev);
1404 switch (tun->flags & TUN_TYPE_MASK) {
1406 dev->netdev_ops = &tun_netdev_ops;
1408 /* Point-to-Point TUN Device */
1409 dev->hard_header_len = 0;
1413 /* Zero header length */
1414 dev->type = ARPHRD_NONE;
1415 dev->flags = IFF_POINTOPOINT | IFF_NOARP | IFF_MULTICAST;
1419 dev->netdev_ops = &tap_netdev_ops;
1420 /* Ethernet TAP Device */
1422 dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1423 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
1425 eth_hw_addr_random(dev);
1430 dev->min_mtu = MIN_MTU;
1431 dev->max_mtu = MAX_MTU - dev->hard_header_len;
1434 /* Character device part */
1437 static __poll_t tun_chr_poll(struct file *file, poll_table *wait)
1439 struct tun_file *tfile = file->private_data;
1440 struct tun_struct *tun = tun_get(tfile);
1447 sk = tfile->socket.sk;
1449 tun_debug(KERN_INFO, tun, "tun_chr_poll\n");
1451 poll_wait(file, sk_sleep(sk), wait);
1453 if (!ptr_ring_empty(&tfile->tx_ring))
1454 mask |= EPOLLIN | EPOLLRDNORM;
1456 if (tun->dev->flags & IFF_UP &&
1457 (sock_writeable(sk) ||
1458 (!test_and_set_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags) &&
1459 sock_writeable(sk))))
1460 mask |= EPOLLOUT | EPOLLWRNORM;
1462 if (tun->dev->reg_state != NETREG_REGISTERED)
1469 static struct sk_buff *tun_napi_alloc_frags(struct tun_file *tfile,
1471 const struct iov_iter *it)
1473 struct sk_buff *skb;
1478 if (it->nr_segs > MAX_SKB_FRAGS + 1)
1479 return ERR_PTR(-ENOMEM);
1482 skb = napi_get_frags(&tfile->napi);
1485 return ERR_PTR(-ENOMEM);
1487 linear = iov_iter_single_seg_count(it);
1488 err = __skb_grow(skb, linear);
1493 skb->data_len = len - linear;
1494 skb->truesize += skb->data_len;
1496 for (i = 1; i < it->nr_segs; i++) {
1497 struct page_frag *pfrag = ¤t->task_frag;
1498 size_t fragsz = it->iov[i].iov_len;
1500 if (fragsz == 0 || fragsz > PAGE_SIZE) {
1505 if (!skb_page_frag_refill(fragsz, pfrag, GFP_KERNEL)) {
1510 skb_fill_page_desc(skb, i - 1, pfrag->page,
1511 pfrag->offset, fragsz);
1512 page_ref_inc(pfrag->page);
1513 pfrag->offset += fragsz;
1518 /* frees skb and all frags allocated with napi_alloc_frag() */
1519 napi_free_frags(&tfile->napi);
1520 return ERR_PTR(err);
1523 /* prepad is the amount to reserve at front. len is length after that.
1524 * linear is a hint as to how much to copy (usually headers). */
1525 static struct sk_buff *tun_alloc_skb(struct tun_file *tfile,
1526 size_t prepad, size_t len,
1527 size_t linear, int noblock)
1529 struct sock *sk = tfile->socket.sk;
1530 struct sk_buff *skb;
1533 /* Under a page? Don't bother with paged skb. */
1534 if (prepad + len < PAGE_SIZE || !linear)
1537 skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
1540 return ERR_PTR(err);
1542 skb_reserve(skb, prepad);
1543 skb_put(skb, linear);
1544 skb->data_len = len - linear;
1545 skb->len += len - linear;
1550 static void tun_rx_batched(struct tun_struct *tun, struct tun_file *tfile,
1551 struct sk_buff *skb, int more)
1553 struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
1554 struct sk_buff_head process_queue;
1555 u32 rx_batched = tun->rx_batched;
1558 if (!rx_batched || (!more && skb_queue_empty(queue))) {
1560 netif_receive_skb(skb);
1565 spin_lock(&queue->lock);
1566 if (!more || skb_queue_len(queue) == rx_batched) {
1567 __skb_queue_head_init(&process_queue);
1568 skb_queue_splice_tail_init(queue, &process_queue);
1571 __skb_queue_tail(queue, skb);
1573 spin_unlock(&queue->lock);
1576 struct sk_buff *nskb;
1579 while ((nskb = __skb_dequeue(&process_queue)))
1580 netif_receive_skb(nskb);
1581 netif_receive_skb(skb);
1586 static bool tun_can_build_skb(struct tun_struct *tun, struct tun_file *tfile,
1587 int len, int noblock, bool zerocopy)
1589 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
1592 if (tfile->socket.sk->sk_sndbuf != INT_MAX)
1601 if (SKB_DATA_ALIGN(len + TUN_RX_PAD) +
1602 SKB_DATA_ALIGN(sizeof(struct skb_shared_info)) > PAGE_SIZE)
1608 static struct sk_buff *tun_build_skb(struct tun_struct *tun,
1609 struct tun_file *tfile,
1610 struct iov_iter *from,
1611 struct virtio_net_hdr *hdr,
1612 int len, int *skb_xdp)
1614 struct page_frag *alloc_frag = ¤t->task_frag;
1615 struct sk_buff *skb;
1616 struct bpf_prog *xdp_prog;
1617 int buflen = SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
1618 unsigned int delta = 0;
1621 int err, pad = TUN_RX_PAD;
1624 xdp_prog = rcu_dereference(tun->xdp_prog);
1626 pad += TUN_HEADROOM;
1627 buflen += SKB_DATA_ALIGN(len + pad);
1630 alloc_frag->offset = ALIGN((u64)alloc_frag->offset, SMP_CACHE_BYTES);
1631 if (unlikely(!skb_page_frag_refill(buflen, alloc_frag, GFP_KERNEL)))
1632 return ERR_PTR(-ENOMEM);
1634 buf = (char *)page_address(alloc_frag->page) + alloc_frag->offset;
1635 copied = copy_page_from_iter(alloc_frag->page,
1636 alloc_frag->offset + pad,
1639 return ERR_PTR(-EFAULT);
1641 /* There's a small window that XDP may be set after the check
1642 * of xdp_prog above, this should be rare and for simplicity
1643 * we do XDP on skb in case the headroom is not enough.
1645 if (hdr->gso_type || !xdp_prog)
1652 xdp_prog = rcu_dereference(tun->xdp_prog);
1653 if (xdp_prog && !*skb_xdp) {
1654 struct xdp_buff xdp;
1658 xdp.data_hard_start = buf;
1659 xdp.data = buf + pad;
1660 xdp_set_data_meta_invalid(&xdp);
1661 xdp.data_end = xdp.data + len;
1662 xdp.rxq = &tfile->xdp_rxq;
1663 orig_data = xdp.data;
1664 act = bpf_prog_run_xdp(xdp_prog, &xdp);
1668 get_page(alloc_frag->page);
1669 alloc_frag->offset += buflen;
1670 err = xdp_do_redirect(tun->dev, &xdp, xdp_prog);
1678 get_page(alloc_frag->page);
1679 alloc_frag->offset += buflen;
1680 if (tun_xdp_tx(tun->dev, &xdp))
1682 tun_xdp_flush(tun->dev);
1687 delta = orig_data - xdp.data;
1688 len = xdp.data_end - xdp.data;
1691 bpf_warn_invalid_xdp_action(act);
1694 trace_xdp_exception(tun->dev, xdp_prog, act);
1701 skb = build_skb(buf, buflen);
1705 return ERR_PTR(-ENOMEM);
1708 skb_reserve(skb, pad - delta);
1710 get_page(alloc_frag->page);
1711 alloc_frag->offset += buflen;
1719 put_page(alloc_frag->page);
1723 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1727 /* Get packet from user space buffer */
1728 static ssize_t tun_get_user(struct tun_struct *tun, struct tun_file *tfile,
1729 void *msg_control, struct iov_iter *from,
1730 int noblock, bool more)
1732 struct tun_pi pi = { 0, cpu_to_be16(ETH_P_IP) };
1733 struct sk_buff *skb;
1734 size_t total_len = iov_iter_count(from);
1735 size_t len = total_len, align = tun->align, linear;
1736 struct virtio_net_hdr gso = { 0 };
1737 struct tun_pcpu_stats *stats;
1740 bool zerocopy = false;
1744 bool frags = tun_napi_frags_enabled(tun);
1746 if (!(tun->dev->flags & IFF_UP))
1749 if (!(tun->flags & IFF_NO_PI)) {
1750 if (len < sizeof(pi))
1754 if (!copy_from_iter_full(&pi, sizeof(pi), from))
1758 if (tun->flags & IFF_VNET_HDR) {
1759 int vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
1761 if (len < vnet_hdr_sz)
1765 if (!copy_from_iter_full(&gso, sizeof(gso), from))
1768 if ((gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
1769 tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2 > tun16_to_cpu(tun, gso.hdr_len))
1770 gso.hdr_len = cpu_to_tun16(tun, tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2);
1772 if (tun16_to_cpu(tun, gso.hdr_len) > len)
1774 iov_iter_advance(from, vnet_hdr_sz - sizeof(gso));
1777 if ((tun->flags & TUN_TYPE_MASK) == IFF_TAP) {
1778 align += NET_IP_ALIGN;
1779 if (unlikely(len < ETH_HLEN ||
1780 (gso.hdr_len && tun16_to_cpu(tun, gso.hdr_len) < ETH_HLEN)))
1784 good_linear = SKB_MAX_HEAD(align);
1787 struct iov_iter i = *from;
1789 /* There are 256 bytes to be copied in skb, so there is
1790 * enough room for skb expand head in case it is used.
1791 * The rest of the buffer is mapped from userspace.
1793 copylen = gso.hdr_len ? tun16_to_cpu(tun, gso.hdr_len) : GOODCOPY_LEN;
1794 if (copylen > good_linear)
1795 copylen = good_linear;
1797 iov_iter_advance(&i, copylen);
1798 if (iov_iter_npages(&i, INT_MAX) <= MAX_SKB_FRAGS)
1802 if (!frags && tun_can_build_skb(tun, tfile, len, noblock, zerocopy)) {
1803 /* For the packet that is not easy to be processed
1804 * (e.g gso or jumbo packet), we will do it at after
1805 * skb was created with generic XDP routine.
1807 skb = tun_build_skb(tun, tfile, from, &gso, len, &skb_xdp);
1809 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1810 return PTR_ERR(skb);
1817 if (tun16_to_cpu(tun, gso.hdr_len) > good_linear)
1818 linear = good_linear;
1820 linear = tun16_to_cpu(tun, gso.hdr_len);
1824 mutex_lock(&tfile->napi_mutex);
1825 skb = tun_napi_alloc_frags(tfile, copylen, from);
1826 /* tun_napi_alloc_frags() enforces a layout for the skb.
1827 * If zerocopy is enabled, then this layout will be
1828 * overwritten by zerocopy_sg_from_iter().
1832 skb = tun_alloc_skb(tfile, align, copylen, linear,
1837 if (PTR_ERR(skb) != -EAGAIN)
1838 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1840 mutex_unlock(&tfile->napi_mutex);
1841 return PTR_ERR(skb);
1845 err = zerocopy_sg_from_iter(skb, from);
1847 err = skb_copy_datagram_from_iter(skb, 0, from, len);
1850 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1853 tfile->napi.skb = NULL;
1854 mutex_unlock(&tfile->napi_mutex);
1861 if (virtio_net_hdr_to_skb(skb, &gso, tun_is_little_endian(tun))) {
1862 this_cpu_inc(tun->pcpu_stats->rx_frame_errors);
1865 tfile->napi.skb = NULL;
1866 mutex_unlock(&tfile->napi_mutex);
1872 switch (tun->flags & TUN_TYPE_MASK) {
1874 if (tun->flags & IFF_NO_PI) {
1875 u8 ip_version = skb->len ? (skb->data[0] >> 4) : 0;
1877 switch (ip_version) {
1879 pi.proto = htons(ETH_P_IP);
1882 pi.proto = htons(ETH_P_IPV6);
1885 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1891 skb_reset_mac_header(skb);
1892 skb->protocol = pi.proto;
1893 skb->dev = tun->dev;
1897 skb->protocol = eth_type_trans(skb, tun->dev);
1901 /* copy skb_ubuf_info for callback when skb has no error */
1903 skb_shinfo(skb)->destructor_arg = msg_control;
1904 skb_shinfo(skb)->tx_flags |= SKBTX_DEV_ZEROCOPY;
1905 skb_shinfo(skb)->tx_flags |= SKBTX_SHARED_FRAG;
1906 } else if (msg_control) {
1907 struct ubuf_info *uarg = msg_control;
1908 uarg->callback(uarg, false);
1911 skb_reset_network_header(skb);
1912 skb_probe_transport_header(skb, 0);
1915 struct bpf_prog *xdp_prog;
1919 xdp_prog = rcu_dereference(tun->xdp_prog);
1921 ret = do_xdp_generic(xdp_prog, skb);
1922 if (ret != XDP_PASS) {
1930 /* Compute the costly rx hash only if needed for flow updates.
1931 * We may get a very small possibility of OOO during switching, not
1932 * worth to optimize.
1934 if (!rcu_access_pointer(tun->steering_prog) && tun->numqueues > 1 &&
1936 rxhash = __skb_get_hash_symmetric(skb);
1939 /* Exercise flow dissector code path. */
1940 u32 headlen = eth_get_headlen(skb->data, skb_headlen(skb));
1942 if (unlikely(headlen > skb_headlen(skb))) {
1943 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1944 napi_free_frags(&tfile->napi);
1945 mutex_unlock(&tfile->napi_mutex);
1951 napi_gro_frags(&tfile->napi);
1953 mutex_unlock(&tfile->napi_mutex);
1954 } else if (tfile->napi_enabled) {
1955 struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
1958 spin_lock_bh(&queue->lock);
1959 __skb_queue_tail(queue, skb);
1960 queue_len = skb_queue_len(queue);
1961 spin_unlock(&queue->lock);
1963 if (!more || queue_len > NAPI_POLL_WEIGHT)
1964 napi_schedule(&tfile->napi);
1967 } else if (!IS_ENABLED(CONFIG_4KSTACKS)) {
1968 tun_rx_batched(tun, tfile, skb, more);
1973 stats = get_cpu_ptr(tun->pcpu_stats);
1974 u64_stats_update_begin(&stats->syncp);
1975 stats->rx_packets++;
1976 stats->rx_bytes += len;
1977 u64_stats_update_end(&stats->syncp);
1981 tun_flow_update(tun, rxhash, tfile);
1986 static ssize_t tun_chr_write_iter(struct kiocb *iocb, struct iov_iter *from)
1988 struct file *file = iocb->ki_filp;
1989 struct tun_file *tfile = file->private_data;
1990 struct tun_struct *tun = tun_get(tfile);
1996 result = tun_get_user(tun, tfile, NULL, from,
1997 file->f_flags & O_NONBLOCK, false);
2003 static ssize_t tun_put_user_xdp(struct tun_struct *tun,
2004 struct tun_file *tfile,
2005 struct xdp_frame *xdp_frame,
2006 struct iov_iter *iter)
2008 int vnet_hdr_sz = 0;
2009 size_t size = xdp_frame->len;
2010 struct tun_pcpu_stats *stats;
2013 if (tun->flags & IFF_VNET_HDR) {
2014 struct virtio_net_hdr gso = { 0 };
2016 vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
2017 if (unlikely(iov_iter_count(iter) < vnet_hdr_sz))
2019 if (unlikely(copy_to_iter(&gso, sizeof(gso), iter) !=
2022 iov_iter_advance(iter, vnet_hdr_sz - sizeof(gso));
2025 ret = copy_to_iter(xdp_frame->data, size, iter) + vnet_hdr_sz;
2027 stats = get_cpu_ptr(tun->pcpu_stats);
2028 u64_stats_update_begin(&stats->syncp);
2029 stats->tx_packets++;
2030 stats->tx_bytes += ret;
2031 u64_stats_update_end(&stats->syncp);
2032 put_cpu_ptr(tun->pcpu_stats);
2037 /* Put packet to the user space buffer */
2038 static ssize_t tun_put_user(struct tun_struct *tun,
2039 struct tun_file *tfile,
2040 struct sk_buff *skb,
2041 struct iov_iter *iter)
2043 struct tun_pi pi = { 0, skb->protocol };
2044 struct tun_pcpu_stats *stats;
2046 int vlan_offset = 0;
2048 int vnet_hdr_sz = 0;
2050 if (skb_vlan_tag_present(skb))
2051 vlan_hlen = VLAN_HLEN;
2053 if (tun->flags & IFF_VNET_HDR)
2054 vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
2056 total = skb->len + vlan_hlen + vnet_hdr_sz;
2058 if (!(tun->flags & IFF_NO_PI)) {
2059 if (iov_iter_count(iter) < sizeof(pi))
2062 total += sizeof(pi);
2063 if (iov_iter_count(iter) < total) {
2064 /* Packet will be striped */
2065 pi.flags |= TUN_PKT_STRIP;
2068 if (copy_to_iter(&pi, sizeof(pi), iter) != sizeof(pi))
2073 struct virtio_net_hdr gso;
2075 if (iov_iter_count(iter) < vnet_hdr_sz)
2078 if (virtio_net_hdr_from_skb(skb, &gso,
2079 tun_is_little_endian(tun), true)) {
2080 struct skb_shared_info *sinfo = skb_shinfo(skb);
2081 pr_err("unexpected GSO type: "
2082 "0x%x, gso_size %d, hdr_len %d\n",
2083 sinfo->gso_type, tun16_to_cpu(tun, gso.gso_size),
2084 tun16_to_cpu(tun, gso.hdr_len));
2085 print_hex_dump(KERN_ERR, "tun: ",
2088 min((int)tun16_to_cpu(tun, gso.hdr_len), 64), true);
2093 if (copy_to_iter(&gso, sizeof(gso), iter) != sizeof(gso))
2096 iov_iter_advance(iter, vnet_hdr_sz - sizeof(gso));
2103 veth.h_vlan_proto = skb->vlan_proto;
2104 veth.h_vlan_TCI = htons(skb_vlan_tag_get(skb));
2106 vlan_offset = offsetof(struct vlan_ethhdr, h_vlan_proto);
2108 ret = skb_copy_datagram_iter(skb, 0, iter, vlan_offset);
2109 if (ret || !iov_iter_count(iter))
2112 ret = copy_to_iter(&veth, sizeof(veth), iter);
2113 if (ret != sizeof(veth) || !iov_iter_count(iter))
2117 skb_copy_datagram_iter(skb, vlan_offset, iter, skb->len - vlan_offset);
2120 /* caller is in process context, */
2121 stats = get_cpu_ptr(tun->pcpu_stats);
2122 u64_stats_update_begin(&stats->syncp);
2123 stats->tx_packets++;
2124 stats->tx_bytes += skb->len + vlan_hlen;
2125 u64_stats_update_end(&stats->syncp);
2126 put_cpu_ptr(tun->pcpu_stats);
2131 static void *tun_ring_recv(struct tun_file *tfile, int noblock, int *err)
2133 DECLARE_WAITQUEUE(wait, current);
2137 ptr = ptr_ring_consume(&tfile->tx_ring);
2145 add_wait_queue(&tfile->wq.wait, &wait);
2146 current->state = TASK_INTERRUPTIBLE;
2149 ptr = ptr_ring_consume(&tfile->tx_ring);
2152 if (signal_pending(current)) {
2153 error = -ERESTARTSYS;
2156 if (tfile->socket.sk->sk_shutdown & RCV_SHUTDOWN) {
2164 current->state = TASK_RUNNING;
2165 remove_wait_queue(&tfile->wq.wait, &wait);
2172 static ssize_t tun_do_read(struct tun_struct *tun, struct tun_file *tfile,
2173 struct iov_iter *to,
2174 int noblock, void *ptr)
2179 tun_debug(KERN_INFO, tun, "tun_do_read\n");
2181 if (!iov_iter_count(to)) {
2187 /* Read frames from ring */
2188 ptr = tun_ring_recv(tfile, noblock, &err);
2193 if (tun_is_xdp_frame(ptr)) {
2194 struct xdp_frame *xdpf = tun_ptr_to_xdp(ptr);
2196 ret = tun_put_user_xdp(tun, tfile, xdpf, to);
2197 xdp_return_frame(xdpf);
2199 struct sk_buff *skb = ptr;
2201 ret = tun_put_user(tun, tfile, skb, to);
2202 if (unlikely(ret < 0))
2211 static ssize_t tun_chr_read_iter(struct kiocb *iocb, struct iov_iter *to)
2213 struct file *file = iocb->ki_filp;
2214 struct tun_file *tfile = file->private_data;
2215 struct tun_struct *tun = tun_get(tfile);
2216 ssize_t len = iov_iter_count(to), ret;
2220 ret = tun_do_read(tun, tfile, to, file->f_flags & O_NONBLOCK, NULL);
2221 ret = min_t(ssize_t, ret, len);
2228 static void tun_prog_free(struct rcu_head *rcu)
2230 struct tun_prog *prog = container_of(rcu, struct tun_prog, rcu);
2232 bpf_prog_destroy(prog->prog);
2236 static int __tun_set_ebpf(struct tun_struct *tun,
2237 struct tun_prog __rcu **prog_p,
2238 struct bpf_prog *prog)
2240 struct tun_prog *old, *new = NULL;
2243 new = kmalloc(sizeof(*new), GFP_KERNEL);
2249 spin_lock_bh(&tun->lock);
2250 old = rcu_dereference_protected(*prog_p,
2251 lockdep_is_held(&tun->lock));
2252 rcu_assign_pointer(*prog_p, new);
2253 spin_unlock_bh(&tun->lock);
2256 call_rcu(&old->rcu, tun_prog_free);
2261 static void tun_free_netdev(struct net_device *dev)
2263 struct tun_struct *tun = netdev_priv(dev);
2265 BUG_ON(!(list_empty(&tun->disabled)));
2266 free_percpu(tun->pcpu_stats);
2267 tun_flow_uninit(tun);
2268 security_tun_dev_free_security(tun->security);
2269 __tun_set_ebpf(tun, &tun->steering_prog, NULL);
2270 __tun_set_ebpf(tun, &tun->filter_prog, NULL);
2273 static void tun_setup(struct net_device *dev)
2275 struct tun_struct *tun = netdev_priv(dev);
2277 tun->owner = INVALID_UID;
2278 tun->group = INVALID_GID;
2280 dev->ethtool_ops = &tun_ethtool_ops;
2281 dev->needs_free_netdev = true;
2282 dev->priv_destructor = tun_free_netdev;
2283 /* We prefer our own queue length */
2284 dev->tx_queue_len = TUN_READQ_SIZE;
2287 /* Trivial set of netlink ops to allow deleting tun or tap
2288 * device with netlink.
2290 static int tun_validate(struct nlattr *tb[], struct nlattr *data[],
2291 struct netlink_ext_ack *extack)
2296 static size_t tun_get_size(const struct net_device *dev)
2298 BUILD_BUG_ON(sizeof(u32) != sizeof(uid_t));
2299 BUILD_BUG_ON(sizeof(u32) != sizeof(gid_t));
2301 return nla_total_size(sizeof(uid_t)) + /* OWNER */
2302 nla_total_size(sizeof(gid_t)) + /* GROUP */
2303 nla_total_size(sizeof(u8)) + /* TYPE */
2304 nla_total_size(sizeof(u8)) + /* PI */
2305 nla_total_size(sizeof(u8)) + /* VNET_HDR */
2306 nla_total_size(sizeof(u8)) + /* PERSIST */
2307 nla_total_size(sizeof(u8)) + /* MULTI_QUEUE */
2308 nla_total_size(sizeof(u32)) + /* NUM_QUEUES */
2309 nla_total_size(sizeof(u32)) + /* NUM_DISABLED_QUEUES */
2313 static int tun_fill_info(struct sk_buff *skb, const struct net_device *dev)
2315 struct tun_struct *tun = netdev_priv(dev);
2317 if (nla_put_u8(skb, IFLA_TUN_TYPE, tun->flags & TUN_TYPE_MASK))
2318 goto nla_put_failure;
2319 if (uid_valid(tun->owner) &&
2320 nla_put_u32(skb, IFLA_TUN_OWNER,
2321 from_kuid_munged(current_user_ns(), tun->owner)))
2322 goto nla_put_failure;
2323 if (gid_valid(tun->group) &&
2324 nla_put_u32(skb, IFLA_TUN_GROUP,
2325 from_kgid_munged(current_user_ns(), tun->group)))
2326 goto nla_put_failure;
2327 if (nla_put_u8(skb, IFLA_TUN_PI, !(tun->flags & IFF_NO_PI)))
2328 goto nla_put_failure;
2329 if (nla_put_u8(skb, IFLA_TUN_VNET_HDR, !!(tun->flags & IFF_VNET_HDR)))
2330 goto nla_put_failure;
2331 if (nla_put_u8(skb, IFLA_TUN_PERSIST, !!(tun->flags & IFF_PERSIST)))
2332 goto nla_put_failure;
2333 if (nla_put_u8(skb, IFLA_TUN_MULTI_QUEUE,
2334 !!(tun->flags & IFF_MULTI_QUEUE)))
2335 goto nla_put_failure;
2336 if (tun->flags & IFF_MULTI_QUEUE) {
2337 if (nla_put_u32(skb, IFLA_TUN_NUM_QUEUES, tun->numqueues))
2338 goto nla_put_failure;
2339 if (nla_put_u32(skb, IFLA_TUN_NUM_DISABLED_QUEUES,
2341 goto nla_put_failure;
2350 static struct rtnl_link_ops tun_link_ops __read_mostly = {
2352 .priv_size = sizeof(struct tun_struct),
2354 .validate = tun_validate,
2355 .get_size = tun_get_size,
2356 .fill_info = tun_fill_info,
2359 static void tun_sock_write_space(struct sock *sk)
2361 struct tun_file *tfile;
2362 wait_queue_head_t *wqueue;
2364 if (!sock_writeable(sk))
2367 if (!test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags))
2370 wqueue = sk_sleep(sk);
2371 if (wqueue && waitqueue_active(wqueue))
2372 wake_up_interruptible_sync_poll(wqueue, EPOLLOUT |
2373 EPOLLWRNORM | EPOLLWRBAND);
2375 tfile = container_of(sk, struct tun_file, sk);
2376 kill_fasync(&tfile->fasync, SIGIO, POLL_OUT);
2379 static int tun_sendmsg(struct socket *sock, struct msghdr *m, size_t total_len)
2382 struct tun_file *tfile = container_of(sock, struct tun_file, socket);
2383 struct tun_struct *tun = tun_get(tfile);
2388 ret = tun_get_user(tun, tfile, m->msg_control, &m->msg_iter,
2389 m->msg_flags & MSG_DONTWAIT,
2390 m->msg_flags & MSG_MORE);
2395 static int tun_recvmsg(struct socket *sock, struct msghdr *m, size_t total_len,
2398 struct tun_file *tfile = container_of(sock, struct tun_file, socket);
2399 struct tun_struct *tun = tun_get(tfile);
2400 void *ptr = m->msg_control;
2408 if (flags & ~(MSG_DONTWAIT|MSG_TRUNC|MSG_ERRQUEUE)) {
2412 if (flags & MSG_ERRQUEUE) {
2413 ret = sock_recv_errqueue(sock->sk, m, total_len,
2414 SOL_PACKET, TUN_TX_TIMESTAMP);
2417 ret = tun_do_read(tun, tfile, &m->msg_iter, flags & MSG_DONTWAIT, ptr);
2418 if (ret > (ssize_t)total_len) {
2419 m->msg_flags |= MSG_TRUNC;
2420 ret = flags & MSG_TRUNC ? ret : total_len;
2433 static int tun_ptr_peek_len(void *ptr)
2436 if (tun_is_xdp_frame(ptr)) {
2437 struct xdp_frame *xdpf = tun_ptr_to_xdp(ptr);
2441 return __skb_array_len_with_tag(ptr);
2447 static int tun_peek_len(struct socket *sock)
2449 struct tun_file *tfile = container_of(sock, struct tun_file, socket);
2450 struct tun_struct *tun;
2453 tun = tun_get(tfile);
2457 ret = PTR_RING_PEEK_CALL(&tfile->tx_ring, tun_ptr_peek_len);
2463 /* Ops structure to mimic raw sockets with tun */
2464 static const struct proto_ops tun_socket_ops = {
2465 .peek_len = tun_peek_len,
2466 .sendmsg = tun_sendmsg,
2467 .recvmsg = tun_recvmsg,
2470 static struct proto tun_proto = {
2472 .owner = THIS_MODULE,
2473 .obj_size = sizeof(struct tun_file),
2476 static int tun_flags(struct tun_struct *tun)
2478 return tun->flags & (TUN_FEATURES | IFF_PERSIST | IFF_TUN | IFF_TAP);
2481 static ssize_t tun_show_flags(struct device *dev, struct device_attribute *attr,
2484 struct tun_struct *tun = netdev_priv(to_net_dev(dev));
2485 return sprintf(buf, "0x%x\n", tun_flags(tun));
2488 static ssize_t tun_show_owner(struct device *dev, struct device_attribute *attr,
2491 struct tun_struct *tun = netdev_priv(to_net_dev(dev));
2492 return uid_valid(tun->owner)?
2493 sprintf(buf, "%u\n",
2494 from_kuid_munged(current_user_ns(), tun->owner)):
2495 sprintf(buf, "-1\n");
2498 static ssize_t tun_show_group(struct device *dev, struct device_attribute *attr,
2501 struct tun_struct *tun = netdev_priv(to_net_dev(dev));
2502 return gid_valid(tun->group) ?
2503 sprintf(buf, "%u\n",
2504 from_kgid_munged(current_user_ns(), tun->group)):
2505 sprintf(buf, "-1\n");
2508 static DEVICE_ATTR(tun_flags, 0444, tun_show_flags, NULL);
2509 static DEVICE_ATTR(owner, 0444, tun_show_owner, NULL);
2510 static DEVICE_ATTR(group, 0444, tun_show_group, NULL);
2512 static struct attribute *tun_dev_attrs[] = {
2513 &dev_attr_tun_flags.attr,
2514 &dev_attr_owner.attr,
2515 &dev_attr_group.attr,
2519 static const struct attribute_group tun_attr_group = {
2520 .attrs = tun_dev_attrs
2523 static int tun_set_iff(struct net *net, struct file *file, struct ifreq *ifr)
2525 struct tun_struct *tun;
2526 struct tun_file *tfile = file->private_data;
2527 struct net_device *dev;
2530 if (tfile->detached)
2533 if ((ifr->ifr_flags & IFF_NAPI_FRAGS)) {
2534 if (!capable(CAP_NET_ADMIN))
2537 if (!(ifr->ifr_flags & IFF_NAPI) ||
2538 (ifr->ifr_flags & TUN_TYPE_MASK) != IFF_TAP)
2542 dev = __dev_get_by_name(net, ifr->ifr_name);
2544 if (ifr->ifr_flags & IFF_TUN_EXCL)
2546 if ((ifr->ifr_flags & IFF_TUN) && dev->netdev_ops == &tun_netdev_ops)
2547 tun = netdev_priv(dev);
2548 else if ((ifr->ifr_flags & IFF_TAP) && dev->netdev_ops == &tap_netdev_ops)
2549 tun = netdev_priv(dev);
2553 if (!!(ifr->ifr_flags & IFF_MULTI_QUEUE) !=
2554 !!(tun->flags & IFF_MULTI_QUEUE))
2557 if (tun_not_capable(tun))
2559 err = security_tun_dev_open(tun->security);
2563 err = tun_attach(tun, file, ifr->ifr_flags & IFF_NOFILTER,
2564 ifr->ifr_flags & IFF_NAPI);
2568 if (tun->flags & IFF_MULTI_QUEUE &&
2569 (tun->numqueues + tun->numdisabled > 1)) {
2570 /* One or more queue has already been attached, no need
2571 * to initialize the device again.
2573 netdev_state_change(dev);
2577 tun->flags = (tun->flags & ~TUN_FEATURES) |
2578 (ifr->ifr_flags & TUN_FEATURES);
2580 netdev_state_change(dev);
2583 unsigned long flags = 0;
2584 int queues = ifr->ifr_flags & IFF_MULTI_QUEUE ?
2587 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
2589 err = security_tun_dev_create();
2594 if (ifr->ifr_flags & IFF_TUN) {
2598 } else if (ifr->ifr_flags & IFF_TAP) {
2606 name = ifr->ifr_name;
2608 dev = alloc_netdev_mqs(sizeof(struct tun_struct), name,
2609 NET_NAME_UNKNOWN, tun_setup, queues,
2614 err = dev_get_valid_name(net, dev, name);
2618 dev_net_set(dev, net);
2619 dev->rtnl_link_ops = &tun_link_ops;
2620 dev->ifindex = tfile->ifindex;
2621 dev->sysfs_groups[0] = &tun_attr_group;
2623 tun = netdev_priv(dev);
2626 tun->txflt.count = 0;
2627 tun->vnet_hdr_sz = sizeof(struct virtio_net_hdr);
2629 tun->align = NET_SKB_PAD;
2630 tun->filter_attached = false;
2631 tun->sndbuf = tfile->socket.sk->sk_sndbuf;
2632 tun->rx_batched = 0;
2633 RCU_INIT_POINTER(tun->steering_prog, NULL);
2635 tun->pcpu_stats = netdev_alloc_pcpu_stats(struct tun_pcpu_stats);
2636 if (!tun->pcpu_stats) {
2641 spin_lock_init(&tun->lock);
2643 err = security_tun_dev_alloc_security(&tun->security);
2650 dev->hw_features = NETIF_F_SG | NETIF_F_FRAGLIST |
2651 TUN_USER_FEATURES | NETIF_F_HW_VLAN_CTAG_TX |
2652 NETIF_F_HW_VLAN_STAG_TX;
2653 dev->features = dev->hw_features | NETIF_F_LLTX;
2654 dev->vlan_features = dev->features &
2655 ~(NETIF_F_HW_VLAN_CTAG_TX |
2656 NETIF_F_HW_VLAN_STAG_TX);
2658 tun->flags = (tun->flags & ~TUN_FEATURES) |
2659 (ifr->ifr_flags & TUN_FEATURES);
2661 INIT_LIST_HEAD(&tun->disabled);
2662 err = tun_attach(tun, file, false, ifr->ifr_flags & IFF_NAPI);
2666 err = register_netdevice(tun->dev);
2671 netif_carrier_on(tun->dev);
2673 tun_debug(KERN_INFO, tun, "tun_set_iff\n");
2675 /* Make sure persistent devices do not get stuck in
2678 if (netif_running(tun->dev))
2679 netif_tx_wake_all_queues(tun->dev);
2681 strcpy(ifr->ifr_name, tun->dev->name);
2685 tun_detach_all(dev);
2686 /* register_netdevice() already called tun_free_netdev() */
2690 tun_flow_uninit(tun);
2691 security_tun_dev_free_security(tun->security);
2693 free_percpu(tun->pcpu_stats);
2699 static void tun_get_iff(struct net *net, struct tun_struct *tun,
2702 tun_debug(KERN_INFO, tun, "tun_get_iff\n");
2704 strcpy(ifr->ifr_name, tun->dev->name);
2706 ifr->ifr_flags = tun_flags(tun);
2710 /* This is like a cut-down ethtool ops, except done via tun fd so no
2711 * privs required. */
2712 static int set_offload(struct tun_struct *tun, unsigned long arg)
2714 netdev_features_t features = 0;
2716 if (arg & TUN_F_CSUM) {
2717 features |= NETIF_F_HW_CSUM;
2720 if (arg & (TUN_F_TSO4|TUN_F_TSO6)) {
2721 if (arg & TUN_F_TSO_ECN) {
2722 features |= NETIF_F_TSO_ECN;
2723 arg &= ~TUN_F_TSO_ECN;
2725 if (arg & TUN_F_TSO4)
2726 features |= NETIF_F_TSO;
2727 if (arg & TUN_F_TSO6)
2728 features |= NETIF_F_TSO6;
2729 arg &= ~(TUN_F_TSO4|TUN_F_TSO6);
2735 /* This gives the user a way to test for new features in future by
2736 * trying to set them. */
2740 tun->set_features = features;
2741 tun->dev->wanted_features &= ~TUN_USER_FEATURES;
2742 tun->dev->wanted_features |= features;
2743 netdev_update_features(tun->dev);
2748 static void tun_detach_filter(struct tun_struct *tun, int n)
2751 struct tun_file *tfile;
2753 for (i = 0; i < n; i++) {
2754 tfile = rtnl_dereference(tun->tfiles[i]);
2755 lock_sock(tfile->socket.sk);
2756 sk_detach_filter(tfile->socket.sk);
2757 release_sock(tfile->socket.sk);
2760 tun->filter_attached = false;
2763 static int tun_attach_filter(struct tun_struct *tun)
2766 struct tun_file *tfile;
2768 for (i = 0; i < tun->numqueues; i++) {
2769 tfile = rtnl_dereference(tun->tfiles[i]);
2770 lock_sock(tfile->socket.sk);
2771 ret = sk_attach_filter(&tun->fprog, tfile->socket.sk);
2772 release_sock(tfile->socket.sk);
2774 tun_detach_filter(tun, i);
2779 tun->filter_attached = true;
2783 static void tun_set_sndbuf(struct tun_struct *tun)
2785 struct tun_file *tfile;
2788 for (i = 0; i < tun->numqueues; i++) {
2789 tfile = rtnl_dereference(tun->tfiles[i]);
2790 tfile->socket.sk->sk_sndbuf = tun->sndbuf;
2794 static int tun_set_queue(struct file *file, struct ifreq *ifr)
2796 struct tun_file *tfile = file->private_data;
2797 struct tun_struct *tun;
2802 if (ifr->ifr_flags & IFF_ATTACH_QUEUE) {
2803 tun = tfile->detached;
2808 ret = security_tun_dev_attach_queue(tun->security);
2811 ret = tun_attach(tun, file, false, tun->flags & IFF_NAPI);
2812 } else if (ifr->ifr_flags & IFF_DETACH_QUEUE) {
2813 tun = rtnl_dereference(tfile->tun);
2814 if (!tun || !(tun->flags & IFF_MULTI_QUEUE) || tfile->detached)
2817 __tun_detach(tfile, false);
2822 netdev_state_change(tun->dev);
2829 static int tun_set_ebpf(struct tun_struct *tun, struct tun_prog **prog_p,
2832 struct bpf_prog *prog;
2835 if (copy_from_user(&fd, data, sizeof(fd)))
2841 prog = bpf_prog_get_type(fd, BPF_PROG_TYPE_SOCKET_FILTER);
2843 return PTR_ERR(prog);
2846 return __tun_set_ebpf(tun, prog_p, prog);
2849 static long __tun_chr_ioctl(struct file *file, unsigned int cmd,
2850 unsigned long arg, int ifreq_len)
2852 struct tun_file *tfile = file->private_data;
2853 struct net *net = sock_net(&tfile->sk);
2854 struct tun_struct *tun;
2855 void __user* argp = (void __user*)arg;
2861 unsigned int ifindex;
2864 bool do_notify = false;
2866 if (cmd == TUNSETIFF || cmd == TUNSETQUEUE ||
2867 (_IOC_TYPE(cmd) == SOCK_IOC_TYPE && cmd != SIOCGSKNS)) {
2868 if (copy_from_user(&ifr, argp, ifreq_len))
2871 memset(&ifr, 0, sizeof(ifr));
2873 if (cmd == TUNGETFEATURES) {
2874 /* Currently this just means: "what IFF flags are valid?".
2875 * This is needed because we never checked for invalid flags on
2878 return put_user(IFF_TUN | IFF_TAP | TUN_FEATURES,
2879 (unsigned int __user*)argp);
2880 } else if (cmd == TUNSETQUEUE) {
2881 return tun_set_queue(file, &ifr);
2882 } else if (cmd == SIOCGSKNS) {
2883 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
2885 return open_related_ns(&net->ns, get_net_ns);
2891 tun = tun_get(tfile);
2892 if (cmd == TUNSETIFF) {
2897 ifr.ifr_name[IFNAMSIZ-1] = '\0';
2899 ret = tun_set_iff(net, file, &ifr);
2904 if (copy_to_user(argp, &ifr, ifreq_len))
2908 if (cmd == TUNSETIFINDEX) {
2914 if (copy_from_user(&ifindex, argp, sizeof(ifindex)))
2918 tfile->ifindex = ifindex;
2926 tun_debug(KERN_INFO, tun, "tun_chr_ioctl cmd %u\n", cmd);
2931 tun_get_iff(current->nsproxy->net_ns, tun, &ifr);
2933 if (tfile->detached)
2934 ifr.ifr_flags |= IFF_DETACH_QUEUE;
2935 if (!tfile->socket.sk->sk_filter)
2936 ifr.ifr_flags |= IFF_NOFILTER;
2938 if (copy_to_user(argp, &ifr, ifreq_len))
2943 /* Disable/Enable checksum */
2945 /* [unimplemented] */
2946 tun_debug(KERN_INFO, tun, "ignored: set checksum %s\n",
2947 arg ? "disabled" : "enabled");
2951 /* Disable/Enable persist mode. Keep an extra reference to the
2952 * module to prevent the module being unprobed.
2954 if (arg && !(tun->flags & IFF_PERSIST)) {
2955 tun->flags |= IFF_PERSIST;
2956 __module_get(THIS_MODULE);
2959 if (!arg && (tun->flags & IFF_PERSIST)) {
2960 tun->flags &= ~IFF_PERSIST;
2961 module_put(THIS_MODULE);
2965 tun_debug(KERN_INFO, tun, "persist %s\n",
2966 arg ? "enabled" : "disabled");
2970 /* Set owner of the device */
2971 owner = make_kuid(current_user_ns(), arg);
2972 if (!uid_valid(owner)) {
2978 tun_debug(KERN_INFO, tun, "owner set to %u\n",
2979 from_kuid(&init_user_ns, tun->owner));
2983 /* Set group of the device */
2984 group = make_kgid(current_user_ns(), arg);
2985 if (!gid_valid(group)) {
2991 tun_debug(KERN_INFO, tun, "group set to %u\n",
2992 from_kgid(&init_user_ns, tun->group));
2996 /* Only allow setting the type when the interface is down */
2997 if (tun->dev->flags & IFF_UP) {
2998 tun_debug(KERN_INFO, tun,
2999 "Linktype set failed because interface is up\n");
3002 tun->dev->type = (int) arg;
3003 tun_debug(KERN_INFO, tun, "linktype set to %d\n",
3015 ret = set_offload(tun, arg);
3018 case TUNSETTXFILTER:
3019 /* Can be set only for TAPs */
3021 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3023 ret = update_filter(&tun->txflt, (void __user *)arg);
3027 /* Get hw address */
3028 memcpy(ifr.ifr_hwaddr.sa_data, tun->dev->dev_addr, ETH_ALEN);
3029 ifr.ifr_hwaddr.sa_family = tun->dev->type;
3030 if (copy_to_user(argp, &ifr, ifreq_len))
3035 /* Set hw address */
3036 tun_debug(KERN_DEBUG, tun, "set hw address: %pM\n",
3037 ifr.ifr_hwaddr.sa_data);
3039 ret = dev_set_mac_address(tun->dev, &ifr.ifr_hwaddr);
3043 sndbuf = tfile->socket.sk->sk_sndbuf;
3044 if (copy_to_user(argp, &sndbuf, sizeof(sndbuf)))
3049 if (copy_from_user(&sndbuf, argp, sizeof(sndbuf))) {
3058 tun->sndbuf = sndbuf;
3059 tun_set_sndbuf(tun);
3062 case TUNGETVNETHDRSZ:
3063 vnet_hdr_sz = tun->vnet_hdr_sz;
3064 if (copy_to_user(argp, &vnet_hdr_sz, sizeof(vnet_hdr_sz)))
3068 case TUNSETVNETHDRSZ:
3069 if (copy_from_user(&vnet_hdr_sz, argp, sizeof(vnet_hdr_sz))) {
3073 if (vnet_hdr_sz < (int)sizeof(struct virtio_net_hdr)) {
3078 tun->vnet_hdr_sz = vnet_hdr_sz;
3082 le = !!(tun->flags & TUN_VNET_LE);
3083 if (put_user(le, (int __user *)argp))
3088 if (get_user(le, (int __user *)argp)) {
3093 tun->flags |= TUN_VNET_LE;
3095 tun->flags &= ~TUN_VNET_LE;
3099 ret = tun_get_vnet_be(tun, argp);
3103 ret = tun_set_vnet_be(tun, argp);
3106 case TUNATTACHFILTER:
3107 /* Can be set only for TAPs */
3109 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3112 if (copy_from_user(&tun->fprog, argp, sizeof(tun->fprog)))
3115 ret = tun_attach_filter(tun);
3118 case TUNDETACHFILTER:
3119 /* Can be set only for TAPs */
3121 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3124 tun_detach_filter(tun, tun->numqueues);
3129 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3132 if (copy_to_user(argp, &tun->fprog, sizeof(tun->fprog)))
3137 case TUNSETSTEERINGEBPF:
3138 ret = tun_set_ebpf(tun, &tun->steering_prog, argp);
3141 case TUNSETFILTEREBPF:
3142 ret = tun_set_ebpf(tun, &tun->filter_prog, argp);
3151 netdev_state_change(tun->dev);
3160 static long tun_chr_ioctl(struct file *file,
3161 unsigned int cmd, unsigned long arg)
3163 return __tun_chr_ioctl(file, cmd, arg, sizeof (struct ifreq));
3166 #ifdef CONFIG_COMPAT
3167 static long tun_chr_compat_ioctl(struct file *file,
3168 unsigned int cmd, unsigned long arg)
3173 case TUNSETTXFILTER:
3178 arg = (unsigned long)compat_ptr(arg);
3181 arg = (compat_ulong_t)arg;
3186 * compat_ifreq is shorter than ifreq, so we must not access beyond
3187 * the end of that structure. All fields that are used in this
3188 * driver are compatible though, we don't need to convert the
3191 return __tun_chr_ioctl(file, cmd, arg, sizeof(struct compat_ifreq));
3193 #endif /* CONFIG_COMPAT */
3195 static int tun_chr_fasync(int fd, struct file *file, int on)
3197 struct tun_file *tfile = file->private_data;
3200 if ((ret = fasync_helper(fd, file, on, &tfile->fasync)) < 0)
3204 __f_setown(file, task_pid(current), PIDTYPE_PID, 0);
3205 tfile->flags |= TUN_FASYNC;
3207 tfile->flags &= ~TUN_FASYNC;
3213 static int tun_chr_open(struct inode *inode, struct file * file)
3215 struct net *net = current->nsproxy->net_ns;
3216 struct tun_file *tfile;
3218 DBG1(KERN_INFO, "tunX: tun_chr_open\n");
3220 tfile = (struct tun_file *)sk_alloc(net, AF_UNSPEC, GFP_KERNEL,
3224 RCU_INIT_POINTER(tfile->tun, NULL);
3228 init_waitqueue_head(&tfile->wq.wait);
3229 RCU_INIT_POINTER(tfile->socket.wq, &tfile->wq);
3231 tfile->socket.file = file;
3232 tfile->socket.ops = &tun_socket_ops;
3234 sock_init_data(&tfile->socket, &tfile->sk);
3236 tfile->sk.sk_write_space = tun_sock_write_space;
3237 tfile->sk.sk_sndbuf = INT_MAX;
3239 file->private_data = tfile;
3240 INIT_LIST_HEAD(&tfile->next);
3242 sock_set_flag(&tfile->sk, SOCK_ZEROCOPY);
3244 memset(&tfile->tx_ring, 0, sizeof(tfile->tx_ring));
3249 static int tun_chr_close(struct inode *inode, struct file *file)
3251 struct tun_file *tfile = file->private_data;
3253 tun_detach(tfile, true);
3258 #ifdef CONFIG_PROC_FS
3259 static void tun_chr_show_fdinfo(struct seq_file *m, struct file *file)
3261 struct tun_file *tfile = file->private_data;
3262 struct tun_struct *tun;
3265 memset(&ifr, 0, sizeof(ifr));
3268 tun = tun_get(tfile);
3270 tun_get_iff(current->nsproxy->net_ns, tun, &ifr);
3276 seq_printf(m, "iff:\t%s\n", ifr.ifr_name);
3280 static const struct file_operations tun_fops = {
3281 .owner = THIS_MODULE,
3282 .llseek = no_llseek,
3283 .read_iter = tun_chr_read_iter,
3284 .write_iter = tun_chr_write_iter,
3285 .poll = tun_chr_poll,
3286 .unlocked_ioctl = tun_chr_ioctl,
3287 #ifdef CONFIG_COMPAT
3288 .compat_ioctl = tun_chr_compat_ioctl,
3290 .open = tun_chr_open,
3291 .release = tun_chr_close,
3292 .fasync = tun_chr_fasync,
3293 #ifdef CONFIG_PROC_FS
3294 .show_fdinfo = tun_chr_show_fdinfo,
3298 static struct miscdevice tun_miscdev = {
3301 .nodename = "net/tun",
3305 /* ethtool interface */
3307 static int tun_get_link_ksettings(struct net_device *dev,
3308 struct ethtool_link_ksettings *cmd)
3310 ethtool_link_ksettings_zero_link_mode(cmd, supported);
3311 ethtool_link_ksettings_zero_link_mode(cmd, advertising);
3312 cmd->base.speed = SPEED_10;
3313 cmd->base.duplex = DUPLEX_FULL;
3314 cmd->base.port = PORT_TP;
3315 cmd->base.phy_address = 0;
3316 cmd->base.autoneg = AUTONEG_DISABLE;
3320 static void tun_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
3322 struct tun_struct *tun = netdev_priv(dev);
3324 strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
3325 strlcpy(info->version, DRV_VERSION, sizeof(info->version));
3327 switch (tun->flags & TUN_TYPE_MASK) {
3329 strlcpy(info->bus_info, "tun", sizeof(info->bus_info));
3332 strlcpy(info->bus_info, "tap", sizeof(info->bus_info));
3337 static u32 tun_get_msglevel(struct net_device *dev)
3340 struct tun_struct *tun = netdev_priv(dev);
3347 static void tun_set_msglevel(struct net_device *dev, u32 value)
3350 struct tun_struct *tun = netdev_priv(dev);
3355 static int tun_get_coalesce(struct net_device *dev,
3356 struct ethtool_coalesce *ec)
3358 struct tun_struct *tun = netdev_priv(dev);
3360 ec->rx_max_coalesced_frames = tun->rx_batched;
3365 static int tun_set_coalesce(struct net_device *dev,
3366 struct ethtool_coalesce *ec)
3368 struct tun_struct *tun = netdev_priv(dev);
3370 if (ec->rx_max_coalesced_frames > NAPI_POLL_WEIGHT)
3371 tun->rx_batched = NAPI_POLL_WEIGHT;
3373 tun->rx_batched = ec->rx_max_coalesced_frames;
3378 static const struct ethtool_ops tun_ethtool_ops = {
3379 .get_drvinfo = tun_get_drvinfo,
3380 .get_msglevel = tun_get_msglevel,
3381 .set_msglevel = tun_set_msglevel,
3382 .get_link = ethtool_op_get_link,
3383 .get_ts_info = ethtool_op_get_ts_info,
3384 .get_coalesce = tun_get_coalesce,
3385 .set_coalesce = tun_set_coalesce,
3386 .get_link_ksettings = tun_get_link_ksettings,
3389 static int tun_queue_resize(struct tun_struct *tun)
3391 struct net_device *dev = tun->dev;
3392 struct tun_file *tfile;
3393 struct ptr_ring **rings;
3394 int n = tun->numqueues + tun->numdisabled;
3397 rings = kmalloc_array(n, sizeof(*rings), GFP_KERNEL);
3401 for (i = 0; i < tun->numqueues; i++) {
3402 tfile = rtnl_dereference(tun->tfiles[i]);
3403 rings[i] = &tfile->tx_ring;
3405 list_for_each_entry(tfile, &tun->disabled, next)
3406 rings[i++] = &tfile->tx_ring;
3408 ret = ptr_ring_resize_multiple(rings, n,
3409 dev->tx_queue_len, GFP_KERNEL,
3416 static int tun_device_event(struct notifier_block *unused,
3417 unsigned long event, void *ptr)
3419 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
3420 struct tun_struct *tun = netdev_priv(dev);
3422 if (dev->rtnl_link_ops != &tun_link_ops)
3426 case NETDEV_CHANGE_TX_QUEUE_LEN:
3427 if (tun_queue_resize(tun))
3437 static struct notifier_block tun_notifier_block __read_mostly = {
3438 .notifier_call = tun_device_event,
3441 static int __init tun_init(void)
3445 pr_info("%s, %s\n", DRV_DESCRIPTION, DRV_VERSION);
3447 ret = rtnl_link_register(&tun_link_ops);
3449 pr_err("Can't register link_ops\n");
3453 ret = misc_register(&tun_miscdev);
3455 pr_err("Can't register misc device %d\n", TUN_MINOR);
3459 ret = register_netdevice_notifier(&tun_notifier_block);
3461 pr_err("Can't register netdevice notifier\n");
3468 misc_deregister(&tun_miscdev);
3470 rtnl_link_unregister(&tun_link_ops);
3475 static void tun_cleanup(void)
3477 misc_deregister(&tun_miscdev);
3478 rtnl_link_unregister(&tun_link_ops);
3479 unregister_netdevice_notifier(&tun_notifier_block);
3482 /* Get an underlying socket object from tun file. Returns error unless file is
3483 * attached to a device. The returned object works like a packet socket, it
3484 * can be used for sock_sendmsg/sock_recvmsg. The caller is responsible for
3485 * holding a reference to the file for as long as the socket is in use. */
3486 struct socket *tun_get_socket(struct file *file)
3488 struct tun_file *tfile;
3489 if (file->f_op != &tun_fops)
3490 return ERR_PTR(-EINVAL);
3491 tfile = file->private_data;
3493 return ERR_PTR(-EBADFD);
3494 return &tfile->socket;
3496 EXPORT_SYMBOL_GPL(tun_get_socket);
3498 struct ptr_ring *tun_get_tx_ring(struct file *file)
3500 struct tun_file *tfile;
3502 if (file->f_op != &tun_fops)
3503 return ERR_PTR(-EINVAL);
3504 tfile = file->private_data;
3506 return ERR_PTR(-EBADFD);
3507 return &tfile->tx_ring;
3509 EXPORT_SYMBOL_GPL(tun_get_tx_ring);
3511 module_init(tun_init);
3512 module_exit(tun_cleanup);
3513 MODULE_DESCRIPTION(DRV_DESCRIPTION);
3514 MODULE_AUTHOR(DRV_COPYRIGHT);
3515 MODULE_LICENSE("GPL");
3516 MODULE_ALIAS_MISCDEV(TUN_MINOR);
3517 MODULE_ALIAS("devname:net/tun");