xref: /openbmc/linux/drivers/net/tun.c (revision 60772e48)
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 <linux/seq_file.h>
74 #include <linux/uio.h>
75 #include <linux/skb_array.h>
76 #include <linux/bpf.h>
77 #include <linux/bpf_trace.h>
78 #include <linux/mutex.h>
79 
80 #include <linux/uaccess.h>
81 #include <linux/proc_fs.h>
82 
83 /* Uncomment to enable debugging */
84 /* #define TUN_DEBUG 1 */
85 
86 #ifdef TUN_DEBUG
87 static int debug;
88 
89 #define tun_debug(level, tun, fmt, args...)			\
90 do {								\
91 	if (tun->debug)						\
92 		netdev_printk(level, tun->dev, fmt, ##args);	\
93 } while (0)
94 #define DBG1(level, fmt, args...)				\
95 do {								\
96 	if (debug == 2)						\
97 		printk(level fmt, ##args);			\
98 } while (0)
99 #else
100 #define tun_debug(level, tun, fmt, args...)			\
101 do {								\
102 	if (0)							\
103 		netdev_printk(level, tun->dev, fmt, ##args);	\
104 } while (0)
105 #define DBG1(level, fmt, args...)				\
106 do {								\
107 	if (0)							\
108 		printk(level fmt, ##args);			\
109 } while (0)
110 #endif
111 
112 #define TUN_HEADROOM 256
113 #define TUN_RX_PAD (NET_IP_ALIGN + NET_SKB_PAD)
114 
115 /* TUN device flags */
116 
117 /* IFF_ATTACH_QUEUE is never stored in device flags,
118  * overload it to mean fasync when stored there.
119  */
120 #define TUN_FASYNC	IFF_ATTACH_QUEUE
121 /* High bits in flags field are unused. */
122 #define TUN_VNET_LE     0x80000000
123 #define TUN_VNET_BE     0x40000000
124 
125 #define TUN_FEATURES (IFF_NO_PI | IFF_ONE_QUEUE | IFF_VNET_HDR | \
126 		      IFF_MULTI_QUEUE | IFF_NAPI | IFF_NAPI_FRAGS)
127 
128 #define GOODCOPY_LEN 128
129 
130 #define FLT_EXACT_COUNT 8
131 struct tap_filter {
132 	unsigned int    count;    /* Number of addrs. Zero means disabled */
133 	u32             mask[2];  /* Mask of the hashed addrs */
134 	unsigned char	addr[FLT_EXACT_COUNT][ETH_ALEN];
135 };
136 
137 /* MAX_TAP_QUEUES 256 is chosen to allow rx/tx queues to be equal
138  * to max number of VCPUs in guest. */
139 #define MAX_TAP_QUEUES 256
140 #define MAX_TAP_FLOWS  4096
141 
142 #define TUN_FLOW_EXPIRE (3 * HZ)
143 
144 struct tun_pcpu_stats {
145 	u64 rx_packets;
146 	u64 rx_bytes;
147 	u64 tx_packets;
148 	u64 tx_bytes;
149 	struct u64_stats_sync syncp;
150 	u32 rx_dropped;
151 	u32 tx_dropped;
152 	u32 rx_frame_errors;
153 };
154 
155 /* A tun_file connects an open character device to a tuntap netdevice. It
156  * also contains all socket related structures (except sock_fprog and tap_filter)
157  * to serve as one transmit queue for tuntap device. The sock_fprog and
158  * tap_filter were kept in tun_struct since they were used for filtering for the
159  * netdevice not for a specific queue (at least I didn't see the requirement for
160  * this).
161  *
162  * RCU usage:
163  * The tun_file and tun_struct are loosely coupled, the pointer from one to the
164  * other can only be read while rcu_read_lock or rtnl_lock is held.
165  */
166 struct tun_file {
167 	struct sock sk;
168 	struct socket socket;
169 	struct socket_wq wq;
170 	struct tun_struct __rcu *tun;
171 	struct fasync_struct *fasync;
172 	/* only used for fasnyc */
173 	unsigned int flags;
174 	union {
175 		u16 queue_index;
176 		unsigned int ifindex;
177 	};
178 	struct napi_struct napi;
179 	bool napi_enabled;
180 	struct mutex napi_mutex;	/* Protects access to the above napi */
181 	struct list_head next;
182 	struct tun_struct *detached;
183 	struct ptr_ring tx_ring;
184 	struct xdp_rxq_info xdp_rxq;
185 	int xdp_pending_pkts;
186 };
187 
188 struct tun_flow_entry {
189 	struct hlist_node hash_link;
190 	struct rcu_head rcu;
191 	struct tun_struct *tun;
192 
193 	u32 rxhash;
194 	u32 rps_rxhash;
195 	int queue_index;
196 	unsigned long updated;
197 };
198 
199 #define TUN_NUM_FLOW_ENTRIES 1024
200 
201 struct tun_prog {
202 	struct rcu_head rcu;
203 	struct bpf_prog *prog;
204 };
205 
206 /* Since the socket were moved to tun_file, to preserve the behavior of persist
207  * device, socket filter, sndbuf and vnet header size were restore when the
208  * file were attached to a persist device.
209  */
210 struct tun_struct {
211 	struct tun_file __rcu	*tfiles[MAX_TAP_QUEUES];
212 	unsigned int            numqueues;
213 	unsigned int 		flags;
214 	kuid_t			owner;
215 	kgid_t			group;
216 
217 	struct net_device	*dev;
218 	netdev_features_t	set_features;
219 #define TUN_USER_FEATURES (NETIF_F_HW_CSUM|NETIF_F_TSO_ECN|NETIF_F_TSO| \
220 			  NETIF_F_TSO6)
221 
222 	int			align;
223 	int			vnet_hdr_sz;
224 	int			sndbuf;
225 	struct tap_filter	txflt;
226 	struct sock_fprog	fprog;
227 	/* protected by rtnl lock */
228 	bool			filter_attached;
229 #ifdef TUN_DEBUG
230 	int debug;
231 #endif
232 	spinlock_t lock;
233 	struct hlist_head flows[TUN_NUM_FLOW_ENTRIES];
234 	struct timer_list flow_gc_timer;
235 	unsigned long ageing_time;
236 	unsigned int numdisabled;
237 	struct list_head disabled;
238 	void *security;
239 	u32 flow_count;
240 	u32 rx_batched;
241 	struct tun_pcpu_stats __percpu *pcpu_stats;
242 	struct bpf_prog __rcu *xdp_prog;
243 	struct tun_prog __rcu *steering_prog;
244 	struct tun_prog __rcu *filter_prog;
245 };
246 
247 struct veth {
248 	__be16 h_vlan_proto;
249 	__be16 h_vlan_TCI;
250 };
251 
252 bool tun_is_xdp_buff(void *ptr)
253 {
254 	return (unsigned long)ptr & TUN_XDP_FLAG;
255 }
256 EXPORT_SYMBOL(tun_is_xdp_buff);
257 
258 void *tun_xdp_to_ptr(void *ptr)
259 {
260 	return (void *)((unsigned long)ptr | TUN_XDP_FLAG);
261 }
262 EXPORT_SYMBOL(tun_xdp_to_ptr);
263 
264 void *tun_ptr_to_xdp(void *ptr)
265 {
266 	return (void *)((unsigned long)ptr & ~TUN_XDP_FLAG);
267 }
268 EXPORT_SYMBOL(tun_ptr_to_xdp);
269 
270 static int tun_napi_receive(struct napi_struct *napi, int budget)
271 {
272 	struct tun_file *tfile = container_of(napi, struct tun_file, napi);
273 	struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
274 	struct sk_buff_head process_queue;
275 	struct sk_buff *skb;
276 	int received = 0;
277 
278 	__skb_queue_head_init(&process_queue);
279 
280 	spin_lock(&queue->lock);
281 	skb_queue_splice_tail_init(queue, &process_queue);
282 	spin_unlock(&queue->lock);
283 
284 	while (received < budget && (skb = __skb_dequeue(&process_queue))) {
285 		napi_gro_receive(napi, skb);
286 		++received;
287 	}
288 
289 	if (!skb_queue_empty(&process_queue)) {
290 		spin_lock(&queue->lock);
291 		skb_queue_splice(&process_queue, queue);
292 		spin_unlock(&queue->lock);
293 	}
294 
295 	return received;
296 }
297 
298 static int tun_napi_poll(struct napi_struct *napi, int budget)
299 {
300 	unsigned int received;
301 
302 	received = tun_napi_receive(napi, budget);
303 
304 	if (received < budget)
305 		napi_complete_done(napi, received);
306 
307 	return received;
308 }
309 
310 static void tun_napi_init(struct tun_struct *tun, struct tun_file *tfile,
311 			  bool napi_en)
312 {
313 	tfile->napi_enabled = napi_en;
314 	if (napi_en) {
315 		netif_napi_add(tun->dev, &tfile->napi, tun_napi_poll,
316 			       NAPI_POLL_WEIGHT);
317 		napi_enable(&tfile->napi);
318 		mutex_init(&tfile->napi_mutex);
319 	}
320 }
321 
322 static void tun_napi_disable(struct tun_struct *tun, struct tun_file *tfile)
323 {
324 	if (tfile->napi_enabled)
325 		napi_disable(&tfile->napi);
326 }
327 
328 static void tun_napi_del(struct tun_struct *tun, struct tun_file *tfile)
329 {
330 	if (tfile->napi_enabled)
331 		netif_napi_del(&tfile->napi);
332 }
333 
334 static bool tun_napi_frags_enabled(const struct tun_struct *tun)
335 {
336 	return READ_ONCE(tun->flags) & IFF_NAPI_FRAGS;
337 }
338 
339 #ifdef CONFIG_TUN_VNET_CROSS_LE
340 static inline bool tun_legacy_is_little_endian(struct tun_struct *tun)
341 {
342 	return tun->flags & TUN_VNET_BE ? false :
343 		virtio_legacy_is_little_endian();
344 }
345 
346 static long tun_get_vnet_be(struct tun_struct *tun, int __user *argp)
347 {
348 	int be = !!(tun->flags & TUN_VNET_BE);
349 
350 	if (put_user(be, argp))
351 		return -EFAULT;
352 
353 	return 0;
354 }
355 
356 static long tun_set_vnet_be(struct tun_struct *tun, int __user *argp)
357 {
358 	int be;
359 
360 	if (get_user(be, argp))
361 		return -EFAULT;
362 
363 	if (be)
364 		tun->flags |= TUN_VNET_BE;
365 	else
366 		tun->flags &= ~TUN_VNET_BE;
367 
368 	return 0;
369 }
370 #else
371 static inline bool tun_legacy_is_little_endian(struct tun_struct *tun)
372 {
373 	return virtio_legacy_is_little_endian();
374 }
375 
376 static long tun_get_vnet_be(struct tun_struct *tun, int __user *argp)
377 {
378 	return -EINVAL;
379 }
380 
381 static long tun_set_vnet_be(struct tun_struct *tun, int __user *argp)
382 {
383 	return -EINVAL;
384 }
385 #endif /* CONFIG_TUN_VNET_CROSS_LE */
386 
387 static inline bool tun_is_little_endian(struct tun_struct *tun)
388 {
389 	return tun->flags & TUN_VNET_LE ||
390 		tun_legacy_is_little_endian(tun);
391 }
392 
393 static inline u16 tun16_to_cpu(struct tun_struct *tun, __virtio16 val)
394 {
395 	return __virtio16_to_cpu(tun_is_little_endian(tun), val);
396 }
397 
398 static inline __virtio16 cpu_to_tun16(struct tun_struct *tun, u16 val)
399 {
400 	return __cpu_to_virtio16(tun_is_little_endian(tun), val);
401 }
402 
403 static inline u32 tun_hashfn(u32 rxhash)
404 {
405 	return rxhash & 0x3ff;
406 }
407 
408 static struct tun_flow_entry *tun_flow_find(struct hlist_head *head, u32 rxhash)
409 {
410 	struct tun_flow_entry *e;
411 
412 	hlist_for_each_entry_rcu(e, head, hash_link) {
413 		if (e->rxhash == rxhash)
414 			return e;
415 	}
416 	return NULL;
417 }
418 
419 static struct tun_flow_entry *tun_flow_create(struct tun_struct *tun,
420 					      struct hlist_head *head,
421 					      u32 rxhash, u16 queue_index)
422 {
423 	struct tun_flow_entry *e = kmalloc(sizeof(*e), GFP_ATOMIC);
424 
425 	if (e) {
426 		tun_debug(KERN_INFO, tun, "create flow: hash %u index %u\n",
427 			  rxhash, queue_index);
428 		e->updated = jiffies;
429 		e->rxhash = rxhash;
430 		e->rps_rxhash = 0;
431 		e->queue_index = queue_index;
432 		e->tun = tun;
433 		hlist_add_head_rcu(&e->hash_link, head);
434 		++tun->flow_count;
435 	}
436 	return e;
437 }
438 
439 static void tun_flow_delete(struct tun_struct *tun, struct tun_flow_entry *e)
440 {
441 	tun_debug(KERN_INFO, tun, "delete flow: hash %u index %u\n",
442 		  e->rxhash, e->queue_index);
443 	hlist_del_rcu(&e->hash_link);
444 	kfree_rcu(e, rcu);
445 	--tun->flow_count;
446 }
447 
448 static void tun_flow_flush(struct tun_struct *tun)
449 {
450 	int i;
451 
452 	spin_lock_bh(&tun->lock);
453 	for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
454 		struct tun_flow_entry *e;
455 		struct hlist_node *n;
456 
457 		hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link)
458 			tun_flow_delete(tun, e);
459 	}
460 	spin_unlock_bh(&tun->lock);
461 }
462 
463 static void tun_flow_delete_by_queue(struct tun_struct *tun, u16 queue_index)
464 {
465 	int i;
466 
467 	spin_lock_bh(&tun->lock);
468 	for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
469 		struct tun_flow_entry *e;
470 		struct hlist_node *n;
471 
472 		hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
473 			if (e->queue_index == queue_index)
474 				tun_flow_delete(tun, e);
475 		}
476 	}
477 	spin_unlock_bh(&tun->lock);
478 }
479 
480 static void tun_flow_cleanup(struct timer_list *t)
481 {
482 	struct tun_struct *tun = from_timer(tun, t, flow_gc_timer);
483 	unsigned long delay = tun->ageing_time;
484 	unsigned long next_timer = jiffies + delay;
485 	unsigned long count = 0;
486 	int i;
487 
488 	tun_debug(KERN_INFO, tun, "tun_flow_cleanup\n");
489 
490 	spin_lock(&tun->lock);
491 	for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
492 		struct tun_flow_entry *e;
493 		struct hlist_node *n;
494 
495 		hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
496 			unsigned long this_timer;
497 
498 			this_timer = e->updated + delay;
499 			if (time_before_eq(this_timer, jiffies)) {
500 				tun_flow_delete(tun, e);
501 				continue;
502 			}
503 			count++;
504 			if (time_before(this_timer, next_timer))
505 				next_timer = this_timer;
506 		}
507 	}
508 
509 	if (count)
510 		mod_timer(&tun->flow_gc_timer, round_jiffies_up(next_timer));
511 	spin_unlock(&tun->lock);
512 }
513 
514 static void tun_flow_update(struct tun_struct *tun, u32 rxhash,
515 			    struct tun_file *tfile)
516 {
517 	struct hlist_head *head;
518 	struct tun_flow_entry *e;
519 	unsigned long delay = tun->ageing_time;
520 	u16 queue_index = tfile->queue_index;
521 
522 	if (!rxhash)
523 		return;
524 	else
525 		head = &tun->flows[tun_hashfn(rxhash)];
526 
527 	rcu_read_lock();
528 
529 	/* We may get a very small possibility of OOO during switching, not
530 	 * worth to optimize.*/
531 	if (tun->numqueues == 1 || tfile->detached)
532 		goto unlock;
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 unlock:
553 	rcu_read_unlock();
554 }
555 
556 /**
557  * Save the hash received in the stack receive path and update the
558  * flow_hash table accordingly.
559  */
560 static inline void tun_flow_save_rps_rxhash(struct tun_flow_entry *e, u32 hash)
561 {
562 	if (unlikely(e->rps_rxhash != hash))
563 		e->rps_rxhash = hash;
564 }
565 
566 /* We try to identify a flow through its rxhash first. The reason that
567  * we do not check rxq no. is because some cards(e.g 82599), chooses
568  * the rxq based on the txq where the last packet of the flow comes. As
569  * the userspace application move between processors, we may get a
570  * different rxq no. here. If we could not get rxhash, then we would
571  * hope the rxq no. may help here.
572  */
573 static u16 tun_automq_select_queue(struct tun_struct *tun, struct sk_buff *skb)
574 {
575 	struct tun_flow_entry *e;
576 	u32 txq = 0;
577 	u32 numqueues = 0;
578 
579 	numqueues = READ_ONCE(tun->numqueues);
580 
581 	txq = __skb_get_hash_symmetric(skb);
582 	if (txq) {
583 		e = tun_flow_find(&tun->flows[tun_hashfn(txq)], txq);
584 		if (e) {
585 			tun_flow_save_rps_rxhash(e, txq);
586 			txq = e->queue_index;
587 		} else
588 			/* use multiply and shift instead of expensive divide */
589 			txq = ((u64)txq * numqueues) >> 32;
590 	} else if (likely(skb_rx_queue_recorded(skb))) {
591 		txq = skb_get_rx_queue(skb);
592 		while (unlikely(txq >= numqueues))
593 			txq -= numqueues;
594 	}
595 
596 	return txq;
597 }
598 
599 static u16 tun_ebpf_select_queue(struct tun_struct *tun, struct sk_buff *skb)
600 {
601 	struct tun_prog *prog;
602 	u16 ret = 0;
603 
604 	prog = rcu_dereference(tun->steering_prog);
605 	if (prog)
606 		ret = bpf_prog_run_clear_cb(prog->prog, skb);
607 
608 	return ret % tun->numqueues;
609 }
610 
611 static u16 tun_select_queue(struct net_device *dev, struct sk_buff *skb,
612 			    void *accel_priv, 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 static void tun_ptr_free(void *ptr)
661 {
662 	if (!ptr)
663 		return;
664 	if (tun_is_xdp_buff(ptr)) {
665 		struct xdp_buff *xdp = tun_ptr_to_xdp(ptr);
666 
667 		put_page(virt_to_head_page(xdp->data));
668 	} else {
669 		__skb_array_destroy_skb(ptr);
670 	}
671 }
672 
673 static void tun_queue_purge(struct tun_file *tfile)
674 {
675 	void *ptr;
676 
677 	while ((ptr = ptr_ring_consume(&tfile->tx_ring)) != NULL)
678 		tun_ptr_free(ptr);
679 
680 	skb_queue_purge(&tfile->sk.sk_write_queue);
681 	skb_queue_purge(&tfile->sk.sk_error_queue);
682 }
683 
684 static void tun_cleanup_tx_ring(struct tun_file *tfile)
685 {
686 	if (tfile->tx_ring.queue) {
687 		ptr_ring_cleanup(&tfile->tx_ring, tun_ptr_free);
688 		xdp_rxq_info_unreg(&tfile->xdp_rxq);
689 		memset(&tfile->tx_ring, 0, sizeof(tfile->tx_ring));
690 	}
691 }
692 
693 static void __tun_detach(struct tun_file *tfile, bool clean)
694 {
695 	struct tun_file *ntfile;
696 	struct tun_struct *tun;
697 
698 	tun = rtnl_dereference(tfile->tun);
699 
700 	if (tun && clean) {
701 		tun_napi_disable(tun, tfile);
702 		tun_napi_del(tun, tfile);
703 	}
704 
705 	if (tun && !tfile->detached) {
706 		u16 index = tfile->queue_index;
707 		BUG_ON(index >= tun->numqueues);
708 
709 		rcu_assign_pointer(tun->tfiles[index],
710 				   tun->tfiles[tun->numqueues - 1]);
711 		ntfile = rtnl_dereference(tun->tfiles[index]);
712 		ntfile->queue_index = index;
713 
714 		--tun->numqueues;
715 		if (clean) {
716 			RCU_INIT_POINTER(tfile->tun, NULL);
717 			sock_put(&tfile->sk);
718 		} else
719 			tun_disable_queue(tun, tfile);
720 
721 		synchronize_net();
722 		tun_flow_delete_by_queue(tun, tun->numqueues + 1);
723 		/* Drop read queue */
724 		tun_queue_purge(tfile);
725 		tun_set_real_num_queues(tun);
726 	} else if (tfile->detached && clean) {
727 		tun = tun_enable_queue(tfile);
728 		sock_put(&tfile->sk);
729 	}
730 
731 	if (clean) {
732 		if (tun && tun->numqueues == 0 && tun->numdisabled == 0) {
733 			netif_carrier_off(tun->dev);
734 
735 			if (!(tun->flags & IFF_PERSIST) &&
736 			    tun->dev->reg_state == NETREG_REGISTERED)
737 				unregister_netdevice(tun->dev);
738 		}
739 		tun_cleanup_tx_ring(tfile);
740 		sock_put(&tfile->sk);
741 	}
742 }
743 
744 static void tun_detach(struct tun_file *tfile, bool clean)
745 {
746 	rtnl_lock();
747 	__tun_detach(tfile, clean);
748 	rtnl_unlock();
749 }
750 
751 static void tun_detach_all(struct net_device *dev)
752 {
753 	struct tun_struct *tun = netdev_priv(dev);
754 	struct tun_file *tfile, *tmp;
755 	int i, n = tun->numqueues;
756 
757 	for (i = 0; i < n; i++) {
758 		tfile = rtnl_dereference(tun->tfiles[i]);
759 		BUG_ON(!tfile);
760 		tun_napi_disable(tun, tfile);
761 		tfile->socket.sk->sk_shutdown = RCV_SHUTDOWN;
762 		tfile->socket.sk->sk_data_ready(tfile->socket.sk);
763 		RCU_INIT_POINTER(tfile->tun, NULL);
764 		--tun->numqueues;
765 	}
766 	list_for_each_entry(tfile, &tun->disabled, next) {
767 		tfile->socket.sk->sk_shutdown = RCV_SHUTDOWN;
768 		tfile->socket.sk->sk_data_ready(tfile->socket.sk);
769 		RCU_INIT_POINTER(tfile->tun, NULL);
770 	}
771 	BUG_ON(tun->numqueues != 0);
772 
773 	synchronize_net();
774 	for (i = 0; i < n; i++) {
775 		tfile = rtnl_dereference(tun->tfiles[i]);
776 		tun_napi_del(tun, tfile);
777 		/* Drop read queue */
778 		tun_queue_purge(tfile);
779 		sock_put(&tfile->sk);
780 		tun_cleanup_tx_ring(tfile);
781 	}
782 	list_for_each_entry_safe(tfile, tmp, &tun->disabled, next) {
783 		tun_enable_queue(tfile);
784 		tun_queue_purge(tfile);
785 		sock_put(&tfile->sk);
786 		tun_cleanup_tx_ring(tfile);
787 	}
788 	BUG_ON(tun->numdisabled != 0);
789 
790 	if (tun->flags & IFF_PERSIST)
791 		module_put(THIS_MODULE);
792 }
793 
794 static int tun_attach(struct tun_struct *tun, struct file *file,
795 		      bool skip_filter, bool napi)
796 {
797 	struct tun_file *tfile = file->private_data;
798 	struct net_device *dev = tun->dev;
799 	int err;
800 
801 	err = security_tun_dev_attach(tfile->socket.sk, tun->security);
802 	if (err < 0)
803 		goto out;
804 
805 	err = -EINVAL;
806 	if (rtnl_dereference(tfile->tun) && !tfile->detached)
807 		goto out;
808 
809 	err = -EBUSY;
810 	if (!(tun->flags & IFF_MULTI_QUEUE) && tun->numqueues == 1)
811 		goto out;
812 
813 	err = -E2BIG;
814 	if (!tfile->detached &&
815 	    tun->numqueues + tun->numdisabled == MAX_TAP_QUEUES)
816 		goto out;
817 
818 	err = 0;
819 
820 	/* Re-attach the filter to persist device */
821 	if (!skip_filter && (tun->filter_attached == true)) {
822 		lock_sock(tfile->socket.sk);
823 		err = sk_attach_filter(&tun->fprog, tfile->socket.sk);
824 		release_sock(tfile->socket.sk);
825 		if (!err)
826 			goto out;
827 	}
828 
829 	if (!tfile->detached &&
830 	    ptr_ring_init(&tfile->tx_ring, dev->tx_queue_len, GFP_KERNEL)) {
831 		err = -ENOMEM;
832 		goto out;
833 	}
834 
835 	tfile->queue_index = tun->numqueues;
836 	tfile->socket.sk->sk_shutdown &= ~RCV_SHUTDOWN;
837 
838 	if (tfile->detached) {
839 		/* Re-attach detached tfile, updating XDP queue_index */
840 		WARN_ON(!xdp_rxq_info_is_reg(&tfile->xdp_rxq));
841 
842 		if (tfile->xdp_rxq.queue_index    != tfile->queue_index)
843 			tfile->xdp_rxq.queue_index = tfile->queue_index;
844 	} else {
845 		/* Setup XDP RX-queue info, for new tfile getting attached */
846 		err = xdp_rxq_info_reg(&tfile->xdp_rxq,
847 				       tun->dev, tfile->queue_index);
848 		if (err < 0)
849 			goto out;
850 		err = 0;
851 	}
852 
853 	rcu_assign_pointer(tfile->tun, tun);
854 	rcu_assign_pointer(tun->tfiles[tun->numqueues], tfile);
855 	tun->numqueues++;
856 
857 	if (tfile->detached) {
858 		tun_enable_queue(tfile);
859 	} else {
860 		sock_hold(&tfile->sk);
861 		tun_napi_init(tun, tfile, napi);
862 	}
863 
864 	tun_set_real_num_queues(tun);
865 
866 	/* device is allowed to go away first, so no need to hold extra
867 	 * refcnt.
868 	 */
869 
870 out:
871 	return err;
872 }
873 
874 static struct tun_struct *tun_get(struct tun_file *tfile)
875 {
876 	struct tun_struct *tun;
877 
878 	rcu_read_lock();
879 	tun = rcu_dereference(tfile->tun);
880 	if (tun)
881 		dev_hold(tun->dev);
882 	rcu_read_unlock();
883 
884 	return tun;
885 }
886 
887 static void tun_put(struct tun_struct *tun)
888 {
889 	dev_put(tun->dev);
890 }
891 
892 /* TAP filtering */
893 static void addr_hash_set(u32 *mask, const u8 *addr)
894 {
895 	int n = ether_crc(ETH_ALEN, addr) >> 26;
896 	mask[n >> 5] |= (1 << (n & 31));
897 }
898 
899 static unsigned int addr_hash_test(const u32 *mask, const u8 *addr)
900 {
901 	int n = ether_crc(ETH_ALEN, addr) >> 26;
902 	return mask[n >> 5] & (1 << (n & 31));
903 }
904 
905 static int update_filter(struct tap_filter *filter, void __user *arg)
906 {
907 	struct { u8 u[ETH_ALEN]; } *addr;
908 	struct tun_filter uf;
909 	int err, alen, n, nexact;
910 
911 	if (copy_from_user(&uf, arg, sizeof(uf)))
912 		return -EFAULT;
913 
914 	if (!uf.count) {
915 		/* Disabled */
916 		filter->count = 0;
917 		return 0;
918 	}
919 
920 	alen = ETH_ALEN * uf.count;
921 	addr = memdup_user(arg + sizeof(uf), alen);
922 	if (IS_ERR(addr))
923 		return PTR_ERR(addr);
924 
925 	/* The filter is updated without holding any locks. Which is
926 	 * perfectly safe. We disable it first and in the worst
927 	 * case we'll accept a few undesired packets. */
928 	filter->count = 0;
929 	wmb();
930 
931 	/* Use first set of addresses as an exact filter */
932 	for (n = 0; n < uf.count && n < FLT_EXACT_COUNT; n++)
933 		memcpy(filter->addr[n], addr[n].u, ETH_ALEN);
934 
935 	nexact = n;
936 
937 	/* Remaining multicast addresses are hashed,
938 	 * unicast will leave the filter disabled. */
939 	memset(filter->mask, 0, sizeof(filter->mask));
940 	for (; n < uf.count; n++) {
941 		if (!is_multicast_ether_addr(addr[n].u)) {
942 			err = 0; /* no filter */
943 			goto free_addr;
944 		}
945 		addr_hash_set(filter->mask, addr[n].u);
946 	}
947 
948 	/* For ALLMULTI just set the mask to all ones.
949 	 * This overrides the mask populated above. */
950 	if ((uf.flags & TUN_FLT_ALLMULTI))
951 		memset(filter->mask, ~0, sizeof(filter->mask));
952 
953 	/* Now enable the filter */
954 	wmb();
955 	filter->count = nexact;
956 
957 	/* Return the number of exact filters */
958 	err = nexact;
959 free_addr:
960 	kfree(addr);
961 	return err;
962 }
963 
964 /* Returns: 0 - drop, !=0 - accept */
965 static int run_filter(struct tap_filter *filter, const struct sk_buff *skb)
966 {
967 	/* Cannot use eth_hdr(skb) here because skb_mac_hdr() is incorrect
968 	 * at this point. */
969 	struct ethhdr *eh = (struct ethhdr *) skb->data;
970 	int i;
971 
972 	/* Exact match */
973 	for (i = 0; i < filter->count; i++)
974 		if (ether_addr_equal(eh->h_dest, filter->addr[i]))
975 			return 1;
976 
977 	/* Inexact match (multicast only) */
978 	if (is_multicast_ether_addr(eh->h_dest))
979 		return addr_hash_test(filter->mask, eh->h_dest);
980 
981 	return 0;
982 }
983 
984 /*
985  * Checks whether the packet is accepted or not.
986  * Returns: 0 - drop, !=0 - accept
987  */
988 static int check_filter(struct tap_filter *filter, const struct sk_buff *skb)
989 {
990 	if (!filter->count)
991 		return 1;
992 
993 	return run_filter(filter, skb);
994 }
995 
996 /* Network device part of the driver */
997 
998 static const struct ethtool_ops tun_ethtool_ops;
999 
1000 /* Net device detach from fd. */
1001 static void tun_net_uninit(struct net_device *dev)
1002 {
1003 	tun_detach_all(dev);
1004 }
1005 
1006 /* Net device open. */
1007 static int tun_net_open(struct net_device *dev)
1008 {
1009 	struct tun_struct *tun = netdev_priv(dev);
1010 	int i;
1011 
1012 	netif_tx_start_all_queues(dev);
1013 
1014 	for (i = 0; i < tun->numqueues; i++) {
1015 		struct tun_file *tfile;
1016 
1017 		tfile = rtnl_dereference(tun->tfiles[i]);
1018 		tfile->socket.sk->sk_write_space(tfile->socket.sk);
1019 	}
1020 
1021 	return 0;
1022 }
1023 
1024 /* Net device close. */
1025 static int tun_net_close(struct net_device *dev)
1026 {
1027 	netif_tx_stop_all_queues(dev);
1028 	return 0;
1029 }
1030 
1031 /* Net device start xmit */
1032 static void tun_automq_xmit(struct tun_struct *tun, struct sk_buff *skb)
1033 {
1034 #ifdef CONFIG_RPS
1035 	if (tun->numqueues == 1 && static_key_false(&rps_needed)) {
1036 		/* Select queue was not called for the skbuff, so we extract the
1037 		 * RPS hash and save it into the flow_table here.
1038 		 */
1039 		__u32 rxhash;
1040 
1041 		rxhash = __skb_get_hash_symmetric(skb);
1042 		if (rxhash) {
1043 			struct tun_flow_entry *e;
1044 			e = tun_flow_find(&tun->flows[tun_hashfn(rxhash)],
1045 					rxhash);
1046 			if (e)
1047 				tun_flow_save_rps_rxhash(e, rxhash);
1048 		}
1049 	}
1050 #endif
1051 }
1052 
1053 static unsigned int run_ebpf_filter(struct tun_struct *tun,
1054 				    struct sk_buff *skb,
1055 				    int len)
1056 {
1057 	struct tun_prog *prog = rcu_dereference(tun->filter_prog);
1058 
1059 	if (prog)
1060 		len = bpf_prog_run_clear_cb(prog->prog, skb);
1061 
1062 	return len;
1063 }
1064 
1065 /* Net device start xmit */
1066 static netdev_tx_t tun_net_xmit(struct sk_buff *skb, struct net_device *dev)
1067 {
1068 	struct tun_struct *tun = netdev_priv(dev);
1069 	int txq = skb->queue_mapping;
1070 	struct tun_file *tfile;
1071 	int len = skb->len;
1072 
1073 	rcu_read_lock();
1074 	tfile = rcu_dereference(tun->tfiles[txq]);
1075 
1076 	/* Drop packet if interface is not attached */
1077 	if (txq >= tun->numqueues)
1078 		goto drop;
1079 
1080 	if (!rcu_dereference(tun->steering_prog))
1081 		tun_automq_xmit(tun, skb);
1082 
1083 	tun_debug(KERN_INFO, tun, "tun_net_xmit %d\n", skb->len);
1084 
1085 	BUG_ON(!tfile);
1086 
1087 	/* Drop if the filter does not like it.
1088 	 * This is a noop if the filter is disabled.
1089 	 * Filter can be enabled only for the TAP devices. */
1090 	if (!check_filter(&tun->txflt, skb))
1091 		goto drop;
1092 
1093 	if (tfile->socket.sk->sk_filter &&
1094 	    sk_filter(tfile->socket.sk, skb))
1095 		goto drop;
1096 
1097 	len = run_ebpf_filter(tun, skb, len);
1098 
1099 	/* Trim extra bytes since we may insert vlan proto & TCI
1100 	 * in tun_put_user().
1101 	 */
1102 	len -= skb_vlan_tag_present(skb) ? sizeof(struct veth) : 0;
1103 	if (len <= 0 || pskb_trim(skb, len))
1104 		goto drop;
1105 
1106 	if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC)))
1107 		goto drop;
1108 
1109 	skb_tx_timestamp(skb);
1110 
1111 	/* Orphan the skb - required as we might hang on to it
1112 	 * for indefinite time.
1113 	 */
1114 	skb_orphan(skb);
1115 
1116 	nf_reset(skb);
1117 
1118 	if (ptr_ring_produce(&tfile->tx_ring, skb))
1119 		goto drop;
1120 
1121 	/* Notify and wake up reader process */
1122 	if (tfile->flags & TUN_FASYNC)
1123 		kill_fasync(&tfile->fasync, SIGIO, POLL_IN);
1124 	tfile->socket.sk->sk_data_ready(tfile->socket.sk);
1125 
1126 	rcu_read_unlock();
1127 	return NETDEV_TX_OK;
1128 
1129 drop:
1130 	this_cpu_inc(tun->pcpu_stats->tx_dropped);
1131 	skb_tx_error(skb);
1132 	kfree_skb(skb);
1133 	rcu_read_unlock();
1134 	return NET_XMIT_DROP;
1135 }
1136 
1137 static void tun_net_mclist(struct net_device *dev)
1138 {
1139 	/*
1140 	 * This callback is supposed to deal with mc filter in
1141 	 * _rx_ path and has nothing to do with the _tx_ path.
1142 	 * In rx path we always accept everything userspace gives us.
1143 	 */
1144 }
1145 
1146 static netdev_features_t tun_net_fix_features(struct net_device *dev,
1147 	netdev_features_t features)
1148 {
1149 	struct tun_struct *tun = netdev_priv(dev);
1150 
1151 	return (features & tun->set_features) | (features & ~TUN_USER_FEATURES);
1152 }
1153 #ifdef CONFIG_NET_POLL_CONTROLLER
1154 static void tun_poll_controller(struct net_device *dev)
1155 {
1156 	/*
1157 	 * Tun only receives frames when:
1158 	 * 1) the char device endpoint gets data from user space
1159 	 * 2) the tun socket gets a sendmsg call from user space
1160 	 * If NAPI is not enabled, since both of those are synchronous
1161 	 * operations, we are guaranteed never to have pending data when we poll
1162 	 * for it so there is nothing to do here but return.
1163 	 * We need this though so netpoll recognizes us as an interface that
1164 	 * supports polling, which enables bridge devices in virt setups to
1165 	 * still use netconsole
1166 	 * If NAPI is enabled, however, we need to schedule polling for all
1167 	 * queues unless we are using napi_gro_frags(), which we call in
1168 	 * process context and not in NAPI context.
1169 	 */
1170 	struct tun_struct *tun = netdev_priv(dev);
1171 
1172 	if (tun->flags & IFF_NAPI) {
1173 		struct tun_file *tfile;
1174 		int i;
1175 
1176 		if (tun_napi_frags_enabled(tun))
1177 			return;
1178 
1179 		rcu_read_lock();
1180 		for (i = 0; i < tun->numqueues; i++) {
1181 			tfile = rcu_dereference(tun->tfiles[i]);
1182 			if (tfile->napi_enabled)
1183 				napi_schedule(&tfile->napi);
1184 		}
1185 		rcu_read_unlock();
1186 	}
1187 	return;
1188 }
1189 #endif
1190 
1191 static void tun_set_headroom(struct net_device *dev, int new_hr)
1192 {
1193 	struct tun_struct *tun = netdev_priv(dev);
1194 
1195 	if (new_hr < NET_SKB_PAD)
1196 		new_hr = NET_SKB_PAD;
1197 
1198 	tun->align = new_hr;
1199 }
1200 
1201 static void
1202 tun_net_get_stats64(struct net_device *dev, struct rtnl_link_stats64 *stats)
1203 {
1204 	u32 rx_dropped = 0, tx_dropped = 0, rx_frame_errors = 0;
1205 	struct tun_struct *tun = netdev_priv(dev);
1206 	struct tun_pcpu_stats *p;
1207 	int i;
1208 
1209 	for_each_possible_cpu(i) {
1210 		u64 rxpackets, rxbytes, txpackets, txbytes;
1211 		unsigned int start;
1212 
1213 		p = per_cpu_ptr(tun->pcpu_stats, i);
1214 		do {
1215 			start = u64_stats_fetch_begin(&p->syncp);
1216 			rxpackets	= p->rx_packets;
1217 			rxbytes		= p->rx_bytes;
1218 			txpackets	= p->tx_packets;
1219 			txbytes		= p->tx_bytes;
1220 		} while (u64_stats_fetch_retry(&p->syncp, start));
1221 
1222 		stats->rx_packets	+= rxpackets;
1223 		stats->rx_bytes		+= rxbytes;
1224 		stats->tx_packets	+= txpackets;
1225 		stats->tx_bytes		+= txbytes;
1226 
1227 		/* u32 counters */
1228 		rx_dropped	+= p->rx_dropped;
1229 		rx_frame_errors	+= p->rx_frame_errors;
1230 		tx_dropped	+= p->tx_dropped;
1231 	}
1232 	stats->rx_dropped  = rx_dropped;
1233 	stats->rx_frame_errors = rx_frame_errors;
1234 	stats->tx_dropped = tx_dropped;
1235 }
1236 
1237 static int tun_xdp_set(struct net_device *dev, struct bpf_prog *prog,
1238 		       struct netlink_ext_ack *extack)
1239 {
1240 	struct tun_struct *tun = netdev_priv(dev);
1241 	struct bpf_prog *old_prog;
1242 
1243 	old_prog = rtnl_dereference(tun->xdp_prog);
1244 	rcu_assign_pointer(tun->xdp_prog, prog);
1245 	if (old_prog)
1246 		bpf_prog_put(old_prog);
1247 
1248 	return 0;
1249 }
1250 
1251 static u32 tun_xdp_query(struct net_device *dev)
1252 {
1253 	struct tun_struct *tun = netdev_priv(dev);
1254 	const struct bpf_prog *xdp_prog;
1255 
1256 	xdp_prog = rtnl_dereference(tun->xdp_prog);
1257 	if (xdp_prog)
1258 		return xdp_prog->aux->id;
1259 
1260 	return 0;
1261 }
1262 
1263 static int tun_xdp(struct net_device *dev, struct netdev_bpf *xdp)
1264 {
1265 	switch (xdp->command) {
1266 	case XDP_SETUP_PROG:
1267 		return tun_xdp_set(dev, xdp->prog, xdp->extack);
1268 	case XDP_QUERY_PROG:
1269 		xdp->prog_id = tun_xdp_query(dev);
1270 		xdp->prog_attached = !!xdp->prog_id;
1271 		return 0;
1272 	default:
1273 		return -EINVAL;
1274 	}
1275 }
1276 
1277 static const struct net_device_ops tun_netdev_ops = {
1278 	.ndo_uninit		= tun_net_uninit,
1279 	.ndo_open		= tun_net_open,
1280 	.ndo_stop		= tun_net_close,
1281 	.ndo_start_xmit		= tun_net_xmit,
1282 	.ndo_fix_features	= tun_net_fix_features,
1283 	.ndo_select_queue	= tun_select_queue,
1284 #ifdef CONFIG_NET_POLL_CONTROLLER
1285 	.ndo_poll_controller	= tun_poll_controller,
1286 #endif
1287 	.ndo_set_rx_headroom	= tun_set_headroom,
1288 	.ndo_get_stats64	= tun_net_get_stats64,
1289 };
1290 
1291 static int tun_xdp_xmit(struct net_device *dev, struct xdp_buff *xdp)
1292 {
1293 	struct tun_struct *tun = netdev_priv(dev);
1294 	struct xdp_buff *buff = xdp->data_hard_start;
1295 	int headroom = xdp->data - xdp->data_hard_start;
1296 	struct tun_file *tfile;
1297 	u32 numqueues;
1298 	int ret = 0;
1299 
1300 	/* Assure headroom is available and buff is properly aligned */
1301 	if (unlikely(headroom < sizeof(*xdp) || tun_is_xdp_buff(xdp)))
1302 		return -ENOSPC;
1303 
1304 	*buff = *xdp;
1305 
1306 	rcu_read_lock();
1307 
1308 	numqueues = READ_ONCE(tun->numqueues);
1309 	if (!numqueues) {
1310 		ret = -ENOSPC;
1311 		goto out;
1312 	}
1313 
1314 	tfile = rcu_dereference(tun->tfiles[smp_processor_id() %
1315 					    numqueues]);
1316 	/* Encode the XDP flag into lowest bit for consumer to differ
1317 	 * XDP buffer from sk_buff.
1318 	 */
1319 	if (ptr_ring_produce(&tfile->tx_ring, tun_xdp_to_ptr(buff))) {
1320 		this_cpu_inc(tun->pcpu_stats->tx_dropped);
1321 		ret = -ENOSPC;
1322 	}
1323 
1324 out:
1325 	rcu_read_unlock();
1326 	return ret;
1327 }
1328 
1329 static void tun_xdp_flush(struct net_device *dev)
1330 {
1331 	struct tun_struct *tun = netdev_priv(dev);
1332 	struct tun_file *tfile;
1333 	u32 numqueues;
1334 
1335 	rcu_read_lock();
1336 
1337 	numqueues = READ_ONCE(tun->numqueues);
1338 	if (!numqueues)
1339 		goto out;
1340 
1341 	tfile = rcu_dereference(tun->tfiles[smp_processor_id() %
1342 					    numqueues]);
1343 	/* Notify and wake up reader process */
1344 	if (tfile->flags & TUN_FASYNC)
1345 		kill_fasync(&tfile->fasync, SIGIO, POLL_IN);
1346 	tfile->socket.sk->sk_data_ready(tfile->socket.sk);
1347 
1348 out:
1349 	rcu_read_unlock();
1350 }
1351 
1352 static const struct net_device_ops tap_netdev_ops = {
1353 	.ndo_uninit		= tun_net_uninit,
1354 	.ndo_open		= tun_net_open,
1355 	.ndo_stop		= tun_net_close,
1356 	.ndo_start_xmit		= tun_net_xmit,
1357 	.ndo_fix_features	= tun_net_fix_features,
1358 	.ndo_set_rx_mode	= tun_net_mclist,
1359 	.ndo_set_mac_address	= eth_mac_addr,
1360 	.ndo_validate_addr	= eth_validate_addr,
1361 	.ndo_select_queue	= tun_select_queue,
1362 #ifdef CONFIG_NET_POLL_CONTROLLER
1363 	.ndo_poll_controller	= tun_poll_controller,
1364 #endif
1365 	.ndo_features_check	= passthru_features_check,
1366 	.ndo_set_rx_headroom	= tun_set_headroom,
1367 	.ndo_get_stats64	= tun_net_get_stats64,
1368 	.ndo_bpf		= tun_xdp,
1369 	.ndo_xdp_xmit		= tun_xdp_xmit,
1370 	.ndo_xdp_flush		= tun_xdp_flush,
1371 };
1372 
1373 static void tun_flow_init(struct tun_struct *tun)
1374 {
1375 	int i;
1376 
1377 	for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++)
1378 		INIT_HLIST_HEAD(&tun->flows[i]);
1379 
1380 	tun->ageing_time = TUN_FLOW_EXPIRE;
1381 	timer_setup(&tun->flow_gc_timer, tun_flow_cleanup, 0);
1382 	mod_timer(&tun->flow_gc_timer,
1383 		  round_jiffies_up(jiffies + tun->ageing_time));
1384 }
1385 
1386 static void tun_flow_uninit(struct tun_struct *tun)
1387 {
1388 	del_timer_sync(&tun->flow_gc_timer);
1389 	tun_flow_flush(tun);
1390 }
1391 
1392 #define MIN_MTU 68
1393 #define MAX_MTU 65535
1394 
1395 /* Initialize net device. */
1396 static void tun_net_init(struct net_device *dev)
1397 {
1398 	struct tun_struct *tun = netdev_priv(dev);
1399 
1400 	switch (tun->flags & TUN_TYPE_MASK) {
1401 	case IFF_TUN:
1402 		dev->netdev_ops = &tun_netdev_ops;
1403 
1404 		/* Point-to-Point TUN Device */
1405 		dev->hard_header_len = 0;
1406 		dev->addr_len = 0;
1407 		dev->mtu = 1500;
1408 
1409 		/* Zero header length */
1410 		dev->type = ARPHRD_NONE;
1411 		dev->flags = IFF_POINTOPOINT | IFF_NOARP | IFF_MULTICAST;
1412 		break;
1413 
1414 	case IFF_TAP:
1415 		dev->netdev_ops = &tap_netdev_ops;
1416 		/* Ethernet TAP Device */
1417 		ether_setup(dev);
1418 		dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1419 		dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
1420 
1421 		eth_hw_addr_random(dev);
1422 
1423 		break;
1424 	}
1425 
1426 	dev->min_mtu = MIN_MTU;
1427 	dev->max_mtu = MAX_MTU - dev->hard_header_len;
1428 }
1429 
1430 /* Character device part */
1431 
1432 /* Poll */
1433 static __poll_t tun_chr_poll(struct file *file, poll_table *wait)
1434 {
1435 	struct tun_file *tfile = file->private_data;
1436 	struct tun_struct *tun = tun_get(tfile);
1437 	struct sock *sk;
1438 	__poll_t mask = 0;
1439 
1440 	if (!tun)
1441 		return EPOLLERR;
1442 
1443 	sk = tfile->socket.sk;
1444 
1445 	tun_debug(KERN_INFO, tun, "tun_chr_poll\n");
1446 
1447 	poll_wait(file, sk_sleep(sk), wait);
1448 
1449 	if (!ptr_ring_empty(&tfile->tx_ring))
1450 		mask |= EPOLLIN | EPOLLRDNORM;
1451 
1452 	if (tun->dev->flags & IFF_UP &&
1453 	    (sock_writeable(sk) ||
1454 	     (!test_and_set_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags) &&
1455 	      sock_writeable(sk))))
1456 		mask |= EPOLLOUT | EPOLLWRNORM;
1457 
1458 	if (tun->dev->reg_state != NETREG_REGISTERED)
1459 		mask = EPOLLERR;
1460 
1461 	tun_put(tun);
1462 	return mask;
1463 }
1464 
1465 static struct sk_buff *tun_napi_alloc_frags(struct tun_file *tfile,
1466 					    size_t len,
1467 					    const struct iov_iter *it)
1468 {
1469 	struct sk_buff *skb;
1470 	size_t linear;
1471 	int err;
1472 	int i;
1473 
1474 	if (it->nr_segs > MAX_SKB_FRAGS + 1)
1475 		return ERR_PTR(-ENOMEM);
1476 
1477 	local_bh_disable();
1478 	skb = napi_get_frags(&tfile->napi);
1479 	local_bh_enable();
1480 	if (!skb)
1481 		return ERR_PTR(-ENOMEM);
1482 
1483 	linear = iov_iter_single_seg_count(it);
1484 	err = __skb_grow(skb, linear);
1485 	if (err)
1486 		goto free;
1487 
1488 	skb->len = len;
1489 	skb->data_len = len - linear;
1490 	skb->truesize += skb->data_len;
1491 
1492 	for (i = 1; i < it->nr_segs; i++) {
1493 		struct page_frag *pfrag = &current->task_frag;
1494 		size_t fragsz = it->iov[i].iov_len;
1495 
1496 		if (fragsz == 0 || fragsz > PAGE_SIZE) {
1497 			err = -EINVAL;
1498 			goto free;
1499 		}
1500 
1501 		if (!skb_page_frag_refill(fragsz, pfrag, GFP_KERNEL)) {
1502 			err = -ENOMEM;
1503 			goto free;
1504 		}
1505 
1506 		skb_fill_page_desc(skb, i - 1, pfrag->page,
1507 				   pfrag->offset, fragsz);
1508 		page_ref_inc(pfrag->page);
1509 		pfrag->offset += fragsz;
1510 	}
1511 
1512 	return skb;
1513 free:
1514 	/* frees skb and all frags allocated with napi_alloc_frag() */
1515 	napi_free_frags(&tfile->napi);
1516 	return ERR_PTR(err);
1517 }
1518 
1519 /* prepad is the amount to reserve at front.  len is length after that.
1520  * linear is a hint as to how much to copy (usually headers). */
1521 static struct sk_buff *tun_alloc_skb(struct tun_file *tfile,
1522 				     size_t prepad, size_t len,
1523 				     size_t linear, int noblock)
1524 {
1525 	struct sock *sk = tfile->socket.sk;
1526 	struct sk_buff *skb;
1527 	int err;
1528 
1529 	/* Under a page?  Don't bother with paged skb. */
1530 	if (prepad + len < PAGE_SIZE || !linear)
1531 		linear = len;
1532 
1533 	skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
1534 				   &err, 0);
1535 	if (!skb)
1536 		return ERR_PTR(err);
1537 
1538 	skb_reserve(skb, prepad);
1539 	skb_put(skb, linear);
1540 	skb->data_len = len - linear;
1541 	skb->len += len - linear;
1542 
1543 	return skb;
1544 }
1545 
1546 static void tun_rx_batched(struct tun_struct *tun, struct tun_file *tfile,
1547 			   struct sk_buff *skb, int more)
1548 {
1549 	struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
1550 	struct sk_buff_head process_queue;
1551 	u32 rx_batched = tun->rx_batched;
1552 	bool rcv = false;
1553 
1554 	if (!rx_batched || (!more && skb_queue_empty(queue))) {
1555 		local_bh_disable();
1556 		netif_receive_skb(skb);
1557 		local_bh_enable();
1558 		return;
1559 	}
1560 
1561 	spin_lock(&queue->lock);
1562 	if (!more || skb_queue_len(queue) == rx_batched) {
1563 		__skb_queue_head_init(&process_queue);
1564 		skb_queue_splice_tail_init(queue, &process_queue);
1565 		rcv = true;
1566 	} else {
1567 		__skb_queue_tail(queue, skb);
1568 	}
1569 	spin_unlock(&queue->lock);
1570 
1571 	if (rcv) {
1572 		struct sk_buff *nskb;
1573 
1574 		local_bh_disable();
1575 		while ((nskb = __skb_dequeue(&process_queue)))
1576 			netif_receive_skb(nskb);
1577 		netif_receive_skb(skb);
1578 		local_bh_enable();
1579 	}
1580 }
1581 
1582 static bool tun_can_build_skb(struct tun_struct *tun, struct tun_file *tfile,
1583 			      int len, int noblock, bool zerocopy)
1584 {
1585 	if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
1586 		return false;
1587 
1588 	if (tfile->socket.sk->sk_sndbuf != INT_MAX)
1589 		return false;
1590 
1591 	if (!noblock)
1592 		return false;
1593 
1594 	if (zerocopy)
1595 		return false;
1596 
1597 	if (SKB_DATA_ALIGN(len + TUN_RX_PAD) +
1598 	    SKB_DATA_ALIGN(sizeof(struct skb_shared_info)) > PAGE_SIZE)
1599 		return false;
1600 
1601 	return true;
1602 }
1603 
1604 static struct sk_buff *tun_build_skb(struct tun_struct *tun,
1605 				     struct tun_file *tfile,
1606 				     struct iov_iter *from,
1607 				     struct virtio_net_hdr *hdr,
1608 				     int len, int *skb_xdp)
1609 {
1610 	struct page_frag *alloc_frag = &current->task_frag;
1611 	struct sk_buff *skb;
1612 	struct bpf_prog *xdp_prog;
1613 	int buflen = SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
1614 	unsigned int delta = 0;
1615 	char *buf;
1616 	size_t copied;
1617 	bool xdp_xmit = false;
1618 	int err, pad = TUN_RX_PAD;
1619 
1620 	rcu_read_lock();
1621 	xdp_prog = rcu_dereference(tun->xdp_prog);
1622 	if (xdp_prog)
1623 		pad += TUN_HEADROOM;
1624 	buflen += SKB_DATA_ALIGN(len + pad);
1625 	rcu_read_unlock();
1626 
1627 	alloc_frag->offset = ALIGN((u64)alloc_frag->offset, SMP_CACHE_BYTES);
1628 	if (unlikely(!skb_page_frag_refill(buflen, alloc_frag, GFP_KERNEL)))
1629 		return ERR_PTR(-ENOMEM);
1630 
1631 	buf = (char *)page_address(alloc_frag->page) + alloc_frag->offset;
1632 	copied = copy_page_from_iter(alloc_frag->page,
1633 				     alloc_frag->offset + pad,
1634 				     len, from);
1635 	if (copied != len)
1636 		return ERR_PTR(-EFAULT);
1637 
1638 	/* There's a small window that XDP may be set after the check
1639 	 * of xdp_prog above, this should be rare and for simplicity
1640 	 * we do XDP on skb in case the headroom is not enough.
1641 	 */
1642 	if (hdr->gso_type || !xdp_prog)
1643 		*skb_xdp = 1;
1644 	else
1645 		*skb_xdp = 0;
1646 
1647 	rcu_read_lock();
1648 	xdp_prog = rcu_dereference(tun->xdp_prog);
1649 	if (xdp_prog && !*skb_xdp) {
1650 		struct xdp_buff xdp;
1651 		void *orig_data;
1652 		u32 act;
1653 
1654 		xdp.data_hard_start = buf;
1655 		xdp.data = buf + pad;
1656 		xdp_set_data_meta_invalid(&xdp);
1657 		xdp.data_end = xdp.data + len;
1658 		xdp.rxq = &tfile->xdp_rxq;
1659 		orig_data = xdp.data;
1660 		act = bpf_prog_run_xdp(xdp_prog, &xdp);
1661 
1662 		switch (act) {
1663 		case XDP_REDIRECT:
1664 			get_page(alloc_frag->page);
1665 			alloc_frag->offset += buflen;
1666 			++tfile->xdp_pending_pkts;
1667 			err = xdp_do_redirect(tun->dev, &xdp, xdp_prog);
1668 			if (err)
1669 				goto err_redirect;
1670 			rcu_read_unlock();
1671 			return NULL;
1672 		case XDP_TX:
1673 			xdp_xmit = true;
1674 			/* fall through */
1675 		case XDP_PASS:
1676 			delta = orig_data - xdp.data;
1677 			break;
1678 		default:
1679 			bpf_warn_invalid_xdp_action(act);
1680 			/* fall through */
1681 		case XDP_ABORTED:
1682 			trace_xdp_exception(tun->dev, xdp_prog, act);
1683 			/* fall through */
1684 		case XDP_DROP:
1685 			goto err_xdp;
1686 		}
1687 	}
1688 
1689 	skb = build_skb(buf, buflen);
1690 	if (!skb) {
1691 		rcu_read_unlock();
1692 		return ERR_PTR(-ENOMEM);
1693 	}
1694 
1695 	skb_reserve(skb, pad - delta);
1696 	skb_put(skb, len + delta);
1697 	get_page(alloc_frag->page);
1698 	alloc_frag->offset += buflen;
1699 
1700 	if (xdp_xmit) {
1701 		skb->dev = tun->dev;
1702 		generic_xdp_tx(skb, xdp_prog);
1703 		rcu_read_unlock();
1704 		return NULL;
1705 	}
1706 
1707 	rcu_read_unlock();
1708 
1709 	return skb;
1710 
1711 err_redirect:
1712 	put_page(alloc_frag->page);
1713 err_xdp:
1714 	rcu_read_unlock();
1715 	this_cpu_inc(tun->pcpu_stats->rx_dropped);
1716 	return NULL;
1717 }
1718 
1719 /* Get packet from user space buffer */
1720 static ssize_t tun_get_user(struct tun_struct *tun, struct tun_file *tfile,
1721 			    void *msg_control, struct iov_iter *from,
1722 			    int noblock, bool more)
1723 {
1724 	struct tun_pi pi = { 0, cpu_to_be16(ETH_P_IP) };
1725 	struct sk_buff *skb;
1726 	size_t total_len = iov_iter_count(from);
1727 	size_t len = total_len, align = tun->align, linear;
1728 	struct virtio_net_hdr gso = { 0 };
1729 	struct tun_pcpu_stats *stats;
1730 	int good_linear;
1731 	int copylen;
1732 	bool zerocopy = false;
1733 	int err;
1734 	u32 rxhash = 0;
1735 	int skb_xdp = 1;
1736 	bool frags = tun_napi_frags_enabled(tun);
1737 
1738 	if (!(tun->dev->flags & IFF_UP))
1739 		return -EIO;
1740 
1741 	if (!(tun->flags & IFF_NO_PI)) {
1742 		if (len < sizeof(pi))
1743 			return -EINVAL;
1744 		len -= sizeof(pi);
1745 
1746 		if (!copy_from_iter_full(&pi, sizeof(pi), from))
1747 			return -EFAULT;
1748 	}
1749 
1750 	if (tun->flags & IFF_VNET_HDR) {
1751 		int vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
1752 
1753 		if (len < vnet_hdr_sz)
1754 			return -EINVAL;
1755 		len -= vnet_hdr_sz;
1756 
1757 		if (!copy_from_iter_full(&gso, sizeof(gso), from))
1758 			return -EFAULT;
1759 
1760 		if ((gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
1761 		    tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2 > tun16_to_cpu(tun, gso.hdr_len))
1762 			gso.hdr_len = cpu_to_tun16(tun, tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2);
1763 
1764 		if (tun16_to_cpu(tun, gso.hdr_len) > len)
1765 			return -EINVAL;
1766 		iov_iter_advance(from, vnet_hdr_sz - sizeof(gso));
1767 	}
1768 
1769 	if ((tun->flags & TUN_TYPE_MASK) == IFF_TAP) {
1770 		align += NET_IP_ALIGN;
1771 		if (unlikely(len < ETH_HLEN ||
1772 			     (gso.hdr_len && tun16_to_cpu(tun, gso.hdr_len) < ETH_HLEN)))
1773 			return -EINVAL;
1774 	}
1775 
1776 	good_linear = SKB_MAX_HEAD(align);
1777 
1778 	if (msg_control) {
1779 		struct iov_iter i = *from;
1780 
1781 		/* There are 256 bytes to be copied in skb, so there is
1782 		 * enough room for skb expand head in case it is used.
1783 		 * The rest of the buffer is mapped from userspace.
1784 		 */
1785 		copylen = gso.hdr_len ? tun16_to_cpu(tun, gso.hdr_len) : GOODCOPY_LEN;
1786 		if (copylen > good_linear)
1787 			copylen = good_linear;
1788 		linear = copylen;
1789 		iov_iter_advance(&i, copylen);
1790 		if (iov_iter_npages(&i, INT_MAX) <= MAX_SKB_FRAGS)
1791 			zerocopy = true;
1792 	}
1793 
1794 	if (!frags && tun_can_build_skb(tun, tfile, len, noblock, zerocopy)) {
1795 		/* For the packet that is not easy to be processed
1796 		 * (e.g gso or jumbo packet), we will do it at after
1797 		 * skb was created with generic XDP routine.
1798 		 */
1799 		skb = tun_build_skb(tun, tfile, from, &gso, len, &skb_xdp);
1800 		if (IS_ERR(skb)) {
1801 			this_cpu_inc(tun->pcpu_stats->rx_dropped);
1802 			return PTR_ERR(skb);
1803 		}
1804 		if (!skb)
1805 			return total_len;
1806 	} else {
1807 		if (!zerocopy) {
1808 			copylen = len;
1809 			if (tun16_to_cpu(tun, gso.hdr_len) > good_linear)
1810 				linear = good_linear;
1811 			else
1812 				linear = tun16_to_cpu(tun, gso.hdr_len);
1813 		}
1814 
1815 		if (frags) {
1816 			mutex_lock(&tfile->napi_mutex);
1817 			skb = tun_napi_alloc_frags(tfile, copylen, from);
1818 			/* tun_napi_alloc_frags() enforces a layout for the skb.
1819 			 * If zerocopy is enabled, then this layout will be
1820 			 * overwritten by zerocopy_sg_from_iter().
1821 			 */
1822 			zerocopy = false;
1823 		} else {
1824 			skb = tun_alloc_skb(tfile, align, copylen, linear,
1825 					    noblock);
1826 		}
1827 
1828 		if (IS_ERR(skb)) {
1829 			if (PTR_ERR(skb) != -EAGAIN)
1830 				this_cpu_inc(tun->pcpu_stats->rx_dropped);
1831 			if (frags)
1832 				mutex_unlock(&tfile->napi_mutex);
1833 			return PTR_ERR(skb);
1834 		}
1835 
1836 		if (zerocopy)
1837 			err = zerocopy_sg_from_iter(skb, from);
1838 		else
1839 			err = skb_copy_datagram_from_iter(skb, 0, from, len);
1840 
1841 		if (err) {
1842 			this_cpu_inc(tun->pcpu_stats->rx_dropped);
1843 			kfree_skb(skb);
1844 			if (frags) {
1845 				tfile->napi.skb = NULL;
1846 				mutex_unlock(&tfile->napi_mutex);
1847 			}
1848 
1849 			return -EFAULT;
1850 		}
1851 	}
1852 
1853 	if (virtio_net_hdr_to_skb(skb, &gso, tun_is_little_endian(tun))) {
1854 		this_cpu_inc(tun->pcpu_stats->rx_frame_errors);
1855 		kfree_skb(skb);
1856 		if (frags) {
1857 			tfile->napi.skb = NULL;
1858 			mutex_unlock(&tfile->napi_mutex);
1859 		}
1860 
1861 		return -EINVAL;
1862 	}
1863 
1864 	switch (tun->flags & TUN_TYPE_MASK) {
1865 	case IFF_TUN:
1866 		if (tun->flags & IFF_NO_PI) {
1867 			u8 ip_version = skb->len ? (skb->data[0] >> 4) : 0;
1868 
1869 			switch (ip_version) {
1870 			case 4:
1871 				pi.proto = htons(ETH_P_IP);
1872 				break;
1873 			case 6:
1874 				pi.proto = htons(ETH_P_IPV6);
1875 				break;
1876 			default:
1877 				this_cpu_inc(tun->pcpu_stats->rx_dropped);
1878 				kfree_skb(skb);
1879 				return -EINVAL;
1880 			}
1881 		}
1882 
1883 		skb_reset_mac_header(skb);
1884 		skb->protocol = pi.proto;
1885 		skb->dev = tun->dev;
1886 		break;
1887 	case IFF_TAP:
1888 		if (!frags)
1889 			skb->protocol = eth_type_trans(skb, tun->dev);
1890 		break;
1891 	}
1892 
1893 	/* copy skb_ubuf_info for callback when skb has no error */
1894 	if (zerocopy) {
1895 		skb_shinfo(skb)->destructor_arg = msg_control;
1896 		skb_shinfo(skb)->tx_flags |= SKBTX_DEV_ZEROCOPY;
1897 		skb_shinfo(skb)->tx_flags |= SKBTX_SHARED_FRAG;
1898 	} else if (msg_control) {
1899 		struct ubuf_info *uarg = msg_control;
1900 		uarg->callback(uarg, false);
1901 	}
1902 
1903 	skb_reset_network_header(skb);
1904 	skb_probe_transport_header(skb, 0);
1905 
1906 	if (skb_xdp) {
1907 		struct bpf_prog *xdp_prog;
1908 		int ret;
1909 
1910 		rcu_read_lock();
1911 		xdp_prog = rcu_dereference(tun->xdp_prog);
1912 		if (xdp_prog) {
1913 			ret = do_xdp_generic(xdp_prog, skb);
1914 			if (ret != XDP_PASS) {
1915 				rcu_read_unlock();
1916 				return total_len;
1917 			}
1918 		}
1919 		rcu_read_unlock();
1920 	}
1921 
1922 	rcu_read_lock();
1923 	if (!rcu_dereference(tun->steering_prog))
1924 		rxhash = __skb_get_hash_symmetric(skb);
1925 	rcu_read_unlock();
1926 
1927 	if (frags) {
1928 		/* Exercise flow dissector code path. */
1929 		u32 headlen = eth_get_headlen(skb->data, skb_headlen(skb));
1930 
1931 		if (unlikely(headlen > skb_headlen(skb))) {
1932 			this_cpu_inc(tun->pcpu_stats->rx_dropped);
1933 			napi_free_frags(&tfile->napi);
1934 			mutex_unlock(&tfile->napi_mutex);
1935 			WARN_ON(1);
1936 			return -ENOMEM;
1937 		}
1938 
1939 		local_bh_disable();
1940 		napi_gro_frags(&tfile->napi);
1941 		local_bh_enable();
1942 		mutex_unlock(&tfile->napi_mutex);
1943 	} else if (tfile->napi_enabled) {
1944 		struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
1945 		int queue_len;
1946 
1947 		spin_lock_bh(&queue->lock);
1948 		__skb_queue_tail(queue, skb);
1949 		queue_len = skb_queue_len(queue);
1950 		spin_unlock(&queue->lock);
1951 
1952 		if (!more || queue_len > NAPI_POLL_WEIGHT)
1953 			napi_schedule(&tfile->napi);
1954 
1955 		local_bh_enable();
1956 	} else if (!IS_ENABLED(CONFIG_4KSTACKS)) {
1957 		tun_rx_batched(tun, tfile, skb, more);
1958 	} else {
1959 		netif_rx_ni(skb);
1960 	}
1961 
1962 	stats = get_cpu_ptr(tun->pcpu_stats);
1963 	u64_stats_update_begin(&stats->syncp);
1964 	stats->rx_packets++;
1965 	stats->rx_bytes += len;
1966 	u64_stats_update_end(&stats->syncp);
1967 	put_cpu_ptr(stats);
1968 
1969 	if (rxhash)
1970 		tun_flow_update(tun, rxhash, tfile);
1971 
1972 	return total_len;
1973 }
1974 
1975 static ssize_t tun_chr_write_iter(struct kiocb *iocb, struct iov_iter *from)
1976 {
1977 	struct file *file = iocb->ki_filp;
1978 	struct tun_file *tfile = file->private_data;
1979 	struct tun_struct *tun = tun_get(tfile);
1980 	ssize_t result;
1981 
1982 	if (!tun)
1983 		return -EBADFD;
1984 
1985 	result = tun_get_user(tun, tfile, NULL, from,
1986 			      file->f_flags & O_NONBLOCK, false);
1987 
1988 	if (tfile->xdp_pending_pkts) {
1989 		tfile->xdp_pending_pkts = 0;
1990 		xdp_do_flush_map();
1991 	}
1992 
1993 	tun_put(tun);
1994 	return result;
1995 }
1996 
1997 static ssize_t tun_put_user_xdp(struct tun_struct *tun,
1998 				struct tun_file *tfile,
1999 				struct xdp_buff *xdp,
2000 				struct iov_iter *iter)
2001 {
2002 	int vnet_hdr_sz = 0;
2003 	size_t size = xdp->data_end - xdp->data;
2004 	struct tun_pcpu_stats *stats;
2005 	size_t ret;
2006 
2007 	if (tun->flags & IFF_VNET_HDR) {
2008 		struct virtio_net_hdr gso = { 0 };
2009 
2010 		vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
2011 		if (unlikely(iov_iter_count(iter) < vnet_hdr_sz))
2012 			return -EINVAL;
2013 		if (unlikely(copy_to_iter(&gso, sizeof(gso), iter) !=
2014 			     sizeof(gso)))
2015 			return -EFAULT;
2016 		iov_iter_advance(iter, vnet_hdr_sz - sizeof(gso));
2017 	}
2018 
2019 	ret = copy_to_iter(xdp->data, size, iter) + vnet_hdr_sz;
2020 
2021 	stats = get_cpu_ptr(tun->pcpu_stats);
2022 	u64_stats_update_begin(&stats->syncp);
2023 	stats->tx_packets++;
2024 	stats->tx_bytes += ret;
2025 	u64_stats_update_end(&stats->syncp);
2026 	put_cpu_ptr(tun->pcpu_stats);
2027 
2028 	return ret;
2029 }
2030 
2031 /* Put packet to the user space buffer */
2032 static ssize_t tun_put_user(struct tun_struct *tun,
2033 			    struct tun_file *tfile,
2034 			    struct sk_buff *skb,
2035 			    struct iov_iter *iter)
2036 {
2037 	struct tun_pi pi = { 0, skb->protocol };
2038 	struct tun_pcpu_stats *stats;
2039 	ssize_t total;
2040 	int vlan_offset = 0;
2041 	int vlan_hlen = 0;
2042 	int vnet_hdr_sz = 0;
2043 
2044 	if (skb_vlan_tag_present(skb))
2045 		vlan_hlen = VLAN_HLEN;
2046 
2047 	if (tun->flags & IFF_VNET_HDR)
2048 		vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
2049 
2050 	total = skb->len + vlan_hlen + vnet_hdr_sz;
2051 
2052 	if (!(tun->flags & IFF_NO_PI)) {
2053 		if (iov_iter_count(iter) < sizeof(pi))
2054 			return -EINVAL;
2055 
2056 		total += sizeof(pi);
2057 		if (iov_iter_count(iter) < total) {
2058 			/* Packet will be striped */
2059 			pi.flags |= TUN_PKT_STRIP;
2060 		}
2061 
2062 		if (copy_to_iter(&pi, sizeof(pi), iter) != sizeof(pi))
2063 			return -EFAULT;
2064 	}
2065 
2066 	if (vnet_hdr_sz) {
2067 		struct virtio_net_hdr gso;
2068 
2069 		if (iov_iter_count(iter) < vnet_hdr_sz)
2070 			return -EINVAL;
2071 
2072 		if (virtio_net_hdr_from_skb(skb, &gso,
2073 					    tun_is_little_endian(tun), true)) {
2074 			struct skb_shared_info *sinfo = skb_shinfo(skb);
2075 			pr_err("unexpected GSO type: "
2076 			       "0x%x, gso_size %d, hdr_len %d\n",
2077 			       sinfo->gso_type, tun16_to_cpu(tun, gso.gso_size),
2078 			       tun16_to_cpu(tun, gso.hdr_len));
2079 			print_hex_dump(KERN_ERR, "tun: ",
2080 				       DUMP_PREFIX_NONE,
2081 				       16, 1, skb->head,
2082 				       min((int)tun16_to_cpu(tun, gso.hdr_len), 64), true);
2083 			WARN_ON_ONCE(1);
2084 			return -EINVAL;
2085 		}
2086 
2087 		if (copy_to_iter(&gso, sizeof(gso), iter) != sizeof(gso))
2088 			return -EFAULT;
2089 
2090 		iov_iter_advance(iter, vnet_hdr_sz - sizeof(gso));
2091 	}
2092 
2093 	if (vlan_hlen) {
2094 		int ret;
2095 		struct veth veth;
2096 
2097 		veth.h_vlan_proto = skb->vlan_proto;
2098 		veth.h_vlan_TCI = htons(skb_vlan_tag_get(skb));
2099 
2100 		vlan_offset = offsetof(struct vlan_ethhdr, h_vlan_proto);
2101 
2102 		ret = skb_copy_datagram_iter(skb, 0, iter, vlan_offset);
2103 		if (ret || !iov_iter_count(iter))
2104 			goto done;
2105 
2106 		ret = copy_to_iter(&veth, sizeof(veth), iter);
2107 		if (ret != sizeof(veth) || !iov_iter_count(iter))
2108 			goto done;
2109 	}
2110 
2111 	skb_copy_datagram_iter(skb, vlan_offset, iter, skb->len - vlan_offset);
2112 
2113 done:
2114 	/* caller is in process context, */
2115 	stats = get_cpu_ptr(tun->pcpu_stats);
2116 	u64_stats_update_begin(&stats->syncp);
2117 	stats->tx_packets++;
2118 	stats->tx_bytes += skb->len + vlan_hlen;
2119 	u64_stats_update_end(&stats->syncp);
2120 	put_cpu_ptr(tun->pcpu_stats);
2121 
2122 	return total;
2123 }
2124 
2125 static void *tun_ring_recv(struct tun_file *tfile, int noblock, int *err)
2126 {
2127 	DECLARE_WAITQUEUE(wait, current);
2128 	void *ptr = NULL;
2129 	int error = 0;
2130 
2131 	ptr = ptr_ring_consume(&tfile->tx_ring);
2132 	if (ptr)
2133 		goto out;
2134 	if (noblock) {
2135 		error = -EAGAIN;
2136 		goto out;
2137 	}
2138 
2139 	add_wait_queue(&tfile->wq.wait, &wait);
2140 	current->state = TASK_INTERRUPTIBLE;
2141 
2142 	while (1) {
2143 		ptr = ptr_ring_consume(&tfile->tx_ring);
2144 		if (ptr)
2145 			break;
2146 		if (signal_pending(current)) {
2147 			error = -ERESTARTSYS;
2148 			break;
2149 		}
2150 		if (tfile->socket.sk->sk_shutdown & RCV_SHUTDOWN) {
2151 			error = -EFAULT;
2152 			break;
2153 		}
2154 
2155 		schedule();
2156 	}
2157 
2158 	current->state = TASK_RUNNING;
2159 	remove_wait_queue(&tfile->wq.wait, &wait);
2160 
2161 out:
2162 	*err = error;
2163 	return ptr;
2164 }
2165 
2166 static ssize_t tun_do_read(struct tun_struct *tun, struct tun_file *tfile,
2167 			   struct iov_iter *to,
2168 			   int noblock, void *ptr)
2169 {
2170 	ssize_t ret;
2171 	int err;
2172 
2173 	tun_debug(KERN_INFO, tun, "tun_do_read\n");
2174 
2175 	if (!iov_iter_count(to)) {
2176 		tun_ptr_free(ptr);
2177 		return 0;
2178 	}
2179 
2180 	if (!ptr) {
2181 		/* Read frames from ring */
2182 		ptr = tun_ring_recv(tfile, noblock, &err);
2183 		if (!ptr)
2184 			return err;
2185 	}
2186 
2187 	if (tun_is_xdp_buff(ptr)) {
2188 		struct xdp_buff *xdp = tun_ptr_to_xdp(ptr);
2189 
2190 		ret = tun_put_user_xdp(tun, tfile, xdp, to);
2191 		put_page(virt_to_head_page(xdp->data));
2192 	} else {
2193 		struct sk_buff *skb = ptr;
2194 
2195 		ret = tun_put_user(tun, tfile, skb, to);
2196 		if (unlikely(ret < 0))
2197 			kfree_skb(skb);
2198 		else
2199 			consume_skb(skb);
2200 	}
2201 
2202 	return ret;
2203 }
2204 
2205 static ssize_t tun_chr_read_iter(struct kiocb *iocb, struct iov_iter *to)
2206 {
2207 	struct file *file = iocb->ki_filp;
2208 	struct tun_file *tfile = file->private_data;
2209 	struct tun_struct *tun = tun_get(tfile);
2210 	ssize_t len = iov_iter_count(to), ret;
2211 
2212 	if (!tun)
2213 		return -EBADFD;
2214 	ret = tun_do_read(tun, tfile, to, file->f_flags & O_NONBLOCK, NULL);
2215 	ret = min_t(ssize_t, ret, len);
2216 	if (ret > 0)
2217 		iocb->ki_pos = ret;
2218 	tun_put(tun);
2219 	return ret;
2220 }
2221 
2222 static void tun_prog_free(struct rcu_head *rcu)
2223 {
2224 	struct tun_prog *prog = container_of(rcu, struct tun_prog, rcu);
2225 
2226 	bpf_prog_destroy(prog->prog);
2227 	kfree(prog);
2228 }
2229 
2230 static int __tun_set_ebpf(struct tun_struct *tun,
2231 			  struct tun_prog __rcu **prog_p,
2232 			  struct bpf_prog *prog)
2233 {
2234 	struct tun_prog *old, *new = NULL;
2235 
2236 	if (prog) {
2237 		new = kmalloc(sizeof(*new), GFP_KERNEL);
2238 		if (!new)
2239 			return -ENOMEM;
2240 		new->prog = prog;
2241 	}
2242 
2243 	spin_lock_bh(&tun->lock);
2244 	old = rcu_dereference_protected(*prog_p,
2245 					lockdep_is_held(&tun->lock));
2246 	rcu_assign_pointer(*prog_p, new);
2247 	spin_unlock_bh(&tun->lock);
2248 
2249 	if (old)
2250 		call_rcu(&old->rcu, tun_prog_free);
2251 
2252 	return 0;
2253 }
2254 
2255 static void tun_free_netdev(struct net_device *dev)
2256 {
2257 	struct tun_struct *tun = netdev_priv(dev);
2258 
2259 	BUG_ON(!(list_empty(&tun->disabled)));
2260 	free_percpu(tun->pcpu_stats);
2261 	tun_flow_uninit(tun);
2262 	security_tun_dev_free_security(tun->security);
2263 	__tun_set_ebpf(tun, &tun->steering_prog, NULL);
2264 	__tun_set_ebpf(tun, &tun->filter_prog, NULL);
2265 }
2266 
2267 static void tun_setup(struct net_device *dev)
2268 {
2269 	struct tun_struct *tun = netdev_priv(dev);
2270 
2271 	tun->owner = INVALID_UID;
2272 	tun->group = INVALID_GID;
2273 
2274 	dev->ethtool_ops = &tun_ethtool_ops;
2275 	dev->needs_free_netdev = true;
2276 	dev->priv_destructor = tun_free_netdev;
2277 	/* We prefer our own queue length */
2278 	dev->tx_queue_len = TUN_READQ_SIZE;
2279 }
2280 
2281 /* Trivial set of netlink ops to allow deleting tun or tap
2282  * device with netlink.
2283  */
2284 static int tun_validate(struct nlattr *tb[], struct nlattr *data[],
2285 			struct netlink_ext_ack *extack)
2286 {
2287 	return -EINVAL;
2288 }
2289 
2290 static size_t tun_get_size(const struct net_device *dev)
2291 {
2292 	BUILD_BUG_ON(sizeof(u32) != sizeof(uid_t));
2293 	BUILD_BUG_ON(sizeof(u32) != sizeof(gid_t));
2294 
2295 	return nla_total_size(sizeof(uid_t)) + /* OWNER */
2296 	       nla_total_size(sizeof(gid_t)) + /* GROUP */
2297 	       nla_total_size(sizeof(u8)) + /* TYPE */
2298 	       nla_total_size(sizeof(u8)) + /* PI */
2299 	       nla_total_size(sizeof(u8)) + /* VNET_HDR */
2300 	       nla_total_size(sizeof(u8)) + /* PERSIST */
2301 	       nla_total_size(sizeof(u8)) + /* MULTI_QUEUE */
2302 	       nla_total_size(sizeof(u32)) + /* NUM_QUEUES */
2303 	       nla_total_size(sizeof(u32)) + /* NUM_DISABLED_QUEUES */
2304 	       0;
2305 }
2306 
2307 static int tun_fill_info(struct sk_buff *skb, const struct net_device *dev)
2308 {
2309 	struct tun_struct *tun = netdev_priv(dev);
2310 
2311 	if (nla_put_u8(skb, IFLA_TUN_TYPE, tun->flags & TUN_TYPE_MASK))
2312 		goto nla_put_failure;
2313 	if (uid_valid(tun->owner) &&
2314 	    nla_put_u32(skb, IFLA_TUN_OWNER,
2315 			from_kuid_munged(current_user_ns(), tun->owner)))
2316 		goto nla_put_failure;
2317 	if (gid_valid(tun->group) &&
2318 	    nla_put_u32(skb, IFLA_TUN_GROUP,
2319 			from_kgid_munged(current_user_ns(), tun->group)))
2320 		goto nla_put_failure;
2321 	if (nla_put_u8(skb, IFLA_TUN_PI, !(tun->flags & IFF_NO_PI)))
2322 		goto nla_put_failure;
2323 	if (nla_put_u8(skb, IFLA_TUN_VNET_HDR, !!(tun->flags & IFF_VNET_HDR)))
2324 		goto nla_put_failure;
2325 	if (nla_put_u8(skb, IFLA_TUN_PERSIST, !!(tun->flags & IFF_PERSIST)))
2326 		goto nla_put_failure;
2327 	if (nla_put_u8(skb, IFLA_TUN_MULTI_QUEUE,
2328 		       !!(tun->flags & IFF_MULTI_QUEUE)))
2329 		goto nla_put_failure;
2330 	if (tun->flags & IFF_MULTI_QUEUE) {
2331 		if (nla_put_u32(skb, IFLA_TUN_NUM_QUEUES, tun->numqueues))
2332 			goto nla_put_failure;
2333 		if (nla_put_u32(skb, IFLA_TUN_NUM_DISABLED_QUEUES,
2334 				tun->numdisabled))
2335 			goto nla_put_failure;
2336 	}
2337 
2338 	return 0;
2339 
2340 nla_put_failure:
2341 	return -EMSGSIZE;
2342 }
2343 
2344 static struct rtnl_link_ops tun_link_ops __read_mostly = {
2345 	.kind		= DRV_NAME,
2346 	.priv_size	= sizeof(struct tun_struct),
2347 	.setup		= tun_setup,
2348 	.validate	= tun_validate,
2349 	.get_size       = tun_get_size,
2350 	.fill_info      = tun_fill_info,
2351 };
2352 
2353 static void tun_sock_write_space(struct sock *sk)
2354 {
2355 	struct tun_file *tfile;
2356 	wait_queue_head_t *wqueue;
2357 
2358 	if (!sock_writeable(sk))
2359 		return;
2360 
2361 	if (!test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags))
2362 		return;
2363 
2364 	wqueue = sk_sleep(sk);
2365 	if (wqueue && waitqueue_active(wqueue))
2366 		wake_up_interruptible_sync_poll(wqueue, EPOLLOUT |
2367 						EPOLLWRNORM | EPOLLWRBAND);
2368 
2369 	tfile = container_of(sk, struct tun_file, sk);
2370 	kill_fasync(&tfile->fasync, SIGIO, POLL_OUT);
2371 }
2372 
2373 static int tun_sendmsg(struct socket *sock, struct msghdr *m, size_t total_len)
2374 {
2375 	int ret;
2376 	struct tun_file *tfile = container_of(sock, struct tun_file, socket);
2377 	struct tun_struct *tun = tun_get(tfile);
2378 
2379 	if (!tun)
2380 		return -EBADFD;
2381 
2382 	ret = tun_get_user(tun, tfile, m->msg_control, &m->msg_iter,
2383 			   m->msg_flags & MSG_DONTWAIT,
2384 			   m->msg_flags & MSG_MORE);
2385 
2386 	if (tfile->xdp_pending_pkts >= NAPI_POLL_WEIGHT ||
2387 	    !(m->msg_flags & MSG_MORE)) {
2388 		tfile->xdp_pending_pkts = 0;
2389 		xdp_do_flush_map();
2390 	}
2391 
2392 	tun_put(tun);
2393 	return ret;
2394 }
2395 
2396 static int tun_recvmsg(struct socket *sock, struct msghdr *m, size_t total_len,
2397 		       int flags)
2398 {
2399 	struct tun_file *tfile = container_of(sock, struct tun_file, socket);
2400 	struct tun_struct *tun = tun_get(tfile);
2401 	void *ptr = m->msg_control;
2402 	int ret;
2403 
2404 	if (!tun) {
2405 		ret = -EBADFD;
2406 		goto out_free;
2407 	}
2408 
2409 	if (flags & ~(MSG_DONTWAIT|MSG_TRUNC|MSG_ERRQUEUE)) {
2410 		ret = -EINVAL;
2411 		goto out_put_tun;
2412 	}
2413 	if (flags & MSG_ERRQUEUE) {
2414 		ret = sock_recv_errqueue(sock->sk, m, total_len,
2415 					 SOL_PACKET, TUN_TX_TIMESTAMP);
2416 		goto out;
2417 	}
2418 	ret = tun_do_read(tun, tfile, &m->msg_iter, flags & MSG_DONTWAIT, ptr);
2419 	if (ret > (ssize_t)total_len) {
2420 		m->msg_flags |= MSG_TRUNC;
2421 		ret = flags & MSG_TRUNC ? ret : total_len;
2422 	}
2423 out:
2424 	tun_put(tun);
2425 	return ret;
2426 
2427 out_put_tun:
2428 	tun_put(tun);
2429 out_free:
2430 	tun_ptr_free(ptr);
2431 	return ret;
2432 }
2433 
2434 static int tun_ptr_peek_len(void *ptr)
2435 {
2436 	if (likely(ptr)) {
2437 		if (tun_is_xdp_buff(ptr)) {
2438 			struct xdp_buff *xdp = tun_ptr_to_xdp(ptr);
2439 
2440 			return xdp->data_end - xdp->data;
2441 		}
2442 		return __skb_array_len_with_tag(ptr);
2443 	} else {
2444 		return 0;
2445 	}
2446 }
2447 
2448 static int tun_peek_len(struct socket *sock)
2449 {
2450 	struct tun_file *tfile = container_of(sock, struct tun_file, socket);
2451 	struct tun_struct *tun;
2452 	int ret = 0;
2453 
2454 	tun = tun_get(tfile);
2455 	if (!tun)
2456 		return 0;
2457 
2458 	ret = PTR_RING_PEEK_CALL(&tfile->tx_ring, tun_ptr_peek_len);
2459 	tun_put(tun);
2460 
2461 	return ret;
2462 }
2463 
2464 /* Ops structure to mimic raw sockets with tun */
2465 static const struct proto_ops tun_socket_ops = {
2466 	.peek_len = tun_peek_len,
2467 	.sendmsg = tun_sendmsg,
2468 	.recvmsg = tun_recvmsg,
2469 };
2470 
2471 static struct proto tun_proto = {
2472 	.name		= "tun",
2473 	.owner		= THIS_MODULE,
2474 	.obj_size	= sizeof(struct tun_file),
2475 };
2476 
2477 static int tun_flags(struct tun_struct *tun)
2478 {
2479 	return tun->flags & (TUN_FEATURES | IFF_PERSIST | IFF_TUN | IFF_TAP);
2480 }
2481 
2482 static ssize_t tun_show_flags(struct device *dev, struct device_attribute *attr,
2483 			      char *buf)
2484 {
2485 	struct tun_struct *tun = netdev_priv(to_net_dev(dev));
2486 	return sprintf(buf, "0x%x\n", tun_flags(tun));
2487 }
2488 
2489 static ssize_t tun_show_owner(struct device *dev, struct device_attribute *attr,
2490 			      char *buf)
2491 {
2492 	struct tun_struct *tun = netdev_priv(to_net_dev(dev));
2493 	return uid_valid(tun->owner)?
2494 		sprintf(buf, "%u\n",
2495 			from_kuid_munged(current_user_ns(), tun->owner)):
2496 		sprintf(buf, "-1\n");
2497 }
2498 
2499 static ssize_t tun_show_group(struct device *dev, struct device_attribute *attr,
2500 			      char *buf)
2501 {
2502 	struct tun_struct *tun = netdev_priv(to_net_dev(dev));
2503 	return gid_valid(tun->group) ?
2504 		sprintf(buf, "%u\n",
2505 			from_kgid_munged(current_user_ns(), tun->group)):
2506 		sprintf(buf, "-1\n");
2507 }
2508 
2509 static DEVICE_ATTR(tun_flags, 0444, tun_show_flags, NULL);
2510 static DEVICE_ATTR(owner, 0444, tun_show_owner, NULL);
2511 static DEVICE_ATTR(group, 0444, tun_show_group, NULL);
2512 
2513 static struct attribute *tun_dev_attrs[] = {
2514 	&dev_attr_tun_flags.attr,
2515 	&dev_attr_owner.attr,
2516 	&dev_attr_group.attr,
2517 	NULL
2518 };
2519 
2520 static const struct attribute_group tun_attr_group = {
2521 	.attrs = tun_dev_attrs
2522 };
2523 
2524 static int tun_set_iff(struct net *net, struct file *file, struct ifreq *ifr)
2525 {
2526 	struct tun_struct *tun;
2527 	struct tun_file *tfile = file->private_data;
2528 	struct net_device *dev;
2529 	int err;
2530 
2531 	if (tfile->detached)
2532 		return -EINVAL;
2533 
2534 	if ((ifr->ifr_flags & IFF_NAPI_FRAGS)) {
2535 		if (!capable(CAP_NET_ADMIN))
2536 			return -EPERM;
2537 
2538 		if (!(ifr->ifr_flags & IFF_NAPI) ||
2539 		    (ifr->ifr_flags & TUN_TYPE_MASK) != IFF_TAP)
2540 			return -EINVAL;
2541 	}
2542 
2543 	dev = __dev_get_by_name(net, ifr->ifr_name);
2544 	if (dev) {
2545 		if (ifr->ifr_flags & IFF_TUN_EXCL)
2546 			return -EBUSY;
2547 		if ((ifr->ifr_flags & IFF_TUN) && dev->netdev_ops == &tun_netdev_ops)
2548 			tun = netdev_priv(dev);
2549 		else if ((ifr->ifr_flags & IFF_TAP) && dev->netdev_ops == &tap_netdev_ops)
2550 			tun = netdev_priv(dev);
2551 		else
2552 			return -EINVAL;
2553 
2554 		if (!!(ifr->ifr_flags & IFF_MULTI_QUEUE) !=
2555 		    !!(tun->flags & IFF_MULTI_QUEUE))
2556 			return -EINVAL;
2557 
2558 		if (tun_not_capable(tun))
2559 			return -EPERM;
2560 		err = security_tun_dev_open(tun->security);
2561 		if (err < 0)
2562 			return err;
2563 
2564 		err = tun_attach(tun, file, ifr->ifr_flags & IFF_NOFILTER,
2565 				 ifr->ifr_flags & IFF_NAPI);
2566 		if (err < 0)
2567 			return err;
2568 
2569 		if (tun->flags & IFF_MULTI_QUEUE &&
2570 		    (tun->numqueues + tun->numdisabled > 1)) {
2571 			/* One or more queue has already been attached, no need
2572 			 * to initialize the device again.
2573 			 */
2574 			return 0;
2575 		}
2576 	}
2577 	else {
2578 		char *name;
2579 		unsigned long flags = 0;
2580 		int queues = ifr->ifr_flags & IFF_MULTI_QUEUE ?
2581 			     MAX_TAP_QUEUES : 1;
2582 
2583 		if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
2584 			return -EPERM;
2585 		err = security_tun_dev_create();
2586 		if (err < 0)
2587 			return err;
2588 
2589 		/* Set dev type */
2590 		if (ifr->ifr_flags & IFF_TUN) {
2591 			/* TUN device */
2592 			flags |= IFF_TUN;
2593 			name = "tun%d";
2594 		} else if (ifr->ifr_flags & IFF_TAP) {
2595 			/* TAP device */
2596 			flags |= IFF_TAP;
2597 			name = "tap%d";
2598 		} else
2599 			return -EINVAL;
2600 
2601 		if (*ifr->ifr_name)
2602 			name = ifr->ifr_name;
2603 
2604 		dev = alloc_netdev_mqs(sizeof(struct tun_struct), name,
2605 				       NET_NAME_UNKNOWN, tun_setup, queues,
2606 				       queues);
2607 
2608 		if (!dev)
2609 			return -ENOMEM;
2610 		err = dev_get_valid_name(net, dev, name);
2611 		if (err < 0)
2612 			goto err_free_dev;
2613 
2614 		dev_net_set(dev, net);
2615 		dev->rtnl_link_ops = &tun_link_ops;
2616 		dev->ifindex = tfile->ifindex;
2617 		dev->sysfs_groups[0] = &tun_attr_group;
2618 
2619 		tun = netdev_priv(dev);
2620 		tun->dev = dev;
2621 		tun->flags = flags;
2622 		tun->txflt.count = 0;
2623 		tun->vnet_hdr_sz = sizeof(struct virtio_net_hdr);
2624 
2625 		tun->align = NET_SKB_PAD;
2626 		tun->filter_attached = false;
2627 		tun->sndbuf = tfile->socket.sk->sk_sndbuf;
2628 		tun->rx_batched = 0;
2629 		RCU_INIT_POINTER(tun->steering_prog, NULL);
2630 
2631 		tun->pcpu_stats = netdev_alloc_pcpu_stats(struct tun_pcpu_stats);
2632 		if (!tun->pcpu_stats) {
2633 			err = -ENOMEM;
2634 			goto err_free_dev;
2635 		}
2636 
2637 		spin_lock_init(&tun->lock);
2638 
2639 		err = security_tun_dev_alloc_security(&tun->security);
2640 		if (err < 0)
2641 			goto err_free_stat;
2642 
2643 		tun_net_init(dev);
2644 		tun_flow_init(tun);
2645 
2646 		dev->hw_features = NETIF_F_SG | NETIF_F_FRAGLIST |
2647 				   TUN_USER_FEATURES | NETIF_F_HW_VLAN_CTAG_TX |
2648 				   NETIF_F_HW_VLAN_STAG_TX;
2649 		dev->features = dev->hw_features | NETIF_F_LLTX;
2650 		dev->vlan_features = dev->features &
2651 				     ~(NETIF_F_HW_VLAN_CTAG_TX |
2652 				       NETIF_F_HW_VLAN_STAG_TX);
2653 
2654 		INIT_LIST_HEAD(&tun->disabled);
2655 		err = tun_attach(tun, file, false, ifr->ifr_flags & IFF_NAPI);
2656 		if (err < 0)
2657 			goto err_free_flow;
2658 
2659 		err = register_netdevice(tun->dev);
2660 		if (err < 0)
2661 			goto err_detach;
2662 	}
2663 
2664 	netif_carrier_on(tun->dev);
2665 
2666 	tun_debug(KERN_INFO, tun, "tun_set_iff\n");
2667 
2668 	tun->flags = (tun->flags & ~TUN_FEATURES) |
2669 		(ifr->ifr_flags & TUN_FEATURES);
2670 
2671 	/* Make sure persistent devices do not get stuck in
2672 	 * xoff state.
2673 	 */
2674 	if (netif_running(tun->dev))
2675 		netif_tx_wake_all_queues(tun->dev);
2676 
2677 	strcpy(ifr->ifr_name, tun->dev->name);
2678 	return 0;
2679 
2680 err_detach:
2681 	tun_detach_all(dev);
2682 	/* register_netdevice() already called tun_free_netdev() */
2683 	goto err_free_dev;
2684 
2685 err_free_flow:
2686 	tun_flow_uninit(tun);
2687 	security_tun_dev_free_security(tun->security);
2688 err_free_stat:
2689 	free_percpu(tun->pcpu_stats);
2690 err_free_dev:
2691 	free_netdev(dev);
2692 	return err;
2693 }
2694 
2695 static void tun_get_iff(struct net *net, struct tun_struct *tun,
2696 		       struct ifreq *ifr)
2697 {
2698 	tun_debug(KERN_INFO, tun, "tun_get_iff\n");
2699 
2700 	strcpy(ifr->ifr_name, tun->dev->name);
2701 
2702 	ifr->ifr_flags = tun_flags(tun);
2703 
2704 }
2705 
2706 /* This is like a cut-down ethtool ops, except done via tun fd so no
2707  * privs required. */
2708 static int set_offload(struct tun_struct *tun, unsigned long arg)
2709 {
2710 	netdev_features_t features = 0;
2711 
2712 	if (arg & TUN_F_CSUM) {
2713 		features |= NETIF_F_HW_CSUM;
2714 		arg &= ~TUN_F_CSUM;
2715 
2716 		if (arg & (TUN_F_TSO4|TUN_F_TSO6)) {
2717 			if (arg & TUN_F_TSO_ECN) {
2718 				features |= NETIF_F_TSO_ECN;
2719 				arg &= ~TUN_F_TSO_ECN;
2720 			}
2721 			if (arg & TUN_F_TSO4)
2722 				features |= NETIF_F_TSO;
2723 			if (arg & TUN_F_TSO6)
2724 				features |= NETIF_F_TSO6;
2725 			arg &= ~(TUN_F_TSO4|TUN_F_TSO6);
2726 		}
2727 
2728 		arg &= ~TUN_F_UFO;
2729 	}
2730 
2731 	/* This gives the user a way to test for new features in future by
2732 	 * trying to set them. */
2733 	if (arg)
2734 		return -EINVAL;
2735 
2736 	tun->set_features = features;
2737 	tun->dev->wanted_features &= ~TUN_USER_FEATURES;
2738 	tun->dev->wanted_features |= features;
2739 	netdev_update_features(tun->dev);
2740 
2741 	return 0;
2742 }
2743 
2744 static void tun_detach_filter(struct tun_struct *tun, int n)
2745 {
2746 	int i;
2747 	struct tun_file *tfile;
2748 
2749 	for (i = 0; i < n; i++) {
2750 		tfile = rtnl_dereference(tun->tfiles[i]);
2751 		lock_sock(tfile->socket.sk);
2752 		sk_detach_filter(tfile->socket.sk);
2753 		release_sock(tfile->socket.sk);
2754 	}
2755 
2756 	tun->filter_attached = false;
2757 }
2758 
2759 static int tun_attach_filter(struct tun_struct *tun)
2760 {
2761 	int i, ret = 0;
2762 	struct tun_file *tfile;
2763 
2764 	for (i = 0; i < tun->numqueues; i++) {
2765 		tfile = rtnl_dereference(tun->tfiles[i]);
2766 		lock_sock(tfile->socket.sk);
2767 		ret = sk_attach_filter(&tun->fprog, tfile->socket.sk);
2768 		release_sock(tfile->socket.sk);
2769 		if (ret) {
2770 			tun_detach_filter(tun, i);
2771 			return ret;
2772 		}
2773 	}
2774 
2775 	tun->filter_attached = true;
2776 	return ret;
2777 }
2778 
2779 static void tun_set_sndbuf(struct tun_struct *tun)
2780 {
2781 	struct tun_file *tfile;
2782 	int i;
2783 
2784 	for (i = 0; i < tun->numqueues; i++) {
2785 		tfile = rtnl_dereference(tun->tfiles[i]);
2786 		tfile->socket.sk->sk_sndbuf = tun->sndbuf;
2787 	}
2788 }
2789 
2790 static int tun_set_queue(struct file *file, struct ifreq *ifr)
2791 {
2792 	struct tun_file *tfile = file->private_data;
2793 	struct tun_struct *tun;
2794 	int ret = 0;
2795 
2796 	rtnl_lock();
2797 
2798 	if (ifr->ifr_flags & IFF_ATTACH_QUEUE) {
2799 		tun = tfile->detached;
2800 		if (!tun) {
2801 			ret = -EINVAL;
2802 			goto unlock;
2803 		}
2804 		ret = security_tun_dev_attach_queue(tun->security);
2805 		if (ret < 0)
2806 			goto unlock;
2807 		ret = tun_attach(tun, file, false, tun->flags & IFF_NAPI);
2808 	} else if (ifr->ifr_flags & IFF_DETACH_QUEUE) {
2809 		tun = rtnl_dereference(tfile->tun);
2810 		if (!tun || !(tun->flags & IFF_MULTI_QUEUE) || tfile->detached)
2811 			ret = -EINVAL;
2812 		else
2813 			__tun_detach(tfile, false);
2814 	} else
2815 		ret = -EINVAL;
2816 
2817 unlock:
2818 	rtnl_unlock();
2819 	return ret;
2820 }
2821 
2822 static int tun_set_ebpf(struct tun_struct *tun, struct tun_prog **prog_p,
2823 			void __user *data)
2824 {
2825 	struct bpf_prog *prog;
2826 	int fd;
2827 
2828 	if (copy_from_user(&fd, data, sizeof(fd)))
2829 		return -EFAULT;
2830 
2831 	if (fd == -1) {
2832 		prog = NULL;
2833 	} else {
2834 		prog = bpf_prog_get_type(fd, BPF_PROG_TYPE_SOCKET_FILTER);
2835 		if (IS_ERR(prog))
2836 			return PTR_ERR(prog);
2837 	}
2838 
2839 	return __tun_set_ebpf(tun, prog_p, prog);
2840 }
2841 
2842 static long __tun_chr_ioctl(struct file *file, unsigned int cmd,
2843 			    unsigned long arg, int ifreq_len)
2844 {
2845 	struct tun_file *tfile = file->private_data;
2846 	struct tun_struct *tun;
2847 	void __user* argp = (void __user*)arg;
2848 	struct ifreq ifr;
2849 	struct net *net;
2850 	kuid_t owner;
2851 	kgid_t group;
2852 	int sndbuf;
2853 	int vnet_hdr_sz;
2854 	unsigned int ifindex;
2855 	int le;
2856 	int ret;
2857 
2858 	if (cmd == TUNSETIFF || cmd == TUNSETQUEUE ||
2859 	    (_IOC_TYPE(cmd) == SOCK_IOC_TYPE && cmd != SIOCGSKNS)) {
2860 		if (copy_from_user(&ifr, argp, ifreq_len))
2861 			return -EFAULT;
2862 	} else {
2863 		memset(&ifr, 0, sizeof(ifr));
2864 	}
2865 	if (cmd == TUNGETFEATURES) {
2866 		/* Currently this just means: "what IFF flags are valid?".
2867 		 * This is needed because we never checked for invalid flags on
2868 		 * TUNSETIFF.
2869 		 */
2870 		return put_user(IFF_TUN | IFF_TAP | TUN_FEATURES,
2871 				(unsigned int __user*)argp);
2872 	} else if (cmd == TUNSETQUEUE)
2873 		return tun_set_queue(file, &ifr);
2874 
2875 	ret = 0;
2876 	rtnl_lock();
2877 
2878 	tun = tun_get(tfile);
2879 	net = sock_net(&tfile->sk);
2880 	if (cmd == TUNSETIFF) {
2881 		ret = -EEXIST;
2882 		if (tun)
2883 			goto unlock;
2884 
2885 		ifr.ifr_name[IFNAMSIZ-1] = '\0';
2886 
2887 		ret = tun_set_iff(net, file, &ifr);
2888 
2889 		if (ret)
2890 			goto unlock;
2891 
2892 		if (copy_to_user(argp, &ifr, ifreq_len))
2893 			ret = -EFAULT;
2894 		goto unlock;
2895 	}
2896 	if (cmd == TUNSETIFINDEX) {
2897 		ret = -EPERM;
2898 		if (tun)
2899 			goto unlock;
2900 
2901 		ret = -EFAULT;
2902 		if (copy_from_user(&ifindex, argp, sizeof(ifindex)))
2903 			goto unlock;
2904 
2905 		ret = 0;
2906 		tfile->ifindex = ifindex;
2907 		goto unlock;
2908 	}
2909 	if (cmd == SIOCGSKNS) {
2910 		ret = -EPERM;
2911 		if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
2912 			goto unlock;
2913 
2914 		ret = open_related_ns(&net->ns, get_net_ns);
2915 		goto unlock;
2916 	}
2917 
2918 	ret = -EBADFD;
2919 	if (!tun)
2920 		goto unlock;
2921 
2922 	tun_debug(KERN_INFO, tun, "tun_chr_ioctl cmd %u\n", cmd);
2923 
2924 	ret = 0;
2925 	switch (cmd) {
2926 	case TUNGETIFF:
2927 		tun_get_iff(current->nsproxy->net_ns, tun, &ifr);
2928 
2929 		if (tfile->detached)
2930 			ifr.ifr_flags |= IFF_DETACH_QUEUE;
2931 		if (!tfile->socket.sk->sk_filter)
2932 			ifr.ifr_flags |= IFF_NOFILTER;
2933 
2934 		if (copy_to_user(argp, &ifr, ifreq_len))
2935 			ret = -EFAULT;
2936 		break;
2937 
2938 	case TUNSETNOCSUM:
2939 		/* Disable/Enable checksum */
2940 
2941 		/* [unimplemented] */
2942 		tun_debug(KERN_INFO, tun, "ignored: set checksum %s\n",
2943 			  arg ? "disabled" : "enabled");
2944 		break;
2945 
2946 	case TUNSETPERSIST:
2947 		/* Disable/Enable persist mode. Keep an extra reference to the
2948 		 * module to prevent the module being unprobed.
2949 		 */
2950 		if (arg && !(tun->flags & IFF_PERSIST)) {
2951 			tun->flags |= IFF_PERSIST;
2952 			__module_get(THIS_MODULE);
2953 		}
2954 		if (!arg && (tun->flags & IFF_PERSIST)) {
2955 			tun->flags &= ~IFF_PERSIST;
2956 			module_put(THIS_MODULE);
2957 		}
2958 
2959 		tun_debug(KERN_INFO, tun, "persist %s\n",
2960 			  arg ? "enabled" : "disabled");
2961 		break;
2962 
2963 	case TUNSETOWNER:
2964 		/* Set owner of the device */
2965 		owner = make_kuid(current_user_ns(), arg);
2966 		if (!uid_valid(owner)) {
2967 			ret = -EINVAL;
2968 			break;
2969 		}
2970 		tun->owner = owner;
2971 		tun_debug(KERN_INFO, tun, "owner set to %u\n",
2972 			  from_kuid(&init_user_ns, tun->owner));
2973 		break;
2974 
2975 	case TUNSETGROUP:
2976 		/* Set group of the device */
2977 		group = make_kgid(current_user_ns(), arg);
2978 		if (!gid_valid(group)) {
2979 			ret = -EINVAL;
2980 			break;
2981 		}
2982 		tun->group = group;
2983 		tun_debug(KERN_INFO, tun, "group set to %u\n",
2984 			  from_kgid(&init_user_ns, tun->group));
2985 		break;
2986 
2987 	case TUNSETLINK:
2988 		/* Only allow setting the type when the interface is down */
2989 		if (tun->dev->flags & IFF_UP) {
2990 			tun_debug(KERN_INFO, tun,
2991 				  "Linktype set failed because interface is up\n");
2992 			ret = -EBUSY;
2993 		} else {
2994 			tun->dev->type = (int) arg;
2995 			tun_debug(KERN_INFO, tun, "linktype set to %d\n",
2996 				  tun->dev->type);
2997 			ret = 0;
2998 		}
2999 		break;
3000 
3001 #ifdef TUN_DEBUG
3002 	case TUNSETDEBUG:
3003 		tun->debug = arg;
3004 		break;
3005 #endif
3006 	case TUNSETOFFLOAD:
3007 		ret = set_offload(tun, arg);
3008 		break;
3009 
3010 	case TUNSETTXFILTER:
3011 		/* Can be set only for TAPs */
3012 		ret = -EINVAL;
3013 		if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3014 			break;
3015 		ret = update_filter(&tun->txflt, (void __user *)arg);
3016 		break;
3017 
3018 	case SIOCGIFHWADDR:
3019 		/* Get hw address */
3020 		memcpy(ifr.ifr_hwaddr.sa_data, tun->dev->dev_addr, ETH_ALEN);
3021 		ifr.ifr_hwaddr.sa_family = tun->dev->type;
3022 		if (copy_to_user(argp, &ifr, ifreq_len))
3023 			ret = -EFAULT;
3024 		break;
3025 
3026 	case SIOCSIFHWADDR:
3027 		/* Set hw address */
3028 		tun_debug(KERN_DEBUG, tun, "set hw address: %pM\n",
3029 			  ifr.ifr_hwaddr.sa_data);
3030 
3031 		ret = dev_set_mac_address(tun->dev, &ifr.ifr_hwaddr);
3032 		break;
3033 
3034 	case TUNGETSNDBUF:
3035 		sndbuf = tfile->socket.sk->sk_sndbuf;
3036 		if (copy_to_user(argp, &sndbuf, sizeof(sndbuf)))
3037 			ret = -EFAULT;
3038 		break;
3039 
3040 	case TUNSETSNDBUF:
3041 		if (copy_from_user(&sndbuf, argp, sizeof(sndbuf))) {
3042 			ret = -EFAULT;
3043 			break;
3044 		}
3045 		if (sndbuf <= 0) {
3046 			ret = -EINVAL;
3047 			break;
3048 		}
3049 
3050 		tun->sndbuf = sndbuf;
3051 		tun_set_sndbuf(tun);
3052 		break;
3053 
3054 	case TUNGETVNETHDRSZ:
3055 		vnet_hdr_sz = tun->vnet_hdr_sz;
3056 		if (copy_to_user(argp, &vnet_hdr_sz, sizeof(vnet_hdr_sz)))
3057 			ret = -EFAULT;
3058 		break;
3059 
3060 	case TUNSETVNETHDRSZ:
3061 		if (copy_from_user(&vnet_hdr_sz, argp, sizeof(vnet_hdr_sz))) {
3062 			ret = -EFAULT;
3063 			break;
3064 		}
3065 		if (vnet_hdr_sz < (int)sizeof(struct virtio_net_hdr)) {
3066 			ret = -EINVAL;
3067 			break;
3068 		}
3069 
3070 		tun->vnet_hdr_sz = vnet_hdr_sz;
3071 		break;
3072 
3073 	case TUNGETVNETLE:
3074 		le = !!(tun->flags & TUN_VNET_LE);
3075 		if (put_user(le, (int __user *)argp))
3076 			ret = -EFAULT;
3077 		break;
3078 
3079 	case TUNSETVNETLE:
3080 		if (get_user(le, (int __user *)argp)) {
3081 			ret = -EFAULT;
3082 			break;
3083 		}
3084 		if (le)
3085 			tun->flags |= TUN_VNET_LE;
3086 		else
3087 			tun->flags &= ~TUN_VNET_LE;
3088 		break;
3089 
3090 	case TUNGETVNETBE:
3091 		ret = tun_get_vnet_be(tun, argp);
3092 		break;
3093 
3094 	case TUNSETVNETBE:
3095 		ret = tun_set_vnet_be(tun, argp);
3096 		break;
3097 
3098 	case TUNATTACHFILTER:
3099 		/* Can be set only for TAPs */
3100 		ret = -EINVAL;
3101 		if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3102 			break;
3103 		ret = -EFAULT;
3104 		if (copy_from_user(&tun->fprog, argp, sizeof(tun->fprog)))
3105 			break;
3106 
3107 		ret = tun_attach_filter(tun);
3108 		break;
3109 
3110 	case TUNDETACHFILTER:
3111 		/* Can be set only for TAPs */
3112 		ret = -EINVAL;
3113 		if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3114 			break;
3115 		ret = 0;
3116 		tun_detach_filter(tun, tun->numqueues);
3117 		break;
3118 
3119 	case TUNGETFILTER:
3120 		ret = -EINVAL;
3121 		if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3122 			break;
3123 		ret = -EFAULT;
3124 		if (copy_to_user(argp, &tun->fprog, sizeof(tun->fprog)))
3125 			break;
3126 		ret = 0;
3127 		break;
3128 
3129 	case TUNSETSTEERINGEBPF:
3130 		ret = tun_set_ebpf(tun, &tun->steering_prog, argp);
3131 		break;
3132 
3133 	case TUNSETFILTEREBPF:
3134 		ret = tun_set_ebpf(tun, &tun->filter_prog, argp);
3135 		break;
3136 
3137 	default:
3138 		ret = -EINVAL;
3139 		break;
3140 	}
3141 
3142 unlock:
3143 	rtnl_unlock();
3144 	if (tun)
3145 		tun_put(tun);
3146 	return ret;
3147 }
3148 
3149 static long tun_chr_ioctl(struct file *file,
3150 			  unsigned int cmd, unsigned long arg)
3151 {
3152 	return __tun_chr_ioctl(file, cmd, arg, sizeof (struct ifreq));
3153 }
3154 
3155 #ifdef CONFIG_COMPAT
3156 static long tun_chr_compat_ioctl(struct file *file,
3157 			 unsigned int cmd, unsigned long arg)
3158 {
3159 	switch (cmd) {
3160 	case TUNSETIFF:
3161 	case TUNGETIFF:
3162 	case TUNSETTXFILTER:
3163 	case TUNGETSNDBUF:
3164 	case TUNSETSNDBUF:
3165 	case SIOCGIFHWADDR:
3166 	case SIOCSIFHWADDR:
3167 		arg = (unsigned long)compat_ptr(arg);
3168 		break;
3169 	default:
3170 		arg = (compat_ulong_t)arg;
3171 		break;
3172 	}
3173 
3174 	/*
3175 	 * compat_ifreq is shorter than ifreq, so we must not access beyond
3176 	 * the end of that structure. All fields that are used in this
3177 	 * driver are compatible though, we don't need to convert the
3178 	 * contents.
3179 	 */
3180 	return __tun_chr_ioctl(file, cmd, arg, sizeof(struct compat_ifreq));
3181 }
3182 #endif /* CONFIG_COMPAT */
3183 
3184 static int tun_chr_fasync(int fd, struct file *file, int on)
3185 {
3186 	struct tun_file *tfile = file->private_data;
3187 	int ret;
3188 
3189 	if ((ret = fasync_helper(fd, file, on, &tfile->fasync)) < 0)
3190 		goto out;
3191 
3192 	if (on) {
3193 		__f_setown(file, task_pid(current), PIDTYPE_PID, 0);
3194 		tfile->flags |= TUN_FASYNC;
3195 	} else
3196 		tfile->flags &= ~TUN_FASYNC;
3197 	ret = 0;
3198 out:
3199 	return ret;
3200 }
3201 
3202 static int tun_chr_open(struct inode *inode, struct file * file)
3203 {
3204 	struct net *net = current->nsproxy->net_ns;
3205 	struct tun_file *tfile;
3206 
3207 	DBG1(KERN_INFO, "tunX: tun_chr_open\n");
3208 
3209 	tfile = (struct tun_file *)sk_alloc(net, AF_UNSPEC, GFP_KERNEL,
3210 					    &tun_proto, 0);
3211 	if (!tfile)
3212 		return -ENOMEM;
3213 	RCU_INIT_POINTER(tfile->tun, NULL);
3214 	tfile->flags = 0;
3215 	tfile->ifindex = 0;
3216 
3217 	init_waitqueue_head(&tfile->wq.wait);
3218 	RCU_INIT_POINTER(tfile->socket.wq, &tfile->wq);
3219 
3220 	tfile->socket.file = file;
3221 	tfile->socket.ops = &tun_socket_ops;
3222 
3223 	sock_init_data(&tfile->socket, &tfile->sk);
3224 
3225 	tfile->sk.sk_write_space = tun_sock_write_space;
3226 	tfile->sk.sk_sndbuf = INT_MAX;
3227 
3228 	file->private_data = tfile;
3229 	INIT_LIST_HEAD(&tfile->next);
3230 
3231 	sock_set_flag(&tfile->sk, SOCK_ZEROCOPY);
3232 
3233 	memset(&tfile->tx_ring, 0, sizeof(tfile->tx_ring));
3234 	tfile->xdp_pending_pkts = 0;
3235 
3236 	return 0;
3237 }
3238 
3239 static int tun_chr_close(struct inode *inode, struct file *file)
3240 {
3241 	struct tun_file *tfile = file->private_data;
3242 
3243 	tun_detach(tfile, true);
3244 
3245 	return 0;
3246 }
3247 
3248 #ifdef CONFIG_PROC_FS
3249 static void tun_chr_show_fdinfo(struct seq_file *m, struct file *file)
3250 {
3251 	struct tun_file *tfile = file->private_data;
3252 	struct tun_struct *tun;
3253 	struct ifreq ifr;
3254 
3255 	memset(&ifr, 0, sizeof(ifr));
3256 
3257 	rtnl_lock();
3258 	tun = tun_get(tfile);
3259 	if (tun)
3260 		tun_get_iff(current->nsproxy->net_ns, tun, &ifr);
3261 	rtnl_unlock();
3262 
3263 	if (tun)
3264 		tun_put(tun);
3265 
3266 	seq_printf(m, "iff:\t%s\n", ifr.ifr_name);
3267 }
3268 #endif
3269 
3270 static const struct file_operations tun_fops = {
3271 	.owner	= THIS_MODULE,
3272 	.llseek = no_llseek,
3273 	.read_iter  = tun_chr_read_iter,
3274 	.write_iter = tun_chr_write_iter,
3275 	.poll	= tun_chr_poll,
3276 	.unlocked_ioctl	= tun_chr_ioctl,
3277 #ifdef CONFIG_COMPAT
3278 	.compat_ioctl = tun_chr_compat_ioctl,
3279 #endif
3280 	.open	= tun_chr_open,
3281 	.release = tun_chr_close,
3282 	.fasync = tun_chr_fasync,
3283 #ifdef CONFIG_PROC_FS
3284 	.show_fdinfo = tun_chr_show_fdinfo,
3285 #endif
3286 };
3287 
3288 static struct miscdevice tun_miscdev = {
3289 	.minor = TUN_MINOR,
3290 	.name = "tun",
3291 	.nodename = "net/tun",
3292 	.fops = &tun_fops,
3293 };
3294 
3295 /* ethtool interface */
3296 
3297 static int tun_get_link_ksettings(struct net_device *dev,
3298 				  struct ethtool_link_ksettings *cmd)
3299 {
3300 	ethtool_link_ksettings_zero_link_mode(cmd, supported);
3301 	ethtool_link_ksettings_zero_link_mode(cmd, advertising);
3302 	cmd->base.speed		= SPEED_10;
3303 	cmd->base.duplex	= DUPLEX_FULL;
3304 	cmd->base.port		= PORT_TP;
3305 	cmd->base.phy_address	= 0;
3306 	cmd->base.autoneg	= AUTONEG_DISABLE;
3307 	return 0;
3308 }
3309 
3310 static void tun_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
3311 {
3312 	struct tun_struct *tun = netdev_priv(dev);
3313 
3314 	strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
3315 	strlcpy(info->version, DRV_VERSION, sizeof(info->version));
3316 
3317 	switch (tun->flags & TUN_TYPE_MASK) {
3318 	case IFF_TUN:
3319 		strlcpy(info->bus_info, "tun", sizeof(info->bus_info));
3320 		break;
3321 	case IFF_TAP:
3322 		strlcpy(info->bus_info, "tap", sizeof(info->bus_info));
3323 		break;
3324 	}
3325 }
3326 
3327 static u32 tun_get_msglevel(struct net_device *dev)
3328 {
3329 #ifdef TUN_DEBUG
3330 	struct tun_struct *tun = netdev_priv(dev);
3331 	return tun->debug;
3332 #else
3333 	return -EOPNOTSUPP;
3334 #endif
3335 }
3336 
3337 static void tun_set_msglevel(struct net_device *dev, u32 value)
3338 {
3339 #ifdef TUN_DEBUG
3340 	struct tun_struct *tun = netdev_priv(dev);
3341 	tun->debug = value;
3342 #endif
3343 }
3344 
3345 static int tun_get_coalesce(struct net_device *dev,
3346 			    struct ethtool_coalesce *ec)
3347 {
3348 	struct tun_struct *tun = netdev_priv(dev);
3349 
3350 	ec->rx_max_coalesced_frames = tun->rx_batched;
3351 
3352 	return 0;
3353 }
3354 
3355 static int tun_set_coalesce(struct net_device *dev,
3356 			    struct ethtool_coalesce *ec)
3357 {
3358 	struct tun_struct *tun = netdev_priv(dev);
3359 
3360 	if (ec->rx_max_coalesced_frames > NAPI_POLL_WEIGHT)
3361 		tun->rx_batched = NAPI_POLL_WEIGHT;
3362 	else
3363 		tun->rx_batched = ec->rx_max_coalesced_frames;
3364 
3365 	return 0;
3366 }
3367 
3368 static const struct ethtool_ops tun_ethtool_ops = {
3369 	.get_drvinfo	= tun_get_drvinfo,
3370 	.get_msglevel	= tun_get_msglevel,
3371 	.set_msglevel	= tun_set_msglevel,
3372 	.get_link	= ethtool_op_get_link,
3373 	.get_ts_info	= ethtool_op_get_ts_info,
3374 	.get_coalesce   = tun_get_coalesce,
3375 	.set_coalesce   = tun_set_coalesce,
3376 	.get_link_ksettings = tun_get_link_ksettings,
3377 };
3378 
3379 static int tun_queue_resize(struct tun_struct *tun)
3380 {
3381 	struct net_device *dev = tun->dev;
3382 	struct tun_file *tfile;
3383 	struct ptr_ring **rings;
3384 	int n = tun->numqueues + tun->numdisabled;
3385 	int ret, i;
3386 
3387 	rings = kmalloc_array(n, sizeof(*rings), GFP_KERNEL);
3388 	if (!rings)
3389 		return -ENOMEM;
3390 
3391 	for (i = 0; i < tun->numqueues; i++) {
3392 		tfile = rtnl_dereference(tun->tfiles[i]);
3393 		rings[i] = &tfile->tx_ring;
3394 	}
3395 	list_for_each_entry(tfile, &tun->disabled, next)
3396 		rings[i++] = &tfile->tx_ring;
3397 
3398 	ret = ptr_ring_resize_multiple(rings, n,
3399 				       dev->tx_queue_len, GFP_KERNEL,
3400 				       tun_ptr_free);
3401 
3402 	kfree(rings);
3403 	return ret;
3404 }
3405 
3406 static int tun_device_event(struct notifier_block *unused,
3407 			    unsigned long event, void *ptr)
3408 {
3409 	struct net_device *dev = netdev_notifier_info_to_dev(ptr);
3410 	struct tun_struct *tun = netdev_priv(dev);
3411 
3412 	if (dev->rtnl_link_ops != &tun_link_ops)
3413 		return NOTIFY_DONE;
3414 
3415 	switch (event) {
3416 	case NETDEV_CHANGE_TX_QUEUE_LEN:
3417 		if (tun_queue_resize(tun))
3418 			return NOTIFY_BAD;
3419 		break;
3420 	default:
3421 		break;
3422 	}
3423 
3424 	return NOTIFY_DONE;
3425 }
3426 
3427 static struct notifier_block tun_notifier_block __read_mostly = {
3428 	.notifier_call	= tun_device_event,
3429 };
3430 
3431 static int __init tun_init(void)
3432 {
3433 	int ret = 0;
3434 
3435 	pr_info("%s, %s\n", DRV_DESCRIPTION, DRV_VERSION);
3436 
3437 	ret = rtnl_link_register(&tun_link_ops);
3438 	if (ret) {
3439 		pr_err("Can't register link_ops\n");
3440 		goto err_linkops;
3441 	}
3442 
3443 	ret = misc_register(&tun_miscdev);
3444 	if (ret) {
3445 		pr_err("Can't register misc device %d\n", TUN_MINOR);
3446 		goto err_misc;
3447 	}
3448 
3449 	ret = register_netdevice_notifier(&tun_notifier_block);
3450 	if (ret) {
3451 		pr_err("Can't register netdevice notifier\n");
3452 		goto err_notifier;
3453 	}
3454 
3455 	return  0;
3456 
3457 err_notifier:
3458 	misc_deregister(&tun_miscdev);
3459 err_misc:
3460 	rtnl_link_unregister(&tun_link_ops);
3461 err_linkops:
3462 	return ret;
3463 }
3464 
3465 static void tun_cleanup(void)
3466 {
3467 	misc_deregister(&tun_miscdev);
3468 	rtnl_link_unregister(&tun_link_ops);
3469 	unregister_netdevice_notifier(&tun_notifier_block);
3470 }
3471 
3472 /* Get an underlying socket object from tun file.  Returns error unless file is
3473  * attached to a device.  The returned object works like a packet socket, it
3474  * can be used for sock_sendmsg/sock_recvmsg.  The caller is responsible for
3475  * holding a reference to the file for as long as the socket is in use. */
3476 struct socket *tun_get_socket(struct file *file)
3477 {
3478 	struct tun_file *tfile;
3479 	if (file->f_op != &tun_fops)
3480 		return ERR_PTR(-EINVAL);
3481 	tfile = file->private_data;
3482 	if (!tfile)
3483 		return ERR_PTR(-EBADFD);
3484 	return &tfile->socket;
3485 }
3486 EXPORT_SYMBOL_GPL(tun_get_socket);
3487 
3488 struct ptr_ring *tun_get_tx_ring(struct file *file)
3489 {
3490 	struct tun_file *tfile;
3491 
3492 	if (file->f_op != &tun_fops)
3493 		return ERR_PTR(-EINVAL);
3494 	tfile = file->private_data;
3495 	if (!tfile)
3496 		return ERR_PTR(-EBADFD);
3497 	return &tfile->tx_ring;
3498 }
3499 EXPORT_SYMBOL_GPL(tun_get_tx_ring);
3500 
3501 module_init(tun_init);
3502 module_exit(tun_cleanup);
3503 MODULE_DESCRIPTION(DRV_DESCRIPTION);
3504 MODULE_AUTHOR(DRV_COPYRIGHT);
3505 MODULE_LICENSE("GPL");
3506 MODULE_ALIAS_MISCDEV(TUN_MINOR);
3507 MODULE_ALIAS("devname:net/tun");
3508