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 void __tun_xdp_flush_tfile(struct tun_file *tfile)
1290 /* Notify and wake up reader process */
1291 if (tfile->flags & TUN_FASYNC)
1292 kill_fasync(&tfile->fasync, SIGIO, POLL_IN);
1293 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
1296 static int tun_xdp_xmit(struct net_device *dev, int n,
1297 struct xdp_frame **frames, u32 flags)
1299 struct tun_struct *tun = netdev_priv(dev);
1300 struct tun_file *tfile;
1306 if (unlikely(flags & ~XDP_XMIT_FLAGS_MASK))
1311 numqueues = READ_ONCE(tun->numqueues);
1314 return -ENXIO; /* Caller will free/return all frames */
1317 tfile = rcu_dereference(tun->tfiles[smp_processor_id() %
1320 spin_lock(&tfile->tx_ring.producer_lock);
1321 for (i = 0; i < n; i++) {
1322 struct xdp_frame *xdp = frames[i];
1323 /* Encode the XDP flag into lowest bit for consumer to differ
1324 * XDP buffer from sk_buff.
1326 void *frame = tun_xdp_to_ptr(xdp);
1328 if (__ptr_ring_produce(&tfile->tx_ring, frame)) {
1329 this_cpu_inc(tun->pcpu_stats->tx_dropped);
1330 xdp_return_frame_rx_napi(xdp);
1334 spin_unlock(&tfile->tx_ring.producer_lock);
1336 if (flags & XDP_XMIT_FLUSH)
1337 __tun_xdp_flush_tfile(tfile);
1343 static int tun_xdp_tx(struct net_device *dev, struct xdp_buff *xdp)
1345 struct xdp_frame *frame = convert_to_xdp_frame(xdp);
1347 if (unlikely(!frame))
1350 return tun_xdp_xmit(dev, 1, &frame, 0);
1353 static void tun_xdp_flush(struct net_device *dev)
1355 struct tun_struct *tun = netdev_priv(dev);
1356 struct tun_file *tfile;
1361 numqueues = READ_ONCE(tun->numqueues);
1365 tfile = rcu_dereference(tun->tfiles[smp_processor_id() %
1367 __tun_xdp_flush_tfile(tfile);
1372 static const struct net_device_ops tap_netdev_ops = {
1373 .ndo_uninit = tun_net_uninit,
1374 .ndo_open = tun_net_open,
1375 .ndo_stop = tun_net_close,
1376 .ndo_start_xmit = tun_net_xmit,
1377 .ndo_fix_features = tun_net_fix_features,
1378 .ndo_set_rx_mode = tun_net_mclist,
1379 .ndo_set_mac_address = eth_mac_addr,
1380 .ndo_validate_addr = eth_validate_addr,
1381 .ndo_select_queue = tun_select_queue,
1382 #ifdef CONFIG_NET_POLL_CONTROLLER
1383 .ndo_poll_controller = tun_poll_controller,
1385 .ndo_features_check = passthru_features_check,
1386 .ndo_set_rx_headroom = tun_set_headroom,
1387 .ndo_get_stats64 = tun_net_get_stats64,
1389 .ndo_xdp_xmit = tun_xdp_xmit,
1390 .ndo_xdp_flush = tun_xdp_flush,
1393 static void tun_flow_init(struct tun_struct *tun)
1397 for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++)
1398 INIT_HLIST_HEAD(&tun->flows[i]);
1400 tun->ageing_time = TUN_FLOW_EXPIRE;
1401 timer_setup(&tun->flow_gc_timer, tun_flow_cleanup, 0);
1402 mod_timer(&tun->flow_gc_timer,
1403 round_jiffies_up(jiffies + tun->ageing_time));
1406 static void tun_flow_uninit(struct tun_struct *tun)
1408 del_timer_sync(&tun->flow_gc_timer);
1409 tun_flow_flush(tun);
1413 #define MAX_MTU 65535
1415 /* Initialize net device. */
1416 static void tun_net_init(struct net_device *dev)
1418 struct tun_struct *tun = netdev_priv(dev);
1420 switch (tun->flags & TUN_TYPE_MASK) {
1422 dev->netdev_ops = &tun_netdev_ops;
1424 /* Point-to-Point TUN Device */
1425 dev->hard_header_len = 0;
1429 /* Zero header length */
1430 dev->type = ARPHRD_NONE;
1431 dev->flags = IFF_POINTOPOINT | IFF_NOARP | IFF_MULTICAST;
1435 dev->netdev_ops = &tap_netdev_ops;
1436 /* Ethernet TAP Device */
1438 dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1439 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
1441 eth_hw_addr_random(dev);
1446 dev->min_mtu = MIN_MTU;
1447 dev->max_mtu = MAX_MTU - dev->hard_header_len;
1450 static bool tun_sock_writeable(struct tun_struct *tun, struct tun_file *tfile)
1452 struct sock *sk = tfile->socket.sk;
1454 return (tun->dev->flags & IFF_UP) && sock_writeable(sk);
1457 /* Character device part */
1460 static __poll_t tun_chr_poll(struct file *file, poll_table *wait)
1462 struct tun_file *tfile = file->private_data;
1463 struct tun_struct *tun = tun_get(tfile);
1470 sk = tfile->socket.sk;
1472 tun_debug(KERN_INFO, tun, "tun_chr_poll\n");
1474 poll_wait(file, sk_sleep(sk), wait);
1476 if (!ptr_ring_empty(&tfile->tx_ring))
1477 mask |= EPOLLIN | EPOLLRDNORM;
1479 /* Make sure SOCKWQ_ASYNC_NOSPACE is set if not writable to
1480 * guarantee EPOLLOUT to be raised by either here or
1481 * tun_sock_write_space(). Then process could get notification
1482 * after it writes to a down device and meets -EIO.
1484 if (tun_sock_writeable(tun, tfile) ||
1485 (!test_and_set_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags) &&
1486 tun_sock_writeable(tun, tfile)))
1487 mask |= EPOLLOUT | EPOLLWRNORM;
1489 if (tun->dev->reg_state != NETREG_REGISTERED)
1496 static struct sk_buff *tun_napi_alloc_frags(struct tun_file *tfile,
1498 const struct iov_iter *it)
1500 struct sk_buff *skb;
1505 if (it->nr_segs > MAX_SKB_FRAGS + 1)
1506 return ERR_PTR(-ENOMEM);
1509 skb = napi_get_frags(&tfile->napi);
1512 return ERR_PTR(-ENOMEM);
1514 linear = iov_iter_single_seg_count(it);
1515 err = __skb_grow(skb, linear);
1520 skb->data_len = len - linear;
1521 skb->truesize += skb->data_len;
1523 for (i = 1; i < it->nr_segs; i++) {
1524 struct page_frag *pfrag = ¤t->task_frag;
1525 size_t fragsz = it->iov[i].iov_len;
1527 if (fragsz == 0 || fragsz > PAGE_SIZE) {
1532 if (!skb_page_frag_refill(fragsz, pfrag, GFP_KERNEL)) {
1537 skb_fill_page_desc(skb, i - 1, pfrag->page,
1538 pfrag->offset, fragsz);
1539 page_ref_inc(pfrag->page);
1540 pfrag->offset += fragsz;
1545 /* frees skb and all frags allocated with napi_alloc_frag() */
1546 napi_free_frags(&tfile->napi);
1547 return ERR_PTR(err);
1550 /* prepad is the amount to reserve at front. len is length after that.
1551 * linear is a hint as to how much to copy (usually headers). */
1552 static struct sk_buff *tun_alloc_skb(struct tun_file *tfile,
1553 size_t prepad, size_t len,
1554 size_t linear, int noblock)
1556 struct sock *sk = tfile->socket.sk;
1557 struct sk_buff *skb;
1560 /* Under a page? Don't bother with paged skb. */
1561 if (prepad + len < PAGE_SIZE || !linear)
1564 skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
1567 return ERR_PTR(err);
1569 skb_reserve(skb, prepad);
1570 skb_put(skb, linear);
1571 skb->data_len = len - linear;
1572 skb->len += len - linear;
1577 static void tun_rx_batched(struct tun_struct *tun, struct tun_file *tfile,
1578 struct sk_buff *skb, int more)
1580 struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
1581 struct sk_buff_head process_queue;
1582 u32 rx_batched = tun->rx_batched;
1585 if (!rx_batched || (!more && skb_queue_empty(queue))) {
1587 netif_receive_skb(skb);
1592 spin_lock(&queue->lock);
1593 if (!more || skb_queue_len(queue) == rx_batched) {
1594 __skb_queue_head_init(&process_queue);
1595 skb_queue_splice_tail_init(queue, &process_queue);
1598 __skb_queue_tail(queue, skb);
1600 spin_unlock(&queue->lock);
1603 struct sk_buff *nskb;
1606 while ((nskb = __skb_dequeue(&process_queue)))
1607 netif_receive_skb(nskb);
1608 netif_receive_skb(skb);
1613 static bool tun_can_build_skb(struct tun_struct *tun, struct tun_file *tfile,
1614 int len, int noblock, bool zerocopy)
1616 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
1619 if (tfile->socket.sk->sk_sndbuf != INT_MAX)
1628 if (SKB_DATA_ALIGN(len + TUN_RX_PAD) +
1629 SKB_DATA_ALIGN(sizeof(struct skb_shared_info)) > PAGE_SIZE)
1635 static struct sk_buff *tun_build_skb(struct tun_struct *tun,
1636 struct tun_file *tfile,
1637 struct iov_iter *from,
1638 struct virtio_net_hdr *hdr,
1639 int len, int *skb_xdp)
1641 struct page_frag *alloc_frag = ¤t->task_frag;
1642 struct sk_buff *skb;
1643 struct bpf_prog *xdp_prog;
1644 int buflen = SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
1645 unsigned int delta = 0;
1648 int err, pad = TUN_RX_PAD;
1651 xdp_prog = rcu_dereference(tun->xdp_prog);
1653 pad += TUN_HEADROOM;
1654 buflen += SKB_DATA_ALIGN(len + pad);
1657 alloc_frag->offset = ALIGN((u64)alloc_frag->offset, SMP_CACHE_BYTES);
1658 if (unlikely(!skb_page_frag_refill(buflen, alloc_frag, GFP_KERNEL)))
1659 return ERR_PTR(-ENOMEM);
1661 buf = (char *)page_address(alloc_frag->page) + alloc_frag->offset;
1662 copied = copy_page_from_iter(alloc_frag->page,
1663 alloc_frag->offset + pad,
1666 return ERR_PTR(-EFAULT);
1668 /* There's a small window that XDP may be set after the check
1669 * of xdp_prog above, this should be rare and for simplicity
1670 * we do XDP on skb in case the headroom is not enough.
1672 if (hdr->gso_type || !xdp_prog)
1679 xdp_prog = rcu_dereference(tun->xdp_prog);
1680 if (xdp_prog && !*skb_xdp) {
1681 struct xdp_buff xdp;
1685 xdp.data_hard_start = buf;
1686 xdp.data = buf + pad;
1687 xdp_set_data_meta_invalid(&xdp);
1688 xdp.data_end = xdp.data + len;
1689 xdp.rxq = &tfile->xdp_rxq;
1690 orig_data = xdp.data;
1691 act = bpf_prog_run_xdp(xdp_prog, &xdp);
1695 get_page(alloc_frag->page);
1696 alloc_frag->offset += buflen;
1697 err = xdp_do_redirect(tun->dev, &xdp, xdp_prog);
1705 get_page(alloc_frag->page);
1706 alloc_frag->offset += buflen;
1707 if (tun_xdp_tx(tun->dev, &xdp))
1709 tun_xdp_flush(tun->dev);
1714 delta = orig_data - xdp.data;
1715 len = xdp.data_end - xdp.data;
1718 bpf_warn_invalid_xdp_action(act);
1721 trace_xdp_exception(tun->dev, xdp_prog, act);
1728 skb = build_skb(buf, buflen);
1732 return ERR_PTR(-ENOMEM);
1735 skb_reserve(skb, pad - delta);
1737 get_page(alloc_frag->page);
1738 alloc_frag->offset += buflen;
1746 put_page(alloc_frag->page);
1750 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1754 /* Get packet from user space buffer */
1755 static ssize_t tun_get_user(struct tun_struct *tun, struct tun_file *tfile,
1756 void *msg_control, struct iov_iter *from,
1757 int noblock, bool more)
1759 struct tun_pi pi = { 0, cpu_to_be16(ETH_P_IP) };
1760 struct sk_buff *skb;
1761 size_t total_len = iov_iter_count(from);
1762 size_t len = total_len, align = tun->align, linear;
1763 struct virtio_net_hdr gso = { 0 };
1764 struct tun_pcpu_stats *stats;
1767 bool zerocopy = false;
1771 bool frags = tun_napi_frags_enabled(tun);
1773 if (!(tun->dev->flags & IFF_UP))
1776 if (!(tun->flags & IFF_NO_PI)) {
1777 if (len < sizeof(pi))
1781 if (!copy_from_iter_full(&pi, sizeof(pi), from))
1785 if (tun->flags & IFF_VNET_HDR) {
1786 int vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
1788 if (len < vnet_hdr_sz)
1792 if (!copy_from_iter_full(&gso, sizeof(gso), from))
1795 if ((gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
1796 tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2 > tun16_to_cpu(tun, gso.hdr_len))
1797 gso.hdr_len = cpu_to_tun16(tun, tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2);
1799 if (tun16_to_cpu(tun, gso.hdr_len) > len)
1801 iov_iter_advance(from, vnet_hdr_sz - sizeof(gso));
1804 if ((tun->flags & TUN_TYPE_MASK) == IFF_TAP) {
1805 align += NET_IP_ALIGN;
1806 if (unlikely(len < ETH_HLEN ||
1807 (gso.hdr_len && tun16_to_cpu(tun, gso.hdr_len) < ETH_HLEN)))
1811 good_linear = SKB_MAX_HEAD(align);
1814 struct iov_iter i = *from;
1816 /* There are 256 bytes to be copied in skb, so there is
1817 * enough room for skb expand head in case it is used.
1818 * The rest of the buffer is mapped from userspace.
1820 copylen = gso.hdr_len ? tun16_to_cpu(tun, gso.hdr_len) : GOODCOPY_LEN;
1821 if (copylen > good_linear)
1822 copylen = good_linear;
1824 iov_iter_advance(&i, copylen);
1825 if (iov_iter_npages(&i, INT_MAX) <= MAX_SKB_FRAGS)
1829 if (!frags && tun_can_build_skb(tun, tfile, len, noblock, zerocopy)) {
1830 /* For the packet that is not easy to be processed
1831 * (e.g gso or jumbo packet), we will do it at after
1832 * skb was created with generic XDP routine.
1834 skb = tun_build_skb(tun, tfile, from, &gso, len, &skb_xdp);
1836 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1837 return PTR_ERR(skb);
1844 if (tun16_to_cpu(tun, gso.hdr_len) > good_linear)
1845 linear = good_linear;
1847 linear = tun16_to_cpu(tun, gso.hdr_len);
1851 mutex_lock(&tfile->napi_mutex);
1852 skb = tun_napi_alloc_frags(tfile, copylen, from);
1853 /* tun_napi_alloc_frags() enforces a layout for the skb.
1854 * If zerocopy is enabled, then this layout will be
1855 * overwritten by zerocopy_sg_from_iter().
1859 skb = tun_alloc_skb(tfile, align, copylen, linear,
1864 if (PTR_ERR(skb) != -EAGAIN)
1865 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1867 mutex_unlock(&tfile->napi_mutex);
1868 return PTR_ERR(skb);
1872 err = zerocopy_sg_from_iter(skb, from);
1874 err = skb_copy_datagram_from_iter(skb, 0, from, len);
1877 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1880 tfile->napi.skb = NULL;
1881 mutex_unlock(&tfile->napi_mutex);
1888 if (virtio_net_hdr_to_skb(skb, &gso, tun_is_little_endian(tun))) {
1889 this_cpu_inc(tun->pcpu_stats->rx_frame_errors);
1892 tfile->napi.skb = NULL;
1893 mutex_unlock(&tfile->napi_mutex);
1899 switch (tun->flags & TUN_TYPE_MASK) {
1901 if (tun->flags & IFF_NO_PI) {
1902 u8 ip_version = skb->len ? (skb->data[0] >> 4) : 0;
1904 switch (ip_version) {
1906 pi.proto = htons(ETH_P_IP);
1909 pi.proto = htons(ETH_P_IPV6);
1912 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1918 skb_reset_mac_header(skb);
1919 skb->protocol = pi.proto;
1920 skb->dev = tun->dev;
1924 skb->protocol = eth_type_trans(skb, tun->dev);
1928 /* copy skb_ubuf_info for callback when skb has no error */
1930 skb_shinfo(skb)->destructor_arg = msg_control;
1931 skb_shinfo(skb)->tx_flags |= SKBTX_DEV_ZEROCOPY;
1932 skb_shinfo(skb)->tx_flags |= SKBTX_SHARED_FRAG;
1933 } else if (msg_control) {
1934 struct ubuf_info *uarg = msg_control;
1935 uarg->callback(uarg, false);
1938 skb_reset_network_header(skb);
1939 skb_probe_transport_header(skb, 0);
1942 struct bpf_prog *xdp_prog;
1946 xdp_prog = rcu_dereference(tun->xdp_prog);
1948 ret = do_xdp_generic(xdp_prog, skb);
1949 if (ret != XDP_PASS) {
1957 /* Compute the costly rx hash only if needed for flow updates.
1958 * We may get a very small possibility of OOO during switching, not
1959 * worth to optimize.
1961 if (!rcu_access_pointer(tun->steering_prog) && tun->numqueues > 1 &&
1963 rxhash = __skb_get_hash_symmetric(skb);
1966 /* Exercise flow dissector code path. */
1967 u32 headlen = eth_get_headlen(skb->data, skb_headlen(skb));
1969 if (unlikely(headlen > skb_headlen(skb))) {
1970 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1971 napi_free_frags(&tfile->napi);
1972 mutex_unlock(&tfile->napi_mutex);
1978 napi_gro_frags(&tfile->napi);
1980 mutex_unlock(&tfile->napi_mutex);
1981 } else if (tfile->napi_enabled) {
1982 struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
1985 spin_lock_bh(&queue->lock);
1986 __skb_queue_tail(queue, skb);
1987 queue_len = skb_queue_len(queue);
1988 spin_unlock(&queue->lock);
1990 if (!more || queue_len > NAPI_POLL_WEIGHT)
1991 napi_schedule(&tfile->napi);
1994 } else if (!IS_ENABLED(CONFIG_4KSTACKS)) {
1995 tun_rx_batched(tun, tfile, skb, more);
2000 stats = get_cpu_ptr(tun->pcpu_stats);
2001 u64_stats_update_begin(&stats->syncp);
2002 stats->rx_packets++;
2003 stats->rx_bytes += len;
2004 u64_stats_update_end(&stats->syncp);
2008 tun_flow_update(tun, rxhash, tfile);
2013 static ssize_t tun_chr_write_iter(struct kiocb *iocb, struct iov_iter *from)
2015 struct file *file = iocb->ki_filp;
2016 struct tun_file *tfile = file->private_data;
2017 struct tun_struct *tun = tun_get(tfile);
2023 result = tun_get_user(tun, tfile, NULL, from,
2024 file->f_flags & O_NONBLOCK, false);
2030 static ssize_t tun_put_user_xdp(struct tun_struct *tun,
2031 struct tun_file *tfile,
2032 struct xdp_frame *xdp_frame,
2033 struct iov_iter *iter)
2035 int vnet_hdr_sz = 0;
2036 size_t size = xdp_frame->len;
2037 struct tun_pcpu_stats *stats;
2040 if (tun->flags & IFF_VNET_HDR) {
2041 struct virtio_net_hdr gso = { 0 };
2043 vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
2044 if (unlikely(iov_iter_count(iter) < vnet_hdr_sz))
2046 if (unlikely(copy_to_iter(&gso, sizeof(gso), iter) !=
2049 iov_iter_advance(iter, vnet_hdr_sz - sizeof(gso));
2052 ret = copy_to_iter(xdp_frame->data, size, iter) + vnet_hdr_sz;
2054 stats = get_cpu_ptr(tun->pcpu_stats);
2055 u64_stats_update_begin(&stats->syncp);
2056 stats->tx_packets++;
2057 stats->tx_bytes += ret;
2058 u64_stats_update_end(&stats->syncp);
2059 put_cpu_ptr(tun->pcpu_stats);
2064 /* Put packet to the user space buffer */
2065 static ssize_t tun_put_user(struct tun_struct *tun,
2066 struct tun_file *tfile,
2067 struct sk_buff *skb,
2068 struct iov_iter *iter)
2070 struct tun_pi pi = { 0, skb->protocol };
2071 struct tun_pcpu_stats *stats;
2073 int vlan_offset = 0;
2075 int vnet_hdr_sz = 0;
2077 if (skb_vlan_tag_present(skb))
2078 vlan_hlen = VLAN_HLEN;
2080 if (tun->flags & IFF_VNET_HDR)
2081 vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
2083 total = skb->len + vlan_hlen + vnet_hdr_sz;
2085 if (!(tun->flags & IFF_NO_PI)) {
2086 if (iov_iter_count(iter) < sizeof(pi))
2089 total += sizeof(pi);
2090 if (iov_iter_count(iter) < total) {
2091 /* Packet will be striped */
2092 pi.flags |= TUN_PKT_STRIP;
2095 if (copy_to_iter(&pi, sizeof(pi), iter) != sizeof(pi))
2100 struct virtio_net_hdr gso;
2102 if (iov_iter_count(iter) < vnet_hdr_sz)
2105 if (virtio_net_hdr_from_skb(skb, &gso,
2106 tun_is_little_endian(tun), true)) {
2107 struct skb_shared_info *sinfo = skb_shinfo(skb);
2108 pr_err("unexpected GSO type: "
2109 "0x%x, gso_size %d, hdr_len %d\n",
2110 sinfo->gso_type, tun16_to_cpu(tun, gso.gso_size),
2111 tun16_to_cpu(tun, gso.hdr_len));
2112 print_hex_dump(KERN_ERR, "tun: ",
2115 min((int)tun16_to_cpu(tun, gso.hdr_len), 64), true);
2120 if (copy_to_iter(&gso, sizeof(gso), iter) != sizeof(gso))
2123 iov_iter_advance(iter, vnet_hdr_sz - sizeof(gso));
2130 veth.h_vlan_proto = skb->vlan_proto;
2131 veth.h_vlan_TCI = htons(skb_vlan_tag_get(skb));
2133 vlan_offset = offsetof(struct vlan_ethhdr, h_vlan_proto);
2135 ret = skb_copy_datagram_iter(skb, 0, iter, vlan_offset);
2136 if (ret || !iov_iter_count(iter))
2139 ret = copy_to_iter(&veth, sizeof(veth), iter);
2140 if (ret != sizeof(veth) || !iov_iter_count(iter))
2144 skb_copy_datagram_iter(skb, vlan_offset, iter, skb->len - vlan_offset);
2147 /* caller is in process context, */
2148 stats = get_cpu_ptr(tun->pcpu_stats);
2149 u64_stats_update_begin(&stats->syncp);
2150 stats->tx_packets++;
2151 stats->tx_bytes += skb->len + vlan_hlen;
2152 u64_stats_update_end(&stats->syncp);
2153 put_cpu_ptr(tun->pcpu_stats);
2158 static void *tun_ring_recv(struct tun_file *tfile, int noblock, int *err)
2160 DECLARE_WAITQUEUE(wait, current);
2164 ptr = ptr_ring_consume(&tfile->tx_ring);
2172 add_wait_queue(&tfile->wq.wait, &wait);
2173 current->state = TASK_INTERRUPTIBLE;
2176 ptr = ptr_ring_consume(&tfile->tx_ring);
2179 if (signal_pending(current)) {
2180 error = -ERESTARTSYS;
2183 if (tfile->socket.sk->sk_shutdown & RCV_SHUTDOWN) {
2191 current->state = TASK_RUNNING;
2192 remove_wait_queue(&tfile->wq.wait, &wait);
2199 static ssize_t tun_do_read(struct tun_struct *tun, struct tun_file *tfile,
2200 struct iov_iter *to,
2201 int noblock, void *ptr)
2206 tun_debug(KERN_INFO, tun, "tun_do_read\n");
2208 if (!iov_iter_count(to)) {
2214 /* Read frames from ring */
2215 ptr = tun_ring_recv(tfile, noblock, &err);
2220 if (tun_is_xdp_frame(ptr)) {
2221 struct xdp_frame *xdpf = tun_ptr_to_xdp(ptr);
2223 ret = tun_put_user_xdp(tun, tfile, xdpf, to);
2224 xdp_return_frame(xdpf);
2226 struct sk_buff *skb = ptr;
2228 ret = tun_put_user(tun, tfile, skb, to);
2229 if (unlikely(ret < 0))
2238 static ssize_t tun_chr_read_iter(struct kiocb *iocb, struct iov_iter *to)
2240 struct file *file = iocb->ki_filp;
2241 struct tun_file *tfile = file->private_data;
2242 struct tun_struct *tun = tun_get(tfile);
2243 ssize_t len = iov_iter_count(to), ret;
2247 ret = tun_do_read(tun, tfile, to, file->f_flags & O_NONBLOCK, NULL);
2248 ret = min_t(ssize_t, ret, len);
2255 static void tun_prog_free(struct rcu_head *rcu)
2257 struct tun_prog *prog = container_of(rcu, struct tun_prog, rcu);
2259 bpf_prog_destroy(prog->prog);
2263 static int __tun_set_ebpf(struct tun_struct *tun,
2264 struct tun_prog __rcu **prog_p,
2265 struct bpf_prog *prog)
2267 struct tun_prog *old, *new = NULL;
2270 new = kmalloc(sizeof(*new), GFP_KERNEL);
2276 spin_lock_bh(&tun->lock);
2277 old = rcu_dereference_protected(*prog_p,
2278 lockdep_is_held(&tun->lock));
2279 rcu_assign_pointer(*prog_p, new);
2280 spin_unlock_bh(&tun->lock);
2283 call_rcu(&old->rcu, tun_prog_free);
2288 static void tun_free_netdev(struct net_device *dev)
2290 struct tun_struct *tun = netdev_priv(dev);
2292 BUG_ON(!(list_empty(&tun->disabled)));
2293 free_percpu(tun->pcpu_stats);
2294 tun_flow_uninit(tun);
2295 security_tun_dev_free_security(tun->security);
2296 __tun_set_ebpf(tun, &tun->steering_prog, NULL);
2297 __tun_set_ebpf(tun, &tun->filter_prog, NULL);
2300 static void tun_setup(struct net_device *dev)
2302 struct tun_struct *tun = netdev_priv(dev);
2304 tun->owner = INVALID_UID;
2305 tun->group = INVALID_GID;
2307 dev->ethtool_ops = &tun_ethtool_ops;
2308 dev->needs_free_netdev = true;
2309 dev->priv_destructor = tun_free_netdev;
2310 /* We prefer our own queue length */
2311 dev->tx_queue_len = TUN_READQ_SIZE;
2314 /* Trivial set of netlink ops to allow deleting tun or tap
2315 * device with netlink.
2317 static int tun_validate(struct nlattr *tb[], struct nlattr *data[],
2318 struct netlink_ext_ack *extack)
2323 static size_t tun_get_size(const struct net_device *dev)
2325 BUILD_BUG_ON(sizeof(u32) != sizeof(uid_t));
2326 BUILD_BUG_ON(sizeof(u32) != sizeof(gid_t));
2328 return nla_total_size(sizeof(uid_t)) + /* OWNER */
2329 nla_total_size(sizeof(gid_t)) + /* GROUP */
2330 nla_total_size(sizeof(u8)) + /* TYPE */
2331 nla_total_size(sizeof(u8)) + /* PI */
2332 nla_total_size(sizeof(u8)) + /* VNET_HDR */
2333 nla_total_size(sizeof(u8)) + /* PERSIST */
2334 nla_total_size(sizeof(u8)) + /* MULTI_QUEUE */
2335 nla_total_size(sizeof(u32)) + /* NUM_QUEUES */
2336 nla_total_size(sizeof(u32)) + /* NUM_DISABLED_QUEUES */
2340 static int tun_fill_info(struct sk_buff *skb, const struct net_device *dev)
2342 struct tun_struct *tun = netdev_priv(dev);
2344 if (nla_put_u8(skb, IFLA_TUN_TYPE, tun->flags & TUN_TYPE_MASK))
2345 goto nla_put_failure;
2346 if (uid_valid(tun->owner) &&
2347 nla_put_u32(skb, IFLA_TUN_OWNER,
2348 from_kuid_munged(current_user_ns(), tun->owner)))
2349 goto nla_put_failure;
2350 if (gid_valid(tun->group) &&
2351 nla_put_u32(skb, IFLA_TUN_GROUP,
2352 from_kgid_munged(current_user_ns(), tun->group)))
2353 goto nla_put_failure;
2354 if (nla_put_u8(skb, IFLA_TUN_PI, !(tun->flags & IFF_NO_PI)))
2355 goto nla_put_failure;
2356 if (nla_put_u8(skb, IFLA_TUN_VNET_HDR, !!(tun->flags & IFF_VNET_HDR)))
2357 goto nla_put_failure;
2358 if (nla_put_u8(skb, IFLA_TUN_PERSIST, !!(tun->flags & IFF_PERSIST)))
2359 goto nla_put_failure;
2360 if (nla_put_u8(skb, IFLA_TUN_MULTI_QUEUE,
2361 !!(tun->flags & IFF_MULTI_QUEUE)))
2362 goto nla_put_failure;
2363 if (tun->flags & IFF_MULTI_QUEUE) {
2364 if (nla_put_u32(skb, IFLA_TUN_NUM_QUEUES, tun->numqueues))
2365 goto nla_put_failure;
2366 if (nla_put_u32(skb, IFLA_TUN_NUM_DISABLED_QUEUES,
2368 goto nla_put_failure;
2377 static struct rtnl_link_ops tun_link_ops __read_mostly = {
2379 .priv_size = sizeof(struct tun_struct),
2381 .validate = tun_validate,
2382 .get_size = tun_get_size,
2383 .fill_info = tun_fill_info,
2386 static void tun_sock_write_space(struct sock *sk)
2388 struct tun_file *tfile;
2389 wait_queue_head_t *wqueue;
2391 if (!sock_writeable(sk))
2394 if (!test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags))
2397 wqueue = sk_sleep(sk);
2398 if (wqueue && waitqueue_active(wqueue))
2399 wake_up_interruptible_sync_poll(wqueue, EPOLLOUT |
2400 EPOLLWRNORM | EPOLLWRBAND);
2402 tfile = container_of(sk, struct tun_file, sk);
2403 kill_fasync(&tfile->fasync, SIGIO, POLL_OUT);
2406 static int tun_sendmsg(struct socket *sock, struct msghdr *m, size_t total_len)
2409 struct tun_file *tfile = container_of(sock, struct tun_file, socket);
2410 struct tun_struct *tun = tun_get(tfile);
2415 ret = tun_get_user(tun, tfile, m->msg_control, &m->msg_iter,
2416 m->msg_flags & MSG_DONTWAIT,
2417 m->msg_flags & MSG_MORE);
2422 static int tun_recvmsg(struct socket *sock, struct msghdr *m, size_t total_len,
2425 struct tun_file *tfile = container_of(sock, struct tun_file, socket);
2426 struct tun_struct *tun = tun_get(tfile);
2427 void *ptr = m->msg_control;
2435 if (flags & ~(MSG_DONTWAIT|MSG_TRUNC|MSG_ERRQUEUE)) {
2439 if (flags & MSG_ERRQUEUE) {
2440 ret = sock_recv_errqueue(sock->sk, m, total_len,
2441 SOL_PACKET, TUN_TX_TIMESTAMP);
2444 ret = tun_do_read(tun, tfile, &m->msg_iter, flags & MSG_DONTWAIT, ptr);
2445 if (ret > (ssize_t)total_len) {
2446 m->msg_flags |= MSG_TRUNC;
2447 ret = flags & MSG_TRUNC ? ret : total_len;
2460 static int tun_ptr_peek_len(void *ptr)
2463 if (tun_is_xdp_frame(ptr)) {
2464 struct xdp_frame *xdpf = tun_ptr_to_xdp(ptr);
2468 return __skb_array_len_with_tag(ptr);
2474 static int tun_peek_len(struct socket *sock)
2476 struct tun_file *tfile = container_of(sock, struct tun_file, socket);
2477 struct tun_struct *tun;
2480 tun = tun_get(tfile);
2484 ret = PTR_RING_PEEK_CALL(&tfile->tx_ring, tun_ptr_peek_len);
2490 /* Ops structure to mimic raw sockets with tun */
2491 static const struct proto_ops tun_socket_ops = {
2492 .peek_len = tun_peek_len,
2493 .sendmsg = tun_sendmsg,
2494 .recvmsg = tun_recvmsg,
2497 static struct proto tun_proto = {
2499 .owner = THIS_MODULE,
2500 .obj_size = sizeof(struct tun_file),
2503 static int tun_flags(struct tun_struct *tun)
2505 return tun->flags & (TUN_FEATURES | IFF_PERSIST | IFF_TUN | IFF_TAP);
2508 static ssize_t tun_show_flags(struct device *dev, struct device_attribute *attr,
2511 struct tun_struct *tun = netdev_priv(to_net_dev(dev));
2512 return sprintf(buf, "0x%x\n", tun_flags(tun));
2515 static ssize_t tun_show_owner(struct device *dev, struct device_attribute *attr,
2518 struct tun_struct *tun = netdev_priv(to_net_dev(dev));
2519 return uid_valid(tun->owner)?
2520 sprintf(buf, "%u\n",
2521 from_kuid_munged(current_user_ns(), tun->owner)):
2522 sprintf(buf, "-1\n");
2525 static ssize_t tun_show_group(struct device *dev, struct device_attribute *attr,
2528 struct tun_struct *tun = netdev_priv(to_net_dev(dev));
2529 return gid_valid(tun->group) ?
2530 sprintf(buf, "%u\n",
2531 from_kgid_munged(current_user_ns(), tun->group)):
2532 sprintf(buf, "-1\n");
2535 static DEVICE_ATTR(tun_flags, 0444, tun_show_flags, NULL);
2536 static DEVICE_ATTR(owner, 0444, tun_show_owner, NULL);
2537 static DEVICE_ATTR(group, 0444, tun_show_group, NULL);
2539 static struct attribute *tun_dev_attrs[] = {
2540 &dev_attr_tun_flags.attr,
2541 &dev_attr_owner.attr,
2542 &dev_attr_group.attr,
2546 static const struct attribute_group tun_attr_group = {
2547 .attrs = tun_dev_attrs
2550 static int tun_set_iff(struct net *net, struct file *file, struct ifreq *ifr)
2552 struct tun_struct *tun;
2553 struct tun_file *tfile = file->private_data;
2554 struct net_device *dev;
2557 if (tfile->detached)
2560 if ((ifr->ifr_flags & IFF_NAPI_FRAGS)) {
2561 if (!capable(CAP_NET_ADMIN))
2564 if (!(ifr->ifr_flags & IFF_NAPI) ||
2565 (ifr->ifr_flags & TUN_TYPE_MASK) != IFF_TAP)
2569 dev = __dev_get_by_name(net, ifr->ifr_name);
2571 if (ifr->ifr_flags & IFF_TUN_EXCL)
2573 if ((ifr->ifr_flags & IFF_TUN) && dev->netdev_ops == &tun_netdev_ops)
2574 tun = netdev_priv(dev);
2575 else if ((ifr->ifr_flags & IFF_TAP) && dev->netdev_ops == &tap_netdev_ops)
2576 tun = netdev_priv(dev);
2580 if (!!(ifr->ifr_flags & IFF_MULTI_QUEUE) !=
2581 !!(tun->flags & IFF_MULTI_QUEUE))
2584 if (tun_not_capable(tun))
2586 err = security_tun_dev_open(tun->security);
2590 err = tun_attach(tun, file, ifr->ifr_flags & IFF_NOFILTER,
2591 ifr->ifr_flags & IFF_NAPI);
2595 if (tun->flags & IFF_MULTI_QUEUE &&
2596 (tun->numqueues + tun->numdisabled > 1)) {
2597 /* One or more queue has already been attached, no need
2598 * to initialize the device again.
2600 netdev_state_change(dev);
2604 tun->flags = (tun->flags & ~TUN_FEATURES) |
2605 (ifr->ifr_flags & TUN_FEATURES);
2607 netdev_state_change(dev);
2610 unsigned long flags = 0;
2611 int queues = ifr->ifr_flags & IFF_MULTI_QUEUE ?
2614 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
2616 err = security_tun_dev_create();
2621 if (ifr->ifr_flags & IFF_TUN) {
2625 } else if (ifr->ifr_flags & IFF_TAP) {
2633 name = ifr->ifr_name;
2635 dev = alloc_netdev_mqs(sizeof(struct tun_struct), name,
2636 NET_NAME_UNKNOWN, tun_setup, queues,
2641 err = dev_get_valid_name(net, dev, name);
2645 dev_net_set(dev, net);
2646 dev->rtnl_link_ops = &tun_link_ops;
2647 dev->ifindex = tfile->ifindex;
2648 dev->sysfs_groups[0] = &tun_attr_group;
2650 tun = netdev_priv(dev);
2653 tun->txflt.count = 0;
2654 tun->vnet_hdr_sz = sizeof(struct virtio_net_hdr);
2656 tun->align = NET_SKB_PAD;
2657 tun->filter_attached = false;
2658 tun->sndbuf = tfile->socket.sk->sk_sndbuf;
2659 tun->rx_batched = 0;
2660 RCU_INIT_POINTER(tun->steering_prog, NULL);
2662 tun->pcpu_stats = netdev_alloc_pcpu_stats(struct tun_pcpu_stats);
2663 if (!tun->pcpu_stats) {
2668 spin_lock_init(&tun->lock);
2670 err = security_tun_dev_alloc_security(&tun->security);
2677 dev->hw_features = NETIF_F_SG | NETIF_F_FRAGLIST |
2678 TUN_USER_FEATURES | NETIF_F_HW_VLAN_CTAG_TX |
2679 NETIF_F_HW_VLAN_STAG_TX;
2680 dev->features = dev->hw_features | NETIF_F_LLTX;
2681 dev->vlan_features = dev->features &
2682 ~(NETIF_F_HW_VLAN_CTAG_TX |
2683 NETIF_F_HW_VLAN_STAG_TX);
2685 tun->flags = (tun->flags & ~TUN_FEATURES) |
2686 (ifr->ifr_flags & TUN_FEATURES);
2688 INIT_LIST_HEAD(&tun->disabled);
2689 err = tun_attach(tun, file, false, ifr->ifr_flags & IFF_NAPI);
2693 err = register_netdevice(tun->dev);
2698 netif_carrier_on(tun->dev);
2700 tun_debug(KERN_INFO, tun, "tun_set_iff\n");
2702 /* Make sure persistent devices do not get stuck in
2705 if (netif_running(tun->dev))
2706 netif_tx_wake_all_queues(tun->dev);
2708 strcpy(ifr->ifr_name, tun->dev->name);
2712 tun_detach_all(dev);
2713 /* register_netdevice() already called tun_free_netdev() */
2717 tun_flow_uninit(tun);
2718 security_tun_dev_free_security(tun->security);
2720 free_percpu(tun->pcpu_stats);
2726 static void tun_get_iff(struct net *net, struct tun_struct *tun,
2729 tun_debug(KERN_INFO, tun, "tun_get_iff\n");
2731 strcpy(ifr->ifr_name, tun->dev->name);
2733 ifr->ifr_flags = tun_flags(tun);
2737 /* This is like a cut-down ethtool ops, except done via tun fd so no
2738 * privs required. */
2739 static int set_offload(struct tun_struct *tun, unsigned long arg)
2741 netdev_features_t features = 0;
2743 if (arg & TUN_F_CSUM) {
2744 features |= NETIF_F_HW_CSUM;
2747 if (arg & (TUN_F_TSO4|TUN_F_TSO6)) {
2748 if (arg & TUN_F_TSO_ECN) {
2749 features |= NETIF_F_TSO_ECN;
2750 arg &= ~TUN_F_TSO_ECN;
2752 if (arg & TUN_F_TSO4)
2753 features |= NETIF_F_TSO;
2754 if (arg & TUN_F_TSO6)
2755 features |= NETIF_F_TSO6;
2756 arg &= ~(TUN_F_TSO4|TUN_F_TSO6);
2762 /* This gives the user a way to test for new features in future by
2763 * trying to set them. */
2767 tun->set_features = features;
2768 tun->dev->wanted_features &= ~TUN_USER_FEATURES;
2769 tun->dev->wanted_features |= features;
2770 netdev_update_features(tun->dev);
2775 static void tun_detach_filter(struct tun_struct *tun, int n)
2778 struct tun_file *tfile;
2780 for (i = 0; i < n; i++) {
2781 tfile = rtnl_dereference(tun->tfiles[i]);
2782 lock_sock(tfile->socket.sk);
2783 sk_detach_filter(tfile->socket.sk);
2784 release_sock(tfile->socket.sk);
2787 tun->filter_attached = false;
2790 static int tun_attach_filter(struct tun_struct *tun)
2793 struct tun_file *tfile;
2795 for (i = 0; i < tun->numqueues; i++) {
2796 tfile = rtnl_dereference(tun->tfiles[i]);
2797 lock_sock(tfile->socket.sk);
2798 ret = sk_attach_filter(&tun->fprog, tfile->socket.sk);
2799 release_sock(tfile->socket.sk);
2801 tun_detach_filter(tun, i);
2806 tun->filter_attached = true;
2810 static void tun_set_sndbuf(struct tun_struct *tun)
2812 struct tun_file *tfile;
2815 for (i = 0; i < tun->numqueues; i++) {
2816 tfile = rtnl_dereference(tun->tfiles[i]);
2817 tfile->socket.sk->sk_sndbuf = tun->sndbuf;
2821 static int tun_set_queue(struct file *file, struct ifreq *ifr)
2823 struct tun_file *tfile = file->private_data;
2824 struct tun_struct *tun;
2829 if (ifr->ifr_flags & IFF_ATTACH_QUEUE) {
2830 tun = tfile->detached;
2835 ret = security_tun_dev_attach_queue(tun->security);
2838 ret = tun_attach(tun, file, false, tun->flags & IFF_NAPI);
2839 } else if (ifr->ifr_flags & IFF_DETACH_QUEUE) {
2840 tun = rtnl_dereference(tfile->tun);
2841 if (!tun || !(tun->flags & IFF_MULTI_QUEUE) || tfile->detached)
2844 __tun_detach(tfile, false);
2849 netdev_state_change(tun->dev);
2856 static int tun_set_ebpf(struct tun_struct *tun, struct tun_prog **prog_p,
2859 struct bpf_prog *prog;
2862 if (copy_from_user(&fd, data, sizeof(fd)))
2868 prog = bpf_prog_get_type(fd, BPF_PROG_TYPE_SOCKET_FILTER);
2870 return PTR_ERR(prog);
2873 return __tun_set_ebpf(tun, prog_p, prog);
2876 static long __tun_chr_ioctl(struct file *file, unsigned int cmd,
2877 unsigned long arg, int ifreq_len)
2879 struct tun_file *tfile = file->private_data;
2880 struct net *net = sock_net(&tfile->sk);
2881 struct tun_struct *tun;
2882 void __user* argp = (void __user*)arg;
2888 unsigned int ifindex;
2891 bool do_notify = false;
2893 if (cmd == TUNSETIFF || cmd == TUNSETQUEUE ||
2894 (_IOC_TYPE(cmd) == SOCK_IOC_TYPE && cmd != SIOCGSKNS)) {
2895 if (copy_from_user(&ifr, argp, ifreq_len))
2898 memset(&ifr, 0, sizeof(ifr));
2900 if (cmd == TUNGETFEATURES) {
2901 /* Currently this just means: "what IFF flags are valid?".
2902 * This is needed because we never checked for invalid flags on
2905 return put_user(IFF_TUN | IFF_TAP | TUN_FEATURES,
2906 (unsigned int __user*)argp);
2907 } else if (cmd == TUNSETQUEUE) {
2908 return tun_set_queue(file, &ifr);
2909 } else if (cmd == SIOCGSKNS) {
2910 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
2912 return open_related_ns(&net->ns, get_net_ns);
2918 tun = tun_get(tfile);
2919 if (cmd == TUNSETIFF) {
2924 ifr.ifr_name[IFNAMSIZ-1] = '\0';
2926 ret = tun_set_iff(net, file, &ifr);
2931 if (copy_to_user(argp, &ifr, ifreq_len))
2935 if (cmd == TUNSETIFINDEX) {
2941 if (copy_from_user(&ifindex, argp, sizeof(ifindex)))
2945 tfile->ifindex = ifindex;
2953 tun_debug(KERN_INFO, tun, "tun_chr_ioctl cmd %u\n", cmd);
2958 tun_get_iff(current->nsproxy->net_ns, tun, &ifr);
2960 if (tfile->detached)
2961 ifr.ifr_flags |= IFF_DETACH_QUEUE;
2962 if (!tfile->socket.sk->sk_filter)
2963 ifr.ifr_flags |= IFF_NOFILTER;
2965 if (copy_to_user(argp, &ifr, ifreq_len))
2970 /* Disable/Enable checksum */
2972 /* [unimplemented] */
2973 tun_debug(KERN_INFO, tun, "ignored: set checksum %s\n",
2974 arg ? "disabled" : "enabled");
2978 /* Disable/Enable persist mode. Keep an extra reference to the
2979 * module to prevent the module being unprobed.
2981 if (arg && !(tun->flags & IFF_PERSIST)) {
2982 tun->flags |= IFF_PERSIST;
2983 __module_get(THIS_MODULE);
2986 if (!arg && (tun->flags & IFF_PERSIST)) {
2987 tun->flags &= ~IFF_PERSIST;
2988 module_put(THIS_MODULE);
2992 tun_debug(KERN_INFO, tun, "persist %s\n",
2993 arg ? "enabled" : "disabled");
2997 /* Set owner of the device */
2998 owner = make_kuid(current_user_ns(), arg);
2999 if (!uid_valid(owner)) {
3005 tun_debug(KERN_INFO, tun, "owner set to %u\n",
3006 from_kuid(&init_user_ns, tun->owner));
3010 /* Set group of the device */
3011 group = make_kgid(current_user_ns(), arg);
3012 if (!gid_valid(group)) {
3018 tun_debug(KERN_INFO, tun, "group set to %u\n",
3019 from_kgid(&init_user_ns, tun->group));
3023 /* Only allow setting the type when the interface is down */
3024 if (tun->dev->flags & IFF_UP) {
3025 tun_debug(KERN_INFO, tun,
3026 "Linktype set failed because interface is up\n");
3029 tun->dev->type = (int) arg;
3030 tun_debug(KERN_INFO, tun, "linktype set to %d\n",
3042 ret = set_offload(tun, arg);
3045 case TUNSETTXFILTER:
3046 /* Can be set only for TAPs */
3048 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3050 ret = update_filter(&tun->txflt, (void __user *)arg);
3054 /* Get hw address */
3055 memcpy(ifr.ifr_hwaddr.sa_data, tun->dev->dev_addr, ETH_ALEN);
3056 ifr.ifr_hwaddr.sa_family = tun->dev->type;
3057 if (copy_to_user(argp, &ifr, ifreq_len))
3062 /* Set hw address */
3063 tun_debug(KERN_DEBUG, tun, "set hw address: %pM\n",
3064 ifr.ifr_hwaddr.sa_data);
3066 ret = dev_set_mac_address(tun->dev, &ifr.ifr_hwaddr);
3070 sndbuf = tfile->socket.sk->sk_sndbuf;
3071 if (copy_to_user(argp, &sndbuf, sizeof(sndbuf)))
3076 if (copy_from_user(&sndbuf, argp, sizeof(sndbuf))) {
3085 tun->sndbuf = sndbuf;
3086 tun_set_sndbuf(tun);
3089 case TUNGETVNETHDRSZ:
3090 vnet_hdr_sz = tun->vnet_hdr_sz;
3091 if (copy_to_user(argp, &vnet_hdr_sz, sizeof(vnet_hdr_sz)))
3095 case TUNSETVNETHDRSZ:
3096 if (copy_from_user(&vnet_hdr_sz, argp, sizeof(vnet_hdr_sz))) {
3100 if (vnet_hdr_sz < (int)sizeof(struct virtio_net_hdr)) {
3105 tun->vnet_hdr_sz = vnet_hdr_sz;
3109 le = !!(tun->flags & TUN_VNET_LE);
3110 if (put_user(le, (int __user *)argp))
3115 if (get_user(le, (int __user *)argp)) {
3120 tun->flags |= TUN_VNET_LE;
3122 tun->flags &= ~TUN_VNET_LE;
3126 ret = tun_get_vnet_be(tun, argp);
3130 ret = tun_set_vnet_be(tun, argp);
3133 case TUNATTACHFILTER:
3134 /* Can be set only for TAPs */
3136 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3139 if (copy_from_user(&tun->fprog, argp, sizeof(tun->fprog)))
3142 ret = tun_attach_filter(tun);
3145 case TUNDETACHFILTER:
3146 /* Can be set only for TAPs */
3148 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3151 tun_detach_filter(tun, tun->numqueues);
3156 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3159 if (copy_to_user(argp, &tun->fprog, sizeof(tun->fprog)))
3164 case TUNSETSTEERINGEBPF:
3165 ret = tun_set_ebpf(tun, &tun->steering_prog, argp);
3168 case TUNSETFILTEREBPF:
3169 ret = tun_set_ebpf(tun, &tun->filter_prog, argp);
3178 netdev_state_change(tun->dev);
3187 static long tun_chr_ioctl(struct file *file,
3188 unsigned int cmd, unsigned long arg)
3190 return __tun_chr_ioctl(file, cmd, arg, sizeof (struct ifreq));
3193 #ifdef CONFIG_COMPAT
3194 static long tun_chr_compat_ioctl(struct file *file,
3195 unsigned int cmd, unsigned long arg)
3200 case TUNSETTXFILTER:
3205 arg = (unsigned long)compat_ptr(arg);
3208 arg = (compat_ulong_t)arg;
3213 * compat_ifreq is shorter than ifreq, so we must not access beyond
3214 * the end of that structure. All fields that are used in this
3215 * driver are compatible though, we don't need to convert the
3218 return __tun_chr_ioctl(file, cmd, arg, sizeof(struct compat_ifreq));
3220 #endif /* CONFIG_COMPAT */
3222 static int tun_chr_fasync(int fd, struct file *file, int on)
3224 struct tun_file *tfile = file->private_data;
3227 if ((ret = fasync_helper(fd, file, on, &tfile->fasync)) < 0)
3231 __f_setown(file, task_pid(current), PIDTYPE_PID, 0);
3232 tfile->flags |= TUN_FASYNC;
3234 tfile->flags &= ~TUN_FASYNC;
3240 static int tun_chr_open(struct inode *inode, struct file * file)
3242 struct net *net = current->nsproxy->net_ns;
3243 struct tun_file *tfile;
3245 DBG1(KERN_INFO, "tunX: tun_chr_open\n");
3247 tfile = (struct tun_file *)sk_alloc(net, AF_UNSPEC, GFP_KERNEL,
3251 if (ptr_ring_init(&tfile->tx_ring, 0, GFP_KERNEL)) {
3252 sk_free(&tfile->sk);
3256 RCU_INIT_POINTER(tfile->tun, NULL);
3260 init_waitqueue_head(&tfile->wq.wait);
3261 RCU_INIT_POINTER(tfile->socket.wq, &tfile->wq);
3263 tfile->socket.file = file;
3264 tfile->socket.ops = &tun_socket_ops;
3266 sock_init_data(&tfile->socket, &tfile->sk);
3268 tfile->sk.sk_write_space = tun_sock_write_space;
3269 tfile->sk.sk_sndbuf = INT_MAX;
3271 file->private_data = tfile;
3272 INIT_LIST_HEAD(&tfile->next);
3274 sock_set_flag(&tfile->sk, SOCK_ZEROCOPY);
3279 static int tun_chr_close(struct inode *inode, struct file *file)
3281 struct tun_file *tfile = file->private_data;
3283 tun_detach(tfile, true);
3288 #ifdef CONFIG_PROC_FS
3289 static void tun_chr_show_fdinfo(struct seq_file *m, struct file *file)
3291 struct tun_file *tfile = file->private_data;
3292 struct tun_struct *tun;
3295 memset(&ifr, 0, sizeof(ifr));
3298 tun = tun_get(tfile);
3300 tun_get_iff(current->nsproxy->net_ns, tun, &ifr);
3306 seq_printf(m, "iff:\t%s\n", ifr.ifr_name);
3310 static const struct file_operations tun_fops = {
3311 .owner = THIS_MODULE,
3312 .llseek = no_llseek,
3313 .read_iter = tun_chr_read_iter,
3314 .write_iter = tun_chr_write_iter,
3315 .poll = tun_chr_poll,
3316 .unlocked_ioctl = tun_chr_ioctl,
3317 #ifdef CONFIG_COMPAT
3318 .compat_ioctl = tun_chr_compat_ioctl,
3320 .open = tun_chr_open,
3321 .release = tun_chr_close,
3322 .fasync = tun_chr_fasync,
3323 #ifdef CONFIG_PROC_FS
3324 .show_fdinfo = tun_chr_show_fdinfo,
3328 static struct miscdevice tun_miscdev = {
3331 .nodename = "net/tun",
3335 /* ethtool interface */
3337 static int tun_get_link_ksettings(struct net_device *dev,
3338 struct ethtool_link_ksettings *cmd)
3340 ethtool_link_ksettings_zero_link_mode(cmd, supported);
3341 ethtool_link_ksettings_zero_link_mode(cmd, advertising);
3342 cmd->base.speed = SPEED_10;
3343 cmd->base.duplex = DUPLEX_FULL;
3344 cmd->base.port = PORT_TP;
3345 cmd->base.phy_address = 0;
3346 cmd->base.autoneg = AUTONEG_DISABLE;
3350 static void tun_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
3352 struct tun_struct *tun = netdev_priv(dev);
3354 strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
3355 strlcpy(info->version, DRV_VERSION, sizeof(info->version));
3357 switch (tun->flags & TUN_TYPE_MASK) {
3359 strlcpy(info->bus_info, "tun", sizeof(info->bus_info));
3362 strlcpy(info->bus_info, "tap", sizeof(info->bus_info));
3367 static u32 tun_get_msglevel(struct net_device *dev)
3370 struct tun_struct *tun = netdev_priv(dev);
3377 static void tun_set_msglevel(struct net_device *dev, u32 value)
3380 struct tun_struct *tun = netdev_priv(dev);
3385 static int tun_get_coalesce(struct net_device *dev,
3386 struct ethtool_coalesce *ec)
3388 struct tun_struct *tun = netdev_priv(dev);
3390 ec->rx_max_coalesced_frames = tun->rx_batched;
3395 static int tun_set_coalesce(struct net_device *dev,
3396 struct ethtool_coalesce *ec)
3398 struct tun_struct *tun = netdev_priv(dev);
3400 if (ec->rx_max_coalesced_frames > NAPI_POLL_WEIGHT)
3401 tun->rx_batched = NAPI_POLL_WEIGHT;
3403 tun->rx_batched = ec->rx_max_coalesced_frames;
3408 static const struct ethtool_ops tun_ethtool_ops = {
3409 .get_drvinfo = tun_get_drvinfo,
3410 .get_msglevel = tun_get_msglevel,
3411 .set_msglevel = tun_set_msglevel,
3412 .get_link = ethtool_op_get_link,
3413 .get_ts_info = ethtool_op_get_ts_info,
3414 .get_coalesce = tun_get_coalesce,
3415 .set_coalesce = tun_set_coalesce,
3416 .get_link_ksettings = tun_get_link_ksettings,
3419 static int tun_queue_resize(struct tun_struct *tun)
3421 struct net_device *dev = tun->dev;
3422 struct tun_file *tfile;
3423 struct ptr_ring **rings;
3424 int n = tun->numqueues + tun->numdisabled;
3427 rings = kmalloc_array(n, sizeof(*rings), GFP_KERNEL);
3431 for (i = 0; i < tun->numqueues; i++) {
3432 tfile = rtnl_dereference(tun->tfiles[i]);
3433 rings[i] = &tfile->tx_ring;
3435 list_for_each_entry(tfile, &tun->disabled, next)
3436 rings[i++] = &tfile->tx_ring;
3438 ret = ptr_ring_resize_multiple(rings, n,
3439 dev->tx_queue_len, GFP_KERNEL,
3446 static int tun_device_event(struct notifier_block *unused,
3447 unsigned long event, void *ptr)
3449 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
3450 struct tun_struct *tun = netdev_priv(dev);
3452 if (dev->rtnl_link_ops != &tun_link_ops)
3456 case NETDEV_CHANGE_TX_QUEUE_LEN:
3457 if (tun_queue_resize(tun))
3467 static struct notifier_block tun_notifier_block __read_mostly = {
3468 .notifier_call = tun_device_event,
3471 static int __init tun_init(void)
3475 pr_info("%s, %s\n", DRV_DESCRIPTION, DRV_VERSION);
3477 ret = rtnl_link_register(&tun_link_ops);
3479 pr_err("Can't register link_ops\n");
3483 ret = misc_register(&tun_miscdev);
3485 pr_err("Can't register misc device %d\n", TUN_MINOR);
3489 ret = register_netdevice_notifier(&tun_notifier_block);
3491 pr_err("Can't register netdevice notifier\n");
3498 misc_deregister(&tun_miscdev);
3500 rtnl_link_unregister(&tun_link_ops);
3505 static void tun_cleanup(void)
3507 misc_deregister(&tun_miscdev);
3508 rtnl_link_unregister(&tun_link_ops);
3509 unregister_netdevice_notifier(&tun_notifier_block);
3512 /* Get an underlying socket object from tun file. Returns error unless file is
3513 * attached to a device. The returned object works like a packet socket, it
3514 * can be used for sock_sendmsg/sock_recvmsg. The caller is responsible for
3515 * holding a reference to the file for as long as the socket is in use. */
3516 struct socket *tun_get_socket(struct file *file)
3518 struct tun_file *tfile;
3519 if (file->f_op != &tun_fops)
3520 return ERR_PTR(-EINVAL);
3521 tfile = file->private_data;
3523 return ERR_PTR(-EBADFD);
3524 return &tfile->socket;
3526 EXPORT_SYMBOL_GPL(tun_get_socket);
3528 struct ptr_ring *tun_get_tx_ring(struct file *file)
3530 struct tun_file *tfile;
3532 if (file->f_op != &tun_fops)
3533 return ERR_PTR(-EINVAL);
3534 tfile = file->private_data;
3536 return ERR_PTR(-EBADFD);
3537 return &tfile->tx_ring;
3539 EXPORT_SYMBOL_GPL(tun_get_tx_ring);
3541 module_init(tun_init);
3542 module_exit(tun_cleanup);
3543 MODULE_DESCRIPTION(DRV_DESCRIPTION);
3544 MODULE_AUTHOR(DRV_COPYRIGHT);
3545 MODULE_LICENSE("GPL");
3546 MODULE_ALIAS_MISCDEV(TUN_MINOR);
3547 MODULE_ALIAS("devname:net/tun");