2 * TUN - Universal TUN/TAP device driver.
3 * Copyright (C) 1999-2002 Maxim Krasnyansky <maxk@qualcomm.com>
5 * This program is free software; you can redistribute it and/or modify
6 * it under the terms of the GNU General Public License as published by
7 * the Free Software Foundation; either version 2 of the License, or
8 * (at your option) any later version.
10 * This program is distributed in the hope that it will be useful,
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
13 * GNU General Public License for more details.
15 * $Id: tun.c,v 1.15 2002/03/01 02:44:24 maxk Exp $
21 * Mike Kershaw <dragorn@kismetwireless.net> 2005/08/14
22 * Add TUNSETLINK ioctl to set the link encapsulation
24 * Mark Smith <markzzzsmith@yahoo.com.au>
25 * Use eth_random_addr() for tap MAC address.
27 * Harald Roelle <harald.roelle@ifi.lmu.de> 2004/04/20
28 * Fixes in packet dropping, queue length setting and queue wakeup.
29 * Increased default tx queue length.
33 * Daniel Podlejski <underley@underley.eu.org>
34 * Modifications for 2.3.99-pre5 kernel.
37 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
39 #define DRV_NAME "tun"
40 #define DRV_VERSION "1.6"
41 #define DRV_DESCRIPTION "Universal TUN/TAP device driver"
42 #define DRV_COPYRIGHT "(C) 1999-2004 Max Krasnyansky <maxk@qualcomm.com>"
44 #include <linux/module.h>
45 #include <linux/errno.h>
46 #include <linux/kernel.h>
47 #include <linux/sched/signal.h>
48 #include <linux/major.h>
49 #include <linux/slab.h>
50 #include <linux/poll.h>
51 #include <linux/fcntl.h>
52 #include <linux/init.h>
53 #include <linux/skbuff.h>
54 #include <linux/netdevice.h>
55 #include <linux/etherdevice.h>
56 #include <linux/miscdevice.h>
57 #include <linux/ethtool.h>
58 #include <linux/rtnetlink.h>
59 #include <linux/compat.h>
61 #include <linux/if_arp.h>
62 #include <linux/if_ether.h>
63 #include <linux/if_tun.h>
64 #include <linux/if_vlan.h>
65 #include <linux/crc32.h>
66 #include <linux/nsproxy.h>
67 #include <linux/virtio_net.h>
68 #include <linux/rcupdate.h>
69 #include <net/net_namespace.h>
70 #include <net/netns/generic.h>
71 #include <net/rtnetlink.h>
73 #include <linux/seq_file.h>
74 #include <linux/uio.h>
75 #include <linux/skb_array.h>
76 #include <linux/bpf.h>
77 #include <linux/bpf_trace.h>
78 #include <linux/mutex.h>
80 #include <linux/uaccess.h>
81 #include <linux/proc_fs.h>
83 /* Uncomment to enable debugging */
84 /* #define TUN_DEBUG 1 */
89 #define tun_debug(level, tun, fmt, args...) \
92 netdev_printk(level, tun->dev, fmt, ##args); \
94 #define DBG1(level, fmt, args...) \
97 printk(level fmt, ##args); \
100 #define tun_debug(level, tun, fmt, args...) \
103 netdev_printk(level, tun->dev, fmt, ##args); \
105 #define DBG1(level, fmt, args...) \
108 printk(level fmt, ##args); \
112 #define TUN_HEADROOM 256
113 #define TUN_RX_PAD (NET_IP_ALIGN + NET_SKB_PAD)
115 /* TUN device flags */
117 /* IFF_ATTACH_QUEUE is never stored in device flags,
118 * overload it to mean fasync when stored there.
120 #define TUN_FASYNC IFF_ATTACH_QUEUE
121 /* High bits in flags field are unused. */
122 #define TUN_VNET_LE 0x80000000
123 #define TUN_VNET_BE 0x40000000
125 #define TUN_FEATURES (IFF_NO_PI | IFF_ONE_QUEUE | IFF_VNET_HDR | \
126 IFF_MULTI_QUEUE | IFF_NAPI | IFF_NAPI_FRAGS)
128 #define GOODCOPY_LEN 128
130 #define FLT_EXACT_COUNT 8
132 unsigned int count; /* Number of addrs. Zero means disabled */
133 u32 mask[2]; /* Mask of the hashed addrs */
134 unsigned char addr[FLT_EXACT_COUNT][ETH_ALEN];
137 /* MAX_TAP_QUEUES 256 is chosen to allow rx/tx queues to be equal
138 * to max number of VCPUs in guest. */
139 #define MAX_TAP_QUEUES 256
140 #define MAX_TAP_FLOWS 4096
142 #define TUN_FLOW_EXPIRE (3 * HZ)
144 struct tun_pcpu_stats {
149 struct u64_stats_sync syncp;
155 /* A tun_file connects an open character device to a tuntap netdevice. It
156 * also contains all socket related structures (except sock_fprog and tap_filter)
157 * to serve as one transmit queue for tuntap device. The sock_fprog and
158 * tap_filter were kept in tun_struct since they were used for filtering for the
159 * netdevice not for a specific queue (at least I didn't see the requirement for
163 * The tun_file and tun_struct are loosely coupled, the pointer from one to the
164 * other can only be read while rcu_read_lock or rtnl_lock is held.
168 struct socket socket;
170 struct tun_struct __rcu *tun;
171 struct fasync_struct *fasync;
172 /* only used for fasnyc */
176 unsigned int ifindex;
178 struct napi_struct napi;
180 struct mutex napi_mutex; /* Protects access to the above napi */
181 struct list_head next;
182 struct tun_struct *detached;
183 struct ptr_ring tx_ring;
184 struct xdp_rxq_info xdp_rxq;
185 int xdp_pending_pkts;
188 struct tun_flow_entry {
189 struct hlist_node hash_link;
191 struct tun_struct *tun;
196 unsigned long updated;
199 #define TUN_NUM_FLOW_ENTRIES 1024
203 struct bpf_prog *prog;
206 /* Since the socket were moved to tun_file, to preserve the behavior of persist
207 * device, socket filter, sndbuf and vnet header size were restore when the
208 * file were attached to a persist device.
211 struct tun_file __rcu *tfiles[MAX_TAP_QUEUES];
212 unsigned int numqueues;
217 struct net_device *dev;
218 netdev_features_t set_features;
219 #define TUN_USER_FEATURES (NETIF_F_HW_CSUM|NETIF_F_TSO_ECN|NETIF_F_TSO| \
225 struct tap_filter txflt;
226 struct sock_fprog fprog;
227 /* protected by rtnl lock */
228 bool filter_attached;
233 struct hlist_head flows[TUN_NUM_FLOW_ENTRIES];
234 struct timer_list flow_gc_timer;
235 unsigned long ageing_time;
236 unsigned int numdisabled;
237 struct list_head disabled;
241 struct tun_pcpu_stats __percpu *pcpu_stats;
242 struct bpf_prog __rcu *xdp_prog;
243 struct tun_prog __rcu *steering_prog;
244 struct tun_prog __rcu *filter_prog;
252 bool tun_is_xdp_buff(void *ptr)
254 return (unsigned long)ptr & TUN_XDP_FLAG;
256 EXPORT_SYMBOL(tun_is_xdp_buff);
258 void *tun_xdp_to_ptr(void *ptr)
260 return (void *)((unsigned long)ptr | TUN_XDP_FLAG);
262 EXPORT_SYMBOL(tun_xdp_to_ptr);
264 void *tun_ptr_to_xdp(void *ptr)
266 return (void *)((unsigned long)ptr & ~TUN_XDP_FLAG);
268 EXPORT_SYMBOL(tun_ptr_to_xdp);
270 static int tun_napi_receive(struct napi_struct *napi, int budget)
272 struct tun_file *tfile = container_of(napi, struct tun_file, napi);
273 struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
274 struct sk_buff_head process_queue;
278 __skb_queue_head_init(&process_queue);
280 spin_lock(&queue->lock);
281 skb_queue_splice_tail_init(queue, &process_queue);
282 spin_unlock(&queue->lock);
284 while (received < budget && (skb = __skb_dequeue(&process_queue))) {
285 napi_gro_receive(napi, skb);
289 if (!skb_queue_empty(&process_queue)) {
290 spin_lock(&queue->lock);
291 skb_queue_splice(&process_queue, queue);
292 spin_unlock(&queue->lock);
298 static int tun_napi_poll(struct napi_struct *napi, int budget)
300 unsigned int received;
302 received = tun_napi_receive(napi, budget);
304 if (received < budget)
305 napi_complete_done(napi, received);
310 static void tun_napi_init(struct tun_struct *tun, struct tun_file *tfile,
313 tfile->napi_enabled = napi_en;
315 netif_napi_add(tun->dev, &tfile->napi, tun_napi_poll,
317 napi_enable(&tfile->napi);
318 mutex_init(&tfile->napi_mutex);
322 static void tun_napi_disable(struct tun_struct *tun, struct tun_file *tfile)
324 if (tfile->napi_enabled)
325 napi_disable(&tfile->napi);
328 static void tun_napi_del(struct tun_struct *tun, struct tun_file *tfile)
330 if (tfile->napi_enabled)
331 netif_napi_del(&tfile->napi);
334 static bool tun_napi_frags_enabled(const struct tun_struct *tun)
336 return READ_ONCE(tun->flags) & IFF_NAPI_FRAGS;
339 #ifdef CONFIG_TUN_VNET_CROSS_LE
340 static inline bool tun_legacy_is_little_endian(struct tun_struct *tun)
342 return tun->flags & TUN_VNET_BE ? false :
343 virtio_legacy_is_little_endian();
346 static long tun_get_vnet_be(struct tun_struct *tun, int __user *argp)
348 int be = !!(tun->flags & TUN_VNET_BE);
350 if (put_user(be, argp))
356 static long tun_set_vnet_be(struct tun_struct *tun, int __user *argp)
360 if (get_user(be, argp))
364 tun->flags |= TUN_VNET_BE;
366 tun->flags &= ~TUN_VNET_BE;
371 static inline bool tun_legacy_is_little_endian(struct tun_struct *tun)
373 return virtio_legacy_is_little_endian();
376 static long tun_get_vnet_be(struct tun_struct *tun, int __user *argp)
381 static long tun_set_vnet_be(struct tun_struct *tun, int __user *argp)
385 #endif /* CONFIG_TUN_VNET_CROSS_LE */
387 static inline bool tun_is_little_endian(struct tun_struct *tun)
389 return tun->flags & TUN_VNET_LE ||
390 tun_legacy_is_little_endian(tun);
393 static inline u16 tun16_to_cpu(struct tun_struct *tun, __virtio16 val)
395 return __virtio16_to_cpu(tun_is_little_endian(tun), val);
398 static inline __virtio16 cpu_to_tun16(struct tun_struct *tun, u16 val)
400 return __cpu_to_virtio16(tun_is_little_endian(tun), val);
403 static inline u32 tun_hashfn(u32 rxhash)
405 return rxhash & 0x3ff;
408 static struct tun_flow_entry *tun_flow_find(struct hlist_head *head, u32 rxhash)
410 struct tun_flow_entry *e;
412 hlist_for_each_entry_rcu(e, head, hash_link) {
413 if (e->rxhash == rxhash)
419 static struct tun_flow_entry *tun_flow_create(struct tun_struct *tun,
420 struct hlist_head *head,
421 u32 rxhash, u16 queue_index)
423 struct tun_flow_entry *e = kmalloc(sizeof(*e), GFP_ATOMIC);
426 tun_debug(KERN_INFO, tun, "create flow: hash %u index %u\n",
427 rxhash, queue_index);
428 e->updated = jiffies;
431 e->queue_index = queue_index;
433 hlist_add_head_rcu(&e->hash_link, head);
439 static void tun_flow_delete(struct tun_struct *tun, struct tun_flow_entry *e)
441 tun_debug(KERN_INFO, tun, "delete flow: hash %u index %u\n",
442 e->rxhash, e->queue_index);
443 hlist_del_rcu(&e->hash_link);
448 static void tun_flow_flush(struct tun_struct *tun)
452 spin_lock_bh(&tun->lock);
453 for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
454 struct tun_flow_entry *e;
455 struct hlist_node *n;
457 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link)
458 tun_flow_delete(tun, e);
460 spin_unlock_bh(&tun->lock);
463 static void tun_flow_delete_by_queue(struct tun_struct *tun, u16 queue_index)
467 spin_lock_bh(&tun->lock);
468 for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
469 struct tun_flow_entry *e;
470 struct hlist_node *n;
472 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
473 if (e->queue_index == queue_index)
474 tun_flow_delete(tun, e);
477 spin_unlock_bh(&tun->lock);
480 static void tun_flow_cleanup(struct timer_list *t)
482 struct tun_struct *tun = from_timer(tun, t, flow_gc_timer);
483 unsigned long delay = tun->ageing_time;
484 unsigned long next_timer = jiffies + delay;
485 unsigned long count = 0;
488 tun_debug(KERN_INFO, tun, "tun_flow_cleanup\n");
490 spin_lock(&tun->lock);
491 for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
492 struct tun_flow_entry *e;
493 struct hlist_node *n;
495 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
496 unsigned long this_timer;
498 this_timer = e->updated + delay;
499 if (time_before_eq(this_timer, jiffies)) {
500 tun_flow_delete(tun, e);
504 if (time_before(this_timer, next_timer))
505 next_timer = this_timer;
510 mod_timer(&tun->flow_gc_timer, round_jiffies_up(next_timer));
511 spin_unlock(&tun->lock);
514 static void tun_flow_update(struct tun_struct *tun, u32 rxhash,
515 struct tun_file *tfile)
517 struct hlist_head *head;
518 struct tun_flow_entry *e;
519 unsigned long delay = tun->ageing_time;
520 u16 queue_index = tfile->queue_index;
525 head = &tun->flows[tun_hashfn(rxhash)];
529 /* We may get a very small possibility of OOO during switching, not
530 * worth to optimize.*/
531 if (tun->numqueues == 1 || tfile->detached)
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);
557 * Save the hash received in the stack receive path and update the
558 * flow_hash table accordingly.
560 static inline void tun_flow_save_rps_rxhash(struct tun_flow_entry *e, u32 hash)
562 if (unlikely(e->rps_rxhash != hash))
563 e->rps_rxhash = hash;
566 /* We try to identify a flow through its rxhash first. The reason that
567 * we do not check rxq no. is because some cards(e.g 82599), chooses
568 * the rxq based on the txq where the last packet of the flow comes. As
569 * the userspace application move between processors, we may get a
570 * different rxq no. here. If we could not get rxhash, then we would
571 * hope the rxq no. may help here.
573 static u16 tun_automq_select_queue(struct tun_struct *tun, struct sk_buff *skb)
575 struct tun_flow_entry *e;
579 numqueues = READ_ONCE(tun->numqueues);
581 txq = __skb_get_hash_symmetric(skb);
583 e = tun_flow_find(&tun->flows[tun_hashfn(txq)], txq);
585 tun_flow_save_rps_rxhash(e, txq);
586 txq = e->queue_index;
588 /* use multiply and shift instead of expensive divide */
589 txq = ((u64)txq * numqueues) >> 32;
590 } else if (likely(skb_rx_queue_recorded(skb))) {
591 txq = skb_get_rx_queue(skb);
592 while (unlikely(txq >= numqueues))
599 static u16 tun_ebpf_select_queue(struct tun_struct *tun, struct sk_buff *skb)
601 struct tun_prog *prog;
604 prog = rcu_dereference(tun->steering_prog);
606 ret = bpf_prog_run_clear_cb(prog->prog, skb);
608 return ret % tun->numqueues;
611 static u16 tun_select_queue(struct net_device *dev, struct sk_buff *skb,
612 void *accel_priv, 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 static void tun_ptr_free(void *ptr)
664 if (tun_is_xdp_buff(ptr)) {
665 struct xdp_buff *xdp = tun_ptr_to_xdp(ptr);
667 put_page(virt_to_head_page(xdp->data));
669 __skb_array_destroy_skb(ptr);
673 static void tun_queue_purge(struct tun_file *tfile)
677 while ((ptr = ptr_ring_consume(&tfile->tx_ring)) != NULL)
680 skb_queue_purge(&tfile->sk.sk_write_queue);
681 skb_queue_purge(&tfile->sk.sk_error_queue);
684 static void tun_cleanup_tx_ring(struct tun_file *tfile)
686 if (tfile->tx_ring.queue) {
687 ptr_ring_cleanup(&tfile->tx_ring, tun_ptr_free);
688 xdp_rxq_info_unreg(&tfile->xdp_rxq);
689 memset(&tfile->tx_ring, 0, sizeof(tfile->tx_ring));
693 static void __tun_detach(struct tun_file *tfile, bool clean)
695 struct tun_file *ntfile;
696 struct tun_struct *tun;
698 tun = rtnl_dereference(tfile->tun);
701 tun_napi_disable(tun, tfile);
702 tun_napi_del(tun, tfile);
705 if (tun && !tfile->detached) {
706 u16 index = tfile->queue_index;
707 BUG_ON(index >= tun->numqueues);
709 rcu_assign_pointer(tun->tfiles[index],
710 tun->tfiles[tun->numqueues - 1]);
711 ntfile = rtnl_dereference(tun->tfiles[index]);
712 ntfile->queue_index = index;
716 RCU_INIT_POINTER(tfile->tun, NULL);
717 sock_put(&tfile->sk);
719 tun_disable_queue(tun, tfile);
722 tun_flow_delete_by_queue(tun, tun->numqueues + 1);
723 /* Drop read queue */
724 tun_queue_purge(tfile);
725 tun_set_real_num_queues(tun);
726 } else if (tfile->detached && clean) {
727 tun = tun_enable_queue(tfile);
728 sock_put(&tfile->sk);
732 if (tun && tun->numqueues == 0 && tun->numdisabled == 0) {
733 netif_carrier_off(tun->dev);
735 if (!(tun->flags & IFF_PERSIST) &&
736 tun->dev->reg_state == NETREG_REGISTERED)
737 unregister_netdevice(tun->dev);
739 tun_cleanup_tx_ring(tfile);
740 sock_put(&tfile->sk);
744 static void tun_detach(struct tun_file *tfile, bool clean)
747 __tun_detach(tfile, clean);
751 static void tun_detach_all(struct net_device *dev)
753 struct tun_struct *tun = netdev_priv(dev);
754 struct tun_file *tfile, *tmp;
755 int i, n = tun->numqueues;
757 for (i = 0; i < n; i++) {
758 tfile = rtnl_dereference(tun->tfiles[i]);
760 tun_napi_disable(tun, tfile);
761 tfile->socket.sk->sk_shutdown = RCV_SHUTDOWN;
762 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
763 RCU_INIT_POINTER(tfile->tun, NULL);
766 list_for_each_entry(tfile, &tun->disabled, next) {
767 tfile->socket.sk->sk_shutdown = RCV_SHUTDOWN;
768 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
769 RCU_INIT_POINTER(tfile->tun, NULL);
771 BUG_ON(tun->numqueues != 0);
774 for (i = 0; i < n; i++) {
775 tfile = rtnl_dereference(tun->tfiles[i]);
776 tun_napi_del(tun, tfile);
777 /* Drop read queue */
778 tun_queue_purge(tfile);
779 sock_put(&tfile->sk);
780 tun_cleanup_tx_ring(tfile);
782 list_for_each_entry_safe(tfile, tmp, &tun->disabled, next) {
783 tun_enable_queue(tfile);
784 tun_queue_purge(tfile);
785 sock_put(&tfile->sk);
786 tun_cleanup_tx_ring(tfile);
788 BUG_ON(tun->numdisabled != 0);
790 if (tun->flags & IFF_PERSIST)
791 module_put(THIS_MODULE);
794 static int tun_attach(struct tun_struct *tun, struct file *file,
795 bool skip_filter, bool napi)
797 struct tun_file *tfile = file->private_data;
798 struct net_device *dev = tun->dev;
801 err = security_tun_dev_attach(tfile->socket.sk, tun->security);
806 if (rtnl_dereference(tfile->tun) && !tfile->detached)
810 if (!(tun->flags & IFF_MULTI_QUEUE) && tun->numqueues == 1)
814 if (!tfile->detached &&
815 tun->numqueues + tun->numdisabled == MAX_TAP_QUEUES)
820 /* Re-attach the filter to persist device */
821 if (!skip_filter && (tun->filter_attached == true)) {
822 lock_sock(tfile->socket.sk);
823 err = sk_attach_filter(&tun->fprog, tfile->socket.sk);
824 release_sock(tfile->socket.sk);
829 if (!tfile->detached &&
830 ptr_ring_init(&tfile->tx_ring, dev->tx_queue_len, GFP_KERNEL)) {
835 tfile->queue_index = tun->numqueues;
836 tfile->socket.sk->sk_shutdown &= ~RCV_SHUTDOWN;
838 if (tfile->detached) {
839 /* Re-attach detached tfile, updating XDP queue_index */
840 WARN_ON(!xdp_rxq_info_is_reg(&tfile->xdp_rxq));
842 if (tfile->xdp_rxq.queue_index != tfile->queue_index)
843 tfile->xdp_rxq.queue_index = tfile->queue_index;
845 /* Setup XDP RX-queue info, for new tfile getting attached */
846 err = xdp_rxq_info_reg(&tfile->xdp_rxq,
847 tun->dev, tfile->queue_index);
853 rcu_assign_pointer(tfile->tun, tun);
854 rcu_assign_pointer(tun->tfiles[tun->numqueues], tfile);
857 if (tfile->detached) {
858 tun_enable_queue(tfile);
860 sock_hold(&tfile->sk);
861 tun_napi_init(tun, tfile, napi);
864 tun_set_real_num_queues(tun);
866 /* device is allowed to go away first, so no need to hold extra
874 static struct tun_struct *tun_get(struct tun_file *tfile)
876 struct tun_struct *tun;
879 tun = rcu_dereference(tfile->tun);
887 static void tun_put(struct tun_struct *tun)
893 static void addr_hash_set(u32 *mask, const u8 *addr)
895 int n = ether_crc(ETH_ALEN, addr) >> 26;
896 mask[n >> 5] |= (1 << (n & 31));
899 static unsigned int addr_hash_test(const u32 *mask, const u8 *addr)
901 int n = ether_crc(ETH_ALEN, addr) >> 26;
902 return mask[n >> 5] & (1 << (n & 31));
905 static int update_filter(struct tap_filter *filter, void __user *arg)
907 struct { u8 u[ETH_ALEN]; } *addr;
908 struct tun_filter uf;
909 int err, alen, n, nexact;
911 if (copy_from_user(&uf, arg, sizeof(uf)))
920 alen = ETH_ALEN * uf.count;
921 addr = memdup_user(arg + sizeof(uf), alen);
923 return PTR_ERR(addr);
925 /* The filter is updated without holding any locks. Which is
926 * perfectly safe. We disable it first and in the worst
927 * case we'll accept a few undesired packets. */
931 /* Use first set of addresses as an exact filter */
932 for (n = 0; n < uf.count && n < FLT_EXACT_COUNT; n++)
933 memcpy(filter->addr[n], addr[n].u, ETH_ALEN);
937 /* Remaining multicast addresses are hashed,
938 * unicast will leave the filter disabled. */
939 memset(filter->mask, 0, sizeof(filter->mask));
940 for (; n < uf.count; n++) {
941 if (!is_multicast_ether_addr(addr[n].u)) {
942 err = 0; /* no filter */
945 addr_hash_set(filter->mask, addr[n].u);
948 /* For ALLMULTI just set the mask to all ones.
949 * This overrides the mask populated above. */
950 if ((uf.flags & TUN_FLT_ALLMULTI))
951 memset(filter->mask, ~0, sizeof(filter->mask));
953 /* Now enable the filter */
955 filter->count = nexact;
957 /* Return the number of exact filters */
964 /* Returns: 0 - drop, !=0 - accept */
965 static int run_filter(struct tap_filter *filter, const struct sk_buff *skb)
967 /* Cannot use eth_hdr(skb) here because skb_mac_hdr() is incorrect
969 struct ethhdr *eh = (struct ethhdr *) skb->data;
973 for (i = 0; i < filter->count; i++)
974 if (ether_addr_equal(eh->h_dest, filter->addr[i]))
977 /* Inexact match (multicast only) */
978 if (is_multicast_ether_addr(eh->h_dest))
979 return addr_hash_test(filter->mask, eh->h_dest);
985 * Checks whether the packet is accepted or not.
986 * Returns: 0 - drop, !=0 - accept
988 static int check_filter(struct tap_filter *filter, const struct sk_buff *skb)
993 return run_filter(filter, skb);
996 /* Network device part of the driver */
998 static const struct ethtool_ops tun_ethtool_ops;
1000 /* Net device detach from fd. */
1001 static void tun_net_uninit(struct net_device *dev)
1003 tun_detach_all(dev);
1006 /* Net device open. */
1007 static int tun_net_open(struct net_device *dev)
1009 struct tun_struct *tun = netdev_priv(dev);
1012 netif_tx_start_all_queues(dev);
1014 for (i = 0; i < tun->numqueues; i++) {
1015 struct tun_file *tfile;
1017 tfile = rtnl_dereference(tun->tfiles[i]);
1018 tfile->socket.sk->sk_write_space(tfile->socket.sk);
1024 /* Net device close. */
1025 static int tun_net_close(struct net_device *dev)
1027 netif_tx_stop_all_queues(dev);
1031 /* Net device start xmit */
1032 static void tun_automq_xmit(struct tun_struct *tun, struct sk_buff *skb)
1035 if (tun->numqueues == 1 && static_key_false(&rps_needed)) {
1036 /* Select queue was not called for the skbuff, so we extract the
1037 * RPS hash and save it into the flow_table here.
1041 rxhash = __skb_get_hash_symmetric(skb);
1043 struct tun_flow_entry *e;
1044 e = tun_flow_find(&tun->flows[tun_hashfn(rxhash)],
1047 tun_flow_save_rps_rxhash(e, rxhash);
1053 static unsigned int run_ebpf_filter(struct tun_struct *tun,
1054 struct sk_buff *skb,
1057 struct tun_prog *prog = rcu_dereference(tun->filter_prog);
1060 len = bpf_prog_run_clear_cb(prog->prog, skb);
1065 /* Net device start xmit */
1066 static netdev_tx_t tun_net_xmit(struct sk_buff *skb, struct net_device *dev)
1068 struct tun_struct *tun = netdev_priv(dev);
1069 int txq = skb->queue_mapping;
1070 struct tun_file *tfile;
1074 tfile = rcu_dereference(tun->tfiles[txq]);
1076 /* Drop packet if interface is not attached */
1077 if (txq >= tun->numqueues)
1080 if (!rcu_dereference(tun->steering_prog))
1081 tun_automq_xmit(tun, skb);
1083 tun_debug(KERN_INFO, tun, "tun_net_xmit %d\n", skb->len);
1087 /* Drop if the filter does not like it.
1088 * This is a noop if the filter is disabled.
1089 * Filter can be enabled only for the TAP devices. */
1090 if (!check_filter(&tun->txflt, skb))
1093 if (tfile->socket.sk->sk_filter &&
1094 sk_filter(tfile->socket.sk, skb))
1097 len = run_ebpf_filter(tun, skb, len);
1099 /* Trim extra bytes since we may insert vlan proto & TCI
1100 * in tun_put_user().
1102 len -= skb_vlan_tag_present(skb) ? sizeof(struct veth) : 0;
1103 if (len <= 0 || pskb_trim(skb, len))
1106 if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC)))
1109 skb_tx_timestamp(skb);
1111 /* Orphan the skb - required as we might hang on to it
1112 * for indefinite time.
1118 if (ptr_ring_produce(&tfile->tx_ring, skb))
1121 /* Notify and wake up reader process */
1122 if (tfile->flags & TUN_FASYNC)
1123 kill_fasync(&tfile->fasync, SIGIO, POLL_IN);
1124 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
1127 return NETDEV_TX_OK;
1130 this_cpu_inc(tun->pcpu_stats->tx_dropped);
1134 return NET_XMIT_DROP;
1137 static void tun_net_mclist(struct net_device *dev)
1140 * This callback is supposed to deal with mc filter in
1141 * _rx_ path and has nothing to do with the _tx_ path.
1142 * In rx path we always accept everything userspace gives us.
1146 static netdev_features_t tun_net_fix_features(struct net_device *dev,
1147 netdev_features_t features)
1149 struct tun_struct *tun = netdev_priv(dev);
1151 return (features & tun->set_features) | (features & ~TUN_USER_FEATURES);
1153 #ifdef CONFIG_NET_POLL_CONTROLLER
1154 static void tun_poll_controller(struct net_device *dev)
1157 * Tun only receives frames when:
1158 * 1) the char device endpoint gets data from user space
1159 * 2) the tun socket gets a sendmsg call from user space
1160 * If NAPI is not enabled, since both of those are synchronous
1161 * operations, we are guaranteed never to have pending data when we poll
1162 * for it so there is nothing to do here but return.
1163 * We need this though so netpoll recognizes us as an interface that
1164 * supports polling, which enables bridge devices in virt setups to
1165 * still use netconsole
1166 * If NAPI is enabled, however, we need to schedule polling for all
1167 * queues unless we are using napi_gro_frags(), which we call in
1168 * process context and not in NAPI context.
1170 struct tun_struct *tun = netdev_priv(dev);
1172 if (tun->flags & IFF_NAPI) {
1173 struct tun_file *tfile;
1176 if (tun_napi_frags_enabled(tun))
1180 for (i = 0; i < tun->numqueues; i++) {
1181 tfile = rcu_dereference(tun->tfiles[i]);
1182 if (tfile->napi_enabled)
1183 napi_schedule(&tfile->napi);
1191 static void tun_set_headroom(struct net_device *dev, int new_hr)
1193 struct tun_struct *tun = netdev_priv(dev);
1195 if (new_hr < NET_SKB_PAD)
1196 new_hr = NET_SKB_PAD;
1198 tun->align = new_hr;
1202 tun_net_get_stats64(struct net_device *dev, struct rtnl_link_stats64 *stats)
1204 u32 rx_dropped = 0, tx_dropped = 0, rx_frame_errors = 0;
1205 struct tun_struct *tun = netdev_priv(dev);
1206 struct tun_pcpu_stats *p;
1209 for_each_possible_cpu(i) {
1210 u64 rxpackets, rxbytes, txpackets, txbytes;
1213 p = per_cpu_ptr(tun->pcpu_stats, i);
1215 start = u64_stats_fetch_begin(&p->syncp);
1216 rxpackets = p->rx_packets;
1217 rxbytes = p->rx_bytes;
1218 txpackets = p->tx_packets;
1219 txbytes = p->tx_bytes;
1220 } while (u64_stats_fetch_retry(&p->syncp, start));
1222 stats->rx_packets += rxpackets;
1223 stats->rx_bytes += rxbytes;
1224 stats->tx_packets += txpackets;
1225 stats->tx_bytes += txbytes;
1228 rx_dropped += p->rx_dropped;
1229 rx_frame_errors += p->rx_frame_errors;
1230 tx_dropped += p->tx_dropped;
1232 stats->rx_dropped = rx_dropped;
1233 stats->rx_frame_errors = rx_frame_errors;
1234 stats->tx_dropped = tx_dropped;
1237 static int tun_xdp_set(struct net_device *dev, struct bpf_prog *prog,
1238 struct netlink_ext_ack *extack)
1240 struct tun_struct *tun = netdev_priv(dev);
1241 struct bpf_prog *old_prog;
1243 old_prog = rtnl_dereference(tun->xdp_prog);
1244 rcu_assign_pointer(tun->xdp_prog, prog);
1246 bpf_prog_put(old_prog);
1251 static u32 tun_xdp_query(struct net_device *dev)
1253 struct tun_struct *tun = netdev_priv(dev);
1254 const struct bpf_prog *xdp_prog;
1256 xdp_prog = rtnl_dereference(tun->xdp_prog);
1258 return xdp_prog->aux->id;
1263 static int tun_xdp(struct net_device *dev, struct netdev_bpf *xdp)
1265 switch (xdp->command) {
1266 case XDP_SETUP_PROG:
1267 return tun_xdp_set(dev, xdp->prog, xdp->extack);
1268 case XDP_QUERY_PROG:
1269 xdp->prog_id = tun_xdp_query(dev);
1270 xdp->prog_attached = !!xdp->prog_id;
1277 static const struct net_device_ops tun_netdev_ops = {
1278 .ndo_uninit = tun_net_uninit,
1279 .ndo_open = tun_net_open,
1280 .ndo_stop = tun_net_close,
1281 .ndo_start_xmit = tun_net_xmit,
1282 .ndo_fix_features = tun_net_fix_features,
1283 .ndo_select_queue = tun_select_queue,
1284 #ifdef CONFIG_NET_POLL_CONTROLLER
1285 .ndo_poll_controller = tun_poll_controller,
1287 .ndo_set_rx_headroom = tun_set_headroom,
1288 .ndo_get_stats64 = tun_net_get_stats64,
1291 static int tun_xdp_xmit(struct net_device *dev, struct xdp_buff *xdp)
1293 struct tun_struct *tun = netdev_priv(dev);
1294 struct xdp_buff *buff = xdp->data_hard_start;
1295 int headroom = xdp->data - xdp->data_hard_start;
1296 struct tun_file *tfile;
1300 /* Assure headroom is available and buff is properly aligned */
1301 if (unlikely(headroom < sizeof(*xdp) || tun_is_xdp_buff(xdp)))
1308 numqueues = READ_ONCE(tun->numqueues);
1314 tfile = rcu_dereference(tun->tfiles[smp_processor_id() %
1316 /* Encode the XDP flag into lowest bit for consumer to differ
1317 * XDP buffer from sk_buff.
1319 if (ptr_ring_produce(&tfile->tx_ring, tun_xdp_to_ptr(buff))) {
1320 this_cpu_inc(tun->pcpu_stats->tx_dropped);
1329 static void tun_xdp_flush(struct net_device *dev)
1331 struct tun_struct *tun = netdev_priv(dev);
1332 struct tun_file *tfile;
1337 numqueues = READ_ONCE(tun->numqueues);
1341 tfile = rcu_dereference(tun->tfiles[smp_processor_id() %
1343 /* Notify and wake up reader process */
1344 if (tfile->flags & TUN_FASYNC)
1345 kill_fasync(&tfile->fasync, SIGIO, POLL_IN);
1346 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
1352 static const struct net_device_ops tap_netdev_ops = {
1353 .ndo_uninit = tun_net_uninit,
1354 .ndo_open = tun_net_open,
1355 .ndo_stop = tun_net_close,
1356 .ndo_start_xmit = tun_net_xmit,
1357 .ndo_fix_features = tun_net_fix_features,
1358 .ndo_set_rx_mode = tun_net_mclist,
1359 .ndo_set_mac_address = eth_mac_addr,
1360 .ndo_validate_addr = eth_validate_addr,
1361 .ndo_select_queue = tun_select_queue,
1362 #ifdef CONFIG_NET_POLL_CONTROLLER
1363 .ndo_poll_controller = tun_poll_controller,
1365 .ndo_features_check = passthru_features_check,
1366 .ndo_set_rx_headroom = tun_set_headroom,
1367 .ndo_get_stats64 = tun_net_get_stats64,
1369 .ndo_xdp_xmit = tun_xdp_xmit,
1370 .ndo_xdp_flush = tun_xdp_flush,
1373 static void tun_flow_init(struct tun_struct *tun)
1377 for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++)
1378 INIT_HLIST_HEAD(&tun->flows[i]);
1380 tun->ageing_time = TUN_FLOW_EXPIRE;
1381 timer_setup(&tun->flow_gc_timer, tun_flow_cleanup, 0);
1382 mod_timer(&tun->flow_gc_timer,
1383 round_jiffies_up(jiffies + tun->ageing_time));
1386 static void tun_flow_uninit(struct tun_struct *tun)
1388 del_timer_sync(&tun->flow_gc_timer);
1389 tun_flow_flush(tun);
1393 #define MAX_MTU 65535
1395 /* Initialize net device. */
1396 static void tun_net_init(struct net_device *dev)
1398 struct tun_struct *tun = netdev_priv(dev);
1400 switch (tun->flags & TUN_TYPE_MASK) {
1402 dev->netdev_ops = &tun_netdev_ops;
1404 /* Point-to-Point TUN Device */
1405 dev->hard_header_len = 0;
1409 /* Zero header length */
1410 dev->type = ARPHRD_NONE;
1411 dev->flags = IFF_POINTOPOINT | IFF_NOARP | IFF_MULTICAST;
1415 dev->netdev_ops = &tap_netdev_ops;
1416 /* Ethernet TAP Device */
1418 dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1419 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
1421 eth_hw_addr_random(dev);
1426 dev->min_mtu = MIN_MTU;
1427 dev->max_mtu = MAX_MTU - dev->hard_header_len;
1430 /* Character device part */
1433 static __poll_t tun_chr_poll(struct file *file, poll_table *wait)
1435 struct tun_file *tfile = file->private_data;
1436 struct tun_struct *tun = tun_get(tfile);
1443 sk = tfile->socket.sk;
1445 tun_debug(KERN_INFO, tun, "tun_chr_poll\n");
1447 poll_wait(file, sk_sleep(sk), wait);
1449 if (!ptr_ring_empty(&tfile->tx_ring))
1450 mask |= EPOLLIN | EPOLLRDNORM;
1452 if (tun->dev->flags & IFF_UP &&
1453 (sock_writeable(sk) ||
1454 (!test_and_set_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags) &&
1455 sock_writeable(sk))))
1456 mask |= EPOLLOUT | EPOLLWRNORM;
1458 if (tun->dev->reg_state != NETREG_REGISTERED)
1465 static struct sk_buff *tun_napi_alloc_frags(struct tun_file *tfile,
1467 const struct iov_iter *it)
1469 struct sk_buff *skb;
1474 if (it->nr_segs > MAX_SKB_FRAGS + 1)
1475 return ERR_PTR(-ENOMEM);
1478 skb = napi_get_frags(&tfile->napi);
1481 return ERR_PTR(-ENOMEM);
1483 linear = iov_iter_single_seg_count(it);
1484 err = __skb_grow(skb, linear);
1489 skb->data_len = len - linear;
1490 skb->truesize += skb->data_len;
1492 for (i = 1; i < it->nr_segs; i++) {
1493 size_t fragsz = it->iov[i].iov_len;
1494 unsigned long offset;
1498 if (fragsz == 0 || fragsz > PAGE_SIZE) {
1504 data = napi_alloc_frag(fragsz);
1511 page = virt_to_head_page(data);
1512 offset = data - page_address(page);
1513 skb_fill_page_desc(skb, i - 1, page, offset, fragsz);
1518 /* frees skb and all frags allocated with napi_alloc_frag() */
1519 napi_free_frags(&tfile->napi);
1520 return ERR_PTR(err);
1523 /* prepad is the amount to reserve at front. len is length after that.
1524 * linear is a hint as to how much to copy (usually headers). */
1525 static struct sk_buff *tun_alloc_skb(struct tun_file *tfile,
1526 size_t prepad, size_t len,
1527 size_t linear, int noblock)
1529 struct sock *sk = tfile->socket.sk;
1530 struct sk_buff *skb;
1533 /* Under a page? Don't bother with paged skb. */
1534 if (prepad + len < PAGE_SIZE || !linear)
1537 skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
1540 return ERR_PTR(err);
1542 skb_reserve(skb, prepad);
1543 skb_put(skb, linear);
1544 skb->data_len = len - linear;
1545 skb->len += len - linear;
1550 static void tun_rx_batched(struct tun_struct *tun, struct tun_file *tfile,
1551 struct sk_buff *skb, int more)
1553 struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
1554 struct sk_buff_head process_queue;
1555 u32 rx_batched = tun->rx_batched;
1558 if (!rx_batched || (!more && skb_queue_empty(queue))) {
1560 netif_receive_skb(skb);
1565 spin_lock(&queue->lock);
1566 if (!more || skb_queue_len(queue) == rx_batched) {
1567 __skb_queue_head_init(&process_queue);
1568 skb_queue_splice_tail_init(queue, &process_queue);
1571 __skb_queue_tail(queue, skb);
1573 spin_unlock(&queue->lock);
1576 struct sk_buff *nskb;
1579 while ((nskb = __skb_dequeue(&process_queue)))
1580 netif_receive_skb(nskb);
1581 netif_receive_skb(skb);
1586 static bool tun_can_build_skb(struct tun_struct *tun, struct tun_file *tfile,
1587 int len, int noblock, bool zerocopy)
1589 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
1592 if (tfile->socket.sk->sk_sndbuf != INT_MAX)
1601 if (SKB_DATA_ALIGN(len + TUN_RX_PAD) +
1602 SKB_DATA_ALIGN(sizeof(struct skb_shared_info)) > PAGE_SIZE)
1608 static struct sk_buff *tun_build_skb(struct tun_struct *tun,
1609 struct tun_file *tfile,
1610 struct iov_iter *from,
1611 struct virtio_net_hdr *hdr,
1612 int len, int *skb_xdp)
1614 struct page_frag *alloc_frag = ¤t->task_frag;
1615 struct sk_buff *skb;
1616 struct bpf_prog *xdp_prog;
1617 int buflen = SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
1618 unsigned int delta = 0;
1621 bool xdp_xmit = false;
1622 int err, pad = TUN_RX_PAD;
1625 xdp_prog = rcu_dereference(tun->xdp_prog);
1627 pad += TUN_HEADROOM;
1628 buflen += SKB_DATA_ALIGN(len + pad);
1631 alloc_frag->offset = ALIGN((u64)alloc_frag->offset, SMP_CACHE_BYTES);
1632 if (unlikely(!skb_page_frag_refill(buflen, alloc_frag, GFP_KERNEL)))
1633 return ERR_PTR(-ENOMEM);
1635 buf = (char *)page_address(alloc_frag->page) + alloc_frag->offset;
1636 copied = copy_page_from_iter(alloc_frag->page,
1637 alloc_frag->offset + pad,
1640 return ERR_PTR(-EFAULT);
1642 /* There's a small window that XDP may be set after the check
1643 * of xdp_prog above, this should be rare and for simplicity
1644 * we do XDP on skb in case the headroom is not enough.
1646 if (hdr->gso_type || !xdp_prog)
1652 xdp_prog = rcu_dereference(tun->xdp_prog);
1653 if (xdp_prog && !*skb_xdp) {
1654 struct xdp_buff xdp;
1658 xdp.data_hard_start = buf;
1659 xdp.data = buf + pad;
1660 xdp_set_data_meta_invalid(&xdp);
1661 xdp.data_end = xdp.data + len;
1662 xdp.rxq = &tfile->xdp_rxq;
1663 orig_data = xdp.data;
1664 act = bpf_prog_run_xdp(xdp_prog, &xdp);
1668 get_page(alloc_frag->page);
1669 alloc_frag->offset += buflen;
1670 ++tfile->xdp_pending_pkts;
1671 err = xdp_do_redirect(tun->dev, &xdp, xdp_prog);
1680 delta = orig_data - xdp.data;
1683 bpf_warn_invalid_xdp_action(act);
1686 trace_xdp_exception(tun->dev, xdp_prog, act);
1693 skb = build_skb(buf, buflen);
1696 return ERR_PTR(-ENOMEM);
1699 skb_reserve(skb, pad - delta);
1700 skb_put(skb, len + delta);
1701 get_page(alloc_frag->page);
1702 alloc_frag->offset += buflen;
1705 skb->dev = tun->dev;
1706 generic_xdp_tx(skb, xdp_prog);
1716 put_page(alloc_frag->page);
1719 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1723 /* Get packet from user space buffer */
1724 static ssize_t tun_get_user(struct tun_struct *tun, struct tun_file *tfile,
1725 void *msg_control, struct iov_iter *from,
1726 int noblock, bool more)
1728 struct tun_pi pi = { 0, cpu_to_be16(ETH_P_IP) };
1729 struct sk_buff *skb;
1730 size_t total_len = iov_iter_count(from);
1731 size_t len = total_len, align = tun->align, linear;
1732 struct virtio_net_hdr gso = { 0 };
1733 struct tun_pcpu_stats *stats;
1736 bool zerocopy = false;
1740 bool frags = tun_napi_frags_enabled(tun);
1742 if (!(tun->dev->flags & IFF_UP))
1745 if (!(tun->flags & IFF_NO_PI)) {
1746 if (len < sizeof(pi))
1750 if (!copy_from_iter_full(&pi, sizeof(pi), from))
1754 if (tun->flags & IFF_VNET_HDR) {
1755 int vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
1757 if (len < vnet_hdr_sz)
1761 if (!copy_from_iter_full(&gso, sizeof(gso), from))
1764 if ((gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
1765 tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2 > tun16_to_cpu(tun, gso.hdr_len))
1766 gso.hdr_len = cpu_to_tun16(tun, tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2);
1768 if (tun16_to_cpu(tun, gso.hdr_len) > len)
1770 iov_iter_advance(from, vnet_hdr_sz - sizeof(gso));
1773 if ((tun->flags & TUN_TYPE_MASK) == IFF_TAP) {
1774 align += NET_IP_ALIGN;
1775 if (unlikely(len < ETH_HLEN ||
1776 (gso.hdr_len && tun16_to_cpu(tun, gso.hdr_len) < ETH_HLEN)))
1780 good_linear = SKB_MAX_HEAD(align);
1783 struct iov_iter i = *from;
1785 /* There are 256 bytes to be copied in skb, so there is
1786 * enough room for skb expand head in case it is used.
1787 * The rest of the buffer is mapped from userspace.
1789 copylen = gso.hdr_len ? tun16_to_cpu(tun, gso.hdr_len) : GOODCOPY_LEN;
1790 if (copylen > good_linear)
1791 copylen = good_linear;
1793 iov_iter_advance(&i, copylen);
1794 if (iov_iter_npages(&i, INT_MAX) <= MAX_SKB_FRAGS)
1798 if (!frags && tun_can_build_skb(tun, tfile, len, noblock, zerocopy)) {
1799 /* For the packet that is not easy to be processed
1800 * (e.g gso or jumbo packet), we will do it at after
1801 * skb was created with generic XDP routine.
1803 skb = tun_build_skb(tun, tfile, from, &gso, len, &skb_xdp);
1805 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1806 return PTR_ERR(skb);
1813 if (tun16_to_cpu(tun, gso.hdr_len) > good_linear)
1814 linear = good_linear;
1816 linear = tun16_to_cpu(tun, gso.hdr_len);
1820 mutex_lock(&tfile->napi_mutex);
1821 skb = tun_napi_alloc_frags(tfile, copylen, from);
1822 /* tun_napi_alloc_frags() enforces a layout for the skb.
1823 * If zerocopy is enabled, then this layout will be
1824 * overwritten by zerocopy_sg_from_iter().
1828 skb = tun_alloc_skb(tfile, align, copylen, linear,
1833 if (PTR_ERR(skb) != -EAGAIN)
1834 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1836 mutex_unlock(&tfile->napi_mutex);
1837 return PTR_ERR(skb);
1841 err = zerocopy_sg_from_iter(skb, from);
1843 err = skb_copy_datagram_from_iter(skb, 0, from, len);
1846 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1849 tfile->napi.skb = NULL;
1850 mutex_unlock(&tfile->napi_mutex);
1857 if (virtio_net_hdr_to_skb(skb, &gso, tun_is_little_endian(tun))) {
1858 this_cpu_inc(tun->pcpu_stats->rx_frame_errors);
1861 tfile->napi.skb = NULL;
1862 mutex_unlock(&tfile->napi_mutex);
1868 switch (tun->flags & TUN_TYPE_MASK) {
1870 if (tun->flags & IFF_NO_PI) {
1871 u8 ip_version = skb->len ? (skb->data[0] >> 4) : 0;
1873 switch (ip_version) {
1875 pi.proto = htons(ETH_P_IP);
1878 pi.proto = htons(ETH_P_IPV6);
1881 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1887 skb_reset_mac_header(skb);
1888 skb->protocol = pi.proto;
1889 skb->dev = tun->dev;
1893 skb->protocol = eth_type_trans(skb, tun->dev);
1897 /* copy skb_ubuf_info for callback when skb has no error */
1899 skb_shinfo(skb)->destructor_arg = msg_control;
1900 skb_shinfo(skb)->tx_flags |= SKBTX_DEV_ZEROCOPY;
1901 skb_shinfo(skb)->tx_flags |= SKBTX_SHARED_FRAG;
1902 } else if (msg_control) {
1903 struct ubuf_info *uarg = msg_control;
1904 uarg->callback(uarg, false);
1907 skb_reset_network_header(skb);
1908 skb_probe_transport_header(skb, 0);
1911 struct bpf_prog *xdp_prog;
1915 xdp_prog = rcu_dereference(tun->xdp_prog);
1917 ret = do_xdp_generic(xdp_prog, skb);
1918 if (ret != XDP_PASS) {
1927 if (!rcu_dereference(tun->steering_prog))
1928 rxhash = __skb_get_hash_symmetric(skb);
1932 /* Exercise flow dissector code path. */
1933 u32 headlen = eth_get_headlen(skb->data, skb_headlen(skb));
1935 if (unlikely(headlen > skb_headlen(skb))) {
1936 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1937 napi_free_frags(&tfile->napi);
1938 mutex_unlock(&tfile->napi_mutex);
1944 napi_gro_frags(&tfile->napi);
1946 mutex_unlock(&tfile->napi_mutex);
1947 } else if (tfile->napi_enabled) {
1948 struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
1951 spin_lock_bh(&queue->lock);
1952 __skb_queue_tail(queue, skb);
1953 queue_len = skb_queue_len(queue);
1954 spin_unlock(&queue->lock);
1956 if (!more || queue_len > NAPI_POLL_WEIGHT)
1957 napi_schedule(&tfile->napi);
1960 } else if (!IS_ENABLED(CONFIG_4KSTACKS)) {
1961 tun_rx_batched(tun, tfile, skb, more);
1966 stats = get_cpu_ptr(tun->pcpu_stats);
1967 u64_stats_update_begin(&stats->syncp);
1968 stats->rx_packets++;
1969 stats->rx_bytes += len;
1970 u64_stats_update_end(&stats->syncp);
1974 tun_flow_update(tun, rxhash, tfile);
1979 static ssize_t tun_chr_write_iter(struct kiocb *iocb, struct iov_iter *from)
1981 struct file *file = iocb->ki_filp;
1982 struct tun_file *tfile = file->private_data;
1983 struct tun_struct *tun = tun_get(tfile);
1989 result = tun_get_user(tun, tfile, NULL, from,
1990 file->f_flags & O_NONBLOCK, false);
1992 if (tfile->xdp_pending_pkts) {
1993 tfile->xdp_pending_pkts = 0;
2001 static ssize_t tun_put_user_xdp(struct tun_struct *tun,
2002 struct tun_file *tfile,
2003 struct xdp_buff *xdp,
2004 struct iov_iter *iter)
2006 int vnet_hdr_sz = 0;
2007 size_t size = xdp->data_end - xdp->data;
2008 struct tun_pcpu_stats *stats;
2011 if (tun->flags & IFF_VNET_HDR) {
2012 struct virtio_net_hdr gso = { 0 };
2014 vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
2015 if (unlikely(iov_iter_count(iter) < vnet_hdr_sz))
2017 if (unlikely(copy_to_iter(&gso, sizeof(gso), iter) !=
2020 iov_iter_advance(iter, vnet_hdr_sz - sizeof(gso));
2023 ret = copy_to_iter(xdp->data, size, iter) + vnet_hdr_sz;
2025 stats = get_cpu_ptr(tun->pcpu_stats);
2026 u64_stats_update_begin(&stats->syncp);
2027 stats->tx_packets++;
2028 stats->tx_bytes += ret;
2029 u64_stats_update_end(&stats->syncp);
2030 put_cpu_ptr(tun->pcpu_stats);
2035 /* Put packet to the user space buffer */
2036 static ssize_t tun_put_user(struct tun_struct *tun,
2037 struct tun_file *tfile,
2038 struct sk_buff *skb,
2039 struct iov_iter *iter)
2041 struct tun_pi pi = { 0, skb->protocol };
2042 struct tun_pcpu_stats *stats;
2044 int vlan_offset = 0;
2046 int vnet_hdr_sz = 0;
2048 if (skb_vlan_tag_present(skb))
2049 vlan_hlen = VLAN_HLEN;
2051 if (tun->flags & IFF_VNET_HDR)
2052 vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
2054 total = skb->len + vlan_hlen + vnet_hdr_sz;
2056 if (!(tun->flags & IFF_NO_PI)) {
2057 if (iov_iter_count(iter) < sizeof(pi))
2060 total += sizeof(pi);
2061 if (iov_iter_count(iter) < total) {
2062 /* Packet will be striped */
2063 pi.flags |= TUN_PKT_STRIP;
2066 if (copy_to_iter(&pi, sizeof(pi), iter) != sizeof(pi))
2071 struct virtio_net_hdr gso;
2073 if (iov_iter_count(iter) < vnet_hdr_sz)
2076 if (virtio_net_hdr_from_skb(skb, &gso,
2077 tun_is_little_endian(tun), true)) {
2078 struct skb_shared_info *sinfo = skb_shinfo(skb);
2079 pr_err("unexpected GSO type: "
2080 "0x%x, gso_size %d, hdr_len %d\n",
2081 sinfo->gso_type, tun16_to_cpu(tun, gso.gso_size),
2082 tun16_to_cpu(tun, gso.hdr_len));
2083 print_hex_dump(KERN_ERR, "tun: ",
2086 min((int)tun16_to_cpu(tun, gso.hdr_len), 64), true);
2091 if (copy_to_iter(&gso, sizeof(gso), iter) != sizeof(gso))
2094 iov_iter_advance(iter, vnet_hdr_sz - sizeof(gso));
2101 veth.h_vlan_proto = skb->vlan_proto;
2102 veth.h_vlan_TCI = htons(skb_vlan_tag_get(skb));
2104 vlan_offset = offsetof(struct vlan_ethhdr, h_vlan_proto);
2106 ret = skb_copy_datagram_iter(skb, 0, iter, vlan_offset);
2107 if (ret || !iov_iter_count(iter))
2110 ret = copy_to_iter(&veth, sizeof(veth), iter);
2111 if (ret != sizeof(veth) || !iov_iter_count(iter))
2115 skb_copy_datagram_iter(skb, vlan_offset, iter, skb->len - vlan_offset);
2118 /* caller is in process context, */
2119 stats = get_cpu_ptr(tun->pcpu_stats);
2120 u64_stats_update_begin(&stats->syncp);
2121 stats->tx_packets++;
2122 stats->tx_bytes += skb->len + vlan_hlen;
2123 u64_stats_update_end(&stats->syncp);
2124 put_cpu_ptr(tun->pcpu_stats);
2129 static void *tun_ring_recv(struct tun_file *tfile, int noblock, int *err)
2131 DECLARE_WAITQUEUE(wait, current);
2135 ptr = ptr_ring_consume(&tfile->tx_ring);
2143 add_wait_queue(&tfile->wq.wait, &wait);
2144 current->state = TASK_INTERRUPTIBLE;
2147 ptr = ptr_ring_consume(&tfile->tx_ring);
2150 if (signal_pending(current)) {
2151 error = -ERESTARTSYS;
2154 if (tfile->socket.sk->sk_shutdown & RCV_SHUTDOWN) {
2162 current->state = TASK_RUNNING;
2163 remove_wait_queue(&tfile->wq.wait, &wait);
2170 static ssize_t tun_do_read(struct tun_struct *tun, struct tun_file *tfile,
2171 struct iov_iter *to,
2172 int noblock, void *ptr)
2177 tun_debug(KERN_INFO, tun, "tun_do_read\n");
2179 if (!iov_iter_count(to)) {
2185 /* Read frames from ring */
2186 ptr = tun_ring_recv(tfile, noblock, &err);
2191 if (tun_is_xdp_buff(ptr)) {
2192 struct xdp_buff *xdp = tun_ptr_to_xdp(ptr);
2194 ret = tun_put_user_xdp(tun, tfile, xdp, to);
2195 put_page(virt_to_head_page(xdp->data));
2197 struct sk_buff *skb = ptr;
2199 ret = tun_put_user(tun, tfile, skb, to);
2200 if (unlikely(ret < 0))
2209 static ssize_t tun_chr_read_iter(struct kiocb *iocb, struct iov_iter *to)
2211 struct file *file = iocb->ki_filp;
2212 struct tun_file *tfile = file->private_data;
2213 struct tun_struct *tun = tun_get(tfile);
2214 ssize_t len = iov_iter_count(to), ret;
2218 ret = tun_do_read(tun, tfile, to, file->f_flags & O_NONBLOCK, NULL);
2219 ret = min_t(ssize_t, ret, len);
2226 static void tun_prog_free(struct rcu_head *rcu)
2228 struct tun_prog *prog = container_of(rcu, struct tun_prog, rcu);
2230 bpf_prog_destroy(prog->prog);
2234 static int __tun_set_ebpf(struct tun_struct *tun,
2235 struct tun_prog __rcu **prog_p,
2236 struct bpf_prog *prog)
2238 struct tun_prog *old, *new = NULL;
2241 new = kmalloc(sizeof(*new), GFP_KERNEL);
2247 spin_lock_bh(&tun->lock);
2248 old = rcu_dereference_protected(*prog_p,
2249 lockdep_is_held(&tun->lock));
2250 rcu_assign_pointer(*prog_p, new);
2251 spin_unlock_bh(&tun->lock);
2254 call_rcu(&old->rcu, tun_prog_free);
2259 static void tun_free_netdev(struct net_device *dev)
2261 struct tun_struct *tun = netdev_priv(dev);
2263 BUG_ON(!(list_empty(&tun->disabled)));
2264 free_percpu(tun->pcpu_stats);
2265 tun_flow_uninit(tun);
2266 security_tun_dev_free_security(tun->security);
2267 __tun_set_ebpf(tun, &tun->steering_prog, NULL);
2268 __tun_set_ebpf(tun, &tun->filter_prog, NULL);
2271 static void tun_setup(struct net_device *dev)
2273 struct tun_struct *tun = netdev_priv(dev);
2275 tun->owner = INVALID_UID;
2276 tun->group = INVALID_GID;
2278 dev->ethtool_ops = &tun_ethtool_ops;
2279 dev->needs_free_netdev = true;
2280 dev->priv_destructor = tun_free_netdev;
2281 /* We prefer our own queue length */
2282 dev->tx_queue_len = TUN_READQ_SIZE;
2285 /* Trivial set of netlink ops to allow deleting tun or tap
2286 * device with netlink.
2288 static int tun_validate(struct nlattr *tb[], struct nlattr *data[],
2289 struct netlink_ext_ack *extack)
2294 static size_t tun_get_size(const struct net_device *dev)
2296 BUILD_BUG_ON(sizeof(u32) != sizeof(uid_t));
2297 BUILD_BUG_ON(sizeof(u32) != sizeof(gid_t));
2299 return nla_total_size(sizeof(uid_t)) + /* OWNER */
2300 nla_total_size(sizeof(gid_t)) + /* GROUP */
2301 nla_total_size(sizeof(u8)) + /* TYPE */
2302 nla_total_size(sizeof(u8)) + /* PI */
2303 nla_total_size(sizeof(u8)) + /* VNET_HDR */
2304 nla_total_size(sizeof(u8)) + /* PERSIST */
2305 nla_total_size(sizeof(u8)) + /* MULTI_QUEUE */
2306 nla_total_size(sizeof(u32)) + /* NUM_QUEUES */
2307 nla_total_size(sizeof(u32)) + /* NUM_DISABLED_QUEUES */
2311 static int tun_fill_info(struct sk_buff *skb, const struct net_device *dev)
2313 struct tun_struct *tun = netdev_priv(dev);
2315 if (nla_put_u8(skb, IFLA_TUN_TYPE, tun->flags & TUN_TYPE_MASK))
2316 goto nla_put_failure;
2317 if (uid_valid(tun->owner) &&
2318 nla_put_u32(skb, IFLA_TUN_OWNER,
2319 from_kuid_munged(current_user_ns(), tun->owner)))
2320 goto nla_put_failure;
2321 if (gid_valid(tun->group) &&
2322 nla_put_u32(skb, IFLA_TUN_GROUP,
2323 from_kgid_munged(current_user_ns(), tun->group)))
2324 goto nla_put_failure;
2325 if (nla_put_u8(skb, IFLA_TUN_PI, !(tun->flags & IFF_NO_PI)))
2326 goto nla_put_failure;
2327 if (nla_put_u8(skb, IFLA_TUN_VNET_HDR, !!(tun->flags & IFF_VNET_HDR)))
2328 goto nla_put_failure;
2329 if (nla_put_u8(skb, IFLA_TUN_PERSIST, !!(tun->flags & IFF_PERSIST)))
2330 goto nla_put_failure;
2331 if (nla_put_u8(skb, IFLA_TUN_MULTI_QUEUE,
2332 !!(tun->flags & IFF_MULTI_QUEUE)))
2333 goto nla_put_failure;
2334 if (tun->flags & IFF_MULTI_QUEUE) {
2335 if (nla_put_u32(skb, IFLA_TUN_NUM_QUEUES, tun->numqueues))
2336 goto nla_put_failure;
2337 if (nla_put_u32(skb, IFLA_TUN_NUM_DISABLED_QUEUES,
2339 goto nla_put_failure;
2348 static struct rtnl_link_ops tun_link_ops __read_mostly = {
2350 .priv_size = sizeof(struct tun_struct),
2352 .validate = tun_validate,
2353 .get_size = tun_get_size,
2354 .fill_info = tun_fill_info,
2357 static void tun_sock_write_space(struct sock *sk)
2359 struct tun_file *tfile;
2360 wait_queue_head_t *wqueue;
2362 if (!sock_writeable(sk))
2365 if (!test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags))
2368 wqueue = sk_sleep(sk);
2369 if (wqueue && waitqueue_active(wqueue))
2370 wake_up_interruptible_sync_poll(wqueue, EPOLLOUT |
2371 EPOLLWRNORM | EPOLLWRBAND);
2373 tfile = container_of(sk, struct tun_file, sk);
2374 kill_fasync(&tfile->fasync, SIGIO, POLL_OUT);
2377 static int tun_sendmsg(struct socket *sock, struct msghdr *m, size_t total_len)
2380 struct tun_file *tfile = container_of(sock, struct tun_file, socket);
2381 struct tun_struct *tun = tun_get(tfile);
2386 ret = tun_get_user(tun, tfile, m->msg_control, &m->msg_iter,
2387 m->msg_flags & MSG_DONTWAIT,
2388 m->msg_flags & MSG_MORE);
2390 if (tfile->xdp_pending_pkts >= NAPI_POLL_WEIGHT ||
2391 !(m->msg_flags & MSG_MORE)) {
2392 tfile->xdp_pending_pkts = 0;
2400 static int tun_recvmsg(struct socket *sock, struct msghdr *m, size_t total_len,
2403 struct tun_file *tfile = container_of(sock, struct tun_file, socket);
2404 struct tun_struct *tun = tun_get(tfile);
2405 void *ptr = m->msg_control;
2413 if (flags & ~(MSG_DONTWAIT|MSG_TRUNC|MSG_ERRQUEUE)) {
2417 if (flags & MSG_ERRQUEUE) {
2418 ret = sock_recv_errqueue(sock->sk, m, total_len,
2419 SOL_PACKET, TUN_TX_TIMESTAMP);
2422 ret = tun_do_read(tun, tfile, &m->msg_iter, flags & MSG_DONTWAIT, ptr);
2423 if (ret > (ssize_t)total_len) {
2424 m->msg_flags |= MSG_TRUNC;
2425 ret = flags & MSG_TRUNC ? ret : total_len;
2438 static int tun_ptr_peek_len(void *ptr)
2441 if (tun_is_xdp_buff(ptr)) {
2442 struct xdp_buff *xdp = tun_ptr_to_xdp(ptr);
2444 return xdp->data_end - xdp->data;
2446 return __skb_array_len_with_tag(ptr);
2452 static int tun_peek_len(struct socket *sock)
2454 struct tun_file *tfile = container_of(sock, struct tun_file, socket);
2455 struct tun_struct *tun;
2458 tun = tun_get(tfile);
2462 ret = PTR_RING_PEEK_CALL(&tfile->tx_ring, tun_ptr_peek_len);
2468 /* Ops structure to mimic raw sockets with tun */
2469 static const struct proto_ops tun_socket_ops = {
2470 .peek_len = tun_peek_len,
2471 .sendmsg = tun_sendmsg,
2472 .recvmsg = tun_recvmsg,
2475 static struct proto tun_proto = {
2477 .owner = THIS_MODULE,
2478 .obj_size = sizeof(struct tun_file),
2481 static int tun_flags(struct tun_struct *tun)
2483 return tun->flags & (TUN_FEATURES | IFF_PERSIST | IFF_TUN | IFF_TAP);
2486 static ssize_t tun_show_flags(struct device *dev, struct device_attribute *attr,
2489 struct tun_struct *tun = netdev_priv(to_net_dev(dev));
2490 return sprintf(buf, "0x%x\n", tun_flags(tun));
2493 static ssize_t tun_show_owner(struct device *dev, struct device_attribute *attr,
2496 struct tun_struct *tun = netdev_priv(to_net_dev(dev));
2497 return uid_valid(tun->owner)?
2498 sprintf(buf, "%u\n",
2499 from_kuid_munged(current_user_ns(), tun->owner)):
2500 sprintf(buf, "-1\n");
2503 static ssize_t tun_show_group(struct device *dev, struct device_attribute *attr,
2506 struct tun_struct *tun = netdev_priv(to_net_dev(dev));
2507 return gid_valid(tun->group) ?
2508 sprintf(buf, "%u\n",
2509 from_kgid_munged(current_user_ns(), tun->group)):
2510 sprintf(buf, "-1\n");
2513 static DEVICE_ATTR(tun_flags, 0444, tun_show_flags, NULL);
2514 static DEVICE_ATTR(owner, 0444, tun_show_owner, NULL);
2515 static DEVICE_ATTR(group, 0444, tun_show_group, NULL);
2517 static struct attribute *tun_dev_attrs[] = {
2518 &dev_attr_tun_flags.attr,
2519 &dev_attr_owner.attr,
2520 &dev_attr_group.attr,
2524 static const struct attribute_group tun_attr_group = {
2525 .attrs = tun_dev_attrs
2528 static int tun_set_iff(struct net *net, struct file *file, struct ifreq *ifr)
2530 struct tun_struct *tun;
2531 struct tun_file *tfile = file->private_data;
2532 struct net_device *dev;
2535 if (tfile->detached)
2538 if ((ifr->ifr_flags & IFF_NAPI_FRAGS)) {
2539 if (!capable(CAP_NET_ADMIN))
2542 if (!(ifr->ifr_flags & IFF_NAPI) ||
2543 (ifr->ifr_flags & TUN_TYPE_MASK) != IFF_TAP)
2547 dev = __dev_get_by_name(net, ifr->ifr_name);
2549 if (ifr->ifr_flags & IFF_TUN_EXCL)
2551 if ((ifr->ifr_flags & IFF_TUN) && dev->netdev_ops == &tun_netdev_ops)
2552 tun = netdev_priv(dev);
2553 else if ((ifr->ifr_flags & IFF_TAP) && dev->netdev_ops == &tap_netdev_ops)
2554 tun = netdev_priv(dev);
2558 if (!!(ifr->ifr_flags & IFF_MULTI_QUEUE) !=
2559 !!(tun->flags & IFF_MULTI_QUEUE))
2562 if (tun_not_capable(tun))
2564 err = security_tun_dev_open(tun->security);
2568 err = tun_attach(tun, file, ifr->ifr_flags & IFF_NOFILTER,
2569 ifr->ifr_flags & IFF_NAPI);
2573 if (tun->flags & IFF_MULTI_QUEUE &&
2574 (tun->numqueues + tun->numdisabled > 1)) {
2575 /* One or more queue has already been attached, no need
2576 * to initialize the device again.
2583 unsigned long flags = 0;
2584 int queues = ifr->ifr_flags & IFF_MULTI_QUEUE ?
2587 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
2589 err = security_tun_dev_create();
2594 if (ifr->ifr_flags & IFF_TUN) {
2598 } else if (ifr->ifr_flags & IFF_TAP) {
2606 name = ifr->ifr_name;
2608 dev = alloc_netdev_mqs(sizeof(struct tun_struct), name,
2609 NET_NAME_UNKNOWN, tun_setup, queues,
2614 err = dev_get_valid_name(net, dev, name);
2618 dev_net_set(dev, net);
2619 dev->rtnl_link_ops = &tun_link_ops;
2620 dev->ifindex = tfile->ifindex;
2621 dev->sysfs_groups[0] = &tun_attr_group;
2623 tun = netdev_priv(dev);
2626 tun->txflt.count = 0;
2627 tun->vnet_hdr_sz = sizeof(struct virtio_net_hdr);
2629 tun->align = NET_SKB_PAD;
2630 tun->filter_attached = false;
2631 tun->sndbuf = tfile->socket.sk->sk_sndbuf;
2632 tun->rx_batched = 0;
2633 RCU_INIT_POINTER(tun->steering_prog, NULL);
2635 tun->pcpu_stats = netdev_alloc_pcpu_stats(struct tun_pcpu_stats);
2636 if (!tun->pcpu_stats) {
2641 spin_lock_init(&tun->lock);
2643 err = security_tun_dev_alloc_security(&tun->security);
2650 dev->hw_features = NETIF_F_SG | NETIF_F_FRAGLIST |
2651 TUN_USER_FEATURES | NETIF_F_HW_VLAN_CTAG_TX |
2652 NETIF_F_HW_VLAN_STAG_TX;
2653 dev->features = dev->hw_features | NETIF_F_LLTX;
2654 dev->vlan_features = dev->features &
2655 ~(NETIF_F_HW_VLAN_CTAG_TX |
2656 NETIF_F_HW_VLAN_STAG_TX);
2658 INIT_LIST_HEAD(&tun->disabled);
2659 err = tun_attach(tun, file, false, ifr->ifr_flags & IFF_NAPI);
2663 err = register_netdevice(tun->dev);
2668 netif_carrier_on(tun->dev);
2670 tun_debug(KERN_INFO, tun, "tun_set_iff\n");
2672 tun->flags = (tun->flags & ~TUN_FEATURES) |
2673 (ifr->ifr_flags & TUN_FEATURES);
2675 /* Make sure persistent devices do not get stuck in
2678 if (netif_running(tun->dev))
2679 netif_tx_wake_all_queues(tun->dev);
2681 strcpy(ifr->ifr_name, tun->dev->name);
2685 tun_detach_all(dev);
2686 /* register_netdevice() already called tun_free_netdev() */
2690 tun_flow_uninit(tun);
2691 security_tun_dev_free_security(tun->security);
2693 free_percpu(tun->pcpu_stats);
2699 static void tun_get_iff(struct net *net, struct tun_struct *tun,
2702 tun_debug(KERN_INFO, tun, "tun_get_iff\n");
2704 strcpy(ifr->ifr_name, tun->dev->name);
2706 ifr->ifr_flags = tun_flags(tun);
2710 /* This is like a cut-down ethtool ops, except done via tun fd so no
2711 * privs required. */
2712 static int set_offload(struct tun_struct *tun, unsigned long arg)
2714 netdev_features_t features = 0;
2716 if (arg & TUN_F_CSUM) {
2717 features |= NETIF_F_HW_CSUM;
2720 if (arg & (TUN_F_TSO4|TUN_F_TSO6)) {
2721 if (arg & TUN_F_TSO_ECN) {
2722 features |= NETIF_F_TSO_ECN;
2723 arg &= ~TUN_F_TSO_ECN;
2725 if (arg & TUN_F_TSO4)
2726 features |= NETIF_F_TSO;
2727 if (arg & TUN_F_TSO6)
2728 features |= NETIF_F_TSO6;
2729 arg &= ~(TUN_F_TSO4|TUN_F_TSO6);
2735 /* This gives the user a way to test for new features in future by
2736 * trying to set them. */
2740 tun->set_features = features;
2741 tun->dev->wanted_features &= ~TUN_USER_FEATURES;
2742 tun->dev->wanted_features |= features;
2743 netdev_update_features(tun->dev);
2748 static void tun_detach_filter(struct tun_struct *tun, int n)
2751 struct tun_file *tfile;
2753 for (i = 0; i < n; i++) {
2754 tfile = rtnl_dereference(tun->tfiles[i]);
2755 lock_sock(tfile->socket.sk);
2756 sk_detach_filter(tfile->socket.sk);
2757 release_sock(tfile->socket.sk);
2760 tun->filter_attached = false;
2763 static int tun_attach_filter(struct tun_struct *tun)
2766 struct tun_file *tfile;
2768 for (i = 0; i < tun->numqueues; i++) {
2769 tfile = rtnl_dereference(tun->tfiles[i]);
2770 lock_sock(tfile->socket.sk);
2771 ret = sk_attach_filter(&tun->fprog, tfile->socket.sk);
2772 release_sock(tfile->socket.sk);
2774 tun_detach_filter(tun, i);
2779 tun->filter_attached = true;
2783 static void tun_set_sndbuf(struct tun_struct *tun)
2785 struct tun_file *tfile;
2788 for (i = 0; i < tun->numqueues; i++) {
2789 tfile = rtnl_dereference(tun->tfiles[i]);
2790 tfile->socket.sk->sk_sndbuf = tun->sndbuf;
2794 static int tun_set_queue(struct file *file, struct ifreq *ifr)
2796 struct tun_file *tfile = file->private_data;
2797 struct tun_struct *tun;
2802 if (ifr->ifr_flags & IFF_ATTACH_QUEUE) {
2803 tun = tfile->detached;
2808 ret = security_tun_dev_attach_queue(tun->security);
2811 ret = tun_attach(tun, file, false, tun->flags & IFF_NAPI);
2812 } else if (ifr->ifr_flags & IFF_DETACH_QUEUE) {
2813 tun = rtnl_dereference(tfile->tun);
2814 if (!tun || !(tun->flags & IFF_MULTI_QUEUE) || tfile->detached)
2817 __tun_detach(tfile, false);
2826 static int tun_set_ebpf(struct tun_struct *tun, struct tun_prog **prog_p,
2829 struct bpf_prog *prog;
2832 if (copy_from_user(&fd, data, sizeof(fd)))
2838 prog = bpf_prog_get_type(fd, BPF_PROG_TYPE_SOCKET_FILTER);
2840 return PTR_ERR(prog);
2843 return __tun_set_ebpf(tun, prog_p, prog);
2846 static long __tun_chr_ioctl(struct file *file, unsigned int cmd,
2847 unsigned long arg, int ifreq_len)
2849 struct tun_file *tfile = file->private_data;
2850 struct tun_struct *tun;
2851 void __user* argp = (void __user*)arg;
2858 unsigned int ifindex;
2862 if (cmd == TUNSETIFF || cmd == TUNSETQUEUE ||
2863 (_IOC_TYPE(cmd) == SOCK_IOC_TYPE && cmd != SIOCGSKNS)) {
2864 if (copy_from_user(&ifr, argp, ifreq_len))
2867 memset(&ifr, 0, sizeof(ifr));
2869 if (cmd == TUNGETFEATURES) {
2870 /* Currently this just means: "what IFF flags are valid?".
2871 * This is needed because we never checked for invalid flags on
2874 return put_user(IFF_TUN | IFF_TAP | TUN_FEATURES,
2875 (unsigned int __user*)argp);
2876 } else if (cmd == TUNSETQUEUE)
2877 return tun_set_queue(file, &ifr);
2882 tun = tun_get(tfile);
2883 net = sock_net(&tfile->sk);
2884 if (cmd == TUNSETIFF) {
2889 ifr.ifr_name[IFNAMSIZ-1] = '\0';
2891 ret = tun_set_iff(net, file, &ifr);
2896 if (copy_to_user(argp, &ifr, ifreq_len))
2900 if (cmd == TUNSETIFINDEX) {
2906 if (copy_from_user(&ifindex, argp, sizeof(ifindex)))
2910 tfile->ifindex = ifindex;
2913 if (cmd == SIOCGSKNS) {
2915 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
2918 ret = open_related_ns(&net->ns, get_net_ns);
2926 tun_debug(KERN_INFO, tun, "tun_chr_ioctl cmd %u\n", cmd);
2931 tun_get_iff(current->nsproxy->net_ns, tun, &ifr);
2933 if (tfile->detached)
2934 ifr.ifr_flags |= IFF_DETACH_QUEUE;
2935 if (!tfile->socket.sk->sk_filter)
2936 ifr.ifr_flags |= IFF_NOFILTER;
2938 if (copy_to_user(argp, &ifr, ifreq_len))
2943 /* Disable/Enable checksum */
2945 /* [unimplemented] */
2946 tun_debug(KERN_INFO, tun, "ignored: set checksum %s\n",
2947 arg ? "disabled" : "enabled");
2951 /* Disable/Enable persist mode. Keep an extra reference to the
2952 * module to prevent the module being unprobed.
2954 if (arg && !(tun->flags & IFF_PERSIST)) {
2955 tun->flags |= IFF_PERSIST;
2956 __module_get(THIS_MODULE);
2958 if (!arg && (tun->flags & IFF_PERSIST)) {
2959 tun->flags &= ~IFF_PERSIST;
2960 module_put(THIS_MODULE);
2963 tun_debug(KERN_INFO, tun, "persist %s\n",
2964 arg ? "enabled" : "disabled");
2968 /* Set owner of the device */
2969 owner = make_kuid(current_user_ns(), arg);
2970 if (!uid_valid(owner)) {
2975 tun_debug(KERN_INFO, tun, "owner set to %u\n",
2976 from_kuid(&init_user_ns, tun->owner));
2980 /* Set group of the device */
2981 group = make_kgid(current_user_ns(), arg);
2982 if (!gid_valid(group)) {
2987 tun_debug(KERN_INFO, tun, "group set to %u\n",
2988 from_kgid(&init_user_ns, tun->group));
2992 /* Only allow setting the type when the interface is down */
2993 if (tun->dev->flags & IFF_UP) {
2994 tun_debug(KERN_INFO, tun,
2995 "Linktype set failed because interface is up\n");
2998 tun->dev->type = (int) arg;
2999 tun_debug(KERN_INFO, tun, "linktype set to %d\n",
3011 ret = set_offload(tun, arg);
3014 case TUNSETTXFILTER:
3015 /* Can be set only for TAPs */
3017 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3019 ret = update_filter(&tun->txflt, (void __user *)arg);
3023 /* Get hw address */
3024 memcpy(ifr.ifr_hwaddr.sa_data, tun->dev->dev_addr, ETH_ALEN);
3025 ifr.ifr_hwaddr.sa_family = tun->dev->type;
3026 if (copy_to_user(argp, &ifr, ifreq_len))
3031 /* Set hw address */
3032 tun_debug(KERN_DEBUG, tun, "set hw address: %pM\n",
3033 ifr.ifr_hwaddr.sa_data);
3035 ret = dev_set_mac_address(tun->dev, &ifr.ifr_hwaddr);
3039 sndbuf = tfile->socket.sk->sk_sndbuf;
3040 if (copy_to_user(argp, &sndbuf, sizeof(sndbuf)))
3045 if (copy_from_user(&sndbuf, argp, sizeof(sndbuf))) {
3054 tun->sndbuf = sndbuf;
3055 tun_set_sndbuf(tun);
3058 case TUNGETVNETHDRSZ:
3059 vnet_hdr_sz = tun->vnet_hdr_sz;
3060 if (copy_to_user(argp, &vnet_hdr_sz, sizeof(vnet_hdr_sz)))
3064 case TUNSETVNETHDRSZ:
3065 if (copy_from_user(&vnet_hdr_sz, argp, sizeof(vnet_hdr_sz))) {
3069 if (vnet_hdr_sz < (int)sizeof(struct virtio_net_hdr)) {
3074 tun->vnet_hdr_sz = vnet_hdr_sz;
3078 le = !!(tun->flags & TUN_VNET_LE);
3079 if (put_user(le, (int __user *)argp))
3084 if (get_user(le, (int __user *)argp)) {
3089 tun->flags |= TUN_VNET_LE;
3091 tun->flags &= ~TUN_VNET_LE;
3095 ret = tun_get_vnet_be(tun, argp);
3099 ret = tun_set_vnet_be(tun, argp);
3102 case TUNATTACHFILTER:
3103 /* Can be set only for TAPs */
3105 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3108 if (copy_from_user(&tun->fprog, argp, sizeof(tun->fprog)))
3111 ret = tun_attach_filter(tun);
3114 case TUNDETACHFILTER:
3115 /* Can be set only for TAPs */
3117 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3120 tun_detach_filter(tun, tun->numqueues);
3125 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3128 if (copy_to_user(argp, &tun->fprog, sizeof(tun->fprog)))
3133 case TUNSETSTEERINGEBPF:
3134 ret = tun_set_ebpf(tun, &tun->steering_prog, argp);
3137 case TUNSETFILTEREBPF:
3138 ret = tun_set_ebpf(tun, &tun->filter_prog, argp);
3153 static long tun_chr_ioctl(struct file *file,
3154 unsigned int cmd, unsigned long arg)
3156 return __tun_chr_ioctl(file, cmd, arg, sizeof (struct ifreq));
3159 #ifdef CONFIG_COMPAT
3160 static long tun_chr_compat_ioctl(struct file *file,
3161 unsigned int cmd, unsigned long arg)
3166 case TUNSETTXFILTER:
3171 arg = (unsigned long)compat_ptr(arg);
3174 arg = (compat_ulong_t)arg;
3179 * compat_ifreq is shorter than ifreq, so we must not access beyond
3180 * the end of that structure. All fields that are used in this
3181 * driver are compatible though, we don't need to convert the
3184 return __tun_chr_ioctl(file, cmd, arg, sizeof(struct compat_ifreq));
3186 #endif /* CONFIG_COMPAT */
3188 static int tun_chr_fasync(int fd, struct file *file, int on)
3190 struct tun_file *tfile = file->private_data;
3193 if ((ret = fasync_helper(fd, file, on, &tfile->fasync)) < 0)
3197 __f_setown(file, task_pid(current), PIDTYPE_PID, 0);
3198 tfile->flags |= TUN_FASYNC;
3200 tfile->flags &= ~TUN_FASYNC;
3206 static int tun_chr_open(struct inode *inode, struct file * file)
3208 struct net *net = current->nsproxy->net_ns;
3209 struct tun_file *tfile;
3211 DBG1(KERN_INFO, "tunX: tun_chr_open\n");
3213 tfile = (struct tun_file *)sk_alloc(net, AF_UNSPEC, GFP_KERNEL,
3217 RCU_INIT_POINTER(tfile->tun, NULL);
3221 init_waitqueue_head(&tfile->wq.wait);
3222 RCU_INIT_POINTER(tfile->socket.wq, &tfile->wq);
3224 tfile->socket.file = file;
3225 tfile->socket.ops = &tun_socket_ops;
3227 sock_init_data(&tfile->socket, &tfile->sk);
3229 tfile->sk.sk_write_space = tun_sock_write_space;
3230 tfile->sk.sk_sndbuf = INT_MAX;
3232 file->private_data = tfile;
3233 INIT_LIST_HEAD(&tfile->next);
3235 sock_set_flag(&tfile->sk, SOCK_ZEROCOPY);
3237 memset(&tfile->tx_ring, 0, sizeof(tfile->tx_ring));
3238 tfile->xdp_pending_pkts = 0;
3243 static int tun_chr_close(struct inode *inode, struct file *file)
3245 struct tun_file *tfile = file->private_data;
3247 tun_detach(tfile, true);
3252 #ifdef CONFIG_PROC_FS
3253 static void tun_chr_show_fdinfo(struct seq_file *m, struct file *file)
3255 struct tun_file *tfile = file->private_data;
3256 struct tun_struct *tun;
3259 memset(&ifr, 0, sizeof(ifr));
3262 tun = tun_get(tfile);
3264 tun_get_iff(current->nsproxy->net_ns, tun, &ifr);
3270 seq_printf(m, "iff:\t%s\n", ifr.ifr_name);
3274 static const struct file_operations tun_fops = {
3275 .owner = THIS_MODULE,
3276 .llseek = no_llseek,
3277 .read_iter = tun_chr_read_iter,
3278 .write_iter = tun_chr_write_iter,
3279 .poll = tun_chr_poll,
3280 .unlocked_ioctl = tun_chr_ioctl,
3281 #ifdef CONFIG_COMPAT
3282 .compat_ioctl = tun_chr_compat_ioctl,
3284 .open = tun_chr_open,
3285 .release = tun_chr_close,
3286 .fasync = tun_chr_fasync,
3287 #ifdef CONFIG_PROC_FS
3288 .show_fdinfo = tun_chr_show_fdinfo,
3292 static struct miscdevice tun_miscdev = {
3295 .nodename = "net/tun",
3299 /* ethtool interface */
3301 static int tun_get_link_ksettings(struct net_device *dev,
3302 struct ethtool_link_ksettings *cmd)
3304 ethtool_link_ksettings_zero_link_mode(cmd, supported);
3305 ethtool_link_ksettings_zero_link_mode(cmd, advertising);
3306 cmd->base.speed = SPEED_10;
3307 cmd->base.duplex = DUPLEX_FULL;
3308 cmd->base.port = PORT_TP;
3309 cmd->base.phy_address = 0;
3310 cmd->base.autoneg = AUTONEG_DISABLE;
3314 static void tun_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
3316 struct tun_struct *tun = netdev_priv(dev);
3318 strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
3319 strlcpy(info->version, DRV_VERSION, sizeof(info->version));
3321 switch (tun->flags & TUN_TYPE_MASK) {
3323 strlcpy(info->bus_info, "tun", sizeof(info->bus_info));
3326 strlcpy(info->bus_info, "tap", sizeof(info->bus_info));
3331 static u32 tun_get_msglevel(struct net_device *dev)
3334 struct tun_struct *tun = netdev_priv(dev);
3341 static void tun_set_msglevel(struct net_device *dev, u32 value)
3344 struct tun_struct *tun = netdev_priv(dev);
3349 static int tun_get_coalesce(struct net_device *dev,
3350 struct ethtool_coalesce *ec)
3352 struct tun_struct *tun = netdev_priv(dev);
3354 ec->rx_max_coalesced_frames = tun->rx_batched;
3359 static int tun_set_coalesce(struct net_device *dev,
3360 struct ethtool_coalesce *ec)
3362 struct tun_struct *tun = netdev_priv(dev);
3364 if (ec->rx_max_coalesced_frames > NAPI_POLL_WEIGHT)
3365 tun->rx_batched = NAPI_POLL_WEIGHT;
3367 tun->rx_batched = ec->rx_max_coalesced_frames;
3372 static const struct ethtool_ops tun_ethtool_ops = {
3373 .get_drvinfo = tun_get_drvinfo,
3374 .get_msglevel = tun_get_msglevel,
3375 .set_msglevel = tun_set_msglevel,
3376 .get_link = ethtool_op_get_link,
3377 .get_ts_info = ethtool_op_get_ts_info,
3378 .get_coalesce = tun_get_coalesce,
3379 .set_coalesce = tun_set_coalesce,
3380 .get_link_ksettings = tun_get_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");