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 static void tun_default_link_ksettings(struct net_device *dev,
85 struct ethtool_link_ksettings *cmd);
87 /* Uncomment to enable debugging */
88 /* #define TUN_DEBUG 1 */
93 #define tun_debug(level, tun, fmt, args...) \
96 netdev_printk(level, tun->dev, fmt, ##args); \
98 #define DBG1(level, fmt, args...) \
101 printk(level fmt, ##args); \
104 #define tun_debug(level, tun, fmt, args...) \
107 netdev_printk(level, tun->dev, fmt, ##args); \
109 #define DBG1(level, fmt, args...) \
112 printk(level fmt, ##args); \
116 #define TUN_HEADROOM 256
117 #define TUN_RX_PAD (NET_IP_ALIGN + NET_SKB_PAD)
119 /* TUN device flags */
121 /* IFF_ATTACH_QUEUE is never stored in device flags,
122 * overload it to mean fasync when stored there.
124 #define TUN_FASYNC IFF_ATTACH_QUEUE
125 /* High bits in flags field are unused. */
126 #define TUN_VNET_LE 0x80000000
127 #define TUN_VNET_BE 0x40000000
129 #define TUN_FEATURES (IFF_NO_PI | IFF_ONE_QUEUE | IFF_VNET_HDR | \
130 IFF_MULTI_QUEUE | IFF_NAPI | IFF_NAPI_FRAGS)
132 #define GOODCOPY_LEN 128
134 #define FLT_EXACT_COUNT 8
136 unsigned int count; /* Number of addrs. Zero means disabled */
137 u32 mask[2]; /* Mask of the hashed addrs */
138 unsigned char addr[FLT_EXACT_COUNT][ETH_ALEN];
141 /* MAX_TAP_QUEUES 256 is chosen to allow rx/tx queues to be equal
142 * to max number of VCPUs in guest. */
143 #define MAX_TAP_QUEUES 256
144 #define MAX_TAP_FLOWS 4096
146 #define TUN_FLOW_EXPIRE (3 * HZ)
148 struct tun_pcpu_stats {
153 struct u64_stats_sync syncp;
159 /* A tun_file connects an open character device to a tuntap netdevice. It
160 * also contains all socket related structures (except sock_fprog and tap_filter)
161 * to serve as one transmit queue for tuntap device. The sock_fprog and
162 * tap_filter were kept in tun_struct since they were used for filtering for the
163 * netdevice not for a specific queue (at least I didn't see the requirement for
167 * The tun_file and tun_struct are loosely coupled, the pointer from one to the
168 * other can only be read while rcu_read_lock or rtnl_lock is held.
172 struct socket socket;
174 struct tun_struct __rcu *tun;
175 struct fasync_struct *fasync;
176 /* only used for fasnyc */
180 unsigned int ifindex;
182 struct napi_struct napi;
184 struct mutex napi_mutex; /* Protects access to the above napi */
185 struct list_head next;
186 struct tun_struct *detached;
187 struct ptr_ring tx_ring;
188 struct xdp_rxq_info xdp_rxq;
191 struct tun_flow_entry {
192 struct hlist_node hash_link;
194 struct tun_struct *tun;
199 unsigned long updated;
202 #define TUN_NUM_FLOW_ENTRIES 1024
203 #define TUN_MASK_FLOW_ENTRIES (TUN_NUM_FLOW_ENTRIES - 1)
207 struct bpf_prog *prog;
210 /* Since the socket were moved to tun_file, to preserve the behavior of persist
211 * device, socket filter, sndbuf and vnet header size were restore when the
212 * file were attached to a persist device.
215 struct tun_file __rcu *tfiles[MAX_TAP_QUEUES];
216 unsigned int numqueues;
221 struct net_device *dev;
222 netdev_features_t set_features;
223 #define TUN_USER_FEATURES (NETIF_F_HW_CSUM|NETIF_F_TSO_ECN|NETIF_F_TSO| \
229 struct tap_filter txflt;
230 struct sock_fprog fprog;
231 /* protected by rtnl lock */
232 bool filter_attached;
237 struct hlist_head flows[TUN_NUM_FLOW_ENTRIES];
238 struct timer_list flow_gc_timer;
239 unsigned long ageing_time;
240 unsigned int numdisabled;
241 struct list_head disabled;
245 struct tun_pcpu_stats __percpu *pcpu_stats;
246 struct bpf_prog __rcu *xdp_prog;
247 struct tun_prog __rcu *steering_prog;
248 struct tun_prog __rcu *filter_prog;
249 struct ethtool_link_ksettings link_ksettings;
257 bool tun_is_xdp_frame(void *ptr)
259 return (unsigned long)ptr & TUN_XDP_FLAG;
261 EXPORT_SYMBOL(tun_is_xdp_frame);
263 void *tun_xdp_to_ptr(void *ptr)
265 return (void *)((unsigned long)ptr | TUN_XDP_FLAG);
267 EXPORT_SYMBOL(tun_xdp_to_ptr);
269 void *tun_ptr_to_xdp(void *ptr)
271 return (void *)((unsigned long)ptr & ~TUN_XDP_FLAG);
273 EXPORT_SYMBOL(tun_ptr_to_xdp);
275 static int tun_napi_receive(struct napi_struct *napi, int budget)
277 struct tun_file *tfile = container_of(napi, struct tun_file, napi);
278 struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
279 struct sk_buff_head process_queue;
283 __skb_queue_head_init(&process_queue);
285 spin_lock(&queue->lock);
286 skb_queue_splice_tail_init(queue, &process_queue);
287 spin_unlock(&queue->lock);
289 while (received < budget && (skb = __skb_dequeue(&process_queue))) {
290 napi_gro_receive(napi, skb);
294 if (!skb_queue_empty(&process_queue)) {
295 spin_lock(&queue->lock);
296 skb_queue_splice(&process_queue, queue);
297 spin_unlock(&queue->lock);
303 static int tun_napi_poll(struct napi_struct *napi, int budget)
305 unsigned int received;
307 received = tun_napi_receive(napi, budget);
309 if (received < budget)
310 napi_complete_done(napi, received);
315 static void tun_napi_init(struct tun_struct *tun, struct tun_file *tfile,
318 tfile->napi_enabled = napi_en;
320 netif_napi_add(tun->dev, &tfile->napi, tun_napi_poll,
322 napi_enable(&tfile->napi);
323 mutex_init(&tfile->napi_mutex);
327 static void tun_napi_disable(struct tun_file *tfile)
329 if (tfile->napi_enabled)
330 napi_disable(&tfile->napi);
333 static void tun_napi_del(struct tun_file *tfile)
335 if (tfile->napi_enabled)
336 netif_napi_del(&tfile->napi);
339 static bool tun_napi_frags_enabled(const struct tun_struct *tun)
341 return READ_ONCE(tun->flags) & IFF_NAPI_FRAGS;
344 #ifdef CONFIG_TUN_VNET_CROSS_LE
345 static inline bool tun_legacy_is_little_endian(struct tun_struct *tun)
347 return tun->flags & TUN_VNET_BE ? false :
348 virtio_legacy_is_little_endian();
351 static long tun_get_vnet_be(struct tun_struct *tun, int __user *argp)
353 int be = !!(tun->flags & TUN_VNET_BE);
355 if (put_user(be, argp))
361 static long tun_set_vnet_be(struct tun_struct *tun, int __user *argp)
365 if (get_user(be, argp))
369 tun->flags |= TUN_VNET_BE;
371 tun->flags &= ~TUN_VNET_BE;
376 static inline bool tun_legacy_is_little_endian(struct tun_struct *tun)
378 return virtio_legacy_is_little_endian();
381 static long tun_get_vnet_be(struct tun_struct *tun, int __user *argp)
386 static long tun_set_vnet_be(struct tun_struct *tun, int __user *argp)
390 #endif /* CONFIG_TUN_VNET_CROSS_LE */
392 static inline bool tun_is_little_endian(struct tun_struct *tun)
394 return tun->flags & TUN_VNET_LE ||
395 tun_legacy_is_little_endian(tun);
398 static inline u16 tun16_to_cpu(struct tun_struct *tun, __virtio16 val)
400 return __virtio16_to_cpu(tun_is_little_endian(tun), val);
403 static inline __virtio16 cpu_to_tun16(struct tun_struct *tun, u16 val)
405 return __cpu_to_virtio16(tun_is_little_endian(tun), val);
408 static inline u32 tun_hashfn(u32 rxhash)
410 return rxhash & TUN_MASK_FLOW_ENTRIES;
413 static struct tun_flow_entry *tun_flow_find(struct hlist_head *head, u32 rxhash)
415 struct tun_flow_entry *e;
417 hlist_for_each_entry_rcu(e, head, hash_link) {
418 if (e->rxhash == rxhash)
424 static struct tun_flow_entry *tun_flow_create(struct tun_struct *tun,
425 struct hlist_head *head,
426 u32 rxhash, u16 queue_index)
428 struct tun_flow_entry *e = kmalloc(sizeof(*e), GFP_ATOMIC);
431 tun_debug(KERN_INFO, tun, "create flow: hash %u index %u\n",
432 rxhash, queue_index);
433 e->updated = jiffies;
436 e->queue_index = queue_index;
438 hlist_add_head_rcu(&e->hash_link, head);
444 static void tun_flow_delete(struct tun_struct *tun, struct tun_flow_entry *e)
446 tun_debug(KERN_INFO, tun, "delete flow: hash %u index %u\n",
447 e->rxhash, e->queue_index);
448 hlist_del_rcu(&e->hash_link);
453 static void tun_flow_flush(struct tun_struct *tun)
457 spin_lock_bh(&tun->lock);
458 for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
459 struct tun_flow_entry *e;
460 struct hlist_node *n;
462 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link)
463 tun_flow_delete(tun, e);
465 spin_unlock_bh(&tun->lock);
468 static void tun_flow_delete_by_queue(struct tun_struct *tun, u16 queue_index)
472 spin_lock_bh(&tun->lock);
473 for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
474 struct tun_flow_entry *e;
475 struct hlist_node *n;
477 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
478 if (e->queue_index == queue_index)
479 tun_flow_delete(tun, e);
482 spin_unlock_bh(&tun->lock);
485 static void tun_flow_cleanup(struct timer_list *t)
487 struct tun_struct *tun = from_timer(tun, t, flow_gc_timer);
488 unsigned long delay = tun->ageing_time;
489 unsigned long next_timer = jiffies + delay;
490 unsigned long count = 0;
493 tun_debug(KERN_INFO, tun, "tun_flow_cleanup\n");
495 spin_lock(&tun->lock);
496 for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
497 struct tun_flow_entry *e;
498 struct hlist_node *n;
500 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
501 unsigned long this_timer;
503 this_timer = e->updated + delay;
504 if (time_before_eq(this_timer, jiffies)) {
505 tun_flow_delete(tun, e);
509 if (time_before(this_timer, next_timer))
510 next_timer = this_timer;
515 mod_timer(&tun->flow_gc_timer, round_jiffies_up(next_timer));
516 spin_unlock(&tun->lock);
519 static void tun_flow_update(struct tun_struct *tun, u32 rxhash,
520 struct tun_file *tfile)
522 struct hlist_head *head;
523 struct tun_flow_entry *e;
524 unsigned long delay = tun->ageing_time;
525 u16 queue_index = tfile->queue_index;
530 head = &tun->flows[tun_hashfn(rxhash)];
534 e = tun_flow_find(head, rxhash);
536 /* TODO: keep queueing to old queue until it's empty? */
537 e->queue_index = queue_index;
538 e->updated = jiffies;
539 sock_rps_record_flow_hash(e->rps_rxhash);
541 spin_lock_bh(&tun->lock);
542 if (!tun_flow_find(head, rxhash) &&
543 tun->flow_count < MAX_TAP_FLOWS)
544 tun_flow_create(tun, head, rxhash, queue_index);
546 if (!timer_pending(&tun->flow_gc_timer))
547 mod_timer(&tun->flow_gc_timer,
548 round_jiffies_up(jiffies + delay));
549 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 struct net_device *sb_dev,
612 select_queue_fallback_t fallback)
614 struct tun_struct *tun = netdev_priv(dev);
618 if (rcu_dereference(tun->steering_prog))
619 ret = tun_ebpf_select_queue(tun, skb);
621 ret = tun_automq_select_queue(tun, skb);
627 static inline bool tun_not_capable(struct tun_struct *tun)
629 const struct cred *cred = current_cred();
630 struct net *net = dev_net(tun->dev);
632 return ((uid_valid(tun->owner) && !uid_eq(cred->euid, tun->owner)) ||
633 (gid_valid(tun->group) && !in_egroup_p(tun->group))) &&
634 !ns_capable(net->user_ns, CAP_NET_ADMIN);
637 static void tun_set_real_num_queues(struct tun_struct *tun)
639 netif_set_real_num_tx_queues(tun->dev, tun->numqueues);
640 netif_set_real_num_rx_queues(tun->dev, tun->numqueues);
643 static void tun_disable_queue(struct tun_struct *tun, struct tun_file *tfile)
645 tfile->detached = tun;
646 list_add_tail(&tfile->next, &tun->disabled);
650 static struct tun_struct *tun_enable_queue(struct tun_file *tfile)
652 struct tun_struct *tun = tfile->detached;
654 tfile->detached = NULL;
655 list_del_init(&tfile->next);
660 void tun_ptr_free(void *ptr)
664 if (tun_is_xdp_frame(ptr)) {
665 struct xdp_frame *xdpf = tun_ptr_to_xdp(ptr);
667 xdp_return_frame(xdpf);
669 __skb_array_destroy_skb(ptr);
672 EXPORT_SYMBOL_GPL(tun_ptr_free);
674 static void tun_queue_purge(struct tun_file *tfile)
678 while ((ptr = ptr_ring_consume(&tfile->tx_ring)) != NULL)
681 skb_queue_purge(&tfile->sk.sk_write_queue);
682 skb_queue_purge(&tfile->sk.sk_error_queue);
685 static void __tun_detach(struct tun_file *tfile, bool clean)
687 struct tun_file *ntfile;
688 struct tun_struct *tun;
690 tun = rtnl_dereference(tfile->tun);
693 tun_napi_disable(tfile);
697 if (tun && !tfile->detached) {
698 u16 index = tfile->queue_index;
699 BUG_ON(index >= tun->numqueues);
701 rcu_assign_pointer(tun->tfiles[index],
702 tun->tfiles[tun->numqueues - 1]);
703 ntfile = rtnl_dereference(tun->tfiles[index]);
704 ntfile->queue_index = index;
708 RCU_INIT_POINTER(tfile->tun, NULL);
709 sock_put(&tfile->sk);
711 tun_disable_queue(tun, tfile);
714 tun_flow_delete_by_queue(tun, tun->numqueues + 1);
715 /* Drop read queue */
716 tun_queue_purge(tfile);
717 tun_set_real_num_queues(tun);
718 } else if (tfile->detached && clean) {
719 tun = tun_enable_queue(tfile);
720 sock_put(&tfile->sk);
724 if (tun && tun->numqueues == 0 && tun->numdisabled == 0) {
725 netif_carrier_off(tun->dev);
727 if (!(tun->flags & IFF_PERSIST) &&
728 tun->dev->reg_state == NETREG_REGISTERED)
729 unregister_netdevice(tun->dev);
732 xdp_rxq_info_unreg(&tfile->xdp_rxq);
733 ptr_ring_cleanup(&tfile->tx_ring, tun_ptr_free);
734 sock_put(&tfile->sk);
738 static void tun_detach(struct tun_file *tfile, bool clean)
740 struct tun_struct *tun;
741 struct net_device *dev;
744 tun = rtnl_dereference(tfile->tun);
745 dev = tun ? tun->dev : NULL;
746 __tun_detach(tfile, clean);
748 netdev_state_change(dev);
752 static void tun_detach_all(struct net_device *dev)
754 struct tun_struct *tun = netdev_priv(dev);
755 struct tun_file *tfile, *tmp;
756 int i, n = tun->numqueues;
758 for (i = 0; i < n; i++) {
759 tfile = rtnl_dereference(tun->tfiles[i]);
761 tun_napi_disable(tfile);
762 tfile->socket.sk->sk_shutdown = RCV_SHUTDOWN;
763 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
764 RCU_INIT_POINTER(tfile->tun, NULL);
767 list_for_each_entry(tfile, &tun->disabled, next) {
768 tfile->socket.sk->sk_shutdown = RCV_SHUTDOWN;
769 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
770 RCU_INIT_POINTER(tfile->tun, NULL);
772 BUG_ON(tun->numqueues != 0);
775 for (i = 0; i < n; i++) {
776 tfile = rtnl_dereference(tun->tfiles[i]);
778 /* Drop read queue */
779 tun_queue_purge(tfile);
780 xdp_rxq_info_unreg(&tfile->xdp_rxq);
781 sock_put(&tfile->sk);
783 list_for_each_entry_safe(tfile, tmp, &tun->disabled, next) {
784 tun_enable_queue(tfile);
785 tun_queue_purge(tfile);
786 xdp_rxq_info_unreg(&tfile->xdp_rxq);
787 sock_put(&tfile->sk);
789 BUG_ON(tun->numdisabled != 0);
791 if (tun->flags & IFF_PERSIST)
792 module_put(THIS_MODULE);
795 static int tun_attach(struct tun_struct *tun, struct file *file,
796 bool skip_filter, bool napi)
798 struct tun_file *tfile = file->private_data;
799 struct net_device *dev = tun->dev;
802 err = security_tun_dev_attach(tfile->socket.sk, tun->security);
807 if (rtnl_dereference(tfile->tun) && !tfile->detached)
811 if (!(tun->flags & IFF_MULTI_QUEUE) && tun->numqueues == 1)
815 if (!tfile->detached &&
816 tun->numqueues + tun->numdisabled == MAX_TAP_QUEUES)
821 /* Re-attach the filter to persist device */
822 if (!skip_filter && (tun->filter_attached == true)) {
823 lock_sock(tfile->socket.sk);
824 err = sk_attach_filter(&tun->fprog, tfile->socket.sk);
825 release_sock(tfile->socket.sk);
830 if (!tfile->detached &&
831 ptr_ring_resize(&tfile->tx_ring, dev->tx_queue_len,
832 GFP_KERNEL, tun_ptr_free)) {
837 tfile->queue_index = tun->numqueues;
838 tfile->socket.sk->sk_shutdown &= ~RCV_SHUTDOWN;
840 if (tfile->detached) {
841 /* Re-attach detached tfile, updating XDP queue_index */
842 WARN_ON(!xdp_rxq_info_is_reg(&tfile->xdp_rxq));
844 if (tfile->xdp_rxq.queue_index != tfile->queue_index)
845 tfile->xdp_rxq.queue_index = tfile->queue_index;
847 /* Setup XDP RX-queue info, for new tfile getting attached */
848 err = xdp_rxq_info_reg(&tfile->xdp_rxq,
849 tun->dev, tfile->queue_index);
852 err = xdp_rxq_info_reg_mem_model(&tfile->xdp_rxq,
853 MEM_TYPE_PAGE_SHARED, NULL);
855 xdp_rxq_info_unreg(&tfile->xdp_rxq);
861 rcu_assign_pointer(tfile->tun, tun);
862 rcu_assign_pointer(tun->tfiles[tun->numqueues], tfile);
865 if (tfile->detached) {
866 tun_enable_queue(tfile);
868 sock_hold(&tfile->sk);
869 tun_napi_init(tun, tfile, napi);
872 tun_set_real_num_queues(tun);
874 /* device is allowed to go away first, so no need to hold extra
882 static struct tun_struct *tun_get(struct tun_file *tfile)
884 struct tun_struct *tun;
887 tun = rcu_dereference(tfile->tun);
895 static void tun_put(struct tun_struct *tun)
901 static void addr_hash_set(u32 *mask, const u8 *addr)
903 int n = ether_crc(ETH_ALEN, addr) >> 26;
904 mask[n >> 5] |= (1 << (n & 31));
907 static unsigned int addr_hash_test(const u32 *mask, const u8 *addr)
909 int n = ether_crc(ETH_ALEN, addr) >> 26;
910 return mask[n >> 5] & (1 << (n & 31));
913 static int update_filter(struct tap_filter *filter, void __user *arg)
915 struct { u8 u[ETH_ALEN]; } *addr;
916 struct tun_filter uf;
917 int err, alen, n, nexact;
919 if (copy_from_user(&uf, arg, sizeof(uf)))
928 alen = ETH_ALEN * uf.count;
929 addr = memdup_user(arg + sizeof(uf), alen);
931 return PTR_ERR(addr);
933 /* The filter is updated without holding any locks. Which is
934 * perfectly safe. We disable it first and in the worst
935 * case we'll accept a few undesired packets. */
939 /* Use first set of addresses as an exact filter */
940 for (n = 0; n < uf.count && n < FLT_EXACT_COUNT; n++)
941 memcpy(filter->addr[n], addr[n].u, ETH_ALEN);
945 /* Remaining multicast addresses are hashed,
946 * unicast will leave the filter disabled. */
947 memset(filter->mask, 0, sizeof(filter->mask));
948 for (; n < uf.count; n++) {
949 if (!is_multicast_ether_addr(addr[n].u)) {
950 err = 0; /* no filter */
953 addr_hash_set(filter->mask, addr[n].u);
956 /* For ALLMULTI just set the mask to all ones.
957 * This overrides the mask populated above. */
958 if ((uf.flags & TUN_FLT_ALLMULTI))
959 memset(filter->mask, ~0, sizeof(filter->mask));
961 /* Now enable the filter */
963 filter->count = nexact;
965 /* Return the number of exact filters */
972 /* Returns: 0 - drop, !=0 - accept */
973 static int run_filter(struct tap_filter *filter, const struct sk_buff *skb)
975 /* Cannot use eth_hdr(skb) here because skb_mac_hdr() is incorrect
977 struct ethhdr *eh = (struct ethhdr *) skb->data;
981 for (i = 0; i < filter->count; i++)
982 if (ether_addr_equal(eh->h_dest, filter->addr[i]))
985 /* Inexact match (multicast only) */
986 if (is_multicast_ether_addr(eh->h_dest))
987 return addr_hash_test(filter->mask, eh->h_dest);
993 * Checks whether the packet is accepted or not.
994 * Returns: 0 - drop, !=0 - accept
996 static int check_filter(struct tap_filter *filter, const struct sk_buff *skb)
1001 return run_filter(filter, skb);
1004 /* Network device part of the driver */
1006 static const struct ethtool_ops tun_ethtool_ops;
1008 /* Net device detach from fd. */
1009 static void tun_net_uninit(struct net_device *dev)
1011 tun_detach_all(dev);
1014 /* Net device open. */
1015 static int tun_net_open(struct net_device *dev)
1017 struct tun_struct *tun = netdev_priv(dev);
1020 netif_tx_start_all_queues(dev);
1022 for (i = 0; i < tun->numqueues; i++) {
1023 struct tun_file *tfile;
1025 tfile = rtnl_dereference(tun->tfiles[i]);
1026 tfile->socket.sk->sk_write_space(tfile->socket.sk);
1032 /* Net device close. */
1033 static int tun_net_close(struct net_device *dev)
1035 netif_tx_stop_all_queues(dev);
1039 /* Net device start xmit */
1040 static void tun_automq_xmit(struct tun_struct *tun, struct sk_buff *skb)
1043 if (tun->numqueues == 1 && static_key_false(&rps_needed)) {
1044 /* Select queue was not called for the skbuff, so we extract the
1045 * RPS hash and save it into the flow_table here.
1049 rxhash = __skb_get_hash_symmetric(skb);
1051 struct tun_flow_entry *e;
1052 e = tun_flow_find(&tun->flows[tun_hashfn(rxhash)],
1055 tun_flow_save_rps_rxhash(e, rxhash);
1061 static unsigned int run_ebpf_filter(struct tun_struct *tun,
1062 struct sk_buff *skb,
1065 struct tun_prog *prog = rcu_dereference(tun->filter_prog);
1068 len = bpf_prog_run_clear_cb(prog->prog, skb);
1073 /* Net device start xmit */
1074 static netdev_tx_t tun_net_xmit(struct sk_buff *skb, struct net_device *dev)
1076 struct tun_struct *tun = netdev_priv(dev);
1077 int txq = skb->queue_mapping;
1078 struct tun_file *tfile;
1082 tfile = rcu_dereference(tun->tfiles[txq]);
1084 /* Drop packet if interface is not attached */
1085 if (txq >= tun->numqueues)
1088 if (!rcu_dereference(tun->steering_prog))
1089 tun_automq_xmit(tun, skb);
1091 tun_debug(KERN_INFO, tun, "tun_net_xmit %d\n", skb->len);
1095 /* Drop if the filter does not like it.
1096 * This is a noop if the filter is disabled.
1097 * Filter can be enabled only for the TAP devices. */
1098 if (!check_filter(&tun->txflt, skb))
1101 if (tfile->socket.sk->sk_filter &&
1102 sk_filter(tfile->socket.sk, skb))
1105 len = run_ebpf_filter(tun, skb, len);
1106 if (len == 0 || pskb_trim(skb, len))
1109 if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC)))
1112 skb_tx_timestamp(skb);
1114 /* Orphan the skb - required as we might hang on to it
1115 * for indefinite time.
1121 if (ptr_ring_produce(&tfile->tx_ring, skb))
1124 /* Notify and wake up reader process */
1125 if (tfile->flags & TUN_FASYNC)
1126 kill_fasync(&tfile->fasync, SIGIO, POLL_IN);
1127 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
1130 return NETDEV_TX_OK;
1133 this_cpu_inc(tun->pcpu_stats->tx_dropped);
1137 return NET_XMIT_DROP;
1140 static void tun_net_mclist(struct net_device *dev)
1143 * This callback is supposed to deal with mc filter in
1144 * _rx_ path and has nothing to do with the _tx_ path.
1145 * In rx path we always accept everything userspace gives us.
1149 static netdev_features_t tun_net_fix_features(struct net_device *dev,
1150 netdev_features_t features)
1152 struct tun_struct *tun = netdev_priv(dev);
1154 return (features & tun->set_features) | (features & ~TUN_USER_FEATURES);
1157 static void tun_set_headroom(struct net_device *dev, int new_hr)
1159 struct tun_struct *tun = netdev_priv(dev);
1161 if (new_hr < NET_SKB_PAD)
1162 new_hr = NET_SKB_PAD;
1164 tun->align = new_hr;
1168 tun_net_get_stats64(struct net_device *dev, struct rtnl_link_stats64 *stats)
1170 u32 rx_dropped = 0, tx_dropped = 0, rx_frame_errors = 0;
1171 struct tun_struct *tun = netdev_priv(dev);
1172 struct tun_pcpu_stats *p;
1175 for_each_possible_cpu(i) {
1176 u64 rxpackets, rxbytes, txpackets, txbytes;
1179 p = per_cpu_ptr(tun->pcpu_stats, i);
1181 start = u64_stats_fetch_begin(&p->syncp);
1182 rxpackets = p->rx_packets;
1183 rxbytes = p->rx_bytes;
1184 txpackets = p->tx_packets;
1185 txbytes = p->tx_bytes;
1186 } while (u64_stats_fetch_retry(&p->syncp, start));
1188 stats->rx_packets += rxpackets;
1189 stats->rx_bytes += rxbytes;
1190 stats->tx_packets += txpackets;
1191 stats->tx_bytes += txbytes;
1194 rx_dropped += p->rx_dropped;
1195 rx_frame_errors += p->rx_frame_errors;
1196 tx_dropped += p->tx_dropped;
1198 stats->rx_dropped = rx_dropped;
1199 stats->rx_frame_errors = rx_frame_errors;
1200 stats->tx_dropped = tx_dropped;
1203 static int tun_xdp_set(struct net_device *dev, struct bpf_prog *prog,
1204 struct netlink_ext_ack *extack)
1206 struct tun_struct *tun = netdev_priv(dev);
1207 struct bpf_prog *old_prog;
1209 old_prog = rtnl_dereference(tun->xdp_prog);
1210 rcu_assign_pointer(tun->xdp_prog, prog);
1212 bpf_prog_put(old_prog);
1217 static u32 tun_xdp_query(struct net_device *dev)
1219 struct tun_struct *tun = netdev_priv(dev);
1220 const struct bpf_prog *xdp_prog;
1222 xdp_prog = rtnl_dereference(tun->xdp_prog);
1224 return xdp_prog->aux->id;
1229 static int tun_xdp(struct net_device *dev, struct netdev_bpf *xdp)
1231 switch (xdp->command) {
1232 case XDP_SETUP_PROG:
1233 return tun_xdp_set(dev, xdp->prog, xdp->extack);
1234 case XDP_QUERY_PROG:
1235 xdp->prog_id = tun_xdp_query(dev);
1242 static const struct net_device_ops tun_netdev_ops = {
1243 .ndo_uninit = tun_net_uninit,
1244 .ndo_open = tun_net_open,
1245 .ndo_stop = tun_net_close,
1246 .ndo_start_xmit = tun_net_xmit,
1247 .ndo_fix_features = tun_net_fix_features,
1248 .ndo_select_queue = tun_select_queue,
1249 .ndo_set_rx_headroom = tun_set_headroom,
1250 .ndo_get_stats64 = tun_net_get_stats64,
1253 static void __tun_xdp_flush_tfile(struct tun_file *tfile)
1255 /* Notify and wake up reader process */
1256 if (tfile->flags & TUN_FASYNC)
1257 kill_fasync(&tfile->fasync, SIGIO, POLL_IN);
1258 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
1261 static int tun_xdp_xmit(struct net_device *dev, int n,
1262 struct xdp_frame **frames, u32 flags)
1264 struct tun_struct *tun = netdev_priv(dev);
1265 struct tun_file *tfile;
1271 if (unlikely(flags & ~XDP_XMIT_FLAGS_MASK))
1276 numqueues = READ_ONCE(tun->numqueues);
1279 return -ENXIO; /* Caller will free/return all frames */
1282 tfile = rcu_dereference(tun->tfiles[smp_processor_id() %
1285 spin_lock(&tfile->tx_ring.producer_lock);
1286 for (i = 0; i < n; i++) {
1287 struct xdp_frame *xdp = frames[i];
1288 /* Encode the XDP flag into lowest bit for consumer to differ
1289 * XDP buffer from sk_buff.
1291 void *frame = tun_xdp_to_ptr(xdp);
1293 if (__ptr_ring_produce(&tfile->tx_ring, frame)) {
1294 this_cpu_inc(tun->pcpu_stats->tx_dropped);
1295 xdp_return_frame_rx_napi(xdp);
1299 spin_unlock(&tfile->tx_ring.producer_lock);
1301 if (flags & XDP_XMIT_FLUSH)
1302 __tun_xdp_flush_tfile(tfile);
1308 static int tun_xdp_tx(struct net_device *dev, struct xdp_buff *xdp)
1310 struct xdp_frame *frame = convert_to_xdp_frame(xdp);
1312 if (unlikely(!frame))
1315 return tun_xdp_xmit(dev, 1, &frame, XDP_XMIT_FLUSH);
1318 static const struct net_device_ops tap_netdev_ops = {
1319 .ndo_uninit = tun_net_uninit,
1320 .ndo_open = tun_net_open,
1321 .ndo_stop = tun_net_close,
1322 .ndo_start_xmit = tun_net_xmit,
1323 .ndo_fix_features = tun_net_fix_features,
1324 .ndo_set_rx_mode = tun_net_mclist,
1325 .ndo_set_mac_address = eth_mac_addr,
1326 .ndo_validate_addr = eth_validate_addr,
1327 .ndo_select_queue = tun_select_queue,
1328 .ndo_features_check = passthru_features_check,
1329 .ndo_set_rx_headroom = tun_set_headroom,
1330 .ndo_get_stats64 = tun_net_get_stats64,
1332 .ndo_xdp_xmit = tun_xdp_xmit,
1335 static void tun_flow_init(struct tun_struct *tun)
1339 for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++)
1340 INIT_HLIST_HEAD(&tun->flows[i]);
1342 tun->ageing_time = TUN_FLOW_EXPIRE;
1343 timer_setup(&tun->flow_gc_timer, tun_flow_cleanup, 0);
1344 mod_timer(&tun->flow_gc_timer,
1345 round_jiffies_up(jiffies + tun->ageing_time));
1348 static void tun_flow_uninit(struct tun_struct *tun)
1350 del_timer_sync(&tun->flow_gc_timer);
1351 tun_flow_flush(tun);
1355 #define MAX_MTU 65535
1357 /* Initialize net device. */
1358 static void tun_net_init(struct net_device *dev)
1360 struct tun_struct *tun = netdev_priv(dev);
1362 switch (tun->flags & TUN_TYPE_MASK) {
1364 dev->netdev_ops = &tun_netdev_ops;
1366 /* Point-to-Point TUN Device */
1367 dev->hard_header_len = 0;
1371 /* Zero header length */
1372 dev->type = ARPHRD_NONE;
1373 dev->flags = IFF_POINTOPOINT | IFF_NOARP | IFF_MULTICAST;
1377 dev->netdev_ops = &tap_netdev_ops;
1378 /* Ethernet TAP Device */
1380 dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1381 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
1383 eth_hw_addr_random(dev);
1388 dev->min_mtu = MIN_MTU;
1389 dev->max_mtu = MAX_MTU - dev->hard_header_len;
1392 static bool tun_sock_writeable(struct tun_struct *tun, struct tun_file *tfile)
1394 struct sock *sk = tfile->socket.sk;
1396 return (tun->dev->flags & IFF_UP) && sock_writeable(sk);
1399 /* Character device part */
1402 static __poll_t tun_chr_poll(struct file *file, poll_table *wait)
1404 struct tun_file *tfile = file->private_data;
1405 struct tun_struct *tun = tun_get(tfile);
1412 sk = tfile->socket.sk;
1414 tun_debug(KERN_INFO, tun, "tun_chr_poll\n");
1416 poll_wait(file, sk_sleep(sk), wait);
1418 if (!ptr_ring_empty(&tfile->tx_ring))
1419 mask |= EPOLLIN | EPOLLRDNORM;
1421 /* Make sure SOCKWQ_ASYNC_NOSPACE is set if not writable to
1422 * guarantee EPOLLOUT to be raised by either here or
1423 * tun_sock_write_space(). Then process could get notification
1424 * after it writes to a down device and meets -EIO.
1426 if (tun_sock_writeable(tun, tfile) ||
1427 (!test_and_set_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags) &&
1428 tun_sock_writeable(tun, tfile)))
1429 mask |= EPOLLOUT | EPOLLWRNORM;
1431 if (tun->dev->reg_state != NETREG_REGISTERED)
1438 static struct sk_buff *tun_napi_alloc_frags(struct tun_file *tfile,
1440 const struct iov_iter *it)
1442 struct sk_buff *skb;
1447 if (it->nr_segs > MAX_SKB_FRAGS + 1)
1448 return ERR_PTR(-ENOMEM);
1451 skb = napi_get_frags(&tfile->napi);
1454 return ERR_PTR(-ENOMEM);
1456 linear = iov_iter_single_seg_count(it);
1457 err = __skb_grow(skb, linear);
1462 skb->data_len = len - linear;
1463 skb->truesize += skb->data_len;
1465 for (i = 1; i < it->nr_segs; i++) {
1466 struct page_frag *pfrag = ¤t->task_frag;
1467 size_t fragsz = it->iov[i].iov_len;
1469 if (fragsz == 0 || fragsz > PAGE_SIZE) {
1474 if (!skb_page_frag_refill(fragsz, pfrag, GFP_KERNEL)) {
1479 skb_fill_page_desc(skb, i - 1, pfrag->page,
1480 pfrag->offset, fragsz);
1481 page_ref_inc(pfrag->page);
1482 pfrag->offset += fragsz;
1487 /* frees skb and all frags allocated with napi_alloc_frag() */
1488 napi_free_frags(&tfile->napi);
1489 return ERR_PTR(err);
1492 /* prepad is the amount to reserve at front. len is length after that.
1493 * linear is a hint as to how much to copy (usually headers). */
1494 static struct sk_buff *tun_alloc_skb(struct tun_file *tfile,
1495 size_t prepad, size_t len,
1496 size_t linear, int noblock)
1498 struct sock *sk = tfile->socket.sk;
1499 struct sk_buff *skb;
1502 /* Under a page? Don't bother with paged skb. */
1503 if (prepad + len < PAGE_SIZE || !linear)
1506 skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
1509 return ERR_PTR(err);
1511 skb_reserve(skb, prepad);
1512 skb_put(skb, linear);
1513 skb->data_len = len - linear;
1514 skb->len += len - linear;
1519 static void tun_rx_batched(struct tun_struct *tun, struct tun_file *tfile,
1520 struct sk_buff *skb, int more)
1522 struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
1523 struct sk_buff_head process_queue;
1524 u32 rx_batched = tun->rx_batched;
1527 if (!rx_batched || (!more && skb_queue_empty(queue))) {
1529 netif_receive_skb(skb);
1534 spin_lock(&queue->lock);
1535 if (!more || skb_queue_len(queue) == rx_batched) {
1536 __skb_queue_head_init(&process_queue);
1537 skb_queue_splice_tail_init(queue, &process_queue);
1540 __skb_queue_tail(queue, skb);
1542 spin_unlock(&queue->lock);
1545 struct sk_buff *nskb;
1548 while ((nskb = __skb_dequeue(&process_queue)))
1549 netif_receive_skb(nskb);
1550 netif_receive_skb(skb);
1555 static bool tun_can_build_skb(struct tun_struct *tun, struct tun_file *tfile,
1556 int len, int noblock, bool zerocopy)
1558 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
1561 if (tfile->socket.sk->sk_sndbuf != INT_MAX)
1570 if (SKB_DATA_ALIGN(len + TUN_RX_PAD) +
1571 SKB_DATA_ALIGN(sizeof(struct skb_shared_info)) > PAGE_SIZE)
1577 static struct sk_buff *tun_build_skb(struct tun_struct *tun,
1578 struct tun_file *tfile,
1579 struct iov_iter *from,
1580 struct virtio_net_hdr *hdr,
1581 int len, int *skb_xdp)
1583 struct page_frag *alloc_frag = ¤t->task_frag;
1584 struct sk_buff *skb;
1585 struct bpf_prog *xdp_prog;
1586 int buflen = SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
1587 unsigned int delta = 0;
1590 int err, pad = TUN_RX_PAD;
1593 xdp_prog = rcu_dereference(tun->xdp_prog);
1595 pad += TUN_HEADROOM;
1596 buflen += SKB_DATA_ALIGN(len + pad);
1599 alloc_frag->offset = ALIGN((u64)alloc_frag->offset, SMP_CACHE_BYTES);
1600 if (unlikely(!skb_page_frag_refill(buflen, alloc_frag, GFP_KERNEL)))
1601 return ERR_PTR(-ENOMEM);
1603 buf = (char *)page_address(alloc_frag->page) + alloc_frag->offset;
1604 copied = copy_page_from_iter(alloc_frag->page,
1605 alloc_frag->offset + pad,
1608 return ERR_PTR(-EFAULT);
1610 /* There's a small window that XDP may be set after the check
1611 * of xdp_prog above, this should be rare and for simplicity
1612 * we do XDP on skb in case the headroom is not enough.
1614 if (hdr->gso_type || !xdp_prog)
1621 xdp_prog = rcu_dereference(tun->xdp_prog);
1622 if (xdp_prog && !*skb_xdp) {
1623 struct xdp_buff xdp;
1627 xdp.data_hard_start = buf;
1628 xdp.data = buf + pad;
1629 xdp_set_data_meta_invalid(&xdp);
1630 xdp.data_end = xdp.data + len;
1631 xdp.rxq = &tfile->xdp_rxq;
1632 orig_data = xdp.data;
1633 act = bpf_prog_run_xdp(xdp_prog, &xdp);
1637 get_page(alloc_frag->page);
1638 alloc_frag->offset += buflen;
1639 err = xdp_do_redirect(tun->dev, &xdp, xdp_prog);
1647 get_page(alloc_frag->page);
1648 alloc_frag->offset += buflen;
1649 if (tun_xdp_tx(tun->dev, &xdp) < 0)
1655 delta = orig_data - xdp.data;
1656 len = xdp.data_end - xdp.data;
1659 bpf_warn_invalid_xdp_action(act);
1662 trace_xdp_exception(tun->dev, xdp_prog, act);
1669 skb = build_skb(buf, buflen);
1673 return ERR_PTR(-ENOMEM);
1676 skb_reserve(skb, pad - delta);
1678 get_page(alloc_frag->page);
1679 alloc_frag->offset += buflen;
1687 put_page(alloc_frag->page);
1691 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1695 /* Get packet from user space buffer */
1696 static ssize_t tun_get_user(struct tun_struct *tun, struct tun_file *tfile,
1697 void *msg_control, struct iov_iter *from,
1698 int noblock, bool more)
1700 struct tun_pi pi = { 0, cpu_to_be16(ETH_P_IP) };
1701 struct sk_buff *skb;
1702 size_t total_len = iov_iter_count(from);
1703 size_t len = total_len, align = tun->align, linear;
1704 struct virtio_net_hdr gso = { 0 };
1705 struct tun_pcpu_stats *stats;
1708 bool zerocopy = false;
1712 bool frags = tun_napi_frags_enabled(tun);
1714 if (!(tun->dev->flags & IFF_UP))
1717 if (!(tun->flags & IFF_NO_PI)) {
1718 if (len < sizeof(pi))
1722 if (!copy_from_iter_full(&pi, sizeof(pi), from))
1726 if (tun->flags & IFF_VNET_HDR) {
1727 int vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
1729 if (len < vnet_hdr_sz)
1733 if (!copy_from_iter_full(&gso, sizeof(gso), from))
1736 if ((gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
1737 tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2 > tun16_to_cpu(tun, gso.hdr_len))
1738 gso.hdr_len = cpu_to_tun16(tun, tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2);
1740 if (tun16_to_cpu(tun, gso.hdr_len) > len)
1742 iov_iter_advance(from, vnet_hdr_sz - sizeof(gso));
1745 if ((tun->flags & TUN_TYPE_MASK) == IFF_TAP) {
1746 align += NET_IP_ALIGN;
1747 if (unlikely(len < ETH_HLEN ||
1748 (gso.hdr_len && tun16_to_cpu(tun, gso.hdr_len) < ETH_HLEN)))
1752 good_linear = SKB_MAX_HEAD(align);
1755 struct iov_iter i = *from;
1757 /* There are 256 bytes to be copied in skb, so there is
1758 * enough room for skb expand head in case it is used.
1759 * The rest of the buffer is mapped from userspace.
1761 copylen = gso.hdr_len ? tun16_to_cpu(tun, gso.hdr_len) : GOODCOPY_LEN;
1762 if (copylen > good_linear)
1763 copylen = good_linear;
1765 iov_iter_advance(&i, copylen);
1766 if (iov_iter_npages(&i, INT_MAX) <= MAX_SKB_FRAGS)
1770 if (!frags && tun_can_build_skb(tun, tfile, len, noblock, zerocopy)) {
1771 /* For the packet that is not easy to be processed
1772 * (e.g gso or jumbo packet), we will do it at after
1773 * skb was created with generic XDP routine.
1775 skb = tun_build_skb(tun, tfile, from, &gso, len, &skb_xdp);
1777 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1778 return PTR_ERR(skb);
1785 if (tun16_to_cpu(tun, gso.hdr_len) > good_linear)
1786 linear = good_linear;
1788 linear = tun16_to_cpu(tun, gso.hdr_len);
1792 mutex_lock(&tfile->napi_mutex);
1793 skb = tun_napi_alloc_frags(tfile, copylen, from);
1794 /* tun_napi_alloc_frags() enforces a layout for the skb.
1795 * If zerocopy is enabled, then this layout will be
1796 * overwritten by zerocopy_sg_from_iter().
1800 skb = tun_alloc_skb(tfile, align, copylen, linear,
1805 if (PTR_ERR(skb) != -EAGAIN)
1806 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1808 mutex_unlock(&tfile->napi_mutex);
1809 return PTR_ERR(skb);
1813 err = zerocopy_sg_from_iter(skb, from);
1815 err = skb_copy_datagram_from_iter(skb, 0, from, len);
1818 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1821 tfile->napi.skb = NULL;
1822 mutex_unlock(&tfile->napi_mutex);
1829 if (virtio_net_hdr_to_skb(skb, &gso, tun_is_little_endian(tun))) {
1830 this_cpu_inc(tun->pcpu_stats->rx_frame_errors);
1833 tfile->napi.skb = NULL;
1834 mutex_unlock(&tfile->napi_mutex);
1840 switch (tun->flags & TUN_TYPE_MASK) {
1842 if (tun->flags & IFF_NO_PI) {
1843 u8 ip_version = skb->len ? (skb->data[0] >> 4) : 0;
1845 switch (ip_version) {
1847 pi.proto = htons(ETH_P_IP);
1850 pi.proto = htons(ETH_P_IPV6);
1853 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1859 skb_reset_mac_header(skb);
1860 skb->protocol = pi.proto;
1861 skb->dev = tun->dev;
1865 skb->protocol = eth_type_trans(skb, tun->dev);
1869 /* copy skb_ubuf_info for callback when skb has no error */
1871 skb_shinfo(skb)->destructor_arg = msg_control;
1872 skb_shinfo(skb)->tx_flags |= SKBTX_DEV_ZEROCOPY;
1873 skb_shinfo(skb)->tx_flags |= SKBTX_SHARED_FRAG;
1874 } else if (msg_control) {
1875 struct ubuf_info *uarg = msg_control;
1876 uarg->callback(uarg, false);
1879 skb_reset_network_header(skb);
1880 skb_probe_transport_header(skb, 0);
1883 struct bpf_prog *xdp_prog;
1888 xdp_prog = rcu_dereference(tun->xdp_prog);
1890 ret = do_xdp_generic(xdp_prog, skb);
1891 if (ret != XDP_PASS) {
1901 /* Compute the costly rx hash only if needed for flow updates.
1902 * We may get a very small possibility of OOO during switching, not
1903 * worth to optimize.
1905 if (!rcu_access_pointer(tun->steering_prog) && tun->numqueues > 1 &&
1907 rxhash = __skb_get_hash_symmetric(skb);
1910 /* Exercise flow dissector code path. */
1911 u32 headlen = eth_get_headlen(skb->data, skb_headlen(skb));
1913 if (unlikely(headlen > skb_headlen(skb))) {
1914 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1915 napi_free_frags(&tfile->napi);
1916 mutex_unlock(&tfile->napi_mutex);
1922 napi_gro_frags(&tfile->napi);
1924 mutex_unlock(&tfile->napi_mutex);
1925 } else if (tfile->napi_enabled) {
1926 struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
1929 spin_lock_bh(&queue->lock);
1930 __skb_queue_tail(queue, skb);
1931 queue_len = skb_queue_len(queue);
1932 spin_unlock(&queue->lock);
1934 if (!more || queue_len > NAPI_POLL_WEIGHT)
1935 napi_schedule(&tfile->napi);
1938 } else if (!IS_ENABLED(CONFIG_4KSTACKS)) {
1939 tun_rx_batched(tun, tfile, skb, more);
1944 stats = get_cpu_ptr(tun->pcpu_stats);
1945 u64_stats_update_begin(&stats->syncp);
1946 stats->rx_packets++;
1947 stats->rx_bytes += len;
1948 u64_stats_update_end(&stats->syncp);
1952 tun_flow_update(tun, rxhash, tfile);
1957 static ssize_t tun_chr_write_iter(struct kiocb *iocb, struct iov_iter *from)
1959 struct file *file = iocb->ki_filp;
1960 struct tun_file *tfile = file->private_data;
1961 struct tun_struct *tun = tun_get(tfile);
1967 result = tun_get_user(tun, tfile, NULL, from,
1968 file->f_flags & O_NONBLOCK, false);
1974 static ssize_t tun_put_user_xdp(struct tun_struct *tun,
1975 struct tun_file *tfile,
1976 struct xdp_frame *xdp_frame,
1977 struct iov_iter *iter)
1979 int vnet_hdr_sz = 0;
1980 size_t size = xdp_frame->len;
1981 struct tun_pcpu_stats *stats;
1984 if (tun->flags & IFF_VNET_HDR) {
1985 struct virtio_net_hdr gso = { 0 };
1987 vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
1988 if (unlikely(iov_iter_count(iter) < vnet_hdr_sz))
1990 if (unlikely(copy_to_iter(&gso, sizeof(gso), iter) !=
1993 iov_iter_advance(iter, vnet_hdr_sz - sizeof(gso));
1996 ret = copy_to_iter(xdp_frame->data, size, iter) + vnet_hdr_sz;
1998 stats = get_cpu_ptr(tun->pcpu_stats);
1999 u64_stats_update_begin(&stats->syncp);
2000 stats->tx_packets++;
2001 stats->tx_bytes += ret;
2002 u64_stats_update_end(&stats->syncp);
2003 put_cpu_ptr(tun->pcpu_stats);
2008 /* Put packet to the user space buffer */
2009 static ssize_t tun_put_user(struct tun_struct *tun,
2010 struct tun_file *tfile,
2011 struct sk_buff *skb,
2012 struct iov_iter *iter)
2014 struct tun_pi pi = { 0, skb->protocol };
2015 struct tun_pcpu_stats *stats;
2017 int vlan_offset = 0;
2019 int vnet_hdr_sz = 0;
2021 if (skb_vlan_tag_present(skb))
2022 vlan_hlen = VLAN_HLEN;
2024 if (tun->flags & IFF_VNET_HDR)
2025 vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
2027 total = skb->len + vlan_hlen + vnet_hdr_sz;
2029 if (!(tun->flags & IFF_NO_PI)) {
2030 if (iov_iter_count(iter) < sizeof(pi))
2033 total += sizeof(pi);
2034 if (iov_iter_count(iter) < total) {
2035 /* Packet will be striped */
2036 pi.flags |= TUN_PKT_STRIP;
2039 if (copy_to_iter(&pi, sizeof(pi), iter) != sizeof(pi))
2044 struct virtio_net_hdr gso;
2046 if (iov_iter_count(iter) < vnet_hdr_sz)
2049 if (virtio_net_hdr_from_skb(skb, &gso,
2050 tun_is_little_endian(tun), true,
2052 struct skb_shared_info *sinfo = skb_shinfo(skb);
2053 pr_err("unexpected GSO type: "
2054 "0x%x, gso_size %d, hdr_len %d\n",
2055 sinfo->gso_type, tun16_to_cpu(tun, gso.gso_size),
2056 tun16_to_cpu(tun, gso.hdr_len));
2057 print_hex_dump(KERN_ERR, "tun: ",
2060 min((int)tun16_to_cpu(tun, gso.hdr_len), 64), true);
2065 if (copy_to_iter(&gso, sizeof(gso), iter) != sizeof(gso))
2068 iov_iter_advance(iter, vnet_hdr_sz - sizeof(gso));
2075 veth.h_vlan_proto = skb->vlan_proto;
2076 veth.h_vlan_TCI = htons(skb_vlan_tag_get(skb));
2078 vlan_offset = offsetof(struct vlan_ethhdr, h_vlan_proto);
2080 ret = skb_copy_datagram_iter(skb, 0, iter, vlan_offset);
2081 if (ret || !iov_iter_count(iter))
2084 ret = copy_to_iter(&veth, sizeof(veth), iter);
2085 if (ret != sizeof(veth) || !iov_iter_count(iter))
2089 skb_copy_datagram_iter(skb, vlan_offset, iter, skb->len - vlan_offset);
2092 /* caller is in process context, */
2093 stats = get_cpu_ptr(tun->pcpu_stats);
2094 u64_stats_update_begin(&stats->syncp);
2095 stats->tx_packets++;
2096 stats->tx_bytes += skb->len + vlan_hlen;
2097 u64_stats_update_end(&stats->syncp);
2098 put_cpu_ptr(tun->pcpu_stats);
2103 static void *tun_ring_recv(struct tun_file *tfile, int noblock, int *err)
2105 DECLARE_WAITQUEUE(wait, current);
2109 ptr = ptr_ring_consume(&tfile->tx_ring);
2117 add_wait_queue(&tfile->wq.wait, &wait);
2118 current->state = TASK_INTERRUPTIBLE;
2121 ptr = ptr_ring_consume(&tfile->tx_ring);
2124 if (signal_pending(current)) {
2125 error = -ERESTARTSYS;
2128 if (tfile->socket.sk->sk_shutdown & RCV_SHUTDOWN) {
2136 current->state = TASK_RUNNING;
2137 remove_wait_queue(&tfile->wq.wait, &wait);
2144 static ssize_t tun_do_read(struct tun_struct *tun, struct tun_file *tfile,
2145 struct iov_iter *to,
2146 int noblock, void *ptr)
2151 tun_debug(KERN_INFO, tun, "tun_do_read\n");
2153 if (!iov_iter_count(to)) {
2159 /* Read frames from ring */
2160 ptr = tun_ring_recv(tfile, noblock, &err);
2165 if (tun_is_xdp_frame(ptr)) {
2166 struct xdp_frame *xdpf = tun_ptr_to_xdp(ptr);
2168 ret = tun_put_user_xdp(tun, tfile, xdpf, to);
2169 xdp_return_frame(xdpf);
2171 struct sk_buff *skb = ptr;
2173 ret = tun_put_user(tun, tfile, skb, to);
2174 if (unlikely(ret < 0))
2183 static ssize_t tun_chr_read_iter(struct kiocb *iocb, struct iov_iter *to)
2185 struct file *file = iocb->ki_filp;
2186 struct tun_file *tfile = file->private_data;
2187 struct tun_struct *tun = tun_get(tfile);
2188 ssize_t len = iov_iter_count(to), ret;
2192 ret = tun_do_read(tun, tfile, to, file->f_flags & O_NONBLOCK, NULL);
2193 ret = min_t(ssize_t, ret, len);
2200 static void tun_prog_free(struct rcu_head *rcu)
2202 struct tun_prog *prog = container_of(rcu, struct tun_prog, rcu);
2204 bpf_prog_destroy(prog->prog);
2208 static int __tun_set_ebpf(struct tun_struct *tun,
2209 struct tun_prog __rcu **prog_p,
2210 struct bpf_prog *prog)
2212 struct tun_prog *old, *new = NULL;
2215 new = kmalloc(sizeof(*new), GFP_KERNEL);
2221 spin_lock_bh(&tun->lock);
2222 old = rcu_dereference_protected(*prog_p,
2223 lockdep_is_held(&tun->lock));
2224 rcu_assign_pointer(*prog_p, new);
2225 spin_unlock_bh(&tun->lock);
2228 call_rcu(&old->rcu, tun_prog_free);
2233 static void tun_free_netdev(struct net_device *dev)
2235 struct tun_struct *tun = netdev_priv(dev);
2237 BUG_ON(!(list_empty(&tun->disabled)));
2238 free_percpu(tun->pcpu_stats);
2239 tun_flow_uninit(tun);
2240 security_tun_dev_free_security(tun->security);
2241 __tun_set_ebpf(tun, &tun->steering_prog, NULL);
2242 __tun_set_ebpf(tun, &tun->filter_prog, NULL);
2245 static void tun_setup(struct net_device *dev)
2247 struct tun_struct *tun = netdev_priv(dev);
2249 tun->owner = INVALID_UID;
2250 tun->group = INVALID_GID;
2251 tun_default_link_ksettings(dev, &tun->link_ksettings);
2253 dev->ethtool_ops = &tun_ethtool_ops;
2254 dev->needs_free_netdev = true;
2255 dev->priv_destructor = tun_free_netdev;
2256 /* We prefer our own queue length */
2257 dev->tx_queue_len = TUN_READQ_SIZE;
2260 /* Trivial set of netlink ops to allow deleting tun or tap
2261 * device with netlink.
2263 static int tun_validate(struct nlattr *tb[], struct nlattr *data[],
2264 struct netlink_ext_ack *extack)
2269 static size_t tun_get_size(const struct net_device *dev)
2271 BUILD_BUG_ON(sizeof(u32) != sizeof(uid_t));
2272 BUILD_BUG_ON(sizeof(u32) != sizeof(gid_t));
2274 return nla_total_size(sizeof(uid_t)) + /* OWNER */
2275 nla_total_size(sizeof(gid_t)) + /* GROUP */
2276 nla_total_size(sizeof(u8)) + /* TYPE */
2277 nla_total_size(sizeof(u8)) + /* PI */
2278 nla_total_size(sizeof(u8)) + /* VNET_HDR */
2279 nla_total_size(sizeof(u8)) + /* PERSIST */
2280 nla_total_size(sizeof(u8)) + /* MULTI_QUEUE */
2281 nla_total_size(sizeof(u32)) + /* NUM_QUEUES */
2282 nla_total_size(sizeof(u32)) + /* NUM_DISABLED_QUEUES */
2286 static int tun_fill_info(struct sk_buff *skb, const struct net_device *dev)
2288 struct tun_struct *tun = netdev_priv(dev);
2290 if (nla_put_u8(skb, IFLA_TUN_TYPE, tun->flags & TUN_TYPE_MASK))
2291 goto nla_put_failure;
2292 if (uid_valid(tun->owner) &&
2293 nla_put_u32(skb, IFLA_TUN_OWNER,
2294 from_kuid_munged(current_user_ns(), tun->owner)))
2295 goto nla_put_failure;
2296 if (gid_valid(tun->group) &&
2297 nla_put_u32(skb, IFLA_TUN_GROUP,
2298 from_kgid_munged(current_user_ns(), tun->group)))
2299 goto nla_put_failure;
2300 if (nla_put_u8(skb, IFLA_TUN_PI, !(tun->flags & IFF_NO_PI)))
2301 goto nla_put_failure;
2302 if (nla_put_u8(skb, IFLA_TUN_VNET_HDR, !!(tun->flags & IFF_VNET_HDR)))
2303 goto nla_put_failure;
2304 if (nla_put_u8(skb, IFLA_TUN_PERSIST, !!(tun->flags & IFF_PERSIST)))
2305 goto nla_put_failure;
2306 if (nla_put_u8(skb, IFLA_TUN_MULTI_QUEUE,
2307 !!(tun->flags & IFF_MULTI_QUEUE)))
2308 goto nla_put_failure;
2309 if (tun->flags & IFF_MULTI_QUEUE) {
2310 if (nla_put_u32(skb, IFLA_TUN_NUM_QUEUES, tun->numqueues))
2311 goto nla_put_failure;
2312 if (nla_put_u32(skb, IFLA_TUN_NUM_DISABLED_QUEUES,
2314 goto nla_put_failure;
2323 static struct rtnl_link_ops tun_link_ops __read_mostly = {
2325 .priv_size = sizeof(struct tun_struct),
2327 .validate = tun_validate,
2328 .get_size = tun_get_size,
2329 .fill_info = tun_fill_info,
2332 static void tun_sock_write_space(struct sock *sk)
2334 struct tun_file *tfile;
2335 wait_queue_head_t *wqueue;
2337 if (!sock_writeable(sk))
2340 if (!test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags))
2343 wqueue = sk_sleep(sk);
2344 if (wqueue && waitqueue_active(wqueue))
2345 wake_up_interruptible_sync_poll(wqueue, EPOLLOUT |
2346 EPOLLWRNORM | EPOLLWRBAND);
2348 tfile = container_of(sk, struct tun_file, sk);
2349 kill_fasync(&tfile->fasync, SIGIO, POLL_OUT);
2352 static int tun_sendmsg(struct socket *sock, struct msghdr *m, size_t total_len)
2355 struct tun_file *tfile = container_of(sock, struct tun_file, socket);
2356 struct tun_struct *tun = tun_get(tfile);
2361 ret = tun_get_user(tun, tfile, m->msg_control, &m->msg_iter,
2362 m->msg_flags & MSG_DONTWAIT,
2363 m->msg_flags & MSG_MORE);
2368 static int tun_recvmsg(struct socket *sock, struct msghdr *m, size_t total_len,
2371 struct tun_file *tfile = container_of(sock, struct tun_file, socket);
2372 struct tun_struct *tun = tun_get(tfile);
2373 void *ptr = m->msg_control;
2381 if (flags & ~(MSG_DONTWAIT|MSG_TRUNC|MSG_ERRQUEUE)) {
2385 if (flags & MSG_ERRQUEUE) {
2386 ret = sock_recv_errqueue(sock->sk, m, total_len,
2387 SOL_PACKET, TUN_TX_TIMESTAMP);
2390 ret = tun_do_read(tun, tfile, &m->msg_iter, flags & MSG_DONTWAIT, ptr);
2391 if (ret > (ssize_t)total_len) {
2392 m->msg_flags |= MSG_TRUNC;
2393 ret = flags & MSG_TRUNC ? ret : total_len;
2406 static int tun_ptr_peek_len(void *ptr)
2409 if (tun_is_xdp_frame(ptr)) {
2410 struct xdp_frame *xdpf = tun_ptr_to_xdp(ptr);
2414 return __skb_array_len_with_tag(ptr);
2420 static int tun_peek_len(struct socket *sock)
2422 struct tun_file *tfile = container_of(sock, struct tun_file, socket);
2423 struct tun_struct *tun;
2426 tun = tun_get(tfile);
2430 ret = PTR_RING_PEEK_CALL(&tfile->tx_ring, tun_ptr_peek_len);
2436 /* Ops structure to mimic raw sockets with tun */
2437 static const struct proto_ops tun_socket_ops = {
2438 .peek_len = tun_peek_len,
2439 .sendmsg = tun_sendmsg,
2440 .recvmsg = tun_recvmsg,
2443 static struct proto tun_proto = {
2445 .owner = THIS_MODULE,
2446 .obj_size = sizeof(struct tun_file),
2449 static int tun_flags(struct tun_struct *tun)
2451 return tun->flags & (TUN_FEATURES | IFF_PERSIST | IFF_TUN | IFF_TAP);
2454 static ssize_t tun_show_flags(struct device *dev, struct device_attribute *attr,
2457 struct tun_struct *tun = netdev_priv(to_net_dev(dev));
2458 return sprintf(buf, "0x%x\n", tun_flags(tun));
2461 static ssize_t tun_show_owner(struct device *dev, struct device_attribute *attr,
2464 struct tun_struct *tun = netdev_priv(to_net_dev(dev));
2465 return uid_valid(tun->owner)?
2466 sprintf(buf, "%u\n",
2467 from_kuid_munged(current_user_ns(), tun->owner)):
2468 sprintf(buf, "-1\n");
2471 static ssize_t tun_show_group(struct device *dev, struct device_attribute *attr,
2474 struct tun_struct *tun = netdev_priv(to_net_dev(dev));
2475 return gid_valid(tun->group) ?
2476 sprintf(buf, "%u\n",
2477 from_kgid_munged(current_user_ns(), tun->group)):
2478 sprintf(buf, "-1\n");
2481 static DEVICE_ATTR(tun_flags, 0444, tun_show_flags, NULL);
2482 static DEVICE_ATTR(owner, 0444, tun_show_owner, NULL);
2483 static DEVICE_ATTR(group, 0444, tun_show_group, NULL);
2485 static struct attribute *tun_dev_attrs[] = {
2486 &dev_attr_tun_flags.attr,
2487 &dev_attr_owner.attr,
2488 &dev_attr_group.attr,
2492 static const struct attribute_group tun_attr_group = {
2493 .attrs = tun_dev_attrs
2496 static int tun_set_iff(struct net *net, struct file *file, struct ifreq *ifr)
2498 struct tun_struct *tun;
2499 struct tun_file *tfile = file->private_data;
2500 struct net_device *dev;
2503 if (tfile->detached)
2506 if ((ifr->ifr_flags & IFF_NAPI_FRAGS)) {
2507 if (!capable(CAP_NET_ADMIN))
2510 if (!(ifr->ifr_flags & IFF_NAPI) ||
2511 (ifr->ifr_flags & TUN_TYPE_MASK) != IFF_TAP)
2515 dev = __dev_get_by_name(net, ifr->ifr_name);
2517 if (ifr->ifr_flags & IFF_TUN_EXCL)
2519 if ((ifr->ifr_flags & IFF_TUN) && dev->netdev_ops == &tun_netdev_ops)
2520 tun = netdev_priv(dev);
2521 else if ((ifr->ifr_flags & IFF_TAP) && dev->netdev_ops == &tap_netdev_ops)
2522 tun = netdev_priv(dev);
2526 if (!!(ifr->ifr_flags & IFF_MULTI_QUEUE) !=
2527 !!(tun->flags & IFF_MULTI_QUEUE))
2530 if (tun_not_capable(tun))
2532 err = security_tun_dev_open(tun->security);
2536 err = tun_attach(tun, file, ifr->ifr_flags & IFF_NOFILTER,
2537 ifr->ifr_flags & IFF_NAPI);
2541 if (tun->flags & IFF_MULTI_QUEUE &&
2542 (tun->numqueues + tun->numdisabled > 1)) {
2543 /* One or more queue has already been attached, no need
2544 * to initialize the device again.
2546 netdev_state_change(dev);
2550 tun->flags = (tun->flags & ~TUN_FEATURES) |
2551 (ifr->ifr_flags & TUN_FEATURES);
2553 netdev_state_change(dev);
2556 unsigned long flags = 0;
2557 int queues = ifr->ifr_flags & IFF_MULTI_QUEUE ?
2560 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
2562 err = security_tun_dev_create();
2567 if (ifr->ifr_flags & IFF_TUN) {
2571 } else if (ifr->ifr_flags & IFF_TAP) {
2579 name = ifr->ifr_name;
2581 dev = alloc_netdev_mqs(sizeof(struct tun_struct), name,
2582 NET_NAME_UNKNOWN, tun_setup, queues,
2587 err = dev_get_valid_name(net, dev, name);
2591 dev_net_set(dev, net);
2592 dev->rtnl_link_ops = &tun_link_ops;
2593 dev->ifindex = tfile->ifindex;
2594 dev->sysfs_groups[0] = &tun_attr_group;
2596 tun = netdev_priv(dev);
2599 tun->txflt.count = 0;
2600 tun->vnet_hdr_sz = sizeof(struct virtio_net_hdr);
2602 tun->align = NET_SKB_PAD;
2603 tun->filter_attached = false;
2604 tun->sndbuf = tfile->socket.sk->sk_sndbuf;
2605 tun->rx_batched = 0;
2606 RCU_INIT_POINTER(tun->steering_prog, NULL);
2608 tun->pcpu_stats = netdev_alloc_pcpu_stats(struct tun_pcpu_stats);
2609 if (!tun->pcpu_stats) {
2614 spin_lock_init(&tun->lock);
2616 err = security_tun_dev_alloc_security(&tun->security);
2623 dev->hw_features = NETIF_F_SG | NETIF_F_FRAGLIST |
2624 TUN_USER_FEATURES | NETIF_F_HW_VLAN_CTAG_TX |
2625 NETIF_F_HW_VLAN_STAG_TX;
2626 dev->features = dev->hw_features | NETIF_F_LLTX;
2627 dev->vlan_features = dev->features &
2628 ~(NETIF_F_HW_VLAN_CTAG_TX |
2629 NETIF_F_HW_VLAN_STAG_TX);
2631 tun->flags = (tun->flags & ~TUN_FEATURES) |
2632 (ifr->ifr_flags & TUN_FEATURES);
2634 INIT_LIST_HEAD(&tun->disabled);
2635 err = tun_attach(tun, file, false, ifr->ifr_flags & IFF_NAPI);
2639 err = register_netdevice(tun->dev);
2644 netif_carrier_on(tun->dev);
2646 tun_debug(KERN_INFO, tun, "tun_set_iff\n");
2648 /* Make sure persistent devices do not get stuck in
2651 if (netif_running(tun->dev))
2652 netif_tx_wake_all_queues(tun->dev);
2654 strcpy(ifr->ifr_name, tun->dev->name);
2658 tun_detach_all(dev);
2659 /* register_netdevice() already called tun_free_netdev() */
2663 tun_flow_uninit(tun);
2664 security_tun_dev_free_security(tun->security);
2666 free_percpu(tun->pcpu_stats);
2672 static void tun_get_iff(struct net *net, struct tun_struct *tun,
2675 tun_debug(KERN_INFO, tun, "tun_get_iff\n");
2677 strcpy(ifr->ifr_name, tun->dev->name);
2679 ifr->ifr_flags = tun_flags(tun);
2683 /* This is like a cut-down ethtool ops, except done via tun fd so no
2684 * privs required. */
2685 static int set_offload(struct tun_struct *tun, unsigned long arg)
2687 netdev_features_t features = 0;
2689 if (arg & TUN_F_CSUM) {
2690 features |= NETIF_F_HW_CSUM;
2693 if (arg & (TUN_F_TSO4|TUN_F_TSO6)) {
2694 if (arg & TUN_F_TSO_ECN) {
2695 features |= NETIF_F_TSO_ECN;
2696 arg &= ~TUN_F_TSO_ECN;
2698 if (arg & TUN_F_TSO4)
2699 features |= NETIF_F_TSO;
2700 if (arg & TUN_F_TSO6)
2701 features |= NETIF_F_TSO6;
2702 arg &= ~(TUN_F_TSO4|TUN_F_TSO6);
2708 /* This gives the user a way to test for new features in future by
2709 * trying to set them. */
2713 tun->set_features = features;
2714 tun->dev->wanted_features &= ~TUN_USER_FEATURES;
2715 tun->dev->wanted_features |= features;
2716 netdev_update_features(tun->dev);
2721 static void tun_detach_filter(struct tun_struct *tun, int n)
2724 struct tun_file *tfile;
2726 for (i = 0; i < n; i++) {
2727 tfile = rtnl_dereference(tun->tfiles[i]);
2728 lock_sock(tfile->socket.sk);
2729 sk_detach_filter(tfile->socket.sk);
2730 release_sock(tfile->socket.sk);
2733 tun->filter_attached = false;
2736 static int tun_attach_filter(struct tun_struct *tun)
2739 struct tun_file *tfile;
2741 for (i = 0; i < tun->numqueues; i++) {
2742 tfile = rtnl_dereference(tun->tfiles[i]);
2743 lock_sock(tfile->socket.sk);
2744 ret = sk_attach_filter(&tun->fprog, tfile->socket.sk);
2745 release_sock(tfile->socket.sk);
2747 tun_detach_filter(tun, i);
2752 tun->filter_attached = true;
2756 static void tun_set_sndbuf(struct tun_struct *tun)
2758 struct tun_file *tfile;
2761 for (i = 0; i < tun->numqueues; i++) {
2762 tfile = rtnl_dereference(tun->tfiles[i]);
2763 tfile->socket.sk->sk_sndbuf = tun->sndbuf;
2767 static int tun_set_queue(struct file *file, struct ifreq *ifr)
2769 struct tun_file *tfile = file->private_data;
2770 struct tun_struct *tun;
2775 if (ifr->ifr_flags & IFF_ATTACH_QUEUE) {
2776 tun = tfile->detached;
2781 ret = security_tun_dev_attach_queue(tun->security);
2784 ret = tun_attach(tun, file, false, tun->flags & IFF_NAPI);
2785 } else if (ifr->ifr_flags & IFF_DETACH_QUEUE) {
2786 tun = rtnl_dereference(tfile->tun);
2787 if (!tun || !(tun->flags & IFF_MULTI_QUEUE) || tfile->detached)
2790 __tun_detach(tfile, false);
2795 netdev_state_change(tun->dev);
2802 static int tun_set_ebpf(struct tun_struct *tun, struct tun_prog **prog_p,
2805 struct bpf_prog *prog;
2808 if (copy_from_user(&fd, data, sizeof(fd)))
2814 prog = bpf_prog_get_type(fd, BPF_PROG_TYPE_SOCKET_FILTER);
2816 return PTR_ERR(prog);
2819 return __tun_set_ebpf(tun, prog_p, prog);
2822 static long __tun_chr_ioctl(struct file *file, unsigned int cmd,
2823 unsigned long arg, int ifreq_len)
2825 struct tun_file *tfile = file->private_data;
2826 struct net *net = sock_net(&tfile->sk);
2827 struct tun_struct *tun;
2828 void __user* argp = (void __user*)arg;
2834 unsigned int ifindex;
2837 bool do_notify = false;
2839 if (cmd == TUNSETIFF || cmd == TUNSETQUEUE ||
2840 (_IOC_TYPE(cmd) == SOCK_IOC_TYPE && cmd != SIOCGSKNS)) {
2841 if (copy_from_user(&ifr, argp, ifreq_len))
2844 memset(&ifr, 0, sizeof(ifr));
2846 if (cmd == TUNGETFEATURES) {
2847 /* Currently this just means: "what IFF flags are valid?".
2848 * This is needed because we never checked for invalid flags on
2851 return put_user(IFF_TUN | IFF_TAP | TUN_FEATURES,
2852 (unsigned int __user*)argp);
2853 } else if (cmd == TUNSETQUEUE) {
2854 return tun_set_queue(file, &ifr);
2855 } else if (cmd == SIOCGSKNS) {
2856 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
2858 return open_related_ns(&net->ns, get_net_ns);
2864 tun = tun_get(tfile);
2865 if (cmd == TUNSETIFF) {
2870 ifr.ifr_name[IFNAMSIZ-1] = '\0';
2872 ret = tun_set_iff(net, file, &ifr);
2877 if (copy_to_user(argp, &ifr, ifreq_len))
2881 if (cmd == TUNSETIFINDEX) {
2887 if (copy_from_user(&ifindex, argp, sizeof(ifindex)))
2891 tfile->ifindex = ifindex;
2899 tun_debug(KERN_INFO, tun, "tun_chr_ioctl cmd %u\n", cmd);
2904 tun_get_iff(current->nsproxy->net_ns, tun, &ifr);
2906 if (tfile->detached)
2907 ifr.ifr_flags |= IFF_DETACH_QUEUE;
2908 if (!tfile->socket.sk->sk_filter)
2909 ifr.ifr_flags |= IFF_NOFILTER;
2911 if (copy_to_user(argp, &ifr, ifreq_len))
2916 /* Disable/Enable checksum */
2918 /* [unimplemented] */
2919 tun_debug(KERN_INFO, tun, "ignored: set checksum %s\n",
2920 arg ? "disabled" : "enabled");
2924 /* Disable/Enable persist mode. Keep an extra reference to the
2925 * module to prevent the module being unprobed.
2927 if (arg && !(tun->flags & IFF_PERSIST)) {
2928 tun->flags |= IFF_PERSIST;
2929 __module_get(THIS_MODULE);
2932 if (!arg && (tun->flags & IFF_PERSIST)) {
2933 tun->flags &= ~IFF_PERSIST;
2934 module_put(THIS_MODULE);
2938 tun_debug(KERN_INFO, tun, "persist %s\n",
2939 arg ? "enabled" : "disabled");
2943 /* Set owner of the device */
2944 owner = make_kuid(current_user_ns(), arg);
2945 if (!uid_valid(owner)) {
2951 tun_debug(KERN_INFO, tun, "owner set to %u\n",
2952 from_kuid(&init_user_ns, tun->owner));
2956 /* Set group of the device */
2957 group = make_kgid(current_user_ns(), arg);
2958 if (!gid_valid(group)) {
2964 tun_debug(KERN_INFO, tun, "group set to %u\n",
2965 from_kgid(&init_user_ns, tun->group));
2969 /* Only allow setting the type when the interface is down */
2970 if (tun->dev->flags & IFF_UP) {
2971 tun_debug(KERN_INFO, tun,
2972 "Linktype set failed because interface is up\n");
2975 tun->dev->type = (int) arg;
2976 tun_debug(KERN_INFO, tun, "linktype set to %d\n",
2988 ret = set_offload(tun, arg);
2991 case TUNSETTXFILTER:
2992 /* Can be set only for TAPs */
2994 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
2996 ret = update_filter(&tun->txflt, (void __user *)arg);
3000 /* Get hw address */
3001 memcpy(ifr.ifr_hwaddr.sa_data, tun->dev->dev_addr, ETH_ALEN);
3002 ifr.ifr_hwaddr.sa_family = tun->dev->type;
3003 if (copy_to_user(argp, &ifr, ifreq_len))
3008 /* Set hw address */
3009 tun_debug(KERN_DEBUG, tun, "set hw address: %pM\n",
3010 ifr.ifr_hwaddr.sa_data);
3012 ret = dev_set_mac_address(tun->dev, &ifr.ifr_hwaddr);
3016 sndbuf = tfile->socket.sk->sk_sndbuf;
3017 if (copy_to_user(argp, &sndbuf, sizeof(sndbuf)))
3022 if (copy_from_user(&sndbuf, argp, sizeof(sndbuf))) {
3031 tun->sndbuf = sndbuf;
3032 tun_set_sndbuf(tun);
3035 case TUNGETVNETHDRSZ:
3036 vnet_hdr_sz = tun->vnet_hdr_sz;
3037 if (copy_to_user(argp, &vnet_hdr_sz, sizeof(vnet_hdr_sz)))
3041 case TUNSETVNETHDRSZ:
3042 if (copy_from_user(&vnet_hdr_sz, argp, sizeof(vnet_hdr_sz))) {
3046 if (vnet_hdr_sz < (int)sizeof(struct virtio_net_hdr)) {
3051 tun->vnet_hdr_sz = vnet_hdr_sz;
3055 le = !!(tun->flags & TUN_VNET_LE);
3056 if (put_user(le, (int __user *)argp))
3061 if (get_user(le, (int __user *)argp)) {
3066 tun->flags |= TUN_VNET_LE;
3068 tun->flags &= ~TUN_VNET_LE;
3072 ret = tun_get_vnet_be(tun, argp);
3076 ret = tun_set_vnet_be(tun, argp);
3079 case TUNATTACHFILTER:
3080 /* Can be set only for TAPs */
3082 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3085 if (copy_from_user(&tun->fprog, argp, sizeof(tun->fprog)))
3088 ret = tun_attach_filter(tun);
3091 case TUNDETACHFILTER:
3092 /* Can be set only for TAPs */
3094 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3097 tun_detach_filter(tun, tun->numqueues);
3102 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3105 if (copy_to_user(argp, &tun->fprog, sizeof(tun->fprog)))
3110 case TUNSETSTEERINGEBPF:
3111 ret = tun_set_ebpf(tun, &tun->steering_prog, argp);
3114 case TUNSETFILTEREBPF:
3115 ret = tun_set_ebpf(tun, &tun->filter_prog, argp);
3124 netdev_state_change(tun->dev);
3133 static long tun_chr_ioctl(struct file *file,
3134 unsigned int cmd, unsigned long arg)
3136 return __tun_chr_ioctl(file, cmd, arg, sizeof (struct ifreq));
3139 #ifdef CONFIG_COMPAT
3140 static long tun_chr_compat_ioctl(struct file *file,
3141 unsigned int cmd, unsigned long arg)
3146 case TUNSETTXFILTER:
3151 arg = (unsigned long)compat_ptr(arg);
3154 arg = (compat_ulong_t)arg;
3159 * compat_ifreq is shorter than ifreq, so we must not access beyond
3160 * the end of that structure. All fields that are used in this
3161 * driver are compatible though, we don't need to convert the
3164 return __tun_chr_ioctl(file, cmd, arg, sizeof(struct compat_ifreq));
3166 #endif /* CONFIG_COMPAT */
3168 static int tun_chr_fasync(int fd, struct file *file, int on)
3170 struct tun_file *tfile = file->private_data;
3173 if ((ret = fasync_helper(fd, file, on, &tfile->fasync)) < 0)
3177 __f_setown(file, task_pid(current), PIDTYPE_TGID, 0);
3178 tfile->flags |= TUN_FASYNC;
3180 tfile->flags &= ~TUN_FASYNC;
3186 static int tun_chr_open(struct inode *inode, struct file * file)
3188 struct net *net = current->nsproxy->net_ns;
3189 struct tun_file *tfile;
3191 DBG1(KERN_INFO, "tunX: tun_chr_open\n");
3193 tfile = (struct tun_file *)sk_alloc(net, AF_UNSPEC, GFP_KERNEL,
3197 if (ptr_ring_init(&tfile->tx_ring, 0, GFP_KERNEL)) {
3198 sk_free(&tfile->sk);
3202 RCU_INIT_POINTER(tfile->tun, NULL);
3206 init_waitqueue_head(&tfile->wq.wait);
3207 RCU_INIT_POINTER(tfile->socket.wq, &tfile->wq);
3209 tfile->socket.file = file;
3210 tfile->socket.ops = &tun_socket_ops;
3212 sock_init_data(&tfile->socket, &tfile->sk);
3214 tfile->sk.sk_write_space = tun_sock_write_space;
3215 tfile->sk.sk_sndbuf = INT_MAX;
3217 file->private_data = tfile;
3218 INIT_LIST_HEAD(&tfile->next);
3220 sock_set_flag(&tfile->sk, SOCK_ZEROCOPY);
3225 static int tun_chr_close(struct inode *inode, struct file *file)
3227 struct tun_file *tfile = file->private_data;
3229 tun_detach(tfile, true);
3234 #ifdef CONFIG_PROC_FS
3235 static void tun_chr_show_fdinfo(struct seq_file *m, struct file *file)
3237 struct tun_file *tfile = file->private_data;
3238 struct tun_struct *tun;
3241 memset(&ifr, 0, sizeof(ifr));
3244 tun = tun_get(tfile);
3246 tun_get_iff(current->nsproxy->net_ns, tun, &ifr);
3252 seq_printf(m, "iff:\t%s\n", ifr.ifr_name);
3256 static const struct file_operations tun_fops = {
3257 .owner = THIS_MODULE,
3258 .llseek = no_llseek,
3259 .read_iter = tun_chr_read_iter,
3260 .write_iter = tun_chr_write_iter,
3261 .poll = tun_chr_poll,
3262 .unlocked_ioctl = tun_chr_ioctl,
3263 #ifdef CONFIG_COMPAT
3264 .compat_ioctl = tun_chr_compat_ioctl,
3266 .open = tun_chr_open,
3267 .release = tun_chr_close,
3268 .fasync = tun_chr_fasync,
3269 #ifdef CONFIG_PROC_FS
3270 .show_fdinfo = tun_chr_show_fdinfo,
3274 static struct miscdevice tun_miscdev = {
3277 .nodename = "net/tun",
3281 /* ethtool interface */
3283 static void tun_default_link_ksettings(struct net_device *dev,
3284 struct ethtool_link_ksettings *cmd)
3286 ethtool_link_ksettings_zero_link_mode(cmd, supported);
3287 ethtool_link_ksettings_zero_link_mode(cmd, advertising);
3288 cmd->base.speed = SPEED_10;
3289 cmd->base.duplex = DUPLEX_FULL;
3290 cmd->base.port = PORT_TP;
3291 cmd->base.phy_address = 0;
3292 cmd->base.autoneg = AUTONEG_DISABLE;
3295 static int tun_get_link_ksettings(struct net_device *dev,
3296 struct ethtool_link_ksettings *cmd)
3298 struct tun_struct *tun = netdev_priv(dev);
3300 memcpy(cmd, &tun->link_ksettings, sizeof(*cmd));
3304 static int tun_set_link_ksettings(struct net_device *dev,
3305 const struct ethtool_link_ksettings *cmd)
3307 struct tun_struct *tun = netdev_priv(dev);
3309 memcpy(&tun->link_ksettings, cmd, sizeof(*cmd));
3313 static void tun_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
3315 struct tun_struct *tun = netdev_priv(dev);
3317 strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
3318 strlcpy(info->version, DRV_VERSION, sizeof(info->version));
3320 switch (tun->flags & TUN_TYPE_MASK) {
3322 strlcpy(info->bus_info, "tun", sizeof(info->bus_info));
3325 strlcpy(info->bus_info, "tap", sizeof(info->bus_info));
3330 static u32 tun_get_msglevel(struct net_device *dev)
3333 struct tun_struct *tun = netdev_priv(dev);
3340 static void tun_set_msglevel(struct net_device *dev, u32 value)
3343 struct tun_struct *tun = netdev_priv(dev);
3348 static int tun_get_coalesce(struct net_device *dev,
3349 struct ethtool_coalesce *ec)
3351 struct tun_struct *tun = netdev_priv(dev);
3353 ec->rx_max_coalesced_frames = tun->rx_batched;
3358 static int tun_set_coalesce(struct net_device *dev,
3359 struct ethtool_coalesce *ec)
3361 struct tun_struct *tun = netdev_priv(dev);
3363 if (ec->rx_max_coalesced_frames > NAPI_POLL_WEIGHT)
3364 tun->rx_batched = NAPI_POLL_WEIGHT;
3366 tun->rx_batched = ec->rx_max_coalesced_frames;
3371 static const struct ethtool_ops tun_ethtool_ops = {
3372 .get_drvinfo = tun_get_drvinfo,
3373 .get_msglevel = tun_get_msglevel,
3374 .set_msglevel = tun_set_msglevel,
3375 .get_link = ethtool_op_get_link,
3376 .get_ts_info = ethtool_op_get_ts_info,
3377 .get_coalesce = tun_get_coalesce,
3378 .set_coalesce = tun_set_coalesce,
3379 .get_link_ksettings = tun_get_link_ksettings,
3380 .set_link_ksettings = tun_set_link_ksettings,
3383 static int tun_queue_resize(struct tun_struct *tun)
3385 struct net_device *dev = tun->dev;
3386 struct tun_file *tfile;
3387 struct ptr_ring **rings;
3388 int n = tun->numqueues + tun->numdisabled;
3391 rings = kmalloc_array(n, sizeof(*rings), GFP_KERNEL);
3395 for (i = 0; i < tun->numqueues; i++) {
3396 tfile = rtnl_dereference(tun->tfiles[i]);
3397 rings[i] = &tfile->tx_ring;
3399 list_for_each_entry(tfile, &tun->disabled, next)
3400 rings[i++] = &tfile->tx_ring;
3402 ret = ptr_ring_resize_multiple(rings, n,
3403 dev->tx_queue_len, GFP_KERNEL,
3410 static int tun_device_event(struct notifier_block *unused,
3411 unsigned long event, void *ptr)
3413 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
3414 struct tun_struct *tun = netdev_priv(dev);
3416 if (dev->rtnl_link_ops != &tun_link_ops)
3420 case NETDEV_CHANGE_TX_QUEUE_LEN:
3421 if (tun_queue_resize(tun))
3431 static struct notifier_block tun_notifier_block __read_mostly = {
3432 .notifier_call = tun_device_event,
3435 static int __init tun_init(void)
3439 pr_info("%s, %s\n", DRV_DESCRIPTION, DRV_VERSION);
3441 ret = rtnl_link_register(&tun_link_ops);
3443 pr_err("Can't register link_ops\n");
3447 ret = misc_register(&tun_miscdev);
3449 pr_err("Can't register misc device %d\n", TUN_MINOR);
3453 ret = register_netdevice_notifier(&tun_notifier_block);
3455 pr_err("Can't register netdevice notifier\n");
3462 misc_deregister(&tun_miscdev);
3464 rtnl_link_unregister(&tun_link_ops);
3469 static void tun_cleanup(void)
3471 misc_deregister(&tun_miscdev);
3472 rtnl_link_unregister(&tun_link_ops);
3473 unregister_netdevice_notifier(&tun_notifier_block);
3476 /* Get an underlying socket object from tun file. Returns error unless file is
3477 * attached to a device. The returned object works like a packet socket, it
3478 * can be used for sock_sendmsg/sock_recvmsg. The caller is responsible for
3479 * holding a reference to the file for as long as the socket is in use. */
3480 struct socket *tun_get_socket(struct file *file)
3482 struct tun_file *tfile;
3483 if (file->f_op != &tun_fops)
3484 return ERR_PTR(-EINVAL);
3485 tfile = file->private_data;
3487 return ERR_PTR(-EBADFD);
3488 return &tfile->socket;
3490 EXPORT_SYMBOL_GPL(tun_get_socket);
3492 struct ptr_ring *tun_get_tx_ring(struct file *file)
3494 struct tun_file *tfile;
3496 if (file->f_op != &tun_fops)
3497 return ERR_PTR(-EINVAL);
3498 tfile = file->private_data;
3500 return ERR_PTR(-EBADFD);
3501 return &tfile->tx_ring;
3503 EXPORT_SYMBOL_GPL(tun_get_tx_ring);
3505 module_init(tun_init);
3506 module_exit(tun_cleanup);
3507 MODULE_DESCRIPTION(DRV_DESCRIPTION);
3508 MODULE_AUTHOR(DRV_COPYRIGHT);
3509 MODULE_LICENSE("GPL");
3510 MODULE_ALIAS_MISCDEV(TUN_MINOR);
3511 MODULE_ALIAS("devname:net/tun");