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>
74 #include <linux/seq_file.h>
75 #include <linux/uio.h>
76 #include <linux/skb_array.h>
77 #include <linux/bpf.h>
78 #include <linux/bpf_trace.h>
79 #include <linux/mutex.h>
81 #include <linux/uaccess.h>
82 #include <linux/proc_fs.h>
84 /* Uncomment to enable debugging */
85 /* #define TUN_DEBUG 1 */
90 #define tun_debug(level, tun, fmt, args...) \
93 netdev_printk(level, tun->dev, fmt, ##args); \
95 #define DBG1(level, fmt, args...) \
98 printk(level fmt, ##args); \
101 #define tun_debug(level, tun, fmt, args...) \
104 netdev_printk(level, tun->dev, fmt, ##args); \
106 #define DBG1(level, fmt, args...) \
109 printk(level fmt, ##args); \
113 #define TUN_HEADROOM 256
114 #define TUN_RX_PAD (NET_IP_ALIGN + NET_SKB_PAD)
116 /* TUN device flags */
118 /* IFF_ATTACH_QUEUE is never stored in device flags,
119 * overload it to mean fasync when stored there.
121 #define TUN_FASYNC IFF_ATTACH_QUEUE
122 /* High bits in flags field are unused. */
123 #define TUN_VNET_LE 0x80000000
124 #define TUN_VNET_BE 0x40000000
126 #define TUN_FEATURES (IFF_NO_PI | IFF_ONE_QUEUE | IFF_VNET_HDR | \
127 IFF_MULTI_QUEUE | IFF_NAPI | IFF_NAPI_FRAGS)
129 #define GOODCOPY_LEN 128
131 #define FLT_EXACT_COUNT 8
133 unsigned int count; /* Number of addrs. Zero means disabled */
134 u32 mask[2]; /* Mask of the hashed addrs */
135 unsigned char addr[FLT_EXACT_COUNT][ETH_ALEN];
138 /* MAX_TAP_QUEUES 256 is chosen to allow rx/tx queues to be equal
139 * to max number of VCPUs in guest. */
140 #define MAX_TAP_QUEUES 256
141 #define MAX_TAP_FLOWS 4096
143 #define TUN_FLOW_EXPIRE (3 * HZ)
145 struct tun_pcpu_stats {
150 struct u64_stats_sync syncp;
156 /* A tun_file connects an open character device to a tuntap netdevice. It
157 * also contains all socket related structures (except sock_fprog and tap_filter)
158 * to serve as one transmit queue for tuntap device. The sock_fprog and
159 * tap_filter were kept in tun_struct since they were used for filtering for the
160 * netdevice not for a specific queue (at least I didn't see the requirement for
164 * The tun_file and tun_struct are loosely coupled, the pointer from one to the
165 * other can only be read while rcu_read_lock or rtnl_lock is held.
169 struct socket socket;
171 struct tun_struct __rcu *tun;
172 struct fasync_struct *fasync;
173 /* only used for fasnyc */
177 unsigned int ifindex;
179 struct napi_struct napi;
181 struct mutex napi_mutex; /* Protects access to the above napi */
182 struct list_head next;
183 struct tun_struct *detached;
184 struct ptr_ring tx_ring;
185 struct xdp_rxq_info xdp_rxq;
188 struct tun_flow_entry {
189 struct hlist_node hash_link;
191 struct tun_struct *tun;
196 unsigned long updated;
199 #define TUN_NUM_FLOW_ENTRIES 1024
203 struct bpf_prog *prog;
206 /* Since the socket were moved to tun_file, to preserve the behavior of persist
207 * device, socket filter, sndbuf and vnet header size were restore when the
208 * file were attached to a persist device.
211 struct tun_file __rcu *tfiles[MAX_TAP_QUEUES];
212 unsigned int numqueues;
217 struct net_device *dev;
218 netdev_features_t set_features;
219 #define TUN_USER_FEATURES (NETIF_F_HW_CSUM|NETIF_F_TSO_ECN|NETIF_F_TSO| \
225 struct tap_filter txflt;
226 struct sock_fprog fprog;
227 /* protected by rtnl lock */
228 bool filter_attached;
233 struct hlist_head flows[TUN_NUM_FLOW_ENTRIES];
234 struct timer_list flow_gc_timer;
235 unsigned long ageing_time;
236 unsigned int numdisabled;
237 struct list_head disabled;
241 struct tun_pcpu_stats __percpu *pcpu_stats;
242 struct bpf_prog __rcu *xdp_prog;
243 struct tun_prog __rcu *steering_prog;
244 struct tun_prog __rcu *filter_prog;
252 bool tun_is_xdp_frame(void *ptr)
254 return (unsigned long)ptr & TUN_XDP_FLAG;
256 EXPORT_SYMBOL(tun_is_xdp_frame);
258 void *tun_xdp_to_ptr(void *ptr)
260 return (void *)((unsigned long)ptr | TUN_XDP_FLAG);
262 EXPORT_SYMBOL(tun_xdp_to_ptr);
264 void *tun_ptr_to_xdp(void *ptr)
266 return (void *)((unsigned long)ptr & ~TUN_XDP_FLAG);
268 EXPORT_SYMBOL(tun_ptr_to_xdp);
270 static int tun_napi_receive(struct napi_struct *napi, int budget)
272 struct tun_file *tfile = container_of(napi, struct tun_file, napi);
273 struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
274 struct sk_buff_head process_queue;
278 __skb_queue_head_init(&process_queue);
280 spin_lock(&queue->lock);
281 skb_queue_splice_tail_init(queue, &process_queue);
282 spin_unlock(&queue->lock);
284 while (received < budget && (skb = __skb_dequeue(&process_queue))) {
285 napi_gro_receive(napi, skb);
289 if (!skb_queue_empty(&process_queue)) {
290 spin_lock(&queue->lock);
291 skb_queue_splice(&process_queue, queue);
292 spin_unlock(&queue->lock);
298 static int tun_napi_poll(struct napi_struct *napi, int budget)
300 unsigned int received;
302 received = tun_napi_receive(napi, budget);
304 if (received < budget)
305 napi_complete_done(napi, received);
310 static void tun_napi_init(struct tun_struct *tun, struct tun_file *tfile,
313 tfile->napi_enabled = napi_en;
315 netif_napi_add(tun->dev, &tfile->napi, tun_napi_poll,
317 napi_enable(&tfile->napi);
318 mutex_init(&tfile->napi_mutex);
322 static void tun_napi_disable(struct tun_struct *tun, struct tun_file *tfile)
324 if (tfile->napi_enabled)
325 napi_disable(&tfile->napi);
328 static void tun_napi_del(struct tun_struct *tun, struct tun_file *tfile)
330 if (tfile->napi_enabled)
331 netif_napi_del(&tfile->napi);
334 static bool tun_napi_frags_enabled(const struct tun_struct *tun)
336 return READ_ONCE(tun->flags) & IFF_NAPI_FRAGS;
339 #ifdef CONFIG_TUN_VNET_CROSS_LE
340 static inline bool tun_legacy_is_little_endian(struct tun_struct *tun)
342 return tun->flags & TUN_VNET_BE ? false :
343 virtio_legacy_is_little_endian();
346 static long tun_get_vnet_be(struct tun_struct *tun, int __user *argp)
348 int be = !!(tun->flags & TUN_VNET_BE);
350 if (put_user(be, argp))
356 static long tun_set_vnet_be(struct tun_struct *tun, int __user *argp)
360 if (get_user(be, argp))
364 tun->flags |= TUN_VNET_BE;
366 tun->flags &= ~TUN_VNET_BE;
371 static inline bool tun_legacy_is_little_endian(struct tun_struct *tun)
373 return virtio_legacy_is_little_endian();
376 static long tun_get_vnet_be(struct tun_struct *tun, int __user *argp)
381 static long tun_set_vnet_be(struct tun_struct *tun, int __user *argp)
385 #endif /* CONFIG_TUN_VNET_CROSS_LE */
387 static inline bool tun_is_little_endian(struct tun_struct *tun)
389 return tun->flags & TUN_VNET_LE ||
390 tun_legacy_is_little_endian(tun);
393 static inline u16 tun16_to_cpu(struct tun_struct *tun, __virtio16 val)
395 return __virtio16_to_cpu(tun_is_little_endian(tun), val);
398 static inline __virtio16 cpu_to_tun16(struct tun_struct *tun, u16 val)
400 return __cpu_to_virtio16(tun_is_little_endian(tun), val);
403 static inline u32 tun_hashfn(u32 rxhash)
405 return rxhash & 0x3ff;
408 static struct tun_flow_entry *tun_flow_find(struct hlist_head *head, u32 rxhash)
410 struct tun_flow_entry *e;
412 hlist_for_each_entry_rcu(e, head, hash_link) {
413 if (e->rxhash == rxhash)
419 static struct tun_flow_entry *tun_flow_create(struct tun_struct *tun,
420 struct hlist_head *head,
421 u32 rxhash, u16 queue_index)
423 struct tun_flow_entry *e = kmalloc(sizeof(*e), GFP_ATOMIC);
426 tun_debug(KERN_INFO, tun, "create flow: hash %u index %u\n",
427 rxhash, queue_index);
428 e->updated = jiffies;
431 e->queue_index = queue_index;
433 hlist_add_head_rcu(&e->hash_link, head);
439 static void tun_flow_delete(struct tun_struct *tun, struct tun_flow_entry *e)
441 tun_debug(KERN_INFO, tun, "delete flow: hash %u index %u\n",
442 e->rxhash, e->queue_index);
443 hlist_del_rcu(&e->hash_link);
448 static void tun_flow_flush(struct tun_struct *tun)
452 spin_lock_bh(&tun->lock);
453 for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
454 struct tun_flow_entry *e;
455 struct hlist_node *n;
457 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link)
458 tun_flow_delete(tun, e);
460 spin_unlock_bh(&tun->lock);
463 static void tun_flow_delete_by_queue(struct tun_struct *tun, u16 queue_index)
467 spin_lock_bh(&tun->lock);
468 for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
469 struct tun_flow_entry *e;
470 struct hlist_node *n;
472 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
473 if (e->queue_index == queue_index)
474 tun_flow_delete(tun, e);
477 spin_unlock_bh(&tun->lock);
480 static void tun_flow_cleanup(struct timer_list *t)
482 struct tun_struct *tun = from_timer(tun, t, flow_gc_timer);
483 unsigned long delay = tun->ageing_time;
484 unsigned long next_timer = jiffies + delay;
485 unsigned long count = 0;
488 tun_debug(KERN_INFO, tun, "tun_flow_cleanup\n");
490 spin_lock(&tun->lock);
491 for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
492 struct tun_flow_entry *e;
493 struct hlist_node *n;
495 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
496 unsigned long this_timer;
498 this_timer = e->updated + delay;
499 if (time_before_eq(this_timer, jiffies)) {
500 tun_flow_delete(tun, e);
504 if (time_before(this_timer, next_timer))
505 next_timer = this_timer;
510 mod_timer(&tun->flow_gc_timer, round_jiffies_up(next_timer));
511 spin_unlock(&tun->lock);
514 static void tun_flow_update(struct tun_struct *tun, u32 rxhash,
515 struct tun_file *tfile)
517 struct hlist_head *head;
518 struct tun_flow_entry *e;
519 unsigned long delay = tun->ageing_time;
520 u16 queue_index = tfile->queue_index;
525 head = &tun->flows[tun_hashfn(rxhash)];
529 e = tun_flow_find(head, rxhash);
531 /* TODO: keep queueing to old queue until it's empty? */
532 e->queue_index = queue_index;
533 e->updated = jiffies;
534 sock_rps_record_flow_hash(e->rps_rxhash);
536 spin_lock_bh(&tun->lock);
537 if (!tun_flow_find(head, rxhash) &&
538 tun->flow_count < MAX_TAP_FLOWS)
539 tun_flow_create(tun, head, rxhash, queue_index);
541 if (!timer_pending(&tun->flow_gc_timer))
542 mod_timer(&tun->flow_gc_timer,
543 round_jiffies_up(jiffies + delay));
544 spin_unlock_bh(&tun->lock);
551 * Save the hash received in the stack receive path and update the
552 * flow_hash table accordingly.
554 static inline void tun_flow_save_rps_rxhash(struct tun_flow_entry *e, u32 hash)
556 if (unlikely(e->rps_rxhash != hash))
557 e->rps_rxhash = hash;
560 /* We try to identify a flow through its rxhash first. The reason that
561 * we do not check rxq no. is because some cards(e.g 82599), chooses
562 * the rxq based on the txq where the last packet of the flow comes. As
563 * the userspace application move between processors, we may get a
564 * different rxq no. here. If we could not get rxhash, then we would
565 * hope the rxq no. may help here.
567 static u16 tun_automq_select_queue(struct tun_struct *tun, struct sk_buff *skb)
569 struct tun_flow_entry *e;
573 numqueues = READ_ONCE(tun->numqueues);
575 txq = __skb_get_hash_symmetric(skb);
577 e = tun_flow_find(&tun->flows[tun_hashfn(txq)], txq);
579 tun_flow_save_rps_rxhash(e, txq);
580 txq = e->queue_index;
582 /* use multiply and shift instead of expensive divide */
583 txq = ((u64)txq * numqueues) >> 32;
584 } else if (likely(skb_rx_queue_recorded(skb))) {
585 txq = skb_get_rx_queue(skb);
586 while (unlikely(txq >= numqueues))
593 static u16 tun_ebpf_select_queue(struct tun_struct *tun, struct sk_buff *skb)
595 struct tun_prog *prog;
598 prog = rcu_dereference(tun->steering_prog);
600 ret = bpf_prog_run_clear_cb(prog->prog, skb);
602 return ret % tun->numqueues;
605 static u16 tun_select_queue(struct net_device *dev, struct sk_buff *skb,
606 void *accel_priv, select_queue_fallback_t fallback)
608 struct tun_struct *tun = netdev_priv(dev);
612 if (rcu_dereference(tun->steering_prog))
613 ret = tun_ebpf_select_queue(tun, skb);
615 ret = tun_automq_select_queue(tun, skb);
621 static inline bool tun_not_capable(struct tun_struct *tun)
623 const struct cred *cred = current_cred();
624 struct net *net = dev_net(tun->dev);
626 return ((uid_valid(tun->owner) && !uid_eq(cred->euid, tun->owner)) ||
627 (gid_valid(tun->group) && !in_egroup_p(tun->group))) &&
628 !ns_capable(net->user_ns, CAP_NET_ADMIN);
631 static void tun_set_real_num_queues(struct tun_struct *tun)
633 netif_set_real_num_tx_queues(tun->dev, tun->numqueues);
634 netif_set_real_num_rx_queues(tun->dev, tun->numqueues);
637 static void tun_disable_queue(struct tun_struct *tun, struct tun_file *tfile)
639 tfile->detached = tun;
640 list_add_tail(&tfile->next, &tun->disabled);
644 static struct tun_struct *tun_enable_queue(struct tun_file *tfile)
646 struct tun_struct *tun = tfile->detached;
648 tfile->detached = NULL;
649 list_del_init(&tfile->next);
654 void tun_ptr_free(void *ptr)
658 if (tun_is_xdp_frame(ptr)) {
659 struct xdp_frame *xdpf = tun_ptr_to_xdp(ptr);
661 xdp_return_frame(xdpf);
663 __skb_array_destroy_skb(ptr);
666 EXPORT_SYMBOL_GPL(tun_ptr_free);
668 static void tun_queue_purge(struct tun_file *tfile)
672 while ((ptr = ptr_ring_consume(&tfile->tx_ring)) != NULL)
675 skb_queue_purge(&tfile->sk.sk_write_queue);
676 skb_queue_purge(&tfile->sk.sk_error_queue);
679 static void __tun_detach(struct tun_file *tfile, bool clean)
681 struct tun_file *ntfile;
682 struct tun_struct *tun;
684 tun = rtnl_dereference(tfile->tun);
687 tun_napi_disable(tun, tfile);
688 tun_napi_del(tun, tfile);
691 if (tun && !tfile->detached) {
692 u16 index = tfile->queue_index;
693 BUG_ON(index >= tun->numqueues);
695 rcu_assign_pointer(tun->tfiles[index],
696 tun->tfiles[tun->numqueues - 1]);
697 ntfile = rtnl_dereference(tun->tfiles[index]);
698 ntfile->queue_index = index;
702 RCU_INIT_POINTER(tfile->tun, NULL);
703 sock_put(&tfile->sk);
705 tun_disable_queue(tun, tfile);
708 tun_flow_delete_by_queue(tun, tun->numqueues + 1);
709 /* Drop read queue */
710 tun_queue_purge(tfile);
711 tun_set_real_num_queues(tun);
712 } else if (tfile->detached && clean) {
713 tun = tun_enable_queue(tfile);
714 sock_put(&tfile->sk);
718 if (tun && tun->numqueues == 0 && tun->numdisabled == 0) {
719 netif_carrier_off(tun->dev);
721 if (!(tun->flags & IFF_PERSIST) &&
722 tun->dev->reg_state == NETREG_REGISTERED)
723 unregister_netdevice(tun->dev);
726 xdp_rxq_info_unreg(&tfile->xdp_rxq);
727 ptr_ring_cleanup(&tfile->tx_ring, tun_ptr_free);
728 sock_put(&tfile->sk);
732 static void tun_detach(struct tun_file *tfile, bool clean)
734 struct tun_struct *tun;
735 struct net_device *dev;
738 tun = rtnl_dereference(tfile->tun);
739 dev = tun ? tun->dev : NULL;
740 __tun_detach(tfile, clean);
742 netdev_state_change(dev);
746 static void tun_detach_all(struct net_device *dev)
748 struct tun_struct *tun = netdev_priv(dev);
749 struct tun_file *tfile, *tmp;
750 int i, n = tun->numqueues;
752 for (i = 0; i < n; i++) {
753 tfile = rtnl_dereference(tun->tfiles[i]);
755 tun_napi_disable(tun, tfile);
756 tfile->socket.sk->sk_shutdown = RCV_SHUTDOWN;
757 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
758 RCU_INIT_POINTER(tfile->tun, NULL);
761 list_for_each_entry(tfile, &tun->disabled, next) {
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);
766 BUG_ON(tun->numqueues != 0);
769 for (i = 0; i < n; i++) {
770 tfile = rtnl_dereference(tun->tfiles[i]);
771 tun_napi_del(tun, tfile);
772 /* Drop read queue */
773 tun_queue_purge(tfile);
774 xdp_rxq_info_unreg(&tfile->xdp_rxq);
775 sock_put(&tfile->sk);
777 list_for_each_entry_safe(tfile, tmp, &tun->disabled, next) {
778 tun_enable_queue(tfile);
779 tun_queue_purge(tfile);
780 xdp_rxq_info_unreg(&tfile->xdp_rxq);
781 sock_put(&tfile->sk);
783 BUG_ON(tun->numdisabled != 0);
785 if (tun->flags & IFF_PERSIST)
786 module_put(THIS_MODULE);
789 static int tun_attach(struct tun_struct *tun, struct file *file,
790 bool skip_filter, bool napi)
792 struct tun_file *tfile = file->private_data;
793 struct net_device *dev = tun->dev;
796 err = security_tun_dev_attach(tfile->socket.sk, tun->security);
801 if (rtnl_dereference(tfile->tun) && !tfile->detached)
805 if (!(tun->flags & IFF_MULTI_QUEUE) && tun->numqueues == 1)
809 if (!tfile->detached &&
810 tun->numqueues + tun->numdisabled == MAX_TAP_QUEUES)
815 /* Re-attach the filter to persist device */
816 if (!skip_filter && (tun->filter_attached == true)) {
817 lock_sock(tfile->socket.sk);
818 err = sk_attach_filter(&tun->fprog, tfile->socket.sk);
819 release_sock(tfile->socket.sk);
824 if (!tfile->detached &&
825 ptr_ring_resize(&tfile->tx_ring, dev->tx_queue_len,
826 GFP_KERNEL, tun_ptr_free)) {
831 tfile->queue_index = tun->numqueues;
832 tfile->socket.sk->sk_shutdown &= ~RCV_SHUTDOWN;
834 if (tfile->detached) {
835 /* Re-attach detached tfile, updating XDP queue_index */
836 WARN_ON(!xdp_rxq_info_is_reg(&tfile->xdp_rxq));
838 if (tfile->xdp_rxq.queue_index != tfile->queue_index)
839 tfile->xdp_rxq.queue_index = tfile->queue_index;
841 /* Setup XDP RX-queue info, for new tfile getting attached */
842 err = xdp_rxq_info_reg(&tfile->xdp_rxq,
843 tun->dev, tfile->queue_index);
846 err = xdp_rxq_info_reg_mem_model(&tfile->xdp_rxq,
847 MEM_TYPE_PAGE_SHARED, NULL);
849 xdp_rxq_info_unreg(&tfile->xdp_rxq);
855 rcu_assign_pointer(tfile->tun, tun);
856 rcu_assign_pointer(tun->tfiles[tun->numqueues], tfile);
859 if (tfile->detached) {
860 tun_enable_queue(tfile);
862 sock_hold(&tfile->sk);
863 tun_napi_init(tun, tfile, napi);
866 tun_set_real_num_queues(tun);
868 /* device is allowed to go away first, so no need to hold extra
876 static struct tun_struct *tun_get(struct tun_file *tfile)
878 struct tun_struct *tun;
881 tun = rcu_dereference(tfile->tun);
889 static void tun_put(struct tun_struct *tun)
895 static void addr_hash_set(u32 *mask, const u8 *addr)
897 int n = ether_crc(ETH_ALEN, addr) >> 26;
898 mask[n >> 5] |= (1 << (n & 31));
901 static unsigned int addr_hash_test(const u32 *mask, const u8 *addr)
903 int n = ether_crc(ETH_ALEN, addr) >> 26;
904 return mask[n >> 5] & (1 << (n & 31));
907 static int update_filter(struct tap_filter *filter, void __user *arg)
909 struct { u8 u[ETH_ALEN]; } *addr;
910 struct tun_filter uf;
911 int err, alen, n, nexact;
913 if (copy_from_user(&uf, arg, sizeof(uf)))
922 alen = ETH_ALEN * uf.count;
923 addr = memdup_user(arg + sizeof(uf), alen);
925 return PTR_ERR(addr);
927 /* The filter is updated without holding any locks. Which is
928 * perfectly safe. We disable it first and in the worst
929 * case we'll accept a few undesired packets. */
933 /* Use first set of addresses as an exact filter */
934 for (n = 0; n < uf.count && n < FLT_EXACT_COUNT; n++)
935 memcpy(filter->addr[n], addr[n].u, ETH_ALEN);
939 /* Remaining multicast addresses are hashed,
940 * unicast will leave the filter disabled. */
941 memset(filter->mask, 0, sizeof(filter->mask));
942 for (; n < uf.count; n++) {
943 if (!is_multicast_ether_addr(addr[n].u)) {
944 err = 0; /* no filter */
947 addr_hash_set(filter->mask, addr[n].u);
950 /* For ALLMULTI just set the mask to all ones.
951 * This overrides the mask populated above. */
952 if ((uf.flags & TUN_FLT_ALLMULTI))
953 memset(filter->mask, ~0, sizeof(filter->mask));
955 /* Now enable the filter */
957 filter->count = nexact;
959 /* Return the number of exact filters */
966 /* Returns: 0 - drop, !=0 - accept */
967 static int run_filter(struct tap_filter *filter, const struct sk_buff *skb)
969 /* Cannot use eth_hdr(skb) here because skb_mac_hdr() is incorrect
971 struct ethhdr *eh = (struct ethhdr *) skb->data;
975 for (i = 0; i < filter->count; i++)
976 if (ether_addr_equal(eh->h_dest, filter->addr[i]))
979 /* Inexact match (multicast only) */
980 if (is_multicast_ether_addr(eh->h_dest))
981 return addr_hash_test(filter->mask, eh->h_dest);
987 * Checks whether the packet is accepted or not.
988 * Returns: 0 - drop, !=0 - accept
990 static int check_filter(struct tap_filter *filter, const struct sk_buff *skb)
995 return run_filter(filter, skb);
998 /* Network device part of the driver */
1000 static const struct ethtool_ops tun_ethtool_ops;
1002 /* Net device detach from fd. */
1003 static void tun_net_uninit(struct net_device *dev)
1005 tun_detach_all(dev);
1008 /* Net device open. */
1009 static int tun_net_open(struct net_device *dev)
1011 struct tun_struct *tun = netdev_priv(dev);
1014 netif_tx_start_all_queues(dev);
1016 for (i = 0; i < tun->numqueues; i++) {
1017 struct tun_file *tfile;
1019 tfile = rtnl_dereference(tun->tfiles[i]);
1020 tfile->socket.sk->sk_write_space(tfile->socket.sk);
1026 /* Net device close. */
1027 static int tun_net_close(struct net_device *dev)
1029 netif_tx_stop_all_queues(dev);
1033 /* Net device start xmit */
1034 static void tun_automq_xmit(struct tun_struct *tun, struct sk_buff *skb)
1037 if (tun->numqueues == 1 && static_key_false(&rps_needed)) {
1038 /* Select queue was not called for the skbuff, so we extract the
1039 * RPS hash and save it into the flow_table here.
1043 rxhash = __skb_get_hash_symmetric(skb);
1045 struct tun_flow_entry *e;
1046 e = tun_flow_find(&tun->flows[tun_hashfn(rxhash)],
1049 tun_flow_save_rps_rxhash(e, rxhash);
1055 static unsigned int run_ebpf_filter(struct tun_struct *tun,
1056 struct sk_buff *skb,
1059 struct tun_prog *prog = rcu_dereference(tun->filter_prog);
1062 len = bpf_prog_run_clear_cb(prog->prog, skb);
1067 /* Net device start xmit */
1068 static netdev_tx_t tun_net_xmit(struct sk_buff *skb, struct net_device *dev)
1070 struct tun_struct *tun = netdev_priv(dev);
1071 int txq = skb->queue_mapping;
1072 struct tun_file *tfile;
1076 tfile = rcu_dereference(tun->tfiles[txq]);
1078 /* Drop packet if interface is not attached */
1079 if (txq >= tun->numqueues)
1082 if (!rcu_dereference(tun->steering_prog))
1083 tun_automq_xmit(tun, skb);
1085 tun_debug(KERN_INFO, tun, "tun_net_xmit %d\n", skb->len);
1089 /* Drop if the filter does not like it.
1090 * This is a noop if the filter is disabled.
1091 * Filter can be enabled only for the TAP devices. */
1092 if (!check_filter(&tun->txflt, skb))
1095 if (tfile->socket.sk->sk_filter &&
1096 sk_filter(tfile->socket.sk, skb))
1099 len = run_ebpf_filter(tun, skb, len);
1100 if (len == 0 || pskb_trim(skb, len))
1103 if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC)))
1106 skb_tx_timestamp(skb);
1108 /* Orphan the skb - required as we might hang on to it
1109 * for indefinite time.
1115 if (ptr_ring_produce(&tfile->tx_ring, skb))
1118 /* Notify and wake up reader process */
1119 if (tfile->flags & TUN_FASYNC)
1120 kill_fasync(&tfile->fasync, SIGIO, POLL_IN);
1121 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
1124 return NETDEV_TX_OK;
1127 this_cpu_inc(tun->pcpu_stats->tx_dropped);
1131 return NET_XMIT_DROP;
1134 static void tun_net_mclist(struct net_device *dev)
1137 * This callback is supposed to deal with mc filter in
1138 * _rx_ path and has nothing to do with the _tx_ path.
1139 * In rx path we always accept everything userspace gives us.
1143 static netdev_features_t tun_net_fix_features(struct net_device *dev,
1144 netdev_features_t features)
1146 struct tun_struct *tun = netdev_priv(dev);
1148 return (features & tun->set_features) | (features & ~TUN_USER_FEATURES);
1150 #ifdef CONFIG_NET_POLL_CONTROLLER
1151 static void tun_poll_controller(struct net_device *dev)
1154 * Tun only receives frames when:
1155 * 1) the char device endpoint gets data from user space
1156 * 2) the tun socket gets a sendmsg call from user space
1157 * If NAPI is not enabled, since both of those are synchronous
1158 * operations, we are guaranteed never to have pending data when we poll
1159 * for it so there is nothing to do here but return.
1160 * We need this though so netpoll recognizes us as an interface that
1161 * supports polling, which enables bridge devices in virt setups to
1162 * still use netconsole
1163 * If NAPI is enabled, however, we need to schedule polling for all
1164 * queues unless we are using napi_gro_frags(), which we call in
1165 * process context and not in NAPI context.
1167 struct tun_struct *tun = netdev_priv(dev);
1169 if (tun->flags & IFF_NAPI) {
1170 struct tun_file *tfile;
1173 if (tun_napi_frags_enabled(tun))
1177 for (i = 0; i < tun->numqueues; i++) {
1178 tfile = rcu_dereference(tun->tfiles[i]);
1179 if (tfile->napi_enabled)
1180 napi_schedule(&tfile->napi);
1188 static void tun_set_headroom(struct net_device *dev, int new_hr)
1190 struct tun_struct *tun = netdev_priv(dev);
1192 if (new_hr < NET_SKB_PAD)
1193 new_hr = NET_SKB_PAD;
1195 tun->align = new_hr;
1199 tun_net_get_stats64(struct net_device *dev, struct rtnl_link_stats64 *stats)
1201 u32 rx_dropped = 0, tx_dropped = 0, rx_frame_errors = 0;
1202 struct tun_struct *tun = netdev_priv(dev);
1203 struct tun_pcpu_stats *p;
1206 for_each_possible_cpu(i) {
1207 u64 rxpackets, rxbytes, txpackets, txbytes;
1210 p = per_cpu_ptr(tun->pcpu_stats, i);
1212 start = u64_stats_fetch_begin(&p->syncp);
1213 rxpackets = p->rx_packets;
1214 rxbytes = p->rx_bytes;
1215 txpackets = p->tx_packets;
1216 txbytes = p->tx_bytes;
1217 } while (u64_stats_fetch_retry(&p->syncp, start));
1219 stats->rx_packets += rxpackets;
1220 stats->rx_bytes += rxbytes;
1221 stats->tx_packets += txpackets;
1222 stats->tx_bytes += txbytes;
1225 rx_dropped += p->rx_dropped;
1226 rx_frame_errors += p->rx_frame_errors;
1227 tx_dropped += p->tx_dropped;
1229 stats->rx_dropped = rx_dropped;
1230 stats->rx_frame_errors = rx_frame_errors;
1231 stats->tx_dropped = tx_dropped;
1234 static int tun_xdp_set(struct net_device *dev, struct bpf_prog *prog,
1235 struct netlink_ext_ack *extack)
1237 struct tun_struct *tun = netdev_priv(dev);
1238 struct bpf_prog *old_prog;
1240 old_prog = rtnl_dereference(tun->xdp_prog);
1241 rcu_assign_pointer(tun->xdp_prog, prog);
1243 bpf_prog_put(old_prog);
1248 static u32 tun_xdp_query(struct net_device *dev)
1250 struct tun_struct *tun = netdev_priv(dev);
1251 const struct bpf_prog *xdp_prog;
1253 xdp_prog = rtnl_dereference(tun->xdp_prog);
1255 return xdp_prog->aux->id;
1260 static int tun_xdp(struct net_device *dev, struct netdev_bpf *xdp)
1262 switch (xdp->command) {
1263 case XDP_SETUP_PROG:
1264 return tun_xdp_set(dev, xdp->prog, xdp->extack);
1265 case XDP_QUERY_PROG:
1266 xdp->prog_id = tun_xdp_query(dev);
1267 xdp->prog_attached = !!xdp->prog_id;
1274 static const struct net_device_ops tun_netdev_ops = {
1275 .ndo_uninit = tun_net_uninit,
1276 .ndo_open = tun_net_open,
1277 .ndo_stop = tun_net_close,
1278 .ndo_start_xmit = tun_net_xmit,
1279 .ndo_fix_features = tun_net_fix_features,
1280 .ndo_select_queue = tun_select_queue,
1281 #ifdef CONFIG_NET_POLL_CONTROLLER
1282 .ndo_poll_controller = tun_poll_controller,
1284 .ndo_set_rx_headroom = tun_set_headroom,
1285 .ndo_get_stats64 = tun_net_get_stats64,
1288 static int tun_xdp_xmit(struct net_device *dev, int n, struct xdp_frame **frames)
1290 struct tun_struct *tun = netdev_priv(dev);
1291 struct tun_file *tfile;
1299 numqueues = READ_ONCE(tun->numqueues);
1302 return -ENXIO; /* Caller will free/return all frames */
1305 tfile = rcu_dereference(tun->tfiles[smp_processor_id() %
1308 spin_lock(&tfile->tx_ring.producer_lock);
1309 for (i = 0; i < n; i++) {
1310 struct xdp_frame *xdp = frames[i];
1311 /* Encode the XDP flag into lowest bit for consumer to differ
1312 * XDP buffer from sk_buff.
1314 void *frame = tun_xdp_to_ptr(xdp);
1316 if (__ptr_ring_produce(&tfile->tx_ring, frame)) {
1317 this_cpu_inc(tun->pcpu_stats->tx_dropped);
1318 xdp_return_frame_rx_napi(xdp);
1322 spin_unlock(&tfile->tx_ring.producer_lock);
1328 static int tun_xdp_tx(struct net_device *dev, struct xdp_buff *xdp)
1330 struct xdp_frame *frame = convert_to_xdp_frame(xdp);
1332 if (unlikely(!frame))
1335 return tun_xdp_xmit(dev, 1, &frame);
1338 static void tun_xdp_flush(struct net_device *dev)
1340 struct tun_struct *tun = netdev_priv(dev);
1341 struct tun_file *tfile;
1346 numqueues = READ_ONCE(tun->numqueues);
1350 tfile = rcu_dereference(tun->tfiles[smp_processor_id() %
1352 /* Notify and wake up reader process */
1353 if (tfile->flags & TUN_FASYNC)
1354 kill_fasync(&tfile->fasync, SIGIO, POLL_IN);
1355 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
1361 static const struct net_device_ops tap_netdev_ops = {
1362 .ndo_uninit = tun_net_uninit,
1363 .ndo_open = tun_net_open,
1364 .ndo_stop = tun_net_close,
1365 .ndo_start_xmit = tun_net_xmit,
1366 .ndo_fix_features = tun_net_fix_features,
1367 .ndo_set_rx_mode = tun_net_mclist,
1368 .ndo_set_mac_address = eth_mac_addr,
1369 .ndo_validate_addr = eth_validate_addr,
1370 .ndo_select_queue = tun_select_queue,
1371 #ifdef CONFIG_NET_POLL_CONTROLLER
1372 .ndo_poll_controller = tun_poll_controller,
1374 .ndo_features_check = passthru_features_check,
1375 .ndo_set_rx_headroom = tun_set_headroom,
1376 .ndo_get_stats64 = tun_net_get_stats64,
1378 .ndo_xdp_xmit = tun_xdp_xmit,
1379 .ndo_xdp_flush = tun_xdp_flush,
1382 static void tun_flow_init(struct tun_struct *tun)
1386 for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++)
1387 INIT_HLIST_HEAD(&tun->flows[i]);
1389 tun->ageing_time = TUN_FLOW_EXPIRE;
1390 timer_setup(&tun->flow_gc_timer, tun_flow_cleanup, 0);
1391 mod_timer(&tun->flow_gc_timer,
1392 round_jiffies_up(jiffies + tun->ageing_time));
1395 static void tun_flow_uninit(struct tun_struct *tun)
1397 del_timer_sync(&tun->flow_gc_timer);
1398 tun_flow_flush(tun);
1402 #define MAX_MTU 65535
1404 /* Initialize net device. */
1405 static void tun_net_init(struct net_device *dev)
1407 struct tun_struct *tun = netdev_priv(dev);
1409 switch (tun->flags & TUN_TYPE_MASK) {
1411 dev->netdev_ops = &tun_netdev_ops;
1413 /* Point-to-Point TUN Device */
1414 dev->hard_header_len = 0;
1418 /* Zero header length */
1419 dev->type = ARPHRD_NONE;
1420 dev->flags = IFF_POINTOPOINT | IFF_NOARP | IFF_MULTICAST;
1424 dev->netdev_ops = &tap_netdev_ops;
1425 /* Ethernet TAP Device */
1427 dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1428 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
1430 eth_hw_addr_random(dev);
1435 dev->min_mtu = MIN_MTU;
1436 dev->max_mtu = MAX_MTU - dev->hard_header_len;
1439 static bool tun_sock_writeable(struct tun_struct *tun, struct tun_file *tfile)
1441 struct sock *sk = tfile->socket.sk;
1443 return (tun->dev->flags & IFF_UP) && sock_writeable(sk);
1446 /* Character device part */
1449 static __poll_t tun_chr_poll(struct file *file, poll_table *wait)
1451 struct tun_file *tfile = file->private_data;
1452 struct tun_struct *tun = tun_get(tfile);
1459 sk = tfile->socket.sk;
1461 tun_debug(KERN_INFO, tun, "tun_chr_poll\n");
1463 poll_wait(file, sk_sleep(sk), wait);
1465 if (!ptr_ring_empty(&tfile->tx_ring))
1466 mask |= EPOLLIN | EPOLLRDNORM;
1468 /* Make sure SOCKWQ_ASYNC_NOSPACE is set if not writable to
1469 * guarantee EPOLLOUT to be raised by either here or
1470 * tun_sock_write_space(). Then process could get notification
1471 * after it writes to a down device and meets -EIO.
1473 if (tun_sock_writeable(tun, tfile) ||
1474 (!test_and_set_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags) &&
1475 tun_sock_writeable(tun, tfile)))
1476 mask |= EPOLLOUT | EPOLLWRNORM;
1478 if (tun->dev->reg_state != NETREG_REGISTERED)
1485 static struct sk_buff *tun_napi_alloc_frags(struct tun_file *tfile,
1487 const struct iov_iter *it)
1489 struct sk_buff *skb;
1494 if (it->nr_segs > MAX_SKB_FRAGS + 1)
1495 return ERR_PTR(-ENOMEM);
1498 skb = napi_get_frags(&tfile->napi);
1501 return ERR_PTR(-ENOMEM);
1503 linear = iov_iter_single_seg_count(it);
1504 err = __skb_grow(skb, linear);
1509 skb->data_len = len - linear;
1510 skb->truesize += skb->data_len;
1512 for (i = 1; i < it->nr_segs; i++) {
1513 struct page_frag *pfrag = ¤t->task_frag;
1514 size_t fragsz = it->iov[i].iov_len;
1516 if (fragsz == 0 || fragsz > PAGE_SIZE) {
1521 if (!skb_page_frag_refill(fragsz, pfrag, GFP_KERNEL)) {
1526 skb_fill_page_desc(skb, i - 1, pfrag->page,
1527 pfrag->offset, fragsz);
1528 page_ref_inc(pfrag->page);
1529 pfrag->offset += fragsz;
1534 /* frees skb and all frags allocated with napi_alloc_frag() */
1535 napi_free_frags(&tfile->napi);
1536 return ERR_PTR(err);
1539 /* prepad is the amount to reserve at front. len is length after that.
1540 * linear is a hint as to how much to copy (usually headers). */
1541 static struct sk_buff *tun_alloc_skb(struct tun_file *tfile,
1542 size_t prepad, size_t len,
1543 size_t linear, int noblock)
1545 struct sock *sk = tfile->socket.sk;
1546 struct sk_buff *skb;
1549 /* Under a page? Don't bother with paged skb. */
1550 if (prepad + len < PAGE_SIZE || !linear)
1553 skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
1556 return ERR_PTR(err);
1558 skb_reserve(skb, prepad);
1559 skb_put(skb, linear);
1560 skb->data_len = len - linear;
1561 skb->len += len - linear;
1566 static void tun_rx_batched(struct tun_struct *tun, struct tun_file *tfile,
1567 struct sk_buff *skb, int more)
1569 struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
1570 struct sk_buff_head process_queue;
1571 u32 rx_batched = tun->rx_batched;
1574 if (!rx_batched || (!more && skb_queue_empty(queue))) {
1576 netif_receive_skb(skb);
1581 spin_lock(&queue->lock);
1582 if (!more || skb_queue_len(queue) == rx_batched) {
1583 __skb_queue_head_init(&process_queue);
1584 skb_queue_splice_tail_init(queue, &process_queue);
1587 __skb_queue_tail(queue, skb);
1589 spin_unlock(&queue->lock);
1592 struct sk_buff *nskb;
1595 while ((nskb = __skb_dequeue(&process_queue)))
1596 netif_receive_skb(nskb);
1597 netif_receive_skb(skb);
1602 static bool tun_can_build_skb(struct tun_struct *tun, struct tun_file *tfile,
1603 int len, int noblock, bool zerocopy)
1605 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
1608 if (tfile->socket.sk->sk_sndbuf != INT_MAX)
1617 if (SKB_DATA_ALIGN(len + TUN_RX_PAD) +
1618 SKB_DATA_ALIGN(sizeof(struct skb_shared_info)) > PAGE_SIZE)
1624 static struct sk_buff *tun_build_skb(struct tun_struct *tun,
1625 struct tun_file *tfile,
1626 struct iov_iter *from,
1627 struct virtio_net_hdr *hdr,
1628 int len, int *skb_xdp)
1630 struct page_frag *alloc_frag = ¤t->task_frag;
1631 struct sk_buff *skb;
1632 struct bpf_prog *xdp_prog;
1633 int buflen = SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
1634 unsigned int delta = 0;
1637 int err, pad = TUN_RX_PAD;
1640 xdp_prog = rcu_dereference(tun->xdp_prog);
1642 pad += TUN_HEADROOM;
1643 buflen += SKB_DATA_ALIGN(len + pad);
1646 alloc_frag->offset = ALIGN((u64)alloc_frag->offset, SMP_CACHE_BYTES);
1647 if (unlikely(!skb_page_frag_refill(buflen, alloc_frag, GFP_KERNEL)))
1648 return ERR_PTR(-ENOMEM);
1650 buf = (char *)page_address(alloc_frag->page) + alloc_frag->offset;
1651 copied = copy_page_from_iter(alloc_frag->page,
1652 alloc_frag->offset + pad,
1655 return ERR_PTR(-EFAULT);
1657 /* There's a small window that XDP may be set after the check
1658 * of xdp_prog above, this should be rare and for simplicity
1659 * we do XDP on skb in case the headroom is not enough.
1661 if (hdr->gso_type || !xdp_prog)
1668 xdp_prog = rcu_dereference(tun->xdp_prog);
1669 if (xdp_prog && !*skb_xdp) {
1670 struct xdp_buff xdp;
1674 xdp.data_hard_start = buf;
1675 xdp.data = buf + pad;
1676 xdp_set_data_meta_invalid(&xdp);
1677 xdp.data_end = xdp.data + len;
1678 xdp.rxq = &tfile->xdp_rxq;
1679 orig_data = xdp.data;
1680 act = bpf_prog_run_xdp(xdp_prog, &xdp);
1684 get_page(alloc_frag->page);
1685 alloc_frag->offset += buflen;
1686 err = xdp_do_redirect(tun->dev, &xdp, xdp_prog);
1694 get_page(alloc_frag->page);
1695 alloc_frag->offset += buflen;
1696 if (tun_xdp_tx(tun->dev, &xdp))
1698 tun_xdp_flush(tun->dev);
1703 delta = orig_data - xdp.data;
1704 len = xdp.data_end - xdp.data;
1707 bpf_warn_invalid_xdp_action(act);
1710 trace_xdp_exception(tun->dev, xdp_prog, act);
1717 skb = build_skb(buf, buflen);
1721 return ERR_PTR(-ENOMEM);
1724 skb_reserve(skb, pad - delta);
1726 get_page(alloc_frag->page);
1727 alloc_frag->offset += buflen;
1735 put_page(alloc_frag->page);
1739 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1743 /* Get packet from user space buffer */
1744 static ssize_t tun_get_user(struct tun_struct *tun, struct tun_file *tfile,
1745 void *msg_control, struct iov_iter *from,
1746 int noblock, bool more)
1748 struct tun_pi pi = { 0, cpu_to_be16(ETH_P_IP) };
1749 struct sk_buff *skb;
1750 size_t total_len = iov_iter_count(from);
1751 size_t len = total_len, align = tun->align, linear;
1752 struct virtio_net_hdr gso = { 0 };
1753 struct tun_pcpu_stats *stats;
1756 bool zerocopy = false;
1760 bool frags = tun_napi_frags_enabled(tun);
1762 if (!(tun->dev->flags & IFF_UP))
1765 if (!(tun->flags & IFF_NO_PI)) {
1766 if (len < sizeof(pi))
1770 if (!copy_from_iter_full(&pi, sizeof(pi), from))
1774 if (tun->flags & IFF_VNET_HDR) {
1775 int vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
1777 if (len < vnet_hdr_sz)
1781 if (!copy_from_iter_full(&gso, sizeof(gso), from))
1784 if ((gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
1785 tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2 > tun16_to_cpu(tun, gso.hdr_len))
1786 gso.hdr_len = cpu_to_tun16(tun, tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2);
1788 if (tun16_to_cpu(tun, gso.hdr_len) > len)
1790 iov_iter_advance(from, vnet_hdr_sz - sizeof(gso));
1793 if ((tun->flags & TUN_TYPE_MASK) == IFF_TAP) {
1794 align += NET_IP_ALIGN;
1795 if (unlikely(len < ETH_HLEN ||
1796 (gso.hdr_len && tun16_to_cpu(tun, gso.hdr_len) < ETH_HLEN)))
1800 good_linear = SKB_MAX_HEAD(align);
1803 struct iov_iter i = *from;
1805 /* There are 256 bytes to be copied in skb, so there is
1806 * enough room for skb expand head in case it is used.
1807 * The rest of the buffer is mapped from userspace.
1809 copylen = gso.hdr_len ? tun16_to_cpu(tun, gso.hdr_len) : GOODCOPY_LEN;
1810 if (copylen > good_linear)
1811 copylen = good_linear;
1813 iov_iter_advance(&i, copylen);
1814 if (iov_iter_npages(&i, INT_MAX) <= MAX_SKB_FRAGS)
1818 if (!frags && tun_can_build_skb(tun, tfile, len, noblock, zerocopy)) {
1819 /* For the packet that is not easy to be processed
1820 * (e.g gso or jumbo packet), we will do it at after
1821 * skb was created with generic XDP routine.
1823 skb = tun_build_skb(tun, tfile, from, &gso, len, &skb_xdp);
1825 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1826 return PTR_ERR(skb);
1833 if (tun16_to_cpu(tun, gso.hdr_len) > good_linear)
1834 linear = good_linear;
1836 linear = tun16_to_cpu(tun, gso.hdr_len);
1840 mutex_lock(&tfile->napi_mutex);
1841 skb = tun_napi_alloc_frags(tfile, copylen, from);
1842 /* tun_napi_alloc_frags() enforces a layout for the skb.
1843 * If zerocopy is enabled, then this layout will be
1844 * overwritten by zerocopy_sg_from_iter().
1848 skb = tun_alloc_skb(tfile, align, copylen, linear,
1853 if (PTR_ERR(skb) != -EAGAIN)
1854 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1856 mutex_unlock(&tfile->napi_mutex);
1857 return PTR_ERR(skb);
1861 err = zerocopy_sg_from_iter(skb, from);
1863 err = skb_copy_datagram_from_iter(skb, 0, from, len);
1866 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1869 tfile->napi.skb = NULL;
1870 mutex_unlock(&tfile->napi_mutex);
1877 if (virtio_net_hdr_to_skb(skb, &gso, tun_is_little_endian(tun))) {
1878 this_cpu_inc(tun->pcpu_stats->rx_frame_errors);
1881 tfile->napi.skb = NULL;
1882 mutex_unlock(&tfile->napi_mutex);
1888 switch (tun->flags & TUN_TYPE_MASK) {
1890 if (tun->flags & IFF_NO_PI) {
1891 u8 ip_version = skb->len ? (skb->data[0] >> 4) : 0;
1893 switch (ip_version) {
1895 pi.proto = htons(ETH_P_IP);
1898 pi.proto = htons(ETH_P_IPV6);
1901 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1907 skb_reset_mac_header(skb);
1908 skb->protocol = pi.proto;
1909 skb->dev = tun->dev;
1913 skb->protocol = eth_type_trans(skb, tun->dev);
1917 /* copy skb_ubuf_info for callback when skb has no error */
1919 skb_shinfo(skb)->destructor_arg = msg_control;
1920 skb_shinfo(skb)->tx_flags |= SKBTX_DEV_ZEROCOPY;
1921 skb_shinfo(skb)->tx_flags |= SKBTX_SHARED_FRAG;
1922 } else if (msg_control) {
1923 struct ubuf_info *uarg = msg_control;
1924 uarg->callback(uarg, false);
1927 skb_reset_network_header(skb);
1928 skb_probe_transport_header(skb, 0);
1931 struct bpf_prog *xdp_prog;
1935 xdp_prog = rcu_dereference(tun->xdp_prog);
1937 ret = do_xdp_generic(xdp_prog, skb);
1938 if (ret != XDP_PASS) {
1946 /* Compute the costly rx hash only if needed for flow updates.
1947 * We may get a very small possibility of OOO during switching, not
1948 * worth to optimize.
1950 if (!rcu_access_pointer(tun->steering_prog) && tun->numqueues > 1 &&
1952 rxhash = __skb_get_hash_symmetric(skb);
1955 /* Exercise flow dissector code path. */
1956 u32 headlen = eth_get_headlen(skb->data, skb_headlen(skb));
1958 if (unlikely(headlen > skb_headlen(skb))) {
1959 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1960 napi_free_frags(&tfile->napi);
1961 mutex_unlock(&tfile->napi_mutex);
1967 napi_gro_frags(&tfile->napi);
1969 mutex_unlock(&tfile->napi_mutex);
1970 } else if (tfile->napi_enabled) {
1971 struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
1974 spin_lock_bh(&queue->lock);
1975 __skb_queue_tail(queue, skb);
1976 queue_len = skb_queue_len(queue);
1977 spin_unlock(&queue->lock);
1979 if (!more || queue_len > NAPI_POLL_WEIGHT)
1980 napi_schedule(&tfile->napi);
1983 } else if (!IS_ENABLED(CONFIG_4KSTACKS)) {
1984 tun_rx_batched(tun, tfile, skb, more);
1989 stats = get_cpu_ptr(tun->pcpu_stats);
1990 u64_stats_update_begin(&stats->syncp);
1991 stats->rx_packets++;
1992 stats->rx_bytes += len;
1993 u64_stats_update_end(&stats->syncp);
1997 tun_flow_update(tun, rxhash, tfile);
2002 static ssize_t tun_chr_write_iter(struct kiocb *iocb, struct iov_iter *from)
2004 struct file *file = iocb->ki_filp;
2005 struct tun_file *tfile = file->private_data;
2006 struct tun_struct *tun = tun_get(tfile);
2012 result = tun_get_user(tun, tfile, NULL, from,
2013 file->f_flags & O_NONBLOCK, false);
2019 static ssize_t tun_put_user_xdp(struct tun_struct *tun,
2020 struct tun_file *tfile,
2021 struct xdp_frame *xdp_frame,
2022 struct iov_iter *iter)
2024 int vnet_hdr_sz = 0;
2025 size_t size = xdp_frame->len;
2026 struct tun_pcpu_stats *stats;
2029 if (tun->flags & IFF_VNET_HDR) {
2030 struct virtio_net_hdr gso = { 0 };
2032 vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
2033 if (unlikely(iov_iter_count(iter) < vnet_hdr_sz))
2035 if (unlikely(copy_to_iter(&gso, sizeof(gso), iter) !=
2038 iov_iter_advance(iter, vnet_hdr_sz - sizeof(gso));
2041 ret = copy_to_iter(xdp_frame->data, size, iter) + vnet_hdr_sz;
2043 stats = get_cpu_ptr(tun->pcpu_stats);
2044 u64_stats_update_begin(&stats->syncp);
2045 stats->tx_packets++;
2046 stats->tx_bytes += ret;
2047 u64_stats_update_end(&stats->syncp);
2048 put_cpu_ptr(tun->pcpu_stats);
2053 /* Put packet to the user space buffer */
2054 static ssize_t tun_put_user(struct tun_struct *tun,
2055 struct tun_file *tfile,
2056 struct sk_buff *skb,
2057 struct iov_iter *iter)
2059 struct tun_pi pi = { 0, skb->protocol };
2060 struct tun_pcpu_stats *stats;
2062 int vlan_offset = 0;
2064 int vnet_hdr_sz = 0;
2066 if (skb_vlan_tag_present(skb))
2067 vlan_hlen = VLAN_HLEN;
2069 if (tun->flags & IFF_VNET_HDR)
2070 vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
2072 total = skb->len + vlan_hlen + vnet_hdr_sz;
2074 if (!(tun->flags & IFF_NO_PI)) {
2075 if (iov_iter_count(iter) < sizeof(pi))
2078 total += sizeof(pi);
2079 if (iov_iter_count(iter) < total) {
2080 /* Packet will be striped */
2081 pi.flags |= TUN_PKT_STRIP;
2084 if (copy_to_iter(&pi, sizeof(pi), iter) != sizeof(pi))
2089 struct virtio_net_hdr gso;
2091 if (iov_iter_count(iter) < vnet_hdr_sz)
2094 if (virtio_net_hdr_from_skb(skb, &gso,
2095 tun_is_little_endian(tun), true)) {
2096 struct skb_shared_info *sinfo = skb_shinfo(skb);
2097 pr_err("unexpected GSO type: "
2098 "0x%x, gso_size %d, hdr_len %d\n",
2099 sinfo->gso_type, tun16_to_cpu(tun, gso.gso_size),
2100 tun16_to_cpu(tun, gso.hdr_len));
2101 print_hex_dump(KERN_ERR, "tun: ",
2104 min((int)tun16_to_cpu(tun, gso.hdr_len), 64), true);
2109 if (copy_to_iter(&gso, sizeof(gso), iter) != sizeof(gso))
2112 iov_iter_advance(iter, vnet_hdr_sz - sizeof(gso));
2119 veth.h_vlan_proto = skb->vlan_proto;
2120 veth.h_vlan_TCI = htons(skb_vlan_tag_get(skb));
2122 vlan_offset = offsetof(struct vlan_ethhdr, h_vlan_proto);
2124 ret = skb_copy_datagram_iter(skb, 0, iter, vlan_offset);
2125 if (ret || !iov_iter_count(iter))
2128 ret = copy_to_iter(&veth, sizeof(veth), iter);
2129 if (ret != sizeof(veth) || !iov_iter_count(iter))
2133 skb_copy_datagram_iter(skb, vlan_offset, iter, skb->len - vlan_offset);
2136 /* caller is in process context, */
2137 stats = get_cpu_ptr(tun->pcpu_stats);
2138 u64_stats_update_begin(&stats->syncp);
2139 stats->tx_packets++;
2140 stats->tx_bytes += skb->len + vlan_hlen;
2141 u64_stats_update_end(&stats->syncp);
2142 put_cpu_ptr(tun->pcpu_stats);
2147 static void *tun_ring_recv(struct tun_file *tfile, int noblock, int *err)
2149 DECLARE_WAITQUEUE(wait, current);
2153 ptr = ptr_ring_consume(&tfile->tx_ring);
2161 add_wait_queue(&tfile->wq.wait, &wait);
2162 current->state = TASK_INTERRUPTIBLE;
2165 ptr = ptr_ring_consume(&tfile->tx_ring);
2168 if (signal_pending(current)) {
2169 error = -ERESTARTSYS;
2172 if (tfile->socket.sk->sk_shutdown & RCV_SHUTDOWN) {
2180 current->state = TASK_RUNNING;
2181 remove_wait_queue(&tfile->wq.wait, &wait);
2188 static ssize_t tun_do_read(struct tun_struct *tun, struct tun_file *tfile,
2189 struct iov_iter *to,
2190 int noblock, void *ptr)
2195 tun_debug(KERN_INFO, tun, "tun_do_read\n");
2197 if (!iov_iter_count(to)) {
2203 /* Read frames from ring */
2204 ptr = tun_ring_recv(tfile, noblock, &err);
2209 if (tun_is_xdp_frame(ptr)) {
2210 struct xdp_frame *xdpf = tun_ptr_to_xdp(ptr);
2212 ret = tun_put_user_xdp(tun, tfile, xdpf, to);
2213 xdp_return_frame(xdpf);
2215 struct sk_buff *skb = ptr;
2217 ret = tun_put_user(tun, tfile, skb, to);
2218 if (unlikely(ret < 0))
2227 static ssize_t tun_chr_read_iter(struct kiocb *iocb, struct iov_iter *to)
2229 struct file *file = iocb->ki_filp;
2230 struct tun_file *tfile = file->private_data;
2231 struct tun_struct *tun = tun_get(tfile);
2232 ssize_t len = iov_iter_count(to), ret;
2236 ret = tun_do_read(tun, tfile, to, file->f_flags & O_NONBLOCK, NULL);
2237 ret = min_t(ssize_t, ret, len);
2244 static void tun_prog_free(struct rcu_head *rcu)
2246 struct tun_prog *prog = container_of(rcu, struct tun_prog, rcu);
2248 bpf_prog_destroy(prog->prog);
2252 static int __tun_set_ebpf(struct tun_struct *tun,
2253 struct tun_prog __rcu **prog_p,
2254 struct bpf_prog *prog)
2256 struct tun_prog *old, *new = NULL;
2259 new = kmalloc(sizeof(*new), GFP_KERNEL);
2265 spin_lock_bh(&tun->lock);
2266 old = rcu_dereference_protected(*prog_p,
2267 lockdep_is_held(&tun->lock));
2268 rcu_assign_pointer(*prog_p, new);
2269 spin_unlock_bh(&tun->lock);
2272 call_rcu(&old->rcu, tun_prog_free);
2277 static void tun_free_netdev(struct net_device *dev)
2279 struct tun_struct *tun = netdev_priv(dev);
2281 BUG_ON(!(list_empty(&tun->disabled)));
2282 free_percpu(tun->pcpu_stats);
2283 tun_flow_uninit(tun);
2284 security_tun_dev_free_security(tun->security);
2285 __tun_set_ebpf(tun, &tun->steering_prog, NULL);
2286 __tun_set_ebpf(tun, &tun->filter_prog, NULL);
2289 static void tun_setup(struct net_device *dev)
2291 struct tun_struct *tun = netdev_priv(dev);
2293 tun->owner = INVALID_UID;
2294 tun->group = INVALID_GID;
2296 dev->ethtool_ops = &tun_ethtool_ops;
2297 dev->needs_free_netdev = true;
2298 dev->priv_destructor = tun_free_netdev;
2299 /* We prefer our own queue length */
2300 dev->tx_queue_len = TUN_READQ_SIZE;
2303 /* Trivial set of netlink ops to allow deleting tun or tap
2304 * device with netlink.
2306 static int tun_validate(struct nlattr *tb[], struct nlattr *data[],
2307 struct netlink_ext_ack *extack)
2312 static size_t tun_get_size(const struct net_device *dev)
2314 BUILD_BUG_ON(sizeof(u32) != sizeof(uid_t));
2315 BUILD_BUG_ON(sizeof(u32) != sizeof(gid_t));
2317 return nla_total_size(sizeof(uid_t)) + /* OWNER */
2318 nla_total_size(sizeof(gid_t)) + /* GROUP */
2319 nla_total_size(sizeof(u8)) + /* TYPE */
2320 nla_total_size(sizeof(u8)) + /* PI */
2321 nla_total_size(sizeof(u8)) + /* VNET_HDR */
2322 nla_total_size(sizeof(u8)) + /* PERSIST */
2323 nla_total_size(sizeof(u8)) + /* MULTI_QUEUE */
2324 nla_total_size(sizeof(u32)) + /* NUM_QUEUES */
2325 nla_total_size(sizeof(u32)) + /* NUM_DISABLED_QUEUES */
2329 static int tun_fill_info(struct sk_buff *skb, const struct net_device *dev)
2331 struct tun_struct *tun = netdev_priv(dev);
2333 if (nla_put_u8(skb, IFLA_TUN_TYPE, tun->flags & TUN_TYPE_MASK))
2334 goto nla_put_failure;
2335 if (uid_valid(tun->owner) &&
2336 nla_put_u32(skb, IFLA_TUN_OWNER,
2337 from_kuid_munged(current_user_ns(), tun->owner)))
2338 goto nla_put_failure;
2339 if (gid_valid(tun->group) &&
2340 nla_put_u32(skb, IFLA_TUN_GROUP,
2341 from_kgid_munged(current_user_ns(), tun->group)))
2342 goto nla_put_failure;
2343 if (nla_put_u8(skb, IFLA_TUN_PI, !(tun->flags & IFF_NO_PI)))
2344 goto nla_put_failure;
2345 if (nla_put_u8(skb, IFLA_TUN_VNET_HDR, !!(tun->flags & IFF_VNET_HDR)))
2346 goto nla_put_failure;
2347 if (nla_put_u8(skb, IFLA_TUN_PERSIST, !!(tun->flags & IFF_PERSIST)))
2348 goto nla_put_failure;
2349 if (nla_put_u8(skb, IFLA_TUN_MULTI_QUEUE,
2350 !!(tun->flags & IFF_MULTI_QUEUE)))
2351 goto nla_put_failure;
2352 if (tun->flags & IFF_MULTI_QUEUE) {
2353 if (nla_put_u32(skb, IFLA_TUN_NUM_QUEUES, tun->numqueues))
2354 goto nla_put_failure;
2355 if (nla_put_u32(skb, IFLA_TUN_NUM_DISABLED_QUEUES,
2357 goto nla_put_failure;
2366 static struct rtnl_link_ops tun_link_ops __read_mostly = {
2368 .priv_size = sizeof(struct tun_struct),
2370 .validate = tun_validate,
2371 .get_size = tun_get_size,
2372 .fill_info = tun_fill_info,
2375 static void tun_sock_write_space(struct sock *sk)
2377 struct tun_file *tfile;
2378 wait_queue_head_t *wqueue;
2380 if (!sock_writeable(sk))
2383 if (!test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags))
2386 wqueue = sk_sleep(sk);
2387 if (wqueue && waitqueue_active(wqueue))
2388 wake_up_interruptible_sync_poll(wqueue, EPOLLOUT |
2389 EPOLLWRNORM | EPOLLWRBAND);
2391 tfile = container_of(sk, struct tun_file, sk);
2392 kill_fasync(&tfile->fasync, SIGIO, POLL_OUT);
2395 static int tun_sendmsg(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);
2404 ret = tun_get_user(tun, tfile, m->msg_control, &m->msg_iter,
2405 m->msg_flags & MSG_DONTWAIT,
2406 m->msg_flags & MSG_MORE);
2411 static int tun_recvmsg(struct socket *sock, struct msghdr *m, size_t total_len,
2414 struct tun_file *tfile = container_of(sock, struct tun_file, socket);
2415 struct tun_struct *tun = tun_get(tfile);
2416 void *ptr = m->msg_control;
2424 if (flags & ~(MSG_DONTWAIT|MSG_TRUNC|MSG_ERRQUEUE)) {
2428 if (flags & MSG_ERRQUEUE) {
2429 ret = sock_recv_errqueue(sock->sk, m, total_len,
2430 SOL_PACKET, TUN_TX_TIMESTAMP);
2433 ret = tun_do_read(tun, tfile, &m->msg_iter, flags & MSG_DONTWAIT, ptr);
2434 if (ret > (ssize_t)total_len) {
2435 m->msg_flags |= MSG_TRUNC;
2436 ret = flags & MSG_TRUNC ? ret : total_len;
2449 static int tun_ptr_peek_len(void *ptr)
2452 if (tun_is_xdp_frame(ptr)) {
2453 struct xdp_frame *xdpf = tun_ptr_to_xdp(ptr);
2457 return __skb_array_len_with_tag(ptr);
2463 static int tun_peek_len(struct socket *sock)
2465 struct tun_file *tfile = container_of(sock, struct tun_file, socket);
2466 struct tun_struct *tun;
2469 tun = tun_get(tfile);
2473 ret = PTR_RING_PEEK_CALL(&tfile->tx_ring, tun_ptr_peek_len);
2479 /* Ops structure to mimic raw sockets with tun */
2480 static const struct proto_ops tun_socket_ops = {
2481 .peek_len = tun_peek_len,
2482 .sendmsg = tun_sendmsg,
2483 .recvmsg = tun_recvmsg,
2486 static struct proto tun_proto = {
2488 .owner = THIS_MODULE,
2489 .obj_size = sizeof(struct tun_file),
2492 static int tun_flags(struct tun_struct *tun)
2494 return tun->flags & (TUN_FEATURES | IFF_PERSIST | IFF_TUN | IFF_TAP);
2497 static ssize_t tun_show_flags(struct device *dev, struct device_attribute *attr,
2500 struct tun_struct *tun = netdev_priv(to_net_dev(dev));
2501 return sprintf(buf, "0x%x\n", tun_flags(tun));
2504 static ssize_t tun_show_owner(struct device *dev, struct device_attribute *attr,
2507 struct tun_struct *tun = netdev_priv(to_net_dev(dev));
2508 return uid_valid(tun->owner)?
2509 sprintf(buf, "%u\n",
2510 from_kuid_munged(current_user_ns(), tun->owner)):
2511 sprintf(buf, "-1\n");
2514 static ssize_t tun_show_group(struct device *dev, struct device_attribute *attr,
2517 struct tun_struct *tun = netdev_priv(to_net_dev(dev));
2518 return gid_valid(tun->group) ?
2519 sprintf(buf, "%u\n",
2520 from_kgid_munged(current_user_ns(), tun->group)):
2521 sprintf(buf, "-1\n");
2524 static DEVICE_ATTR(tun_flags, 0444, tun_show_flags, NULL);
2525 static DEVICE_ATTR(owner, 0444, tun_show_owner, NULL);
2526 static DEVICE_ATTR(group, 0444, tun_show_group, NULL);
2528 static struct attribute *tun_dev_attrs[] = {
2529 &dev_attr_tun_flags.attr,
2530 &dev_attr_owner.attr,
2531 &dev_attr_group.attr,
2535 static const struct attribute_group tun_attr_group = {
2536 .attrs = tun_dev_attrs
2539 static int tun_set_iff(struct net *net, struct file *file, struct ifreq *ifr)
2541 struct tun_struct *tun;
2542 struct tun_file *tfile = file->private_data;
2543 struct net_device *dev;
2546 if (tfile->detached)
2549 if ((ifr->ifr_flags & IFF_NAPI_FRAGS)) {
2550 if (!capable(CAP_NET_ADMIN))
2553 if (!(ifr->ifr_flags & IFF_NAPI) ||
2554 (ifr->ifr_flags & TUN_TYPE_MASK) != IFF_TAP)
2558 dev = __dev_get_by_name(net, ifr->ifr_name);
2560 if (ifr->ifr_flags & IFF_TUN_EXCL)
2562 if ((ifr->ifr_flags & IFF_TUN) && dev->netdev_ops == &tun_netdev_ops)
2563 tun = netdev_priv(dev);
2564 else if ((ifr->ifr_flags & IFF_TAP) && dev->netdev_ops == &tap_netdev_ops)
2565 tun = netdev_priv(dev);
2569 if (!!(ifr->ifr_flags & IFF_MULTI_QUEUE) !=
2570 !!(tun->flags & IFF_MULTI_QUEUE))
2573 if (tun_not_capable(tun))
2575 err = security_tun_dev_open(tun->security);
2579 err = tun_attach(tun, file, ifr->ifr_flags & IFF_NOFILTER,
2580 ifr->ifr_flags & IFF_NAPI);
2584 if (tun->flags & IFF_MULTI_QUEUE &&
2585 (tun->numqueues + tun->numdisabled > 1)) {
2586 /* One or more queue has already been attached, no need
2587 * to initialize the device again.
2589 netdev_state_change(dev);
2593 tun->flags = (tun->flags & ~TUN_FEATURES) |
2594 (ifr->ifr_flags & TUN_FEATURES);
2596 netdev_state_change(dev);
2599 unsigned long flags = 0;
2600 int queues = ifr->ifr_flags & IFF_MULTI_QUEUE ?
2603 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
2605 err = security_tun_dev_create();
2610 if (ifr->ifr_flags & IFF_TUN) {
2614 } else if (ifr->ifr_flags & IFF_TAP) {
2622 name = ifr->ifr_name;
2624 dev = alloc_netdev_mqs(sizeof(struct tun_struct), name,
2625 NET_NAME_UNKNOWN, tun_setup, queues,
2630 err = dev_get_valid_name(net, dev, name);
2634 dev_net_set(dev, net);
2635 dev->rtnl_link_ops = &tun_link_ops;
2636 dev->ifindex = tfile->ifindex;
2637 dev->sysfs_groups[0] = &tun_attr_group;
2639 tun = netdev_priv(dev);
2642 tun->txflt.count = 0;
2643 tun->vnet_hdr_sz = sizeof(struct virtio_net_hdr);
2645 tun->align = NET_SKB_PAD;
2646 tun->filter_attached = false;
2647 tun->sndbuf = tfile->socket.sk->sk_sndbuf;
2648 tun->rx_batched = 0;
2649 RCU_INIT_POINTER(tun->steering_prog, NULL);
2651 tun->pcpu_stats = netdev_alloc_pcpu_stats(struct tun_pcpu_stats);
2652 if (!tun->pcpu_stats) {
2657 spin_lock_init(&tun->lock);
2659 err = security_tun_dev_alloc_security(&tun->security);
2666 dev->hw_features = NETIF_F_SG | NETIF_F_FRAGLIST |
2667 TUN_USER_FEATURES | NETIF_F_HW_VLAN_CTAG_TX |
2668 NETIF_F_HW_VLAN_STAG_TX;
2669 dev->features = dev->hw_features | NETIF_F_LLTX;
2670 dev->vlan_features = dev->features &
2671 ~(NETIF_F_HW_VLAN_CTAG_TX |
2672 NETIF_F_HW_VLAN_STAG_TX);
2674 tun->flags = (tun->flags & ~TUN_FEATURES) |
2675 (ifr->ifr_flags & TUN_FEATURES);
2677 INIT_LIST_HEAD(&tun->disabled);
2678 err = tun_attach(tun, file, false, ifr->ifr_flags & IFF_NAPI);
2682 err = register_netdevice(tun->dev);
2687 netif_carrier_on(tun->dev);
2689 tun_debug(KERN_INFO, tun, "tun_set_iff\n");
2691 /* Make sure persistent devices do not get stuck in
2694 if (netif_running(tun->dev))
2695 netif_tx_wake_all_queues(tun->dev);
2697 strcpy(ifr->ifr_name, tun->dev->name);
2701 tun_detach_all(dev);
2702 /* register_netdevice() already called tun_free_netdev() */
2706 tun_flow_uninit(tun);
2707 security_tun_dev_free_security(tun->security);
2709 free_percpu(tun->pcpu_stats);
2715 static void tun_get_iff(struct net *net, struct tun_struct *tun,
2718 tun_debug(KERN_INFO, tun, "tun_get_iff\n");
2720 strcpy(ifr->ifr_name, tun->dev->name);
2722 ifr->ifr_flags = tun_flags(tun);
2726 /* This is like a cut-down ethtool ops, except done via tun fd so no
2727 * privs required. */
2728 static int set_offload(struct tun_struct *tun, unsigned long arg)
2730 netdev_features_t features = 0;
2732 if (arg & TUN_F_CSUM) {
2733 features |= NETIF_F_HW_CSUM;
2736 if (arg & (TUN_F_TSO4|TUN_F_TSO6)) {
2737 if (arg & TUN_F_TSO_ECN) {
2738 features |= NETIF_F_TSO_ECN;
2739 arg &= ~TUN_F_TSO_ECN;
2741 if (arg & TUN_F_TSO4)
2742 features |= NETIF_F_TSO;
2743 if (arg & TUN_F_TSO6)
2744 features |= NETIF_F_TSO6;
2745 arg &= ~(TUN_F_TSO4|TUN_F_TSO6);
2751 /* This gives the user a way to test for new features in future by
2752 * trying to set them. */
2756 tun->set_features = features;
2757 tun->dev->wanted_features &= ~TUN_USER_FEATURES;
2758 tun->dev->wanted_features |= features;
2759 netdev_update_features(tun->dev);
2764 static void tun_detach_filter(struct tun_struct *tun, int n)
2767 struct tun_file *tfile;
2769 for (i = 0; i < n; i++) {
2770 tfile = rtnl_dereference(tun->tfiles[i]);
2771 lock_sock(tfile->socket.sk);
2772 sk_detach_filter(tfile->socket.sk);
2773 release_sock(tfile->socket.sk);
2776 tun->filter_attached = false;
2779 static int tun_attach_filter(struct tun_struct *tun)
2782 struct tun_file *tfile;
2784 for (i = 0; i < tun->numqueues; i++) {
2785 tfile = rtnl_dereference(tun->tfiles[i]);
2786 lock_sock(tfile->socket.sk);
2787 ret = sk_attach_filter(&tun->fprog, tfile->socket.sk);
2788 release_sock(tfile->socket.sk);
2790 tun_detach_filter(tun, i);
2795 tun->filter_attached = true;
2799 static void tun_set_sndbuf(struct tun_struct *tun)
2801 struct tun_file *tfile;
2804 for (i = 0; i < tun->numqueues; i++) {
2805 tfile = rtnl_dereference(tun->tfiles[i]);
2806 tfile->socket.sk->sk_sndbuf = tun->sndbuf;
2810 static int tun_set_queue(struct file *file, struct ifreq *ifr)
2812 struct tun_file *tfile = file->private_data;
2813 struct tun_struct *tun;
2818 if (ifr->ifr_flags & IFF_ATTACH_QUEUE) {
2819 tun = tfile->detached;
2824 ret = security_tun_dev_attach_queue(tun->security);
2827 ret = tun_attach(tun, file, false, tun->flags & IFF_NAPI);
2828 } else if (ifr->ifr_flags & IFF_DETACH_QUEUE) {
2829 tun = rtnl_dereference(tfile->tun);
2830 if (!tun || !(tun->flags & IFF_MULTI_QUEUE) || tfile->detached)
2833 __tun_detach(tfile, false);
2838 netdev_state_change(tun->dev);
2845 static int tun_set_ebpf(struct tun_struct *tun, struct tun_prog **prog_p,
2848 struct bpf_prog *prog;
2851 if (copy_from_user(&fd, data, sizeof(fd)))
2857 prog = bpf_prog_get_type(fd, BPF_PROG_TYPE_SOCKET_FILTER);
2859 return PTR_ERR(prog);
2862 return __tun_set_ebpf(tun, prog_p, prog);
2865 static long __tun_chr_ioctl(struct file *file, unsigned int cmd,
2866 unsigned long arg, int ifreq_len)
2868 struct tun_file *tfile = file->private_data;
2869 struct net *net = sock_net(&tfile->sk);
2870 struct tun_struct *tun;
2871 void __user* argp = (void __user*)arg;
2877 unsigned int ifindex;
2880 bool do_notify = false;
2882 if (cmd == TUNSETIFF || cmd == TUNSETQUEUE ||
2883 (_IOC_TYPE(cmd) == SOCK_IOC_TYPE && cmd != SIOCGSKNS)) {
2884 if (copy_from_user(&ifr, argp, ifreq_len))
2887 memset(&ifr, 0, sizeof(ifr));
2889 if (cmd == TUNGETFEATURES) {
2890 /* Currently this just means: "what IFF flags are valid?".
2891 * This is needed because we never checked for invalid flags on
2894 return put_user(IFF_TUN | IFF_TAP | TUN_FEATURES,
2895 (unsigned int __user*)argp);
2896 } else if (cmd == TUNSETQUEUE) {
2897 return tun_set_queue(file, &ifr);
2898 } else if (cmd == SIOCGSKNS) {
2899 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
2901 return open_related_ns(&net->ns, get_net_ns);
2907 tun = tun_get(tfile);
2908 if (cmd == TUNSETIFF) {
2913 ifr.ifr_name[IFNAMSIZ-1] = '\0';
2915 ret = tun_set_iff(net, file, &ifr);
2920 if (copy_to_user(argp, &ifr, ifreq_len))
2924 if (cmd == TUNSETIFINDEX) {
2930 if (copy_from_user(&ifindex, argp, sizeof(ifindex)))
2934 tfile->ifindex = ifindex;
2942 tun_debug(KERN_INFO, tun, "tun_chr_ioctl cmd %u\n", cmd);
2947 tun_get_iff(current->nsproxy->net_ns, tun, &ifr);
2949 if (tfile->detached)
2950 ifr.ifr_flags |= IFF_DETACH_QUEUE;
2951 if (!tfile->socket.sk->sk_filter)
2952 ifr.ifr_flags |= IFF_NOFILTER;
2954 if (copy_to_user(argp, &ifr, ifreq_len))
2959 /* Disable/Enable checksum */
2961 /* [unimplemented] */
2962 tun_debug(KERN_INFO, tun, "ignored: set checksum %s\n",
2963 arg ? "disabled" : "enabled");
2967 /* Disable/Enable persist mode. Keep an extra reference to the
2968 * module to prevent the module being unprobed.
2970 if (arg && !(tun->flags & IFF_PERSIST)) {
2971 tun->flags |= IFF_PERSIST;
2972 __module_get(THIS_MODULE);
2975 if (!arg && (tun->flags & IFF_PERSIST)) {
2976 tun->flags &= ~IFF_PERSIST;
2977 module_put(THIS_MODULE);
2981 tun_debug(KERN_INFO, tun, "persist %s\n",
2982 arg ? "enabled" : "disabled");
2986 /* Set owner of the device */
2987 owner = make_kuid(current_user_ns(), arg);
2988 if (!uid_valid(owner)) {
2994 tun_debug(KERN_INFO, tun, "owner set to %u\n",
2995 from_kuid(&init_user_ns, tun->owner));
2999 /* Set group of the device */
3000 group = make_kgid(current_user_ns(), arg);
3001 if (!gid_valid(group)) {
3007 tun_debug(KERN_INFO, tun, "group set to %u\n",
3008 from_kgid(&init_user_ns, tun->group));
3012 /* Only allow setting the type when the interface is down */
3013 if (tun->dev->flags & IFF_UP) {
3014 tun_debug(KERN_INFO, tun,
3015 "Linktype set failed because interface is up\n");
3018 tun->dev->type = (int) arg;
3019 tun_debug(KERN_INFO, tun, "linktype set to %d\n",
3031 ret = set_offload(tun, arg);
3034 case TUNSETTXFILTER:
3035 /* Can be set only for TAPs */
3037 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3039 ret = update_filter(&tun->txflt, (void __user *)arg);
3043 /* Get hw address */
3044 memcpy(ifr.ifr_hwaddr.sa_data, tun->dev->dev_addr, ETH_ALEN);
3045 ifr.ifr_hwaddr.sa_family = tun->dev->type;
3046 if (copy_to_user(argp, &ifr, ifreq_len))
3051 /* Set hw address */
3052 tun_debug(KERN_DEBUG, tun, "set hw address: %pM\n",
3053 ifr.ifr_hwaddr.sa_data);
3055 ret = dev_set_mac_address(tun->dev, &ifr.ifr_hwaddr);
3059 sndbuf = tfile->socket.sk->sk_sndbuf;
3060 if (copy_to_user(argp, &sndbuf, sizeof(sndbuf)))
3065 if (copy_from_user(&sndbuf, argp, sizeof(sndbuf))) {
3074 tun->sndbuf = sndbuf;
3075 tun_set_sndbuf(tun);
3078 case TUNGETVNETHDRSZ:
3079 vnet_hdr_sz = tun->vnet_hdr_sz;
3080 if (copy_to_user(argp, &vnet_hdr_sz, sizeof(vnet_hdr_sz)))
3084 case TUNSETVNETHDRSZ:
3085 if (copy_from_user(&vnet_hdr_sz, argp, sizeof(vnet_hdr_sz))) {
3089 if (vnet_hdr_sz < (int)sizeof(struct virtio_net_hdr)) {
3094 tun->vnet_hdr_sz = vnet_hdr_sz;
3098 le = !!(tun->flags & TUN_VNET_LE);
3099 if (put_user(le, (int __user *)argp))
3104 if (get_user(le, (int __user *)argp)) {
3109 tun->flags |= TUN_VNET_LE;
3111 tun->flags &= ~TUN_VNET_LE;
3115 ret = tun_get_vnet_be(tun, argp);
3119 ret = tun_set_vnet_be(tun, argp);
3122 case TUNATTACHFILTER:
3123 /* Can be set only for TAPs */
3125 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3128 if (copy_from_user(&tun->fprog, argp, sizeof(tun->fprog)))
3131 ret = tun_attach_filter(tun);
3134 case TUNDETACHFILTER:
3135 /* Can be set only for TAPs */
3137 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3140 tun_detach_filter(tun, tun->numqueues);
3145 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3148 if (copy_to_user(argp, &tun->fprog, sizeof(tun->fprog)))
3153 case TUNSETSTEERINGEBPF:
3154 ret = tun_set_ebpf(tun, &tun->steering_prog, argp);
3157 case TUNSETFILTEREBPF:
3158 ret = tun_set_ebpf(tun, &tun->filter_prog, argp);
3167 netdev_state_change(tun->dev);
3176 static long tun_chr_ioctl(struct file *file,
3177 unsigned int cmd, unsigned long arg)
3179 return __tun_chr_ioctl(file, cmd, arg, sizeof (struct ifreq));
3182 #ifdef CONFIG_COMPAT
3183 static long tun_chr_compat_ioctl(struct file *file,
3184 unsigned int cmd, unsigned long arg)
3189 case TUNSETTXFILTER:
3194 arg = (unsigned long)compat_ptr(arg);
3197 arg = (compat_ulong_t)arg;
3202 * compat_ifreq is shorter than ifreq, so we must not access beyond
3203 * the end of that structure. All fields that are used in this
3204 * driver are compatible though, we don't need to convert the
3207 return __tun_chr_ioctl(file, cmd, arg, sizeof(struct compat_ifreq));
3209 #endif /* CONFIG_COMPAT */
3211 static int tun_chr_fasync(int fd, struct file *file, int on)
3213 struct tun_file *tfile = file->private_data;
3216 if ((ret = fasync_helper(fd, file, on, &tfile->fasync)) < 0)
3220 __f_setown(file, task_pid(current), PIDTYPE_PID, 0);
3221 tfile->flags |= TUN_FASYNC;
3223 tfile->flags &= ~TUN_FASYNC;
3229 static int tun_chr_open(struct inode *inode, struct file * file)
3231 struct net *net = current->nsproxy->net_ns;
3232 struct tun_file *tfile;
3234 DBG1(KERN_INFO, "tunX: tun_chr_open\n");
3236 tfile = (struct tun_file *)sk_alloc(net, AF_UNSPEC, GFP_KERNEL,
3240 if (ptr_ring_init(&tfile->tx_ring, 0, GFP_KERNEL)) {
3241 sk_free(&tfile->sk);
3245 RCU_INIT_POINTER(tfile->tun, NULL);
3249 init_waitqueue_head(&tfile->wq.wait);
3250 RCU_INIT_POINTER(tfile->socket.wq, &tfile->wq);
3252 tfile->socket.file = file;
3253 tfile->socket.ops = &tun_socket_ops;
3255 sock_init_data(&tfile->socket, &tfile->sk);
3257 tfile->sk.sk_write_space = tun_sock_write_space;
3258 tfile->sk.sk_sndbuf = INT_MAX;
3260 file->private_data = tfile;
3261 INIT_LIST_HEAD(&tfile->next);
3263 sock_set_flag(&tfile->sk, SOCK_ZEROCOPY);
3268 static int tun_chr_close(struct inode *inode, struct file *file)
3270 struct tun_file *tfile = file->private_data;
3272 tun_detach(tfile, true);
3277 #ifdef CONFIG_PROC_FS
3278 static void tun_chr_show_fdinfo(struct seq_file *m, struct file *file)
3280 struct tun_file *tfile = file->private_data;
3281 struct tun_struct *tun;
3284 memset(&ifr, 0, sizeof(ifr));
3287 tun = tun_get(tfile);
3289 tun_get_iff(current->nsproxy->net_ns, tun, &ifr);
3295 seq_printf(m, "iff:\t%s\n", ifr.ifr_name);
3299 static const struct file_operations tun_fops = {
3300 .owner = THIS_MODULE,
3301 .llseek = no_llseek,
3302 .read_iter = tun_chr_read_iter,
3303 .write_iter = tun_chr_write_iter,
3304 .poll = tun_chr_poll,
3305 .unlocked_ioctl = tun_chr_ioctl,
3306 #ifdef CONFIG_COMPAT
3307 .compat_ioctl = tun_chr_compat_ioctl,
3309 .open = tun_chr_open,
3310 .release = tun_chr_close,
3311 .fasync = tun_chr_fasync,
3312 #ifdef CONFIG_PROC_FS
3313 .show_fdinfo = tun_chr_show_fdinfo,
3317 static struct miscdevice tun_miscdev = {
3320 .nodename = "net/tun",
3324 /* ethtool interface */
3326 static int tun_get_link_ksettings(struct net_device *dev,
3327 struct ethtool_link_ksettings *cmd)
3329 ethtool_link_ksettings_zero_link_mode(cmd, supported);
3330 ethtool_link_ksettings_zero_link_mode(cmd, advertising);
3331 cmd->base.speed = SPEED_10;
3332 cmd->base.duplex = DUPLEX_FULL;
3333 cmd->base.port = PORT_TP;
3334 cmd->base.phy_address = 0;
3335 cmd->base.autoneg = AUTONEG_DISABLE;
3339 static void tun_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
3341 struct tun_struct *tun = netdev_priv(dev);
3343 strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
3344 strlcpy(info->version, DRV_VERSION, sizeof(info->version));
3346 switch (tun->flags & TUN_TYPE_MASK) {
3348 strlcpy(info->bus_info, "tun", sizeof(info->bus_info));
3351 strlcpy(info->bus_info, "tap", sizeof(info->bus_info));
3356 static u32 tun_get_msglevel(struct net_device *dev)
3359 struct tun_struct *tun = netdev_priv(dev);
3366 static void tun_set_msglevel(struct net_device *dev, u32 value)
3369 struct tun_struct *tun = netdev_priv(dev);
3374 static int tun_get_coalesce(struct net_device *dev,
3375 struct ethtool_coalesce *ec)
3377 struct tun_struct *tun = netdev_priv(dev);
3379 ec->rx_max_coalesced_frames = tun->rx_batched;
3384 static int tun_set_coalesce(struct net_device *dev,
3385 struct ethtool_coalesce *ec)
3387 struct tun_struct *tun = netdev_priv(dev);
3389 if (ec->rx_max_coalesced_frames > NAPI_POLL_WEIGHT)
3390 tun->rx_batched = NAPI_POLL_WEIGHT;
3392 tun->rx_batched = ec->rx_max_coalesced_frames;
3397 static const struct ethtool_ops tun_ethtool_ops = {
3398 .get_drvinfo = tun_get_drvinfo,
3399 .get_msglevel = tun_get_msglevel,
3400 .set_msglevel = tun_set_msglevel,
3401 .get_link = ethtool_op_get_link,
3402 .get_ts_info = ethtool_op_get_ts_info,
3403 .get_coalesce = tun_get_coalesce,
3404 .set_coalesce = tun_set_coalesce,
3405 .get_link_ksettings = tun_get_link_ksettings,
3408 static int tun_queue_resize(struct tun_struct *tun)
3410 struct net_device *dev = tun->dev;
3411 struct tun_file *tfile;
3412 struct ptr_ring **rings;
3413 int n = tun->numqueues + tun->numdisabled;
3416 rings = kmalloc_array(n, sizeof(*rings), GFP_KERNEL);
3420 for (i = 0; i < tun->numqueues; i++) {
3421 tfile = rtnl_dereference(tun->tfiles[i]);
3422 rings[i] = &tfile->tx_ring;
3424 list_for_each_entry(tfile, &tun->disabled, next)
3425 rings[i++] = &tfile->tx_ring;
3427 ret = ptr_ring_resize_multiple(rings, n,
3428 dev->tx_queue_len, GFP_KERNEL,
3435 static int tun_device_event(struct notifier_block *unused,
3436 unsigned long event, void *ptr)
3438 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
3439 struct tun_struct *tun = netdev_priv(dev);
3441 if (dev->rtnl_link_ops != &tun_link_ops)
3445 case NETDEV_CHANGE_TX_QUEUE_LEN:
3446 if (tun_queue_resize(tun))
3456 static struct notifier_block tun_notifier_block __read_mostly = {
3457 .notifier_call = tun_device_event,
3460 static int __init tun_init(void)
3464 pr_info("%s, %s\n", DRV_DESCRIPTION, DRV_VERSION);
3466 ret = rtnl_link_register(&tun_link_ops);
3468 pr_err("Can't register link_ops\n");
3472 ret = misc_register(&tun_miscdev);
3474 pr_err("Can't register misc device %d\n", TUN_MINOR);
3478 ret = register_netdevice_notifier(&tun_notifier_block);
3480 pr_err("Can't register netdevice notifier\n");
3487 misc_deregister(&tun_miscdev);
3489 rtnl_link_unregister(&tun_link_ops);
3494 static void tun_cleanup(void)
3496 misc_deregister(&tun_miscdev);
3497 rtnl_link_unregister(&tun_link_ops);
3498 unregister_netdevice_notifier(&tun_notifier_block);
3501 /* Get an underlying socket object from tun file. Returns error unless file is
3502 * attached to a device. The returned object works like a packet socket, it
3503 * can be used for sock_sendmsg/sock_recvmsg. The caller is responsible for
3504 * holding a reference to the file for as long as the socket is in use. */
3505 struct socket *tun_get_socket(struct file *file)
3507 struct tun_file *tfile;
3508 if (file->f_op != &tun_fops)
3509 return ERR_PTR(-EINVAL);
3510 tfile = file->private_data;
3512 return ERR_PTR(-EBADFD);
3513 return &tfile->socket;
3515 EXPORT_SYMBOL_GPL(tun_get_socket);
3517 struct ptr_ring *tun_get_tx_ring(struct file *file)
3519 struct tun_file *tfile;
3521 if (file->f_op != &tun_fops)
3522 return ERR_PTR(-EINVAL);
3523 tfile = file->private_data;
3525 return ERR_PTR(-EBADFD);
3526 return &tfile->tx_ring;
3528 EXPORT_SYMBOL_GPL(tun_get_tx_ring);
3530 module_init(tun_init);
3531 module_exit(tun_cleanup);
3532 MODULE_DESCRIPTION(DRV_DESCRIPTION);
3533 MODULE_AUTHOR(DRV_COPYRIGHT);
3534 MODULE_LICENSE("GPL");
3535 MODULE_ALIAS_MISCDEV(TUN_MINOR);
3536 MODULE_ALIAS("devname:net/tun");