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);
326 static void tun_napi_disable(struct tun_file *tfile)
328 if (tfile->napi_enabled)
329 napi_disable(&tfile->napi);
332 static void tun_napi_del(struct tun_file *tfile)
334 if (tfile->napi_enabled)
335 netif_napi_del(&tfile->napi);
338 static bool tun_napi_frags_enabled(const struct tun_struct *tun)
340 return READ_ONCE(tun->flags) & IFF_NAPI_FRAGS;
343 #ifdef CONFIG_TUN_VNET_CROSS_LE
344 static inline bool tun_legacy_is_little_endian(struct tun_struct *tun)
346 return tun->flags & TUN_VNET_BE ? false :
347 virtio_legacy_is_little_endian();
350 static long tun_get_vnet_be(struct tun_struct *tun, int __user *argp)
352 int be = !!(tun->flags & TUN_VNET_BE);
354 if (put_user(be, argp))
360 static long tun_set_vnet_be(struct tun_struct *tun, int __user *argp)
364 if (get_user(be, argp))
368 tun->flags |= TUN_VNET_BE;
370 tun->flags &= ~TUN_VNET_BE;
375 static inline bool tun_legacy_is_little_endian(struct tun_struct *tun)
377 return virtio_legacy_is_little_endian();
380 static long tun_get_vnet_be(struct tun_struct *tun, int __user *argp)
385 static long tun_set_vnet_be(struct tun_struct *tun, int __user *argp)
389 #endif /* CONFIG_TUN_VNET_CROSS_LE */
391 static inline bool tun_is_little_endian(struct tun_struct *tun)
393 return tun->flags & TUN_VNET_LE ||
394 tun_legacy_is_little_endian(tun);
397 static inline u16 tun16_to_cpu(struct tun_struct *tun, __virtio16 val)
399 return __virtio16_to_cpu(tun_is_little_endian(tun), val);
402 static inline __virtio16 cpu_to_tun16(struct tun_struct *tun, u16 val)
404 return __cpu_to_virtio16(tun_is_little_endian(tun), val);
407 static inline u32 tun_hashfn(u32 rxhash)
409 return rxhash & TUN_MASK_FLOW_ENTRIES;
412 static struct tun_flow_entry *tun_flow_find(struct hlist_head *head, u32 rxhash)
414 struct tun_flow_entry *e;
416 hlist_for_each_entry_rcu(e, head, hash_link) {
417 if (e->rxhash == rxhash)
423 static struct tun_flow_entry *tun_flow_create(struct tun_struct *tun,
424 struct hlist_head *head,
425 u32 rxhash, u16 queue_index)
427 struct tun_flow_entry *e = kmalloc(sizeof(*e), GFP_ATOMIC);
430 tun_debug(KERN_INFO, tun, "create flow: hash %u index %u\n",
431 rxhash, queue_index);
432 e->updated = jiffies;
435 e->queue_index = queue_index;
437 hlist_add_head_rcu(&e->hash_link, head);
443 static void tun_flow_delete(struct tun_struct *tun, struct tun_flow_entry *e)
445 tun_debug(KERN_INFO, tun, "delete flow: hash %u index %u\n",
446 e->rxhash, e->queue_index);
447 hlist_del_rcu(&e->hash_link);
452 static void tun_flow_flush(struct tun_struct *tun)
456 spin_lock_bh(&tun->lock);
457 for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
458 struct tun_flow_entry *e;
459 struct hlist_node *n;
461 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link)
462 tun_flow_delete(tun, e);
464 spin_unlock_bh(&tun->lock);
467 static void tun_flow_delete_by_queue(struct tun_struct *tun, u16 queue_index)
471 spin_lock_bh(&tun->lock);
472 for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
473 struct tun_flow_entry *e;
474 struct hlist_node *n;
476 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
477 if (e->queue_index == queue_index)
478 tun_flow_delete(tun, e);
481 spin_unlock_bh(&tun->lock);
484 static void tun_flow_cleanup(struct timer_list *t)
486 struct tun_struct *tun = from_timer(tun, t, flow_gc_timer);
487 unsigned long delay = tun->ageing_time;
488 unsigned long next_timer = jiffies + delay;
489 unsigned long count = 0;
492 tun_debug(KERN_INFO, tun, "tun_flow_cleanup\n");
494 spin_lock(&tun->lock);
495 for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
496 struct tun_flow_entry *e;
497 struct hlist_node *n;
499 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
500 unsigned long this_timer;
502 this_timer = e->updated + delay;
503 if (time_before_eq(this_timer, jiffies)) {
504 tun_flow_delete(tun, e);
508 if (time_before(this_timer, next_timer))
509 next_timer = this_timer;
514 mod_timer(&tun->flow_gc_timer, round_jiffies_up(next_timer));
515 spin_unlock(&tun->lock);
518 static void tun_flow_update(struct tun_struct *tun, u32 rxhash,
519 struct tun_file *tfile)
521 struct hlist_head *head;
522 struct tun_flow_entry *e;
523 unsigned long delay = tun->ageing_time;
524 u16 queue_index = tfile->queue_index;
529 head = &tun->flows[tun_hashfn(rxhash)];
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);
555 * Save the hash received in the stack receive path and update the
556 * flow_hash table accordingly.
558 static inline void tun_flow_save_rps_rxhash(struct tun_flow_entry *e, u32 hash)
560 if (unlikely(e->rps_rxhash != hash))
561 e->rps_rxhash = hash;
564 /* We try to identify a flow through its rxhash first. The reason that
565 * we do not check rxq no. is because some cards(e.g 82599), chooses
566 * the rxq based on the txq where the last packet of the flow comes. As
567 * the userspace application move between processors, we may get a
568 * different rxq no. here. If we could not get rxhash, then we would
569 * hope the rxq no. may help here.
571 static u16 tun_automq_select_queue(struct tun_struct *tun, struct sk_buff *skb)
573 struct tun_flow_entry *e;
577 numqueues = READ_ONCE(tun->numqueues);
579 txq = __skb_get_hash_symmetric(skb);
581 e = tun_flow_find(&tun->flows[tun_hashfn(txq)], txq);
583 tun_flow_save_rps_rxhash(e, txq);
584 txq = e->queue_index;
586 /* use multiply and shift instead of expensive divide */
587 txq = ((u64)txq * numqueues) >> 32;
588 } else if (likely(skb_rx_queue_recorded(skb))) {
589 txq = skb_get_rx_queue(skb);
590 while (unlikely(txq >= numqueues))
597 static u16 tun_ebpf_select_queue(struct tun_struct *tun, struct sk_buff *skb)
599 struct tun_prog *prog;
602 prog = rcu_dereference(tun->steering_prog);
604 ret = bpf_prog_run_clear_cb(prog->prog, skb);
606 return ret % tun->numqueues;
609 static u16 tun_select_queue(struct net_device *dev, struct sk_buff *skb,
610 struct net_device *sb_dev,
611 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_frame(ptr)) {
664 struct xdp_frame *xdpf = tun_ptr_to_xdp(ptr);
666 xdp_return_frame(xdpf);
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_detach(struct tun_file *tfile, bool clean)
686 struct tun_file *ntfile;
687 struct tun_struct *tun;
689 tun = rtnl_dereference(tfile->tun);
692 tun_napi_disable(tfile);
696 if (tun && !tfile->detached) {
697 u16 index = tfile->queue_index;
698 BUG_ON(index >= tun->numqueues);
700 rcu_assign_pointer(tun->tfiles[index],
701 tun->tfiles[tun->numqueues - 1]);
702 ntfile = rtnl_dereference(tun->tfiles[index]);
703 ntfile->queue_index = index;
707 RCU_INIT_POINTER(tfile->tun, NULL);
708 sock_put(&tfile->sk);
710 tun_disable_queue(tun, tfile);
713 tun_flow_delete_by_queue(tun, tun->numqueues + 1);
714 /* Drop read queue */
715 tun_queue_purge(tfile);
716 tun_set_real_num_queues(tun);
717 } else if (tfile->detached && clean) {
718 tun = tun_enable_queue(tfile);
719 sock_put(&tfile->sk);
723 if (tun && tun->numqueues == 0 && tun->numdisabled == 0) {
724 netif_carrier_off(tun->dev);
726 if (!(tun->flags & IFF_PERSIST) &&
727 tun->dev->reg_state == NETREG_REGISTERED)
728 unregister_netdevice(tun->dev);
731 xdp_rxq_info_unreg(&tfile->xdp_rxq);
732 ptr_ring_cleanup(&tfile->tx_ring, tun_ptr_free);
733 sock_put(&tfile->sk);
737 static void tun_detach(struct tun_file *tfile, bool clean)
739 struct tun_struct *tun;
740 struct net_device *dev;
743 tun = rtnl_dereference(tfile->tun);
744 dev = tun ? tun->dev : NULL;
745 __tun_detach(tfile, clean);
747 netdev_state_change(dev);
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(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]);
777 /* Drop read queue */
778 tun_queue_purge(tfile);
779 xdp_rxq_info_unreg(&tfile->xdp_rxq);
780 sock_put(&tfile->sk);
782 list_for_each_entry_safe(tfile, tmp, &tun->disabled, next) {
783 tun_enable_queue(tfile);
784 tun_queue_purge(tfile);
785 xdp_rxq_info_unreg(&tfile->xdp_rxq);
786 sock_put(&tfile->sk);
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_resize(&tfile->tx_ring, dev->tx_queue_len,
831 GFP_KERNEL, tun_ptr_free)) {
836 tfile->queue_index = tun->numqueues;
837 tfile->socket.sk->sk_shutdown &= ~RCV_SHUTDOWN;
839 if (tfile->detached) {
840 /* Re-attach detached tfile, updating XDP queue_index */
841 WARN_ON(!xdp_rxq_info_is_reg(&tfile->xdp_rxq));
843 if (tfile->xdp_rxq.queue_index != tfile->queue_index)
844 tfile->xdp_rxq.queue_index = tfile->queue_index;
846 /* Setup XDP RX-queue info, for new tfile getting attached */
847 err = xdp_rxq_info_reg(&tfile->xdp_rxq,
848 tun->dev, tfile->queue_index);
851 err = xdp_rxq_info_reg_mem_model(&tfile->xdp_rxq,
852 MEM_TYPE_PAGE_SHARED, NULL);
854 xdp_rxq_info_unreg(&tfile->xdp_rxq);
860 rcu_assign_pointer(tfile->tun, tun);
861 rcu_assign_pointer(tun->tfiles[tun->numqueues], tfile);
864 if (tfile->detached) {
865 tun_enable_queue(tfile);
867 sock_hold(&tfile->sk);
868 tun_napi_init(tun, tfile, napi);
871 tun_set_real_num_queues(tun);
873 /* device is allowed to go away first, so no need to hold extra
881 static struct tun_struct *tun_get(struct tun_file *tfile)
883 struct tun_struct *tun;
886 tun = rcu_dereference(tfile->tun);
894 static void tun_put(struct tun_struct *tun)
900 static void addr_hash_set(u32 *mask, const u8 *addr)
902 int n = ether_crc(ETH_ALEN, addr) >> 26;
903 mask[n >> 5] |= (1 << (n & 31));
906 static unsigned int addr_hash_test(const u32 *mask, const u8 *addr)
908 int n = ether_crc(ETH_ALEN, addr) >> 26;
909 return mask[n >> 5] & (1 << (n & 31));
912 static int update_filter(struct tap_filter *filter, void __user *arg)
914 struct { u8 u[ETH_ALEN]; } *addr;
915 struct tun_filter uf;
916 int err, alen, n, nexact;
918 if (copy_from_user(&uf, arg, sizeof(uf)))
927 alen = ETH_ALEN * uf.count;
928 addr = memdup_user(arg + sizeof(uf), alen);
930 return PTR_ERR(addr);
932 /* The filter is updated without holding any locks. Which is
933 * perfectly safe. We disable it first and in the worst
934 * case we'll accept a few undesired packets. */
938 /* Use first set of addresses as an exact filter */
939 for (n = 0; n < uf.count && n < FLT_EXACT_COUNT; n++)
940 memcpy(filter->addr[n], addr[n].u, ETH_ALEN);
944 /* Remaining multicast addresses are hashed,
945 * unicast will leave the filter disabled. */
946 memset(filter->mask, 0, sizeof(filter->mask));
947 for (; n < uf.count; n++) {
948 if (!is_multicast_ether_addr(addr[n].u)) {
949 err = 0; /* no filter */
952 addr_hash_set(filter->mask, addr[n].u);
955 /* For ALLMULTI just set the mask to all ones.
956 * This overrides the mask populated above. */
957 if ((uf.flags & TUN_FLT_ALLMULTI))
958 memset(filter->mask, ~0, sizeof(filter->mask));
960 /* Now enable the filter */
962 filter->count = nexact;
964 /* Return the number of exact filters */
971 /* Returns: 0 - drop, !=0 - accept */
972 static int run_filter(struct tap_filter *filter, const struct sk_buff *skb)
974 /* Cannot use eth_hdr(skb) here because skb_mac_hdr() is incorrect
976 struct ethhdr *eh = (struct ethhdr *) skb->data;
980 for (i = 0; i < filter->count; i++)
981 if (ether_addr_equal(eh->h_dest, filter->addr[i]))
984 /* Inexact match (multicast only) */
985 if (is_multicast_ether_addr(eh->h_dest))
986 return addr_hash_test(filter->mask, eh->h_dest);
992 * Checks whether the packet is accepted or not.
993 * Returns: 0 - drop, !=0 - accept
995 static int check_filter(struct tap_filter *filter, const struct sk_buff *skb)
1000 return run_filter(filter, skb);
1003 /* Network device part of the driver */
1005 static const struct ethtool_ops tun_ethtool_ops;
1007 /* Net device detach from fd. */
1008 static void tun_net_uninit(struct net_device *dev)
1010 tun_detach_all(dev);
1013 /* Net device open. */
1014 static int tun_net_open(struct net_device *dev)
1016 struct tun_struct *tun = netdev_priv(dev);
1019 netif_tx_start_all_queues(dev);
1021 for (i = 0; i < tun->numqueues; i++) {
1022 struct tun_file *tfile;
1024 tfile = rtnl_dereference(tun->tfiles[i]);
1025 tfile->socket.sk->sk_write_space(tfile->socket.sk);
1031 /* Net device close. */
1032 static int tun_net_close(struct net_device *dev)
1034 netif_tx_stop_all_queues(dev);
1038 /* Net device start xmit */
1039 static void tun_automq_xmit(struct tun_struct *tun, struct sk_buff *skb)
1042 if (tun->numqueues == 1 && static_key_false(&rps_needed)) {
1043 /* Select queue was not called for the skbuff, so we extract the
1044 * RPS hash and save it into the flow_table here.
1048 rxhash = __skb_get_hash_symmetric(skb);
1050 struct tun_flow_entry *e;
1051 e = tun_flow_find(&tun->flows[tun_hashfn(rxhash)],
1054 tun_flow_save_rps_rxhash(e, rxhash);
1060 static unsigned int run_ebpf_filter(struct tun_struct *tun,
1061 struct sk_buff *skb,
1064 struct tun_prog *prog = rcu_dereference(tun->filter_prog);
1067 len = bpf_prog_run_clear_cb(prog->prog, skb);
1072 /* Net device start xmit */
1073 static netdev_tx_t tun_net_xmit(struct sk_buff *skb, struct net_device *dev)
1075 struct tun_struct *tun = netdev_priv(dev);
1076 int txq = skb->queue_mapping;
1077 struct tun_file *tfile;
1081 tfile = rcu_dereference(tun->tfiles[txq]);
1083 /* Drop packet if interface is not attached */
1084 if (txq >= tun->numqueues)
1087 if (!rcu_dereference(tun->steering_prog))
1088 tun_automq_xmit(tun, skb);
1090 tun_debug(KERN_INFO, tun, "tun_net_xmit %d\n", skb->len);
1094 /* Drop if the filter does not like it.
1095 * This is a noop if the filter is disabled.
1096 * Filter can be enabled only for the TAP devices. */
1097 if (!check_filter(&tun->txflt, skb))
1100 if (tfile->socket.sk->sk_filter &&
1101 sk_filter(tfile->socket.sk, skb))
1104 len = run_ebpf_filter(tun, skb, len);
1105 if (len == 0 || pskb_trim(skb, len))
1108 if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC)))
1111 skb_tx_timestamp(skb);
1113 /* Orphan the skb - required as we might hang on to it
1114 * for indefinite time.
1120 if (ptr_ring_produce(&tfile->tx_ring, skb))
1123 /* Notify and wake up reader process */
1124 if (tfile->flags & TUN_FASYNC)
1125 kill_fasync(&tfile->fasync, SIGIO, POLL_IN);
1126 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
1129 return NETDEV_TX_OK;
1132 this_cpu_inc(tun->pcpu_stats->tx_dropped);
1136 return NET_XMIT_DROP;
1139 static void tun_net_mclist(struct net_device *dev)
1142 * This callback is supposed to deal with mc filter in
1143 * _rx_ path and has nothing to do with the _tx_ path.
1144 * In rx path we always accept everything userspace gives us.
1148 static netdev_features_t tun_net_fix_features(struct net_device *dev,
1149 netdev_features_t features)
1151 struct tun_struct *tun = netdev_priv(dev);
1153 return (features & tun->set_features) | (features & ~TUN_USER_FEATURES);
1156 static void tun_set_headroom(struct net_device *dev, int new_hr)
1158 struct tun_struct *tun = netdev_priv(dev);
1160 if (new_hr < NET_SKB_PAD)
1161 new_hr = NET_SKB_PAD;
1163 tun->align = new_hr;
1167 tun_net_get_stats64(struct net_device *dev, struct rtnl_link_stats64 *stats)
1169 u32 rx_dropped = 0, tx_dropped = 0, rx_frame_errors = 0;
1170 struct tun_struct *tun = netdev_priv(dev);
1171 struct tun_pcpu_stats *p;
1174 for_each_possible_cpu(i) {
1175 u64 rxpackets, rxbytes, txpackets, txbytes;
1178 p = per_cpu_ptr(tun->pcpu_stats, i);
1180 start = u64_stats_fetch_begin(&p->syncp);
1181 rxpackets = p->rx_packets;
1182 rxbytes = p->rx_bytes;
1183 txpackets = p->tx_packets;
1184 txbytes = p->tx_bytes;
1185 } while (u64_stats_fetch_retry(&p->syncp, start));
1187 stats->rx_packets += rxpackets;
1188 stats->rx_bytes += rxbytes;
1189 stats->tx_packets += txpackets;
1190 stats->tx_bytes += txbytes;
1193 rx_dropped += p->rx_dropped;
1194 rx_frame_errors += p->rx_frame_errors;
1195 tx_dropped += p->tx_dropped;
1197 stats->rx_dropped = rx_dropped;
1198 stats->rx_frame_errors = rx_frame_errors;
1199 stats->tx_dropped = tx_dropped;
1202 static int tun_xdp_set(struct net_device *dev, struct bpf_prog *prog,
1203 struct netlink_ext_ack *extack)
1205 struct tun_struct *tun = netdev_priv(dev);
1206 struct bpf_prog *old_prog;
1208 old_prog = rtnl_dereference(tun->xdp_prog);
1209 rcu_assign_pointer(tun->xdp_prog, prog);
1211 bpf_prog_put(old_prog);
1216 static u32 tun_xdp_query(struct net_device *dev)
1218 struct tun_struct *tun = netdev_priv(dev);
1219 const struct bpf_prog *xdp_prog;
1221 xdp_prog = rtnl_dereference(tun->xdp_prog);
1223 return xdp_prog->aux->id;
1228 static int tun_xdp(struct net_device *dev, struct netdev_bpf *xdp)
1230 switch (xdp->command) {
1231 case XDP_SETUP_PROG:
1232 return tun_xdp_set(dev, xdp->prog, xdp->extack);
1233 case XDP_QUERY_PROG:
1234 xdp->prog_id = tun_xdp_query(dev);
1241 static const struct net_device_ops tun_netdev_ops = {
1242 .ndo_uninit = tun_net_uninit,
1243 .ndo_open = tun_net_open,
1244 .ndo_stop = tun_net_close,
1245 .ndo_start_xmit = tun_net_xmit,
1246 .ndo_fix_features = tun_net_fix_features,
1247 .ndo_select_queue = tun_select_queue,
1248 .ndo_set_rx_headroom = tun_set_headroom,
1249 .ndo_get_stats64 = tun_net_get_stats64,
1252 static void __tun_xdp_flush_tfile(struct tun_file *tfile)
1254 /* Notify and wake up reader process */
1255 if (tfile->flags & TUN_FASYNC)
1256 kill_fasync(&tfile->fasync, SIGIO, POLL_IN);
1257 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
1260 static int tun_xdp_xmit(struct net_device *dev, int n,
1261 struct xdp_frame **frames, u32 flags)
1263 struct tun_struct *tun = netdev_priv(dev);
1264 struct tun_file *tfile;
1270 if (unlikely(flags & ~XDP_XMIT_FLAGS_MASK))
1275 numqueues = READ_ONCE(tun->numqueues);
1278 return -ENXIO; /* Caller will free/return all frames */
1281 tfile = rcu_dereference(tun->tfiles[smp_processor_id() %
1284 spin_lock(&tfile->tx_ring.producer_lock);
1285 for (i = 0; i < n; i++) {
1286 struct xdp_frame *xdp = frames[i];
1287 /* Encode the XDP flag into lowest bit for consumer to differ
1288 * XDP buffer from sk_buff.
1290 void *frame = tun_xdp_to_ptr(xdp);
1292 if (__ptr_ring_produce(&tfile->tx_ring, frame)) {
1293 this_cpu_inc(tun->pcpu_stats->tx_dropped);
1294 xdp_return_frame_rx_napi(xdp);
1298 spin_unlock(&tfile->tx_ring.producer_lock);
1300 if (flags & XDP_XMIT_FLUSH)
1301 __tun_xdp_flush_tfile(tfile);
1307 static int tun_xdp_tx(struct net_device *dev, struct xdp_buff *xdp)
1309 struct xdp_frame *frame = convert_to_xdp_frame(xdp);
1311 if (unlikely(!frame))
1314 return tun_xdp_xmit(dev, 1, &frame, XDP_XMIT_FLUSH);
1317 static const struct net_device_ops tap_netdev_ops = {
1318 .ndo_uninit = tun_net_uninit,
1319 .ndo_open = tun_net_open,
1320 .ndo_stop = tun_net_close,
1321 .ndo_start_xmit = tun_net_xmit,
1322 .ndo_fix_features = tun_net_fix_features,
1323 .ndo_set_rx_mode = tun_net_mclist,
1324 .ndo_set_mac_address = eth_mac_addr,
1325 .ndo_validate_addr = eth_validate_addr,
1326 .ndo_select_queue = tun_select_queue,
1327 .ndo_features_check = passthru_features_check,
1328 .ndo_set_rx_headroom = tun_set_headroom,
1329 .ndo_get_stats64 = tun_net_get_stats64,
1331 .ndo_xdp_xmit = tun_xdp_xmit,
1334 static void tun_flow_init(struct tun_struct *tun)
1338 for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++)
1339 INIT_HLIST_HEAD(&tun->flows[i]);
1341 tun->ageing_time = TUN_FLOW_EXPIRE;
1342 timer_setup(&tun->flow_gc_timer, tun_flow_cleanup, 0);
1343 mod_timer(&tun->flow_gc_timer,
1344 round_jiffies_up(jiffies + tun->ageing_time));
1347 static void tun_flow_uninit(struct tun_struct *tun)
1349 del_timer_sync(&tun->flow_gc_timer);
1350 tun_flow_flush(tun);
1354 #define MAX_MTU 65535
1356 /* Initialize net device. */
1357 static void tun_net_init(struct net_device *dev)
1359 struct tun_struct *tun = netdev_priv(dev);
1361 switch (tun->flags & TUN_TYPE_MASK) {
1363 dev->netdev_ops = &tun_netdev_ops;
1365 /* Point-to-Point TUN Device */
1366 dev->hard_header_len = 0;
1370 /* Zero header length */
1371 dev->type = ARPHRD_NONE;
1372 dev->flags = IFF_POINTOPOINT | IFF_NOARP | IFF_MULTICAST;
1376 dev->netdev_ops = &tap_netdev_ops;
1377 /* Ethernet TAP Device */
1379 dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1380 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
1382 eth_hw_addr_random(dev);
1387 dev->min_mtu = MIN_MTU;
1388 dev->max_mtu = MAX_MTU - dev->hard_header_len;
1391 static bool tun_sock_writeable(struct tun_struct *tun, struct tun_file *tfile)
1393 struct sock *sk = tfile->socket.sk;
1395 return (tun->dev->flags & IFF_UP) && sock_writeable(sk);
1398 /* Character device part */
1401 static __poll_t tun_chr_poll(struct file *file, poll_table *wait)
1403 struct tun_file *tfile = file->private_data;
1404 struct tun_struct *tun = tun_get(tfile);
1411 sk = tfile->socket.sk;
1413 tun_debug(KERN_INFO, tun, "tun_chr_poll\n");
1415 poll_wait(file, sk_sleep(sk), wait);
1417 if (!ptr_ring_empty(&tfile->tx_ring))
1418 mask |= EPOLLIN | EPOLLRDNORM;
1420 /* Make sure SOCKWQ_ASYNC_NOSPACE is set if not writable to
1421 * guarantee EPOLLOUT to be raised by either here or
1422 * tun_sock_write_space(). Then process could get notification
1423 * after it writes to a down device and meets -EIO.
1425 if (tun_sock_writeable(tun, tfile) ||
1426 (!test_and_set_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags) &&
1427 tun_sock_writeable(tun, tfile)))
1428 mask |= EPOLLOUT | EPOLLWRNORM;
1430 if (tun->dev->reg_state != NETREG_REGISTERED)
1437 static struct sk_buff *tun_napi_alloc_frags(struct tun_file *tfile,
1439 const struct iov_iter *it)
1441 struct sk_buff *skb;
1446 if (it->nr_segs > MAX_SKB_FRAGS + 1)
1447 return ERR_PTR(-ENOMEM);
1450 skb = napi_get_frags(&tfile->napi);
1453 return ERR_PTR(-ENOMEM);
1455 linear = iov_iter_single_seg_count(it);
1456 err = __skb_grow(skb, linear);
1461 skb->data_len = len - linear;
1462 skb->truesize += skb->data_len;
1464 for (i = 1; i < it->nr_segs; i++) {
1465 struct page_frag *pfrag = ¤t->task_frag;
1466 size_t fragsz = it->iov[i].iov_len;
1468 if (fragsz == 0 || fragsz > PAGE_SIZE) {
1473 if (!skb_page_frag_refill(fragsz, pfrag, GFP_KERNEL)) {
1478 skb_fill_page_desc(skb, i - 1, pfrag->page,
1479 pfrag->offset, fragsz);
1480 page_ref_inc(pfrag->page);
1481 pfrag->offset += fragsz;
1486 /* frees skb and all frags allocated with napi_alloc_frag() */
1487 napi_free_frags(&tfile->napi);
1488 return ERR_PTR(err);
1491 /* prepad is the amount to reserve at front. len is length after that.
1492 * linear is a hint as to how much to copy (usually headers). */
1493 static struct sk_buff *tun_alloc_skb(struct tun_file *tfile,
1494 size_t prepad, size_t len,
1495 size_t linear, int noblock)
1497 struct sock *sk = tfile->socket.sk;
1498 struct sk_buff *skb;
1501 /* Under a page? Don't bother with paged skb. */
1502 if (prepad + len < PAGE_SIZE || !linear)
1505 skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
1508 return ERR_PTR(err);
1510 skb_reserve(skb, prepad);
1511 skb_put(skb, linear);
1512 skb->data_len = len - linear;
1513 skb->len += len - linear;
1518 static void tun_rx_batched(struct tun_struct *tun, struct tun_file *tfile,
1519 struct sk_buff *skb, int more)
1521 struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
1522 struct sk_buff_head process_queue;
1523 u32 rx_batched = tun->rx_batched;
1526 if (!rx_batched || (!more && skb_queue_empty(queue))) {
1528 netif_receive_skb(skb);
1533 spin_lock(&queue->lock);
1534 if (!more || skb_queue_len(queue) == rx_batched) {
1535 __skb_queue_head_init(&process_queue);
1536 skb_queue_splice_tail_init(queue, &process_queue);
1539 __skb_queue_tail(queue, skb);
1541 spin_unlock(&queue->lock);
1544 struct sk_buff *nskb;
1547 while ((nskb = __skb_dequeue(&process_queue)))
1548 netif_receive_skb(nskb);
1549 netif_receive_skb(skb);
1554 static bool tun_can_build_skb(struct tun_struct *tun, struct tun_file *tfile,
1555 int len, int noblock, bool zerocopy)
1557 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
1560 if (tfile->socket.sk->sk_sndbuf != INT_MAX)
1569 if (SKB_DATA_ALIGN(len + TUN_RX_PAD) +
1570 SKB_DATA_ALIGN(sizeof(struct skb_shared_info)) > PAGE_SIZE)
1576 static struct sk_buff *tun_build_skb(struct tun_struct *tun,
1577 struct tun_file *tfile,
1578 struct iov_iter *from,
1579 struct virtio_net_hdr *hdr,
1580 int len, int *skb_xdp)
1582 struct page_frag *alloc_frag = ¤t->task_frag;
1583 struct sk_buff *skb;
1584 struct bpf_prog *xdp_prog;
1585 int buflen = SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
1586 unsigned int delta = 0;
1589 int err, pad = TUN_RX_PAD;
1592 xdp_prog = rcu_dereference(tun->xdp_prog);
1594 pad += TUN_HEADROOM;
1595 buflen += SKB_DATA_ALIGN(len + pad);
1598 alloc_frag->offset = ALIGN((u64)alloc_frag->offset, SMP_CACHE_BYTES);
1599 if (unlikely(!skb_page_frag_refill(buflen, alloc_frag, GFP_KERNEL)))
1600 return ERR_PTR(-ENOMEM);
1602 buf = (char *)page_address(alloc_frag->page) + alloc_frag->offset;
1603 copied = copy_page_from_iter(alloc_frag->page,
1604 alloc_frag->offset + pad,
1607 return ERR_PTR(-EFAULT);
1609 /* There's a small window that XDP may be set after the check
1610 * of xdp_prog above, this should be rare and for simplicity
1611 * we do XDP on skb in case the headroom is not enough.
1613 if (hdr->gso_type || !xdp_prog)
1620 xdp_prog = rcu_dereference(tun->xdp_prog);
1621 if (xdp_prog && !*skb_xdp) {
1622 struct xdp_buff xdp;
1626 xdp.data_hard_start = buf;
1627 xdp.data = buf + pad;
1628 xdp_set_data_meta_invalid(&xdp);
1629 xdp.data_end = xdp.data + len;
1630 xdp.rxq = &tfile->xdp_rxq;
1631 orig_data = xdp.data;
1632 act = bpf_prog_run_xdp(xdp_prog, &xdp);
1636 get_page(alloc_frag->page);
1637 alloc_frag->offset += buflen;
1638 err = xdp_do_redirect(tun->dev, &xdp, xdp_prog);
1646 get_page(alloc_frag->page);
1647 alloc_frag->offset += buflen;
1648 if (tun_xdp_tx(tun->dev, &xdp) < 0)
1654 delta = orig_data - xdp.data;
1655 len = xdp.data_end - xdp.data;
1658 bpf_warn_invalid_xdp_action(act);
1661 trace_xdp_exception(tun->dev, xdp_prog, act);
1668 skb = build_skb(buf, buflen);
1672 return ERR_PTR(-ENOMEM);
1675 skb_reserve(skb, pad - delta);
1677 get_page(alloc_frag->page);
1678 alloc_frag->offset += buflen;
1686 put_page(alloc_frag->page);
1690 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1694 /* Get packet from user space buffer */
1695 static ssize_t tun_get_user(struct tun_struct *tun, struct tun_file *tfile,
1696 void *msg_control, struct iov_iter *from,
1697 int noblock, bool more)
1699 struct tun_pi pi = { 0, cpu_to_be16(ETH_P_IP) };
1700 struct sk_buff *skb;
1701 size_t total_len = iov_iter_count(from);
1702 size_t len = total_len, align = tun->align, linear;
1703 struct virtio_net_hdr gso = { 0 };
1704 struct tun_pcpu_stats *stats;
1707 bool zerocopy = false;
1711 bool frags = tun_napi_frags_enabled(tun);
1713 if (!(tun->dev->flags & IFF_UP))
1716 if (!(tun->flags & IFF_NO_PI)) {
1717 if (len < sizeof(pi))
1721 if (!copy_from_iter_full(&pi, sizeof(pi), from))
1725 if (tun->flags & IFF_VNET_HDR) {
1726 int vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
1728 if (len < vnet_hdr_sz)
1732 if (!copy_from_iter_full(&gso, sizeof(gso), from))
1735 if ((gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
1736 tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2 > tun16_to_cpu(tun, gso.hdr_len))
1737 gso.hdr_len = cpu_to_tun16(tun, tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2);
1739 if (tun16_to_cpu(tun, gso.hdr_len) > len)
1741 iov_iter_advance(from, vnet_hdr_sz - sizeof(gso));
1744 if ((tun->flags & TUN_TYPE_MASK) == IFF_TAP) {
1745 align += NET_IP_ALIGN;
1746 if (unlikely(len < ETH_HLEN ||
1747 (gso.hdr_len && tun16_to_cpu(tun, gso.hdr_len) < ETH_HLEN)))
1751 good_linear = SKB_MAX_HEAD(align);
1754 struct iov_iter i = *from;
1756 /* There are 256 bytes to be copied in skb, so there is
1757 * enough room for skb expand head in case it is used.
1758 * The rest of the buffer is mapped from userspace.
1760 copylen = gso.hdr_len ? tun16_to_cpu(tun, gso.hdr_len) : GOODCOPY_LEN;
1761 if (copylen > good_linear)
1762 copylen = good_linear;
1764 iov_iter_advance(&i, copylen);
1765 if (iov_iter_npages(&i, INT_MAX) <= MAX_SKB_FRAGS)
1769 if (!frags && tun_can_build_skb(tun, tfile, len, noblock, zerocopy)) {
1770 /* For the packet that is not easy to be processed
1771 * (e.g gso or jumbo packet), we will do it at after
1772 * skb was created with generic XDP routine.
1774 skb = tun_build_skb(tun, tfile, from, &gso, len, &skb_xdp);
1776 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1777 return PTR_ERR(skb);
1784 if (tun16_to_cpu(tun, gso.hdr_len) > good_linear)
1785 linear = good_linear;
1787 linear = tun16_to_cpu(tun, gso.hdr_len);
1791 mutex_lock(&tfile->napi_mutex);
1792 skb = tun_napi_alloc_frags(tfile, copylen, from);
1793 /* tun_napi_alloc_frags() enforces a layout for the skb.
1794 * If zerocopy is enabled, then this layout will be
1795 * overwritten by zerocopy_sg_from_iter().
1799 skb = tun_alloc_skb(tfile, align, copylen, linear,
1804 if (PTR_ERR(skb) != -EAGAIN)
1805 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1807 mutex_unlock(&tfile->napi_mutex);
1808 return PTR_ERR(skb);
1812 err = zerocopy_sg_from_iter(skb, from);
1814 err = skb_copy_datagram_from_iter(skb, 0, from, len);
1817 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1820 tfile->napi.skb = NULL;
1821 mutex_unlock(&tfile->napi_mutex);
1828 if (virtio_net_hdr_to_skb(skb, &gso, tun_is_little_endian(tun))) {
1829 this_cpu_inc(tun->pcpu_stats->rx_frame_errors);
1832 tfile->napi.skb = NULL;
1833 mutex_unlock(&tfile->napi_mutex);
1839 switch (tun->flags & TUN_TYPE_MASK) {
1841 if (tun->flags & IFF_NO_PI) {
1842 u8 ip_version = skb->len ? (skb->data[0] >> 4) : 0;
1844 switch (ip_version) {
1846 pi.proto = htons(ETH_P_IP);
1849 pi.proto = htons(ETH_P_IPV6);
1852 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1858 skb_reset_mac_header(skb);
1859 skb->protocol = pi.proto;
1860 skb->dev = tun->dev;
1864 skb->protocol = eth_type_trans(skb, tun->dev);
1868 /* copy skb_ubuf_info for callback when skb has no error */
1870 skb_shinfo(skb)->destructor_arg = msg_control;
1871 skb_shinfo(skb)->tx_flags |= SKBTX_DEV_ZEROCOPY;
1872 skb_shinfo(skb)->tx_flags |= SKBTX_SHARED_FRAG;
1873 } else if (msg_control) {
1874 struct ubuf_info *uarg = msg_control;
1875 uarg->callback(uarg, false);
1878 skb_reset_network_header(skb);
1879 skb_probe_transport_header(skb, 0);
1882 struct bpf_prog *xdp_prog;
1887 xdp_prog = rcu_dereference(tun->xdp_prog);
1889 ret = do_xdp_generic(xdp_prog, skb);
1890 if (ret != XDP_PASS) {
1900 /* Compute the costly rx hash only if needed for flow updates.
1901 * We may get a very small possibility of OOO during switching, not
1902 * worth to optimize.
1904 if (!rcu_access_pointer(tun->steering_prog) && tun->numqueues > 1 &&
1906 rxhash = __skb_get_hash_symmetric(skb);
1909 /* Exercise flow dissector code path. */
1910 u32 headlen = eth_get_headlen(skb->data, skb_headlen(skb));
1912 if (unlikely(headlen > skb_headlen(skb))) {
1913 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1914 napi_free_frags(&tfile->napi);
1915 mutex_unlock(&tfile->napi_mutex);
1921 napi_gro_frags(&tfile->napi);
1923 mutex_unlock(&tfile->napi_mutex);
1924 } else if (tfile->napi_enabled) {
1925 struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
1928 spin_lock_bh(&queue->lock);
1929 __skb_queue_tail(queue, skb);
1930 queue_len = skb_queue_len(queue);
1931 spin_unlock(&queue->lock);
1933 if (!more || queue_len > NAPI_POLL_WEIGHT)
1934 napi_schedule(&tfile->napi);
1937 } else if (!IS_ENABLED(CONFIG_4KSTACKS)) {
1938 tun_rx_batched(tun, tfile, skb, more);
1943 stats = get_cpu_ptr(tun->pcpu_stats);
1944 u64_stats_update_begin(&stats->syncp);
1945 stats->rx_packets++;
1946 stats->rx_bytes += len;
1947 u64_stats_update_end(&stats->syncp);
1951 tun_flow_update(tun, rxhash, tfile);
1956 static ssize_t tun_chr_write_iter(struct kiocb *iocb, struct iov_iter *from)
1958 struct file *file = iocb->ki_filp;
1959 struct tun_file *tfile = file->private_data;
1960 struct tun_struct *tun = tun_get(tfile);
1966 result = tun_get_user(tun, tfile, NULL, from,
1967 file->f_flags & O_NONBLOCK, false);
1973 static ssize_t tun_put_user_xdp(struct tun_struct *tun,
1974 struct tun_file *tfile,
1975 struct xdp_frame *xdp_frame,
1976 struct iov_iter *iter)
1978 int vnet_hdr_sz = 0;
1979 size_t size = xdp_frame->len;
1980 struct tun_pcpu_stats *stats;
1983 if (tun->flags & IFF_VNET_HDR) {
1984 struct virtio_net_hdr gso = { 0 };
1986 vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
1987 if (unlikely(iov_iter_count(iter) < vnet_hdr_sz))
1989 if (unlikely(copy_to_iter(&gso, sizeof(gso), iter) !=
1992 iov_iter_advance(iter, vnet_hdr_sz - sizeof(gso));
1995 ret = copy_to_iter(xdp_frame->data, size, iter) + vnet_hdr_sz;
1997 stats = get_cpu_ptr(tun->pcpu_stats);
1998 u64_stats_update_begin(&stats->syncp);
1999 stats->tx_packets++;
2000 stats->tx_bytes += ret;
2001 u64_stats_update_end(&stats->syncp);
2002 put_cpu_ptr(tun->pcpu_stats);
2007 /* Put packet to the user space buffer */
2008 static ssize_t tun_put_user(struct tun_struct *tun,
2009 struct tun_file *tfile,
2010 struct sk_buff *skb,
2011 struct iov_iter *iter)
2013 struct tun_pi pi = { 0, skb->protocol };
2014 struct tun_pcpu_stats *stats;
2016 int vlan_offset = 0;
2018 int vnet_hdr_sz = 0;
2020 if (skb_vlan_tag_present(skb))
2021 vlan_hlen = VLAN_HLEN;
2023 if (tun->flags & IFF_VNET_HDR)
2024 vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
2026 total = skb->len + vlan_hlen + vnet_hdr_sz;
2028 if (!(tun->flags & IFF_NO_PI)) {
2029 if (iov_iter_count(iter) < sizeof(pi))
2032 total += sizeof(pi);
2033 if (iov_iter_count(iter) < total) {
2034 /* Packet will be striped */
2035 pi.flags |= TUN_PKT_STRIP;
2038 if (copy_to_iter(&pi, sizeof(pi), iter) != sizeof(pi))
2043 struct virtio_net_hdr gso;
2045 if (iov_iter_count(iter) < vnet_hdr_sz)
2048 if (virtio_net_hdr_from_skb(skb, &gso,
2049 tun_is_little_endian(tun), true,
2051 struct skb_shared_info *sinfo = skb_shinfo(skb);
2052 pr_err("unexpected GSO type: "
2053 "0x%x, gso_size %d, hdr_len %d\n",
2054 sinfo->gso_type, tun16_to_cpu(tun, gso.gso_size),
2055 tun16_to_cpu(tun, gso.hdr_len));
2056 print_hex_dump(KERN_ERR, "tun: ",
2059 min((int)tun16_to_cpu(tun, gso.hdr_len), 64), true);
2064 if (copy_to_iter(&gso, sizeof(gso), iter) != sizeof(gso))
2067 iov_iter_advance(iter, vnet_hdr_sz - sizeof(gso));
2074 veth.h_vlan_proto = skb->vlan_proto;
2075 veth.h_vlan_TCI = htons(skb_vlan_tag_get(skb));
2077 vlan_offset = offsetof(struct vlan_ethhdr, h_vlan_proto);
2079 ret = skb_copy_datagram_iter(skb, 0, iter, vlan_offset);
2080 if (ret || !iov_iter_count(iter))
2083 ret = copy_to_iter(&veth, sizeof(veth), iter);
2084 if (ret != sizeof(veth) || !iov_iter_count(iter))
2088 skb_copy_datagram_iter(skb, vlan_offset, iter, skb->len - vlan_offset);
2091 /* caller is in process context, */
2092 stats = get_cpu_ptr(tun->pcpu_stats);
2093 u64_stats_update_begin(&stats->syncp);
2094 stats->tx_packets++;
2095 stats->tx_bytes += skb->len + vlan_hlen;
2096 u64_stats_update_end(&stats->syncp);
2097 put_cpu_ptr(tun->pcpu_stats);
2102 static void *tun_ring_recv(struct tun_file *tfile, int noblock, int *err)
2104 DECLARE_WAITQUEUE(wait, current);
2108 ptr = ptr_ring_consume(&tfile->tx_ring);
2116 add_wait_queue(&tfile->wq.wait, &wait);
2117 current->state = TASK_INTERRUPTIBLE;
2120 ptr = ptr_ring_consume(&tfile->tx_ring);
2123 if (signal_pending(current)) {
2124 error = -ERESTARTSYS;
2127 if (tfile->socket.sk->sk_shutdown & RCV_SHUTDOWN) {
2135 current->state = TASK_RUNNING;
2136 remove_wait_queue(&tfile->wq.wait, &wait);
2143 static ssize_t tun_do_read(struct tun_struct *tun, struct tun_file *tfile,
2144 struct iov_iter *to,
2145 int noblock, void *ptr)
2150 tun_debug(KERN_INFO, tun, "tun_do_read\n");
2152 if (!iov_iter_count(to)) {
2158 /* Read frames from ring */
2159 ptr = tun_ring_recv(tfile, noblock, &err);
2164 if (tun_is_xdp_frame(ptr)) {
2165 struct xdp_frame *xdpf = tun_ptr_to_xdp(ptr);
2167 ret = tun_put_user_xdp(tun, tfile, xdpf, to);
2168 xdp_return_frame(xdpf);
2170 struct sk_buff *skb = ptr;
2172 ret = tun_put_user(tun, tfile, skb, to);
2173 if (unlikely(ret < 0))
2182 static ssize_t tun_chr_read_iter(struct kiocb *iocb, struct iov_iter *to)
2184 struct file *file = iocb->ki_filp;
2185 struct tun_file *tfile = file->private_data;
2186 struct tun_struct *tun = tun_get(tfile);
2187 ssize_t len = iov_iter_count(to), ret;
2191 ret = tun_do_read(tun, tfile, to, file->f_flags & O_NONBLOCK, NULL);
2192 ret = min_t(ssize_t, ret, len);
2199 static void tun_prog_free(struct rcu_head *rcu)
2201 struct tun_prog *prog = container_of(rcu, struct tun_prog, rcu);
2203 bpf_prog_destroy(prog->prog);
2207 static int __tun_set_ebpf(struct tun_struct *tun,
2208 struct tun_prog __rcu **prog_p,
2209 struct bpf_prog *prog)
2211 struct tun_prog *old, *new = NULL;
2214 new = kmalloc(sizeof(*new), GFP_KERNEL);
2220 spin_lock_bh(&tun->lock);
2221 old = rcu_dereference_protected(*prog_p,
2222 lockdep_is_held(&tun->lock));
2223 rcu_assign_pointer(*prog_p, new);
2224 spin_unlock_bh(&tun->lock);
2227 call_rcu(&old->rcu, tun_prog_free);
2232 static void tun_free_netdev(struct net_device *dev)
2234 struct tun_struct *tun = netdev_priv(dev);
2236 BUG_ON(!(list_empty(&tun->disabled)));
2237 free_percpu(tun->pcpu_stats);
2238 tun_flow_uninit(tun);
2239 security_tun_dev_free_security(tun->security);
2240 __tun_set_ebpf(tun, &tun->steering_prog, NULL);
2241 __tun_set_ebpf(tun, &tun->filter_prog, NULL);
2244 static void tun_setup(struct net_device *dev)
2246 struct tun_struct *tun = netdev_priv(dev);
2248 tun->owner = INVALID_UID;
2249 tun->group = INVALID_GID;
2250 tun_default_link_ksettings(dev, &tun->link_ksettings);
2252 dev->ethtool_ops = &tun_ethtool_ops;
2253 dev->needs_free_netdev = true;
2254 dev->priv_destructor = tun_free_netdev;
2255 /* We prefer our own queue length */
2256 dev->tx_queue_len = TUN_READQ_SIZE;
2259 /* Trivial set of netlink ops to allow deleting tun or tap
2260 * device with netlink.
2262 static int tun_validate(struct nlattr *tb[], struct nlattr *data[],
2263 struct netlink_ext_ack *extack)
2268 static size_t tun_get_size(const struct net_device *dev)
2270 BUILD_BUG_ON(sizeof(u32) != sizeof(uid_t));
2271 BUILD_BUG_ON(sizeof(u32) != sizeof(gid_t));
2273 return nla_total_size(sizeof(uid_t)) + /* OWNER */
2274 nla_total_size(sizeof(gid_t)) + /* GROUP */
2275 nla_total_size(sizeof(u8)) + /* TYPE */
2276 nla_total_size(sizeof(u8)) + /* PI */
2277 nla_total_size(sizeof(u8)) + /* VNET_HDR */
2278 nla_total_size(sizeof(u8)) + /* PERSIST */
2279 nla_total_size(sizeof(u8)) + /* MULTI_QUEUE */
2280 nla_total_size(sizeof(u32)) + /* NUM_QUEUES */
2281 nla_total_size(sizeof(u32)) + /* NUM_DISABLED_QUEUES */
2285 static int tun_fill_info(struct sk_buff *skb, const struct net_device *dev)
2287 struct tun_struct *tun = netdev_priv(dev);
2289 if (nla_put_u8(skb, IFLA_TUN_TYPE, tun->flags & TUN_TYPE_MASK))
2290 goto nla_put_failure;
2291 if (uid_valid(tun->owner) &&
2292 nla_put_u32(skb, IFLA_TUN_OWNER,
2293 from_kuid_munged(current_user_ns(), tun->owner)))
2294 goto nla_put_failure;
2295 if (gid_valid(tun->group) &&
2296 nla_put_u32(skb, IFLA_TUN_GROUP,
2297 from_kgid_munged(current_user_ns(), tun->group)))
2298 goto nla_put_failure;
2299 if (nla_put_u8(skb, IFLA_TUN_PI, !(tun->flags & IFF_NO_PI)))
2300 goto nla_put_failure;
2301 if (nla_put_u8(skb, IFLA_TUN_VNET_HDR, !!(tun->flags & IFF_VNET_HDR)))
2302 goto nla_put_failure;
2303 if (nla_put_u8(skb, IFLA_TUN_PERSIST, !!(tun->flags & IFF_PERSIST)))
2304 goto nla_put_failure;
2305 if (nla_put_u8(skb, IFLA_TUN_MULTI_QUEUE,
2306 !!(tun->flags & IFF_MULTI_QUEUE)))
2307 goto nla_put_failure;
2308 if (tun->flags & IFF_MULTI_QUEUE) {
2309 if (nla_put_u32(skb, IFLA_TUN_NUM_QUEUES, tun->numqueues))
2310 goto nla_put_failure;
2311 if (nla_put_u32(skb, IFLA_TUN_NUM_DISABLED_QUEUES,
2313 goto nla_put_failure;
2322 static struct rtnl_link_ops tun_link_ops __read_mostly = {
2324 .priv_size = sizeof(struct tun_struct),
2326 .validate = tun_validate,
2327 .get_size = tun_get_size,
2328 .fill_info = tun_fill_info,
2331 static void tun_sock_write_space(struct sock *sk)
2333 struct tun_file *tfile;
2334 wait_queue_head_t *wqueue;
2336 if (!sock_writeable(sk))
2339 if (!test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags))
2342 wqueue = sk_sleep(sk);
2343 if (wqueue && waitqueue_active(wqueue))
2344 wake_up_interruptible_sync_poll(wqueue, EPOLLOUT |
2345 EPOLLWRNORM | EPOLLWRBAND);
2347 tfile = container_of(sk, struct tun_file, sk);
2348 kill_fasync(&tfile->fasync, SIGIO, POLL_OUT);
2351 static int tun_sendmsg(struct socket *sock, struct msghdr *m, size_t total_len)
2354 struct tun_file *tfile = container_of(sock, struct tun_file, socket);
2355 struct tun_struct *tun = tun_get(tfile);
2360 ret = tun_get_user(tun, tfile, m->msg_control, &m->msg_iter,
2361 m->msg_flags & MSG_DONTWAIT,
2362 m->msg_flags & MSG_MORE);
2367 static int tun_recvmsg(struct socket *sock, struct msghdr *m, size_t total_len,
2370 struct tun_file *tfile = container_of(sock, struct tun_file, socket);
2371 struct tun_struct *tun = tun_get(tfile);
2372 void *ptr = m->msg_control;
2380 if (flags & ~(MSG_DONTWAIT|MSG_TRUNC|MSG_ERRQUEUE)) {
2384 if (flags & MSG_ERRQUEUE) {
2385 ret = sock_recv_errqueue(sock->sk, m, total_len,
2386 SOL_PACKET, TUN_TX_TIMESTAMP);
2389 ret = tun_do_read(tun, tfile, &m->msg_iter, flags & MSG_DONTWAIT, ptr);
2390 if (ret > (ssize_t)total_len) {
2391 m->msg_flags |= MSG_TRUNC;
2392 ret = flags & MSG_TRUNC ? ret : total_len;
2405 static int tun_ptr_peek_len(void *ptr)
2408 if (tun_is_xdp_frame(ptr)) {
2409 struct xdp_frame *xdpf = tun_ptr_to_xdp(ptr);
2413 return __skb_array_len_with_tag(ptr);
2419 static int tun_peek_len(struct socket *sock)
2421 struct tun_file *tfile = container_of(sock, struct tun_file, socket);
2422 struct tun_struct *tun;
2425 tun = tun_get(tfile);
2429 ret = PTR_RING_PEEK_CALL(&tfile->tx_ring, tun_ptr_peek_len);
2435 /* Ops structure to mimic raw sockets with tun */
2436 static const struct proto_ops tun_socket_ops = {
2437 .peek_len = tun_peek_len,
2438 .sendmsg = tun_sendmsg,
2439 .recvmsg = tun_recvmsg,
2442 static struct proto tun_proto = {
2444 .owner = THIS_MODULE,
2445 .obj_size = sizeof(struct tun_file),
2448 static int tun_flags(struct tun_struct *tun)
2450 return tun->flags & (TUN_FEATURES | IFF_PERSIST | IFF_TUN | IFF_TAP);
2453 static ssize_t tun_show_flags(struct device *dev, struct device_attribute *attr,
2456 struct tun_struct *tun = netdev_priv(to_net_dev(dev));
2457 return sprintf(buf, "0x%x\n", tun_flags(tun));
2460 static ssize_t tun_show_owner(struct device *dev, struct device_attribute *attr,
2463 struct tun_struct *tun = netdev_priv(to_net_dev(dev));
2464 return uid_valid(tun->owner)?
2465 sprintf(buf, "%u\n",
2466 from_kuid_munged(current_user_ns(), tun->owner)):
2467 sprintf(buf, "-1\n");
2470 static ssize_t tun_show_group(struct device *dev, struct device_attribute *attr,
2473 struct tun_struct *tun = netdev_priv(to_net_dev(dev));
2474 return gid_valid(tun->group) ?
2475 sprintf(buf, "%u\n",
2476 from_kgid_munged(current_user_ns(), tun->group)):
2477 sprintf(buf, "-1\n");
2480 static DEVICE_ATTR(tun_flags, 0444, tun_show_flags, NULL);
2481 static DEVICE_ATTR(owner, 0444, tun_show_owner, NULL);
2482 static DEVICE_ATTR(group, 0444, tun_show_group, NULL);
2484 static struct attribute *tun_dev_attrs[] = {
2485 &dev_attr_tun_flags.attr,
2486 &dev_attr_owner.attr,
2487 &dev_attr_group.attr,
2491 static const struct attribute_group tun_attr_group = {
2492 .attrs = tun_dev_attrs
2495 static int tun_set_iff(struct net *net, struct file *file, struct ifreq *ifr)
2497 struct tun_struct *tun;
2498 struct tun_file *tfile = file->private_data;
2499 struct net_device *dev;
2502 if (tfile->detached)
2505 if ((ifr->ifr_flags & IFF_NAPI_FRAGS)) {
2506 if (!capable(CAP_NET_ADMIN))
2509 if (!(ifr->ifr_flags & IFF_NAPI) ||
2510 (ifr->ifr_flags & TUN_TYPE_MASK) != IFF_TAP)
2514 dev = __dev_get_by_name(net, ifr->ifr_name);
2516 if (ifr->ifr_flags & IFF_TUN_EXCL)
2518 if ((ifr->ifr_flags & IFF_TUN) && dev->netdev_ops == &tun_netdev_ops)
2519 tun = netdev_priv(dev);
2520 else if ((ifr->ifr_flags & IFF_TAP) && dev->netdev_ops == &tap_netdev_ops)
2521 tun = netdev_priv(dev);
2525 if (!!(ifr->ifr_flags & IFF_MULTI_QUEUE) !=
2526 !!(tun->flags & IFF_MULTI_QUEUE))
2529 if (tun_not_capable(tun))
2531 err = security_tun_dev_open(tun->security);
2535 err = tun_attach(tun, file, ifr->ifr_flags & IFF_NOFILTER,
2536 ifr->ifr_flags & IFF_NAPI);
2540 if (tun->flags & IFF_MULTI_QUEUE &&
2541 (tun->numqueues + tun->numdisabled > 1)) {
2542 /* One or more queue has already been attached, no need
2543 * to initialize the device again.
2545 netdev_state_change(dev);
2549 tun->flags = (tun->flags & ~TUN_FEATURES) |
2550 (ifr->ifr_flags & TUN_FEATURES);
2552 netdev_state_change(dev);
2555 unsigned long flags = 0;
2556 int queues = ifr->ifr_flags & IFF_MULTI_QUEUE ?
2559 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
2561 err = security_tun_dev_create();
2566 if (ifr->ifr_flags & IFF_TUN) {
2570 } else if (ifr->ifr_flags & IFF_TAP) {
2578 name = ifr->ifr_name;
2580 dev = alloc_netdev_mqs(sizeof(struct tun_struct), name,
2581 NET_NAME_UNKNOWN, tun_setup, queues,
2586 err = dev_get_valid_name(net, dev, name);
2590 dev_net_set(dev, net);
2591 dev->rtnl_link_ops = &tun_link_ops;
2592 dev->ifindex = tfile->ifindex;
2593 dev->sysfs_groups[0] = &tun_attr_group;
2595 tun = netdev_priv(dev);
2598 tun->txflt.count = 0;
2599 tun->vnet_hdr_sz = sizeof(struct virtio_net_hdr);
2601 tun->align = NET_SKB_PAD;
2602 tun->filter_attached = false;
2603 tun->sndbuf = tfile->socket.sk->sk_sndbuf;
2604 tun->rx_batched = 0;
2605 RCU_INIT_POINTER(tun->steering_prog, NULL);
2607 tun->pcpu_stats = netdev_alloc_pcpu_stats(struct tun_pcpu_stats);
2608 if (!tun->pcpu_stats) {
2613 spin_lock_init(&tun->lock);
2615 err = security_tun_dev_alloc_security(&tun->security);
2622 dev->hw_features = NETIF_F_SG | NETIF_F_FRAGLIST |
2623 TUN_USER_FEATURES | NETIF_F_HW_VLAN_CTAG_TX |
2624 NETIF_F_HW_VLAN_STAG_TX;
2625 dev->features = dev->hw_features | NETIF_F_LLTX;
2626 dev->vlan_features = dev->features &
2627 ~(NETIF_F_HW_VLAN_CTAG_TX |
2628 NETIF_F_HW_VLAN_STAG_TX);
2630 tun->flags = (tun->flags & ~TUN_FEATURES) |
2631 (ifr->ifr_flags & TUN_FEATURES);
2633 INIT_LIST_HEAD(&tun->disabled);
2634 err = tun_attach(tun, file, false, ifr->ifr_flags & IFF_NAPI);
2638 err = register_netdevice(tun->dev);
2643 netif_carrier_on(tun->dev);
2645 tun_debug(KERN_INFO, tun, "tun_set_iff\n");
2647 /* Make sure persistent devices do not get stuck in
2650 if (netif_running(tun->dev))
2651 netif_tx_wake_all_queues(tun->dev);
2653 strcpy(ifr->ifr_name, tun->dev->name);
2657 tun_detach_all(dev);
2658 /* register_netdevice() already called tun_free_netdev() */
2662 tun_flow_uninit(tun);
2663 security_tun_dev_free_security(tun->security);
2665 free_percpu(tun->pcpu_stats);
2671 static void tun_get_iff(struct net *net, struct tun_struct *tun,
2674 tun_debug(KERN_INFO, tun, "tun_get_iff\n");
2676 strcpy(ifr->ifr_name, tun->dev->name);
2678 ifr->ifr_flags = tun_flags(tun);
2682 /* This is like a cut-down ethtool ops, except done via tun fd so no
2683 * privs required. */
2684 static int set_offload(struct tun_struct *tun, unsigned long arg)
2686 netdev_features_t features = 0;
2688 if (arg & TUN_F_CSUM) {
2689 features |= NETIF_F_HW_CSUM;
2692 if (arg & (TUN_F_TSO4|TUN_F_TSO6)) {
2693 if (arg & TUN_F_TSO_ECN) {
2694 features |= NETIF_F_TSO_ECN;
2695 arg &= ~TUN_F_TSO_ECN;
2697 if (arg & TUN_F_TSO4)
2698 features |= NETIF_F_TSO;
2699 if (arg & TUN_F_TSO6)
2700 features |= NETIF_F_TSO6;
2701 arg &= ~(TUN_F_TSO4|TUN_F_TSO6);
2707 /* This gives the user a way to test for new features in future by
2708 * trying to set them. */
2712 tun->set_features = features;
2713 tun->dev->wanted_features &= ~TUN_USER_FEATURES;
2714 tun->dev->wanted_features |= features;
2715 netdev_update_features(tun->dev);
2720 static void tun_detach_filter(struct tun_struct *tun, int n)
2723 struct tun_file *tfile;
2725 for (i = 0; i < n; i++) {
2726 tfile = rtnl_dereference(tun->tfiles[i]);
2727 lock_sock(tfile->socket.sk);
2728 sk_detach_filter(tfile->socket.sk);
2729 release_sock(tfile->socket.sk);
2732 tun->filter_attached = false;
2735 static int tun_attach_filter(struct tun_struct *tun)
2738 struct tun_file *tfile;
2740 for (i = 0; i < tun->numqueues; i++) {
2741 tfile = rtnl_dereference(tun->tfiles[i]);
2742 lock_sock(tfile->socket.sk);
2743 ret = sk_attach_filter(&tun->fprog, tfile->socket.sk);
2744 release_sock(tfile->socket.sk);
2746 tun_detach_filter(tun, i);
2751 tun->filter_attached = true;
2755 static void tun_set_sndbuf(struct tun_struct *tun)
2757 struct tun_file *tfile;
2760 for (i = 0; i < tun->numqueues; i++) {
2761 tfile = rtnl_dereference(tun->tfiles[i]);
2762 tfile->socket.sk->sk_sndbuf = tun->sndbuf;
2766 static int tun_set_queue(struct file *file, struct ifreq *ifr)
2768 struct tun_file *tfile = file->private_data;
2769 struct tun_struct *tun;
2774 if (ifr->ifr_flags & IFF_ATTACH_QUEUE) {
2775 tun = tfile->detached;
2780 ret = security_tun_dev_attach_queue(tun->security);
2783 ret = tun_attach(tun, file, false, tun->flags & IFF_NAPI);
2784 } else if (ifr->ifr_flags & IFF_DETACH_QUEUE) {
2785 tun = rtnl_dereference(tfile->tun);
2786 if (!tun || !(tun->flags & IFF_MULTI_QUEUE) || tfile->detached)
2789 __tun_detach(tfile, false);
2794 netdev_state_change(tun->dev);
2801 static int tun_set_ebpf(struct tun_struct *tun, struct tun_prog **prog_p,
2804 struct bpf_prog *prog;
2807 if (copy_from_user(&fd, data, sizeof(fd)))
2813 prog = bpf_prog_get_type(fd, BPF_PROG_TYPE_SOCKET_FILTER);
2815 return PTR_ERR(prog);
2818 return __tun_set_ebpf(tun, prog_p, prog);
2821 static long __tun_chr_ioctl(struct file *file, unsigned int cmd,
2822 unsigned long arg, int ifreq_len)
2824 struct tun_file *tfile = file->private_data;
2825 struct net *net = sock_net(&tfile->sk);
2826 struct tun_struct *tun;
2827 void __user* argp = (void __user*)arg;
2833 unsigned int ifindex;
2836 bool do_notify = false;
2838 if (cmd == TUNSETIFF || cmd == TUNSETQUEUE ||
2839 (_IOC_TYPE(cmd) == SOCK_IOC_TYPE && cmd != SIOCGSKNS)) {
2840 if (copy_from_user(&ifr, argp, ifreq_len))
2843 memset(&ifr, 0, sizeof(ifr));
2845 if (cmd == TUNGETFEATURES) {
2846 /* Currently this just means: "what IFF flags are valid?".
2847 * This is needed because we never checked for invalid flags on
2850 return put_user(IFF_TUN | IFF_TAP | TUN_FEATURES,
2851 (unsigned int __user*)argp);
2852 } else if (cmd == TUNSETQUEUE) {
2853 return tun_set_queue(file, &ifr);
2854 } else if (cmd == SIOCGSKNS) {
2855 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
2857 return open_related_ns(&net->ns, get_net_ns);
2863 tun = tun_get(tfile);
2864 if (cmd == TUNSETIFF) {
2869 ifr.ifr_name[IFNAMSIZ-1] = '\0';
2871 ret = tun_set_iff(net, file, &ifr);
2876 if (copy_to_user(argp, &ifr, ifreq_len))
2880 if (cmd == TUNSETIFINDEX) {
2886 if (copy_from_user(&ifindex, argp, sizeof(ifindex)))
2890 tfile->ifindex = ifindex;
2898 tun_debug(KERN_INFO, tun, "tun_chr_ioctl cmd %u\n", cmd);
2903 tun_get_iff(current->nsproxy->net_ns, tun, &ifr);
2905 if (tfile->detached)
2906 ifr.ifr_flags |= IFF_DETACH_QUEUE;
2907 if (!tfile->socket.sk->sk_filter)
2908 ifr.ifr_flags |= IFF_NOFILTER;
2910 if (copy_to_user(argp, &ifr, ifreq_len))
2915 /* Disable/Enable checksum */
2917 /* [unimplemented] */
2918 tun_debug(KERN_INFO, tun, "ignored: set checksum %s\n",
2919 arg ? "disabled" : "enabled");
2923 /* Disable/Enable persist mode. Keep an extra reference to the
2924 * module to prevent the module being unprobed.
2926 if (arg && !(tun->flags & IFF_PERSIST)) {
2927 tun->flags |= IFF_PERSIST;
2928 __module_get(THIS_MODULE);
2931 if (!arg && (tun->flags & IFF_PERSIST)) {
2932 tun->flags &= ~IFF_PERSIST;
2933 module_put(THIS_MODULE);
2937 tun_debug(KERN_INFO, tun, "persist %s\n",
2938 arg ? "enabled" : "disabled");
2942 /* Set owner of the device */
2943 owner = make_kuid(current_user_ns(), arg);
2944 if (!uid_valid(owner)) {
2950 tun_debug(KERN_INFO, tun, "owner set to %u\n",
2951 from_kuid(&init_user_ns, tun->owner));
2955 /* Set group of the device */
2956 group = make_kgid(current_user_ns(), arg);
2957 if (!gid_valid(group)) {
2963 tun_debug(KERN_INFO, tun, "group set to %u\n",
2964 from_kgid(&init_user_ns, tun->group));
2968 /* Only allow setting the type when the interface is down */
2969 if (tun->dev->flags & IFF_UP) {
2970 tun_debug(KERN_INFO, tun,
2971 "Linktype set failed because interface is up\n");
2974 tun->dev->type = (int) arg;
2975 tun_debug(KERN_INFO, tun, "linktype set to %d\n",
2987 ret = set_offload(tun, arg);
2990 case TUNSETTXFILTER:
2991 /* Can be set only for TAPs */
2993 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
2995 ret = update_filter(&tun->txflt, (void __user *)arg);
2999 /* Get hw address */
3000 memcpy(ifr.ifr_hwaddr.sa_data, tun->dev->dev_addr, ETH_ALEN);
3001 ifr.ifr_hwaddr.sa_family = tun->dev->type;
3002 if (copy_to_user(argp, &ifr, ifreq_len))
3007 /* Set hw address */
3008 tun_debug(KERN_DEBUG, tun, "set hw address: %pM\n",
3009 ifr.ifr_hwaddr.sa_data);
3011 ret = dev_set_mac_address(tun->dev, &ifr.ifr_hwaddr);
3015 sndbuf = tfile->socket.sk->sk_sndbuf;
3016 if (copy_to_user(argp, &sndbuf, sizeof(sndbuf)))
3021 if (copy_from_user(&sndbuf, argp, sizeof(sndbuf))) {
3030 tun->sndbuf = sndbuf;
3031 tun_set_sndbuf(tun);
3034 case TUNGETVNETHDRSZ:
3035 vnet_hdr_sz = tun->vnet_hdr_sz;
3036 if (copy_to_user(argp, &vnet_hdr_sz, sizeof(vnet_hdr_sz)))
3040 case TUNSETVNETHDRSZ:
3041 if (copy_from_user(&vnet_hdr_sz, argp, sizeof(vnet_hdr_sz))) {
3045 if (vnet_hdr_sz < (int)sizeof(struct virtio_net_hdr)) {
3050 tun->vnet_hdr_sz = vnet_hdr_sz;
3054 le = !!(tun->flags & TUN_VNET_LE);
3055 if (put_user(le, (int __user *)argp))
3060 if (get_user(le, (int __user *)argp)) {
3065 tun->flags |= TUN_VNET_LE;
3067 tun->flags &= ~TUN_VNET_LE;
3071 ret = tun_get_vnet_be(tun, argp);
3075 ret = tun_set_vnet_be(tun, argp);
3078 case TUNATTACHFILTER:
3079 /* Can be set only for TAPs */
3081 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3084 if (copy_from_user(&tun->fprog, argp, sizeof(tun->fprog)))
3087 ret = tun_attach_filter(tun);
3090 case TUNDETACHFILTER:
3091 /* Can be set only for TAPs */
3093 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3096 tun_detach_filter(tun, tun->numqueues);
3101 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3104 if (copy_to_user(argp, &tun->fprog, sizeof(tun->fprog)))
3109 case TUNSETSTEERINGEBPF:
3110 ret = tun_set_ebpf(tun, &tun->steering_prog, argp);
3113 case TUNSETFILTEREBPF:
3114 ret = tun_set_ebpf(tun, &tun->filter_prog, argp);
3123 netdev_state_change(tun->dev);
3132 static long tun_chr_ioctl(struct file *file,
3133 unsigned int cmd, unsigned long arg)
3135 return __tun_chr_ioctl(file, cmd, arg, sizeof (struct ifreq));
3138 #ifdef CONFIG_COMPAT
3139 static long tun_chr_compat_ioctl(struct file *file,
3140 unsigned int cmd, unsigned long arg)
3145 case TUNSETTXFILTER:
3150 arg = (unsigned long)compat_ptr(arg);
3153 arg = (compat_ulong_t)arg;
3158 * compat_ifreq is shorter than ifreq, so we must not access beyond
3159 * the end of that structure. All fields that are used in this
3160 * driver are compatible though, we don't need to convert the
3163 return __tun_chr_ioctl(file, cmd, arg, sizeof(struct compat_ifreq));
3165 #endif /* CONFIG_COMPAT */
3167 static int tun_chr_fasync(int fd, struct file *file, int on)
3169 struct tun_file *tfile = file->private_data;
3172 if ((ret = fasync_helper(fd, file, on, &tfile->fasync)) < 0)
3176 __f_setown(file, task_pid(current), PIDTYPE_TGID, 0);
3177 tfile->flags |= TUN_FASYNC;
3179 tfile->flags &= ~TUN_FASYNC;
3185 static int tun_chr_open(struct inode *inode, struct file * file)
3187 struct net *net = current->nsproxy->net_ns;
3188 struct tun_file *tfile;
3190 DBG1(KERN_INFO, "tunX: tun_chr_open\n");
3192 tfile = (struct tun_file *)sk_alloc(net, AF_UNSPEC, GFP_KERNEL,
3196 if (ptr_ring_init(&tfile->tx_ring, 0, GFP_KERNEL)) {
3197 sk_free(&tfile->sk);
3201 mutex_init(&tfile->napi_mutex);
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");