2 * TUN - Universal TUN/TAP device driver.
3 * Copyright (C) 1999-2002 Maxim Krasnyansky <maxk@qualcomm.com>
5 * This program is free software; you can redistribute it and/or modify
6 * it under the terms of the GNU General Public License as published by
7 * the Free Software Foundation; either version 2 of the License, or
8 * (at your option) any later version.
10 * This program is distributed in the hope that it will be useful,
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
13 * GNU General Public License for more details.
15 * $Id: tun.c,v 1.15 2002/03/01 02:44:24 maxk Exp $
21 * Mike Kershaw <dragorn@kismetwireless.net> 2005/08/14
22 * Add TUNSETLINK ioctl to set the link encapsulation
24 * Mark Smith <markzzzsmith@yahoo.com.au>
25 * Use eth_random_addr() for tap MAC address.
27 * Harald Roelle <harald.roelle@ifi.lmu.de> 2004/04/20
28 * Fixes in packet dropping, queue length setting and queue wakeup.
29 * Increased default tx queue length.
33 * Daniel Podlejski <underley@underley.eu.org>
34 * Modifications for 2.3.99-pre5 kernel.
37 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
39 #define DRV_NAME "tun"
40 #define DRV_VERSION "1.6"
41 #define DRV_DESCRIPTION "Universal TUN/TAP device driver"
42 #define DRV_COPYRIGHT "(C) 1999-2004 Max Krasnyansky <maxk@qualcomm.com>"
44 #include <linux/module.h>
45 #include <linux/errno.h>
46 #include <linux/kernel.h>
47 #include <linux/sched/signal.h>
48 #include <linux/major.h>
49 #include <linux/slab.h>
50 #include <linux/poll.h>
51 #include <linux/fcntl.h>
52 #include <linux/init.h>
53 #include <linux/skbuff.h>
54 #include <linux/netdevice.h>
55 #include <linux/etherdevice.h>
56 #include <linux/miscdevice.h>
57 #include <linux/ethtool.h>
58 #include <linux/rtnetlink.h>
59 #include <linux/compat.h>
61 #include <linux/if_arp.h>
62 #include <linux/if_ether.h>
63 #include <linux/if_tun.h>
64 #include <linux/if_vlan.h>
65 #include <linux/crc32.h>
66 #include <linux/nsproxy.h>
67 #include <linux/virtio_net.h>
68 #include <linux/rcupdate.h>
69 #include <net/net_namespace.h>
70 #include <net/netns/generic.h>
71 #include <net/rtnetlink.h>
73 #include <linux/seq_file.h>
74 #include <linux/uio.h>
75 #include <linux/skb_array.h>
76 #include <linux/bpf.h>
77 #include <linux/bpf_trace.h>
78 #include <linux/mutex.h>
80 #include <linux/uaccess.h>
81 #include <linux/proc_fs.h>
83 /* Uncomment to enable debugging */
84 /* #define TUN_DEBUG 1 */
89 #define tun_debug(level, tun, fmt, args...) \
92 netdev_printk(level, tun->dev, fmt, ##args); \
94 #define DBG1(level, fmt, args...) \
97 printk(level fmt, ##args); \
100 #define tun_debug(level, tun, fmt, args...) \
103 netdev_printk(level, tun->dev, fmt, ##args); \
105 #define DBG1(level, fmt, args...) \
108 printk(level fmt, ##args); \
112 #define TUN_HEADROOM 256
113 #define TUN_RX_PAD (NET_IP_ALIGN + NET_SKB_PAD)
115 /* TUN device flags */
117 /* IFF_ATTACH_QUEUE is never stored in device flags,
118 * overload it to mean fasync when stored there.
120 #define TUN_FASYNC IFF_ATTACH_QUEUE
121 /* High bits in flags field are unused. */
122 #define TUN_VNET_LE 0x80000000
123 #define TUN_VNET_BE 0x40000000
125 #define TUN_FEATURES (IFF_NO_PI | IFF_ONE_QUEUE | IFF_VNET_HDR | \
126 IFF_MULTI_QUEUE | IFF_NAPI | IFF_NAPI_FRAGS)
128 #define GOODCOPY_LEN 128
130 #define FLT_EXACT_COUNT 8
132 unsigned int count; /* Number of addrs. Zero means disabled */
133 u32 mask[2]; /* Mask of the hashed addrs */
134 unsigned char addr[FLT_EXACT_COUNT][ETH_ALEN];
137 /* MAX_TAP_QUEUES 256 is chosen to allow rx/tx queues to be equal
138 * to max number of VCPUs in guest. */
139 #define MAX_TAP_QUEUES 256
140 #define MAX_TAP_FLOWS 4096
142 #define TUN_FLOW_EXPIRE (3 * HZ)
144 struct tun_pcpu_stats {
149 struct u64_stats_sync syncp;
155 /* A tun_file connects an open character device to a tuntap netdevice. It
156 * also contains all socket related structures (except sock_fprog and tap_filter)
157 * to serve as one transmit queue for tuntap device. The sock_fprog and
158 * tap_filter were kept in tun_struct since they were used for filtering for the
159 * netdevice not for a specific queue (at least I didn't see the requirement for
163 * The tun_file and tun_struct are loosely coupled, the pointer from one to the
164 * other can only be read while rcu_read_lock or rtnl_lock is held.
168 struct socket socket;
170 struct tun_struct __rcu *tun;
171 struct fasync_struct *fasync;
172 /* only used for fasnyc */
176 unsigned int ifindex;
178 struct napi_struct napi;
180 struct mutex napi_mutex; /* Protects access to the above napi */
181 struct list_head next;
182 struct tun_struct *detached;
183 struct ptr_ring tx_ring;
184 struct xdp_rxq_info xdp_rxq;
187 struct tun_flow_entry {
188 struct hlist_node hash_link;
190 struct tun_struct *tun;
195 unsigned long updated;
198 #define TUN_NUM_FLOW_ENTRIES 1024
202 struct bpf_prog *prog;
205 /* Since the socket were moved to tun_file, to preserve the behavior of persist
206 * device, socket filter, sndbuf and vnet header size were restore when the
207 * file were attached to a persist device.
210 struct tun_file __rcu *tfiles[MAX_TAP_QUEUES];
211 unsigned int numqueues;
216 struct net_device *dev;
217 netdev_features_t set_features;
218 #define TUN_USER_FEATURES (NETIF_F_HW_CSUM|NETIF_F_TSO_ECN|NETIF_F_TSO| \
224 struct tap_filter txflt;
225 struct sock_fprog fprog;
226 /* protected by rtnl lock */
227 bool filter_attached;
232 struct hlist_head flows[TUN_NUM_FLOW_ENTRIES];
233 struct timer_list flow_gc_timer;
234 unsigned long ageing_time;
235 unsigned int numdisabled;
236 struct list_head disabled;
240 struct tun_pcpu_stats __percpu *pcpu_stats;
241 struct bpf_prog __rcu *xdp_prog;
242 struct tun_prog __rcu *steering_prog;
243 struct tun_prog __rcu *filter_prog;
251 bool tun_is_xdp_buff(void *ptr)
253 return (unsigned long)ptr & TUN_XDP_FLAG;
255 EXPORT_SYMBOL(tun_is_xdp_buff);
257 void *tun_xdp_to_ptr(void *ptr)
259 return (void *)((unsigned long)ptr | TUN_XDP_FLAG);
261 EXPORT_SYMBOL(tun_xdp_to_ptr);
263 void *tun_ptr_to_xdp(void *ptr)
265 return (void *)((unsigned long)ptr & ~TUN_XDP_FLAG);
267 EXPORT_SYMBOL(tun_ptr_to_xdp);
269 static int tun_napi_receive(struct napi_struct *napi, int budget)
271 struct tun_file *tfile = container_of(napi, struct tun_file, napi);
272 struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
273 struct sk_buff_head process_queue;
277 __skb_queue_head_init(&process_queue);
279 spin_lock(&queue->lock);
280 skb_queue_splice_tail_init(queue, &process_queue);
281 spin_unlock(&queue->lock);
283 while (received < budget && (skb = __skb_dequeue(&process_queue))) {
284 napi_gro_receive(napi, skb);
288 if (!skb_queue_empty(&process_queue)) {
289 spin_lock(&queue->lock);
290 skb_queue_splice(&process_queue, queue);
291 spin_unlock(&queue->lock);
297 static int tun_napi_poll(struct napi_struct *napi, int budget)
299 unsigned int received;
301 received = tun_napi_receive(napi, budget);
303 if (received < budget)
304 napi_complete_done(napi, received);
309 static void tun_napi_init(struct tun_struct *tun, struct tun_file *tfile,
312 tfile->napi_enabled = napi_en;
314 netif_napi_add(tun->dev, &tfile->napi, tun_napi_poll,
316 napi_enable(&tfile->napi);
317 mutex_init(&tfile->napi_mutex);
321 static void tun_napi_disable(struct tun_struct *tun, struct tun_file *tfile)
323 if (tfile->napi_enabled)
324 napi_disable(&tfile->napi);
327 static void tun_napi_del(struct tun_struct *tun, struct tun_file *tfile)
329 if (tfile->napi_enabled)
330 netif_napi_del(&tfile->napi);
333 static bool tun_napi_frags_enabled(const struct tun_struct *tun)
335 return READ_ONCE(tun->flags) & IFF_NAPI_FRAGS;
338 #ifdef CONFIG_TUN_VNET_CROSS_LE
339 static inline bool tun_legacy_is_little_endian(struct tun_struct *tun)
341 return tun->flags & TUN_VNET_BE ? false :
342 virtio_legacy_is_little_endian();
345 static long tun_get_vnet_be(struct tun_struct *tun, int __user *argp)
347 int be = !!(tun->flags & TUN_VNET_BE);
349 if (put_user(be, argp))
355 static long tun_set_vnet_be(struct tun_struct *tun, int __user *argp)
359 if (get_user(be, argp))
363 tun->flags |= TUN_VNET_BE;
365 tun->flags &= ~TUN_VNET_BE;
370 static inline bool tun_legacy_is_little_endian(struct tun_struct *tun)
372 return virtio_legacy_is_little_endian();
375 static long tun_get_vnet_be(struct tun_struct *tun, int __user *argp)
380 static long tun_set_vnet_be(struct tun_struct *tun, int __user *argp)
384 #endif /* CONFIG_TUN_VNET_CROSS_LE */
386 static inline bool tun_is_little_endian(struct tun_struct *tun)
388 return tun->flags & TUN_VNET_LE ||
389 tun_legacy_is_little_endian(tun);
392 static inline u16 tun16_to_cpu(struct tun_struct *tun, __virtio16 val)
394 return __virtio16_to_cpu(tun_is_little_endian(tun), val);
397 static inline __virtio16 cpu_to_tun16(struct tun_struct *tun, u16 val)
399 return __cpu_to_virtio16(tun_is_little_endian(tun), val);
402 static inline u32 tun_hashfn(u32 rxhash)
404 return rxhash & 0x3ff;
407 static struct tun_flow_entry *tun_flow_find(struct hlist_head *head, u32 rxhash)
409 struct tun_flow_entry *e;
411 hlist_for_each_entry_rcu(e, head, hash_link) {
412 if (e->rxhash == rxhash)
418 static struct tun_flow_entry *tun_flow_create(struct tun_struct *tun,
419 struct hlist_head *head,
420 u32 rxhash, u16 queue_index)
422 struct tun_flow_entry *e = kmalloc(sizeof(*e), GFP_ATOMIC);
425 tun_debug(KERN_INFO, tun, "create flow: hash %u index %u\n",
426 rxhash, queue_index);
427 e->updated = jiffies;
430 e->queue_index = queue_index;
432 hlist_add_head_rcu(&e->hash_link, head);
438 static void tun_flow_delete(struct tun_struct *tun, struct tun_flow_entry *e)
440 tun_debug(KERN_INFO, tun, "delete flow: hash %u index %u\n",
441 e->rxhash, e->queue_index);
442 hlist_del_rcu(&e->hash_link);
447 static void tun_flow_flush(struct tun_struct *tun)
451 spin_lock_bh(&tun->lock);
452 for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
453 struct tun_flow_entry *e;
454 struct hlist_node *n;
456 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link)
457 tun_flow_delete(tun, e);
459 spin_unlock_bh(&tun->lock);
462 static void tun_flow_delete_by_queue(struct tun_struct *tun, u16 queue_index)
466 spin_lock_bh(&tun->lock);
467 for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
468 struct tun_flow_entry *e;
469 struct hlist_node *n;
471 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
472 if (e->queue_index == queue_index)
473 tun_flow_delete(tun, e);
476 spin_unlock_bh(&tun->lock);
479 static void tun_flow_cleanup(struct timer_list *t)
481 struct tun_struct *tun = from_timer(tun, t, flow_gc_timer);
482 unsigned long delay = tun->ageing_time;
483 unsigned long next_timer = jiffies + delay;
484 unsigned long count = 0;
487 tun_debug(KERN_INFO, tun, "tun_flow_cleanup\n");
489 spin_lock(&tun->lock);
490 for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
491 struct tun_flow_entry *e;
492 struct hlist_node *n;
494 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
495 unsigned long this_timer;
497 this_timer = e->updated + delay;
498 if (time_before_eq(this_timer, jiffies)) {
499 tun_flow_delete(tun, e);
503 if (time_before(this_timer, next_timer))
504 next_timer = this_timer;
509 mod_timer(&tun->flow_gc_timer, round_jiffies_up(next_timer));
510 spin_unlock(&tun->lock);
513 static void tun_flow_update(struct tun_struct *tun, u32 rxhash,
514 struct tun_file *tfile)
516 struct hlist_head *head;
517 struct tun_flow_entry *e;
518 unsigned long delay = tun->ageing_time;
519 u16 queue_index = tfile->queue_index;
524 head = &tun->flows[tun_hashfn(rxhash)];
528 /* We may get a very small possibility of OOO during switching, not
529 * worth to optimize.*/
530 if (tun->numqueues == 1 || tfile->detached)
533 e = tun_flow_find(head, rxhash);
535 /* TODO: keep queueing to old queue until it's empty? */
536 e->queue_index = queue_index;
537 e->updated = jiffies;
538 sock_rps_record_flow_hash(e->rps_rxhash);
540 spin_lock_bh(&tun->lock);
541 if (!tun_flow_find(head, rxhash) &&
542 tun->flow_count < MAX_TAP_FLOWS)
543 tun_flow_create(tun, head, rxhash, queue_index);
545 if (!timer_pending(&tun->flow_gc_timer))
546 mod_timer(&tun->flow_gc_timer,
547 round_jiffies_up(jiffies + delay));
548 spin_unlock_bh(&tun->lock);
556 * Save the hash received in the stack receive path and update the
557 * flow_hash table accordingly.
559 static inline void tun_flow_save_rps_rxhash(struct tun_flow_entry *e, u32 hash)
561 if (unlikely(e->rps_rxhash != hash))
562 e->rps_rxhash = hash;
565 /* We try to identify a flow through its rxhash first. The reason that
566 * we do not check rxq no. is because some cards(e.g 82599), chooses
567 * the rxq based on the txq where the last packet of the flow comes. As
568 * the userspace application move between processors, we may get a
569 * different rxq no. here. If we could not get rxhash, then we would
570 * hope the rxq no. may help here.
572 static u16 tun_automq_select_queue(struct tun_struct *tun, struct sk_buff *skb)
574 struct tun_flow_entry *e;
578 numqueues = READ_ONCE(tun->numqueues);
580 txq = __skb_get_hash_symmetric(skb);
582 e = tun_flow_find(&tun->flows[tun_hashfn(txq)], txq);
584 tun_flow_save_rps_rxhash(e, txq);
585 txq = e->queue_index;
587 /* use multiply and shift instead of expensive divide */
588 txq = ((u64)txq * numqueues) >> 32;
589 } else if (likely(skb_rx_queue_recorded(skb))) {
590 txq = skb_get_rx_queue(skb);
591 while (unlikely(txq >= numqueues))
598 static u16 tun_ebpf_select_queue(struct tun_struct *tun, struct sk_buff *skb)
600 struct tun_prog *prog;
603 prog = rcu_dereference(tun->steering_prog);
605 ret = bpf_prog_run_clear_cb(prog->prog, skb);
607 return ret % tun->numqueues;
610 static u16 tun_select_queue(struct net_device *dev, struct sk_buff *skb,
611 void *accel_priv, select_queue_fallback_t fallback)
613 struct tun_struct *tun = netdev_priv(dev);
617 if (rcu_dereference(tun->steering_prog))
618 ret = tun_ebpf_select_queue(tun, skb);
620 ret = tun_automq_select_queue(tun, skb);
626 static inline bool tun_not_capable(struct tun_struct *tun)
628 const struct cred *cred = current_cred();
629 struct net *net = dev_net(tun->dev);
631 return ((uid_valid(tun->owner) && !uid_eq(cred->euid, tun->owner)) ||
632 (gid_valid(tun->group) && !in_egroup_p(tun->group))) &&
633 !ns_capable(net->user_ns, CAP_NET_ADMIN);
636 static void tun_set_real_num_queues(struct tun_struct *tun)
638 netif_set_real_num_tx_queues(tun->dev, tun->numqueues);
639 netif_set_real_num_rx_queues(tun->dev, tun->numqueues);
642 static void tun_disable_queue(struct tun_struct *tun, struct tun_file *tfile)
644 tfile->detached = tun;
645 list_add_tail(&tfile->next, &tun->disabled);
649 static struct tun_struct *tun_enable_queue(struct tun_file *tfile)
651 struct tun_struct *tun = tfile->detached;
653 tfile->detached = NULL;
654 list_del_init(&tfile->next);
659 void tun_ptr_free(void *ptr)
663 if (tun_is_xdp_buff(ptr)) {
664 struct xdp_buff *xdp = tun_ptr_to_xdp(ptr);
666 put_page(virt_to_head_page(xdp->data));
668 __skb_array_destroy_skb(ptr);
671 EXPORT_SYMBOL_GPL(tun_ptr_free);
673 static void tun_queue_purge(struct tun_file *tfile)
677 while ((ptr = ptr_ring_consume(&tfile->tx_ring)) != NULL)
680 skb_queue_purge(&tfile->sk.sk_write_queue);
681 skb_queue_purge(&tfile->sk.sk_error_queue);
684 static void tun_cleanup_tx_ring(struct tun_file *tfile)
686 if (tfile->tx_ring.queue) {
687 ptr_ring_cleanup(&tfile->tx_ring, tun_ptr_free);
688 xdp_rxq_info_unreg(&tfile->xdp_rxq);
689 memset(&tfile->tx_ring, 0, sizeof(tfile->tx_ring));
693 static void __tun_detach(struct tun_file *tfile, bool clean)
695 struct tun_file *ntfile;
696 struct tun_struct *tun;
698 tun = rtnl_dereference(tfile->tun);
701 tun_napi_disable(tun, tfile);
702 tun_napi_del(tun, tfile);
705 if (tun && !tfile->detached) {
706 u16 index = tfile->queue_index;
707 BUG_ON(index >= tun->numqueues);
709 rcu_assign_pointer(tun->tfiles[index],
710 tun->tfiles[tun->numqueues - 1]);
711 ntfile = rtnl_dereference(tun->tfiles[index]);
712 ntfile->queue_index = index;
716 RCU_INIT_POINTER(tfile->tun, NULL);
717 sock_put(&tfile->sk);
719 tun_disable_queue(tun, tfile);
722 tun_flow_delete_by_queue(tun, tun->numqueues + 1);
723 /* Drop read queue */
724 tun_queue_purge(tfile);
725 tun_set_real_num_queues(tun);
726 } else if (tfile->detached && clean) {
727 tun = tun_enable_queue(tfile);
728 sock_put(&tfile->sk);
732 if (tun && tun->numqueues == 0 && tun->numdisabled == 0) {
733 netif_carrier_off(tun->dev);
735 if (!(tun->flags & IFF_PERSIST) &&
736 tun->dev->reg_state == NETREG_REGISTERED)
737 unregister_netdevice(tun->dev);
739 tun_cleanup_tx_ring(tfile);
740 sock_put(&tfile->sk);
744 static void tun_detach(struct tun_file *tfile, bool clean)
747 __tun_detach(tfile, clean);
751 static void tun_detach_all(struct net_device *dev)
753 struct tun_struct *tun = netdev_priv(dev);
754 struct tun_file *tfile, *tmp;
755 int i, n = tun->numqueues;
757 for (i = 0; i < n; i++) {
758 tfile = rtnl_dereference(tun->tfiles[i]);
760 tun_napi_disable(tun, tfile);
761 tfile->socket.sk->sk_shutdown = RCV_SHUTDOWN;
762 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
763 RCU_INIT_POINTER(tfile->tun, NULL);
766 list_for_each_entry(tfile, &tun->disabled, next) {
767 tfile->socket.sk->sk_shutdown = RCV_SHUTDOWN;
768 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
769 RCU_INIT_POINTER(tfile->tun, NULL);
771 BUG_ON(tun->numqueues != 0);
774 for (i = 0; i < n; i++) {
775 tfile = rtnl_dereference(tun->tfiles[i]);
776 tun_napi_del(tun, tfile);
777 /* Drop read queue */
778 tun_queue_purge(tfile);
779 sock_put(&tfile->sk);
780 tun_cleanup_tx_ring(tfile);
782 list_for_each_entry_safe(tfile, tmp, &tun->disabled, next) {
783 tun_enable_queue(tfile);
784 tun_queue_purge(tfile);
785 sock_put(&tfile->sk);
786 tun_cleanup_tx_ring(tfile);
788 BUG_ON(tun->numdisabled != 0);
790 if (tun->flags & IFF_PERSIST)
791 module_put(THIS_MODULE);
794 static int tun_attach(struct tun_struct *tun, struct file *file,
795 bool skip_filter, bool napi)
797 struct tun_file *tfile = file->private_data;
798 struct net_device *dev = tun->dev;
801 err = security_tun_dev_attach(tfile->socket.sk, tun->security);
806 if (rtnl_dereference(tfile->tun) && !tfile->detached)
810 if (!(tun->flags & IFF_MULTI_QUEUE) && tun->numqueues == 1)
814 if (!tfile->detached &&
815 tun->numqueues + tun->numdisabled == MAX_TAP_QUEUES)
820 /* Re-attach the filter to persist device */
821 if (!skip_filter && (tun->filter_attached == true)) {
822 lock_sock(tfile->socket.sk);
823 err = sk_attach_filter(&tun->fprog, tfile->socket.sk);
824 release_sock(tfile->socket.sk);
829 if (!tfile->detached &&
830 ptr_ring_init(&tfile->tx_ring, dev->tx_queue_len, GFP_KERNEL)) {
835 tfile->queue_index = tun->numqueues;
836 tfile->socket.sk->sk_shutdown &= ~RCV_SHUTDOWN;
838 if (tfile->detached) {
839 /* Re-attach detached tfile, updating XDP queue_index */
840 WARN_ON(!xdp_rxq_info_is_reg(&tfile->xdp_rxq));
842 if (tfile->xdp_rxq.queue_index != tfile->queue_index)
843 tfile->xdp_rxq.queue_index = tfile->queue_index;
845 /* Setup XDP RX-queue info, for new tfile getting attached */
846 err = xdp_rxq_info_reg(&tfile->xdp_rxq,
847 tun->dev, tfile->queue_index);
853 rcu_assign_pointer(tfile->tun, tun);
854 rcu_assign_pointer(tun->tfiles[tun->numqueues], tfile);
857 if (tfile->detached) {
858 tun_enable_queue(tfile);
860 sock_hold(&tfile->sk);
861 tun_napi_init(tun, tfile, napi);
864 tun_set_real_num_queues(tun);
866 /* device is allowed to go away first, so no need to hold extra
874 static struct tun_struct *tun_get(struct tun_file *tfile)
876 struct tun_struct *tun;
879 tun = rcu_dereference(tfile->tun);
887 static void tun_put(struct tun_struct *tun)
893 static void addr_hash_set(u32 *mask, const u8 *addr)
895 int n = ether_crc(ETH_ALEN, addr) >> 26;
896 mask[n >> 5] |= (1 << (n & 31));
899 static unsigned int addr_hash_test(const u32 *mask, const u8 *addr)
901 int n = ether_crc(ETH_ALEN, addr) >> 26;
902 return mask[n >> 5] & (1 << (n & 31));
905 static int update_filter(struct tap_filter *filter, void __user *arg)
907 struct { u8 u[ETH_ALEN]; } *addr;
908 struct tun_filter uf;
909 int err, alen, n, nexact;
911 if (copy_from_user(&uf, arg, sizeof(uf)))
920 alen = ETH_ALEN * uf.count;
921 addr = memdup_user(arg + sizeof(uf), alen);
923 return PTR_ERR(addr);
925 /* The filter is updated without holding any locks. Which is
926 * perfectly safe. We disable it first and in the worst
927 * case we'll accept a few undesired packets. */
931 /* Use first set of addresses as an exact filter */
932 for (n = 0; n < uf.count && n < FLT_EXACT_COUNT; n++)
933 memcpy(filter->addr[n], addr[n].u, ETH_ALEN);
937 /* Remaining multicast addresses are hashed,
938 * unicast will leave the filter disabled. */
939 memset(filter->mask, 0, sizeof(filter->mask));
940 for (; n < uf.count; n++) {
941 if (!is_multicast_ether_addr(addr[n].u)) {
942 err = 0; /* no filter */
945 addr_hash_set(filter->mask, addr[n].u);
948 /* For ALLMULTI just set the mask to all ones.
949 * This overrides the mask populated above. */
950 if ((uf.flags & TUN_FLT_ALLMULTI))
951 memset(filter->mask, ~0, sizeof(filter->mask));
953 /* Now enable the filter */
955 filter->count = nexact;
957 /* Return the number of exact filters */
964 /* Returns: 0 - drop, !=0 - accept */
965 static int run_filter(struct tap_filter *filter, const struct sk_buff *skb)
967 /* Cannot use eth_hdr(skb) here because skb_mac_hdr() is incorrect
969 struct ethhdr *eh = (struct ethhdr *) skb->data;
973 for (i = 0; i < filter->count; i++)
974 if (ether_addr_equal(eh->h_dest, filter->addr[i]))
977 /* Inexact match (multicast only) */
978 if (is_multicast_ether_addr(eh->h_dest))
979 return addr_hash_test(filter->mask, eh->h_dest);
985 * Checks whether the packet is accepted or not.
986 * Returns: 0 - drop, !=0 - accept
988 static int check_filter(struct tap_filter *filter, const struct sk_buff *skb)
993 return run_filter(filter, skb);
996 /* Network device part of the driver */
998 static const struct ethtool_ops tun_ethtool_ops;
1000 /* Net device detach from fd. */
1001 static void tun_net_uninit(struct net_device *dev)
1003 tun_detach_all(dev);
1006 /* Net device open. */
1007 static int tun_net_open(struct net_device *dev)
1009 struct tun_struct *tun = netdev_priv(dev);
1012 netif_tx_start_all_queues(dev);
1014 for (i = 0; i < tun->numqueues; i++) {
1015 struct tun_file *tfile;
1017 tfile = rtnl_dereference(tun->tfiles[i]);
1018 tfile->socket.sk->sk_write_space(tfile->socket.sk);
1024 /* Net device close. */
1025 static int tun_net_close(struct net_device *dev)
1027 netif_tx_stop_all_queues(dev);
1031 /* Net device start xmit */
1032 static void tun_automq_xmit(struct tun_struct *tun, struct sk_buff *skb)
1035 if (tun->numqueues == 1 && static_key_false(&rps_needed)) {
1036 /* Select queue was not called for the skbuff, so we extract the
1037 * RPS hash and save it into the flow_table here.
1041 rxhash = __skb_get_hash_symmetric(skb);
1043 struct tun_flow_entry *e;
1044 e = tun_flow_find(&tun->flows[tun_hashfn(rxhash)],
1047 tun_flow_save_rps_rxhash(e, rxhash);
1053 static unsigned int run_ebpf_filter(struct tun_struct *tun,
1054 struct sk_buff *skb,
1057 struct tun_prog *prog = rcu_dereference(tun->filter_prog);
1060 len = bpf_prog_run_clear_cb(prog->prog, skb);
1065 /* Net device start xmit */
1066 static netdev_tx_t tun_net_xmit(struct sk_buff *skb, struct net_device *dev)
1068 struct tun_struct *tun = netdev_priv(dev);
1069 int txq = skb->queue_mapping;
1070 struct tun_file *tfile;
1074 tfile = rcu_dereference(tun->tfiles[txq]);
1076 /* Drop packet if interface is not attached */
1077 if (txq >= tun->numqueues)
1080 if (!rcu_dereference(tun->steering_prog))
1081 tun_automq_xmit(tun, skb);
1083 tun_debug(KERN_INFO, tun, "tun_net_xmit %d\n", skb->len);
1087 /* Drop if the filter does not like it.
1088 * This is a noop if the filter is disabled.
1089 * Filter can be enabled only for the TAP devices. */
1090 if (!check_filter(&tun->txflt, skb))
1093 if (tfile->socket.sk->sk_filter &&
1094 sk_filter(tfile->socket.sk, skb))
1097 len = run_ebpf_filter(tun, skb, len);
1099 /* Trim extra bytes since we may insert vlan proto & TCI
1100 * in tun_put_user().
1102 len -= skb_vlan_tag_present(skb) ? sizeof(struct veth) : 0;
1103 if (len <= 0 || pskb_trim(skb, len))
1106 if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC)))
1109 skb_tx_timestamp(skb);
1111 /* Orphan the skb - required as we might hang on to it
1112 * for indefinite time.
1118 if (ptr_ring_produce(&tfile->tx_ring, skb))
1121 /* Notify and wake up reader process */
1122 if (tfile->flags & TUN_FASYNC)
1123 kill_fasync(&tfile->fasync, SIGIO, POLL_IN);
1124 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
1127 return NETDEV_TX_OK;
1130 this_cpu_inc(tun->pcpu_stats->tx_dropped);
1134 return NET_XMIT_DROP;
1137 static void tun_net_mclist(struct net_device *dev)
1140 * This callback is supposed to deal with mc filter in
1141 * _rx_ path and has nothing to do with the _tx_ path.
1142 * In rx path we always accept everything userspace gives us.
1146 static netdev_features_t tun_net_fix_features(struct net_device *dev,
1147 netdev_features_t features)
1149 struct tun_struct *tun = netdev_priv(dev);
1151 return (features & tun->set_features) | (features & ~TUN_USER_FEATURES);
1153 #ifdef CONFIG_NET_POLL_CONTROLLER
1154 static void tun_poll_controller(struct net_device *dev)
1157 * Tun only receives frames when:
1158 * 1) the char device endpoint gets data from user space
1159 * 2) the tun socket gets a sendmsg call from user space
1160 * If NAPI is not enabled, since both of those are synchronous
1161 * operations, we are guaranteed never to have pending data when we poll
1162 * for it so there is nothing to do here but return.
1163 * We need this though so netpoll recognizes us as an interface that
1164 * supports polling, which enables bridge devices in virt setups to
1165 * still use netconsole
1166 * If NAPI is enabled, however, we need to schedule polling for all
1167 * queues unless we are using napi_gro_frags(), which we call in
1168 * process context and not in NAPI context.
1170 struct tun_struct *tun = netdev_priv(dev);
1172 if (tun->flags & IFF_NAPI) {
1173 struct tun_file *tfile;
1176 if (tun_napi_frags_enabled(tun))
1180 for (i = 0; i < tun->numqueues; i++) {
1181 tfile = rcu_dereference(tun->tfiles[i]);
1182 if (tfile->napi_enabled)
1183 napi_schedule(&tfile->napi);
1191 static void tun_set_headroom(struct net_device *dev, int new_hr)
1193 struct tun_struct *tun = netdev_priv(dev);
1195 if (new_hr < NET_SKB_PAD)
1196 new_hr = NET_SKB_PAD;
1198 tun->align = new_hr;
1202 tun_net_get_stats64(struct net_device *dev, struct rtnl_link_stats64 *stats)
1204 u32 rx_dropped = 0, tx_dropped = 0, rx_frame_errors = 0;
1205 struct tun_struct *tun = netdev_priv(dev);
1206 struct tun_pcpu_stats *p;
1209 for_each_possible_cpu(i) {
1210 u64 rxpackets, rxbytes, txpackets, txbytes;
1213 p = per_cpu_ptr(tun->pcpu_stats, i);
1215 start = u64_stats_fetch_begin(&p->syncp);
1216 rxpackets = p->rx_packets;
1217 rxbytes = p->rx_bytes;
1218 txpackets = p->tx_packets;
1219 txbytes = p->tx_bytes;
1220 } while (u64_stats_fetch_retry(&p->syncp, start));
1222 stats->rx_packets += rxpackets;
1223 stats->rx_bytes += rxbytes;
1224 stats->tx_packets += txpackets;
1225 stats->tx_bytes += txbytes;
1228 rx_dropped += p->rx_dropped;
1229 rx_frame_errors += p->rx_frame_errors;
1230 tx_dropped += p->tx_dropped;
1232 stats->rx_dropped = rx_dropped;
1233 stats->rx_frame_errors = rx_frame_errors;
1234 stats->tx_dropped = tx_dropped;
1237 static int tun_xdp_set(struct net_device *dev, struct bpf_prog *prog,
1238 struct netlink_ext_ack *extack)
1240 struct tun_struct *tun = netdev_priv(dev);
1241 struct bpf_prog *old_prog;
1243 old_prog = rtnl_dereference(tun->xdp_prog);
1244 rcu_assign_pointer(tun->xdp_prog, prog);
1246 bpf_prog_put(old_prog);
1251 static u32 tun_xdp_query(struct net_device *dev)
1253 struct tun_struct *tun = netdev_priv(dev);
1254 const struct bpf_prog *xdp_prog;
1256 xdp_prog = rtnl_dereference(tun->xdp_prog);
1258 return xdp_prog->aux->id;
1263 static int tun_xdp(struct net_device *dev, struct netdev_bpf *xdp)
1265 switch (xdp->command) {
1266 case XDP_SETUP_PROG:
1267 return tun_xdp_set(dev, xdp->prog, xdp->extack);
1268 case XDP_QUERY_PROG:
1269 xdp->prog_id = tun_xdp_query(dev);
1270 xdp->prog_attached = !!xdp->prog_id;
1277 static const struct net_device_ops tun_netdev_ops = {
1278 .ndo_uninit = tun_net_uninit,
1279 .ndo_open = tun_net_open,
1280 .ndo_stop = tun_net_close,
1281 .ndo_start_xmit = tun_net_xmit,
1282 .ndo_fix_features = tun_net_fix_features,
1283 .ndo_select_queue = tun_select_queue,
1284 #ifdef CONFIG_NET_POLL_CONTROLLER
1285 .ndo_poll_controller = tun_poll_controller,
1287 .ndo_set_rx_headroom = tun_set_headroom,
1288 .ndo_get_stats64 = tun_net_get_stats64,
1291 static int tun_xdp_xmit(struct net_device *dev, struct xdp_buff *xdp)
1293 struct tun_struct *tun = netdev_priv(dev);
1294 struct xdp_buff *buff = xdp->data_hard_start;
1295 int headroom = xdp->data - xdp->data_hard_start;
1296 struct tun_file *tfile;
1300 /* Assure headroom is available and buff is properly aligned */
1301 if (unlikely(headroom < sizeof(*xdp) || tun_is_xdp_buff(xdp)))
1308 numqueues = READ_ONCE(tun->numqueues);
1314 tfile = rcu_dereference(tun->tfiles[smp_processor_id() %
1316 /* Encode the XDP flag into lowest bit for consumer to differ
1317 * XDP buffer from sk_buff.
1319 if (ptr_ring_produce(&tfile->tx_ring, tun_xdp_to_ptr(buff))) {
1320 this_cpu_inc(tun->pcpu_stats->tx_dropped);
1329 static void tun_xdp_flush(struct net_device *dev)
1331 struct tun_struct *tun = netdev_priv(dev);
1332 struct tun_file *tfile;
1337 numqueues = READ_ONCE(tun->numqueues);
1341 tfile = rcu_dereference(tun->tfiles[smp_processor_id() %
1343 /* Notify and wake up reader process */
1344 if (tfile->flags & TUN_FASYNC)
1345 kill_fasync(&tfile->fasync, SIGIO, POLL_IN);
1346 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
1352 static const struct net_device_ops tap_netdev_ops = {
1353 .ndo_uninit = tun_net_uninit,
1354 .ndo_open = tun_net_open,
1355 .ndo_stop = tun_net_close,
1356 .ndo_start_xmit = tun_net_xmit,
1357 .ndo_fix_features = tun_net_fix_features,
1358 .ndo_set_rx_mode = tun_net_mclist,
1359 .ndo_set_mac_address = eth_mac_addr,
1360 .ndo_validate_addr = eth_validate_addr,
1361 .ndo_select_queue = tun_select_queue,
1362 #ifdef CONFIG_NET_POLL_CONTROLLER
1363 .ndo_poll_controller = tun_poll_controller,
1365 .ndo_features_check = passthru_features_check,
1366 .ndo_set_rx_headroom = tun_set_headroom,
1367 .ndo_get_stats64 = tun_net_get_stats64,
1369 .ndo_xdp_xmit = tun_xdp_xmit,
1370 .ndo_xdp_flush = tun_xdp_flush,
1373 static void tun_flow_init(struct tun_struct *tun)
1377 for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++)
1378 INIT_HLIST_HEAD(&tun->flows[i]);
1380 tun->ageing_time = TUN_FLOW_EXPIRE;
1381 timer_setup(&tun->flow_gc_timer, tun_flow_cleanup, 0);
1382 mod_timer(&tun->flow_gc_timer,
1383 round_jiffies_up(jiffies + tun->ageing_time));
1386 static void tun_flow_uninit(struct tun_struct *tun)
1388 del_timer_sync(&tun->flow_gc_timer);
1389 tun_flow_flush(tun);
1393 #define MAX_MTU 65535
1395 /* Initialize net device. */
1396 static void tun_net_init(struct net_device *dev)
1398 struct tun_struct *tun = netdev_priv(dev);
1400 switch (tun->flags & TUN_TYPE_MASK) {
1402 dev->netdev_ops = &tun_netdev_ops;
1404 /* Point-to-Point TUN Device */
1405 dev->hard_header_len = 0;
1409 /* Zero header length */
1410 dev->type = ARPHRD_NONE;
1411 dev->flags = IFF_POINTOPOINT | IFF_NOARP | IFF_MULTICAST;
1415 dev->netdev_ops = &tap_netdev_ops;
1416 /* Ethernet TAP Device */
1418 dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1419 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
1421 eth_hw_addr_random(dev);
1426 dev->min_mtu = MIN_MTU;
1427 dev->max_mtu = MAX_MTU - dev->hard_header_len;
1430 /* Character device part */
1433 static __poll_t tun_chr_poll(struct file *file, poll_table *wait)
1435 struct tun_file *tfile = file->private_data;
1436 struct tun_struct *tun = tun_get(tfile);
1443 sk = tfile->socket.sk;
1445 tun_debug(KERN_INFO, tun, "tun_chr_poll\n");
1447 poll_wait(file, sk_sleep(sk), wait);
1449 if (!ptr_ring_empty(&tfile->tx_ring))
1450 mask |= EPOLLIN | EPOLLRDNORM;
1452 if (tun->dev->flags & IFF_UP &&
1453 (sock_writeable(sk) ||
1454 (!test_and_set_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags) &&
1455 sock_writeable(sk))))
1456 mask |= EPOLLOUT | EPOLLWRNORM;
1458 if (tun->dev->reg_state != NETREG_REGISTERED)
1465 static struct sk_buff *tun_napi_alloc_frags(struct tun_file *tfile,
1467 const struct iov_iter *it)
1469 struct sk_buff *skb;
1474 if (it->nr_segs > MAX_SKB_FRAGS + 1)
1475 return ERR_PTR(-ENOMEM);
1478 skb = napi_get_frags(&tfile->napi);
1481 return ERR_PTR(-ENOMEM);
1483 linear = iov_iter_single_seg_count(it);
1484 err = __skb_grow(skb, linear);
1489 skb->data_len = len - linear;
1490 skb->truesize += skb->data_len;
1492 for (i = 1; i < it->nr_segs; i++) {
1493 struct page_frag *pfrag = ¤t->task_frag;
1494 size_t fragsz = it->iov[i].iov_len;
1496 if (fragsz == 0 || fragsz > PAGE_SIZE) {
1501 if (!skb_page_frag_refill(fragsz, pfrag, GFP_KERNEL)) {
1506 skb_fill_page_desc(skb, i - 1, pfrag->page,
1507 pfrag->offset, fragsz);
1508 page_ref_inc(pfrag->page);
1509 pfrag->offset += fragsz;
1514 /* frees skb and all frags allocated with napi_alloc_frag() */
1515 napi_free_frags(&tfile->napi);
1516 return ERR_PTR(err);
1519 /* prepad is the amount to reserve at front. len is length after that.
1520 * linear is a hint as to how much to copy (usually headers). */
1521 static struct sk_buff *tun_alloc_skb(struct tun_file *tfile,
1522 size_t prepad, size_t len,
1523 size_t linear, int noblock)
1525 struct sock *sk = tfile->socket.sk;
1526 struct sk_buff *skb;
1529 /* Under a page? Don't bother with paged skb. */
1530 if (prepad + len < PAGE_SIZE || !linear)
1533 skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
1536 return ERR_PTR(err);
1538 skb_reserve(skb, prepad);
1539 skb_put(skb, linear);
1540 skb->data_len = len - linear;
1541 skb->len += len - linear;
1546 static void tun_rx_batched(struct tun_struct *tun, struct tun_file *tfile,
1547 struct sk_buff *skb, int more)
1549 struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
1550 struct sk_buff_head process_queue;
1551 u32 rx_batched = tun->rx_batched;
1554 if (!rx_batched || (!more && skb_queue_empty(queue))) {
1556 netif_receive_skb(skb);
1561 spin_lock(&queue->lock);
1562 if (!more || skb_queue_len(queue) == rx_batched) {
1563 __skb_queue_head_init(&process_queue);
1564 skb_queue_splice_tail_init(queue, &process_queue);
1567 __skb_queue_tail(queue, skb);
1569 spin_unlock(&queue->lock);
1572 struct sk_buff *nskb;
1575 while ((nskb = __skb_dequeue(&process_queue)))
1576 netif_receive_skb(nskb);
1577 netif_receive_skb(skb);
1582 static bool tun_can_build_skb(struct tun_struct *tun, struct tun_file *tfile,
1583 int len, int noblock, bool zerocopy)
1585 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
1588 if (tfile->socket.sk->sk_sndbuf != INT_MAX)
1597 if (SKB_DATA_ALIGN(len + TUN_RX_PAD) +
1598 SKB_DATA_ALIGN(sizeof(struct skb_shared_info)) > PAGE_SIZE)
1604 static struct sk_buff *tun_build_skb(struct tun_struct *tun,
1605 struct tun_file *tfile,
1606 struct iov_iter *from,
1607 struct virtio_net_hdr *hdr,
1608 int len, int *skb_xdp)
1610 struct page_frag *alloc_frag = ¤t->task_frag;
1611 struct sk_buff *skb;
1612 struct bpf_prog *xdp_prog;
1613 int buflen = SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
1614 unsigned int delta = 0;
1617 int err, pad = TUN_RX_PAD;
1620 xdp_prog = rcu_dereference(tun->xdp_prog);
1622 pad += TUN_HEADROOM;
1623 buflen += SKB_DATA_ALIGN(len + pad);
1626 alloc_frag->offset = ALIGN((u64)alloc_frag->offset, SMP_CACHE_BYTES);
1627 if (unlikely(!skb_page_frag_refill(buflen, alloc_frag, GFP_KERNEL)))
1628 return ERR_PTR(-ENOMEM);
1630 buf = (char *)page_address(alloc_frag->page) + alloc_frag->offset;
1631 copied = copy_page_from_iter(alloc_frag->page,
1632 alloc_frag->offset + pad,
1635 return ERR_PTR(-EFAULT);
1637 /* There's a small window that XDP may be set after the check
1638 * of xdp_prog above, this should be rare and for simplicity
1639 * we do XDP on skb in case the headroom is not enough.
1641 if (hdr->gso_type || !xdp_prog)
1648 xdp_prog = rcu_dereference(tun->xdp_prog);
1649 if (xdp_prog && !*skb_xdp) {
1650 struct xdp_buff xdp;
1654 xdp.data_hard_start = buf;
1655 xdp.data = buf + pad;
1656 xdp_set_data_meta_invalid(&xdp);
1657 xdp.data_end = xdp.data + len;
1658 xdp.rxq = &tfile->xdp_rxq;
1659 orig_data = xdp.data;
1660 act = bpf_prog_run_xdp(xdp_prog, &xdp);
1664 get_page(alloc_frag->page);
1665 alloc_frag->offset += buflen;
1666 err = xdp_do_redirect(tun->dev, &xdp, xdp_prog);
1674 get_page(alloc_frag->page);
1675 alloc_frag->offset += buflen;
1676 if (tun_xdp_xmit(tun->dev, &xdp))
1678 tun_xdp_flush(tun->dev);
1683 delta = orig_data - xdp.data;
1686 bpf_warn_invalid_xdp_action(act);
1689 trace_xdp_exception(tun->dev, xdp_prog, act);
1696 skb = build_skb(buf, buflen);
1700 return ERR_PTR(-ENOMEM);
1703 skb_reserve(skb, pad - delta);
1704 skb_put(skb, len + delta);
1705 get_page(alloc_frag->page);
1706 alloc_frag->offset += buflen;
1714 put_page(alloc_frag->page);
1718 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1722 /* Get packet from user space buffer */
1723 static ssize_t tun_get_user(struct tun_struct *tun, struct tun_file *tfile,
1724 void *msg_control, struct iov_iter *from,
1725 int noblock, bool more)
1727 struct tun_pi pi = { 0, cpu_to_be16(ETH_P_IP) };
1728 struct sk_buff *skb;
1729 size_t total_len = iov_iter_count(from);
1730 size_t len = total_len, align = tun->align, linear;
1731 struct virtio_net_hdr gso = { 0 };
1732 struct tun_pcpu_stats *stats;
1735 bool zerocopy = false;
1739 bool frags = tun_napi_frags_enabled(tun);
1741 if (!(tun->dev->flags & IFF_UP))
1744 if (!(tun->flags & IFF_NO_PI)) {
1745 if (len < sizeof(pi))
1749 if (!copy_from_iter_full(&pi, sizeof(pi), from))
1753 if (tun->flags & IFF_VNET_HDR) {
1754 int vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
1756 if (len < vnet_hdr_sz)
1760 if (!copy_from_iter_full(&gso, sizeof(gso), from))
1763 if ((gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
1764 tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2 > tun16_to_cpu(tun, gso.hdr_len))
1765 gso.hdr_len = cpu_to_tun16(tun, tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2);
1767 if (tun16_to_cpu(tun, gso.hdr_len) > len)
1769 iov_iter_advance(from, vnet_hdr_sz - sizeof(gso));
1772 if ((tun->flags & TUN_TYPE_MASK) == IFF_TAP) {
1773 align += NET_IP_ALIGN;
1774 if (unlikely(len < ETH_HLEN ||
1775 (gso.hdr_len && tun16_to_cpu(tun, gso.hdr_len) < ETH_HLEN)))
1779 good_linear = SKB_MAX_HEAD(align);
1782 struct iov_iter i = *from;
1784 /* There are 256 bytes to be copied in skb, so there is
1785 * enough room for skb expand head in case it is used.
1786 * The rest of the buffer is mapped from userspace.
1788 copylen = gso.hdr_len ? tun16_to_cpu(tun, gso.hdr_len) : GOODCOPY_LEN;
1789 if (copylen > good_linear)
1790 copylen = good_linear;
1792 iov_iter_advance(&i, copylen);
1793 if (iov_iter_npages(&i, INT_MAX) <= MAX_SKB_FRAGS)
1797 if (!frags && tun_can_build_skb(tun, tfile, len, noblock, zerocopy)) {
1798 /* For the packet that is not easy to be processed
1799 * (e.g gso or jumbo packet), we will do it at after
1800 * skb was created with generic XDP routine.
1802 skb = tun_build_skb(tun, tfile, from, &gso, len, &skb_xdp);
1804 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1805 return PTR_ERR(skb);
1812 if (tun16_to_cpu(tun, gso.hdr_len) > good_linear)
1813 linear = good_linear;
1815 linear = tun16_to_cpu(tun, gso.hdr_len);
1819 mutex_lock(&tfile->napi_mutex);
1820 skb = tun_napi_alloc_frags(tfile, copylen, from);
1821 /* tun_napi_alloc_frags() enforces a layout for the skb.
1822 * If zerocopy is enabled, then this layout will be
1823 * overwritten by zerocopy_sg_from_iter().
1827 skb = tun_alloc_skb(tfile, align, copylen, linear,
1832 if (PTR_ERR(skb) != -EAGAIN)
1833 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1835 mutex_unlock(&tfile->napi_mutex);
1836 return PTR_ERR(skb);
1840 err = zerocopy_sg_from_iter(skb, from);
1842 err = skb_copy_datagram_from_iter(skb, 0, from, len);
1845 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1848 tfile->napi.skb = NULL;
1849 mutex_unlock(&tfile->napi_mutex);
1856 if (virtio_net_hdr_to_skb(skb, &gso, tun_is_little_endian(tun))) {
1857 this_cpu_inc(tun->pcpu_stats->rx_frame_errors);
1860 tfile->napi.skb = NULL;
1861 mutex_unlock(&tfile->napi_mutex);
1867 switch (tun->flags & TUN_TYPE_MASK) {
1869 if (tun->flags & IFF_NO_PI) {
1870 u8 ip_version = skb->len ? (skb->data[0] >> 4) : 0;
1872 switch (ip_version) {
1874 pi.proto = htons(ETH_P_IP);
1877 pi.proto = htons(ETH_P_IPV6);
1880 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1886 skb_reset_mac_header(skb);
1887 skb->protocol = pi.proto;
1888 skb->dev = tun->dev;
1892 skb->protocol = eth_type_trans(skb, tun->dev);
1896 /* copy skb_ubuf_info for callback when skb has no error */
1898 skb_shinfo(skb)->destructor_arg = msg_control;
1899 skb_shinfo(skb)->tx_flags |= SKBTX_DEV_ZEROCOPY;
1900 skb_shinfo(skb)->tx_flags |= SKBTX_SHARED_FRAG;
1901 } else if (msg_control) {
1902 struct ubuf_info *uarg = msg_control;
1903 uarg->callback(uarg, false);
1906 skb_reset_network_header(skb);
1907 skb_probe_transport_header(skb, 0);
1910 struct bpf_prog *xdp_prog;
1914 xdp_prog = rcu_dereference(tun->xdp_prog);
1916 ret = do_xdp_generic(xdp_prog, skb);
1917 if (ret != XDP_PASS) {
1926 if (!rcu_dereference(tun->steering_prog))
1927 rxhash = __skb_get_hash_symmetric(skb);
1931 /* Exercise flow dissector code path. */
1932 u32 headlen = eth_get_headlen(skb->data, skb_headlen(skb));
1934 if (unlikely(headlen > skb_headlen(skb))) {
1935 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1936 napi_free_frags(&tfile->napi);
1937 mutex_unlock(&tfile->napi_mutex);
1943 napi_gro_frags(&tfile->napi);
1945 mutex_unlock(&tfile->napi_mutex);
1946 } else if (tfile->napi_enabled) {
1947 struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
1950 spin_lock_bh(&queue->lock);
1951 __skb_queue_tail(queue, skb);
1952 queue_len = skb_queue_len(queue);
1953 spin_unlock(&queue->lock);
1955 if (!more || queue_len > NAPI_POLL_WEIGHT)
1956 napi_schedule(&tfile->napi);
1959 } else if (!IS_ENABLED(CONFIG_4KSTACKS)) {
1960 tun_rx_batched(tun, tfile, skb, more);
1965 stats = get_cpu_ptr(tun->pcpu_stats);
1966 u64_stats_update_begin(&stats->syncp);
1967 stats->rx_packets++;
1968 stats->rx_bytes += len;
1969 u64_stats_update_end(&stats->syncp);
1973 tun_flow_update(tun, rxhash, tfile);
1978 static ssize_t tun_chr_write_iter(struct kiocb *iocb, struct iov_iter *from)
1980 struct file *file = iocb->ki_filp;
1981 struct tun_file *tfile = file->private_data;
1982 struct tun_struct *tun = tun_get(tfile);
1988 result = tun_get_user(tun, tfile, NULL, from,
1989 file->f_flags & O_NONBLOCK, false);
1995 static ssize_t tun_put_user_xdp(struct tun_struct *tun,
1996 struct tun_file *tfile,
1997 struct xdp_buff *xdp,
1998 struct iov_iter *iter)
2000 int vnet_hdr_sz = 0;
2001 size_t size = xdp->data_end - xdp->data;
2002 struct tun_pcpu_stats *stats;
2005 if (tun->flags & IFF_VNET_HDR) {
2006 struct virtio_net_hdr gso = { 0 };
2008 vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
2009 if (unlikely(iov_iter_count(iter) < vnet_hdr_sz))
2011 if (unlikely(copy_to_iter(&gso, sizeof(gso), iter) !=
2014 iov_iter_advance(iter, vnet_hdr_sz - sizeof(gso));
2017 ret = copy_to_iter(xdp->data, size, iter) + vnet_hdr_sz;
2019 stats = get_cpu_ptr(tun->pcpu_stats);
2020 u64_stats_update_begin(&stats->syncp);
2021 stats->tx_packets++;
2022 stats->tx_bytes += ret;
2023 u64_stats_update_end(&stats->syncp);
2024 put_cpu_ptr(tun->pcpu_stats);
2029 /* Put packet to the user space buffer */
2030 static ssize_t tun_put_user(struct tun_struct *tun,
2031 struct tun_file *tfile,
2032 struct sk_buff *skb,
2033 struct iov_iter *iter)
2035 struct tun_pi pi = { 0, skb->protocol };
2036 struct tun_pcpu_stats *stats;
2038 int vlan_offset = 0;
2040 int vnet_hdr_sz = 0;
2042 if (skb_vlan_tag_present(skb))
2043 vlan_hlen = VLAN_HLEN;
2045 if (tun->flags & IFF_VNET_HDR)
2046 vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
2048 total = skb->len + vlan_hlen + vnet_hdr_sz;
2050 if (!(tun->flags & IFF_NO_PI)) {
2051 if (iov_iter_count(iter) < sizeof(pi))
2054 total += sizeof(pi);
2055 if (iov_iter_count(iter) < total) {
2056 /* Packet will be striped */
2057 pi.flags |= TUN_PKT_STRIP;
2060 if (copy_to_iter(&pi, sizeof(pi), iter) != sizeof(pi))
2065 struct virtio_net_hdr gso;
2067 if (iov_iter_count(iter) < vnet_hdr_sz)
2070 if (virtio_net_hdr_from_skb(skb, &gso,
2071 tun_is_little_endian(tun), true)) {
2072 struct skb_shared_info *sinfo = skb_shinfo(skb);
2073 pr_err("unexpected GSO type: "
2074 "0x%x, gso_size %d, hdr_len %d\n",
2075 sinfo->gso_type, tun16_to_cpu(tun, gso.gso_size),
2076 tun16_to_cpu(tun, gso.hdr_len));
2077 print_hex_dump(KERN_ERR, "tun: ",
2080 min((int)tun16_to_cpu(tun, gso.hdr_len), 64), true);
2085 if (copy_to_iter(&gso, sizeof(gso), iter) != sizeof(gso))
2088 iov_iter_advance(iter, vnet_hdr_sz - sizeof(gso));
2095 veth.h_vlan_proto = skb->vlan_proto;
2096 veth.h_vlan_TCI = htons(skb_vlan_tag_get(skb));
2098 vlan_offset = offsetof(struct vlan_ethhdr, h_vlan_proto);
2100 ret = skb_copy_datagram_iter(skb, 0, iter, vlan_offset);
2101 if (ret || !iov_iter_count(iter))
2104 ret = copy_to_iter(&veth, sizeof(veth), iter);
2105 if (ret != sizeof(veth) || !iov_iter_count(iter))
2109 skb_copy_datagram_iter(skb, vlan_offset, iter, skb->len - vlan_offset);
2112 /* caller is in process context, */
2113 stats = get_cpu_ptr(tun->pcpu_stats);
2114 u64_stats_update_begin(&stats->syncp);
2115 stats->tx_packets++;
2116 stats->tx_bytes += skb->len + vlan_hlen;
2117 u64_stats_update_end(&stats->syncp);
2118 put_cpu_ptr(tun->pcpu_stats);
2123 static void *tun_ring_recv(struct tun_file *tfile, int noblock, int *err)
2125 DECLARE_WAITQUEUE(wait, current);
2129 ptr = ptr_ring_consume(&tfile->tx_ring);
2137 add_wait_queue(&tfile->wq.wait, &wait);
2138 current->state = TASK_INTERRUPTIBLE;
2141 ptr = ptr_ring_consume(&tfile->tx_ring);
2144 if (signal_pending(current)) {
2145 error = -ERESTARTSYS;
2148 if (tfile->socket.sk->sk_shutdown & RCV_SHUTDOWN) {
2156 current->state = TASK_RUNNING;
2157 remove_wait_queue(&tfile->wq.wait, &wait);
2164 static ssize_t tun_do_read(struct tun_struct *tun, struct tun_file *tfile,
2165 struct iov_iter *to,
2166 int noblock, void *ptr)
2171 tun_debug(KERN_INFO, tun, "tun_do_read\n");
2173 if (!iov_iter_count(to)) {
2179 /* Read frames from ring */
2180 ptr = tun_ring_recv(tfile, noblock, &err);
2185 if (tun_is_xdp_buff(ptr)) {
2186 struct xdp_buff *xdp = tun_ptr_to_xdp(ptr);
2188 ret = tun_put_user_xdp(tun, tfile, xdp, to);
2189 put_page(virt_to_head_page(xdp->data));
2191 struct sk_buff *skb = ptr;
2193 ret = tun_put_user(tun, tfile, skb, to);
2194 if (unlikely(ret < 0))
2203 static ssize_t tun_chr_read_iter(struct kiocb *iocb, struct iov_iter *to)
2205 struct file *file = iocb->ki_filp;
2206 struct tun_file *tfile = file->private_data;
2207 struct tun_struct *tun = tun_get(tfile);
2208 ssize_t len = iov_iter_count(to), ret;
2212 ret = tun_do_read(tun, tfile, to, file->f_flags & O_NONBLOCK, NULL);
2213 ret = min_t(ssize_t, ret, len);
2220 static void tun_prog_free(struct rcu_head *rcu)
2222 struct tun_prog *prog = container_of(rcu, struct tun_prog, rcu);
2224 bpf_prog_destroy(prog->prog);
2228 static int __tun_set_ebpf(struct tun_struct *tun,
2229 struct tun_prog __rcu **prog_p,
2230 struct bpf_prog *prog)
2232 struct tun_prog *old, *new = NULL;
2235 new = kmalloc(sizeof(*new), GFP_KERNEL);
2241 spin_lock_bh(&tun->lock);
2242 old = rcu_dereference_protected(*prog_p,
2243 lockdep_is_held(&tun->lock));
2244 rcu_assign_pointer(*prog_p, new);
2245 spin_unlock_bh(&tun->lock);
2248 call_rcu(&old->rcu, tun_prog_free);
2253 static void tun_free_netdev(struct net_device *dev)
2255 struct tun_struct *tun = netdev_priv(dev);
2257 BUG_ON(!(list_empty(&tun->disabled)));
2258 free_percpu(tun->pcpu_stats);
2259 tun_flow_uninit(tun);
2260 security_tun_dev_free_security(tun->security);
2261 __tun_set_ebpf(tun, &tun->steering_prog, NULL);
2262 __tun_set_ebpf(tun, &tun->filter_prog, NULL);
2265 static void tun_setup(struct net_device *dev)
2267 struct tun_struct *tun = netdev_priv(dev);
2269 tun->owner = INVALID_UID;
2270 tun->group = INVALID_GID;
2272 dev->ethtool_ops = &tun_ethtool_ops;
2273 dev->needs_free_netdev = true;
2274 dev->priv_destructor = tun_free_netdev;
2275 /* We prefer our own queue length */
2276 dev->tx_queue_len = TUN_READQ_SIZE;
2279 /* Trivial set of netlink ops to allow deleting tun or tap
2280 * device with netlink.
2282 static int tun_validate(struct nlattr *tb[], struct nlattr *data[],
2283 struct netlink_ext_ack *extack)
2288 static size_t tun_get_size(const struct net_device *dev)
2290 BUILD_BUG_ON(sizeof(u32) != sizeof(uid_t));
2291 BUILD_BUG_ON(sizeof(u32) != sizeof(gid_t));
2293 return nla_total_size(sizeof(uid_t)) + /* OWNER */
2294 nla_total_size(sizeof(gid_t)) + /* GROUP */
2295 nla_total_size(sizeof(u8)) + /* TYPE */
2296 nla_total_size(sizeof(u8)) + /* PI */
2297 nla_total_size(sizeof(u8)) + /* VNET_HDR */
2298 nla_total_size(sizeof(u8)) + /* PERSIST */
2299 nla_total_size(sizeof(u8)) + /* MULTI_QUEUE */
2300 nla_total_size(sizeof(u32)) + /* NUM_QUEUES */
2301 nla_total_size(sizeof(u32)) + /* NUM_DISABLED_QUEUES */
2305 static int tun_fill_info(struct sk_buff *skb, const struct net_device *dev)
2307 struct tun_struct *tun = netdev_priv(dev);
2309 if (nla_put_u8(skb, IFLA_TUN_TYPE, tun->flags & TUN_TYPE_MASK))
2310 goto nla_put_failure;
2311 if (uid_valid(tun->owner) &&
2312 nla_put_u32(skb, IFLA_TUN_OWNER,
2313 from_kuid_munged(current_user_ns(), tun->owner)))
2314 goto nla_put_failure;
2315 if (gid_valid(tun->group) &&
2316 nla_put_u32(skb, IFLA_TUN_GROUP,
2317 from_kgid_munged(current_user_ns(), tun->group)))
2318 goto nla_put_failure;
2319 if (nla_put_u8(skb, IFLA_TUN_PI, !(tun->flags & IFF_NO_PI)))
2320 goto nla_put_failure;
2321 if (nla_put_u8(skb, IFLA_TUN_VNET_HDR, !!(tun->flags & IFF_VNET_HDR)))
2322 goto nla_put_failure;
2323 if (nla_put_u8(skb, IFLA_TUN_PERSIST, !!(tun->flags & IFF_PERSIST)))
2324 goto nla_put_failure;
2325 if (nla_put_u8(skb, IFLA_TUN_MULTI_QUEUE,
2326 !!(tun->flags & IFF_MULTI_QUEUE)))
2327 goto nla_put_failure;
2328 if (tun->flags & IFF_MULTI_QUEUE) {
2329 if (nla_put_u32(skb, IFLA_TUN_NUM_QUEUES, tun->numqueues))
2330 goto nla_put_failure;
2331 if (nla_put_u32(skb, IFLA_TUN_NUM_DISABLED_QUEUES,
2333 goto nla_put_failure;
2342 static struct rtnl_link_ops tun_link_ops __read_mostly = {
2344 .priv_size = sizeof(struct tun_struct),
2346 .validate = tun_validate,
2347 .get_size = tun_get_size,
2348 .fill_info = tun_fill_info,
2351 static void tun_sock_write_space(struct sock *sk)
2353 struct tun_file *tfile;
2354 wait_queue_head_t *wqueue;
2356 if (!sock_writeable(sk))
2359 if (!test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags))
2362 wqueue = sk_sleep(sk);
2363 if (wqueue && waitqueue_active(wqueue))
2364 wake_up_interruptible_sync_poll(wqueue, EPOLLOUT |
2365 EPOLLWRNORM | EPOLLWRBAND);
2367 tfile = container_of(sk, struct tun_file, sk);
2368 kill_fasync(&tfile->fasync, SIGIO, POLL_OUT);
2371 static int tun_sendmsg(struct socket *sock, struct msghdr *m, size_t total_len)
2374 struct tun_file *tfile = container_of(sock, struct tun_file, socket);
2375 struct tun_struct *tun = tun_get(tfile);
2380 ret = tun_get_user(tun, tfile, m->msg_control, &m->msg_iter,
2381 m->msg_flags & MSG_DONTWAIT,
2382 m->msg_flags & MSG_MORE);
2387 static int tun_recvmsg(struct socket *sock, struct msghdr *m, size_t total_len,
2390 struct tun_file *tfile = container_of(sock, struct tun_file, socket);
2391 struct tun_struct *tun = tun_get(tfile);
2392 void *ptr = m->msg_control;
2400 if (flags & ~(MSG_DONTWAIT|MSG_TRUNC|MSG_ERRQUEUE)) {
2404 if (flags & MSG_ERRQUEUE) {
2405 ret = sock_recv_errqueue(sock->sk, m, total_len,
2406 SOL_PACKET, TUN_TX_TIMESTAMP);
2409 ret = tun_do_read(tun, tfile, &m->msg_iter, flags & MSG_DONTWAIT, ptr);
2410 if (ret > (ssize_t)total_len) {
2411 m->msg_flags |= MSG_TRUNC;
2412 ret = flags & MSG_TRUNC ? ret : total_len;
2425 static int tun_ptr_peek_len(void *ptr)
2428 if (tun_is_xdp_buff(ptr)) {
2429 struct xdp_buff *xdp = tun_ptr_to_xdp(ptr);
2431 return xdp->data_end - xdp->data;
2433 return __skb_array_len_with_tag(ptr);
2439 static int tun_peek_len(struct socket *sock)
2441 struct tun_file *tfile = container_of(sock, struct tun_file, socket);
2442 struct tun_struct *tun;
2445 tun = tun_get(tfile);
2449 ret = PTR_RING_PEEK_CALL(&tfile->tx_ring, tun_ptr_peek_len);
2455 /* Ops structure to mimic raw sockets with tun */
2456 static const struct proto_ops tun_socket_ops = {
2457 .peek_len = tun_peek_len,
2458 .sendmsg = tun_sendmsg,
2459 .recvmsg = tun_recvmsg,
2462 static struct proto tun_proto = {
2464 .owner = THIS_MODULE,
2465 .obj_size = sizeof(struct tun_file),
2468 static int tun_flags(struct tun_struct *tun)
2470 return tun->flags & (TUN_FEATURES | IFF_PERSIST | IFF_TUN | IFF_TAP);
2473 static ssize_t tun_show_flags(struct device *dev, struct device_attribute *attr,
2476 struct tun_struct *tun = netdev_priv(to_net_dev(dev));
2477 return sprintf(buf, "0x%x\n", tun_flags(tun));
2480 static ssize_t tun_show_owner(struct device *dev, struct device_attribute *attr,
2483 struct tun_struct *tun = netdev_priv(to_net_dev(dev));
2484 return uid_valid(tun->owner)?
2485 sprintf(buf, "%u\n",
2486 from_kuid_munged(current_user_ns(), tun->owner)):
2487 sprintf(buf, "-1\n");
2490 static ssize_t tun_show_group(struct device *dev, struct device_attribute *attr,
2493 struct tun_struct *tun = netdev_priv(to_net_dev(dev));
2494 return gid_valid(tun->group) ?
2495 sprintf(buf, "%u\n",
2496 from_kgid_munged(current_user_ns(), tun->group)):
2497 sprintf(buf, "-1\n");
2500 static DEVICE_ATTR(tun_flags, 0444, tun_show_flags, NULL);
2501 static DEVICE_ATTR(owner, 0444, tun_show_owner, NULL);
2502 static DEVICE_ATTR(group, 0444, tun_show_group, NULL);
2504 static struct attribute *tun_dev_attrs[] = {
2505 &dev_attr_tun_flags.attr,
2506 &dev_attr_owner.attr,
2507 &dev_attr_group.attr,
2511 static const struct attribute_group tun_attr_group = {
2512 .attrs = tun_dev_attrs
2515 static int tun_set_iff(struct net *net, struct file *file, struct ifreq *ifr)
2517 struct tun_struct *tun;
2518 struct tun_file *tfile = file->private_data;
2519 struct net_device *dev;
2522 if (tfile->detached)
2525 if ((ifr->ifr_flags & IFF_NAPI_FRAGS)) {
2526 if (!capable(CAP_NET_ADMIN))
2529 if (!(ifr->ifr_flags & IFF_NAPI) ||
2530 (ifr->ifr_flags & TUN_TYPE_MASK) != IFF_TAP)
2534 dev = __dev_get_by_name(net, ifr->ifr_name);
2536 if (ifr->ifr_flags & IFF_TUN_EXCL)
2538 if ((ifr->ifr_flags & IFF_TUN) && dev->netdev_ops == &tun_netdev_ops)
2539 tun = netdev_priv(dev);
2540 else if ((ifr->ifr_flags & IFF_TAP) && dev->netdev_ops == &tap_netdev_ops)
2541 tun = netdev_priv(dev);
2545 if (!!(ifr->ifr_flags & IFF_MULTI_QUEUE) !=
2546 !!(tun->flags & IFF_MULTI_QUEUE))
2549 if (tun_not_capable(tun))
2551 err = security_tun_dev_open(tun->security);
2555 err = tun_attach(tun, file, ifr->ifr_flags & IFF_NOFILTER,
2556 ifr->ifr_flags & IFF_NAPI);
2560 if (tun->flags & IFF_MULTI_QUEUE &&
2561 (tun->numqueues + tun->numdisabled > 1)) {
2562 /* One or more queue has already been attached, no need
2563 * to initialize the device again.
2570 unsigned long flags = 0;
2571 int queues = ifr->ifr_flags & IFF_MULTI_QUEUE ?
2574 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
2576 err = security_tun_dev_create();
2581 if (ifr->ifr_flags & IFF_TUN) {
2585 } else if (ifr->ifr_flags & IFF_TAP) {
2593 name = ifr->ifr_name;
2595 dev = alloc_netdev_mqs(sizeof(struct tun_struct), name,
2596 NET_NAME_UNKNOWN, tun_setup, queues,
2601 err = dev_get_valid_name(net, dev, name);
2605 dev_net_set(dev, net);
2606 dev->rtnl_link_ops = &tun_link_ops;
2607 dev->ifindex = tfile->ifindex;
2608 dev->sysfs_groups[0] = &tun_attr_group;
2610 tun = netdev_priv(dev);
2613 tun->txflt.count = 0;
2614 tun->vnet_hdr_sz = sizeof(struct virtio_net_hdr);
2616 tun->align = NET_SKB_PAD;
2617 tun->filter_attached = false;
2618 tun->sndbuf = tfile->socket.sk->sk_sndbuf;
2619 tun->rx_batched = 0;
2620 RCU_INIT_POINTER(tun->steering_prog, NULL);
2622 tun->pcpu_stats = netdev_alloc_pcpu_stats(struct tun_pcpu_stats);
2623 if (!tun->pcpu_stats) {
2628 spin_lock_init(&tun->lock);
2630 err = security_tun_dev_alloc_security(&tun->security);
2637 dev->hw_features = NETIF_F_SG | NETIF_F_FRAGLIST |
2638 TUN_USER_FEATURES | NETIF_F_HW_VLAN_CTAG_TX |
2639 NETIF_F_HW_VLAN_STAG_TX;
2640 dev->features = dev->hw_features | NETIF_F_LLTX;
2641 dev->vlan_features = dev->features &
2642 ~(NETIF_F_HW_VLAN_CTAG_TX |
2643 NETIF_F_HW_VLAN_STAG_TX);
2645 INIT_LIST_HEAD(&tun->disabled);
2646 err = tun_attach(tun, file, false, ifr->ifr_flags & IFF_NAPI);
2650 err = register_netdevice(tun->dev);
2655 netif_carrier_on(tun->dev);
2657 tun_debug(KERN_INFO, tun, "tun_set_iff\n");
2659 tun->flags = (tun->flags & ~TUN_FEATURES) |
2660 (ifr->ifr_flags & TUN_FEATURES);
2662 /* Make sure persistent devices do not get stuck in
2665 if (netif_running(tun->dev))
2666 netif_tx_wake_all_queues(tun->dev);
2668 strcpy(ifr->ifr_name, tun->dev->name);
2672 tun_detach_all(dev);
2673 /* register_netdevice() already called tun_free_netdev() */
2677 tun_flow_uninit(tun);
2678 security_tun_dev_free_security(tun->security);
2680 free_percpu(tun->pcpu_stats);
2686 static void tun_get_iff(struct net *net, struct tun_struct *tun,
2689 tun_debug(KERN_INFO, tun, "tun_get_iff\n");
2691 strcpy(ifr->ifr_name, tun->dev->name);
2693 ifr->ifr_flags = tun_flags(tun);
2697 /* This is like a cut-down ethtool ops, except done via tun fd so no
2698 * privs required. */
2699 static int set_offload(struct tun_struct *tun, unsigned long arg)
2701 netdev_features_t features = 0;
2703 if (arg & TUN_F_CSUM) {
2704 features |= NETIF_F_HW_CSUM;
2707 if (arg & (TUN_F_TSO4|TUN_F_TSO6)) {
2708 if (arg & TUN_F_TSO_ECN) {
2709 features |= NETIF_F_TSO_ECN;
2710 arg &= ~TUN_F_TSO_ECN;
2712 if (arg & TUN_F_TSO4)
2713 features |= NETIF_F_TSO;
2714 if (arg & TUN_F_TSO6)
2715 features |= NETIF_F_TSO6;
2716 arg &= ~(TUN_F_TSO4|TUN_F_TSO6);
2722 /* This gives the user a way to test for new features in future by
2723 * trying to set them. */
2727 tun->set_features = features;
2728 tun->dev->wanted_features &= ~TUN_USER_FEATURES;
2729 tun->dev->wanted_features |= features;
2730 netdev_update_features(tun->dev);
2735 static void tun_detach_filter(struct tun_struct *tun, int n)
2738 struct tun_file *tfile;
2740 for (i = 0; i < n; i++) {
2741 tfile = rtnl_dereference(tun->tfiles[i]);
2742 lock_sock(tfile->socket.sk);
2743 sk_detach_filter(tfile->socket.sk);
2744 release_sock(tfile->socket.sk);
2747 tun->filter_attached = false;
2750 static int tun_attach_filter(struct tun_struct *tun)
2753 struct tun_file *tfile;
2755 for (i = 0; i < tun->numqueues; i++) {
2756 tfile = rtnl_dereference(tun->tfiles[i]);
2757 lock_sock(tfile->socket.sk);
2758 ret = sk_attach_filter(&tun->fprog, tfile->socket.sk);
2759 release_sock(tfile->socket.sk);
2761 tun_detach_filter(tun, i);
2766 tun->filter_attached = true;
2770 static void tun_set_sndbuf(struct tun_struct *tun)
2772 struct tun_file *tfile;
2775 for (i = 0; i < tun->numqueues; i++) {
2776 tfile = rtnl_dereference(tun->tfiles[i]);
2777 tfile->socket.sk->sk_sndbuf = tun->sndbuf;
2781 static int tun_set_queue(struct file *file, struct ifreq *ifr)
2783 struct tun_file *tfile = file->private_data;
2784 struct tun_struct *tun;
2789 if (ifr->ifr_flags & IFF_ATTACH_QUEUE) {
2790 tun = tfile->detached;
2795 ret = security_tun_dev_attach_queue(tun->security);
2798 ret = tun_attach(tun, file, false, tun->flags & IFF_NAPI);
2799 } else if (ifr->ifr_flags & IFF_DETACH_QUEUE) {
2800 tun = rtnl_dereference(tfile->tun);
2801 if (!tun || !(tun->flags & IFF_MULTI_QUEUE) || tfile->detached)
2804 __tun_detach(tfile, false);
2813 static int tun_set_ebpf(struct tun_struct *tun, struct tun_prog **prog_p,
2816 struct bpf_prog *prog;
2819 if (copy_from_user(&fd, data, sizeof(fd)))
2825 prog = bpf_prog_get_type(fd, BPF_PROG_TYPE_SOCKET_FILTER);
2827 return PTR_ERR(prog);
2830 return __tun_set_ebpf(tun, prog_p, prog);
2833 static long __tun_chr_ioctl(struct file *file, unsigned int cmd,
2834 unsigned long arg, int ifreq_len)
2836 struct tun_file *tfile = file->private_data;
2837 struct tun_struct *tun;
2838 void __user* argp = (void __user*)arg;
2845 unsigned int ifindex;
2849 if (cmd == TUNSETIFF || cmd == TUNSETQUEUE ||
2850 (_IOC_TYPE(cmd) == SOCK_IOC_TYPE && cmd != SIOCGSKNS)) {
2851 if (copy_from_user(&ifr, argp, ifreq_len))
2854 memset(&ifr, 0, sizeof(ifr));
2856 if (cmd == TUNGETFEATURES) {
2857 /* Currently this just means: "what IFF flags are valid?".
2858 * This is needed because we never checked for invalid flags on
2861 return put_user(IFF_TUN | IFF_TAP | TUN_FEATURES,
2862 (unsigned int __user*)argp);
2863 } else if (cmd == TUNSETQUEUE)
2864 return tun_set_queue(file, &ifr);
2869 tun = tun_get(tfile);
2870 net = sock_net(&tfile->sk);
2871 if (cmd == TUNSETIFF) {
2876 ifr.ifr_name[IFNAMSIZ-1] = '\0';
2878 ret = tun_set_iff(net, file, &ifr);
2883 if (copy_to_user(argp, &ifr, ifreq_len))
2887 if (cmd == TUNSETIFINDEX) {
2893 if (copy_from_user(&ifindex, argp, sizeof(ifindex)))
2897 tfile->ifindex = ifindex;
2900 if (cmd == SIOCGSKNS) {
2902 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
2905 ret = open_related_ns(&net->ns, get_net_ns);
2913 tun_debug(KERN_INFO, tun, "tun_chr_ioctl cmd %u\n", cmd);
2918 tun_get_iff(current->nsproxy->net_ns, tun, &ifr);
2920 if (tfile->detached)
2921 ifr.ifr_flags |= IFF_DETACH_QUEUE;
2922 if (!tfile->socket.sk->sk_filter)
2923 ifr.ifr_flags |= IFF_NOFILTER;
2925 if (copy_to_user(argp, &ifr, ifreq_len))
2930 /* Disable/Enable checksum */
2932 /* [unimplemented] */
2933 tun_debug(KERN_INFO, tun, "ignored: set checksum %s\n",
2934 arg ? "disabled" : "enabled");
2938 /* Disable/Enable persist mode. Keep an extra reference to the
2939 * module to prevent the module being unprobed.
2941 if (arg && !(tun->flags & IFF_PERSIST)) {
2942 tun->flags |= IFF_PERSIST;
2943 __module_get(THIS_MODULE);
2945 if (!arg && (tun->flags & IFF_PERSIST)) {
2946 tun->flags &= ~IFF_PERSIST;
2947 module_put(THIS_MODULE);
2950 tun_debug(KERN_INFO, tun, "persist %s\n",
2951 arg ? "enabled" : "disabled");
2955 /* Set owner of the device */
2956 owner = make_kuid(current_user_ns(), arg);
2957 if (!uid_valid(owner)) {
2962 tun_debug(KERN_INFO, tun, "owner set to %u\n",
2963 from_kuid(&init_user_ns, tun->owner));
2967 /* Set group of the device */
2968 group = make_kgid(current_user_ns(), arg);
2969 if (!gid_valid(group)) {
2974 tun_debug(KERN_INFO, tun, "group set to %u\n",
2975 from_kgid(&init_user_ns, tun->group));
2979 /* Only allow setting the type when the interface is down */
2980 if (tun->dev->flags & IFF_UP) {
2981 tun_debug(KERN_INFO, tun,
2982 "Linktype set failed because interface is up\n");
2985 tun->dev->type = (int) arg;
2986 tun_debug(KERN_INFO, tun, "linktype set to %d\n",
2998 ret = set_offload(tun, arg);
3001 case TUNSETTXFILTER:
3002 /* Can be set only for TAPs */
3004 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3006 ret = update_filter(&tun->txflt, (void __user *)arg);
3010 /* Get hw address */
3011 memcpy(ifr.ifr_hwaddr.sa_data, tun->dev->dev_addr, ETH_ALEN);
3012 ifr.ifr_hwaddr.sa_family = tun->dev->type;
3013 if (copy_to_user(argp, &ifr, ifreq_len))
3018 /* Set hw address */
3019 tun_debug(KERN_DEBUG, tun, "set hw address: %pM\n",
3020 ifr.ifr_hwaddr.sa_data);
3022 ret = dev_set_mac_address(tun->dev, &ifr.ifr_hwaddr);
3026 sndbuf = tfile->socket.sk->sk_sndbuf;
3027 if (copy_to_user(argp, &sndbuf, sizeof(sndbuf)))
3032 if (copy_from_user(&sndbuf, argp, sizeof(sndbuf))) {
3041 tun->sndbuf = sndbuf;
3042 tun_set_sndbuf(tun);
3045 case TUNGETVNETHDRSZ:
3046 vnet_hdr_sz = tun->vnet_hdr_sz;
3047 if (copy_to_user(argp, &vnet_hdr_sz, sizeof(vnet_hdr_sz)))
3051 case TUNSETVNETHDRSZ:
3052 if (copy_from_user(&vnet_hdr_sz, argp, sizeof(vnet_hdr_sz))) {
3056 if (vnet_hdr_sz < (int)sizeof(struct virtio_net_hdr)) {
3061 tun->vnet_hdr_sz = vnet_hdr_sz;
3065 le = !!(tun->flags & TUN_VNET_LE);
3066 if (put_user(le, (int __user *)argp))
3071 if (get_user(le, (int __user *)argp)) {
3076 tun->flags |= TUN_VNET_LE;
3078 tun->flags &= ~TUN_VNET_LE;
3082 ret = tun_get_vnet_be(tun, argp);
3086 ret = tun_set_vnet_be(tun, argp);
3089 case TUNATTACHFILTER:
3090 /* Can be set only for TAPs */
3092 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3095 if (copy_from_user(&tun->fprog, argp, sizeof(tun->fprog)))
3098 ret = tun_attach_filter(tun);
3101 case TUNDETACHFILTER:
3102 /* Can be set only for TAPs */
3104 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3107 tun_detach_filter(tun, tun->numqueues);
3112 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3115 if (copy_to_user(argp, &tun->fprog, sizeof(tun->fprog)))
3120 case TUNSETSTEERINGEBPF:
3121 ret = tun_set_ebpf(tun, &tun->steering_prog, argp);
3124 case TUNSETFILTEREBPF:
3125 ret = tun_set_ebpf(tun, &tun->filter_prog, argp);
3140 static long tun_chr_ioctl(struct file *file,
3141 unsigned int cmd, unsigned long arg)
3143 return __tun_chr_ioctl(file, cmd, arg, sizeof (struct ifreq));
3146 #ifdef CONFIG_COMPAT
3147 static long tun_chr_compat_ioctl(struct file *file,
3148 unsigned int cmd, unsigned long arg)
3153 case TUNSETTXFILTER:
3158 arg = (unsigned long)compat_ptr(arg);
3161 arg = (compat_ulong_t)arg;
3166 * compat_ifreq is shorter than ifreq, so we must not access beyond
3167 * the end of that structure. All fields that are used in this
3168 * driver are compatible though, we don't need to convert the
3171 return __tun_chr_ioctl(file, cmd, arg, sizeof(struct compat_ifreq));
3173 #endif /* CONFIG_COMPAT */
3175 static int tun_chr_fasync(int fd, struct file *file, int on)
3177 struct tun_file *tfile = file->private_data;
3180 if ((ret = fasync_helper(fd, file, on, &tfile->fasync)) < 0)
3184 __f_setown(file, task_pid(current), PIDTYPE_PID, 0);
3185 tfile->flags |= TUN_FASYNC;
3187 tfile->flags &= ~TUN_FASYNC;
3193 static int tun_chr_open(struct inode *inode, struct file * file)
3195 struct net *net = current->nsproxy->net_ns;
3196 struct tun_file *tfile;
3198 DBG1(KERN_INFO, "tunX: tun_chr_open\n");
3200 tfile = (struct tun_file *)sk_alloc(net, AF_UNSPEC, GFP_KERNEL,
3204 RCU_INIT_POINTER(tfile->tun, NULL);
3208 init_waitqueue_head(&tfile->wq.wait);
3209 RCU_INIT_POINTER(tfile->socket.wq, &tfile->wq);
3211 tfile->socket.file = file;
3212 tfile->socket.ops = &tun_socket_ops;
3214 sock_init_data(&tfile->socket, &tfile->sk);
3216 tfile->sk.sk_write_space = tun_sock_write_space;
3217 tfile->sk.sk_sndbuf = INT_MAX;
3219 file->private_data = tfile;
3220 INIT_LIST_HEAD(&tfile->next);
3222 sock_set_flag(&tfile->sk, SOCK_ZEROCOPY);
3224 memset(&tfile->tx_ring, 0, sizeof(tfile->tx_ring));
3229 static int tun_chr_close(struct inode *inode, struct file *file)
3231 struct tun_file *tfile = file->private_data;
3233 tun_detach(tfile, true);
3238 #ifdef CONFIG_PROC_FS
3239 static void tun_chr_show_fdinfo(struct seq_file *m, struct file *file)
3241 struct tun_file *tfile = file->private_data;
3242 struct tun_struct *tun;
3245 memset(&ifr, 0, sizeof(ifr));
3248 tun = tun_get(tfile);
3250 tun_get_iff(current->nsproxy->net_ns, tun, &ifr);
3256 seq_printf(m, "iff:\t%s\n", ifr.ifr_name);
3260 static const struct file_operations tun_fops = {
3261 .owner = THIS_MODULE,
3262 .llseek = no_llseek,
3263 .read_iter = tun_chr_read_iter,
3264 .write_iter = tun_chr_write_iter,
3265 .poll = tun_chr_poll,
3266 .unlocked_ioctl = tun_chr_ioctl,
3267 #ifdef CONFIG_COMPAT
3268 .compat_ioctl = tun_chr_compat_ioctl,
3270 .open = tun_chr_open,
3271 .release = tun_chr_close,
3272 .fasync = tun_chr_fasync,
3273 #ifdef CONFIG_PROC_FS
3274 .show_fdinfo = tun_chr_show_fdinfo,
3278 static struct miscdevice tun_miscdev = {
3281 .nodename = "net/tun",
3285 /* ethtool interface */
3287 static int tun_get_link_ksettings(struct net_device *dev,
3288 struct ethtool_link_ksettings *cmd)
3290 ethtool_link_ksettings_zero_link_mode(cmd, supported);
3291 ethtool_link_ksettings_zero_link_mode(cmd, advertising);
3292 cmd->base.speed = SPEED_10;
3293 cmd->base.duplex = DUPLEX_FULL;
3294 cmd->base.port = PORT_TP;
3295 cmd->base.phy_address = 0;
3296 cmd->base.autoneg = AUTONEG_DISABLE;
3300 static void tun_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
3302 struct tun_struct *tun = netdev_priv(dev);
3304 strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
3305 strlcpy(info->version, DRV_VERSION, sizeof(info->version));
3307 switch (tun->flags & TUN_TYPE_MASK) {
3309 strlcpy(info->bus_info, "tun", sizeof(info->bus_info));
3312 strlcpy(info->bus_info, "tap", sizeof(info->bus_info));
3317 static u32 tun_get_msglevel(struct net_device *dev)
3320 struct tun_struct *tun = netdev_priv(dev);
3327 static void tun_set_msglevel(struct net_device *dev, u32 value)
3330 struct tun_struct *tun = netdev_priv(dev);
3335 static int tun_get_coalesce(struct net_device *dev,
3336 struct ethtool_coalesce *ec)
3338 struct tun_struct *tun = netdev_priv(dev);
3340 ec->rx_max_coalesced_frames = tun->rx_batched;
3345 static int tun_set_coalesce(struct net_device *dev,
3346 struct ethtool_coalesce *ec)
3348 struct tun_struct *tun = netdev_priv(dev);
3350 if (ec->rx_max_coalesced_frames > NAPI_POLL_WEIGHT)
3351 tun->rx_batched = NAPI_POLL_WEIGHT;
3353 tun->rx_batched = ec->rx_max_coalesced_frames;
3358 static const struct ethtool_ops tun_ethtool_ops = {
3359 .get_drvinfo = tun_get_drvinfo,
3360 .get_msglevel = tun_get_msglevel,
3361 .set_msglevel = tun_set_msglevel,
3362 .get_link = ethtool_op_get_link,
3363 .get_ts_info = ethtool_op_get_ts_info,
3364 .get_coalesce = tun_get_coalesce,
3365 .set_coalesce = tun_set_coalesce,
3366 .get_link_ksettings = tun_get_link_ksettings,
3369 static int tun_queue_resize(struct tun_struct *tun)
3371 struct net_device *dev = tun->dev;
3372 struct tun_file *tfile;
3373 struct ptr_ring **rings;
3374 int n = tun->numqueues + tun->numdisabled;
3377 rings = kmalloc_array(n, sizeof(*rings), GFP_KERNEL);
3381 for (i = 0; i < tun->numqueues; i++) {
3382 tfile = rtnl_dereference(tun->tfiles[i]);
3383 rings[i] = &tfile->tx_ring;
3385 list_for_each_entry(tfile, &tun->disabled, next)
3386 rings[i++] = &tfile->tx_ring;
3388 ret = ptr_ring_resize_multiple(rings, n,
3389 dev->tx_queue_len, GFP_KERNEL,
3396 static int tun_device_event(struct notifier_block *unused,
3397 unsigned long event, void *ptr)
3399 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
3400 struct tun_struct *tun = netdev_priv(dev);
3402 if (dev->rtnl_link_ops != &tun_link_ops)
3406 case NETDEV_CHANGE_TX_QUEUE_LEN:
3407 if (tun_queue_resize(tun))
3417 static struct notifier_block tun_notifier_block __read_mostly = {
3418 .notifier_call = tun_device_event,
3421 static int __init tun_init(void)
3425 pr_info("%s, %s\n", DRV_DESCRIPTION, DRV_VERSION);
3427 ret = rtnl_link_register(&tun_link_ops);
3429 pr_err("Can't register link_ops\n");
3433 ret = misc_register(&tun_miscdev);
3435 pr_err("Can't register misc device %d\n", TUN_MINOR);
3439 ret = register_netdevice_notifier(&tun_notifier_block);
3441 pr_err("Can't register netdevice notifier\n");
3448 misc_deregister(&tun_miscdev);
3450 rtnl_link_unregister(&tun_link_ops);
3455 static void tun_cleanup(void)
3457 misc_deregister(&tun_miscdev);
3458 rtnl_link_unregister(&tun_link_ops);
3459 unregister_netdevice_notifier(&tun_notifier_block);
3462 /* Get an underlying socket object from tun file. Returns error unless file is
3463 * attached to a device. The returned object works like a packet socket, it
3464 * can be used for sock_sendmsg/sock_recvmsg. The caller is responsible for
3465 * holding a reference to the file for as long as the socket is in use. */
3466 struct socket *tun_get_socket(struct file *file)
3468 struct tun_file *tfile;
3469 if (file->f_op != &tun_fops)
3470 return ERR_PTR(-EINVAL);
3471 tfile = file->private_data;
3473 return ERR_PTR(-EBADFD);
3474 return &tfile->socket;
3476 EXPORT_SYMBOL_GPL(tun_get_socket);
3478 struct ptr_ring *tun_get_tx_ring(struct file *file)
3480 struct tun_file *tfile;
3482 if (file->f_op != &tun_fops)
3483 return ERR_PTR(-EINVAL);
3484 tfile = file->private_data;
3486 return ERR_PTR(-EBADFD);
3487 return &tfile->tx_ring;
3489 EXPORT_SYMBOL_GPL(tun_get_tx_ring);
3491 module_init(tun_init);
3492 module_exit(tun_cleanup);
3493 MODULE_DESCRIPTION(DRV_DESCRIPTION);
3494 MODULE_AUTHOR(DRV_COPYRIGHT);
3495 MODULE_LICENSE("GPL");
3496 MODULE_ALIAS_MISCDEV(TUN_MINOR);
3497 MODULE_ALIAS("devname:net/tun");