xref: /linux/drivers/net/tun.c (revision 2174181019e4273e583a0f0a9795e9db38984784)
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3  *  TUN - Universal TUN/TAP device driver.
4  *  Copyright (C) 1999-2002 Maxim Krasnyansky <maxk@qualcomm.com>
5  *
6  *  $Id: tun.c,v 1.15 2002/03/01 02:44:24 maxk Exp $
7  */
8 
9 /*
10  *  Changes:
11  *
12  *  Mike Kershaw <dragorn@kismetwireless.net> 2005/08/14
13  *    Add TUNSETLINK ioctl to set the link encapsulation
14  *
15  *  Mark Smith <markzzzsmith@yahoo.com.au>
16  *    Use eth_random_addr() for tap MAC address.
17  *
18  *  Harald Roelle <harald.roelle@ifi.lmu.de>  2004/04/20
19  *    Fixes in packet dropping, queue length setting and queue wakeup.
20  *    Increased default tx queue length.
21  *    Added ethtool API.
22  *    Minor cleanups
23  *
24  *  Daniel Podlejski <underley@underley.eu.org>
25  *    Modifications for 2.3.99-pre5 kernel.
26  */
27 
28 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
29 
30 #define DRV_NAME	"tun"
31 #define DRV_VERSION	"1.6"
32 #define DRV_DESCRIPTION	"Universal TUN/TAP device driver"
33 #define DRV_COPYRIGHT	"(C) 1999-2004 Max Krasnyansky <maxk@qualcomm.com>"
34 
35 #include <linux/module.h>
36 #include <linux/errno.h>
37 #include <linux/kernel.h>
38 #include <linux/sched/signal.h>
39 #include <linux/major.h>
40 #include <linux/slab.h>
41 #include <linux/poll.h>
42 #include <linux/fcntl.h>
43 #include <linux/init.h>
44 #include <linux/skbuff.h>
45 #include <linux/netdevice.h>
46 #include <linux/etherdevice.h>
47 #include <linux/miscdevice.h>
48 #include <linux/ethtool.h>
49 #include <linux/rtnetlink.h>
50 #include <linux/compat.h>
51 #include <linux/if.h>
52 #include <linux/if_arp.h>
53 #include <linux/if_ether.h>
54 #include <linux/if_tun.h>
55 #include <linux/if_vlan.h>
56 #include <linux/crc32.h>
57 #include <linux/nsproxy.h>
58 #include <linux/virtio_net.h>
59 #include <linux/rcupdate.h>
60 #include <net/net_namespace.h>
61 #include <net/netns/generic.h>
62 #include <net/rtnetlink.h>
63 #include <net/sock.h>
64 #include <net/xdp.h>
65 #include <net/ip_tunnels.h>
66 #include <linux/seq_file.h>
67 #include <linux/uio.h>
68 #include <linux/skb_array.h>
69 #include <linux/bpf.h>
70 #include <linux/bpf_trace.h>
71 #include <linux/mutex.h>
72 #include <linux/ieee802154.h>
73 #include <linux/if_ltalk.h>
74 #include <uapi/linux/if_fddi.h>
75 #include <uapi/linux/if_hippi.h>
76 #include <uapi/linux/if_fc.h>
77 #include <net/ax25.h>
78 #include <net/rose.h>
79 #include <net/6lowpan.h>
80 
81 #include <linux/uaccess.h>
82 #include <linux/proc_fs.h>
83 
84 static void tun_default_link_ksettings(struct net_device *dev,
85 				       struct ethtool_link_ksettings *cmd);
86 
87 #define TUN_RX_PAD (NET_IP_ALIGN + NET_SKB_PAD)
88 
89 /* TUN device flags */
90 
91 /* IFF_ATTACH_QUEUE is never stored in device flags,
92  * overload it to mean fasync when stored there.
93  */
94 #define TUN_FASYNC	IFF_ATTACH_QUEUE
95 /* High bits in flags field are unused. */
96 #define TUN_VNET_LE     0x80000000
97 #define TUN_VNET_BE     0x40000000
98 
99 #define TUN_FEATURES (IFF_NO_PI | IFF_ONE_QUEUE | IFF_VNET_HDR | \
100 		      IFF_MULTI_QUEUE | IFF_NAPI | IFF_NAPI_FRAGS)
101 
102 #define GOODCOPY_LEN 128
103 
104 #define FLT_EXACT_COUNT 8
105 struct tap_filter {
106 	unsigned int    count;    /* Number of addrs. Zero means disabled */
107 	u32             mask[2];  /* Mask of the hashed addrs */
108 	unsigned char	addr[FLT_EXACT_COUNT][ETH_ALEN];
109 };
110 
111 /* MAX_TAP_QUEUES 256 is chosen to allow rx/tx queues to be equal
112  * to max number of VCPUs in guest. */
113 #define MAX_TAP_QUEUES 256
114 #define MAX_TAP_FLOWS  4096
115 
116 #define TUN_FLOW_EXPIRE (3 * HZ)
117 
118 /* A tun_file connects an open character device to a tuntap netdevice. It
119  * also contains all socket related structures (except sock_fprog and tap_filter)
120  * to serve as one transmit queue for tuntap device. The sock_fprog and
121  * tap_filter were kept in tun_struct since they were used for filtering for the
122  * netdevice not for a specific queue (at least I didn't see the requirement for
123  * this).
124  *
125  * RCU usage:
126  * The tun_file and tun_struct are loosely coupled, the pointer from one to the
127  * other can only be read while rcu_read_lock or rtnl_lock is held.
128  */
129 struct tun_file {
130 	struct sock sk;
131 	struct socket socket;
132 	struct tun_struct __rcu *tun;
133 	struct fasync_struct *fasync;
134 	/* only used for fasnyc */
135 	unsigned int flags;
136 	union {
137 		u16 queue_index;
138 		unsigned int ifindex;
139 	};
140 	struct napi_struct napi;
141 	bool napi_enabled;
142 	bool napi_frags_enabled;
143 	struct mutex napi_mutex;	/* Protects access to the above napi */
144 	struct list_head next;
145 	struct tun_struct *detached;
146 	struct ptr_ring tx_ring;
147 	struct xdp_rxq_info xdp_rxq;
148 };
149 
150 struct tun_page {
151 	struct page *page;
152 	int count;
153 };
154 
155 struct tun_flow_entry {
156 	struct hlist_node hash_link;
157 	struct rcu_head rcu;
158 	struct tun_struct *tun;
159 
160 	u32 rxhash;
161 	u32 rps_rxhash;
162 	int queue_index;
163 	unsigned long updated ____cacheline_aligned_in_smp;
164 };
165 
166 #define TUN_NUM_FLOW_ENTRIES 1024
167 #define TUN_MASK_FLOW_ENTRIES (TUN_NUM_FLOW_ENTRIES - 1)
168 
169 struct tun_prog {
170 	struct rcu_head rcu;
171 	struct bpf_prog *prog;
172 };
173 
174 /* Since the socket were moved to tun_file, to preserve the behavior of persist
175  * device, socket filter, sndbuf and vnet header size were restore when the
176  * file were attached to a persist device.
177  */
178 struct tun_struct {
179 	struct tun_file __rcu	*tfiles[MAX_TAP_QUEUES];
180 	unsigned int            numqueues;
181 	unsigned int 		flags;
182 	kuid_t			owner;
183 	kgid_t			group;
184 
185 	struct net_device	*dev;
186 	netdev_features_t	set_features;
187 #define TUN_USER_FEATURES (NETIF_F_HW_CSUM|NETIF_F_TSO_ECN|NETIF_F_TSO| \
188 			  NETIF_F_TSO6 | NETIF_F_GSO_UDP_L4)
189 
190 	int			align;
191 	int			vnet_hdr_sz;
192 	int			sndbuf;
193 	struct tap_filter	txflt;
194 	struct sock_fprog	fprog;
195 	/* protected by rtnl lock */
196 	bool			filter_attached;
197 	u32			msg_enable;
198 	spinlock_t lock;
199 	struct hlist_head flows[TUN_NUM_FLOW_ENTRIES];
200 	struct timer_list flow_gc_timer;
201 	unsigned long ageing_time;
202 	unsigned int numdisabled;
203 	struct list_head disabled;
204 	void *security;
205 	u32 flow_count;
206 	u32 rx_batched;
207 	atomic_long_t rx_frame_errors;
208 	struct bpf_prog __rcu *xdp_prog;
209 	struct tun_prog __rcu *steering_prog;
210 	struct tun_prog __rcu *filter_prog;
211 	struct ethtool_link_ksettings link_ksettings;
212 	/* init args */
213 	struct file *file;
214 	struct ifreq *ifr;
215 };
216 
217 struct veth {
218 	__be16 h_vlan_proto;
219 	__be16 h_vlan_TCI;
220 };
221 
222 static void tun_flow_init(struct tun_struct *tun);
223 static void tun_flow_uninit(struct tun_struct *tun);
224 
225 static int tun_napi_receive(struct napi_struct *napi, int budget)
226 {
227 	struct tun_file *tfile = container_of(napi, struct tun_file, napi);
228 	struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
229 	struct sk_buff_head process_queue;
230 	struct sk_buff *skb;
231 	int received = 0;
232 
233 	__skb_queue_head_init(&process_queue);
234 
235 	spin_lock(&queue->lock);
236 	skb_queue_splice_tail_init(queue, &process_queue);
237 	spin_unlock(&queue->lock);
238 
239 	while (received < budget && (skb = __skb_dequeue(&process_queue))) {
240 		napi_gro_receive(napi, skb);
241 		++received;
242 	}
243 
244 	if (!skb_queue_empty(&process_queue)) {
245 		spin_lock(&queue->lock);
246 		skb_queue_splice(&process_queue, queue);
247 		spin_unlock(&queue->lock);
248 	}
249 
250 	return received;
251 }
252 
253 static int tun_napi_poll(struct napi_struct *napi, int budget)
254 {
255 	unsigned int received;
256 
257 	received = tun_napi_receive(napi, budget);
258 
259 	if (received < budget)
260 		napi_complete_done(napi, received);
261 
262 	return received;
263 }
264 
265 static void tun_napi_init(struct tun_struct *tun, struct tun_file *tfile,
266 			  bool napi_en, bool napi_frags)
267 {
268 	tfile->napi_enabled = napi_en;
269 	tfile->napi_frags_enabled = napi_en && napi_frags;
270 	if (napi_en) {
271 		netif_napi_add_tx(tun->dev, &tfile->napi, tun_napi_poll);
272 		napi_enable(&tfile->napi);
273 	}
274 }
275 
276 static void tun_napi_enable(struct tun_file *tfile)
277 {
278 	if (tfile->napi_enabled)
279 		napi_enable(&tfile->napi);
280 }
281 
282 static void tun_napi_disable(struct tun_file *tfile)
283 {
284 	if (tfile->napi_enabled)
285 		napi_disable(&tfile->napi);
286 }
287 
288 static void tun_napi_del(struct tun_file *tfile)
289 {
290 	if (tfile->napi_enabled)
291 		netif_napi_del(&tfile->napi);
292 }
293 
294 static bool tun_napi_frags_enabled(const struct tun_file *tfile)
295 {
296 	return tfile->napi_frags_enabled;
297 }
298 
299 #ifdef CONFIG_TUN_VNET_CROSS_LE
300 static inline bool tun_legacy_is_little_endian(struct tun_struct *tun)
301 {
302 	return tun->flags & TUN_VNET_BE ? false :
303 		virtio_legacy_is_little_endian();
304 }
305 
306 static long tun_get_vnet_be(struct tun_struct *tun, int __user *argp)
307 {
308 	int be = !!(tun->flags & TUN_VNET_BE);
309 
310 	if (put_user(be, argp))
311 		return -EFAULT;
312 
313 	return 0;
314 }
315 
316 static long tun_set_vnet_be(struct tun_struct *tun, int __user *argp)
317 {
318 	int be;
319 
320 	if (get_user(be, argp))
321 		return -EFAULT;
322 
323 	if (be)
324 		tun->flags |= TUN_VNET_BE;
325 	else
326 		tun->flags &= ~TUN_VNET_BE;
327 
328 	return 0;
329 }
330 #else
331 static inline bool tun_legacy_is_little_endian(struct tun_struct *tun)
332 {
333 	return virtio_legacy_is_little_endian();
334 }
335 
336 static long tun_get_vnet_be(struct tun_struct *tun, int __user *argp)
337 {
338 	return -EINVAL;
339 }
340 
341 static long tun_set_vnet_be(struct tun_struct *tun, int __user *argp)
342 {
343 	return -EINVAL;
344 }
345 #endif /* CONFIG_TUN_VNET_CROSS_LE */
346 
347 static inline bool tun_is_little_endian(struct tun_struct *tun)
348 {
349 	return tun->flags & TUN_VNET_LE ||
350 		tun_legacy_is_little_endian(tun);
351 }
352 
353 static inline u16 tun16_to_cpu(struct tun_struct *tun, __virtio16 val)
354 {
355 	return __virtio16_to_cpu(tun_is_little_endian(tun), val);
356 }
357 
358 static inline __virtio16 cpu_to_tun16(struct tun_struct *tun, u16 val)
359 {
360 	return __cpu_to_virtio16(tun_is_little_endian(tun), val);
361 }
362 
363 static inline u32 tun_hashfn(u32 rxhash)
364 {
365 	return rxhash & TUN_MASK_FLOW_ENTRIES;
366 }
367 
368 static struct tun_flow_entry *tun_flow_find(struct hlist_head *head, u32 rxhash)
369 {
370 	struct tun_flow_entry *e;
371 
372 	hlist_for_each_entry_rcu(e, head, hash_link) {
373 		if (e->rxhash == rxhash)
374 			return e;
375 	}
376 	return NULL;
377 }
378 
379 static struct tun_flow_entry *tun_flow_create(struct tun_struct *tun,
380 					      struct hlist_head *head,
381 					      u32 rxhash, u16 queue_index)
382 {
383 	struct tun_flow_entry *e = kmalloc(sizeof(*e), GFP_ATOMIC);
384 
385 	if (e) {
386 		netif_info(tun, tx_queued, tun->dev,
387 			   "create flow: hash %u index %u\n",
388 			   rxhash, queue_index);
389 		e->updated = jiffies;
390 		e->rxhash = rxhash;
391 		e->rps_rxhash = 0;
392 		e->queue_index = queue_index;
393 		e->tun = tun;
394 		hlist_add_head_rcu(&e->hash_link, head);
395 		++tun->flow_count;
396 	}
397 	return e;
398 }
399 
400 static void tun_flow_delete(struct tun_struct *tun, struct tun_flow_entry *e)
401 {
402 	netif_info(tun, tx_queued, tun->dev, "delete flow: hash %u index %u\n",
403 		   e->rxhash, e->queue_index);
404 	hlist_del_rcu(&e->hash_link);
405 	kfree_rcu(e, rcu);
406 	--tun->flow_count;
407 }
408 
409 static void tun_flow_flush(struct tun_struct *tun)
410 {
411 	int i;
412 
413 	spin_lock_bh(&tun->lock);
414 	for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
415 		struct tun_flow_entry *e;
416 		struct hlist_node *n;
417 
418 		hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link)
419 			tun_flow_delete(tun, e);
420 	}
421 	spin_unlock_bh(&tun->lock);
422 }
423 
424 static void tun_flow_delete_by_queue(struct tun_struct *tun, u16 queue_index)
425 {
426 	int i;
427 
428 	spin_lock_bh(&tun->lock);
429 	for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
430 		struct tun_flow_entry *e;
431 		struct hlist_node *n;
432 
433 		hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
434 			if (e->queue_index == queue_index)
435 				tun_flow_delete(tun, e);
436 		}
437 	}
438 	spin_unlock_bh(&tun->lock);
439 }
440 
441 static void tun_flow_cleanup(struct timer_list *t)
442 {
443 	struct tun_struct *tun = from_timer(tun, t, flow_gc_timer);
444 	unsigned long delay = tun->ageing_time;
445 	unsigned long next_timer = jiffies + delay;
446 	unsigned long count = 0;
447 	int i;
448 
449 	spin_lock(&tun->lock);
450 	for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
451 		struct tun_flow_entry *e;
452 		struct hlist_node *n;
453 
454 		hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
455 			unsigned long this_timer;
456 
457 			this_timer = e->updated + delay;
458 			if (time_before_eq(this_timer, jiffies)) {
459 				tun_flow_delete(tun, e);
460 				continue;
461 			}
462 			count++;
463 			if (time_before(this_timer, next_timer))
464 				next_timer = this_timer;
465 		}
466 	}
467 
468 	if (count)
469 		mod_timer(&tun->flow_gc_timer, round_jiffies_up(next_timer));
470 	spin_unlock(&tun->lock);
471 }
472 
473 static void tun_flow_update(struct tun_struct *tun, u32 rxhash,
474 			    struct tun_file *tfile)
475 {
476 	struct hlist_head *head;
477 	struct tun_flow_entry *e;
478 	unsigned long delay = tun->ageing_time;
479 	u16 queue_index = tfile->queue_index;
480 
481 	head = &tun->flows[tun_hashfn(rxhash)];
482 
483 	rcu_read_lock();
484 
485 	e = tun_flow_find(head, rxhash);
486 	if (likely(e)) {
487 		/* TODO: keep queueing to old queue until it's empty? */
488 		if (READ_ONCE(e->queue_index) != queue_index)
489 			WRITE_ONCE(e->queue_index, queue_index);
490 		if (e->updated != jiffies)
491 			e->updated = jiffies;
492 		sock_rps_record_flow_hash(e->rps_rxhash);
493 	} else {
494 		spin_lock_bh(&tun->lock);
495 		if (!tun_flow_find(head, rxhash) &&
496 		    tun->flow_count < MAX_TAP_FLOWS)
497 			tun_flow_create(tun, head, rxhash, queue_index);
498 
499 		if (!timer_pending(&tun->flow_gc_timer))
500 			mod_timer(&tun->flow_gc_timer,
501 				  round_jiffies_up(jiffies + delay));
502 		spin_unlock_bh(&tun->lock);
503 	}
504 
505 	rcu_read_unlock();
506 }
507 
508 /* Save the hash received in the stack receive path and update the
509  * flow_hash table accordingly.
510  */
511 static inline void tun_flow_save_rps_rxhash(struct tun_flow_entry *e, u32 hash)
512 {
513 	if (unlikely(e->rps_rxhash != hash))
514 		e->rps_rxhash = hash;
515 }
516 
517 /* We try to identify a flow through its rxhash. The reason that
518  * we do not check rxq no. is because some cards(e.g 82599), chooses
519  * the rxq based on the txq where the last packet of the flow comes. As
520  * the userspace application move between processors, we may get a
521  * different rxq no. here.
522  */
523 static u16 tun_automq_select_queue(struct tun_struct *tun, struct sk_buff *skb)
524 {
525 	struct tun_flow_entry *e;
526 	u32 txq = 0;
527 	u32 numqueues = 0;
528 
529 	numqueues = READ_ONCE(tun->numqueues);
530 
531 	txq = __skb_get_hash_symmetric(skb);
532 	e = tun_flow_find(&tun->flows[tun_hashfn(txq)], txq);
533 	if (e) {
534 		tun_flow_save_rps_rxhash(e, txq);
535 		txq = e->queue_index;
536 	} else {
537 		/* use multiply and shift instead of expensive divide */
538 		txq = ((u64)txq * numqueues) >> 32;
539 	}
540 
541 	return txq;
542 }
543 
544 static u16 tun_ebpf_select_queue(struct tun_struct *tun, struct sk_buff *skb)
545 {
546 	struct tun_prog *prog;
547 	u32 numqueues;
548 	u16 ret = 0;
549 
550 	numqueues = READ_ONCE(tun->numqueues);
551 	if (!numqueues)
552 		return 0;
553 
554 	prog = rcu_dereference(tun->steering_prog);
555 	if (prog)
556 		ret = bpf_prog_run_clear_cb(prog->prog, skb);
557 
558 	return ret % numqueues;
559 }
560 
561 static u16 tun_select_queue(struct net_device *dev, struct sk_buff *skb,
562 			    struct net_device *sb_dev)
563 {
564 	struct tun_struct *tun = netdev_priv(dev);
565 	u16 ret;
566 
567 	rcu_read_lock();
568 	if (rcu_dereference(tun->steering_prog))
569 		ret = tun_ebpf_select_queue(tun, skb);
570 	else
571 		ret = tun_automq_select_queue(tun, skb);
572 	rcu_read_unlock();
573 
574 	return ret;
575 }
576 
577 static inline bool tun_not_capable(struct tun_struct *tun)
578 {
579 	const struct cred *cred = current_cred();
580 	struct net *net = dev_net(tun->dev);
581 
582 	return ((uid_valid(tun->owner) && !uid_eq(cred->euid, tun->owner)) ||
583 		  (gid_valid(tun->group) && !in_egroup_p(tun->group))) &&
584 		!ns_capable(net->user_ns, CAP_NET_ADMIN);
585 }
586 
587 static void tun_set_real_num_queues(struct tun_struct *tun)
588 {
589 	netif_set_real_num_tx_queues(tun->dev, tun->numqueues);
590 	netif_set_real_num_rx_queues(tun->dev, tun->numqueues);
591 }
592 
593 static void tun_disable_queue(struct tun_struct *tun, struct tun_file *tfile)
594 {
595 	tfile->detached = tun;
596 	list_add_tail(&tfile->next, &tun->disabled);
597 	++tun->numdisabled;
598 }
599 
600 static struct tun_struct *tun_enable_queue(struct tun_file *tfile)
601 {
602 	struct tun_struct *tun = tfile->detached;
603 
604 	tfile->detached = NULL;
605 	list_del_init(&tfile->next);
606 	--tun->numdisabled;
607 	return tun;
608 }
609 
610 void tun_ptr_free(void *ptr)
611 {
612 	if (!ptr)
613 		return;
614 	if (tun_is_xdp_frame(ptr)) {
615 		struct xdp_frame *xdpf = tun_ptr_to_xdp(ptr);
616 
617 		xdp_return_frame(xdpf);
618 	} else {
619 		__skb_array_destroy_skb(ptr);
620 	}
621 }
622 EXPORT_SYMBOL_GPL(tun_ptr_free);
623 
624 static void tun_queue_purge(struct tun_file *tfile)
625 {
626 	void *ptr;
627 
628 	while ((ptr = ptr_ring_consume(&tfile->tx_ring)) != NULL)
629 		tun_ptr_free(ptr);
630 
631 	skb_queue_purge(&tfile->sk.sk_write_queue);
632 	skb_queue_purge(&tfile->sk.sk_error_queue);
633 }
634 
635 static void __tun_detach(struct tun_file *tfile, bool clean)
636 {
637 	struct tun_file *ntfile;
638 	struct tun_struct *tun;
639 
640 	tun = rtnl_dereference(tfile->tun);
641 
642 	if (tun && clean) {
643 		if (!tfile->detached)
644 			tun_napi_disable(tfile);
645 		tun_napi_del(tfile);
646 	}
647 
648 	if (tun && !tfile->detached) {
649 		u16 index = tfile->queue_index;
650 		BUG_ON(index >= tun->numqueues);
651 
652 		rcu_assign_pointer(tun->tfiles[index],
653 				   tun->tfiles[tun->numqueues - 1]);
654 		ntfile = rtnl_dereference(tun->tfiles[index]);
655 		ntfile->queue_index = index;
656 		rcu_assign_pointer(tun->tfiles[tun->numqueues - 1],
657 				   NULL);
658 
659 		--tun->numqueues;
660 		if (clean) {
661 			RCU_INIT_POINTER(tfile->tun, NULL);
662 			sock_put(&tfile->sk);
663 		} else {
664 			tun_disable_queue(tun, tfile);
665 			tun_napi_disable(tfile);
666 		}
667 
668 		synchronize_net();
669 		tun_flow_delete_by_queue(tun, tun->numqueues + 1);
670 		/* Drop read queue */
671 		tun_queue_purge(tfile);
672 		tun_set_real_num_queues(tun);
673 	} else if (tfile->detached && clean) {
674 		tun = tun_enable_queue(tfile);
675 		sock_put(&tfile->sk);
676 	}
677 
678 	if (clean) {
679 		if (tun && tun->numqueues == 0 && tun->numdisabled == 0) {
680 			netif_carrier_off(tun->dev);
681 
682 			if (!(tun->flags & IFF_PERSIST) &&
683 			    tun->dev->reg_state == NETREG_REGISTERED)
684 				unregister_netdevice(tun->dev);
685 		}
686 		if (tun)
687 			xdp_rxq_info_unreg(&tfile->xdp_rxq);
688 		ptr_ring_cleanup(&tfile->tx_ring, tun_ptr_free);
689 	}
690 }
691 
692 static void tun_detach(struct tun_file *tfile, bool clean)
693 {
694 	struct tun_struct *tun;
695 	struct net_device *dev;
696 
697 	rtnl_lock();
698 	tun = rtnl_dereference(tfile->tun);
699 	dev = tun ? tun->dev : NULL;
700 	__tun_detach(tfile, clean);
701 	if (dev)
702 		netdev_state_change(dev);
703 	rtnl_unlock();
704 
705 	if (clean)
706 		sock_put(&tfile->sk);
707 }
708 
709 static void tun_detach_all(struct net_device *dev)
710 {
711 	struct tun_struct *tun = netdev_priv(dev);
712 	struct tun_file *tfile, *tmp;
713 	int i, n = tun->numqueues;
714 
715 	for (i = 0; i < n; i++) {
716 		tfile = rtnl_dereference(tun->tfiles[i]);
717 		BUG_ON(!tfile);
718 		tun_napi_disable(tfile);
719 		tfile->socket.sk->sk_shutdown = RCV_SHUTDOWN;
720 		tfile->socket.sk->sk_data_ready(tfile->socket.sk);
721 		RCU_INIT_POINTER(tfile->tun, NULL);
722 		--tun->numqueues;
723 	}
724 	list_for_each_entry(tfile, &tun->disabled, next) {
725 		tfile->socket.sk->sk_shutdown = RCV_SHUTDOWN;
726 		tfile->socket.sk->sk_data_ready(tfile->socket.sk);
727 		RCU_INIT_POINTER(tfile->tun, NULL);
728 	}
729 	BUG_ON(tun->numqueues != 0);
730 
731 	synchronize_net();
732 	for (i = 0; i < n; i++) {
733 		tfile = rtnl_dereference(tun->tfiles[i]);
734 		tun_napi_del(tfile);
735 		/* Drop read queue */
736 		tun_queue_purge(tfile);
737 		xdp_rxq_info_unreg(&tfile->xdp_rxq);
738 		sock_put(&tfile->sk);
739 	}
740 	list_for_each_entry_safe(tfile, tmp, &tun->disabled, next) {
741 		tun_napi_del(tfile);
742 		tun_enable_queue(tfile);
743 		tun_queue_purge(tfile);
744 		xdp_rxq_info_unreg(&tfile->xdp_rxq);
745 		sock_put(&tfile->sk);
746 	}
747 	BUG_ON(tun->numdisabled != 0);
748 
749 	if (tun->flags & IFF_PERSIST)
750 		module_put(THIS_MODULE);
751 }
752 
753 static int tun_attach(struct tun_struct *tun, struct file *file,
754 		      bool skip_filter, bool napi, bool napi_frags,
755 		      bool publish_tun)
756 {
757 	struct tun_file *tfile = file->private_data;
758 	struct net_device *dev = tun->dev;
759 	int err;
760 
761 	err = security_tun_dev_attach(tfile->socket.sk, tun->security);
762 	if (err < 0)
763 		goto out;
764 
765 	err = -EINVAL;
766 	if (rtnl_dereference(tfile->tun) && !tfile->detached)
767 		goto out;
768 
769 	err = -EBUSY;
770 	if (!(tun->flags & IFF_MULTI_QUEUE) && tun->numqueues == 1)
771 		goto out;
772 
773 	err = -E2BIG;
774 	if (!tfile->detached &&
775 	    tun->numqueues + tun->numdisabled == MAX_TAP_QUEUES)
776 		goto out;
777 
778 	err = 0;
779 
780 	/* Re-attach the filter to persist device */
781 	if (!skip_filter && (tun->filter_attached == true)) {
782 		lock_sock(tfile->socket.sk);
783 		err = sk_attach_filter(&tun->fprog, tfile->socket.sk);
784 		release_sock(tfile->socket.sk);
785 		if (!err)
786 			goto out;
787 	}
788 
789 	if (!tfile->detached &&
790 	    ptr_ring_resize(&tfile->tx_ring, dev->tx_queue_len,
791 			    GFP_KERNEL, tun_ptr_free)) {
792 		err = -ENOMEM;
793 		goto out;
794 	}
795 
796 	tfile->queue_index = tun->numqueues;
797 	tfile->socket.sk->sk_shutdown &= ~RCV_SHUTDOWN;
798 
799 	if (tfile->detached) {
800 		/* Re-attach detached tfile, updating XDP queue_index */
801 		WARN_ON(!xdp_rxq_info_is_reg(&tfile->xdp_rxq));
802 
803 		if (tfile->xdp_rxq.queue_index    != tfile->queue_index)
804 			tfile->xdp_rxq.queue_index = tfile->queue_index;
805 	} else {
806 		/* Setup XDP RX-queue info, for new tfile getting attached */
807 		err = xdp_rxq_info_reg(&tfile->xdp_rxq,
808 				       tun->dev, tfile->queue_index, 0);
809 		if (err < 0)
810 			goto out;
811 		err = xdp_rxq_info_reg_mem_model(&tfile->xdp_rxq,
812 						 MEM_TYPE_PAGE_SHARED, NULL);
813 		if (err < 0) {
814 			xdp_rxq_info_unreg(&tfile->xdp_rxq);
815 			goto out;
816 		}
817 		err = 0;
818 	}
819 
820 	if (tfile->detached) {
821 		tun_enable_queue(tfile);
822 		tun_napi_enable(tfile);
823 	} else {
824 		sock_hold(&tfile->sk);
825 		tun_napi_init(tun, tfile, napi, napi_frags);
826 	}
827 
828 	if (rtnl_dereference(tun->xdp_prog))
829 		sock_set_flag(&tfile->sk, SOCK_XDP);
830 
831 	/* device is allowed to go away first, so no need to hold extra
832 	 * refcnt.
833 	 */
834 
835 	/* Publish tfile->tun and tun->tfiles only after we've fully
836 	 * initialized tfile; otherwise we risk using half-initialized
837 	 * object.
838 	 */
839 	if (publish_tun)
840 		rcu_assign_pointer(tfile->tun, tun);
841 	rcu_assign_pointer(tun->tfiles[tun->numqueues], tfile);
842 	tun->numqueues++;
843 	tun_set_real_num_queues(tun);
844 out:
845 	return err;
846 }
847 
848 static struct tun_struct *tun_get(struct tun_file *tfile)
849 {
850 	struct tun_struct *tun;
851 
852 	rcu_read_lock();
853 	tun = rcu_dereference(tfile->tun);
854 	if (tun)
855 		dev_hold(tun->dev);
856 	rcu_read_unlock();
857 
858 	return tun;
859 }
860 
861 static void tun_put(struct tun_struct *tun)
862 {
863 	dev_put(tun->dev);
864 }
865 
866 /* TAP filtering */
867 static void addr_hash_set(u32 *mask, const u8 *addr)
868 {
869 	int n = ether_crc(ETH_ALEN, addr) >> 26;
870 	mask[n >> 5] |= (1 << (n & 31));
871 }
872 
873 static unsigned int addr_hash_test(const u32 *mask, const u8 *addr)
874 {
875 	int n = ether_crc(ETH_ALEN, addr) >> 26;
876 	return mask[n >> 5] & (1 << (n & 31));
877 }
878 
879 static int update_filter(struct tap_filter *filter, void __user *arg)
880 {
881 	struct { u8 u[ETH_ALEN]; } *addr;
882 	struct tun_filter uf;
883 	int err, alen, n, nexact;
884 
885 	if (copy_from_user(&uf, arg, sizeof(uf)))
886 		return -EFAULT;
887 
888 	if (!uf.count) {
889 		/* Disabled */
890 		filter->count = 0;
891 		return 0;
892 	}
893 
894 	alen = ETH_ALEN * uf.count;
895 	addr = memdup_user(arg + sizeof(uf), alen);
896 	if (IS_ERR(addr))
897 		return PTR_ERR(addr);
898 
899 	/* The filter is updated without holding any locks. Which is
900 	 * perfectly safe. We disable it first and in the worst
901 	 * case we'll accept a few undesired packets. */
902 	filter->count = 0;
903 	wmb();
904 
905 	/* Use first set of addresses as an exact filter */
906 	for (n = 0; n < uf.count && n < FLT_EXACT_COUNT; n++)
907 		memcpy(filter->addr[n], addr[n].u, ETH_ALEN);
908 
909 	nexact = n;
910 
911 	/* Remaining multicast addresses are hashed,
912 	 * unicast will leave the filter disabled. */
913 	memset(filter->mask, 0, sizeof(filter->mask));
914 	for (; n < uf.count; n++) {
915 		if (!is_multicast_ether_addr(addr[n].u)) {
916 			err = 0; /* no filter */
917 			goto free_addr;
918 		}
919 		addr_hash_set(filter->mask, addr[n].u);
920 	}
921 
922 	/* For ALLMULTI just set the mask to all ones.
923 	 * This overrides the mask populated above. */
924 	if ((uf.flags & TUN_FLT_ALLMULTI))
925 		memset(filter->mask, ~0, sizeof(filter->mask));
926 
927 	/* Now enable the filter */
928 	wmb();
929 	filter->count = nexact;
930 
931 	/* Return the number of exact filters */
932 	err = nexact;
933 free_addr:
934 	kfree(addr);
935 	return err;
936 }
937 
938 /* Returns: 0 - drop, !=0 - accept */
939 static int run_filter(struct tap_filter *filter, const struct sk_buff *skb)
940 {
941 	/* Cannot use eth_hdr(skb) here because skb_mac_hdr() is incorrect
942 	 * at this point. */
943 	struct ethhdr *eh = (struct ethhdr *) skb->data;
944 	int i;
945 
946 	/* Exact match */
947 	for (i = 0; i < filter->count; i++)
948 		if (ether_addr_equal(eh->h_dest, filter->addr[i]))
949 			return 1;
950 
951 	/* Inexact match (multicast only) */
952 	if (is_multicast_ether_addr(eh->h_dest))
953 		return addr_hash_test(filter->mask, eh->h_dest);
954 
955 	return 0;
956 }
957 
958 /*
959  * Checks whether the packet is accepted or not.
960  * Returns: 0 - drop, !=0 - accept
961  */
962 static int check_filter(struct tap_filter *filter, const struct sk_buff *skb)
963 {
964 	if (!filter->count)
965 		return 1;
966 
967 	return run_filter(filter, skb);
968 }
969 
970 /* Network device part of the driver */
971 
972 static const struct ethtool_ops tun_ethtool_ops;
973 
974 static int tun_net_init(struct net_device *dev)
975 {
976 	struct tun_struct *tun = netdev_priv(dev);
977 	struct ifreq *ifr = tun->ifr;
978 	int err;
979 
980 	dev->tstats = netdev_alloc_pcpu_stats(struct pcpu_sw_netstats);
981 	if (!dev->tstats)
982 		return -ENOMEM;
983 
984 	spin_lock_init(&tun->lock);
985 
986 	err = security_tun_dev_alloc_security(&tun->security);
987 	if (err < 0) {
988 		free_percpu(dev->tstats);
989 		return err;
990 	}
991 
992 	tun_flow_init(tun);
993 
994 	dev->hw_features = NETIF_F_SG | NETIF_F_FRAGLIST |
995 			   TUN_USER_FEATURES | NETIF_F_HW_VLAN_CTAG_TX |
996 			   NETIF_F_HW_VLAN_STAG_TX;
997 	dev->features = dev->hw_features | NETIF_F_LLTX;
998 	dev->vlan_features = dev->features &
999 			     ~(NETIF_F_HW_VLAN_CTAG_TX |
1000 			       NETIF_F_HW_VLAN_STAG_TX);
1001 
1002 	tun->flags = (tun->flags & ~TUN_FEATURES) |
1003 		      (ifr->ifr_flags & TUN_FEATURES);
1004 
1005 	INIT_LIST_HEAD(&tun->disabled);
1006 	err = tun_attach(tun, tun->file, false, ifr->ifr_flags & IFF_NAPI,
1007 			 ifr->ifr_flags & IFF_NAPI_FRAGS, false);
1008 	if (err < 0) {
1009 		tun_flow_uninit(tun);
1010 		security_tun_dev_free_security(tun->security);
1011 		free_percpu(dev->tstats);
1012 		return err;
1013 	}
1014 	return 0;
1015 }
1016 
1017 /* Net device detach from fd. */
1018 static void tun_net_uninit(struct net_device *dev)
1019 {
1020 	tun_detach_all(dev);
1021 }
1022 
1023 /* Net device open. */
1024 static int tun_net_open(struct net_device *dev)
1025 {
1026 	netif_tx_start_all_queues(dev);
1027 
1028 	return 0;
1029 }
1030 
1031 /* Net device close. */
1032 static int tun_net_close(struct net_device *dev)
1033 {
1034 	netif_tx_stop_all_queues(dev);
1035 	return 0;
1036 }
1037 
1038 /* Net device start xmit */
1039 static void tun_automq_xmit(struct tun_struct *tun, struct sk_buff *skb)
1040 {
1041 #ifdef CONFIG_RPS
1042 	if (tun->numqueues == 1 && static_branch_unlikely(&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.
1045 		 */
1046 		struct tun_flow_entry *e;
1047 		__u32 rxhash;
1048 
1049 		rxhash = __skb_get_hash_symmetric(skb);
1050 		e = tun_flow_find(&tun->flows[tun_hashfn(rxhash)], rxhash);
1051 		if (e)
1052 			tun_flow_save_rps_rxhash(e, rxhash);
1053 	}
1054 #endif
1055 }
1056 
1057 static unsigned int run_ebpf_filter(struct tun_struct *tun,
1058 				    struct sk_buff *skb,
1059 				    int len)
1060 {
1061 	struct tun_prog *prog = rcu_dereference(tun->filter_prog);
1062 
1063 	if (prog)
1064 		len = bpf_prog_run_clear_cb(prog->prog, skb);
1065 
1066 	return len;
1067 }
1068 
1069 /* Net device start xmit */
1070 static netdev_tx_t tun_net_xmit(struct sk_buff *skb, struct net_device *dev)
1071 {
1072 	struct tun_struct *tun = netdev_priv(dev);
1073 	enum skb_drop_reason drop_reason;
1074 	int txq = skb->queue_mapping;
1075 	struct netdev_queue *queue;
1076 	struct tun_file *tfile;
1077 	int len = skb->len;
1078 
1079 	rcu_read_lock();
1080 	tfile = rcu_dereference(tun->tfiles[txq]);
1081 
1082 	/* Drop packet if interface is not attached */
1083 	if (!tfile) {
1084 		drop_reason = SKB_DROP_REASON_DEV_READY;
1085 		goto drop;
1086 	}
1087 
1088 	if (!rcu_dereference(tun->steering_prog))
1089 		tun_automq_xmit(tun, skb);
1090 
1091 	netif_info(tun, tx_queued, tun->dev, "%s %d\n", __func__, skb->len);
1092 
1093 	/* Drop if the filter does not like it.
1094 	 * This is a noop if the filter is disabled.
1095 	 * Filter can be enabled only for the TAP devices. */
1096 	if (!check_filter(&tun->txflt, skb)) {
1097 		drop_reason = SKB_DROP_REASON_TAP_TXFILTER;
1098 		goto drop;
1099 	}
1100 
1101 	if (tfile->socket.sk->sk_filter &&
1102 	    sk_filter(tfile->socket.sk, skb)) {
1103 		drop_reason = SKB_DROP_REASON_SOCKET_FILTER;
1104 		goto drop;
1105 	}
1106 
1107 	len = run_ebpf_filter(tun, skb, len);
1108 	if (len == 0) {
1109 		drop_reason = SKB_DROP_REASON_TAP_FILTER;
1110 		goto drop;
1111 	}
1112 
1113 	if (pskb_trim(skb, len)) {
1114 		drop_reason = SKB_DROP_REASON_NOMEM;
1115 		goto drop;
1116 	}
1117 
1118 	if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC))) {
1119 		drop_reason = SKB_DROP_REASON_SKB_UCOPY_FAULT;
1120 		goto drop;
1121 	}
1122 
1123 	skb_tx_timestamp(skb);
1124 
1125 	/* Orphan the skb - required as we might hang on to it
1126 	 * for indefinite time.
1127 	 */
1128 	skb_orphan(skb);
1129 
1130 	nf_reset_ct(skb);
1131 
1132 	if (ptr_ring_produce(&tfile->tx_ring, skb)) {
1133 		drop_reason = SKB_DROP_REASON_FULL_RING;
1134 		goto drop;
1135 	}
1136 
1137 	/* NETIF_F_LLTX requires to do our own update of trans_start */
1138 	queue = netdev_get_tx_queue(dev, txq);
1139 	txq_trans_cond_update(queue);
1140 
1141 	/* Notify and wake up reader process */
1142 	if (tfile->flags & TUN_FASYNC)
1143 		kill_fasync(&tfile->fasync, SIGIO, POLL_IN);
1144 	tfile->socket.sk->sk_data_ready(tfile->socket.sk);
1145 
1146 	rcu_read_unlock();
1147 	return NETDEV_TX_OK;
1148 
1149 drop:
1150 	dev_core_stats_tx_dropped_inc(dev);
1151 	skb_tx_error(skb);
1152 	kfree_skb_reason(skb, drop_reason);
1153 	rcu_read_unlock();
1154 	return NET_XMIT_DROP;
1155 }
1156 
1157 static void tun_net_mclist(struct net_device *dev)
1158 {
1159 	/*
1160 	 * This callback is supposed to deal with mc filter in
1161 	 * _rx_ path and has nothing to do with the _tx_ path.
1162 	 * In rx path we always accept everything userspace gives us.
1163 	 */
1164 }
1165 
1166 static netdev_features_t tun_net_fix_features(struct net_device *dev,
1167 	netdev_features_t features)
1168 {
1169 	struct tun_struct *tun = netdev_priv(dev);
1170 
1171 	return (features & tun->set_features) | (features & ~TUN_USER_FEATURES);
1172 }
1173 
1174 static void tun_set_headroom(struct net_device *dev, int new_hr)
1175 {
1176 	struct tun_struct *tun = netdev_priv(dev);
1177 
1178 	if (new_hr < NET_SKB_PAD)
1179 		new_hr = NET_SKB_PAD;
1180 
1181 	tun->align = new_hr;
1182 }
1183 
1184 static void
1185 tun_net_get_stats64(struct net_device *dev, struct rtnl_link_stats64 *stats)
1186 {
1187 	struct tun_struct *tun = netdev_priv(dev);
1188 
1189 	dev_get_tstats64(dev, stats);
1190 
1191 	stats->rx_frame_errors +=
1192 		(unsigned long)atomic_long_read(&tun->rx_frame_errors);
1193 }
1194 
1195 static int tun_xdp_set(struct net_device *dev, struct bpf_prog *prog,
1196 		       struct netlink_ext_ack *extack)
1197 {
1198 	struct tun_struct *tun = netdev_priv(dev);
1199 	struct tun_file *tfile;
1200 	struct bpf_prog *old_prog;
1201 	int i;
1202 
1203 	old_prog = rtnl_dereference(tun->xdp_prog);
1204 	rcu_assign_pointer(tun->xdp_prog, prog);
1205 	if (old_prog)
1206 		bpf_prog_put(old_prog);
1207 
1208 	for (i = 0; i < tun->numqueues; i++) {
1209 		tfile = rtnl_dereference(tun->tfiles[i]);
1210 		if (prog)
1211 			sock_set_flag(&tfile->sk, SOCK_XDP);
1212 		else
1213 			sock_reset_flag(&tfile->sk, SOCK_XDP);
1214 	}
1215 	list_for_each_entry(tfile, &tun->disabled, next) {
1216 		if (prog)
1217 			sock_set_flag(&tfile->sk, SOCK_XDP);
1218 		else
1219 			sock_reset_flag(&tfile->sk, SOCK_XDP);
1220 	}
1221 
1222 	return 0;
1223 }
1224 
1225 static int tun_xdp(struct net_device *dev, struct netdev_bpf *xdp)
1226 {
1227 	switch (xdp->command) {
1228 	case XDP_SETUP_PROG:
1229 		return tun_xdp_set(dev, xdp->prog, xdp->extack);
1230 	default:
1231 		return -EINVAL;
1232 	}
1233 }
1234 
1235 static int tun_net_change_carrier(struct net_device *dev, bool new_carrier)
1236 {
1237 	if (new_carrier) {
1238 		struct tun_struct *tun = netdev_priv(dev);
1239 
1240 		if (!tun->numqueues)
1241 			return -EPERM;
1242 
1243 		netif_carrier_on(dev);
1244 	} else {
1245 		netif_carrier_off(dev);
1246 	}
1247 	return 0;
1248 }
1249 
1250 static const struct net_device_ops tun_netdev_ops = {
1251 	.ndo_init		= tun_net_init,
1252 	.ndo_uninit		= tun_net_uninit,
1253 	.ndo_open		= tun_net_open,
1254 	.ndo_stop		= tun_net_close,
1255 	.ndo_start_xmit		= tun_net_xmit,
1256 	.ndo_fix_features	= tun_net_fix_features,
1257 	.ndo_select_queue	= tun_select_queue,
1258 	.ndo_set_rx_headroom	= tun_set_headroom,
1259 	.ndo_get_stats64	= tun_net_get_stats64,
1260 	.ndo_change_carrier	= tun_net_change_carrier,
1261 };
1262 
1263 static void __tun_xdp_flush_tfile(struct tun_file *tfile)
1264 {
1265 	/* Notify and wake up reader process */
1266 	if (tfile->flags & TUN_FASYNC)
1267 		kill_fasync(&tfile->fasync, SIGIO, POLL_IN);
1268 	tfile->socket.sk->sk_data_ready(tfile->socket.sk);
1269 }
1270 
1271 static int tun_xdp_xmit(struct net_device *dev, int n,
1272 			struct xdp_frame **frames, u32 flags)
1273 {
1274 	struct tun_struct *tun = netdev_priv(dev);
1275 	struct tun_file *tfile;
1276 	u32 numqueues;
1277 	int nxmit = 0;
1278 	int i;
1279 
1280 	if (unlikely(flags & ~XDP_XMIT_FLAGS_MASK))
1281 		return -EINVAL;
1282 
1283 	rcu_read_lock();
1284 
1285 resample:
1286 	numqueues = READ_ONCE(tun->numqueues);
1287 	if (!numqueues) {
1288 		rcu_read_unlock();
1289 		return -ENXIO; /* Caller will free/return all frames */
1290 	}
1291 
1292 	tfile = rcu_dereference(tun->tfiles[smp_processor_id() %
1293 					    numqueues]);
1294 	if (unlikely(!tfile))
1295 		goto resample;
1296 
1297 	spin_lock(&tfile->tx_ring.producer_lock);
1298 	for (i = 0; i < n; i++) {
1299 		struct xdp_frame *xdp = frames[i];
1300 		/* Encode the XDP flag into lowest bit for consumer to differ
1301 		 * XDP buffer from sk_buff.
1302 		 */
1303 		void *frame = tun_xdp_to_ptr(xdp);
1304 
1305 		if (__ptr_ring_produce(&tfile->tx_ring, frame)) {
1306 			dev_core_stats_tx_dropped_inc(dev);
1307 			break;
1308 		}
1309 		nxmit++;
1310 	}
1311 	spin_unlock(&tfile->tx_ring.producer_lock);
1312 
1313 	if (flags & XDP_XMIT_FLUSH)
1314 		__tun_xdp_flush_tfile(tfile);
1315 
1316 	rcu_read_unlock();
1317 	return nxmit;
1318 }
1319 
1320 static int tun_xdp_tx(struct net_device *dev, struct xdp_buff *xdp)
1321 {
1322 	struct xdp_frame *frame = xdp_convert_buff_to_frame(xdp);
1323 	int nxmit;
1324 
1325 	if (unlikely(!frame))
1326 		return -EOVERFLOW;
1327 
1328 	nxmit = tun_xdp_xmit(dev, 1, &frame, XDP_XMIT_FLUSH);
1329 	if (!nxmit)
1330 		xdp_return_frame_rx_napi(frame);
1331 	return nxmit;
1332 }
1333 
1334 static const struct net_device_ops tap_netdev_ops = {
1335 	.ndo_init		= tun_net_init,
1336 	.ndo_uninit		= tun_net_uninit,
1337 	.ndo_open		= tun_net_open,
1338 	.ndo_stop		= tun_net_close,
1339 	.ndo_start_xmit		= tun_net_xmit,
1340 	.ndo_fix_features	= tun_net_fix_features,
1341 	.ndo_set_rx_mode	= tun_net_mclist,
1342 	.ndo_set_mac_address	= eth_mac_addr,
1343 	.ndo_validate_addr	= eth_validate_addr,
1344 	.ndo_select_queue	= tun_select_queue,
1345 	.ndo_features_check	= passthru_features_check,
1346 	.ndo_set_rx_headroom	= tun_set_headroom,
1347 	.ndo_get_stats64	= dev_get_tstats64,
1348 	.ndo_bpf		= tun_xdp,
1349 	.ndo_xdp_xmit		= tun_xdp_xmit,
1350 	.ndo_change_carrier	= tun_net_change_carrier,
1351 };
1352 
1353 static void tun_flow_init(struct tun_struct *tun)
1354 {
1355 	int i;
1356 
1357 	for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++)
1358 		INIT_HLIST_HEAD(&tun->flows[i]);
1359 
1360 	tun->ageing_time = TUN_FLOW_EXPIRE;
1361 	timer_setup(&tun->flow_gc_timer, tun_flow_cleanup, 0);
1362 	mod_timer(&tun->flow_gc_timer,
1363 		  round_jiffies_up(jiffies + tun->ageing_time));
1364 }
1365 
1366 static void tun_flow_uninit(struct tun_struct *tun)
1367 {
1368 	del_timer_sync(&tun->flow_gc_timer);
1369 	tun_flow_flush(tun);
1370 }
1371 
1372 #define MIN_MTU 68
1373 #define MAX_MTU 65535
1374 
1375 /* Initialize net device. */
1376 static void tun_net_initialize(struct net_device *dev)
1377 {
1378 	struct tun_struct *tun = netdev_priv(dev);
1379 
1380 	switch (tun->flags & TUN_TYPE_MASK) {
1381 	case IFF_TUN:
1382 		dev->netdev_ops = &tun_netdev_ops;
1383 		dev->header_ops = &ip_tunnel_header_ops;
1384 
1385 		/* Point-to-Point TUN Device */
1386 		dev->hard_header_len = 0;
1387 		dev->addr_len = 0;
1388 		dev->mtu = 1500;
1389 
1390 		/* Zero header length */
1391 		dev->type = ARPHRD_NONE;
1392 		dev->flags = IFF_POINTOPOINT | IFF_NOARP | IFF_MULTICAST;
1393 		break;
1394 
1395 	case IFF_TAP:
1396 		dev->netdev_ops = &tap_netdev_ops;
1397 		/* Ethernet TAP Device */
1398 		ether_setup(dev);
1399 		dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1400 		dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
1401 
1402 		eth_hw_addr_random(dev);
1403 
1404 		/* Currently tun does not support XDP, only tap does. */
1405 		dev->xdp_features = NETDEV_XDP_ACT_BASIC |
1406 				    NETDEV_XDP_ACT_REDIRECT |
1407 				    NETDEV_XDP_ACT_NDO_XMIT;
1408 
1409 		break;
1410 	}
1411 
1412 	dev->min_mtu = MIN_MTU;
1413 	dev->max_mtu = MAX_MTU - dev->hard_header_len;
1414 }
1415 
1416 static bool tun_sock_writeable(struct tun_struct *tun, struct tun_file *tfile)
1417 {
1418 	struct sock *sk = tfile->socket.sk;
1419 
1420 	return (tun->dev->flags & IFF_UP) && sock_writeable(sk);
1421 }
1422 
1423 /* Character device part */
1424 
1425 /* Poll */
1426 static __poll_t tun_chr_poll(struct file *file, poll_table *wait)
1427 {
1428 	struct tun_file *tfile = file->private_data;
1429 	struct tun_struct *tun = tun_get(tfile);
1430 	struct sock *sk;
1431 	__poll_t mask = 0;
1432 
1433 	if (!tun)
1434 		return EPOLLERR;
1435 
1436 	sk = tfile->socket.sk;
1437 
1438 	poll_wait(file, sk_sleep(sk), wait);
1439 
1440 	if (!ptr_ring_empty(&tfile->tx_ring))
1441 		mask |= EPOLLIN | EPOLLRDNORM;
1442 
1443 	/* Make sure SOCKWQ_ASYNC_NOSPACE is set if not writable to
1444 	 * guarantee EPOLLOUT to be raised by either here or
1445 	 * tun_sock_write_space(). Then process could get notification
1446 	 * after it writes to a down device and meets -EIO.
1447 	 */
1448 	if (tun_sock_writeable(tun, tfile) ||
1449 	    (!test_and_set_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags) &&
1450 	     tun_sock_writeable(tun, tfile)))
1451 		mask |= EPOLLOUT | EPOLLWRNORM;
1452 
1453 	if (tun->dev->reg_state != NETREG_REGISTERED)
1454 		mask = EPOLLERR;
1455 
1456 	tun_put(tun);
1457 	return mask;
1458 }
1459 
1460 static struct sk_buff *tun_napi_alloc_frags(struct tun_file *tfile,
1461 					    size_t len,
1462 					    const struct iov_iter *it)
1463 {
1464 	struct sk_buff *skb;
1465 	size_t linear;
1466 	int err;
1467 	int i;
1468 
1469 	if (it->nr_segs > MAX_SKB_FRAGS + 1 ||
1470 	    len > (ETH_MAX_MTU - NET_SKB_PAD - NET_IP_ALIGN))
1471 		return ERR_PTR(-EMSGSIZE);
1472 
1473 	local_bh_disable();
1474 	skb = napi_get_frags(&tfile->napi);
1475 	local_bh_enable();
1476 	if (!skb)
1477 		return ERR_PTR(-ENOMEM);
1478 
1479 	linear = iov_iter_single_seg_count(it);
1480 	err = __skb_grow(skb, linear);
1481 	if (err)
1482 		goto free;
1483 
1484 	skb->len = len;
1485 	skb->data_len = len - linear;
1486 	skb->truesize += skb->data_len;
1487 
1488 	for (i = 1; i < it->nr_segs; i++) {
1489 		const struct iovec *iov = iter_iov(it);
1490 		size_t fragsz = iov->iov_len;
1491 		struct page *page;
1492 		void *frag;
1493 
1494 		if (fragsz == 0 || fragsz > PAGE_SIZE) {
1495 			err = -EINVAL;
1496 			goto free;
1497 		}
1498 		frag = netdev_alloc_frag(fragsz);
1499 		if (!frag) {
1500 			err = -ENOMEM;
1501 			goto free;
1502 		}
1503 		page = virt_to_head_page(frag);
1504 		skb_fill_page_desc(skb, i - 1, page,
1505 				   frag - page_address(page), fragsz);
1506 	}
1507 
1508 	return skb;
1509 free:
1510 	/* frees skb and all frags allocated with napi_alloc_frag() */
1511 	napi_free_frags(&tfile->napi);
1512 	return ERR_PTR(err);
1513 }
1514 
1515 /* prepad is the amount to reserve at front.  len is length after that.
1516  * linear is a hint as to how much to copy (usually headers). */
1517 static struct sk_buff *tun_alloc_skb(struct tun_file *tfile,
1518 				     size_t prepad, size_t len,
1519 				     size_t linear, int noblock)
1520 {
1521 	struct sock *sk = tfile->socket.sk;
1522 	struct sk_buff *skb;
1523 	int err;
1524 
1525 	/* Under a page?  Don't bother with paged skb. */
1526 	if (prepad + len < PAGE_SIZE || !linear)
1527 		linear = len;
1528 
1529 	skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
1530 				   &err, 0);
1531 	if (!skb)
1532 		return ERR_PTR(err);
1533 
1534 	skb_reserve(skb, prepad);
1535 	skb_put(skb, linear);
1536 	skb->data_len = len - linear;
1537 	skb->len += len - linear;
1538 
1539 	return skb;
1540 }
1541 
1542 static void tun_rx_batched(struct tun_struct *tun, struct tun_file *tfile,
1543 			   struct sk_buff *skb, int more)
1544 {
1545 	struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
1546 	struct sk_buff_head process_queue;
1547 	u32 rx_batched = tun->rx_batched;
1548 	bool rcv = false;
1549 
1550 	if (!rx_batched || (!more && skb_queue_empty(queue))) {
1551 		local_bh_disable();
1552 		skb_record_rx_queue(skb, tfile->queue_index);
1553 		netif_receive_skb(skb);
1554 		local_bh_enable();
1555 		return;
1556 	}
1557 
1558 	spin_lock(&queue->lock);
1559 	if (!more || skb_queue_len(queue) == rx_batched) {
1560 		__skb_queue_head_init(&process_queue);
1561 		skb_queue_splice_tail_init(queue, &process_queue);
1562 		rcv = true;
1563 	} else {
1564 		__skb_queue_tail(queue, skb);
1565 	}
1566 	spin_unlock(&queue->lock);
1567 
1568 	if (rcv) {
1569 		struct sk_buff *nskb;
1570 
1571 		local_bh_disable();
1572 		while ((nskb = __skb_dequeue(&process_queue))) {
1573 			skb_record_rx_queue(nskb, tfile->queue_index);
1574 			netif_receive_skb(nskb);
1575 		}
1576 		skb_record_rx_queue(skb, tfile->queue_index);
1577 		netif_receive_skb(skb);
1578 		local_bh_enable();
1579 	}
1580 }
1581 
1582 static bool tun_can_build_skb(struct tun_struct *tun, struct tun_file *tfile,
1583 			      int len, int noblock, bool zerocopy)
1584 {
1585 	if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
1586 		return false;
1587 
1588 	if (tfile->socket.sk->sk_sndbuf != INT_MAX)
1589 		return false;
1590 
1591 	if (!noblock)
1592 		return false;
1593 
1594 	if (zerocopy)
1595 		return false;
1596 
1597 	if (SKB_DATA_ALIGN(len + TUN_RX_PAD + XDP_PACKET_HEADROOM) +
1598 	    SKB_DATA_ALIGN(sizeof(struct skb_shared_info)) > PAGE_SIZE)
1599 		return false;
1600 
1601 	return true;
1602 }
1603 
1604 static struct sk_buff *__tun_build_skb(struct tun_file *tfile,
1605 				       struct page_frag *alloc_frag, char *buf,
1606 				       int buflen, int len, int pad)
1607 {
1608 	struct sk_buff *skb = build_skb(buf, buflen);
1609 
1610 	if (!skb)
1611 		return ERR_PTR(-ENOMEM);
1612 
1613 	skb_reserve(skb, pad);
1614 	skb_put(skb, len);
1615 	skb_set_owner_w(skb, tfile->socket.sk);
1616 
1617 	get_page(alloc_frag->page);
1618 	alloc_frag->offset += buflen;
1619 
1620 	return skb;
1621 }
1622 
1623 static int tun_xdp_act(struct tun_struct *tun, struct bpf_prog *xdp_prog,
1624 		       struct xdp_buff *xdp, u32 act)
1625 {
1626 	int err;
1627 
1628 	switch (act) {
1629 	case XDP_REDIRECT:
1630 		err = xdp_do_redirect(tun->dev, xdp, xdp_prog);
1631 		if (err)
1632 			return err;
1633 		break;
1634 	case XDP_TX:
1635 		err = tun_xdp_tx(tun->dev, xdp);
1636 		if (err < 0)
1637 			return err;
1638 		break;
1639 	case XDP_PASS:
1640 		break;
1641 	default:
1642 		bpf_warn_invalid_xdp_action(tun->dev, xdp_prog, act);
1643 		fallthrough;
1644 	case XDP_ABORTED:
1645 		trace_xdp_exception(tun->dev, xdp_prog, act);
1646 		fallthrough;
1647 	case XDP_DROP:
1648 		dev_core_stats_rx_dropped_inc(tun->dev);
1649 		break;
1650 	}
1651 
1652 	return act;
1653 }
1654 
1655 static struct sk_buff *tun_build_skb(struct tun_struct *tun,
1656 				     struct tun_file *tfile,
1657 				     struct iov_iter *from,
1658 				     struct virtio_net_hdr *hdr,
1659 				     int len, int *skb_xdp)
1660 {
1661 	struct page_frag *alloc_frag = &current->task_frag;
1662 	struct bpf_prog *xdp_prog;
1663 	int buflen = SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
1664 	char *buf;
1665 	size_t copied;
1666 	int pad = TUN_RX_PAD;
1667 	int err = 0;
1668 
1669 	rcu_read_lock();
1670 	xdp_prog = rcu_dereference(tun->xdp_prog);
1671 	if (xdp_prog)
1672 		pad += XDP_PACKET_HEADROOM;
1673 	buflen += SKB_DATA_ALIGN(len + pad);
1674 	rcu_read_unlock();
1675 
1676 	alloc_frag->offset = ALIGN((u64)alloc_frag->offset, SMP_CACHE_BYTES);
1677 	if (unlikely(!skb_page_frag_refill(buflen, alloc_frag, GFP_KERNEL)))
1678 		return ERR_PTR(-ENOMEM);
1679 
1680 	buf = (char *)page_address(alloc_frag->page) + alloc_frag->offset;
1681 	copied = copy_page_from_iter(alloc_frag->page,
1682 				     alloc_frag->offset + pad,
1683 				     len, from);
1684 	if (copied != len)
1685 		return ERR_PTR(-EFAULT);
1686 
1687 	/* There's a small window that XDP may be set after the check
1688 	 * of xdp_prog above, this should be rare and for simplicity
1689 	 * we do XDP on skb in case the headroom is not enough.
1690 	 */
1691 	if (hdr->gso_type || !xdp_prog) {
1692 		*skb_xdp = 1;
1693 		return __tun_build_skb(tfile, alloc_frag, buf, buflen, len,
1694 				       pad);
1695 	}
1696 
1697 	*skb_xdp = 0;
1698 
1699 	local_bh_disable();
1700 	rcu_read_lock();
1701 	xdp_prog = rcu_dereference(tun->xdp_prog);
1702 	if (xdp_prog) {
1703 		struct xdp_buff xdp;
1704 		u32 act;
1705 
1706 		xdp_init_buff(&xdp, buflen, &tfile->xdp_rxq);
1707 		xdp_prepare_buff(&xdp, buf, pad, len, false);
1708 
1709 		act = bpf_prog_run_xdp(xdp_prog, &xdp);
1710 		if (act == XDP_REDIRECT || act == XDP_TX) {
1711 			get_page(alloc_frag->page);
1712 			alloc_frag->offset += buflen;
1713 		}
1714 		err = tun_xdp_act(tun, xdp_prog, &xdp, act);
1715 		if (err < 0) {
1716 			if (act == XDP_REDIRECT || act == XDP_TX)
1717 				put_page(alloc_frag->page);
1718 			goto out;
1719 		}
1720 
1721 		if (err == XDP_REDIRECT)
1722 			xdp_do_flush();
1723 		if (err != XDP_PASS)
1724 			goto out;
1725 
1726 		pad = xdp.data - xdp.data_hard_start;
1727 		len = xdp.data_end - xdp.data;
1728 	}
1729 	rcu_read_unlock();
1730 	local_bh_enable();
1731 
1732 	return __tun_build_skb(tfile, alloc_frag, buf, buflen, len, pad);
1733 
1734 out:
1735 	rcu_read_unlock();
1736 	local_bh_enable();
1737 	return NULL;
1738 }
1739 
1740 /* Get packet from user space buffer */
1741 static ssize_t tun_get_user(struct tun_struct *tun, struct tun_file *tfile,
1742 			    void *msg_control, struct iov_iter *from,
1743 			    int noblock, bool more)
1744 {
1745 	struct tun_pi pi = { 0, cpu_to_be16(ETH_P_IP) };
1746 	struct sk_buff *skb;
1747 	size_t total_len = iov_iter_count(from);
1748 	size_t len = total_len, align = tun->align, linear;
1749 	struct virtio_net_hdr gso = { 0 };
1750 	int good_linear;
1751 	int copylen;
1752 	bool zerocopy = false;
1753 	int err;
1754 	u32 rxhash = 0;
1755 	int skb_xdp = 1;
1756 	bool frags = tun_napi_frags_enabled(tfile);
1757 	enum skb_drop_reason drop_reason = SKB_DROP_REASON_NOT_SPECIFIED;
1758 
1759 	if (!(tun->flags & IFF_NO_PI)) {
1760 		if (len < sizeof(pi))
1761 			return -EINVAL;
1762 		len -= sizeof(pi);
1763 
1764 		if (!copy_from_iter_full(&pi, sizeof(pi), from))
1765 			return -EFAULT;
1766 	}
1767 
1768 	if (tun->flags & IFF_VNET_HDR) {
1769 		int vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
1770 
1771 		if (len < vnet_hdr_sz)
1772 			return -EINVAL;
1773 		len -= vnet_hdr_sz;
1774 
1775 		if (!copy_from_iter_full(&gso, sizeof(gso), from))
1776 			return -EFAULT;
1777 
1778 		if ((gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
1779 		    tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2 > tun16_to_cpu(tun, gso.hdr_len))
1780 			gso.hdr_len = cpu_to_tun16(tun, tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2);
1781 
1782 		if (tun16_to_cpu(tun, gso.hdr_len) > len)
1783 			return -EINVAL;
1784 		iov_iter_advance(from, vnet_hdr_sz - sizeof(gso));
1785 	}
1786 
1787 	if ((tun->flags & TUN_TYPE_MASK) == IFF_TAP) {
1788 		align += NET_IP_ALIGN;
1789 		if (unlikely(len < ETH_HLEN ||
1790 			     (gso.hdr_len && tun16_to_cpu(tun, gso.hdr_len) < ETH_HLEN)))
1791 			return -EINVAL;
1792 	}
1793 
1794 	good_linear = SKB_MAX_HEAD(align);
1795 
1796 	if (msg_control) {
1797 		struct iov_iter i = *from;
1798 
1799 		/* There are 256 bytes to be copied in skb, so there is
1800 		 * enough room for skb expand head in case it is used.
1801 		 * The rest of the buffer is mapped from userspace.
1802 		 */
1803 		copylen = gso.hdr_len ? tun16_to_cpu(tun, gso.hdr_len) : GOODCOPY_LEN;
1804 		if (copylen > good_linear)
1805 			copylen = good_linear;
1806 		linear = copylen;
1807 		iov_iter_advance(&i, copylen);
1808 		if (iov_iter_npages(&i, INT_MAX) <= MAX_SKB_FRAGS)
1809 			zerocopy = true;
1810 	}
1811 
1812 	if (!frags && tun_can_build_skb(tun, tfile, len, noblock, zerocopy)) {
1813 		/* For the packet that is not easy to be processed
1814 		 * (e.g gso or jumbo packet), we will do it at after
1815 		 * skb was created with generic XDP routine.
1816 		 */
1817 		skb = tun_build_skb(tun, tfile, from, &gso, len, &skb_xdp);
1818 		err = PTR_ERR_OR_ZERO(skb);
1819 		if (err)
1820 			goto drop;
1821 		if (!skb)
1822 			return total_len;
1823 	} else {
1824 		if (!zerocopy) {
1825 			copylen = len;
1826 			if (tun16_to_cpu(tun, gso.hdr_len) > good_linear)
1827 				linear = good_linear;
1828 			else
1829 				linear = tun16_to_cpu(tun, gso.hdr_len);
1830 		}
1831 
1832 		if (frags) {
1833 			mutex_lock(&tfile->napi_mutex);
1834 			skb = tun_napi_alloc_frags(tfile, copylen, from);
1835 			/* tun_napi_alloc_frags() enforces a layout for the skb.
1836 			 * If zerocopy is enabled, then this layout will be
1837 			 * overwritten by zerocopy_sg_from_iter().
1838 			 */
1839 			zerocopy = false;
1840 		} else {
1841 			skb = tun_alloc_skb(tfile, align, copylen, linear,
1842 					    noblock);
1843 		}
1844 
1845 		err = PTR_ERR_OR_ZERO(skb);
1846 		if (err)
1847 			goto drop;
1848 
1849 		if (zerocopy)
1850 			err = zerocopy_sg_from_iter(skb, from);
1851 		else
1852 			err = skb_copy_datagram_from_iter(skb, 0, from, len);
1853 
1854 		if (err) {
1855 			err = -EFAULT;
1856 			drop_reason = SKB_DROP_REASON_SKB_UCOPY_FAULT;
1857 			goto drop;
1858 		}
1859 	}
1860 
1861 	if (virtio_net_hdr_to_skb(skb, &gso, tun_is_little_endian(tun))) {
1862 		atomic_long_inc(&tun->rx_frame_errors);
1863 		err = -EINVAL;
1864 		goto free_skb;
1865 	}
1866 
1867 	switch (tun->flags & TUN_TYPE_MASK) {
1868 	case IFF_TUN:
1869 		if (tun->flags & IFF_NO_PI) {
1870 			u8 ip_version = skb->len ? (skb->data[0] >> 4) : 0;
1871 
1872 			switch (ip_version) {
1873 			case 4:
1874 				pi.proto = htons(ETH_P_IP);
1875 				break;
1876 			case 6:
1877 				pi.proto = htons(ETH_P_IPV6);
1878 				break;
1879 			default:
1880 				err = -EINVAL;
1881 				goto drop;
1882 			}
1883 		}
1884 
1885 		skb_reset_mac_header(skb);
1886 		skb->protocol = pi.proto;
1887 		skb->dev = tun->dev;
1888 		break;
1889 	case IFF_TAP:
1890 		if (frags && !pskb_may_pull(skb, ETH_HLEN)) {
1891 			err = -ENOMEM;
1892 			drop_reason = SKB_DROP_REASON_HDR_TRUNC;
1893 			goto drop;
1894 		}
1895 		skb->protocol = eth_type_trans(skb, tun->dev);
1896 		break;
1897 	}
1898 
1899 	/* copy skb_ubuf_info for callback when skb has no error */
1900 	if (zerocopy) {
1901 		skb_zcopy_init(skb, msg_control);
1902 	} else if (msg_control) {
1903 		struct ubuf_info *uarg = msg_control;
1904 		uarg->callback(NULL, uarg, false);
1905 	}
1906 
1907 	skb_reset_network_header(skb);
1908 	skb_probe_transport_header(skb);
1909 	skb_record_rx_queue(skb, tfile->queue_index);
1910 
1911 	if (skb_xdp) {
1912 		struct bpf_prog *xdp_prog;
1913 		int ret;
1914 
1915 		local_bh_disable();
1916 		rcu_read_lock();
1917 		xdp_prog = rcu_dereference(tun->xdp_prog);
1918 		if (xdp_prog) {
1919 			ret = do_xdp_generic(xdp_prog, skb);
1920 			if (ret != XDP_PASS) {
1921 				rcu_read_unlock();
1922 				local_bh_enable();
1923 				goto unlock_frags;
1924 			}
1925 		}
1926 		rcu_read_unlock();
1927 		local_bh_enable();
1928 	}
1929 
1930 	/* Compute the costly rx hash only if needed for flow updates.
1931 	 * We may get a very small possibility of OOO during switching, not
1932 	 * worth to optimize.
1933 	 */
1934 	if (!rcu_access_pointer(tun->steering_prog) && tun->numqueues > 1 &&
1935 	    !tfile->detached)
1936 		rxhash = __skb_get_hash_symmetric(skb);
1937 
1938 	rcu_read_lock();
1939 	if (unlikely(!(tun->dev->flags & IFF_UP))) {
1940 		err = -EIO;
1941 		rcu_read_unlock();
1942 		drop_reason = SKB_DROP_REASON_DEV_READY;
1943 		goto drop;
1944 	}
1945 
1946 	if (frags) {
1947 		u32 headlen;
1948 
1949 		/* Exercise flow dissector code path. */
1950 		skb_push(skb, ETH_HLEN);
1951 		headlen = eth_get_headlen(tun->dev, skb->data,
1952 					  skb_headlen(skb));
1953 
1954 		if (unlikely(headlen > skb_headlen(skb))) {
1955 			WARN_ON_ONCE(1);
1956 			err = -ENOMEM;
1957 			dev_core_stats_rx_dropped_inc(tun->dev);
1958 napi_busy:
1959 			napi_free_frags(&tfile->napi);
1960 			rcu_read_unlock();
1961 			mutex_unlock(&tfile->napi_mutex);
1962 			return err;
1963 		}
1964 
1965 		if (likely(napi_schedule_prep(&tfile->napi))) {
1966 			local_bh_disable();
1967 			napi_gro_frags(&tfile->napi);
1968 			napi_complete(&tfile->napi);
1969 			local_bh_enable();
1970 		} else {
1971 			err = -EBUSY;
1972 			goto napi_busy;
1973 		}
1974 		mutex_unlock(&tfile->napi_mutex);
1975 	} else if (tfile->napi_enabled) {
1976 		struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
1977 		int queue_len;
1978 
1979 		spin_lock_bh(&queue->lock);
1980 
1981 		if (unlikely(tfile->detached)) {
1982 			spin_unlock_bh(&queue->lock);
1983 			rcu_read_unlock();
1984 			err = -EBUSY;
1985 			goto free_skb;
1986 		}
1987 
1988 		__skb_queue_tail(queue, skb);
1989 		queue_len = skb_queue_len(queue);
1990 		spin_unlock(&queue->lock);
1991 
1992 		if (!more || queue_len > NAPI_POLL_WEIGHT)
1993 			napi_schedule(&tfile->napi);
1994 
1995 		local_bh_enable();
1996 	} else if (!IS_ENABLED(CONFIG_4KSTACKS)) {
1997 		tun_rx_batched(tun, tfile, skb, more);
1998 	} else {
1999 		netif_rx(skb);
2000 	}
2001 	rcu_read_unlock();
2002 
2003 	preempt_disable();
2004 	dev_sw_netstats_rx_add(tun->dev, len);
2005 	preempt_enable();
2006 
2007 	if (rxhash)
2008 		tun_flow_update(tun, rxhash, tfile);
2009 
2010 	return total_len;
2011 
2012 drop:
2013 	if (err != -EAGAIN)
2014 		dev_core_stats_rx_dropped_inc(tun->dev);
2015 
2016 free_skb:
2017 	if (!IS_ERR_OR_NULL(skb))
2018 		kfree_skb_reason(skb, drop_reason);
2019 
2020 unlock_frags:
2021 	if (frags) {
2022 		tfile->napi.skb = NULL;
2023 		mutex_unlock(&tfile->napi_mutex);
2024 	}
2025 
2026 	return err ?: total_len;
2027 }
2028 
2029 static ssize_t tun_chr_write_iter(struct kiocb *iocb, struct iov_iter *from)
2030 {
2031 	struct file *file = iocb->ki_filp;
2032 	struct tun_file *tfile = file->private_data;
2033 	struct tun_struct *tun = tun_get(tfile);
2034 	ssize_t result;
2035 	int noblock = 0;
2036 
2037 	if (!tun)
2038 		return -EBADFD;
2039 
2040 	if ((file->f_flags & O_NONBLOCK) || (iocb->ki_flags & IOCB_NOWAIT))
2041 		noblock = 1;
2042 
2043 	result = tun_get_user(tun, tfile, NULL, from, noblock, false);
2044 
2045 	tun_put(tun);
2046 	return result;
2047 }
2048 
2049 static ssize_t tun_put_user_xdp(struct tun_struct *tun,
2050 				struct tun_file *tfile,
2051 				struct xdp_frame *xdp_frame,
2052 				struct iov_iter *iter)
2053 {
2054 	int vnet_hdr_sz = 0;
2055 	size_t size = xdp_frame->len;
2056 	size_t ret;
2057 
2058 	if (tun->flags & IFF_VNET_HDR) {
2059 		struct virtio_net_hdr gso = { 0 };
2060 
2061 		vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
2062 		if (unlikely(iov_iter_count(iter) < vnet_hdr_sz))
2063 			return -EINVAL;
2064 		if (unlikely(copy_to_iter(&gso, sizeof(gso), iter) !=
2065 			     sizeof(gso)))
2066 			return -EFAULT;
2067 		iov_iter_advance(iter, vnet_hdr_sz - sizeof(gso));
2068 	}
2069 
2070 	ret = copy_to_iter(xdp_frame->data, size, iter) + vnet_hdr_sz;
2071 
2072 	preempt_disable();
2073 	dev_sw_netstats_tx_add(tun->dev, 1, ret);
2074 	preempt_enable();
2075 
2076 	return ret;
2077 }
2078 
2079 /* Put packet to the user space buffer */
2080 static ssize_t tun_put_user(struct tun_struct *tun,
2081 			    struct tun_file *tfile,
2082 			    struct sk_buff *skb,
2083 			    struct iov_iter *iter)
2084 {
2085 	struct tun_pi pi = { 0, skb->protocol };
2086 	ssize_t total;
2087 	int vlan_offset = 0;
2088 	int vlan_hlen = 0;
2089 	int vnet_hdr_sz = 0;
2090 
2091 	if (skb_vlan_tag_present(skb))
2092 		vlan_hlen = VLAN_HLEN;
2093 
2094 	if (tun->flags & IFF_VNET_HDR)
2095 		vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
2096 
2097 	total = skb->len + vlan_hlen + vnet_hdr_sz;
2098 
2099 	if (!(tun->flags & IFF_NO_PI)) {
2100 		if (iov_iter_count(iter) < sizeof(pi))
2101 			return -EINVAL;
2102 
2103 		total += sizeof(pi);
2104 		if (iov_iter_count(iter) < total) {
2105 			/* Packet will be striped */
2106 			pi.flags |= TUN_PKT_STRIP;
2107 		}
2108 
2109 		if (copy_to_iter(&pi, sizeof(pi), iter) != sizeof(pi))
2110 			return -EFAULT;
2111 	}
2112 
2113 	if (vnet_hdr_sz) {
2114 		struct virtio_net_hdr gso;
2115 
2116 		if (iov_iter_count(iter) < vnet_hdr_sz)
2117 			return -EINVAL;
2118 
2119 		if (virtio_net_hdr_from_skb(skb, &gso,
2120 					    tun_is_little_endian(tun), true,
2121 					    vlan_hlen)) {
2122 			struct skb_shared_info *sinfo = skb_shinfo(skb);
2123 			pr_err("unexpected GSO type: "
2124 			       "0x%x, gso_size %d, hdr_len %d\n",
2125 			       sinfo->gso_type, tun16_to_cpu(tun, gso.gso_size),
2126 			       tun16_to_cpu(tun, gso.hdr_len));
2127 			print_hex_dump(KERN_ERR, "tun: ",
2128 				       DUMP_PREFIX_NONE,
2129 				       16, 1, skb->head,
2130 				       min((int)tun16_to_cpu(tun, gso.hdr_len), 64), true);
2131 			WARN_ON_ONCE(1);
2132 			return -EINVAL;
2133 		}
2134 
2135 		if (copy_to_iter(&gso, sizeof(gso), iter) != sizeof(gso))
2136 			return -EFAULT;
2137 
2138 		iov_iter_advance(iter, vnet_hdr_sz - sizeof(gso));
2139 	}
2140 
2141 	if (vlan_hlen) {
2142 		int ret;
2143 		struct veth veth;
2144 
2145 		veth.h_vlan_proto = skb->vlan_proto;
2146 		veth.h_vlan_TCI = htons(skb_vlan_tag_get(skb));
2147 
2148 		vlan_offset = offsetof(struct vlan_ethhdr, h_vlan_proto);
2149 
2150 		ret = skb_copy_datagram_iter(skb, 0, iter, vlan_offset);
2151 		if (ret || !iov_iter_count(iter))
2152 			goto done;
2153 
2154 		ret = copy_to_iter(&veth, sizeof(veth), iter);
2155 		if (ret != sizeof(veth) || !iov_iter_count(iter))
2156 			goto done;
2157 	}
2158 
2159 	skb_copy_datagram_iter(skb, vlan_offset, iter, skb->len - vlan_offset);
2160 
2161 done:
2162 	/* caller is in process context, */
2163 	preempt_disable();
2164 	dev_sw_netstats_tx_add(tun->dev, 1, skb->len + vlan_hlen);
2165 	preempt_enable();
2166 
2167 	return total;
2168 }
2169 
2170 static void *tun_ring_recv(struct tun_file *tfile, int noblock, int *err)
2171 {
2172 	DECLARE_WAITQUEUE(wait, current);
2173 	void *ptr = NULL;
2174 	int error = 0;
2175 
2176 	ptr = ptr_ring_consume(&tfile->tx_ring);
2177 	if (ptr)
2178 		goto out;
2179 	if (noblock) {
2180 		error = -EAGAIN;
2181 		goto out;
2182 	}
2183 
2184 	add_wait_queue(&tfile->socket.wq.wait, &wait);
2185 
2186 	while (1) {
2187 		set_current_state(TASK_INTERRUPTIBLE);
2188 		ptr = ptr_ring_consume(&tfile->tx_ring);
2189 		if (ptr)
2190 			break;
2191 		if (signal_pending(current)) {
2192 			error = -ERESTARTSYS;
2193 			break;
2194 		}
2195 		if (tfile->socket.sk->sk_shutdown & RCV_SHUTDOWN) {
2196 			error = -EFAULT;
2197 			break;
2198 		}
2199 
2200 		schedule();
2201 	}
2202 
2203 	__set_current_state(TASK_RUNNING);
2204 	remove_wait_queue(&tfile->socket.wq.wait, &wait);
2205 
2206 out:
2207 	*err = error;
2208 	return ptr;
2209 }
2210 
2211 static ssize_t tun_do_read(struct tun_struct *tun, struct tun_file *tfile,
2212 			   struct iov_iter *to,
2213 			   int noblock, void *ptr)
2214 {
2215 	ssize_t ret;
2216 	int err;
2217 
2218 	if (!iov_iter_count(to)) {
2219 		tun_ptr_free(ptr);
2220 		return 0;
2221 	}
2222 
2223 	if (!ptr) {
2224 		/* Read frames from ring */
2225 		ptr = tun_ring_recv(tfile, noblock, &err);
2226 		if (!ptr)
2227 			return err;
2228 	}
2229 
2230 	if (tun_is_xdp_frame(ptr)) {
2231 		struct xdp_frame *xdpf = tun_ptr_to_xdp(ptr);
2232 
2233 		ret = tun_put_user_xdp(tun, tfile, xdpf, to);
2234 		xdp_return_frame(xdpf);
2235 	} else {
2236 		struct sk_buff *skb = ptr;
2237 
2238 		ret = tun_put_user(tun, tfile, skb, to);
2239 		if (unlikely(ret < 0))
2240 			kfree_skb(skb);
2241 		else
2242 			consume_skb(skb);
2243 	}
2244 
2245 	return ret;
2246 }
2247 
2248 static ssize_t tun_chr_read_iter(struct kiocb *iocb, struct iov_iter *to)
2249 {
2250 	struct file *file = iocb->ki_filp;
2251 	struct tun_file *tfile = file->private_data;
2252 	struct tun_struct *tun = tun_get(tfile);
2253 	ssize_t len = iov_iter_count(to), ret;
2254 	int noblock = 0;
2255 
2256 	if (!tun)
2257 		return -EBADFD;
2258 
2259 	if ((file->f_flags & O_NONBLOCK) || (iocb->ki_flags & IOCB_NOWAIT))
2260 		noblock = 1;
2261 
2262 	ret = tun_do_read(tun, tfile, to, noblock, NULL);
2263 	ret = min_t(ssize_t, ret, len);
2264 	if (ret > 0)
2265 		iocb->ki_pos = ret;
2266 	tun_put(tun);
2267 	return ret;
2268 }
2269 
2270 static void tun_prog_free(struct rcu_head *rcu)
2271 {
2272 	struct tun_prog *prog = container_of(rcu, struct tun_prog, rcu);
2273 
2274 	bpf_prog_destroy(prog->prog);
2275 	kfree(prog);
2276 }
2277 
2278 static int __tun_set_ebpf(struct tun_struct *tun,
2279 			  struct tun_prog __rcu **prog_p,
2280 			  struct bpf_prog *prog)
2281 {
2282 	struct tun_prog *old, *new = NULL;
2283 
2284 	if (prog) {
2285 		new = kmalloc(sizeof(*new), GFP_KERNEL);
2286 		if (!new)
2287 			return -ENOMEM;
2288 		new->prog = prog;
2289 	}
2290 
2291 	spin_lock_bh(&tun->lock);
2292 	old = rcu_dereference_protected(*prog_p,
2293 					lockdep_is_held(&tun->lock));
2294 	rcu_assign_pointer(*prog_p, new);
2295 	spin_unlock_bh(&tun->lock);
2296 
2297 	if (old)
2298 		call_rcu(&old->rcu, tun_prog_free);
2299 
2300 	return 0;
2301 }
2302 
2303 static void tun_free_netdev(struct net_device *dev)
2304 {
2305 	struct tun_struct *tun = netdev_priv(dev);
2306 
2307 	BUG_ON(!(list_empty(&tun->disabled)));
2308 
2309 	free_percpu(dev->tstats);
2310 	tun_flow_uninit(tun);
2311 	security_tun_dev_free_security(tun->security);
2312 	__tun_set_ebpf(tun, &tun->steering_prog, NULL);
2313 	__tun_set_ebpf(tun, &tun->filter_prog, NULL);
2314 }
2315 
2316 static void tun_setup(struct net_device *dev)
2317 {
2318 	struct tun_struct *tun = netdev_priv(dev);
2319 
2320 	tun->owner = INVALID_UID;
2321 	tun->group = INVALID_GID;
2322 	tun_default_link_ksettings(dev, &tun->link_ksettings);
2323 
2324 	dev->ethtool_ops = &tun_ethtool_ops;
2325 	dev->needs_free_netdev = true;
2326 	dev->priv_destructor = tun_free_netdev;
2327 	/* We prefer our own queue length */
2328 	dev->tx_queue_len = TUN_READQ_SIZE;
2329 }
2330 
2331 /* Trivial set of netlink ops to allow deleting tun or tap
2332  * device with netlink.
2333  */
2334 static int tun_validate(struct nlattr *tb[], struct nlattr *data[],
2335 			struct netlink_ext_ack *extack)
2336 {
2337 	NL_SET_ERR_MSG(extack,
2338 		       "tun/tap creation via rtnetlink is not supported.");
2339 	return -EOPNOTSUPP;
2340 }
2341 
2342 static size_t tun_get_size(const struct net_device *dev)
2343 {
2344 	BUILD_BUG_ON(sizeof(u32) != sizeof(uid_t));
2345 	BUILD_BUG_ON(sizeof(u32) != sizeof(gid_t));
2346 
2347 	return nla_total_size(sizeof(uid_t)) + /* OWNER */
2348 	       nla_total_size(sizeof(gid_t)) + /* GROUP */
2349 	       nla_total_size(sizeof(u8)) + /* TYPE */
2350 	       nla_total_size(sizeof(u8)) + /* PI */
2351 	       nla_total_size(sizeof(u8)) + /* VNET_HDR */
2352 	       nla_total_size(sizeof(u8)) + /* PERSIST */
2353 	       nla_total_size(sizeof(u8)) + /* MULTI_QUEUE */
2354 	       nla_total_size(sizeof(u32)) + /* NUM_QUEUES */
2355 	       nla_total_size(sizeof(u32)) + /* NUM_DISABLED_QUEUES */
2356 	       0;
2357 }
2358 
2359 static int tun_fill_info(struct sk_buff *skb, const struct net_device *dev)
2360 {
2361 	struct tun_struct *tun = netdev_priv(dev);
2362 
2363 	if (nla_put_u8(skb, IFLA_TUN_TYPE, tun->flags & TUN_TYPE_MASK))
2364 		goto nla_put_failure;
2365 	if (uid_valid(tun->owner) &&
2366 	    nla_put_u32(skb, IFLA_TUN_OWNER,
2367 			from_kuid_munged(current_user_ns(), tun->owner)))
2368 		goto nla_put_failure;
2369 	if (gid_valid(tun->group) &&
2370 	    nla_put_u32(skb, IFLA_TUN_GROUP,
2371 			from_kgid_munged(current_user_ns(), tun->group)))
2372 		goto nla_put_failure;
2373 	if (nla_put_u8(skb, IFLA_TUN_PI, !(tun->flags & IFF_NO_PI)))
2374 		goto nla_put_failure;
2375 	if (nla_put_u8(skb, IFLA_TUN_VNET_HDR, !!(tun->flags & IFF_VNET_HDR)))
2376 		goto nla_put_failure;
2377 	if (nla_put_u8(skb, IFLA_TUN_PERSIST, !!(tun->flags & IFF_PERSIST)))
2378 		goto nla_put_failure;
2379 	if (nla_put_u8(skb, IFLA_TUN_MULTI_QUEUE,
2380 		       !!(tun->flags & IFF_MULTI_QUEUE)))
2381 		goto nla_put_failure;
2382 	if (tun->flags & IFF_MULTI_QUEUE) {
2383 		if (nla_put_u32(skb, IFLA_TUN_NUM_QUEUES, tun->numqueues))
2384 			goto nla_put_failure;
2385 		if (nla_put_u32(skb, IFLA_TUN_NUM_DISABLED_QUEUES,
2386 				tun->numdisabled))
2387 			goto nla_put_failure;
2388 	}
2389 
2390 	return 0;
2391 
2392 nla_put_failure:
2393 	return -EMSGSIZE;
2394 }
2395 
2396 static struct rtnl_link_ops tun_link_ops __read_mostly = {
2397 	.kind		= DRV_NAME,
2398 	.priv_size	= sizeof(struct tun_struct),
2399 	.setup		= tun_setup,
2400 	.validate	= tun_validate,
2401 	.get_size       = tun_get_size,
2402 	.fill_info      = tun_fill_info,
2403 };
2404 
2405 static void tun_sock_write_space(struct sock *sk)
2406 {
2407 	struct tun_file *tfile;
2408 	wait_queue_head_t *wqueue;
2409 
2410 	if (!sock_writeable(sk))
2411 		return;
2412 
2413 	if (!test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags))
2414 		return;
2415 
2416 	wqueue = sk_sleep(sk);
2417 	if (wqueue && waitqueue_active(wqueue))
2418 		wake_up_interruptible_sync_poll(wqueue, EPOLLOUT |
2419 						EPOLLWRNORM | EPOLLWRBAND);
2420 
2421 	tfile = container_of(sk, struct tun_file, sk);
2422 	kill_fasync(&tfile->fasync, SIGIO, POLL_OUT);
2423 }
2424 
2425 static void tun_put_page(struct tun_page *tpage)
2426 {
2427 	if (tpage->page)
2428 		__page_frag_cache_drain(tpage->page, tpage->count);
2429 }
2430 
2431 static int tun_xdp_one(struct tun_struct *tun,
2432 		       struct tun_file *tfile,
2433 		       struct xdp_buff *xdp, int *flush,
2434 		       struct tun_page *tpage)
2435 {
2436 	unsigned int datasize = xdp->data_end - xdp->data;
2437 	struct tun_xdp_hdr *hdr = xdp->data_hard_start;
2438 	struct virtio_net_hdr *gso = &hdr->gso;
2439 	struct bpf_prog *xdp_prog;
2440 	struct sk_buff *skb = NULL;
2441 	struct sk_buff_head *queue;
2442 	u32 rxhash = 0, act;
2443 	int buflen = hdr->buflen;
2444 	int ret = 0;
2445 	bool skb_xdp = false;
2446 	struct page *page;
2447 
2448 	xdp_prog = rcu_dereference(tun->xdp_prog);
2449 	if (xdp_prog) {
2450 		if (gso->gso_type) {
2451 			skb_xdp = true;
2452 			goto build;
2453 		}
2454 
2455 		xdp_init_buff(xdp, buflen, &tfile->xdp_rxq);
2456 		xdp_set_data_meta_invalid(xdp);
2457 
2458 		act = bpf_prog_run_xdp(xdp_prog, xdp);
2459 		ret = tun_xdp_act(tun, xdp_prog, xdp, act);
2460 		if (ret < 0) {
2461 			put_page(virt_to_head_page(xdp->data));
2462 			return ret;
2463 		}
2464 
2465 		switch (ret) {
2466 		case XDP_REDIRECT:
2467 			*flush = true;
2468 			fallthrough;
2469 		case XDP_TX:
2470 			return 0;
2471 		case XDP_PASS:
2472 			break;
2473 		default:
2474 			page = virt_to_head_page(xdp->data);
2475 			if (tpage->page == page) {
2476 				++tpage->count;
2477 			} else {
2478 				tun_put_page(tpage);
2479 				tpage->page = page;
2480 				tpage->count = 1;
2481 			}
2482 			return 0;
2483 		}
2484 	}
2485 
2486 build:
2487 	skb = build_skb(xdp->data_hard_start, buflen);
2488 	if (!skb) {
2489 		ret = -ENOMEM;
2490 		goto out;
2491 	}
2492 
2493 	skb_reserve(skb, xdp->data - xdp->data_hard_start);
2494 	skb_put(skb, xdp->data_end - xdp->data);
2495 
2496 	if (virtio_net_hdr_to_skb(skb, gso, tun_is_little_endian(tun))) {
2497 		atomic_long_inc(&tun->rx_frame_errors);
2498 		kfree_skb(skb);
2499 		ret = -EINVAL;
2500 		goto out;
2501 	}
2502 
2503 	skb->protocol = eth_type_trans(skb, tun->dev);
2504 	skb_reset_network_header(skb);
2505 	skb_probe_transport_header(skb);
2506 	skb_record_rx_queue(skb, tfile->queue_index);
2507 
2508 	if (skb_xdp) {
2509 		ret = do_xdp_generic(xdp_prog, skb);
2510 		if (ret != XDP_PASS) {
2511 			ret = 0;
2512 			goto out;
2513 		}
2514 	}
2515 
2516 	if (!rcu_dereference(tun->steering_prog) && tun->numqueues > 1 &&
2517 	    !tfile->detached)
2518 		rxhash = __skb_get_hash_symmetric(skb);
2519 
2520 	if (tfile->napi_enabled) {
2521 		queue = &tfile->sk.sk_write_queue;
2522 		spin_lock(&queue->lock);
2523 
2524 		if (unlikely(tfile->detached)) {
2525 			spin_unlock(&queue->lock);
2526 			kfree_skb(skb);
2527 			return -EBUSY;
2528 		}
2529 
2530 		__skb_queue_tail(queue, skb);
2531 		spin_unlock(&queue->lock);
2532 		ret = 1;
2533 	} else {
2534 		netif_receive_skb(skb);
2535 		ret = 0;
2536 	}
2537 
2538 	/* No need to disable preemption here since this function is
2539 	 * always called with bh disabled
2540 	 */
2541 	dev_sw_netstats_rx_add(tun->dev, datasize);
2542 
2543 	if (rxhash)
2544 		tun_flow_update(tun, rxhash, tfile);
2545 
2546 out:
2547 	return ret;
2548 }
2549 
2550 static int tun_sendmsg(struct socket *sock, struct msghdr *m, size_t total_len)
2551 {
2552 	int ret, i;
2553 	struct tun_file *tfile = container_of(sock, struct tun_file, socket);
2554 	struct tun_struct *tun = tun_get(tfile);
2555 	struct tun_msg_ctl *ctl = m->msg_control;
2556 	struct xdp_buff *xdp;
2557 
2558 	if (!tun)
2559 		return -EBADFD;
2560 
2561 	if (m->msg_controllen == sizeof(struct tun_msg_ctl) &&
2562 	    ctl && ctl->type == TUN_MSG_PTR) {
2563 		struct tun_page tpage;
2564 		int n = ctl->num;
2565 		int flush = 0, queued = 0;
2566 
2567 		memset(&tpage, 0, sizeof(tpage));
2568 
2569 		local_bh_disable();
2570 		rcu_read_lock();
2571 
2572 		for (i = 0; i < n; i++) {
2573 			xdp = &((struct xdp_buff *)ctl->ptr)[i];
2574 			ret = tun_xdp_one(tun, tfile, xdp, &flush, &tpage);
2575 			if (ret > 0)
2576 				queued += ret;
2577 		}
2578 
2579 		if (flush)
2580 			xdp_do_flush();
2581 
2582 		if (tfile->napi_enabled && queued > 0)
2583 			napi_schedule(&tfile->napi);
2584 
2585 		rcu_read_unlock();
2586 		local_bh_enable();
2587 
2588 		tun_put_page(&tpage);
2589 
2590 		ret = total_len;
2591 		goto out;
2592 	}
2593 
2594 	ret = tun_get_user(tun, tfile, ctl ? ctl->ptr : NULL, &m->msg_iter,
2595 			   m->msg_flags & MSG_DONTWAIT,
2596 			   m->msg_flags & MSG_MORE);
2597 out:
2598 	tun_put(tun);
2599 	return ret;
2600 }
2601 
2602 static int tun_recvmsg(struct socket *sock, struct msghdr *m, size_t total_len,
2603 		       int flags)
2604 {
2605 	struct tun_file *tfile = container_of(sock, struct tun_file, socket);
2606 	struct tun_struct *tun = tun_get(tfile);
2607 	void *ptr = m->msg_control;
2608 	int ret;
2609 
2610 	if (!tun) {
2611 		ret = -EBADFD;
2612 		goto out_free;
2613 	}
2614 
2615 	if (flags & ~(MSG_DONTWAIT|MSG_TRUNC|MSG_ERRQUEUE)) {
2616 		ret = -EINVAL;
2617 		goto out_put_tun;
2618 	}
2619 	if (flags & MSG_ERRQUEUE) {
2620 		ret = sock_recv_errqueue(sock->sk, m, total_len,
2621 					 SOL_PACKET, TUN_TX_TIMESTAMP);
2622 		goto out;
2623 	}
2624 	ret = tun_do_read(tun, tfile, &m->msg_iter, flags & MSG_DONTWAIT, ptr);
2625 	if (ret > (ssize_t)total_len) {
2626 		m->msg_flags |= MSG_TRUNC;
2627 		ret = flags & MSG_TRUNC ? ret : total_len;
2628 	}
2629 out:
2630 	tun_put(tun);
2631 	return ret;
2632 
2633 out_put_tun:
2634 	tun_put(tun);
2635 out_free:
2636 	tun_ptr_free(ptr);
2637 	return ret;
2638 }
2639 
2640 static int tun_ptr_peek_len(void *ptr)
2641 {
2642 	if (likely(ptr)) {
2643 		if (tun_is_xdp_frame(ptr)) {
2644 			struct xdp_frame *xdpf = tun_ptr_to_xdp(ptr);
2645 
2646 			return xdpf->len;
2647 		}
2648 		return __skb_array_len_with_tag(ptr);
2649 	} else {
2650 		return 0;
2651 	}
2652 }
2653 
2654 static int tun_peek_len(struct socket *sock)
2655 {
2656 	struct tun_file *tfile = container_of(sock, struct tun_file, socket);
2657 	struct tun_struct *tun;
2658 	int ret = 0;
2659 
2660 	tun = tun_get(tfile);
2661 	if (!tun)
2662 		return 0;
2663 
2664 	ret = PTR_RING_PEEK_CALL(&tfile->tx_ring, tun_ptr_peek_len);
2665 	tun_put(tun);
2666 
2667 	return ret;
2668 }
2669 
2670 /* Ops structure to mimic raw sockets with tun */
2671 static const struct proto_ops tun_socket_ops = {
2672 	.peek_len = tun_peek_len,
2673 	.sendmsg = tun_sendmsg,
2674 	.recvmsg = tun_recvmsg,
2675 };
2676 
2677 static struct proto tun_proto = {
2678 	.name		= "tun",
2679 	.owner		= THIS_MODULE,
2680 	.obj_size	= sizeof(struct tun_file),
2681 };
2682 
2683 static int tun_flags(struct tun_struct *tun)
2684 {
2685 	return tun->flags & (TUN_FEATURES | IFF_PERSIST | IFF_TUN | IFF_TAP);
2686 }
2687 
2688 static ssize_t tun_flags_show(struct device *dev, struct device_attribute *attr,
2689 			      char *buf)
2690 {
2691 	struct tun_struct *tun = netdev_priv(to_net_dev(dev));
2692 	return sysfs_emit(buf, "0x%x\n", tun_flags(tun));
2693 }
2694 
2695 static ssize_t owner_show(struct device *dev, struct device_attribute *attr,
2696 			  char *buf)
2697 {
2698 	struct tun_struct *tun = netdev_priv(to_net_dev(dev));
2699 	return uid_valid(tun->owner)?
2700 		sysfs_emit(buf, "%u\n",
2701 			   from_kuid_munged(current_user_ns(), tun->owner)) :
2702 		sysfs_emit(buf, "-1\n");
2703 }
2704 
2705 static ssize_t group_show(struct device *dev, struct device_attribute *attr,
2706 			  char *buf)
2707 {
2708 	struct tun_struct *tun = netdev_priv(to_net_dev(dev));
2709 	return gid_valid(tun->group) ?
2710 		sysfs_emit(buf, "%u\n",
2711 			   from_kgid_munged(current_user_ns(), tun->group)) :
2712 		sysfs_emit(buf, "-1\n");
2713 }
2714 
2715 static DEVICE_ATTR_RO(tun_flags);
2716 static DEVICE_ATTR_RO(owner);
2717 static DEVICE_ATTR_RO(group);
2718 
2719 static struct attribute *tun_dev_attrs[] = {
2720 	&dev_attr_tun_flags.attr,
2721 	&dev_attr_owner.attr,
2722 	&dev_attr_group.attr,
2723 	NULL
2724 };
2725 
2726 static const struct attribute_group tun_attr_group = {
2727 	.attrs = tun_dev_attrs
2728 };
2729 
2730 static int tun_set_iff(struct net *net, struct file *file, struct ifreq *ifr)
2731 {
2732 	struct tun_struct *tun;
2733 	struct tun_file *tfile = file->private_data;
2734 	struct net_device *dev;
2735 	int err;
2736 
2737 	if (tfile->detached)
2738 		return -EINVAL;
2739 
2740 	if ((ifr->ifr_flags & IFF_NAPI_FRAGS)) {
2741 		if (!capable(CAP_NET_ADMIN))
2742 			return -EPERM;
2743 
2744 		if (!(ifr->ifr_flags & IFF_NAPI) ||
2745 		    (ifr->ifr_flags & TUN_TYPE_MASK) != IFF_TAP)
2746 			return -EINVAL;
2747 	}
2748 
2749 	dev = __dev_get_by_name(net, ifr->ifr_name);
2750 	if (dev) {
2751 		if (ifr->ifr_flags & IFF_TUN_EXCL)
2752 			return -EBUSY;
2753 		if ((ifr->ifr_flags & IFF_TUN) && dev->netdev_ops == &tun_netdev_ops)
2754 			tun = netdev_priv(dev);
2755 		else if ((ifr->ifr_flags & IFF_TAP) && dev->netdev_ops == &tap_netdev_ops)
2756 			tun = netdev_priv(dev);
2757 		else
2758 			return -EINVAL;
2759 
2760 		if (!!(ifr->ifr_flags & IFF_MULTI_QUEUE) !=
2761 		    !!(tun->flags & IFF_MULTI_QUEUE))
2762 			return -EINVAL;
2763 
2764 		if (tun_not_capable(tun))
2765 			return -EPERM;
2766 		err = security_tun_dev_open(tun->security);
2767 		if (err < 0)
2768 			return err;
2769 
2770 		err = tun_attach(tun, file, ifr->ifr_flags & IFF_NOFILTER,
2771 				 ifr->ifr_flags & IFF_NAPI,
2772 				 ifr->ifr_flags & IFF_NAPI_FRAGS, true);
2773 		if (err < 0)
2774 			return err;
2775 
2776 		if (tun->flags & IFF_MULTI_QUEUE &&
2777 		    (tun->numqueues + tun->numdisabled > 1)) {
2778 			/* One or more queue has already been attached, no need
2779 			 * to initialize the device again.
2780 			 */
2781 			netdev_state_change(dev);
2782 			return 0;
2783 		}
2784 
2785 		tun->flags = (tun->flags & ~TUN_FEATURES) |
2786 			      (ifr->ifr_flags & TUN_FEATURES);
2787 
2788 		netdev_state_change(dev);
2789 	} else {
2790 		char *name;
2791 		unsigned long flags = 0;
2792 		int queues = ifr->ifr_flags & IFF_MULTI_QUEUE ?
2793 			     MAX_TAP_QUEUES : 1;
2794 
2795 		if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
2796 			return -EPERM;
2797 		err = security_tun_dev_create();
2798 		if (err < 0)
2799 			return err;
2800 
2801 		/* Set dev type */
2802 		if (ifr->ifr_flags & IFF_TUN) {
2803 			/* TUN device */
2804 			flags |= IFF_TUN;
2805 			name = "tun%d";
2806 		} else if (ifr->ifr_flags & IFF_TAP) {
2807 			/* TAP device */
2808 			flags |= IFF_TAP;
2809 			name = "tap%d";
2810 		} else
2811 			return -EINVAL;
2812 
2813 		if (*ifr->ifr_name)
2814 			name = ifr->ifr_name;
2815 
2816 		dev = alloc_netdev_mqs(sizeof(struct tun_struct), name,
2817 				       NET_NAME_UNKNOWN, tun_setup, queues,
2818 				       queues);
2819 
2820 		if (!dev)
2821 			return -ENOMEM;
2822 
2823 		dev_net_set(dev, net);
2824 		dev->rtnl_link_ops = &tun_link_ops;
2825 		dev->ifindex = tfile->ifindex;
2826 		dev->sysfs_groups[0] = &tun_attr_group;
2827 
2828 		tun = netdev_priv(dev);
2829 		tun->dev = dev;
2830 		tun->flags = flags;
2831 		tun->txflt.count = 0;
2832 		tun->vnet_hdr_sz = sizeof(struct virtio_net_hdr);
2833 
2834 		tun->align = NET_SKB_PAD;
2835 		tun->filter_attached = false;
2836 		tun->sndbuf = tfile->socket.sk->sk_sndbuf;
2837 		tun->rx_batched = 0;
2838 		RCU_INIT_POINTER(tun->steering_prog, NULL);
2839 
2840 		tun->ifr = ifr;
2841 		tun->file = file;
2842 
2843 		tun_net_initialize(dev);
2844 
2845 		err = register_netdevice(tun->dev);
2846 		if (err < 0) {
2847 			free_netdev(dev);
2848 			return err;
2849 		}
2850 		/* free_netdev() won't check refcnt, to avoid race
2851 		 * with dev_put() we need publish tun after registration.
2852 		 */
2853 		rcu_assign_pointer(tfile->tun, tun);
2854 	}
2855 
2856 	if (ifr->ifr_flags & IFF_NO_CARRIER)
2857 		netif_carrier_off(tun->dev);
2858 	else
2859 		netif_carrier_on(tun->dev);
2860 
2861 	/* Make sure persistent devices do not get stuck in
2862 	 * xoff state.
2863 	 */
2864 	if (netif_running(tun->dev))
2865 		netif_tx_wake_all_queues(tun->dev);
2866 
2867 	strcpy(ifr->ifr_name, tun->dev->name);
2868 	return 0;
2869 }
2870 
2871 static void tun_get_iff(struct tun_struct *tun, struct ifreq *ifr)
2872 {
2873 	strcpy(ifr->ifr_name, tun->dev->name);
2874 
2875 	ifr->ifr_flags = tun_flags(tun);
2876 
2877 }
2878 
2879 /* This is like a cut-down ethtool ops, except done via tun fd so no
2880  * privs required. */
2881 static int set_offload(struct tun_struct *tun, unsigned long arg)
2882 {
2883 	netdev_features_t features = 0;
2884 
2885 	if (arg & TUN_F_CSUM) {
2886 		features |= NETIF_F_HW_CSUM;
2887 		arg &= ~TUN_F_CSUM;
2888 
2889 		if (arg & (TUN_F_TSO4|TUN_F_TSO6)) {
2890 			if (arg & TUN_F_TSO_ECN) {
2891 				features |= NETIF_F_TSO_ECN;
2892 				arg &= ~TUN_F_TSO_ECN;
2893 			}
2894 			if (arg & TUN_F_TSO4)
2895 				features |= NETIF_F_TSO;
2896 			if (arg & TUN_F_TSO6)
2897 				features |= NETIF_F_TSO6;
2898 			arg &= ~(TUN_F_TSO4|TUN_F_TSO6);
2899 		}
2900 
2901 		arg &= ~TUN_F_UFO;
2902 
2903 		/* TODO: for now USO4 and USO6 should work simultaneously */
2904 		if (arg & TUN_F_USO4 && arg & TUN_F_USO6) {
2905 			features |= NETIF_F_GSO_UDP_L4;
2906 			arg &= ~(TUN_F_USO4 | TUN_F_USO6);
2907 		}
2908 	}
2909 
2910 	/* This gives the user a way to test for new features in future by
2911 	 * trying to set them. */
2912 	if (arg)
2913 		return -EINVAL;
2914 
2915 	tun->set_features = features;
2916 	tun->dev->wanted_features &= ~TUN_USER_FEATURES;
2917 	tun->dev->wanted_features |= features;
2918 	netdev_update_features(tun->dev);
2919 
2920 	return 0;
2921 }
2922 
2923 static void tun_detach_filter(struct tun_struct *tun, int n)
2924 {
2925 	int i;
2926 	struct tun_file *tfile;
2927 
2928 	for (i = 0; i < n; i++) {
2929 		tfile = rtnl_dereference(tun->tfiles[i]);
2930 		lock_sock(tfile->socket.sk);
2931 		sk_detach_filter(tfile->socket.sk);
2932 		release_sock(tfile->socket.sk);
2933 	}
2934 
2935 	tun->filter_attached = false;
2936 }
2937 
2938 static int tun_attach_filter(struct tun_struct *tun)
2939 {
2940 	int i, ret = 0;
2941 	struct tun_file *tfile;
2942 
2943 	for (i = 0; i < tun->numqueues; i++) {
2944 		tfile = rtnl_dereference(tun->tfiles[i]);
2945 		lock_sock(tfile->socket.sk);
2946 		ret = sk_attach_filter(&tun->fprog, tfile->socket.sk);
2947 		release_sock(tfile->socket.sk);
2948 		if (ret) {
2949 			tun_detach_filter(tun, i);
2950 			return ret;
2951 		}
2952 	}
2953 
2954 	tun->filter_attached = true;
2955 	return ret;
2956 }
2957 
2958 static void tun_set_sndbuf(struct tun_struct *tun)
2959 {
2960 	struct tun_file *tfile;
2961 	int i;
2962 
2963 	for (i = 0; i < tun->numqueues; i++) {
2964 		tfile = rtnl_dereference(tun->tfiles[i]);
2965 		tfile->socket.sk->sk_sndbuf = tun->sndbuf;
2966 	}
2967 }
2968 
2969 static int tun_set_queue(struct file *file, struct ifreq *ifr)
2970 {
2971 	struct tun_file *tfile = file->private_data;
2972 	struct tun_struct *tun;
2973 	int ret = 0;
2974 
2975 	rtnl_lock();
2976 
2977 	if (ifr->ifr_flags & IFF_ATTACH_QUEUE) {
2978 		tun = tfile->detached;
2979 		if (!tun) {
2980 			ret = -EINVAL;
2981 			goto unlock;
2982 		}
2983 		ret = security_tun_dev_attach_queue(tun->security);
2984 		if (ret < 0)
2985 			goto unlock;
2986 		ret = tun_attach(tun, file, false, tun->flags & IFF_NAPI,
2987 				 tun->flags & IFF_NAPI_FRAGS, true);
2988 	} else if (ifr->ifr_flags & IFF_DETACH_QUEUE) {
2989 		tun = rtnl_dereference(tfile->tun);
2990 		if (!tun || !(tun->flags & IFF_MULTI_QUEUE) || tfile->detached)
2991 			ret = -EINVAL;
2992 		else
2993 			__tun_detach(tfile, false);
2994 	} else
2995 		ret = -EINVAL;
2996 
2997 	if (ret >= 0)
2998 		netdev_state_change(tun->dev);
2999 
3000 unlock:
3001 	rtnl_unlock();
3002 	return ret;
3003 }
3004 
3005 static int tun_set_ebpf(struct tun_struct *tun, struct tun_prog __rcu **prog_p,
3006 			void __user *data)
3007 {
3008 	struct bpf_prog *prog;
3009 	int fd;
3010 
3011 	if (copy_from_user(&fd, data, sizeof(fd)))
3012 		return -EFAULT;
3013 
3014 	if (fd == -1) {
3015 		prog = NULL;
3016 	} else {
3017 		prog = bpf_prog_get_type(fd, BPF_PROG_TYPE_SOCKET_FILTER);
3018 		if (IS_ERR(prog))
3019 			return PTR_ERR(prog);
3020 	}
3021 
3022 	return __tun_set_ebpf(tun, prog_p, prog);
3023 }
3024 
3025 /* Return correct value for tun->dev->addr_len based on tun->dev->type. */
3026 static unsigned char tun_get_addr_len(unsigned short type)
3027 {
3028 	switch (type) {
3029 	case ARPHRD_IP6GRE:
3030 	case ARPHRD_TUNNEL6:
3031 		return sizeof(struct in6_addr);
3032 	case ARPHRD_IPGRE:
3033 	case ARPHRD_TUNNEL:
3034 	case ARPHRD_SIT:
3035 		return 4;
3036 	case ARPHRD_ETHER:
3037 		return ETH_ALEN;
3038 	case ARPHRD_IEEE802154:
3039 	case ARPHRD_IEEE802154_MONITOR:
3040 		return IEEE802154_EXTENDED_ADDR_LEN;
3041 	case ARPHRD_PHONET_PIPE:
3042 	case ARPHRD_PPP:
3043 	case ARPHRD_NONE:
3044 		return 0;
3045 	case ARPHRD_6LOWPAN:
3046 		return EUI64_ADDR_LEN;
3047 	case ARPHRD_FDDI:
3048 		return FDDI_K_ALEN;
3049 	case ARPHRD_HIPPI:
3050 		return HIPPI_ALEN;
3051 	case ARPHRD_IEEE802:
3052 		return FC_ALEN;
3053 	case ARPHRD_ROSE:
3054 		return ROSE_ADDR_LEN;
3055 	case ARPHRD_NETROM:
3056 		return AX25_ADDR_LEN;
3057 	case ARPHRD_LOCALTLK:
3058 		return LTALK_ALEN;
3059 	default:
3060 		return 0;
3061 	}
3062 }
3063 
3064 static long __tun_chr_ioctl(struct file *file, unsigned int cmd,
3065 			    unsigned long arg, int ifreq_len)
3066 {
3067 	struct tun_file *tfile = file->private_data;
3068 	struct net *net = sock_net(&tfile->sk);
3069 	struct tun_struct *tun;
3070 	void __user* argp = (void __user*)arg;
3071 	unsigned int ifindex, carrier;
3072 	struct ifreq ifr;
3073 	kuid_t owner;
3074 	kgid_t group;
3075 	int sndbuf;
3076 	int vnet_hdr_sz;
3077 	int le;
3078 	int ret;
3079 	bool do_notify = false;
3080 
3081 	if (cmd == TUNSETIFF || cmd == TUNSETQUEUE ||
3082 	    (_IOC_TYPE(cmd) == SOCK_IOC_TYPE && cmd != SIOCGSKNS)) {
3083 		if (copy_from_user(&ifr, argp, ifreq_len))
3084 			return -EFAULT;
3085 	} else {
3086 		memset(&ifr, 0, sizeof(ifr));
3087 	}
3088 	if (cmd == TUNGETFEATURES) {
3089 		/* Currently this just means: "what IFF flags are valid?".
3090 		 * This is needed because we never checked for invalid flags on
3091 		 * TUNSETIFF.
3092 		 */
3093 		return put_user(IFF_TUN | IFF_TAP | IFF_NO_CARRIER |
3094 				TUN_FEATURES, (unsigned int __user*)argp);
3095 	} else if (cmd == TUNSETQUEUE) {
3096 		return tun_set_queue(file, &ifr);
3097 	} else if (cmd == SIOCGSKNS) {
3098 		if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
3099 			return -EPERM;
3100 		return open_related_ns(&net->ns, get_net_ns);
3101 	}
3102 
3103 	rtnl_lock();
3104 
3105 	tun = tun_get(tfile);
3106 	if (cmd == TUNSETIFF) {
3107 		ret = -EEXIST;
3108 		if (tun)
3109 			goto unlock;
3110 
3111 		ifr.ifr_name[IFNAMSIZ-1] = '\0';
3112 
3113 		ret = tun_set_iff(net, file, &ifr);
3114 
3115 		if (ret)
3116 			goto unlock;
3117 
3118 		if (copy_to_user(argp, &ifr, ifreq_len))
3119 			ret = -EFAULT;
3120 		goto unlock;
3121 	}
3122 	if (cmd == TUNSETIFINDEX) {
3123 		ret = -EPERM;
3124 		if (tun)
3125 			goto unlock;
3126 
3127 		ret = -EFAULT;
3128 		if (copy_from_user(&ifindex, argp, sizeof(ifindex)))
3129 			goto unlock;
3130 
3131 		ret = 0;
3132 		tfile->ifindex = ifindex;
3133 		goto unlock;
3134 	}
3135 
3136 	ret = -EBADFD;
3137 	if (!tun)
3138 		goto unlock;
3139 
3140 	netif_info(tun, drv, tun->dev, "tun_chr_ioctl cmd %u\n", cmd);
3141 
3142 	net = dev_net(tun->dev);
3143 	ret = 0;
3144 	switch (cmd) {
3145 	case TUNGETIFF:
3146 		tun_get_iff(tun, &ifr);
3147 
3148 		if (tfile->detached)
3149 			ifr.ifr_flags |= IFF_DETACH_QUEUE;
3150 		if (!tfile->socket.sk->sk_filter)
3151 			ifr.ifr_flags |= IFF_NOFILTER;
3152 
3153 		if (copy_to_user(argp, &ifr, ifreq_len))
3154 			ret = -EFAULT;
3155 		break;
3156 
3157 	case TUNSETNOCSUM:
3158 		/* Disable/Enable checksum */
3159 
3160 		/* [unimplemented] */
3161 		netif_info(tun, drv, tun->dev, "ignored: set checksum %s\n",
3162 			   arg ? "disabled" : "enabled");
3163 		break;
3164 
3165 	case TUNSETPERSIST:
3166 		/* Disable/Enable persist mode. Keep an extra reference to the
3167 		 * module to prevent the module being unprobed.
3168 		 */
3169 		if (arg && !(tun->flags & IFF_PERSIST)) {
3170 			tun->flags |= IFF_PERSIST;
3171 			__module_get(THIS_MODULE);
3172 			do_notify = true;
3173 		}
3174 		if (!arg && (tun->flags & IFF_PERSIST)) {
3175 			tun->flags &= ~IFF_PERSIST;
3176 			module_put(THIS_MODULE);
3177 			do_notify = true;
3178 		}
3179 
3180 		netif_info(tun, drv, tun->dev, "persist %s\n",
3181 			   arg ? "enabled" : "disabled");
3182 		break;
3183 
3184 	case TUNSETOWNER:
3185 		/* Set owner of the device */
3186 		owner = make_kuid(current_user_ns(), arg);
3187 		if (!uid_valid(owner)) {
3188 			ret = -EINVAL;
3189 			break;
3190 		}
3191 		tun->owner = owner;
3192 		do_notify = true;
3193 		netif_info(tun, drv, tun->dev, "owner set to %u\n",
3194 			   from_kuid(&init_user_ns, tun->owner));
3195 		break;
3196 
3197 	case TUNSETGROUP:
3198 		/* Set group of the device */
3199 		group = make_kgid(current_user_ns(), arg);
3200 		if (!gid_valid(group)) {
3201 			ret = -EINVAL;
3202 			break;
3203 		}
3204 		tun->group = group;
3205 		do_notify = true;
3206 		netif_info(tun, drv, tun->dev, "group set to %u\n",
3207 			   from_kgid(&init_user_ns, tun->group));
3208 		break;
3209 
3210 	case TUNSETLINK:
3211 		/* Only allow setting the type when the interface is down */
3212 		if (tun->dev->flags & IFF_UP) {
3213 			netif_info(tun, drv, tun->dev,
3214 				   "Linktype set failed because interface is up\n");
3215 			ret = -EBUSY;
3216 		} else {
3217 			ret = call_netdevice_notifiers(NETDEV_PRE_TYPE_CHANGE,
3218 						       tun->dev);
3219 			ret = notifier_to_errno(ret);
3220 			if (ret) {
3221 				netif_info(tun, drv, tun->dev,
3222 					   "Refused to change device type\n");
3223 				break;
3224 			}
3225 			tun->dev->type = (int) arg;
3226 			tun->dev->addr_len = tun_get_addr_len(tun->dev->type);
3227 			netif_info(tun, drv, tun->dev, "linktype set to %d\n",
3228 				   tun->dev->type);
3229 			call_netdevice_notifiers(NETDEV_POST_TYPE_CHANGE,
3230 						 tun->dev);
3231 		}
3232 		break;
3233 
3234 	case TUNSETDEBUG:
3235 		tun->msg_enable = (u32)arg;
3236 		break;
3237 
3238 	case TUNSETOFFLOAD:
3239 		ret = set_offload(tun, arg);
3240 		break;
3241 
3242 	case TUNSETTXFILTER:
3243 		/* Can be set only for TAPs */
3244 		ret = -EINVAL;
3245 		if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3246 			break;
3247 		ret = update_filter(&tun->txflt, (void __user *)arg);
3248 		break;
3249 
3250 	case SIOCGIFHWADDR:
3251 		/* Get hw address */
3252 		dev_get_mac_address(&ifr.ifr_hwaddr, net, tun->dev->name);
3253 		if (copy_to_user(argp, &ifr, ifreq_len))
3254 			ret = -EFAULT;
3255 		break;
3256 
3257 	case SIOCSIFHWADDR:
3258 		/* Set hw address */
3259 		ret = dev_set_mac_address_user(tun->dev, &ifr.ifr_hwaddr, NULL);
3260 		break;
3261 
3262 	case TUNGETSNDBUF:
3263 		sndbuf = tfile->socket.sk->sk_sndbuf;
3264 		if (copy_to_user(argp, &sndbuf, sizeof(sndbuf)))
3265 			ret = -EFAULT;
3266 		break;
3267 
3268 	case TUNSETSNDBUF:
3269 		if (copy_from_user(&sndbuf, argp, sizeof(sndbuf))) {
3270 			ret = -EFAULT;
3271 			break;
3272 		}
3273 		if (sndbuf <= 0) {
3274 			ret = -EINVAL;
3275 			break;
3276 		}
3277 
3278 		tun->sndbuf = sndbuf;
3279 		tun_set_sndbuf(tun);
3280 		break;
3281 
3282 	case TUNGETVNETHDRSZ:
3283 		vnet_hdr_sz = tun->vnet_hdr_sz;
3284 		if (copy_to_user(argp, &vnet_hdr_sz, sizeof(vnet_hdr_sz)))
3285 			ret = -EFAULT;
3286 		break;
3287 
3288 	case TUNSETVNETHDRSZ:
3289 		if (copy_from_user(&vnet_hdr_sz, argp, sizeof(vnet_hdr_sz))) {
3290 			ret = -EFAULT;
3291 			break;
3292 		}
3293 		if (vnet_hdr_sz < (int)sizeof(struct virtio_net_hdr)) {
3294 			ret = -EINVAL;
3295 			break;
3296 		}
3297 
3298 		tun->vnet_hdr_sz = vnet_hdr_sz;
3299 		break;
3300 
3301 	case TUNGETVNETLE:
3302 		le = !!(tun->flags & TUN_VNET_LE);
3303 		if (put_user(le, (int __user *)argp))
3304 			ret = -EFAULT;
3305 		break;
3306 
3307 	case TUNSETVNETLE:
3308 		if (get_user(le, (int __user *)argp)) {
3309 			ret = -EFAULT;
3310 			break;
3311 		}
3312 		if (le)
3313 			tun->flags |= TUN_VNET_LE;
3314 		else
3315 			tun->flags &= ~TUN_VNET_LE;
3316 		break;
3317 
3318 	case TUNGETVNETBE:
3319 		ret = tun_get_vnet_be(tun, argp);
3320 		break;
3321 
3322 	case TUNSETVNETBE:
3323 		ret = tun_set_vnet_be(tun, argp);
3324 		break;
3325 
3326 	case TUNATTACHFILTER:
3327 		/* Can be set only for TAPs */
3328 		ret = -EINVAL;
3329 		if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3330 			break;
3331 		ret = -EFAULT;
3332 		if (copy_from_user(&tun->fprog, argp, sizeof(tun->fprog)))
3333 			break;
3334 
3335 		ret = tun_attach_filter(tun);
3336 		break;
3337 
3338 	case TUNDETACHFILTER:
3339 		/* Can be set only for TAPs */
3340 		ret = -EINVAL;
3341 		if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3342 			break;
3343 		ret = 0;
3344 		tun_detach_filter(tun, tun->numqueues);
3345 		break;
3346 
3347 	case TUNGETFILTER:
3348 		ret = -EINVAL;
3349 		if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3350 			break;
3351 		ret = -EFAULT;
3352 		if (copy_to_user(argp, &tun->fprog, sizeof(tun->fprog)))
3353 			break;
3354 		ret = 0;
3355 		break;
3356 
3357 	case TUNSETSTEERINGEBPF:
3358 		ret = tun_set_ebpf(tun, &tun->steering_prog, argp);
3359 		break;
3360 
3361 	case TUNSETFILTEREBPF:
3362 		ret = tun_set_ebpf(tun, &tun->filter_prog, argp);
3363 		break;
3364 
3365 	case TUNSETCARRIER:
3366 		ret = -EFAULT;
3367 		if (copy_from_user(&carrier, argp, sizeof(carrier)))
3368 			goto unlock;
3369 
3370 		ret = tun_net_change_carrier(tun->dev, (bool)carrier);
3371 		break;
3372 
3373 	case TUNGETDEVNETNS:
3374 		ret = -EPERM;
3375 		if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
3376 			goto unlock;
3377 		ret = open_related_ns(&net->ns, get_net_ns);
3378 		break;
3379 
3380 	default:
3381 		ret = -EINVAL;
3382 		break;
3383 	}
3384 
3385 	if (do_notify)
3386 		netdev_state_change(tun->dev);
3387 
3388 unlock:
3389 	rtnl_unlock();
3390 	if (tun)
3391 		tun_put(tun);
3392 	return ret;
3393 }
3394 
3395 static long tun_chr_ioctl(struct file *file,
3396 			  unsigned int cmd, unsigned long arg)
3397 {
3398 	return __tun_chr_ioctl(file, cmd, arg, sizeof (struct ifreq));
3399 }
3400 
3401 #ifdef CONFIG_COMPAT
3402 static long tun_chr_compat_ioctl(struct file *file,
3403 			 unsigned int cmd, unsigned long arg)
3404 {
3405 	switch (cmd) {
3406 	case TUNSETIFF:
3407 	case TUNGETIFF:
3408 	case TUNSETTXFILTER:
3409 	case TUNGETSNDBUF:
3410 	case TUNSETSNDBUF:
3411 	case SIOCGIFHWADDR:
3412 	case SIOCSIFHWADDR:
3413 		arg = (unsigned long)compat_ptr(arg);
3414 		break;
3415 	default:
3416 		arg = (compat_ulong_t)arg;
3417 		break;
3418 	}
3419 
3420 	/*
3421 	 * compat_ifreq is shorter than ifreq, so we must not access beyond
3422 	 * the end of that structure. All fields that are used in this
3423 	 * driver are compatible though, we don't need to convert the
3424 	 * contents.
3425 	 */
3426 	return __tun_chr_ioctl(file, cmd, arg, sizeof(struct compat_ifreq));
3427 }
3428 #endif /* CONFIG_COMPAT */
3429 
3430 static int tun_chr_fasync(int fd, struct file *file, int on)
3431 {
3432 	struct tun_file *tfile = file->private_data;
3433 	int ret;
3434 
3435 	if ((ret = fasync_helper(fd, file, on, &tfile->fasync)) < 0)
3436 		goto out;
3437 
3438 	if (on) {
3439 		__f_setown(file, task_pid(current), PIDTYPE_TGID, 0);
3440 		tfile->flags |= TUN_FASYNC;
3441 	} else
3442 		tfile->flags &= ~TUN_FASYNC;
3443 	ret = 0;
3444 out:
3445 	return ret;
3446 }
3447 
3448 static int tun_chr_open(struct inode *inode, struct file * file)
3449 {
3450 	struct net *net = current->nsproxy->net_ns;
3451 	struct tun_file *tfile;
3452 
3453 	tfile = (struct tun_file *)sk_alloc(net, AF_UNSPEC, GFP_KERNEL,
3454 					    &tun_proto, 0);
3455 	if (!tfile)
3456 		return -ENOMEM;
3457 	if (ptr_ring_init(&tfile->tx_ring, 0, GFP_KERNEL)) {
3458 		sk_free(&tfile->sk);
3459 		return -ENOMEM;
3460 	}
3461 
3462 	mutex_init(&tfile->napi_mutex);
3463 	RCU_INIT_POINTER(tfile->tun, NULL);
3464 	tfile->flags = 0;
3465 	tfile->ifindex = 0;
3466 
3467 	init_waitqueue_head(&tfile->socket.wq.wait);
3468 
3469 	tfile->socket.file = file;
3470 	tfile->socket.ops = &tun_socket_ops;
3471 
3472 	sock_init_data_uid(&tfile->socket, &tfile->sk, current_fsuid());
3473 
3474 	tfile->sk.sk_write_space = tun_sock_write_space;
3475 	tfile->sk.sk_sndbuf = INT_MAX;
3476 
3477 	file->private_data = tfile;
3478 	INIT_LIST_HEAD(&tfile->next);
3479 
3480 	sock_set_flag(&tfile->sk, SOCK_ZEROCOPY);
3481 
3482 	/* tun groks IOCB_NOWAIT just fine, mark it as such */
3483 	file->f_mode |= FMODE_NOWAIT;
3484 	return 0;
3485 }
3486 
3487 static int tun_chr_close(struct inode *inode, struct file *file)
3488 {
3489 	struct tun_file *tfile = file->private_data;
3490 
3491 	tun_detach(tfile, true);
3492 
3493 	return 0;
3494 }
3495 
3496 #ifdef CONFIG_PROC_FS
3497 static void tun_chr_show_fdinfo(struct seq_file *m, struct file *file)
3498 {
3499 	struct tun_file *tfile = file->private_data;
3500 	struct tun_struct *tun;
3501 	struct ifreq ifr;
3502 
3503 	memset(&ifr, 0, sizeof(ifr));
3504 
3505 	rtnl_lock();
3506 	tun = tun_get(tfile);
3507 	if (tun)
3508 		tun_get_iff(tun, &ifr);
3509 	rtnl_unlock();
3510 
3511 	if (tun)
3512 		tun_put(tun);
3513 
3514 	seq_printf(m, "iff:\t%s\n", ifr.ifr_name);
3515 }
3516 #endif
3517 
3518 static const struct file_operations tun_fops = {
3519 	.owner	= THIS_MODULE,
3520 	.llseek = no_llseek,
3521 	.read_iter  = tun_chr_read_iter,
3522 	.write_iter = tun_chr_write_iter,
3523 	.poll	= tun_chr_poll,
3524 	.unlocked_ioctl	= tun_chr_ioctl,
3525 #ifdef CONFIG_COMPAT
3526 	.compat_ioctl = tun_chr_compat_ioctl,
3527 #endif
3528 	.open	= tun_chr_open,
3529 	.release = tun_chr_close,
3530 	.fasync = tun_chr_fasync,
3531 #ifdef CONFIG_PROC_FS
3532 	.show_fdinfo = tun_chr_show_fdinfo,
3533 #endif
3534 };
3535 
3536 static struct miscdevice tun_miscdev = {
3537 	.minor = TUN_MINOR,
3538 	.name = "tun",
3539 	.nodename = "net/tun",
3540 	.fops = &tun_fops,
3541 };
3542 
3543 /* ethtool interface */
3544 
3545 static void tun_default_link_ksettings(struct net_device *dev,
3546 				       struct ethtool_link_ksettings *cmd)
3547 {
3548 	ethtool_link_ksettings_zero_link_mode(cmd, supported);
3549 	ethtool_link_ksettings_zero_link_mode(cmd, advertising);
3550 	cmd->base.speed		= SPEED_10000;
3551 	cmd->base.duplex	= DUPLEX_FULL;
3552 	cmd->base.port		= PORT_TP;
3553 	cmd->base.phy_address	= 0;
3554 	cmd->base.autoneg	= AUTONEG_DISABLE;
3555 }
3556 
3557 static int tun_get_link_ksettings(struct net_device *dev,
3558 				  struct ethtool_link_ksettings *cmd)
3559 {
3560 	struct tun_struct *tun = netdev_priv(dev);
3561 
3562 	memcpy(cmd, &tun->link_ksettings, sizeof(*cmd));
3563 	return 0;
3564 }
3565 
3566 static int tun_set_link_ksettings(struct net_device *dev,
3567 				  const struct ethtool_link_ksettings *cmd)
3568 {
3569 	struct tun_struct *tun = netdev_priv(dev);
3570 
3571 	memcpy(&tun->link_ksettings, cmd, sizeof(*cmd));
3572 	return 0;
3573 }
3574 
3575 static void tun_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
3576 {
3577 	struct tun_struct *tun = netdev_priv(dev);
3578 
3579 	strscpy(info->driver, DRV_NAME, sizeof(info->driver));
3580 	strscpy(info->version, DRV_VERSION, sizeof(info->version));
3581 
3582 	switch (tun->flags & TUN_TYPE_MASK) {
3583 	case IFF_TUN:
3584 		strscpy(info->bus_info, "tun", sizeof(info->bus_info));
3585 		break;
3586 	case IFF_TAP:
3587 		strscpy(info->bus_info, "tap", sizeof(info->bus_info));
3588 		break;
3589 	}
3590 }
3591 
3592 static u32 tun_get_msglevel(struct net_device *dev)
3593 {
3594 	struct tun_struct *tun = netdev_priv(dev);
3595 
3596 	return tun->msg_enable;
3597 }
3598 
3599 static void tun_set_msglevel(struct net_device *dev, u32 value)
3600 {
3601 	struct tun_struct *tun = netdev_priv(dev);
3602 
3603 	tun->msg_enable = value;
3604 }
3605 
3606 static int tun_get_coalesce(struct net_device *dev,
3607 			    struct ethtool_coalesce *ec,
3608 			    struct kernel_ethtool_coalesce *kernel_coal,
3609 			    struct netlink_ext_ack *extack)
3610 {
3611 	struct tun_struct *tun = netdev_priv(dev);
3612 
3613 	ec->rx_max_coalesced_frames = tun->rx_batched;
3614 
3615 	return 0;
3616 }
3617 
3618 static int tun_set_coalesce(struct net_device *dev,
3619 			    struct ethtool_coalesce *ec,
3620 			    struct kernel_ethtool_coalesce *kernel_coal,
3621 			    struct netlink_ext_ack *extack)
3622 {
3623 	struct tun_struct *tun = netdev_priv(dev);
3624 
3625 	if (ec->rx_max_coalesced_frames > NAPI_POLL_WEIGHT)
3626 		tun->rx_batched = NAPI_POLL_WEIGHT;
3627 	else
3628 		tun->rx_batched = ec->rx_max_coalesced_frames;
3629 
3630 	return 0;
3631 }
3632 
3633 static const struct ethtool_ops tun_ethtool_ops = {
3634 	.supported_coalesce_params = ETHTOOL_COALESCE_RX_MAX_FRAMES,
3635 	.get_drvinfo	= tun_get_drvinfo,
3636 	.get_msglevel	= tun_get_msglevel,
3637 	.set_msglevel	= tun_set_msglevel,
3638 	.get_link	= ethtool_op_get_link,
3639 	.get_ts_info	= ethtool_op_get_ts_info,
3640 	.get_coalesce   = tun_get_coalesce,
3641 	.set_coalesce   = tun_set_coalesce,
3642 	.get_link_ksettings = tun_get_link_ksettings,
3643 	.set_link_ksettings = tun_set_link_ksettings,
3644 };
3645 
3646 static int tun_queue_resize(struct tun_struct *tun)
3647 {
3648 	struct net_device *dev = tun->dev;
3649 	struct tun_file *tfile;
3650 	struct ptr_ring **rings;
3651 	int n = tun->numqueues + tun->numdisabled;
3652 	int ret, i;
3653 
3654 	rings = kmalloc_array(n, sizeof(*rings), GFP_KERNEL);
3655 	if (!rings)
3656 		return -ENOMEM;
3657 
3658 	for (i = 0; i < tun->numqueues; i++) {
3659 		tfile = rtnl_dereference(tun->tfiles[i]);
3660 		rings[i] = &tfile->tx_ring;
3661 	}
3662 	list_for_each_entry(tfile, &tun->disabled, next)
3663 		rings[i++] = &tfile->tx_ring;
3664 
3665 	ret = ptr_ring_resize_multiple(rings, n,
3666 				       dev->tx_queue_len, GFP_KERNEL,
3667 				       tun_ptr_free);
3668 
3669 	kfree(rings);
3670 	return ret;
3671 }
3672 
3673 static int tun_device_event(struct notifier_block *unused,
3674 			    unsigned long event, void *ptr)
3675 {
3676 	struct net_device *dev = netdev_notifier_info_to_dev(ptr);
3677 	struct tun_struct *tun = netdev_priv(dev);
3678 	int i;
3679 
3680 	if (dev->rtnl_link_ops != &tun_link_ops)
3681 		return NOTIFY_DONE;
3682 
3683 	switch (event) {
3684 	case NETDEV_CHANGE_TX_QUEUE_LEN:
3685 		if (tun_queue_resize(tun))
3686 			return NOTIFY_BAD;
3687 		break;
3688 	case NETDEV_UP:
3689 		for (i = 0; i < tun->numqueues; i++) {
3690 			struct tun_file *tfile;
3691 
3692 			tfile = rtnl_dereference(tun->tfiles[i]);
3693 			tfile->socket.sk->sk_write_space(tfile->socket.sk);
3694 		}
3695 		break;
3696 	default:
3697 		break;
3698 	}
3699 
3700 	return NOTIFY_DONE;
3701 }
3702 
3703 static struct notifier_block tun_notifier_block __read_mostly = {
3704 	.notifier_call	= tun_device_event,
3705 };
3706 
3707 static int __init tun_init(void)
3708 {
3709 	int ret = 0;
3710 
3711 	pr_info("%s, %s\n", DRV_DESCRIPTION, DRV_VERSION);
3712 
3713 	ret = rtnl_link_register(&tun_link_ops);
3714 	if (ret) {
3715 		pr_err("Can't register link_ops\n");
3716 		goto err_linkops;
3717 	}
3718 
3719 	ret = misc_register(&tun_miscdev);
3720 	if (ret) {
3721 		pr_err("Can't register misc device %d\n", TUN_MINOR);
3722 		goto err_misc;
3723 	}
3724 
3725 	ret = register_netdevice_notifier(&tun_notifier_block);
3726 	if (ret) {
3727 		pr_err("Can't register netdevice notifier\n");
3728 		goto err_notifier;
3729 	}
3730 
3731 	return  0;
3732 
3733 err_notifier:
3734 	misc_deregister(&tun_miscdev);
3735 err_misc:
3736 	rtnl_link_unregister(&tun_link_ops);
3737 err_linkops:
3738 	return ret;
3739 }
3740 
3741 static void tun_cleanup(void)
3742 {
3743 	misc_deregister(&tun_miscdev);
3744 	rtnl_link_unregister(&tun_link_ops);
3745 	unregister_netdevice_notifier(&tun_notifier_block);
3746 }
3747 
3748 /* Get an underlying socket object from tun file.  Returns error unless file is
3749  * attached to a device.  The returned object works like a packet socket, it
3750  * can be used for sock_sendmsg/sock_recvmsg.  The caller is responsible for
3751  * holding a reference to the file for as long as the socket is in use. */
3752 struct socket *tun_get_socket(struct file *file)
3753 {
3754 	struct tun_file *tfile;
3755 	if (file->f_op != &tun_fops)
3756 		return ERR_PTR(-EINVAL);
3757 	tfile = file->private_data;
3758 	if (!tfile)
3759 		return ERR_PTR(-EBADFD);
3760 	return &tfile->socket;
3761 }
3762 EXPORT_SYMBOL_GPL(tun_get_socket);
3763 
3764 struct ptr_ring *tun_get_tx_ring(struct file *file)
3765 {
3766 	struct tun_file *tfile;
3767 
3768 	if (file->f_op != &tun_fops)
3769 		return ERR_PTR(-EINVAL);
3770 	tfile = file->private_data;
3771 	if (!tfile)
3772 		return ERR_PTR(-EBADFD);
3773 	return &tfile->tx_ring;
3774 }
3775 EXPORT_SYMBOL_GPL(tun_get_tx_ring);
3776 
3777 module_init(tun_init);
3778 module_exit(tun_cleanup);
3779 MODULE_DESCRIPTION(DRV_DESCRIPTION);
3780 MODULE_AUTHOR(DRV_COPYRIGHT);
3781 MODULE_LICENSE("GPL");
3782 MODULE_ALIAS_MISCDEV(TUN_MINOR);
3783 MODULE_ALIAS("devname:net/tun");
3784