xref: /openbmc/linux/drivers/net/tun.c (revision b68fc09be48edbc47de1a0f3d42ef8adf6c0ac55)
1 /*
2  *  TUN - Universal TUN/TAP device driver.
3  *  Copyright (C) 1999-2002 Maxim Krasnyansky <maxk@qualcomm.com>
4  *
5  *  This program is free software; you can redistribute it and/or modify
6  *  it under the terms of the GNU General Public License as published by
7  *  the Free Software Foundation; either version 2 of the License, or
8  *  (at your option) any later version.
9  *
10  *  This program is distributed in the hope that it will be useful,
11  *  but WITHOUT ANY WARRANTY; without even the implied warranty of
12  *  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
13  *  GNU General Public License for more details.
14  *
15  *  $Id: tun.c,v 1.15 2002/03/01 02:44:24 maxk Exp $
16  */
17 
18 /*
19  *  Changes:
20  *
21  *  Mike Kershaw <dragorn@kismetwireless.net> 2005/08/14
22  *    Add TUNSETLINK ioctl to set the link encapsulation
23  *
24  *  Mark Smith <markzzzsmith@yahoo.com.au>
25  *    Use eth_random_addr() for tap MAC address.
26  *
27  *  Harald Roelle <harald.roelle@ifi.lmu.de>  2004/04/20
28  *    Fixes in packet dropping, queue length setting and queue wakeup.
29  *    Increased default tx queue length.
30  *    Added ethtool API.
31  *    Minor cleanups
32  *
33  *  Daniel Podlejski <underley@underley.eu.org>
34  *    Modifications for 2.3.99-pre5 kernel.
35  */
36 
37 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
38 
39 #define DRV_NAME	"tun"
40 #define DRV_VERSION	"1.6"
41 #define DRV_DESCRIPTION	"Universal TUN/TAP device driver"
42 #define DRV_COPYRIGHT	"(C) 1999-2004 Max Krasnyansky <maxk@qualcomm.com>"
43 
44 #include <linux/module.h>
45 #include <linux/errno.h>
46 #include <linux/kernel.h>
47 #include <linux/sched/signal.h>
48 #include <linux/major.h>
49 #include <linux/slab.h>
50 #include <linux/poll.h>
51 #include <linux/fcntl.h>
52 #include <linux/init.h>
53 #include <linux/skbuff.h>
54 #include <linux/netdevice.h>
55 #include <linux/etherdevice.h>
56 #include <linux/miscdevice.h>
57 #include <linux/ethtool.h>
58 #include <linux/rtnetlink.h>
59 #include <linux/compat.h>
60 #include <linux/if.h>
61 #include <linux/if_arp.h>
62 #include <linux/if_ether.h>
63 #include <linux/if_tun.h>
64 #include <linux/if_vlan.h>
65 #include <linux/crc32.h>
66 #include <linux/nsproxy.h>
67 #include <linux/virtio_net.h>
68 #include <linux/rcupdate.h>
69 #include <net/net_namespace.h>
70 #include <net/netns/generic.h>
71 #include <net/rtnetlink.h>
72 #include <net/sock.h>
73 #include <net/xdp.h>
74 #include <linux/seq_file.h>
75 #include <linux/uio.h>
76 #include <linux/skb_array.h>
77 #include <linux/bpf.h>
78 #include <linux/bpf_trace.h>
79 #include <linux/mutex.h>
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 /* Uncomment to enable debugging */
88 /* #define TUN_DEBUG 1 */
89 
90 #ifdef TUN_DEBUG
91 static int debug;
92 
93 #define tun_debug(level, tun, fmt, args...)			\
94 do {								\
95 	if (tun->debug)						\
96 		netdev_printk(level, tun->dev, fmt, ##args);	\
97 } while (0)
98 #define DBG1(level, fmt, args...)				\
99 do {								\
100 	if (debug == 2)						\
101 		printk(level fmt, ##args);			\
102 } while (0)
103 #else
104 #define tun_debug(level, tun, fmt, args...)			\
105 do {								\
106 	if (0)							\
107 		netdev_printk(level, tun->dev, fmt, ##args);	\
108 } while (0)
109 #define DBG1(level, fmt, args...)				\
110 do {								\
111 	if (0)							\
112 		printk(level fmt, ##args);			\
113 } while (0)
114 #endif
115 
116 #define TUN_RX_PAD (NET_IP_ALIGN + NET_SKB_PAD)
117 
118 /* TUN device flags */
119 
120 /* IFF_ATTACH_QUEUE is never stored in device flags,
121  * overload it to mean fasync when stored there.
122  */
123 #define TUN_FASYNC	IFF_ATTACH_QUEUE
124 /* High bits in flags field are unused. */
125 #define TUN_VNET_LE     0x80000000
126 #define TUN_VNET_BE     0x40000000
127 
128 #define TUN_FEATURES (IFF_NO_PI | IFF_ONE_QUEUE | IFF_VNET_HDR | \
129 		      IFF_MULTI_QUEUE | IFF_NAPI | IFF_NAPI_FRAGS)
130 
131 #define GOODCOPY_LEN 128
132 
133 #define FLT_EXACT_COUNT 8
134 struct tap_filter {
135 	unsigned int    count;    /* Number of addrs. Zero means disabled */
136 	u32             mask[2];  /* Mask of the hashed addrs */
137 	unsigned char	addr[FLT_EXACT_COUNT][ETH_ALEN];
138 };
139 
140 /* MAX_TAP_QUEUES 256 is chosen to allow rx/tx queues to be equal
141  * to max number of VCPUs in guest. */
142 #define MAX_TAP_QUEUES 256
143 #define MAX_TAP_FLOWS  4096
144 
145 #define TUN_FLOW_EXPIRE (3 * HZ)
146 
147 struct tun_pcpu_stats {
148 	u64 rx_packets;
149 	u64 rx_bytes;
150 	u64 tx_packets;
151 	u64 tx_bytes;
152 	struct u64_stats_sync syncp;
153 	u32 rx_dropped;
154 	u32 tx_dropped;
155 	u32 rx_frame_errors;
156 };
157 
158 /* A tun_file connects an open character device to a tuntap netdevice. It
159  * also contains all socket related structures (except sock_fprog and tap_filter)
160  * to serve as one transmit queue for tuntap device. The sock_fprog and
161  * tap_filter were kept in tun_struct since they were used for filtering for the
162  * netdevice not for a specific queue (at least I didn't see the requirement for
163  * this).
164  *
165  * RCU usage:
166  * The tun_file and tun_struct are loosely coupled, the pointer from one to the
167  * other can only be read while rcu_read_lock or rtnl_lock is held.
168  */
169 struct tun_file {
170 	struct sock sk;
171 	struct socket socket;
172 	struct socket_wq wq;
173 	struct tun_struct __rcu *tun;
174 	struct fasync_struct *fasync;
175 	/* only used for fasnyc */
176 	unsigned int flags;
177 	union {
178 		u16 queue_index;
179 		unsigned int ifindex;
180 	};
181 	struct napi_struct napi;
182 	bool napi_enabled;
183 	bool napi_frags_enabled;
184 	struct mutex napi_mutex;	/* Protects access to the above napi */
185 	struct list_head next;
186 	struct tun_struct *detached;
187 	struct ptr_ring tx_ring;
188 	struct xdp_rxq_info xdp_rxq;
189 };
190 
191 struct tun_flow_entry {
192 	struct hlist_node hash_link;
193 	struct rcu_head rcu;
194 	struct tun_struct *tun;
195 
196 	u32 rxhash;
197 	u32 rps_rxhash;
198 	int queue_index;
199 	unsigned long updated;
200 };
201 
202 #define TUN_NUM_FLOW_ENTRIES 1024
203 #define TUN_MASK_FLOW_ENTRIES (TUN_NUM_FLOW_ENTRIES - 1)
204 
205 struct tun_prog {
206 	struct rcu_head rcu;
207 	struct bpf_prog *prog;
208 };
209 
210 /* Since the socket were moved to tun_file, to preserve the behavior of persist
211  * device, socket filter, sndbuf and vnet header size were restore when the
212  * file were attached to a persist device.
213  */
214 struct tun_struct {
215 	struct tun_file __rcu	*tfiles[MAX_TAP_QUEUES];
216 	unsigned int            numqueues;
217 	unsigned int 		flags;
218 	kuid_t			owner;
219 	kgid_t			group;
220 
221 	struct net_device	*dev;
222 	netdev_features_t	set_features;
223 #define TUN_USER_FEATURES (NETIF_F_HW_CSUM|NETIF_F_TSO_ECN|NETIF_F_TSO| \
224 			  NETIF_F_TSO6)
225 
226 	int			align;
227 	int			vnet_hdr_sz;
228 	int			sndbuf;
229 	struct tap_filter	txflt;
230 	struct sock_fprog	fprog;
231 	/* protected by rtnl lock */
232 	bool			filter_attached;
233 #ifdef TUN_DEBUG
234 	int debug;
235 #endif
236 	spinlock_t lock;
237 	struct hlist_head flows[TUN_NUM_FLOW_ENTRIES];
238 	struct timer_list flow_gc_timer;
239 	unsigned long ageing_time;
240 	unsigned int numdisabled;
241 	struct list_head disabled;
242 	void *security;
243 	u32 flow_count;
244 	u32 rx_batched;
245 	struct tun_pcpu_stats __percpu *pcpu_stats;
246 	struct bpf_prog __rcu *xdp_prog;
247 	struct tun_prog __rcu *steering_prog;
248 	struct tun_prog __rcu *filter_prog;
249 	struct ethtool_link_ksettings link_ksettings;
250 };
251 
252 struct veth {
253 	__be16 h_vlan_proto;
254 	__be16 h_vlan_TCI;
255 };
256 
257 bool tun_is_xdp_frame(void *ptr)
258 {
259 	return (unsigned long)ptr & TUN_XDP_FLAG;
260 }
261 EXPORT_SYMBOL(tun_is_xdp_frame);
262 
263 void *tun_xdp_to_ptr(void *ptr)
264 {
265 	return (void *)((unsigned long)ptr | TUN_XDP_FLAG);
266 }
267 EXPORT_SYMBOL(tun_xdp_to_ptr);
268 
269 void *tun_ptr_to_xdp(void *ptr)
270 {
271 	return (void *)((unsigned long)ptr & ~TUN_XDP_FLAG);
272 }
273 EXPORT_SYMBOL(tun_ptr_to_xdp);
274 
275 static int tun_napi_receive(struct napi_struct *napi, int budget)
276 {
277 	struct tun_file *tfile = container_of(napi, struct tun_file, napi);
278 	struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
279 	struct sk_buff_head process_queue;
280 	struct sk_buff *skb;
281 	int received = 0;
282 
283 	__skb_queue_head_init(&process_queue);
284 
285 	spin_lock(&queue->lock);
286 	skb_queue_splice_tail_init(queue, &process_queue);
287 	spin_unlock(&queue->lock);
288 
289 	while (received < budget && (skb = __skb_dequeue(&process_queue))) {
290 		napi_gro_receive(napi, skb);
291 		++received;
292 	}
293 
294 	if (!skb_queue_empty(&process_queue)) {
295 		spin_lock(&queue->lock);
296 		skb_queue_splice(&process_queue, queue);
297 		spin_unlock(&queue->lock);
298 	}
299 
300 	return received;
301 }
302 
303 static int tun_napi_poll(struct napi_struct *napi, int budget)
304 {
305 	unsigned int received;
306 
307 	received = tun_napi_receive(napi, budget);
308 
309 	if (received < budget)
310 		napi_complete_done(napi, received);
311 
312 	return received;
313 }
314 
315 static void tun_napi_init(struct tun_struct *tun, struct tun_file *tfile,
316 			  bool napi_en, bool napi_frags)
317 {
318 	tfile->napi_enabled = napi_en;
319 	tfile->napi_frags_enabled = napi_en && napi_frags;
320 	if (napi_en) {
321 		netif_napi_add(tun->dev, &tfile->napi, tun_napi_poll,
322 			       NAPI_POLL_WEIGHT);
323 		napi_enable(&tfile->napi);
324 	}
325 }
326 
327 static void tun_napi_disable(struct tun_file *tfile)
328 {
329 	if (tfile->napi_enabled)
330 		napi_disable(&tfile->napi);
331 }
332 
333 static void tun_napi_del(struct tun_file *tfile)
334 {
335 	if (tfile->napi_enabled)
336 		netif_napi_del(&tfile->napi);
337 }
338 
339 static bool tun_napi_frags_enabled(const struct tun_file *tfile)
340 {
341 	return tfile->napi_frags_enabled;
342 }
343 
344 #ifdef CONFIG_TUN_VNET_CROSS_LE
345 static inline bool tun_legacy_is_little_endian(struct tun_struct *tun)
346 {
347 	return tun->flags & TUN_VNET_BE ? false :
348 		virtio_legacy_is_little_endian();
349 }
350 
351 static long tun_get_vnet_be(struct tun_struct *tun, int __user *argp)
352 {
353 	int be = !!(tun->flags & TUN_VNET_BE);
354 
355 	if (put_user(be, argp))
356 		return -EFAULT;
357 
358 	return 0;
359 }
360 
361 static long tun_set_vnet_be(struct tun_struct *tun, int __user *argp)
362 {
363 	int be;
364 
365 	if (get_user(be, argp))
366 		return -EFAULT;
367 
368 	if (be)
369 		tun->flags |= TUN_VNET_BE;
370 	else
371 		tun->flags &= ~TUN_VNET_BE;
372 
373 	return 0;
374 }
375 #else
376 static inline bool tun_legacy_is_little_endian(struct tun_struct *tun)
377 {
378 	return virtio_legacy_is_little_endian();
379 }
380 
381 static long tun_get_vnet_be(struct tun_struct *tun, int __user *argp)
382 {
383 	return -EINVAL;
384 }
385 
386 static long tun_set_vnet_be(struct tun_struct *tun, int __user *argp)
387 {
388 	return -EINVAL;
389 }
390 #endif /* CONFIG_TUN_VNET_CROSS_LE */
391 
392 static inline bool tun_is_little_endian(struct tun_struct *tun)
393 {
394 	return tun->flags & TUN_VNET_LE ||
395 		tun_legacy_is_little_endian(tun);
396 }
397 
398 static inline u16 tun16_to_cpu(struct tun_struct *tun, __virtio16 val)
399 {
400 	return __virtio16_to_cpu(tun_is_little_endian(tun), val);
401 }
402 
403 static inline __virtio16 cpu_to_tun16(struct tun_struct *tun, u16 val)
404 {
405 	return __cpu_to_virtio16(tun_is_little_endian(tun), val);
406 }
407 
408 static inline u32 tun_hashfn(u32 rxhash)
409 {
410 	return rxhash & TUN_MASK_FLOW_ENTRIES;
411 }
412 
413 static struct tun_flow_entry *tun_flow_find(struct hlist_head *head, u32 rxhash)
414 {
415 	struct tun_flow_entry *e;
416 
417 	hlist_for_each_entry_rcu(e, head, hash_link) {
418 		if (e->rxhash == rxhash)
419 			return e;
420 	}
421 	return NULL;
422 }
423 
424 static struct tun_flow_entry *tun_flow_create(struct tun_struct *tun,
425 					      struct hlist_head *head,
426 					      u32 rxhash, u16 queue_index)
427 {
428 	struct tun_flow_entry *e = kmalloc(sizeof(*e), GFP_ATOMIC);
429 
430 	if (e) {
431 		tun_debug(KERN_INFO, tun, "create flow: hash %u index %u\n",
432 			  rxhash, queue_index);
433 		e->updated = jiffies;
434 		e->rxhash = rxhash;
435 		e->rps_rxhash = 0;
436 		e->queue_index = queue_index;
437 		e->tun = tun;
438 		hlist_add_head_rcu(&e->hash_link, head);
439 		++tun->flow_count;
440 	}
441 	return e;
442 }
443 
444 static void tun_flow_delete(struct tun_struct *tun, struct tun_flow_entry *e)
445 {
446 	tun_debug(KERN_INFO, tun, "delete flow: hash %u index %u\n",
447 		  e->rxhash, e->queue_index);
448 	hlist_del_rcu(&e->hash_link);
449 	kfree_rcu(e, rcu);
450 	--tun->flow_count;
451 }
452 
453 static void tun_flow_flush(struct tun_struct *tun)
454 {
455 	int i;
456 
457 	spin_lock_bh(&tun->lock);
458 	for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
459 		struct tun_flow_entry *e;
460 		struct hlist_node *n;
461 
462 		hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link)
463 			tun_flow_delete(tun, e);
464 	}
465 	spin_unlock_bh(&tun->lock);
466 }
467 
468 static void tun_flow_delete_by_queue(struct tun_struct *tun, u16 queue_index)
469 {
470 	int i;
471 
472 	spin_lock_bh(&tun->lock);
473 	for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
474 		struct tun_flow_entry *e;
475 		struct hlist_node *n;
476 
477 		hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
478 			if (e->queue_index == queue_index)
479 				tun_flow_delete(tun, e);
480 		}
481 	}
482 	spin_unlock_bh(&tun->lock);
483 }
484 
485 static void tun_flow_cleanup(struct timer_list *t)
486 {
487 	struct tun_struct *tun = from_timer(tun, t, flow_gc_timer);
488 	unsigned long delay = tun->ageing_time;
489 	unsigned long next_timer = jiffies + delay;
490 	unsigned long count = 0;
491 	int i;
492 
493 	tun_debug(KERN_INFO, tun, "tun_flow_cleanup\n");
494 
495 	spin_lock(&tun->lock);
496 	for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
497 		struct tun_flow_entry *e;
498 		struct hlist_node *n;
499 
500 		hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
501 			unsigned long this_timer;
502 
503 			this_timer = e->updated + delay;
504 			if (time_before_eq(this_timer, jiffies)) {
505 				tun_flow_delete(tun, e);
506 				continue;
507 			}
508 			count++;
509 			if (time_before(this_timer, next_timer))
510 				next_timer = this_timer;
511 		}
512 	}
513 
514 	if (count)
515 		mod_timer(&tun->flow_gc_timer, round_jiffies_up(next_timer));
516 	spin_unlock(&tun->lock);
517 }
518 
519 static void tun_flow_update(struct tun_struct *tun, u32 rxhash,
520 			    struct tun_file *tfile)
521 {
522 	struct hlist_head *head;
523 	struct tun_flow_entry *e;
524 	unsigned long delay = tun->ageing_time;
525 	u16 queue_index = tfile->queue_index;
526 
527 	if (!rxhash)
528 		return;
529 	else
530 		head = &tun->flows[tun_hashfn(rxhash)];
531 
532 	rcu_read_lock();
533 
534 	e = tun_flow_find(head, rxhash);
535 	if (likely(e)) {
536 		/* TODO: keep queueing to old queue until it's empty? */
537 		e->queue_index = queue_index;
538 		e->updated = jiffies;
539 		sock_rps_record_flow_hash(e->rps_rxhash);
540 	} else {
541 		spin_lock_bh(&tun->lock);
542 		if (!tun_flow_find(head, rxhash) &&
543 		    tun->flow_count < MAX_TAP_FLOWS)
544 			tun_flow_create(tun, head, rxhash, queue_index);
545 
546 		if (!timer_pending(&tun->flow_gc_timer))
547 			mod_timer(&tun->flow_gc_timer,
548 				  round_jiffies_up(jiffies + delay));
549 		spin_unlock_bh(&tun->lock);
550 	}
551 
552 	rcu_read_unlock();
553 }
554 
555 /**
556  * Save the hash received in the stack receive path and update the
557  * flow_hash table accordingly.
558  */
559 static inline void tun_flow_save_rps_rxhash(struct tun_flow_entry *e, u32 hash)
560 {
561 	if (unlikely(e->rps_rxhash != hash))
562 		e->rps_rxhash = hash;
563 }
564 
565 /* We try to identify a flow through its rxhash first. The reason that
566  * we do not check rxq no. is because some cards(e.g 82599), chooses
567  * the rxq based on the txq where the last packet of the flow comes. As
568  * the userspace application move between processors, we may get a
569  * different rxq no. here. If we could not get rxhash, then we would
570  * hope the rxq no. may help here.
571  */
572 static u16 tun_automq_select_queue(struct tun_struct *tun, struct sk_buff *skb)
573 {
574 	struct tun_flow_entry *e;
575 	u32 txq = 0;
576 	u32 numqueues = 0;
577 
578 	numqueues = READ_ONCE(tun->numqueues);
579 
580 	txq = __skb_get_hash_symmetric(skb);
581 	if (txq) {
582 		e = tun_flow_find(&tun->flows[tun_hashfn(txq)], txq);
583 		if (e) {
584 			tun_flow_save_rps_rxhash(e, txq);
585 			txq = e->queue_index;
586 		} else
587 			/* use multiply and shift instead of expensive divide */
588 			txq = ((u64)txq * numqueues) >> 32;
589 	} else if (likely(skb_rx_queue_recorded(skb))) {
590 		txq = skb_get_rx_queue(skb);
591 		while (unlikely(txq >= numqueues))
592 			txq -= numqueues;
593 	}
594 
595 	return txq;
596 }
597 
598 static u16 tun_ebpf_select_queue(struct tun_struct *tun, struct sk_buff *skb)
599 {
600 	struct tun_prog *prog;
601 	u16 ret = 0;
602 
603 	prog = rcu_dereference(tun->steering_prog);
604 	if (prog)
605 		ret = bpf_prog_run_clear_cb(prog->prog, skb);
606 
607 	return ret % tun->numqueues;
608 }
609 
610 static u16 tun_select_queue(struct net_device *dev, struct sk_buff *skb,
611 			    struct net_device *sb_dev,
612 			    select_queue_fallback_t fallback)
613 {
614 	struct tun_struct *tun = netdev_priv(dev);
615 	u16 ret;
616 
617 	rcu_read_lock();
618 	if (rcu_dereference(tun->steering_prog))
619 		ret = tun_ebpf_select_queue(tun, skb);
620 	else
621 		ret = tun_automq_select_queue(tun, skb);
622 	rcu_read_unlock();
623 
624 	return ret;
625 }
626 
627 static inline bool tun_not_capable(struct tun_struct *tun)
628 {
629 	const struct cred *cred = current_cred();
630 	struct net *net = dev_net(tun->dev);
631 
632 	return ((uid_valid(tun->owner) && !uid_eq(cred->euid, tun->owner)) ||
633 		  (gid_valid(tun->group) && !in_egroup_p(tun->group))) &&
634 		!ns_capable(net->user_ns, CAP_NET_ADMIN);
635 }
636 
637 static void tun_set_real_num_queues(struct tun_struct *tun)
638 {
639 	netif_set_real_num_tx_queues(tun->dev, tun->numqueues);
640 	netif_set_real_num_rx_queues(tun->dev, tun->numqueues);
641 }
642 
643 static void tun_disable_queue(struct tun_struct *tun, struct tun_file *tfile)
644 {
645 	tfile->detached = tun;
646 	list_add_tail(&tfile->next, &tun->disabled);
647 	++tun->numdisabled;
648 }
649 
650 static struct tun_struct *tun_enable_queue(struct tun_file *tfile)
651 {
652 	struct tun_struct *tun = tfile->detached;
653 
654 	tfile->detached = NULL;
655 	list_del_init(&tfile->next);
656 	--tun->numdisabled;
657 	return tun;
658 }
659 
660 void tun_ptr_free(void *ptr)
661 {
662 	if (!ptr)
663 		return;
664 	if (tun_is_xdp_frame(ptr)) {
665 		struct xdp_frame *xdpf = tun_ptr_to_xdp(ptr);
666 
667 		xdp_return_frame(xdpf);
668 	} else {
669 		__skb_array_destroy_skb(ptr);
670 	}
671 }
672 EXPORT_SYMBOL_GPL(tun_ptr_free);
673 
674 static void tun_queue_purge(struct tun_file *tfile)
675 {
676 	void *ptr;
677 
678 	while ((ptr = ptr_ring_consume(&tfile->tx_ring)) != NULL)
679 		tun_ptr_free(ptr);
680 
681 	skb_queue_purge(&tfile->sk.sk_write_queue);
682 	skb_queue_purge(&tfile->sk.sk_error_queue);
683 }
684 
685 static void __tun_detach(struct tun_file *tfile, bool clean)
686 {
687 	struct tun_file *ntfile;
688 	struct tun_struct *tun;
689 
690 	tun = rtnl_dereference(tfile->tun);
691 
692 	if (tun && clean) {
693 		tun_napi_disable(tfile);
694 		tun_napi_del(tfile);
695 	}
696 
697 	if (tun && !tfile->detached) {
698 		u16 index = tfile->queue_index;
699 		BUG_ON(index >= tun->numqueues);
700 
701 		rcu_assign_pointer(tun->tfiles[index],
702 				   tun->tfiles[tun->numqueues - 1]);
703 		ntfile = rtnl_dereference(tun->tfiles[index]);
704 		ntfile->queue_index = index;
705 
706 		--tun->numqueues;
707 		if (clean) {
708 			RCU_INIT_POINTER(tfile->tun, NULL);
709 			sock_put(&tfile->sk);
710 		} else
711 			tun_disable_queue(tun, tfile);
712 
713 		synchronize_net();
714 		tun_flow_delete_by_queue(tun, tun->numqueues + 1);
715 		/* Drop read queue */
716 		tun_queue_purge(tfile);
717 		tun_set_real_num_queues(tun);
718 	} else if (tfile->detached && clean) {
719 		tun = tun_enable_queue(tfile);
720 		sock_put(&tfile->sk);
721 	}
722 
723 	if (clean) {
724 		if (tun && tun->numqueues == 0 && tun->numdisabled == 0) {
725 			netif_carrier_off(tun->dev);
726 
727 			if (!(tun->flags & IFF_PERSIST) &&
728 			    tun->dev->reg_state == NETREG_REGISTERED)
729 				unregister_netdevice(tun->dev);
730 		}
731 		if (tun)
732 			xdp_rxq_info_unreg(&tfile->xdp_rxq);
733 		ptr_ring_cleanup(&tfile->tx_ring, tun_ptr_free);
734 		sock_put(&tfile->sk);
735 	}
736 }
737 
738 static void tun_detach(struct tun_file *tfile, bool clean)
739 {
740 	struct tun_struct *tun;
741 	struct net_device *dev;
742 
743 	rtnl_lock();
744 	tun = rtnl_dereference(tfile->tun);
745 	dev = tun ? tun->dev : NULL;
746 	__tun_detach(tfile, clean);
747 	if (dev)
748 		netdev_state_change(dev);
749 	rtnl_unlock();
750 }
751 
752 static void tun_detach_all(struct net_device *dev)
753 {
754 	struct tun_struct *tun = netdev_priv(dev);
755 	struct tun_file *tfile, *tmp;
756 	int i, n = tun->numqueues;
757 
758 	for (i = 0; i < n; i++) {
759 		tfile = rtnl_dereference(tun->tfiles[i]);
760 		BUG_ON(!tfile);
761 		tun_napi_disable(tfile);
762 		tfile->socket.sk->sk_shutdown = RCV_SHUTDOWN;
763 		tfile->socket.sk->sk_data_ready(tfile->socket.sk);
764 		RCU_INIT_POINTER(tfile->tun, NULL);
765 		--tun->numqueues;
766 	}
767 	list_for_each_entry(tfile, &tun->disabled, next) {
768 		tfile->socket.sk->sk_shutdown = RCV_SHUTDOWN;
769 		tfile->socket.sk->sk_data_ready(tfile->socket.sk);
770 		RCU_INIT_POINTER(tfile->tun, NULL);
771 	}
772 	BUG_ON(tun->numqueues != 0);
773 
774 	synchronize_net();
775 	for (i = 0; i < n; i++) {
776 		tfile = rtnl_dereference(tun->tfiles[i]);
777 		tun_napi_del(tfile);
778 		/* Drop read queue */
779 		tun_queue_purge(tfile);
780 		xdp_rxq_info_unreg(&tfile->xdp_rxq);
781 		sock_put(&tfile->sk);
782 	}
783 	list_for_each_entry_safe(tfile, tmp, &tun->disabled, next) {
784 		tun_enable_queue(tfile);
785 		tun_queue_purge(tfile);
786 		xdp_rxq_info_unreg(&tfile->xdp_rxq);
787 		sock_put(&tfile->sk);
788 	}
789 	BUG_ON(tun->numdisabled != 0);
790 
791 	if (tun->flags & IFF_PERSIST)
792 		module_put(THIS_MODULE);
793 }
794 
795 static int tun_attach(struct tun_struct *tun, struct file *file,
796 		      bool skip_filter, bool napi, bool napi_frags)
797 {
798 	struct tun_file *tfile = file->private_data;
799 	struct net_device *dev = tun->dev;
800 	int err;
801 
802 	err = security_tun_dev_attach(tfile->socket.sk, tun->security);
803 	if (err < 0)
804 		goto out;
805 
806 	err = -EINVAL;
807 	if (rtnl_dereference(tfile->tun) && !tfile->detached)
808 		goto out;
809 
810 	err = -EBUSY;
811 	if (!(tun->flags & IFF_MULTI_QUEUE) && tun->numqueues == 1)
812 		goto out;
813 
814 	err = -E2BIG;
815 	if (!tfile->detached &&
816 	    tun->numqueues + tun->numdisabled == MAX_TAP_QUEUES)
817 		goto out;
818 
819 	err = 0;
820 
821 	/* Re-attach the filter to persist device */
822 	if (!skip_filter && (tun->filter_attached == true)) {
823 		lock_sock(tfile->socket.sk);
824 		err = sk_attach_filter(&tun->fprog, tfile->socket.sk);
825 		release_sock(tfile->socket.sk);
826 		if (!err)
827 			goto out;
828 	}
829 
830 	if (!tfile->detached &&
831 	    ptr_ring_resize(&tfile->tx_ring, dev->tx_queue_len,
832 			    GFP_KERNEL, tun_ptr_free)) {
833 		err = -ENOMEM;
834 		goto out;
835 	}
836 
837 	tfile->queue_index = tun->numqueues;
838 	tfile->socket.sk->sk_shutdown &= ~RCV_SHUTDOWN;
839 
840 	if (tfile->detached) {
841 		/* Re-attach detached tfile, updating XDP queue_index */
842 		WARN_ON(!xdp_rxq_info_is_reg(&tfile->xdp_rxq));
843 
844 		if (tfile->xdp_rxq.queue_index    != tfile->queue_index)
845 			tfile->xdp_rxq.queue_index = tfile->queue_index;
846 	} else {
847 		/* Setup XDP RX-queue info, for new tfile getting attached */
848 		err = xdp_rxq_info_reg(&tfile->xdp_rxq,
849 				       tun->dev, tfile->queue_index);
850 		if (err < 0)
851 			goto out;
852 		err = xdp_rxq_info_reg_mem_model(&tfile->xdp_rxq,
853 						 MEM_TYPE_PAGE_SHARED, NULL);
854 		if (err < 0) {
855 			xdp_rxq_info_unreg(&tfile->xdp_rxq);
856 			goto out;
857 		}
858 		err = 0;
859 	}
860 
861 	rcu_assign_pointer(tfile->tun, tun);
862 	rcu_assign_pointer(tun->tfiles[tun->numqueues], tfile);
863 	tun->numqueues++;
864 
865 	if (tfile->detached) {
866 		tun_enable_queue(tfile);
867 	} else {
868 		sock_hold(&tfile->sk);
869 		tun_napi_init(tun, tfile, napi, napi_frags);
870 	}
871 
872 	if (rtnl_dereference(tun->xdp_prog))
873 		sock_set_flag(&tfile->sk, SOCK_XDP);
874 
875 	tun_set_real_num_queues(tun);
876 
877 	/* device is allowed to go away first, so no need to hold extra
878 	 * refcnt.
879 	 */
880 
881 out:
882 	return err;
883 }
884 
885 static struct tun_struct *tun_get(struct tun_file *tfile)
886 {
887 	struct tun_struct *tun;
888 
889 	rcu_read_lock();
890 	tun = rcu_dereference(tfile->tun);
891 	if (tun)
892 		dev_hold(tun->dev);
893 	rcu_read_unlock();
894 
895 	return tun;
896 }
897 
898 static void tun_put(struct tun_struct *tun)
899 {
900 	dev_put(tun->dev);
901 }
902 
903 /* TAP filtering */
904 static void addr_hash_set(u32 *mask, const u8 *addr)
905 {
906 	int n = ether_crc(ETH_ALEN, addr) >> 26;
907 	mask[n >> 5] |= (1 << (n & 31));
908 }
909 
910 static unsigned int addr_hash_test(const u32 *mask, const u8 *addr)
911 {
912 	int n = ether_crc(ETH_ALEN, addr) >> 26;
913 	return mask[n >> 5] & (1 << (n & 31));
914 }
915 
916 static int update_filter(struct tap_filter *filter, void __user *arg)
917 {
918 	struct { u8 u[ETH_ALEN]; } *addr;
919 	struct tun_filter uf;
920 	int err, alen, n, nexact;
921 
922 	if (copy_from_user(&uf, arg, sizeof(uf)))
923 		return -EFAULT;
924 
925 	if (!uf.count) {
926 		/* Disabled */
927 		filter->count = 0;
928 		return 0;
929 	}
930 
931 	alen = ETH_ALEN * uf.count;
932 	addr = memdup_user(arg + sizeof(uf), alen);
933 	if (IS_ERR(addr))
934 		return PTR_ERR(addr);
935 
936 	/* The filter is updated without holding any locks. Which is
937 	 * perfectly safe. We disable it first and in the worst
938 	 * case we'll accept a few undesired packets. */
939 	filter->count = 0;
940 	wmb();
941 
942 	/* Use first set of addresses as an exact filter */
943 	for (n = 0; n < uf.count && n < FLT_EXACT_COUNT; n++)
944 		memcpy(filter->addr[n], addr[n].u, ETH_ALEN);
945 
946 	nexact = n;
947 
948 	/* Remaining multicast addresses are hashed,
949 	 * unicast will leave the filter disabled. */
950 	memset(filter->mask, 0, sizeof(filter->mask));
951 	for (; n < uf.count; n++) {
952 		if (!is_multicast_ether_addr(addr[n].u)) {
953 			err = 0; /* no filter */
954 			goto free_addr;
955 		}
956 		addr_hash_set(filter->mask, addr[n].u);
957 	}
958 
959 	/* For ALLMULTI just set the mask to all ones.
960 	 * This overrides the mask populated above. */
961 	if ((uf.flags & TUN_FLT_ALLMULTI))
962 		memset(filter->mask, ~0, sizeof(filter->mask));
963 
964 	/* Now enable the filter */
965 	wmb();
966 	filter->count = nexact;
967 
968 	/* Return the number of exact filters */
969 	err = nexact;
970 free_addr:
971 	kfree(addr);
972 	return err;
973 }
974 
975 /* Returns: 0 - drop, !=0 - accept */
976 static int run_filter(struct tap_filter *filter, const struct sk_buff *skb)
977 {
978 	/* Cannot use eth_hdr(skb) here because skb_mac_hdr() is incorrect
979 	 * at this point. */
980 	struct ethhdr *eh = (struct ethhdr *) skb->data;
981 	int i;
982 
983 	/* Exact match */
984 	for (i = 0; i < filter->count; i++)
985 		if (ether_addr_equal(eh->h_dest, filter->addr[i]))
986 			return 1;
987 
988 	/* Inexact match (multicast only) */
989 	if (is_multicast_ether_addr(eh->h_dest))
990 		return addr_hash_test(filter->mask, eh->h_dest);
991 
992 	return 0;
993 }
994 
995 /*
996  * Checks whether the packet is accepted or not.
997  * Returns: 0 - drop, !=0 - accept
998  */
999 static int check_filter(struct tap_filter *filter, const struct sk_buff *skb)
1000 {
1001 	if (!filter->count)
1002 		return 1;
1003 
1004 	return run_filter(filter, skb);
1005 }
1006 
1007 /* Network device part of the driver */
1008 
1009 static const struct ethtool_ops tun_ethtool_ops;
1010 
1011 /* Net device detach from fd. */
1012 static void tun_net_uninit(struct net_device *dev)
1013 {
1014 	tun_detach_all(dev);
1015 }
1016 
1017 /* Net device open. */
1018 static int tun_net_open(struct net_device *dev)
1019 {
1020 	struct tun_struct *tun = netdev_priv(dev);
1021 	int i;
1022 
1023 	netif_tx_start_all_queues(dev);
1024 
1025 	for (i = 0; i < tun->numqueues; i++) {
1026 		struct tun_file *tfile;
1027 
1028 		tfile = rtnl_dereference(tun->tfiles[i]);
1029 		tfile->socket.sk->sk_write_space(tfile->socket.sk);
1030 	}
1031 
1032 	return 0;
1033 }
1034 
1035 /* Net device close. */
1036 static int tun_net_close(struct net_device *dev)
1037 {
1038 	netif_tx_stop_all_queues(dev);
1039 	return 0;
1040 }
1041 
1042 /* Net device start xmit */
1043 static void tun_automq_xmit(struct tun_struct *tun, struct sk_buff *skb)
1044 {
1045 #ifdef CONFIG_RPS
1046 	if (tun->numqueues == 1 && static_key_false(&rps_needed)) {
1047 		/* Select queue was not called for the skbuff, so we extract the
1048 		 * RPS hash and save it into the flow_table here.
1049 		 */
1050 		__u32 rxhash;
1051 
1052 		rxhash = __skb_get_hash_symmetric(skb);
1053 		if (rxhash) {
1054 			struct tun_flow_entry *e;
1055 			e = tun_flow_find(&tun->flows[tun_hashfn(rxhash)],
1056 					rxhash);
1057 			if (e)
1058 				tun_flow_save_rps_rxhash(e, rxhash);
1059 		}
1060 	}
1061 #endif
1062 }
1063 
1064 static unsigned int run_ebpf_filter(struct tun_struct *tun,
1065 				    struct sk_buff *skb,
1066 				    int len)
1067 {
1068 	struct tun_prog *prog = rcu_dereference(tun->filter_prog);
1069 
1070 	if (prog)
1071 		len = bpf_prog_run_clear_cb(prog->prog, skb);
1072 
1073 	return len;
1074 }
1075 
1076 /* Net device start xmit */
1077 static netdev_tx_t tun_net_xmit(struct sk_buff *skb, struct net_device *dev)
1078 {
1079 	struct tun_struct *tun = netdev_priv(dev);
1080 	int txq = skb->queue_mapping;
1081 	struct tun_file *tfile;
1082 	int len = skb->len;
1083 
1084 	rcu_read_lock();
1085 	tfile = rcu_dereference(tun->tfiles[txq]);
1086 
1087 	/* Drop packet if interface is not attached */
1088 	if (txq >= tun->numqueues)
1089 		goto drop;
1090 
1091 	if (!rcu_dereference(tun->steering_prog))
1092 		tun_automq_xmit(tun, skb);
1093 
1094 	tun_debug(KERN_INFO, tun, "tun_net_xmit %d\n", skb->len);
1095 
1096 	BUG_ON(!tfile);
1097 
1098 	/* Drop if the filter does not like it.
1099 	 * This is a noop if the filter is disabled.
1100 	 * Filter can be enabled only for the TAP devices. */
1101 	if (!check_filter(&tun->txflt, skb))
1102 		goto drop;
1103 
1104 	if (tfile->socket.sk->sk_filter &&
1105 	    sk_filter(tfile->socket.sk, skb))
1106 		goto drop;
1107 
1108 	len = run_ebpf_filter(tun, skb, len);
1109 	if (len == 0 || pskb_trim(skb, len))
1110 		goto drop;
1111 
1112 	if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC)))
1113 		goto drop;
1114 
1115 	skb_tx_timestamp(skb);
1116 
1117 	/* Orphan the skb - required as we might hang on to it
1118 	 * for indefinite time.
1119 	 */
1120 	skb_orphan(skb);
1121 
1122 	nf_reset(skb);
1123 
1124 	if (ptr_ring_produce(&tfile->tx_ring, skb))
1125 		goto drop;
1126 
1127 	/* Notify and wake up reader process */
1128 	if (tfile->flags & TUN_FASYNC)
1129 		kill_fasync(&tfile->fasync, SIGIO, POLL_IN);
1130 	tfile->socket.sk->sk_data_ready(tfile->socket.sk);
1131 
1132 	rcu_read_unlock();
1133 	return NETDEV_TX_OK;
1134 
1135 drop:
1136 	this_cpu_inc(tun->pcpu_stats->tx_dropped);
1137 	skb_tx_error(skb);
1138 	kfree_skb(skb);
1139 	rcu_read_unlock();
1140 	return NET_XMIT_DROP;
1141 }
1142 
1143 static void tun_net_mclist(struct net_device *dev)
1144 {
1145 	/*
1146 	 * This callback is supposed to deal with mc filter in
1147 	 * _rx_ path and has nothing to do with the _tx_ path.
1148 	 * In rx path we always accept everything userspace gives us.
1149 	 */
1150 }
1151 
1152 static netdev_features_t tun_net_fix_features(struct net_device *dev,
1153 	netdev_features_t features)
1154 {
1155 	struct tun_struct *tun = netdev_priv(dev);
1156 
1157 	return (features & tun->set_features) | (features & ~TUN_USER_FEATURES);
1158 }
1159 
1160 static void tun_set_headroom(struct net_device *dev, int new_hr)
1161 {
1162 	struct tun_struct *tun = netdev_priv(dev);
1163 
1164 	if (new_hr < NET_SKB_PAD)
1165 		new_hr = NET_SKB_PAD;
1166 
1167 	tun->align = new_hr;
1168 }
1169 
1170 static void
1171 tun_net_get_stats64(struct net_device *dev, struct rtnl_link_stats64 *stats)
1172 {
1173 	u32 rx_dropped = 0, tx_dropped = 0, rx_frame_errors = 0;
1174 	struct tun_struct *tun = netdev_priv(dev);
1175 	struct tun_pcpu_stats *p;
1176 	int i;
1177 
1178 	for_each_possible_cpu(i) {
1179 		u64 rxpackets, rxbytes, txpackets, txbytes;
1180 		unsigned int start;
1181 
1182 		p = per_cpu_ptr(tun->pcpu_stats, i);
1183 		do {
1184 			start = u64_stats_fetch_begin(&p->syncp);
1185 			rxpackets	= p->rx_packets;
1186 			rxbytes		= p->rx_bytes;
1187 			txpackets	= p->tx_packets;
1188 			txbytes		= p->tx_bytes;
1189 		} while (u64_stats_fetch_retry(&p->syncp, start));
1190 
1191 		stats->rx_packets	+= rxpackets;
1192 		stats->rx_bytes		+= rxbytes;
1193 		stats->tx_packets	+= txpackets;
1194 		stats->tx_bytes		+= txbytes;
1195 
1196 		/* u32 counters */
1197 		rx_dropped	+= p->rx_dropped;
1198 		rx_frame_errors	+= p->rx_frame_errors;
1199 		tx_dropped	+= p->tx_dropped;
1200 	}
1201 	stats->rx_dropped  = rx_dropped;
1202 	stats->rx_frame_errors = rx_frame_errors;
1203 	stats->tx_dropped = tx_dropped;
1204 }
1205 
1206 static int tun_xdp_set(struct net_device *dev, struct bpf_prog *prog,
1207 		       struct netlink_ext_ack *extack)
1208 {
1209 	struct tun_struct *tun = netdev_priv(dev);
1210 	struct tun_file *tfile;
1211 	struct bpf_prog *old_prog;
1212 	int i;
1213 
1214 	old_prog = rtnl_dereference(tun->xdp_prog);
1215 	rcu_assign_pointer(tun->xdp_prog, prog);
1216 	if (old_prog)
1217 		bpf_prog_put(old_prog);
1218 
1219 	for (i = 0; i < tun->numqueues; i++) {
1220 		tfile = rtnl_dereference(tun->tfiles[i]);
1221 		if (prog)
1222 			sock_set_flag(&tfile->sk, SOCK_XDP);
1223 		else
1224 			sock_reset_flag(&tfile->sk, SOCK_XDP);
1225 	}
1226 	list_for_each_entry(tfile, &tun->disabled, next) {
1227 		if (prog)
1228 			sock_set_flag(&tfile->sk, SOCK_XDP);
1229 		else
1230 			sock_reset_flag(&tfile->sk, SOCK_XDP);
1231 	}
1232 
1233 	return 0;
1234 }
1235 
1236 static u32 tun_xdp_query(struct net_device *dev)
1237 {
1238 	struct tun_struct *tun = netdev_priv(dev);
1239 	const struct bpf_prog *xdp_prog;
1240 
1241 	xdp_prog = rtnl_dereference(tun->xdp_prog);
1242 	if (xdp_prog)
1243 		return xdp_prog->aux->id;
1244 
1245 	return 0;
1246 }
1247 
1248 static int tun_xdp(struct net_device *dev, struct netdev_bpf *xdp)
1249 {
1250 	switch (xdp->command) {
1251 	case XDP_SETUP_PROG:
1252 		return tun_xdp_set(dev, xdp->prog, xdp->extack);
1253 	case XDP_QUERY_PROG:
1254 		xdp->prog_id = tun_xdp_query(dev);
1255 		return 0;
1256 	default:
1257 		return -EINVAL;
1258 	}
1259 }
1260 
1261 static const struct net_device_ops tun_netdev_ops = {
1262 	.ndo_uninit		= tun_net_uninit,
1263 	.ndo_open		= tun_net_open,
1264 	.ndo_stop		= tun_net_close,
1265 	.ndo_start_xmit		= tun_net_xmit,
1266 	.ndo_fix_features	= tun_net_fix_features,
1267 	.ndo_select_queue	= tun_select_queue,
1268 	.ndo_set_rx_headroom	= tun_set_headroom,
1269 	.ndo_get_stats64	= tun_net_get_stats64,
1270 };
1271 
1272 static void __tun_xdp_flush_tfile(struct tun_file *tfile)
1273 {
1274 	/* Notify and wake up reader process */
1275 	if (tfile->flags & TUN_FASYNC)
1276 		kill_fasync(&tfile->fasync, SIGIO, POLL_IN);
1277 	tfile->socket.sk->sk_data_ready(tfile->socket.sk);
1278 }
1279 
1280 static int tun_xdp_xmit(struct net_device *dev, int n,
1281 			struct xdp_frame **frames, u32 flags)
1282 {
1283 	struct tun_struct *tun = netdev_priv(dev);
1284 	struct tun_file *tfile;
1285 	u32 numqueues;
1286 	int drops = 0;
1287 	int cnt = n;
1288 	int i;
1289 
1290 	if (unlikely(flags & ~XDP_XMIT_FLAGS_MASK))
1291 		return -EINVAL;
1292 
1293 	rcu_read_lock();
1294 
1295 	numqueues = READ_ONCE(tun->numqueues);
1296 	if (!numqueues) {
1297 		rcu_read_unlock();
1298 		return -ENXIO; /* Caller will free/return all frames */
1299 	}
1300 
1301 	tfile = rcu_dereference(tun->tfiles[smp_processor_id() %
1302 					    numqueues]);
1303 
1304 	spin_lock(&tfile->tx_ring.producer_lock);
1305 	for (i = 0; i < n; i++) {
1306 		struct xdp_frame *xdp = frames[i];
1307 		/* Encode the XDP flag into lowest bit for consumer to differ
1308 		 * XDP buffer from sk_buff.
1309 		 */
1310 		void *frame = tun_xdp_to_ptr(xdp);
1311 
1312 		if (__ptr_ring_produce(&tfile->tx_ring, frame)) {
1313 			this_cpu_inc(tun->pcpu_stats->tx_dropped);
1314 			xdp_return_frame_rx_napi(xdp);
1315 			drops++;
1316 		}
1317 	}
1318 	spin_unlock(&tfile->tx_ring.producer_lock);
1319 
1320 	if (flags & XDP_XMIT_FLUSH)
1321 		__tun_xdp_flush_tfile(tfile);
1322 
1323 	rcu_read_unlock();
1324 	return cnt - drops;
1325 }
1326 
1327 static int tun_xdp_tx(struct net_device *dev, struct xdp_buff *xdp)
1328 {
1329 	struct xdp_frame *frame = convert_to_xdp_frame(xdp);
1330 
1331 	if (unlikely(!frame))
1332 		return -EOVERFLOW;
1333 
1334 	return tun_xdp_xmit(dev, 1, &frame, XDP_XMIT_FLUSH);
1335 }
1336 
1337 static const struct net_device_ops tap_netdev_ops = {
1338 	.ndo_uninit		= tun_net_uninit,
1339 	.ndo_open		= tun_net_open,
1340 	.ndo_stop		= tun_net_close,
1341 	.ndo_start_xmit		= tun_net_xmit,
1342 	.ndo_fix_features	= tun_net_fix_features,
1343 	.ndo_set_rx_mode	= tun_net_mclist,
1344 	.ndo_set_mac_address	= eth_mac_addr,
1345 	.ndo_validate_addr	= eth_validate_addr,
1346 	.ndo_select_queue	= tun_select_queue,
1347 	.ndo_features_check	= passthru_features_check,
1348 	.ndo_set_rx_headroom	= tun_set_headroom,
1349 	.ndo_get_stats64	= tun_net_get_stats64,
1350 	.ndo_bpf		= tun_xdp,
1351 	.ndo_xdp_xmit		= tun_xdp_xmit,
1352 };
1353 
1354 static void tun_flow_init(struct tun_struct *tun)
1355 {
1356 	int i;
1357 
1358 	for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++)
1359 		INIT_HLIST_HEAD(&tun->flows[i]);
1360 
1361 	tun->ageing_time = TUN_FLOW_EXPIRE;
1362 	timer_setup(&tun->flow_gc_timer, tun_flow_cleanup, 0);
1363 	mod_timer(&tun->flow_gc_timer,
1364 		  round_jiffies_up(jiffies + tun->ageing_time));
1365 }
1366 
1367 static void tun_flow_uninit(struct tun_struct *tun)
1368 {
1369 	del_timer_sync(&tun->flow_gc_timer);
1370 	tun_flow_flush(tun);
1371 }
1372 
1373 #define MIN_MTU 68
1374 #define MAX_MTU 65535
1375 
1376 /* Initialize net device. */
1377 static void tun_net_init(struct net_device *dev)
1378 {
1379 	struct tun_struct *tun = netdev_priv(dev);
1380 
1381 	switch (tun->flags & TUN_TYPE_MASK) {
1382 	case IFF_TUN:
1383 		dev->netdev_ops = &tun_netdev_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 		break;
1405 	}
1406 
1407 	dev->min_mtu = MIN_MTU;
1408 	dev->max_mtu = MAX_MTU - dev->hard_header_len;
1409 }
1410 
1411 static bool tun_sock_writeable(struct tun_struct *tun, struct tun_file *tfile)
1412 {
1413 	struct sock *sk = tfile->socket.sk;
1414 
1415 	return (tun->dev->flags & IFF_UP) && sock_writeable(sk);
1416 }
1417 
1418 /* Character device part */
1419 
1420 /* Poll */
1421 static __poll_t tun_chr_poll(struct file *file, poll_table *wait)
1422 {
1423 	struct tun_file *tfile = file->private_data;
1424 	struct tun_struct *tun = tun_get(tfile);
1425 	struct sock *sk;
1426 	__poll_t mask = 0;
1427 
1428 	if (!tun)
1429 		return EPOLLERR;
1430 
1431 	sk = tfile->socket.sk;
1432 
1433 	tun_debug(KERN_INFO, tun, "tun_chr_poll\n");
1434 
1435 	poll_wait(file, sk_sleep(sk), wait);
1436 
1437 	if (!ptr_ring_empty(&tfile->tx_ring))
1438 		mask |= EPOLLIN | EPOLLRDNORM;
1439 
1440 	/* Make sure SOCKWQ_ASYNC_NOSPACE is set if not writable to
1441 	 * guarantee EPOLLOUT to be raised by either here or
1442 	 * tun_sock_write_space(). Then process could get notification
1443 	 * after it writes to a down device and meets -EIO.
1444 	 */
1445 	if (tun_sock_writeable(tun, tfile) ||
1446 	    (!test_and_set_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags) &&
1447 	     tun_sock_writeable(tun, tfile)))
1448 		mask |= EPOLLOUT | EPOLLWRNORM;
1449 
1450 	if (tun->dev->reg_state != NETREG_REGISTERED)
1451 		mask = EPOLLERR;
1452 
1453 	tun_put(tun);
1454 	return mask;
1455 }
1456 
1457 static struct sk_buff *tun_napi_alloc_frags(struct tun_file *tfile,
1458 					    size_t len,
1459 					    const struct iov_iter *it)
1460 {
1461 	struct sk_buff *skb;
1462 	size_t linear;
1463 	int err;
1464 	int i;
1465 
1466 	if (it->nr_segs > MAX_SKB_FRAGS + 1)
1467 		return ERR_PTR(-ENOMEM);
1468 
1469 	local_bh_disable();
1470 	skb = napi_get_frags(&tfile->napi);
1471 	local_bh_enable();
1472 	if (!skb)
1473 		return ERR_PTR(-ENOMEM);
1474 
1475 	linear = iov_iter_single_seg_count(it);
1476 	err = __skb_grow(skb, linear);
1477 	if (err)
1478 		goto free;
1479 
1480 	skb->len = len;
1481 	skb->data_len = len - linear;
1482 	skb->truesize += skb->data_len;
1483 
1484 	for (i = 1; i < it->nr_segs; i++) {
1485 		struct page_frag *pfrag = &current->task_frag;
1486 		size_t fragsz = it->iov[i].iov_len;
1487 
1488 		if (fragsz == 0 || fragsz > PAGE_SIZE) {
1489 			err = -EINVAL;
1490 			goto free;
1491 		}
1492 
1493 		if (!skb_page_frag_refill(fragsz, pfrag, GFP_KERNEL)) {
1494 			err = -ENOMEM;
1495 			goto free;
1496 		}
1497 
1498 		skb_fill_page_desc(skb, i - 1, pfrag->page,
1499 				   pfrag->offset, fragsz);
1500 		page_ref_inc(pfrag->page);
1501 		pfrag->offset += fragsz;
1502 	}
1503 
1504 	return skb;
1505 free:
1506 	/* frees skb and all frags allocated with napi_alloc_frag() */
1507 	napi_free_frags(&tfile->napi);
1508 	return ERR_PTR(err);
1509 }
1510 
1511 /* prepad is the amount to reserve at front.  len is length after that.
1512  * linear is a hint as to how much to copy (usually headers). */
1513 static struct sk_buff *tun_alloc_skb(struct tun_file *tfile,
1514 				     size_t prepad, size_t len,
1515 				     size_t linear, int noblock)
1516 {
1517 	struct sock *sk = tfile->socket.sk;
1518 	struct sk_buff *skb;
1519 	int err;
1520 
1521 	/* Under a page?  Don't bother with paged skb. */
1522 	if (prepad + len < PAGE_SIZE || !linear)
1523 		linear = len;
1524 
1525 	skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
1526 				   &err, 0);
1527 	if (!skb)
1528 		return ERR_PTR(err);
1529 
1530 	skb_reserve(skb, prepad);
1531 	skb_put(skb, linear);
1532 	skb->data_len = len - linear;
1533 	skb->len += len - linear;
1534 
1535 	return skb;
1536 }
1537 
1538 static void tun_rx_batched(struct tun_struct *tun, struct tun_file *tfile,
1539 			   struct sk_buff *skb, int more)
1540 {
1541 	struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
1542 	struct sk_buff_head process_queue;
1543 	u32 rx_batched = tun->rx_batched;
1544 	bool rcv = false;
1545 
1546 	if (!rx_batched || (!more && skb_queue_empty(queue))) {
1547 		local_bh_disable();
1548 		netif_receive_skb(skb);
1549 		local_bh_enable();
1550 		return;
1551 	}
1552 
1553 	spin_lock(&queue->lock);
1554 	if (!more || skb_queue_len(queue) == rx_batched) {
1555 		__skb_queue_head_init(&process_queue);
1556 		skb_queue_splice_tail_init(queue, &process_queue);
1557 		rcv = true;
1558 	} else {
1559 		__skb_queue_tail(queue, skb);
1560 	}
1561 	spin_unlock(&queue->lock);
1562 
1563 	if (rcv) {
1564 		struct sk_buff *nskb;
1565 
1566 		local_bh_disable();
1567 		while ((nskb = __skb_dequeue(&process_queue)))
1568 			netif_receive_skb(nskb);
1569 		netif_receive_skb(skb);
1570 		local_bh_enable();
1571 	}
1572 }
1573 
1574 static bool tun_can_build_skb(struct tun_struct *tun, struct tun_file *tfile,
1575 			      int len, int noblock, bool zerocopy)
1576 {
1577 	if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
1578 		return false;
1579 
1580 	if (tfile->socket.sk->sk_sndbuf != INT_MAX)
1581 		return false;
1582 
1583 	if (!noblock)
1584 		return false;
1585 
1586 	if (zerocopy)
1587 		return false;
1588 
1589 	if (SKB_DATA_ALIGN(len + TUN_RX_PAD) +
1590 	    SKB_DATA_ALIGN(sizeof(struct skb_shared_info)) > PAGE_SIZE)
1591 		return false;
1592 
1593 	return true;
1594 }
1595 
1596 static struct sk_buff *__tun_build_skb(struct page_frag *alloc_frag, char *buf,
1597 				       int buflen, int len, int pad)
1598 {
1599 	struct sk_buff *skb = build_skb(buf, buflen);
1600 
1601 	if (!skb)
1602 		return ERR_PTR(-ENOMEM);
1603 
1604 	skb_reserve(skb, pad);
1605 	skb_put(skb, len);
1606 
1607 	get_page(alloc_frag->page);
1608 	alloc_frag->offset += buflen;
1609 
1610 	return skb;
1611 }
1612 
1613 static int tun_xdp_act(struct tun_struct *tun, struct bpf_prog *xdp_prog,
1614 		       struct xdp_buff *xdp, u32 act)
1615 {
1616 	int err;
1617 
1618 	switch (act) {
1619 	case XDP_REDIRECT:
1620 		err = xdp_do_redirect(tun->dev, xdp, xdp_prog);
1621 		if (err)
1622 			return err;
1623 		break;
1624 	case XDP_TX:
1625 		err = tun_xdp_tx(tun->dev, xdp);
1626 		if (err < 0)
1627 			return err;
1628 		break;
1629 	case XDP_PASS:
1630 		break;
1631 	default:
1632 		bpf_warn_invalid_xdp_action(act);
1633 		/* fall through */
1634 	case XDP_ABORTED:
1635 		trace_xdp_exception(tun->dev, xdp_prog, act);
1636 		/* fall through */
1637 	case XDP_DROP:
1638 		this_cpu_inc(tun->pcpu_stats->rx_dropped);
1639 		break;
1640 	}
1641 
1642 	return act;
1643 }
1644 
1645 static struct sk_buff *tun_build_skb(struct tun_struct *tun,
1646 				     struct tun_file *tfile,
1647 				     struct iov_iter *from,
1648 				     struct virtio_net_hdr *hdr,
1649 				     int len, int *skb_xdp)
1650 {
1651 	struct page_frag *alloc_frag = &current->task_frag;
1652 	struct bpf_prog *xdp_prog;
1653 	int buflen = SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
1654 	char *buf;
1655 	size_t copied;
1656 	int pad = TUN_RX_PAD;
1657 	int err = 0;
1658 
1659 	rcu_read_lock();
1660 	xdp_prog = rcu_dereference(tun->xdp_prog);
1661 	if (xdp_prog)
1662 		pad += XDP_PACKET_HEADROOM;
1663 	buflen += SKB_DATA_ALIGN(len + pad);
1664 	rcu_read_unlock();
1665 
1666 	alloc_frag->offset = ALIGN((u64)alloc_frag->offset, SMP_CACHE_BYTES);
1667 	if (unlikely(!skb_page_frag_refill(buflen, alloc_frag, GFP_KERNEL)))
1668 		return ERR_PTR(-ENOMEM);
1669 
1670 	buf = (char *)page_address(alloc_frag->page) + alloc_frag->offset;
1671 	copied = copy_page_from_iter(alloc_frag->page,
1672 				     alloc_frag->offset + pad,
1673 				     len, from);
1674 	if (copied != len)
1675 		return ERR_PTR(-EFAULT);
1676 
1677 	/* There's a small window that XDP may be set after the check
1678 	 * of xdp_prog above, this should be rare and for simplicity
1679 	 * we do XDP on skb in case the headroom is not enough.
1680 	 */
1681 	if (hdr->gso_type || !xdp_prog) {
1682 		*skb_xdp = 1;
1683 		return __tun_build_skb(alloc_frag, buf, buflen, len, pad);
1684 	}
1685 
1686 	*skb_xdp = 0;
1687 
1688 	local_bh_disable();
1689 	rcu_read_lock();
1690 	xdp_prog = rcu_dereference(tun->xdp_prog);
1691 	if (xdp_prog) {
1692 		struct xdp_buff xdp;
1693 		u32 act;
1694 
1695 		xdp.data_hard_start = buf;
1696 		xdp.data = buf + pad;
1697 		xdp_set_data_meta_invalid(&xdp);
1698 		xdp.data_end = xdp.data + len;
1699 		xdp.rxq = &tfile->xdp_rxq;
1700 
1701 		act = bpf_prog_run_xdp(xdp_prog, &xdp);
1702 		if (act == XDP_REDIRECT || act == XDP_TX) {
1703 			get_page(alloc_frag->page);
1704 			alloc_frag->offset += buflen;
1705 		}
1706 		err = tun_xdp_act(tun, xdp_prog, &xdp, act);
1707 		if (err < 0)
1708 			goto err_xdp;
1709 		if (err == XDP_REDIRECT)
1710 			xdp_do_flush_map();
1711 		if (err != XDP_PASS)
1712 			goto out;
1713 
1714 		pad = xdp.data - xdp.data_hard_start;
1715 		len = xdp.data_end - xdp.data;
1716 	}
1717 	rcu_read_unlock();
1718 	local_bh_enable();
1719 
1720 	return __tun_build_skb(alloc_frag, buf, buflen, len, pad);
1721 
1722 err_xdp:
1723 	put_page(alloc_frag->page);
1724 out:
1725 	rcu_read_unlock();
1726 	local_bh_enable();
1727 	return NULL;
1728 }
1729 
1730 /* Get packet from user space buffer */
1731 static ssize_t tun_get_user(struct tun_struct *tun, struct tun_file *tfile,
1732 			    void *msg_control, struct iov_iter *from,
1733 			    int noblock, bool more)
1734 {
1735 	struct tun_pi pi = { 0, cpu_to_be16(ETH_P_IP) };
1736 	struct sk_buff *skb;
1737 	size_t total_len = iov_iter_count(from);
1738 	size_t len = total_len, align = tun->align, linear;
1739 	struct virtio_net_hdr gso = { 0 };
1740 	struct tun_pcpu_stats *stats;
1741 	int good_linear;
1742 	int copylen;
1743 	bool zerocopy = false;
1744 	int err;
1745 	u32 rxhash = 0;
1746 	int skb_xdp = 1;
1747 	bool frags = tun_napi_frags_enabled(tfile);
1748 
1749 	if (!(tun->dev->flags & IFF_UP))
1750 		return -EIO;
1751 
1752 	if (!(tun->flags & IFF_NO_PI)) {
1753 		if (len < sizeof(pi))
1754 			return -EINVAL;
1755 		len -= sizeof(pi);
1756 
1757 		if (!copy_from_iter_full(&pi, sizeof(pi), from))
1758 			return -EFAULT;
1759 	}
1760 
1761 	if (tun->flags & IFF_VNET_HDR) {
1762 		int vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
1763 
1764 		if (len < vnet_hdr_sz)
1765 			return -EINVAL;
1766 		len -= vnet_hdr_sz;
1767 
1768 		if (!copy_from_iter_full(&gso, sizeof(gso), from))
1769 			return -EFAULT;
1770 
1771 		if ((gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
1772 		    tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2 > tun16_to_cpu(tun, gso.hdr_len))
1773 			gso.hdr_len = cpu_to_tun16(tun, tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2);
1774 
1775 		if (tun16_to_cpu(tun, gso.hdr_len) > len)
1776 			return -EINVAL;
1777 		iov_iter_advance(from, vnet_hdr_sz - sizeof(gso));
1778 	}
1779 
1780 	if ((tun->flags & TUN_TYPE_MASK) == IFF_TAP) {
1781 		align += NET_IP_ALIGN;
1782 		if (unlikely(len < ETH_HLEN ||
1783 			     (gso.hdr_len && tun16_to_cpu(tun, gso.hdr_len) < ETH_HLEN)))
1784 			return -EINVAL;
1785 	}
1786 
1787 	good_linear = SKB_MAX_HEAD(align);
1788 
1789 	if (msg_control) {
1790 		struct iov_iter i = *from;
1791 
1792 		/* There are 256 bytes to be copied in skb, so there is
1793 		 * enough room for skb expand head in case it is used.
1794 		 * The rest of the buffer is mapped from userspace.
1795 		 */
1796 		copylen = gso.hdr_len ? tun16_to_cpu(tun, gso.hdr_len) : GOODCOPY_LEN;
1797 		if (copylen > good_linear)
1798 			copylen = good_linear;
1799 		linear = copylen;
1800 		iov_iter_advance(&i, copylen);
1801 		if (iov_iter_npages(&i, INT_MAX) <= MAX_SKB_FRAGS)
1802 			zerocopy = true;
1803 	}
1804 
1805 	if (!frags && tun_can_build_skb(tun, tfile, len, noblock, zerocopy)) {
1806 		/* For the packet that is not easy to be processed
1807 		 * (e.g gso or jumbo packet), we will do it at after
1808 		 * skb was created with generic XDP routine.
1809 		 */
1810 		skb = tun_build_skb(tun, tfile, from, &gso, len, &skb_xdp);
1811 		if (IS_ERR(skb)) {
1812 			this_cpu_inc(tun->pcpu_stats->rx_dropped);
1813 			return PTR_ERR(skb);
1814 		}
1815 		if (!skb)
1816 			return total_len;
1817 	} else {
1818 		if (!zerocopy) {
1819 			copylen = len;
1820 			if (tun16_to_cpu(tun, gso.hdr_len) > good_linear)
1821 				linear = good_linear;
1822 			else
1823 				linear = tun16_to_cpu(tun, gso.hdr_len);
1824 		}
1825 
1826 		if (frags) {
1827 			mutex_lock(&tfile->napi_mutex);
1828 			skb = tun_napi_alloc_frags(tfile, copylen, from);
1829 			/* tun_napi_alloc_frags() enforces a layout for the skb.
1830 			 * If zerocopy is enabled, then this layout will be
1831 			 * overwritten by zerocopy_sg_from_iter().
1832 			 */
1833 			zerocopy = false;
1834 		} else {
1835 			skb = tun_alloc_skb(tfile, align, copylen, linear,
1836 					    noblock);
1837 		}
1838 
1839 		if (IS_ERR(skb)) {
1840 			if (PTR_ERR(skb) != -EAGAIN)
1841 				this_cpu_inc(tun->pcpu_stats->rx_dropped);
1842 			if (frags)
1843 				mutex_unlock(&tfile->napi_mutex);
1844 			return PTR_ERR(skb);
1845 		}
1846 
1847 		if (zerocopy)
1848 			err = zerocopy_sg_from_iter(skb, from);
1849 		else
1850 			err = skb_copy_datagram_from_iter(skb, 0, from, len);
1851 
1852 		if (err) {
1853 			this_cpu_inc(tun->pcpu_stats->rx_dropped);
1854 			kfree_skb(skb);
1855 			if (frags) {
1856 				tfile->napi.skb = NULL;
1857 				mutex_unlock(&tfile->napi_mutex);
1858 			}
1859 
1860 			return -EFAULT;
1861 		}
1862 	}
1863 
1864 	if (virtio_net_hdr_to_skb(skb, &gso, tun_is_little_endian(tun))) {
1865 		this_cpu_inc(tun->pcpu_stats->rx_frame_errors);
1866 		kfree_skb(skb);
1867 		if (frags) {
1868 			tfile->napi.skb = NULL;
1869 			mutex_unlock(&tfile->napi_mutex);
1870 		}
1871 
1872 		return -EINVAL;
1873 	}
1874 
1875 	switch (tun->flags & TUN_TYPE_MASK) {
1876 	case IFF_TUN:
1877 		if (tun->flags & IFF_NO_PI) {
1878 			u8 ip_version = skb->len ? (skb->data[0] >> 4) : 0;
1879 
1880 			switch (ip_version) {
1881 			case 4:
1882 				pi.proto = htons(ETH_P_IP);
1883 				break;
1884 			case 6:
1885 				pi.proto = htons(ETH_P_IPV6);
1886 				break;
1887 			default:
1888 				this_cpu_inc(tun->pcpu_stats->rx_dropped);
1889 				kfree_skb(skb);
1890 				return -EINVAL;
1891 			}
1892 		}
1893 
1894 		skb_reset_mac_header(skb);
1895 		skb->protocol = pi.proto;
1896 		skb->dev = tun->dev;
1897 		break;
1898 	case IFF_TAP:
1899 		if (!frags)
1900 			skb->protocol = eth_type_trans(skb, tun->dev);
1901 		break;
1902 	}
1903 
1904 	/* copy skb_ubuf_info for callback when skb has no error */
1905 	if (zerocopy) {
1906 		skb_shinfo(skb)->destructor_arg = msg_control;
1907 		skb_shinfo(skb)->tx_flags |= SKBTX_DEV_ZEROCOPY;
1908 		skb_shinfo(skb)->tx_flags |= SKBTX_SHARED_FRAG;
1909 	} else if (msg_control) {
1910 		struct ubuf_info *uarg = msg_control;
1911 		uarg->callback(uarg, false);
1912 	}
1913 
1914 	skb_reset_network_header(skb);
1915 	skb_probe_transport_header(skb, 0);
1916 
1917 	if (skb_xdp) {
1918 		struct bpf_prog *xdp_prog;
1919 		int ret;
1920 
1921 		local_bh_disable();
1922 		rcu_read_lock();
1923 		xdp_prog = rcu_dereference(tun->xdp_prog);
1924 		if (xdp_prog) {
1925 			ret = do_xdp_generic(xdp_prog, skb);
1926 			if (ret != XDP_PASS) {
1927 				rcu_read_unlock();
1928 				local_bh_enable();
1929 				return total_len;
1930 			}
1931 		}
1932 		rcu_read_unlock();
1933 		local_bh_enable();
1934 	}
1935 
1936 	/* Compute the costly rx hash only if needed for flow updates.
1937 	 * We may get a very small possibility of OOO during switching, not
1938 	 * worth to optimize.
1939 	 */
1940 	if (!rcu_access_pointer(tun->steering_prog) && tun->numqueues > 1 &&
1941 	    !tfile->detached)
1942 		rxhash = __skb_get_hash_symmetric(skb);
1943 
1944 	if (frags) {
1945 		/* Exercise flow dissector code path. */
1946 		u32 headlen = eth_get_headlen(skb->data, skb_headlen(skb));
1947 
1948 		if (unlikely(headlen > skb_headlen(skb))) {
1949 			this_cpu_inc(tun->pcpu_stats->rx_dropped);
1950 			napi_free_frags(&tfile->napi);
1951 			mutex_unlock(&tfile->napi_mutex);
1952 			WARN_ON(1);
1953 			return -ENOMEM;
1954 		}
1955 
1956 		local_bh_disable();
1957 		napi_gro_frags(&tfile->napi);
1958 		local_bh_enable();
1959 		mutex_unlock(&tfile->napi_mutex);
1960 	} else if (tfile->napi_enabled) {
1961 		struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
1962 		int queue_len;
1963 
1964 		spin_lock_bh(&queue->lock);
1965 		__skb_queue_tail(queue, skb);
1966 		queue_len = skb_queue_len(queue);
1967 		spin_unlock(&queue->lock);
1968 
1969 		if (!more || queue_len > NAPI_POLL_WEIGHT)
1970 			napi_schedule(&tfile->napi);
1971 
1972 		local_bh_enable();
1973 	} else if (!IS_ENABLED(CONFIG_4KSTACKS)) {
1974 		tun_rx_batched(tun, tfile, skb, more);
1975 	} else {
1976 		netif_rx_ni(skb);
1977 	}
1978 
1979 	stats = get_cpu_ptr(tun->pcpu_stats);
1980 	u64_stats_update_begin(&stats->syncp);
1981 	stats->rx_packets++;
1982 	stats->rx_bytes += len;
1983 	u64_stats_update_end(&stats->syncp);
1984 	put_cpu_ptr(stats);
1985 
1986 	if (rxhash)
1987 		tun_flow_update(tun, rxhash, tfile);
1988 
1989 	return total_len;
1990 }
1991 
1992 static ssize_t tun_chr_write_iter(struct kiocb *iocb, struct iov_iter *from)
1993 {
1994 	struct file *file = iocb->ki_filp;
1995 	struct tun_file *tfile = file->private_data;
1996 	struct tun_struct *tun = tun_get(tfile);
1997 	ssize_t result;
1998 
1999 	if (!tun)
2000 		return -EBADFD;
2001 
2002 	result = tun_get_user(tun, tfile, NULL, from,
2003 			      file->f_flags & O_NONBLOCK, false);
2004 
2005 	tun_put(tun);
2006 	return result;
2007 }
2008 
2009 static ssize_t tun_put_user_xdp(struct tun_struct *tun,
2010 				struct tun_file *tfile,
2011 				struct xdp_frame *xdp_frame,
2012 				struct iov_iter *iter)
2013 {
2014 	int vnet_hdr_sz = 0;
2015 	size_t size = xdp_frame->len;
2016 	struct tun_pcpu_stats *stats;
2017 	size_t ret;
2018 
2019 	if (tun->flags & IFF_VNET_HDR) {
2020 		struct virtio_net_hdr gso = { 0 };
2021 
2022 		vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
2023 		if (unlikely(iov_iter_count(iter) < vnet_hdr_sz))
2024 			return -EINVAL;
2025 		if (unlikely(copy_to_iter(&gso, sizeof(gso), iter) !=
2026 			     sizeof(gso)))
2027 			return -EFAULT;
2028 		iov_iter_advance(iter, vnet_hdr_sz - sizeof(gso));
2029 	}
2030 
2031 	ret = copy_to_iter(xdp_frame->data, size, iter) + vnet_hdr_sz;
2032 
2033 	stats = get_cpu_ptr(tun->pcpu_stats);
2034 	u64_stats_update_begin(&stats->syncp);
2035 	stats->tx_packets++;
2036 	stats->tx_bytes += ret;
2037 	u64_stats_update_end(&stats->syncp);
2038 	put_cpu_ptr(tun->pcpu_stats);
2039 
2040 	return ret;
2041 }
2042 
2043 /* Put packet to the user space buffer */
2044 static ssize_t tun_put_user(struct tun_struct *tun,
2045 			    struct tun_file *tfile,
2046 			    struct sk_buff *skb,
2047 			    struct iov_iter *iter)
2048 {
2049 	struct tun_pi pi = { 0, skb->protocol };
2050 	struct tun_pcpu_stats *stats;
2051 	ssize_t total;
2052 	int vlan_offset = 0;
2053 	int vlan_hlen = 0;
2054 	int vnet_hdr_sz = 0;
2055 
2056 	if (skb_vlan_tag_present(skb))
2057 		vlan_hlen = VLAN_HLEN;
2058 
2059 	if (tun->flags & IFF_VNET_HDR)
2060 		vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
2061 
2062 	total = skb->len + vlan_hlen + vnet_hdr_sz;
2063 
2064 	if (!(tun->flags & IFF_NO_PI)) {
2065 		if (iov_iter_count(iter) < sizeof(pi))
2066 			return -EINVAL;
2067 
2068 		total += sizeof(pi);
2069 		if (iov_iter_count(iter) < total) {
2070 			/* Packet will be striped */
2071 			pi.flags |= TUN_PKT_STRIP;
2072 		}
2073 
2074 		if (copy_to_iter(&pi, sizeof(pi), iter) != sizeof(pi))
2075 			return -EFAULT;
2076 	}
2077 
2078 	if (vnet_hdr_sz) {
2079 		struct virtio_net_hdr gso;
2080 
2081 		if (iov_iter_count(iter) < vnet_hdr_sz)
2082 			return -EINVAL;
2083 
2084 		if (virtio_net_hdr_from_skb(skb, &gso,
2085 					    tun_is_little_endian(tun), true,
2086 					    vlan_hlen)) {
2087 			struct skb_shared_info *sinfo = skb_shinfo(skb);
2088 			pr_err("unexpected GSO type: "
2089 			       "0x%x, gso_size %d, hdr_len %d\n",
2090 			       sinfo->gso_type, tun16_to_cpu(tun, gso.gso_size),
2091 			       tun16_to_cpu(tun, gso.hdr_len));
2092 			print_hex_dump(KERN_ERR, "tun: ",
2093 				       DUMP_PREFIX_NONE,
2094 				       16, 1, skb->head,
2095 				       min((int)tun16_to_cpu(tun, gso.hdr_len), 64), true);
2096 			WARN_ON_ONCE(1);
2097 			return -EINVAL;
2098 		}
2099 
2100 		if (copy_to_iter(&gso, sizeof(gso), iter) != sizeof(gso))
2101 			return -EFAULT;
2102 
2103 		iov_iter_advance(iter, vnet_hdr_sz - sizeof(gso));
2104 	}
2105 
2106 	if (vlan_hlen) {
2107 		int ret;
2108 		struct veth veth;
2109 
2110 		veth.h_vlan_proto = skb->vlan_proto;
2111 		veth.h_vlan_TCI = htons(skb_vlan_tag_get(skb));
2112 
2113 		vlan_offset = offsetof(struct vlan_ethhdr, h_vlan_proto);
2114 
2115 		ret = skb_copy_datagram_iter(skb, 0, iter, vlan_offset);
2116 		if (ret || !iov_iter_count(iter))
2117 			goto done;
2118 
2119 		ret = copy_to_iter(&veth, sizeof(veth), iter);
2120 		if (ret != sizeof(veth) || !iov_iter_count(iter))
2121 			goto done;
2122 	}
2123 
2124 	skb_copy_datagram_iter(skb, vlan_offset, iter, skb->len - vlan_offset);
2125 
2126 done:
2127 	/* caller is in process context, */
2128 	stats = get_cpu_ptr(tun->pcpu_stats);
2129 	u64_stats_update_begin(&stats->syncp);
2130 	stats->tx_packets++;
2131 	stats->tx_bytes += skb->len + vlan_hlen;
2132 	u64_stats_update_end(&stats->syncp);
2133 	put_cpu_ptr(tun->pcpu_stats);
2134 
2135 	return total;
2136 }
2137 
2138 static void *tun_ring_recv(struct tun_file *tfile, int noblock, int *err)
2139 {
2140 	DECLARE_WAITQUEUE(wait, current);
2141 	void *ptr = NULL;
2142 	int error = 0;
2143 
2144 	ptr = ptr_ring_consume(&tfile->tx_ring);
2145 	if (ptr)
2146 		goto out;
2147 	if (noblock) {
2148 		error = -EAGAIN;
2149 		goto out;
2150 	}
2151 
2152 	add_wait_queue(&tfile->wq.wait, &wait);
2153 	current->state = TASK_INTERRUPTIBLE;
2154 
2155 	while (1) {
2156 		ptr = ptr_ring_consume(&tfile->tx_ring);
2157 		if (ptr)
2158 			break;
2159 		if (signal_pending(current)) {
2160 			error = -ERESTARTSYS;
2161 			break;
2162 		}
2163 		if (tfile->socket.sk->sk_shutdown & RCV_SHUTDOWN) {
2164 			error = -EFAULT;
2165 			break;
2166 		}
2167 
2168 		schedule();
2169 	}
2170 
2171 	current->state = TASK_RUNNING;
2172 	remove_wait_queue(&tfile->wq.wait, &wait);
2173 
2174 out:
2175 	*err = error;
2176 	return ptr;
2177 }
2178 
2179 static ssize_t tun_do_read(struct tun_struct *tun, struct tun_file *tfile,
2180 			   struct iov_iter *to,
2181 			   int noblock, void *ptr)
2182 {
2183 	ssize_t ret;
2184 	int err;
2185 
2186 	tun_debug(KERN_INFO, tun, "tun_do_read\n");
2187 
2188 	if (!iov_iter_count(to)) {
2189 		tun_ptr_free(ptr);
2190 		return 0;
2191 	}
2192 
2193 	if (!ptr) {
2194 		/* Read frames from ring */
2195 		ptr = tun_ring_recv(tfile, noblock, &err);
2196 		if (!ptr)
2197 			return err;
2198 	}
2199 
2200 	if (tun_is_xdp_frame(ptr)) {
2201 		struct xdp_frame *xdpf = tun_ptr_to_xdp(ptr);
2202 
2203 		ret = tun_put_user_xdp(tun, tfile, xdpf, to);
2204 		xdp_return_frame(xdpf);
2205 	} else {
2206 		struct sk_buff *skb = ptr;
2207 
2208 		ret = tun_put_user(tun, tfile, skb, to);
2209 		if (unlikely(ret < 0))
2210 			kfree_skb(skb);
2211 		else
2212 			consume_skb(skb);
2213 	}
2214 
2215 	return ret;
2216 }
2217 
2218 static ssize_t tun_chr_read_iter(struct kiocb *iocb, struct iov_iter *to)
2219 {
2220 	struct file *file = iocb->ki_filp;
2221 	struct tun_file *tfile = file->private_data;
2222 	struct tun_struct *tun = tun_get(tfile);
2223 	ssize_t len = iov_iter_count(to), ret;
2224 
2225 	if (!tun)
2226 		return -EBADFD;
2227 	ret = tun_do_read(tun, tfile, to, file->f_flags & O_NONBLOCK, NULL);
2228 	ret = min_t(ssize_t, ret, len);
2229 	if (ret > 0)
2230 		iocb->ki_pos = ret;
2231 	tun_put(tun);
2232 	return ret;
2233 }
2234 
2235 static void tun_prog_free(struct rcu_head *rcu)
2236 {
2237 	struct tun_prog *prog = container_of(rcu, struct tun_prog, rcu);
2238 
2239 	bpf_prog_destroy(prog->prog);
2240 	kfree(prog);
2241 }
2242 
2243 static int __tun_set_ebpf(struct tun_struct *tun,
2244 			  struct tun_prog __rcu **prog_p,
2245 			  struct bpf_prog *prog)
2246 {
2247 	struct tun_prog *old, *new = NULL;
2248 
2249 	if (prog) {
2250 		new = kmalloc(sizeof(*new), GFP_KERNEL);
2251 		if (!new)
2252 			return -ENOMEM;
2253 		new->prog = prog;
2254 	}
2255 
2256 	spin_lock_bh(&tun->lock);
2257 	old = rcu_dereference_protected(*prog_p,
2258 					lockdep_is_held(&tun->lock));
2259 	rcu_assign_pointer(*prog_p, new);
2260 	spin_unlock_bh(&tun->lock);
2261 
2262 	if (old)
2263 		call_rcu(&old->rcu, tun_prog_free);
2264 
2265 	return 0;
2266 }
2267 
2268 static void tun_free_netdev(struct net_device *dev)
2269 {
2270 	struct tun_struct *tun = netdev_priv(dev);
2271 
2272 	BUG_ON(!(list_empty(&tun->disabled)));
2273 	free_percpu(tun->pcpu_stats);
2274 	tun_flow_uninit(tun);
2275 	security_tun_dev_free_security(tun->security);
2276 	__tun_set_ebpf(tun, &tun->steering_prog, NULL);
2277 	__tun_set_ebpf(tun, &tun->filter_prog, NULL);
2278 }
2279 
2280 static void tun_setup(struct net_device *dev)
2281 {
2282 	struct tun_struct *tun = netdev_priv(dev);
2283 
2284 	tun->owner = INVALID_UID;
2285 	tun->group = INVALID_GID;
2286 	tun_default_link_ksettings(dev, &tun->link_ksettings);
2287 
2288 	dev->ethtool_ops = &tun_ethtool_ops;
2289 	dev->needs_free_netdev = true;
2290 	dev->priv_destructor = tun_free_netdev;
2291 	/* We prefer our own queue length */
2292 	dev->tx_queue_len = TUN_READQ_SIZE;
2293 }
2294 
2295 /* Trivial set of netlink ops to allow deleting tun or tap
2296  * device with netlink.
2297  */
2298 static int tun_validate(struct nlattr *tb[], struct nlattr *data[],
2299 			struct netlink_ext_ack *extack)
2300 {
2301 	return -EINVAL;
2302 }
2303 
2304 static size_t tun_get_size(const struct net_device *dev)
2305 {
2306 	BUILD_BUG_ON(sizeof(u32) != sizeof(uid_t));
2307 	BUILD_BUG_ON(sizeof(u32) != sizeof(gid_t));
2308 
2309 	return nla_total_size(sizeof(uid_t)) + /* OWNER */
2310 	       nla_total_size(sizeof(gid_t)) + /* GROUP */
2311 	       nla_total_size(sizeof(u8)) + /* TYPE */
2312 	       nla_total_size(sizeof(u8)) + /* PI */
2313 	       nla_total_size(sizeof(u8)) + /* VNET_HDR */
2314 	       nla_total_size(sizeof(u8)) + /* PERSIST */
2315 	       nla_total_size(sizeof(u8)) + /* MULTI_QUEUE */
2316 	       nla_total_size(sizeof(u32)) + /* NUM_QUEUES */
2317 	       nla_total_size(sizeof(u32)) + /* NUM_DISABLED_QUEUES */
2318 	       0;
2319 }
2320 
2321 static int tun_fill_info(struct sk_buff *skb, const struct net_device *dev)
2322 {
2323 	struct tun_struct *tun = netdev_priv(dev);
2324 
2325 	if (nla_put_u8(skb, IFLA_TUN_TYPE, tun->flags & TUN_TYPE_MASK))
2326 		goto nla_put_failure;
2327 	if (uid_valid(tun->owner) &&
2328 	    nla_put_u32(skb, IFLA_TUN_OWNER,
2329 			from_kuid_munged(current_user_ns(), tun->owner)))
2330 		goto nla_put_failure;
2331 	if (gid_valid(tun->group) &&
2332 	    nla_put_u32(skb, IFLA_TUN_GROUP,
2333 			from_kgid_munged(current_user_ns(), tun->group)))
2334 		goto nla_put_failure;
2335 	if (nla_put_u8(skb, IFLA_TUN_PI, !(tun->flags & IFF_NO_PI)))
2336 		goto nla_put_failure;
2337 	if (nla_put_u8(skb, IFLA_TUN_VNET_HDR, !!(tun->flags & IFF_VNET_HDR)))
2338 		goto nla_put_failure;
2339 	if (nla_put_u8(skb, IFLA_TUN_PERSIST, !!(tun->flags & IFF_PERSIST)))
2340 		goto nla_put_failure;
2341 	if (nla_put_u8(skb, IFLA_TUN_MULTI_QUEUE,
2342 		       !!(tun->flags & IFF_MULTI_QUEUE)))
2343 		goto nla_put_failure;
2344 	if (tun->flags & IFF_MULTI_QUEUE) {
2345 		if (nla_put_u32(skb, IFLA_TUN_NUM_QUEUES, tun->numqueues))
2346 			goto nla_put_failure;
2347 		if (nla_put_u32(skb, IFLA_TUN_NUM_DISABLED_QUEUES,
2348 				tun->numdisabled))
2349 			goto nla_put_failure;
2350 	}
2351 
2352 	return 0;
2353 
2354 nla_put_failure:
2355 	return -EMSGSIZE;
2356 }
2357 
2358 static struct rtnl_link_ops tun_link_ops __read_mostly = {
2359 	.kind		= DRV_NAME,
2360 	.priv_size	= sizeof(struct tun_struct),
2361 	.setup		= tun_setup,
2362 	.validate	= tun_validate,
2363 	.get_size       = tun_get_size,
2364 	.fill_info      = tun_fill_info,
2365 };
2366 
2367 static void tun_sock_write_space(struct sock *sk)
2368 {
2369 	struct tun_file *tfile;
2370 	wait_queue_head_t *wqueue;
2371 
2372 	if (!sock_writeable(sk))
2373 		return;
2374 
2375 	if (!test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags))
2376 		return;
2377 
2378 	wqueue = sk_sleep(sk);
2379 	if (wqueue && waitqueue_active(wqueue))
2380 		wake_up_interruptible_sync_poll(wqueue, EPOLLOUT |
2381 						EPOLLWRNORM | EPOLLWRBAND);
2382 
2383 	tfile = container_of(sk, struct tun_file, sk);
2384 	kill_fasync(&tfile->fasync, SIGIO, POLL_OUT);
2385 }
2386 
2387 static int tun_xdp_one(struct tun_struct *tun,
2388 		       struct tun_file *tfile,
2389 		       struct xdp_buff *xdp, int *flush)
2390 {
2391 	struct tun_xdp_hdr *hdr = xdp->data_hard_start;
2392 	struct virtio_net_hdr *gso = &hdr->gso;
2393 	struct tun_pcpu_stats *stats;
2394 	struct bpf_prog *xdp_prog;
2395 	struct sk_buff *skb = NULL;
2396 	u32 rxhash = 0, act;
2397 	int buflen = hdr->buflen;
2398 	int err = 0;
2399 	bool skb_xdp = false;
2400 
2401 	xdp_prog = rcu_dereference(tun->xdp_prog);
2402 	if (xdp_prog) {
2403 		if (gso->gso_type) {
2404 			skb_xdp = true;
2405 			goto build;
2406 		}
2407 		xdp_set_data_meta_invalid(xdp);
2408 		xdp->rxq = &tfile->xdp_rxq;
2409 
2410 		act = bpf_prog_run_xdp(xdp_prog, xdp);
2411 		err = tun_xdp_act(tun, xdp_prog, xdp, act);
2412 		if (err < 0) {
2413 			put_page(virt_to_head_page(xdp->data));
2414 			return err;
2415 		}
2416 
2417 		switch (err) {
2418 		case XDP_REDIRECT:
2419 			*flush = true;
2420 			/* fall through */
2421 		case XDP_TX:
2422 			return 0;
2423 		case XDP_PASS:
2424 			break;
2425 		default:
2426 			put_page(virt_to_head_page(xdp->data));
2427 			return 0;
2428 		}
2429 	}
2430 
2431 build:
2432 	skb = build_skb(xdp->data_hard_start, buflen);
2433 	if (!skb) {
2434 		err = -ENOMEM;
2435 		goto out;
2436 	}
2437 
2438 	skb_reserve(skb, xdp->data - xdp->data_hard_start);
2439 	skb_put(skb, xdp->data_end - xdp->data);
2440 
2441 	if (virtio_net_hdr_to_skb(skb, gso, tun_is_little_endian(tun))) {
2442 		this_cpu_inc(tun->pcpu_stats->rx_frame_errors);
2443 		kfree_skb(skb);
2444 		err = -EINVAL;
2445 		goto out;
2446 	}
2447 
2448 	skb->protocol = eth_type_trans(skb, tun->dev);
2449 	skb_reset_network_header(skb);
2450 	skb_probe_transport_header(skb, 0);
2451 
2452 	if (skb_xdp) {
2453 		err = do_xdp_generic(xdp_prog, skb);
2454 		if (err != XDP_PASS)
2455 			goto out;
2456 	}
2457 
2458 	if (!rcu_dereference(tun->steering_prog))
2459 		rxhash = __skb_get_hash_symmetric(skb);
2460 
2461 	netif_receive_skb(skb);
2462 
2463 	stats = get_cpu_ptr(tun->pcpu_stats);
2464 	u64_stats_update_begin(&stats->syncp);
2465 	stats->rx_packets++;
2466 	stats->rx_bytes += skb->len;
2467 	u64_stats_update_end(&stats->syncp);
2468 	put_cpu_ptr(stats);
2469 
2470 	if (rxhash)
2471 		tun_flow_update(tun, rxhash, tfile);
2472 
2473 out:
2474 	return err;
2475 }
2476 
2477 static int tun_sendmsg(struct socket *sock, struct msghdr *m, size_t total_len)
2478 {
2479 	int ret, i;
2480 	struct tun_file *tfile = container_of(sock, struct tun_file, socket);
2481 	struct tun_struct *tun = tun_get(tfile);
2482 	struct tun_msg_ctl *ctl = m->msg_control;
2483 	struct xdp_buff *xdp;
2484 
2485 	if (!tun)
2486 		return -EBADFD;
2487 
2488 	if (ctl && (ctl->type == TUN_MSG_PTR)) {
2489 		int n = ctl->num;
2490 		int flush = 0;
2491 
2492 		local_bh_disable();
2493 		rcu_read_lock();
2494 
2495 		for (i = 0; i < n; i++) {
2496 			xdp = &((struct xdp_buff *)ctl->ptr)[i];
2497 			tun_xdp_one(tun, tfile, xdp, &flush);
2498 		}
2499 
2500 		if (flush)
2501 			xdp_do_flush_map();
2502 
2503 		rcu_read_unlock();
2504 		local_bh_enable();
2505 
2506 		ret = total_len;
2507 		goto out;
2508 	}
2509 
2510 	ret = tun_get_user(tun, tfile, ctl ? ctl->ptr : NULL, &m->msg_iter,
2511 			   m->msg_flags & MSG_DONTWAIT,
2512 			   m->msg_flags & MSG_MORE);
2513 out:
2514 	tun_put(tun);
2515 	return ret;
2516 }
2517 
2518 static int tun_recvmsg(struct socket *sock, struct msghdr *m, size_t total_len,
2519 		       int flags)
2520 {
2521 	struct tun_file *tfile = container_of(sock, struct tun_file, socket);
2522 	struct tun_struct *tun = tun_get(tfile);
2523 	void *ptr = m->msg_control;
2524 	int ret;
2525 
2526 	if (!tun) {
2527 		ret = -EBADFD;
2528 		goto out_free;
2529 	}
2530 
2531 	if (flags & ~(MSG_DONTWAIT|MSG_TRUNC|MSG_ERRQUEUE)) {
2532 		ret = -EINVAL;
2533 		goto out_put_tun;
2534 	}
2535 	if (flags & MSG_ERRQUEUE) {
2536 		ret = sock_recv_errqueue(sock->sk, m, total_len,
2537 					 SOL_PACKET, TUN_TX_TIMESTAMP);
2538 		goto out;
2539 	}
2540 	ret = tun_do_read(tun, tfile, &m->msg_iter, flags & MSG_DONTWAIT, ptr);
2541 	if (ret > (ssize_t)total_len) {
2542 		m->msg_flags |= MSG_TRUNC;
2543 		ret = flags & MSG_TRUNC ? ret : total_len;
2544 	}
2545 out:
2546 	tun_put(tun);
2547 	return ret;
2548 
2549 out_put_tun:
2550 	tun_put(tun);
2551 out_free:
2552 	tun_ptr_free(ptr);
2553 	return ret;
2554 }
2555 
2556 static int tun_ptr_peek_len(void *ptr)
2557 {
2558 	if (likely(ptr)) {
2559 		if (tun_is_xdp_frame(ptr)) {
2560 			struct xdp_frame *xdpf = tun_ptr_to_xdp(ptr);
2561 
2562 			return xdpf->len;
2563 		}
2564 		return __skb_array_len_with_tag(ptr);
2565 	} else {
2566 		return 0;
2567 	}
2568 }
2569 
2570 static int tun_peek_len(struct socket *sock)
2571 {
2572 	struct tun_file *tfile = container_of(sock, struct tun_file, socket);
2573 	struct tun_struct *tun;
2574 	int ret = 0;
2575 
2576 	tun = tun_get(tfile);
2577 	if (!tun)
2578 		return 0;
2579 
2580 	ret = PTR_RING_PEEK_CALL(&tfile->tx_ring, tun_ptr_peek_len);
2581 	tun_put(tun);
2582 
2583 	return ret;
2584 }
2585 
2586 /* Ops structure to mimic raw sockets with tun */
2587 static const struct proto_ops tun_socket_ops = {
2588 	.peek_len = tun_peek_len,
2589 	.sendmsg = tun_sendmsg,
2590 	.recvmsg = tun_recvmsg,
2591 };
2592 
2593 static struct proto tun_proto = {
2594 	.name		= "tun",
2595 	.owner		= THIS_MODULE,
2596 	.obj_size	= sizeof(struct tun_file),
2597 };
2598 
2599 static int tun_flags(struct tun_struct *tun)
2600 {
2601 	return tun->flags & (TUN_FEATURES | IFF_PERSIST | IFF_TUN | IFF_TAP);
2602 }
2603 
2604 static ssize_t tun_show_flags(struct device *dev, struct device_attribute *attr,
2605 			      char *buf)
2606 {
2607 	struct tun_struct *tun = netdev_priv(to_net_dev(dev));
2608 	return sprintf(buf, "0x%x\n", tun_flags(tun));
2609 }
2610 
2611 static ssize_t tun_show_owner(struct device *dev, struct device_attribute *attr,
2612 			      char *buf)
2613 {
2614 	struct tun_struct *tun = netdev_priv(to_net_dev(dev));
2615 	return uid_valid(tun->owner)?
2616 		sprintf(buf, "%u\n",
2617 			from_kuid_munged(current_user_ns(), tun->owner)):
2618 		sprintf(buf, "-1\n");
2619 }
2620 
2621 static ssize_t tun_show_group(struct device *dev, struct device_attribute *attr,
2622 			      char *buf)
2623 {
2624 	struct tun_struct *tun = netdev_priv(to_net_dev(dev));
2625 	return gid_valid(tun->group) ?
2626 		sprintf(buf, "%u\n",
2627 			from_kgid_munged(current_user_ns(), tun->group)):
2628 		sprintf(buf, "-1\n");
2629 }
2630 
2631 static DEVICE_ATTR(tun_flags, 0444, tun_show_flags, NULL);
2632 static DEVICE_ATTR(owner, 0444, tun_show_owner, NULL);
2633 static DEVICE_ATTR(group, 0444, tun_show_group, NULL);
2634 
2635 static struct attribute *tun_dev_attrs[] = {
2636 	&dev_attr_tun_flags.attr,
2637 	&dev_attr_owner.attr,
2638 	&dev_attr_group.attr,
2639 	NULL
2640 };
2641 
2642 static const struct attribute_group tun_attr_group = {
2643 	.attrs = tun_dev_attrs
2644 };
2645 
2646 static int tun_set_iff(struct net *net, struct file *file, struct ifreq *ifr)
2647 {
2648 	struct tun_struct *tun;
2649 	struct tun_file *tfile = file->private_data;
2650 	struct net_device *dev;
2651 	int err;
2652 
2653 	if (tfile->detached)
2654 		return -EINVAL;
2655 
2656 	if ((ifr->ifr_flags & IFF_NAPI_FRAGS)) {
2657 		if (!capable(CAP_NET_ADMIN))
2658 			return -EPERM;
2659 
2660 		if (!(ifr->ifr_flags & IFF_NAPI) ||
2661 		    (ifr->ifr_flags & TUN_TYPE_MASK) != IFF_TAP)
2662 			return -EINVAL;
2663 	}
2664 
2665 	dev = __dev_get_by_name(net, ifr->ifr_name);
2666 	if (dev) {
2667 		if (ifr->ifr_flags & IFF_TUN_EXCL)
2668 			return -EBUSY;
2669 		if ((ifr->ifr_flags & IFF_TUN) && dev->netdev_ops == &tun_netdev_ops)
2670 			tun = netdev_priv(dev);
2671 		else if ((ifr->ifr_flags & IFF_TAP) && dev->netdev_ops == &tap_netdev_ops)
2672 			tun = netdev_priv(dev);
2673 		else
2674 			return -EINVAL;
2675 
2676 		if (!!(ifr->ifr_flags & IFF_MULTI_QUEUE) !=
2677 		    !!(tun->flags & IFF_MULTI_QUEUE))
2678 			return -EINVAL;
2679 
2680 		if (tun_not_capable(tun))
2681 			return -EPERM;
2682 		err = security_tun_dev_open(tun->security);
2683 		if (err < 0)
2684 			return err;
2685 
2686 		err = tun_attach(tun, file, ifr->ifr_flags & IFF_NOFILTER,
2687 				 ifr->ifr_flags & IFF_NAPI,
2688 				 ifr->ifr_flags & IFF_NAPI_FRAGS);
2689 		if (err < 0)
2690 			return err;
2691 
2692 		if (tun->flags & IFF_MULTI_QUEUE &&
2693 		    (tun->numqueues + tun->numdisabled > 1)) {
2694 			/* One or more queue has already been attached, no need
2695 			 * to initialize the device again.
2696 			 */
2697 			netdev_state_change(dev);
2698 			return 0;
2699 		}
2700 
2701 		tun->flags = (tun->flags & ~TUN_FEATURES) |
2702 			      (ifr->ifr_flags & TUN_FEATURES);
2703 
2704 		netdev_state_change(dev);
2705 	} else {
2706 		char *name;
2707 		unsigned long flags = 0;
2708 		int queues = ifr->ifr_flags & IFF_MULTI_QUEUE ?
2709 			     MAX_TAP_QUEUES : 1;
2710 
2711 		if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
2712 			return -EPERM;
2713 		err = security_tun_dev_create();
2714 		if (err < 0)
2715 			return err;
2716 
2717 		/* Set dev type */
2718 		if (ifr->ifr_flags & IFF_TUN) {
2719 			/* TUN device */
2720 			flags |= IFF_TUN;
2721 			name = "tun%d";
2722 		} else if (ifr->ifr_flags & IFF_TAP) {
2723 			/* TAP device */
2724 			flags |= IFF_TAP;
2725 			name = "tap%d";
2726 		} else
2727 			return -EINVAL;
2728 
2729 		if (*ifr->ifr_name)
2730 			name = ifr->ifr_name;
2731 
2732 		dev = alloc_netdev_mqs(sizeof(struct tun_struct), name,
2733 				       NET_NAME_UNKNOWN, tun_setup, queues,
2734 				       queues);
2735 
2736 		if (!dev)
2737 			return -ENOMEM;
2738 		err = dev_get_valid_name(net, dev, name);
2739 		if (err < 0)
2740 			goto err_free_dev;
2741 
2742 		dev_net_set(dev, net);
2743 		dev->rtnl_link_ops = &tun_link_ops;
2744 		dev->ifindex = tfile->ifindex;
2745 		dev->sysfs_groups[0] = &tun_attr_group;
2746 
2747 		tun = netdev_priv(dev);
2748 		tun->dev = dev;
2749 		tun->flags = flags;
2750 		tun->txflt.count = 0;
2751 		tun->vnet_hdr_sz = sizeof(struct virtio_net_hdr);
2752 
2753 		tun->align = NET_SKB_PAD;
2754 		tun->filter_attached = false;
2755 		tun->sndbuf = tfile->socket.sk->sk_sndbuf;
2756 		tun->rx_batched = 0;
2757 		RCU_INIT_POINTER(tun->steering_prog, NULL);
2758 
2759 		tun->pcpu_stats = netdev_alloc_pcpu_stats(struct tun_pcpu_stats);
2760 		if (!tun->pcpu_stats) {
2761 			err = -ENOMEM;
2762 			goto err_free_dev;
2763 		}
2764 
2765 		spin_lock_init(&tun->lock);
2766 
2767 		err = security_tun_dev_alloc_security(&tun->security);
2768 		if (err < 0)
2769 			goto err_free_stat;
2770 
2771 		tun_net_init(dev);
2772 		tun_flow_init(tun);
2773 
2774 		dev->hw_features = NETIF_F_SG | NETIF_F_FRAGLIST |
2775 				   TUN_USER_FEATURES | NETIF_F_HW_VLAN_CTAG_TX |
2776 				   NETIF_F_HW_VLAN_STAG_TX;
2777 		dev->features = dev->hw_features | NETIF_F_LLTX;
2778 		dev->vlan_features = dev->features &
2779 				     ~(NETIF_F_HW_VLAN_CTAG_TX |
2780 				       NETIF_F_HW_VLAN_STAG_TX);
2781 
2782 		tun->flags = (tun->flags & ~TUN_FEATURES) |
2783 			      (ifr->ifr_flags & TUN_FEATURES);
2784 
2785 		INIT_LIST_HEAD(&tun->disabled);
2786 		err = tun_attach(tun, file, false, ifr->ifr_flags & IFF_NAPI,
2787 				 ifr->ifr_flags & IFF_NAPI_FRAGS);
2788 		if (err < 0)
2789 			goto err_free_flow;
2790 
2791 		err = register_netdevice(tun->dev);
2792 		if (err < 0)
2793 			goto err_detach;
2794 	}
2795 
2796 	netif_carrier_on(tun->dev);
2797 
2798 	tun_debug(KERN_INFO, tun, "tun_set_iff\n");
2799 
2800 	/* Make sure persistent devices do not get stuck in
2801 	 * xoff state.
2802 	 */
2803 	if (netif_running(tun->dev))
2804 		netif_tx_wake_all_queues(tun->dev);
2805 
2806 	strcpy(ifr->ifr_name, tun->dev->name);
2807 	return 0;
2808 
2809 err_detach:
2810 	tun_detach_all(dev);
2811 	/* register_netdevice() already called tun_free_netdev() */
2812 	goto err_free_dev;
2813 
2814 err_free_flow:
2815 	tun_flow_uninit(tun);
2816 	security_tun_dev_free_security(tun->security);
2817 err_free_stat:
2818 	free_percpu(tun->pcpu_stats);
2819 err_free_dev:
2820 	free_netdev(dev);
2821 	return err;
2822 }
2823 
2824 static void tun_get_iff(struct net *net, struct tun_struct *tun,
2825 		       struct ifreq *ifr)
2826 {
2827 	tun_debug(KERN_INFO, tun, "tun_get_iff\n");
2828 
2829 	strcpy(ifr->ifr_name, tun->dev->name);
2830 
2831 	ifr->ifr_flags = tun_flags(tun);
2832 
2833 }
2834 
2835 /* This is like a cut-down ethtool ops, except done via tun fd so no
2836  * privs required. */
2837 static int set_offload(struct tun_struct *tun, unsigned long arg)
2838 {
2839 	netdev_features_t features = 0;
2840 
2841 	if (arg & TUN_F_CSUM) {
2842 		features |= NETIF_F_HW_CSUM;
2843 		arg &= ~TUN_F_CSUM;
2844 
2845 		if (arg & (TUN_F_TSO4|TUN_F_TSO6)) {
2846 			if (arg & TUN_F_TSO_ECN) {
2847 				features |= NETIF_F_TSO_ECN;
2848 				arg &= ~TUN_F_TSO_ECN;
2849 			}
2850 			if (arg & TUN_F_TSO4)
2851 				features |= NETIF_F_TSO;
2852 			if (arg & TUN_F_TSO6)
2853 				features |= NETIF_F_TSO6;
2854 			arg &= ~(TUN_F_TSO4|TUN_F_TSO6);
2855 		}
2856 
2857 		arg &= ~TUN_F_UFO;
2858 	}
2859 
2860 	/* This gives the user a way to test for new features in future by
2861 	 * trying to set them. */
2862 	if (arg)
2863 		return -EINVAL;
2864 
2865 	tun->set_features = features;
2866 	tun->dev->wanted_features &= ~TUN_USER_FEATURES;
2867 	tun->dev->wanted_features |= features;
2868 	netdev_update_features(tun->dev);
2869 
2870 	return 0;
2871 }
2872 
2873 static void tun_detach_filter(struct tun_struct *tun, int n)
2874 {
2875 	int i;
2876 	struct tun_file *tfile;
2877 
2878 	for (i = 0; i < n; i++) {
2879 		tfile = rtnl_dereference(tun->tfiles[i]);
2880 		lock_sock(tfile->socket.sk);
2881 		sk_detach_filter(tfile->socket.sk);
2882 		release_sock(tfile->socket.sk);
2883 	}
2884 
2885 	tun->filter_attached = false;
2886 }
2887 
2888 static int tun_attach_filter(struct tun_struct *tun)
2889 {
2890 	int i, ret = 0;
2891 	struct tun_file *tfile;
2892 
2893 	for (i = 0; i < tun->numqueues; i++) {
2894 		tfile = rtnl_dereference(tun->tfiles[i]);
2895 		lock_sock(tfile->socket.sk);
2896 		ret = sk_attach_filter(&tun->fprog, tfile->socket.sk);
2897 		release_sock(tfile->socket.sk);
2898 		if (ret) {
2899 			tun_detach_filter(tun, i);
2900 			return ret;
2901 		}
2902 	}
2903 
2904 	tun->filter_attached = true;
2905 	return ret;
2906 }
2907 
2908 static void tun_set_sndbuf(struct tun_struct *tun)
2909 {
2910 	struct tun_file *tfile;
2911 	int i;
2912 
2913 	for (i = 0; i < tun->numqueues; i++) {
2914 		tfile = rtnl_dereference(tun->tfiles[i]);
2915 		tfile->socket.sk->sk_sndbuf = tun->sndbuf;
2916 	}
2917 }
2918 
2919 static int tun_set_queue(struct file *file, struct ifreq *ifr)
2920 {
2921 	struct tun_file *tfile = file->private_data;
2922 	struct tun_struct *tun;
2923 	int ret = 0;
2924 
2925 	rtnl_lock();
2926 
2927 	if (ifr->ifr_flags & IFF_ATTACH_QUEUE) {
2928 		tun = tfile->detached;
2929 		if (!tun) {
2930 			ret = -EINVAL;
2931 			goto unlock;
2932 		}
2933 		ret = security_tun_dev_attach_queue(tun->security);
2934 		if (ret < 0)
2935 			goto unlock;
2936 		ret = tun_attach(tun, file, false, tun->flags & IFF_NAPI,
2937 				 tun->flags & IFF_NAPI_FRAGS);
2938 	} else if (ifr->ifr_flags & IFF_DETACH_QUEUE) {
2939 		tun = rtnl_dereference(tfile->tun);
2940 		if (!tun || !(tun->flags & IFF_MULTI_QUEUE) || tfile->detached)
2941 			ret = -EINVAL;
2942 		else
2943 			__tun_detach(tfile, false);
2944 	} else
2945 		ret = -EINVAL;
2946 
2947 	if (ret >= 0)
2948 		netdev_state_change(tun->dev);
2949 
2950 unlock:
2951 	rtnl_unlock();
2952 	return ret;
2953 }
2954 
2955 static int tun_set_ebpf(struct tun_struct *tun, struct tun_prog **prog_p,
2956 			void __user *data)
2957 {
2958 	struct bpf_prog *prog;
2959 	int fd;
2960 
2961 	if (copy_from_user(&fd, data, sizeof(fd)))
2962 		return -EFAULT;
2963 
2964 	if (fd == -1) {
2965 		prog = NULL;
2966 	} else {
2967 		prog = bpf_prog_get_type(fd, BPF_PROG_TYPE_SOCKET_FILTER);
2968 		if (IS_ERR(prog))
2969 			return PTR_ERR(prog);
2970 	}
2971 
2972 	return __tun_set_ebpf(tun, prog_p, prog);
2973 }
2974 
2975 static long __tun_chr_ioctl(struct file *file, unsigned int cmd,
2976 			    unsigned long arg, int ifreq_len)
2977 {
2978 	struct tun_file *tfile = file->private_data;
2979 	struct net *net = sock_net(&tfile->sk);
2980 	struct tun_struct *tun;
2981 	void __user* argp = (void __user*)arg;
2982 	struct ifreq ifr;
2983 	kuid_t owner;
2984 	kgid_t group;
2985 	int sndbuf;
2986 	int vnet_hdr_sz;
2987 	unsigned int ifindex;
2988 	int le;
2989 	int ret;
2990 	bool do_notify = false;
2991 
2992 	if (cmd == TUNSETIFF || cmd == TUNSETQUEUE ||
2993 	    (_IOC_TYPE(cmd) == SOCK_IOC_TYPE && cmd != SIOCGSKNS)) {
2994 		if (copy_from_user(&ifr, argp, ifreq_len))
2995 			return -EFAULT;
2996 	} else {
2997 		memset(&ifr, 0, sizeof(ifr));
2998 	}
2999 	if (cmd == TUNGETFEATURES) {
3000 		/* Currently this just means: "what IFF flags are valid?".
3001 		 * This is needed because we never checked for invalid flags on
3002 		 * TUNSETIFF.
3003 		 */
3004 		return put_user(IFF_TUN | IFF_TAP | TUN_FEATURES,
3005 				(unsigned int __user*)argp);
3006 	} else if (cmd == TUNSETQUEUE) {
3007 		return tun_set_queue(file, &ifr);
3008 	} else if (cmd == SIOCGSKNS) {
3009 		if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
3010 			return -EPERM;
3011 		return open_related_ns(&net->ns, get_net_ns);
3012 	}
3013 
3014 	ret = 0;
3015 	rtnl_lock();
3016 
3017 	tun = tun_get(tfile);
3018 	if (cmd == TUNSETIFF) {
3019 		ret = -EEXIST;
3020 		if (tun)
3021 			goto unlock;
3022 
3023 		ifr.ifr_name[IFNAMSIZ-1] = '\0';
3024 
3025 		ret = tun_set_iff(net, file, &ifr);
3026 
3027 		if (ret)
3028 			goto unlock;
3029 
3030 		if (copy_to_user(argp, &ifr, ifreq_len))
3031 			ret = -EFAULT;
3032 		goto unlock;
3033 	}
3034 	if (cmd == TUNSETIFINDEX) {
3035 		ret = -EPERM;
3036 		if (tun)
3037 			goto unlock;
3038 
3039 		ret = -EFAULT;
3040 		if (copy_from_user(&ifindex, argp, sizeof(ifindex)))
3041 			goto unlock;
3042 
3043 		ret = 0;
3044 		tfile->ifindex = ifindex;
3045 		goto unlock;
3046 	}
3047 
3048 	ret = -EBADFD;
3049 	if (!tun)
3050 		goto unlock;
3051 
3052 	tun_debug(KERN_INFO, tun, "tun_chr_ioctl cmd %u\n", cmd);
3053 
3054 	ret = 0;
3055 	switch (cmd) {
3056 	case TUNGETIFF:
3057 		tun_get_iff(current->nsproxy->net_ns, tun, &ifr);
3058 
3059 		if (tfile->detached)
3060 			ifr.ifr_flags |= IFF_DETACH_QUEUE;
3061 		if (!tfile->socket.sk->sk_filter)
3062 			ifr.ifr_flags |= IFF_NOFILTER;
3063 
3064 		if (copy_to_user(argp, &ifr, ifreq_len))
3065 			ret = -EFAULT;
3066 		break;
3067 
3068 	case TUNSETNOCSUM:
3069 		/* Disable/Enable checksum */
3070 
3071 		/* [unimplemented] */
3072 		tun_debug(KERN_INFO, tun, "ignored: set checksum %s\n",
3073 			  arg ? "disabled" : "enabled");
3074 		break;
3075 
3076 	case TUNSETPERSIST:
3077 		/* Disable/Enable persist mode. Keep an extra reference to the
3078 		 * module to prevent the module being unprobed.
3079 		 */
3080 		if (arg && !(tun->flags & IFF_PERSIST)) {
3081 			tun->flags |= IFF_PERSIST;
3082 			__module_get(THIS_MODULE);
3083 			do_notify = true;
3084 		}
3085 		if (!arg && (tun->flags & IFF_PERSIST)) {
3086 			tun->flags &= ~IFF_PERSIST;
3087 			module_put(THIS_MODULE);
3088 			do_notify = true;
3089 		}
3090 
3091 		tun_debug(KERN_INFO, tun, "persist %s\n",
3092 			  arg ? "enabled" : "disabled");
3093 		break;
3094 
3095 	case TUNSETOWNER:
3096 		/* Set owner of the device */
3097 		owner = make_kuid(current_user_ns(), arg);
3098 		if (!uid_valid(owner)) {
3099 			ret = -EINVAL;
3100 			break;
3101 		}
3102 		tun->owner = owner;
3103 		do_notify = true;
3104 		tun_debug(KERN_INFO, tun, "owner set to %u\n",
3105 			  from_kuid(&init_user_ns, tun->owner));
3106 		break;
3107 
3108 	case TUNSETGROUP:
3109 		/* Set group of the device */
3110 		group = make_kgid(current_user_ns(), arg);
3111 		if (!gid_valid(group)) {
3112 			ret = -EINVAL;
3113 			break;
3114 		}
3115 		tun->group = group;
3116 		do_notify = true;
3117 		tun_debug(KERN_INFO, tun, "group set to %u\n",
3118 			  from_kgid(&init_user_ns, tun->group));
3119 		break;
3120 
3121 	case TUNSETLINK:
3122 		/* Only allow setting the type when the interface is down */
3123 		if (tun->dev->flags & IFF_UP) {
3124 			tun_debug(KERN_INFO, tun,
3125 				  "Linktype set failed because interface is up\n");
3126 			ret = -EBUSY;
3127 		} else {
3128 			tun->dev->type = (int) arg;
3129 			tun_debug(KERN_INFO, tun, "linktype set to %d\n",
3130 				  tun->dev->type);
3131 			ret = 0;
3132 		}
3133 		break;
3134 
3135 #ifdef TUN_DEBUG
3136 	case TUNSETDEBUG:
3137 		tun->debug = arg;
3138 		break;
3139 #endif
3140 	case TUNSETOFFLOAD:
3141 		ret = set_offload(tun, arg);
3142 		break;
3143 
3144 	case TUNSETTXFILTER:
3145 		/* Can be set only for TAPs */
3146 		ret = -EINVAL;
3147 		if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3148 			break;
3149 		ret = update_filter(&tun->txflt, (void __user *)arg);
3150 		break;
3151 
3152 	case SIOCGIFHWADDR:
3153 		/* Get hw address */
3154 		memcpy(ifr.ifr_hwaddr.sa_data, tun->dev->dev_addr, ETH_ALEN);
3155 		ifr.ifr_hwaddr.sa_family = tun->dev->type;
3156 		if (copy_to_user(argp, &ifr, ifreq_len))
3157 			ret = -EFAULT;
3158 		break;
3159 
3160 	case SIOCSIFHWADDR:
3161 		/* Set hw address */
3162 		tun_debug(KERN_DEBUG, tun, "set hw address: %pM\n",
3163 			  ifr.ifr_hwaddr.sa_data);
3164 
3165 		ret = dev_set_mac_address(tun->dev, &ifr.ifr_hwaddr);
3166 		break;
3167 
3168 	case TUNGETSNDBUF:
3169 		sndbuf = tfile->socket.sk->sk_sndbuf;
3170 		if (copy_to_user(argp, &sndbuf, sizeof(sndbuf)))
3171 			ret = -EFAULT;
3172 		break;
3173 
3174 	case TUNSETSNDBUF:
3175 		if (copy_from_user(&sndbuf, argp, sizeof(sndbuf))) {
3176 			ret = -EFAULT;
3177 			break;
3178 		}
3179 		if (sndbuf <= 0) {
3180 			ret = -EINVAL;
3181 			break;
3182 		}
3183 
3184 		tun->sndbuf = sndbuf;
3185 		tun_set_sndbuf(tun);
3186 		break;
3187 
3188 	case TUNGETVNETHDRSZ:
3189 		vnet_hdr_sz = tun->vnet_hdr_sz;
3190 		if (copy_to_user(argp, &vnet_hdr_sz, sizeof(vnet_hdr_sz)))
3191 			ret = -EFAULT;
3192 		break;
3193 
3194 	case TUNSETVNETHDRSZ:
3195 		if (copy_from_user(&vnet_hdr_sz, argp, sizeof(vnet_hdr_sz))) {
3196 			ret = -EFAULT;
3197 			break;
3198 		}
3199 		if (vnet_hdr_sz < (int)sizeof(struct virtio_net_hdr)) {
3200 			ret = -EINVAL;
3201 			break;
3202 		}
3203 
3204 		tun->vnet_hdr_sz = vnet_hdr_sz;
3205 		break;
3206 
3207 	case TUNGETVNETLE:
3208 		le = !!(tun->flags & TUN_VNET_LE);
3209 		if (put_user(le, (int __user *)argp))
3210 			ret = -EFAULT;
3211 		break;
3212 
3213 	case TUNSETVNETLE:
3214 		if (get_user(le, (int __user *)argp)) {
3215 			ret = -EFAULT;
3216 			break;
3217 		}
3218 		if (le)
3219 			tun->flags |= TUN_VNET_LE;
3220 		else
3221 			tun->flags &= ~TUN_VNET_LE;
3222 		break;
3223 
3224 	case TUNGETVNETBE:
3225 		ret = tun_get_vnet_be(tun, argp);
3226 		break;
3227 
3228 	case TUNSETVNETBE:
3229 		ret = tun_set_vnet_be(tun, argp);
3230 		break;
3231 
3232 	case TUNATTACHFILTER:
3233 		/* Can be set only for TAPs */
3234 		ret = -EINVAL;
3235 		if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3236 			break;
3237 		ret = -EFAULT;
3238 		if (copy_from_user(&tun->fprog, argp, sizeof(tun->fprog)))
3239 			break;
3240 
3241 		ret = tun_attach_filter(tun);
3242 		break;
3243 
3244 	case TUNDETACHFILTER:
3245 		/* Can be set only for TAPs */
3246 		ret = -EINVAL;
3247 		if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3248 			break;
3249 		ret = 0;
3250 		tun_detach_filter(tun, tun->numqueues);
3251 		break;
3252 
3253 	case TUNGETFILTER:
3254 		ret = -EINVAL;
3255 		if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3256 			break;
3257 		ret = -EFAULT;
3258 		if (copy_to_user(argp, &tun->fprog, sizeof(tun->fprog)))
3259 			break;
3260 		ret = 0;
3261 		break;
3262 
3263 	case TUNSETSTEERINGEBPF:
3264 		ret = tun_set_ebpf(tun, &tun->steering_prog, argp);
3265 		break;
3266 
3267 	case TUNSETFILTEREBPF:
3268 		ret = tun_set_ebpf(tun, &tun->filter_prog, argp);
3269 		break;
3270 
3271 	default:
3272 		ret = -EINVAL;
3273 		break;
3274 	}
3275 
3276 	if (do_notify)
3277 		netdev_state_change(tun->dev);
3278 
3279 unlock:
3280 	rtnl_unlock();
3281 	if (tun)
3282 		tun_put(tun);
3283 	return ret;
3284 }
3285 
3286 static long tun_chr_ioctl(struct file *file,
3287 			  unsigned int cmd, unsigned long arg)
3288 {
3289 	return __tun_chr_ioctl(file, cmd, arg, sizeof (struct ifreq));
3290 }
3291 
3292 #ifdef CONFIG_COMPAT
3293 static long tun_chr_compat_ioctl(struct file *file,
3294 			 unsigned int cmd, unsigned long arg)
3295 {
3296 	switch (cmd) {
3297 	case TUNSETIFF:
3298 	case TUNGETIFF:
3299 	case TUNSETTXFILTER:
3300 	case TUNGETSNDBUF:
3301 	case TUNSETSNDBUF:
3302 	case SIOCGIFHWADDR:
3303 	case SIOCSIFHWADDR:
3304 		arg = (unsigned long)compat_ptr(arg);
3305 		break;
3306 	default:
3307 		arg = (compat_ulong_t)arg;
3308 		break;
3309 	}
3310 
3311 	/*
3312 	 * compat_ifreq is shorter than ifreq, so we must not access beyond
3313 	 * the end of that structure. All fields that are used in this
3314 	 * driver are compatible though, we don't need to convert the
3315 	 * contents.
3316 	 */
3317 	return __tun_chr_ioctl(file, cmd, arg, sizeof(struct compat_ifreq));
3318 }
3319 #endif /* CONFIG_COMPAT */
3320 
3321 static int tun_chr_fasync(int fd, struct file *file, int on)
3322 {
3323 	struct tun_file *tfile = file->private_data;
3324 	int ret;
3325 
3326 	if ((ret = fasync_helper(fd, file, on, &tfile->fasync)) < 0)
3327 		goto out;
3328 
3329 	if (on) {
3330 		__f_setown(file, task_pid(current), PIDTYPE_TGID, 0);
3331 		tfile->flags |= TUN_FASYNC;
3332 	} else
3333 		tfile->flags &= ~TUN_FASYNC;
3334 	ret = 0;
3335 out:
3336 	return ret;
3337 }
3338 
3339 static int tun_chr_open(struct inode *inode, struct file * file)
3340 {
3341 	struct net *net = current->nsproxy->net_ns;
3342 	struct tun_file *tfile;
3343 
3344 	DBG1(KERN_INFO, "tunX: tun_chr_open\n");
3345 
3346 	tfile = (struct tun_file *)sk_alloc(net, AF_UNSPEC, GFP_KERNEL,
3347 					    &tun_proto, 0);
3348 	if (!tfile)
3349 		return -ENOMEM;
3350 	if (ptr_ring_init(&tfile->tx_ring, 0, GFP_KERNEL)) {
3351 		sk_free(&tfile->sk);
3352 		return -ENOMEM;
3353 	}
3354 
3355 	mutex_init(&tfile->napi_mutex);
3356 	RCU_INIT_POINTER(tfile->tun, NULL);
3357 	tfile->flags = 0;
3358 	tfile->ifindex = 0;
3359 
3360 	init_waitqueue_head(&tfile->wq.wait);
3361 	RCU_INIT_POINTER(tfile->socket.wq, &tfile->wq);
3362 
3363 	tfile->socket.file = file;
3364 	tfile->socket.ops = &tun_socket_ops;
3365 
3366 	sock_init_data(&tfile->socket, &tfile->sk);
3367 
3368 	tfile->sk.sk_write_space = tun_sock_write_space;
3369 	tfile->sk.sk_sndbuf = INT_MAX;
3370 
3371 	file->private_data = tfile;
3372 	INIT_LIST_HEAD(&tfile->next);
3373 
3374 	sock_set_flag(&tfile->sk, SOCK_ZEROCOPY);
3375 
3376 	return 0;
3377 }
3378 
3379 static int tun_chr_close(struct inode *inode, struct file *file)
3380 {
3381 	struct tun_file *tfile = file->private_data;
3382 
3383 	tun_detach(tfile, true);
3384 
3385 	return 0;
3386 }
3387 
3388 #ifdef CONFIG_PROC_FS
3389 static void tun_chr_show_fdinfo(struct seq_file *m, struct file *file)
3390 {
3391 	struct tun_file *tfile = file->private_data;
3392 	struct tun_struct *tun;
3393 	struct ifreq ifr;
3394 
3395 	memset(&ifr, 0, sizeof(ifr));
3396 
3397 	rtnl_lock();
3398 	tun = tun_get(tfile);
3399 	if (tun)
3400 		tun_get_iff(current->nsproxy->net_ns, tun, &ifr);
3401 	rtnl_unlock();
3402 
3403 	if (tun)
3404 		tun_put(tun);
3405 
3406 	seq_printf(m, "iff:\t%s\n", ifr.ifr_name);
3407 }
3408 #endif
3409 
3410 static const struct file_operations tun_fops = {
3411 	.owner	= THIS_MODULE,
3412 	.llseek = no_llseek,
3413 	.read_iter  = tun_chr_read_iter,
3414 	.write_iter = tun_chr_write_iter,
3415 	.poll	= tun_chr_poll,
3416 	.unlocked_ioctl	= tun_chr_ioctl,
3417 #ifdef CONFIG_COMPAT
3418 	.compat_ioctl = tun_chr_compat_ioctl,
3419 #endif
3420 	.open	= tun_chr_open,
3421 	.release = tun_chr_close,
3422 	.fasync = tun_chr_fasync,
3423 #ifdef CONFIG_PROC_FS
3424 	.show_fdinfo = tun_chr_show_fdinfo,
3425 #endif
3426 };
3427 
3428 static struct miscdevice tun_miscdev = {
3429 	.minor = TUN_MINOR,
3430 	.name = "tun",
3431 	.nodename = "net/tun",
3432 	.fops = &tun_fops,
3433 };
3434 
3435 /* ethtool interface */
3436 
3437 static void tun_default_link_ksettings(struct net_device *dev,
3438 				       struct ethtool_link_ksettings *cmd)
3439 {
3440 	ethtool_link_ksettings_zero_link_mode(cmd, supported);
3441 	ethtool_link_ksettings_zero_link_mode(cmd, advertising);
3442 	cmd->base.speed		= SPEED_10;
3443 	cmd->base.duplex	= DUPLEX_FULL;
3444 	cmd->base.port		= PORT_TP;
3445 	cmd->base.phy_address	= 0;
3446 	cmd->base.autoneg	= AUTONEG_DISABLE;
3447 }
3448 
3449 static int tun_get_link_ksettings(struct net_device *dev,
3450 				  struct ethtool_link_ksettings *cmd)
3451 {
3452 	struct tun_struct *tun = netdev_priv(dev);
3453 
3454 	memcpy(cmd, &tun->link_ksettings, sizeof(*cmd));
3455 	return 0;
3456 }
3457 
3458 static int tun_set_link_ksettings(struct net_device *dev,
3459 				  const struct ethtool_link_ksettings *cmd)
3460 {
3461 	struct tun_struct *tun = netdev_priv(dev);
3462 
3463 	memcpy(&tun->link_ksettings, cmd, sizeof(*cmd));
3464 	return 0;
3465 }
3466 
3467 static void tun_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
3468 {
3469 	struct tun_struct *tun = netdev_priv(dev);
3470 
3471 	strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
3472 	strlcpy(info->version, DRV_VERSION, sizeof(info->version));
3473 
3474 	switch (tun->flags & TUN_TYPE_MASK) {
3475 	case IFF_TUN:
3476 		strlcpy(info->bus_info, "tun", sizeof(info->bus_info));
3477 		break;
3478 	case IFF_TAP:
3479 		strlcpy(info->bus_info, "tap", sizeof(info->bus_info));
3480 		break;
3481 	}
3482 }
3483 
3484 static u32 tun_get_msglevel(struct net_device *dev)
3485 {
3486 #ifdef TUN_DEBUG
3487 	struct tun_struct *tun = netdev_priv(dev);
3488 	return tun->debug;
3489 #else
3490 	return -EOPNOTSUPP;
3491 #endif
3492 }
3493 
3494 static void tun_set_msglevel(struct net_device *dev, u32 value)
3495 {
3496 #ifdef TUN_DEBUG
3497 	struct tun_struct *tun = netdev_priv(dev);
3498 	tun->debug = value;
3499 #endif
3500 }
3501 
3502 static int tun_get_coalesce(struct net_device *dev,
3503 			    struct ethtool_coalesce *ec)
3504 {
3505 	struct tun_struct *tun = netdev_priv(dev);
3506 
3507 	ec->rx_max_coalesced_frames = tun->rx_batched;
3508 
3509 	return 0;
3510 }
3511 
3512 static int tun_set_coalesce(struct net_device *dev,
3513 			    struct ethtool_coalesce *ec)
3514 {
3515 	struct tun_struct *tun = netdev_priv(dev);
3516 
3517 	if (ec->rx_max_coalesced_frames > NAPI_POLL_WEIGHT)
3518 		tun->rx_batched = NAPI_POLL_WEIGHT;
3519 	else
3520 		tun->rx_batched = ec->rx_max_coalesced_frames;
3521 
3522 	return 0;
3523 }
3524 
3525 static const struct ethtool_ops tun_ethtool_ops = {
3526 	.get_drvinfo	= tun_get_drvinfo,
3527 	.get_msglevel	= tun_get_msglevel,
3528 	.set_msglevel	= tun_set_msglevel,
3529 	.get_link	= ethtool_op_get_link,
3530 	.get_ts_info	= ethtool_op_get_ts_info,
3531 	.get_coalesce   = tun_get_coalesce,
3532 	.set_coalesce   = tun_set_coalesce,
3533 	.get_link_ksettings = tun_get_link_ksettings,
3534 	.set_link_ksettings = tun_set_link_ksettings,
3535 };
3536 
3537 static int tun_queue_resize(struct tun_struct *tun)
3538 {
3539 	struct net_device *dev = tun->dev;
3540 	struct tun_file *tfile;
3541 	struct ptr_ring **rings;
3542 	int n = tun->numqueues + tun->numdisabled;
3543 	int ret, i;
3544 
3545 	rings = kmalloc_array(n, sizeof(*rings), GFP_KERNEL);
3546 	if (!rings)
3547 		return -ENOMEM;
3548 
3549 	for (i = 0; i < tun->numqueues; i++) {
3550 		tfile = rtnl_dereference(tun->tfiles[i]);
3551 		rings[i] = &tfile->tx_ring;
3552 	}
3553 	list_for_each_entry(tfile, &tun->disabled, next)
3554 		rings[i++] = &tfile->tx_ring;
3555 
3556 	ret = ptr_ring_resize_multiple(rings, n,
3557 				       dev->tx_queue_len, GFP_KERNEL,
3558 				       tun_ptr_free);
3559 
3560 	kfree(rings);
3561 	return ret;
3562 }
3563 
3564 static int tun_device_event(struct notifier_block *unused,
3565 			    unsigned long event, void *ptr)
3566 {
3567 	struct net_device *dev = netdev_notifier_info_to_dev(ptr);
3568 	struct tun_struct *tun = netdev_priv(dev);
3569 
3570 	if (dev->rtnl_link_ops != &tun_link_ops)
3571 		return NOTIFY_DONE;
3572 
3573 	switch (event) {
3574 	case NETDEV_CHANGE_TX_QUEUE_LEN:
3575 		if (tun_queue_resize(tun))
3576 			return NOTIFY_BAD;
3577 		break;
3578 	default:
3579 		break;
3580 	}
3581 
3582 	return NOTIFY_DONE;
3583 }
3584 
3585 static struct notifier_block tun_notifier_block __read_mostly = {
3586 	.notifier_call	= tun_device_event,
3587 };
3588 
3589 static int __init tun_init(void)
3590 {
3591 	int ret = 0;
3592 
3593 	pr_info("%s, %s\n", DRV_DESCRIPTION, DRV_VERSION);
3594 
3595 	ret = rtnl_link_register(&tun_link_ops);
3596 	if (ret) {
3597 		pr_err("Can't register link_ops\n");
3598 		goto err_linkops;
3599 	}
3600 
3601 	ret = misc_register(&tun_miscdev);
3602 	if (ret) {
3603 		pr_err("Can't register misc device %d\n", TUN_MINOR);
3604 		goto err_misc;
3605 	}
3606 
3607 	ret = register_netdevice_notifier(&tun_notifier_block);
3608 	if (ret) {
3609 		pr_err("Can't register netdevice notifier\n");
3610 		goto err_notifier;
3611 	}
3612 
3613 	return  0;
3614 
3615 err_notifier:
3616 	misc_deregister(&tun_miscdev);
3617 err_misc:
3618 	rtnl_link_unregister(&tun_link_ops);
3619 err_linkops:
3620 	return ret;
3621 }
3622 
3623 static void tun_cleanup(void)
3624 {
3625 	misc_deregister(&tun_miscdev);
3626 	rtnl_link_unregister(&tun_link_ops);
3627 	unregister_netdevice_notifier(&tun_notifier_block);
3628 }
3629 
3630 /* Get an underlying socket object from tun file.  Returns error unless file is
3631  * attached to a device.  The returned object works like a packet socket, it
3632  * can be used for sock_sendmsg/sock_recvmsg.  The caller is responsible for
3633  * holding a reference to the file for as long as the socket is in use. */
3634 struct socket *tun_get_socket(struct file *file)
3635 {
3636 	struct tun_file *tfile;
3637 	if (file->f_op != &tun_fops)
3638 		return ERR_PTR(-EINVAL);
3639 	tfile = file->private_data;
3640 	if (!tfile)
3641 		return ERR_PTR(-EBADFD);
3642 	return &tfile->socket;
3643 }
3644 EXPORT_SYMBOL_GPL(tun_get_socket);
3645 
3646 struct ptr_ring *tun_get_tx_ring(struct file *file)
3647 {
3648 	struct tun_file *tfile;
3649 
3650 	if (file->f_op != &tun_fops)
3651 		return ERR_PTR(-EINVAL);
3652 	tfile = file->private_data;
3653 	if (!tfile)
3654 		return ERR_PTR(-EBADFD);
3655 	return &tfile->tx_ring;
3656 }
3657 EXPORT_SYMBOL_GPL(tun_get_tx_ring);
3658 
3659 module_init(tun_init);
3660 module_exit(tun_cleanup);
3661 MODULE_DESCRIPTION(DRV_DESCRIPTION);
3662 MODULE_AUTHOR(DRV_COPYRIGHT);
3663 MODULE_LICENSE("GPL");
3664 MODULE_ALIAS_MISCDEV(TUN_MINOR);
3665 MODULE_ALIAS("devname:net/tun");
3666