xref: /openbmc/linux/drivers/net/tun.c (revision 8795a739)
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3  *  TUN - Universal TUN/TAP device driver.
4  *  Copyright (C) 1999-2002 Maxim Krasnyansky <maxk@qualcomm.com>
5  *
6  *  $Id: tun.c,v 1.15 2002/03/01 02:44:24 maxk Exp $
7  */
8 
9 /*
10  *  Changes:
11  *
12  *  Mike Kershaw <dragorn@kismetwireless.net> 2005/08/14
13  *    Add TUNSETLINK ioctl to set the link encapsulation
14  *
15  *  Mark Smith <markzzzsmith@yahoo.com.au>
16  *    Use eth_random_addr() for tap MAC address.
17  *
18  *  Harald Roelle <harald.roelle@ifi.lmu.de>  2004/04/20
19  *    Fixes in packet dropping, queue length setting and queue wakeup.
20  *    Increased default tx queue length.
21  *    Added ethtool API.
22  *    Minor cleanups
23  *
24  *  Daniel Podlejski <underley@underley.eu.org>
25  *    Modifications for 2.3.99-pre5 kernel.
26  */
27 
28 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
29 
30 #define DRV_NAME	"tun"
31 #define DRV_VERSION	"1.6"
32 #define DRV_DESCRIPTION	"Universal TUN/TAP device driver"
33 #define DRV_COPYRIGHT	"(C) 1999-2004 Max Krasnyansky <maxk@qualcomm.com>"
34 
35 #include <linux/module.h>
36 #include <linux/errno.h>
37 #include <linux/kernel.h>
38 #include <linux/sched/signal.h>
39 #include <linux/major.h>
40 #include <linux/slab.h>
41 #include <linux/poll.h>
42 #include <linux/fcntl.h>
43 #include <linux/init.h>
44 #include <linux/skbuff.h>
45 #include <linux/netdevice.h>
46 #include <linux/etherdevice.h>
47 #include <linux/miscdevice.h>
48 #include <linux/ethtool.h>
49 #include <linux/rtnetlink.h>
50 #include <linux/compat.h>
51 #include <linux/if.h>
52 #include <linux/if_arp.h>
53 #include <linux/if_ether.h>
54 #include <linux/if_tun.h>
55 #include <linux/if_vlan.h>
56 #include <linux/crc32.h>
57 #include <linux/nsproxy.h>
58 #include <linux/virtio_net.h>
59 #include <linux/rcupdate.h>
60 #include <net/net_namespace.h>
61 #include <net/netns/generic.h>
62 #include <net/rtnetlink.h>
63 #include <net/sock.h>
64 #include <net/xdp.h>
65 #include <linux/seq_file.h>
66 #include <linux/uio.h>
67 #include <linux/skb_array.h>
68 #include <linux/bpf.h>
69 #include <linux/bpf_trace.h>
70 #include <linux/mutex.h>
71 
72 #include <linux/uaccess.h>
73 #include <linux/proc_fs.h>
74 
75 static void tun_default_link_ksettings(struct net_device *dev,
76 				       struct ethtool_link_ksettings *cmd);
77 
78 /* Uncomment to enable debugging */
79 /* #define TUN_DEBUG 1 */
80 
81 #ifdef TUN_DEBUG
82 static int debug;
83 
84 #define tun_debug(level, tun, fmt, args...)			\
85 do {								\
86 	if (tun->debug)						\
87 		netdev_printk(level, tun->dev, fmt, ##args);	\
88 } while (0)
89 #define DBG1(level, fmt, args...)				\
90 do {								\
91 	if (debug == 2)						\
92 		printk(level fmt, ##args);			\
93 } while (0)
94 #else
95 #define tun_debug(level, tun, fmt, args...)			\
96 do {								\
97 	if (0)							\
98 		netdev_printk(level, tun->dev, fmt, ##args);	\
99 } while (0)
100 #define DBG1(level, fmt, args...)				\
101 do {								\
102 	if (0)							\
103 		printk(level fmt, ##args);			\
104 } while (0)
105 #endif
106 
107 #define TUN_RX_PAD (NET_IP_ALIGN + NET_SKB_PAD)
108 
109 /* TUN device flags */
110 
111 /* IFF_ATTACH_QUEUE is never stored in device flags,
112  * overload it to mean fasync when stored there.
113  */
114 #define TUN_FASYNC	IFF_ATTACH_QUEUE
115 /* High bits in flags field are unused. */
116 #define TUN_VNET_LE     0x80000000
117 #define TUN_VNET_BE     0x40000000
118 
119 #define TUN_FEATURES (IFF_NO_PI | IFF_ONE_QUEUE | IFF_VNET_HDR | \
120 		      IFF_MULTI_QUEUE | IFF_NAPI | IFF_NAPI_FRAGS)
121 
122 #define GOODCOPY_LEN 128
123 
124 #define FLT_EXACT_COUNT 8
125 struct tap_filter {
126 	unsigned int    count;    /* Number of addrs. Zero means disabled */
127 	u32             mask[2];  /* Mask of the hashed addrs */
128 	unsigned char	addr[FLT_EXACT_COUNT][ETH_ALEN];
129 };
130 
131 /* MAX_TAP_QUEUES 256 is chosen to allow rx/tx queues to be equal
132  * to max number of VCPUs in guest. */
133 #define MAX_TAP_QUEUES 256
134 #define MAX_TAP_FLOWS  4096
135 
136 #define TUN_FLOW_EXPIRE (3 * HZ)
137 
138 struct tun_pcpu_stats {
139 	u64 rx_packets;
140 	u64 rx_bytes;
141 	u64 tx_packets;
142 	u64 tx_bytes;
143 	struct u64_stats_sync syncp;
144 	u32 rx_dropped;
145 	u32 tx_dropped;
146 	u32 rx_frame_errors;
147 };
148 
149 /* A tun_file connects an open character device to a tuntap netdevice. It
150  * also contains all socket related structures (except sock_fprog and tap_filter)
151  * to serve as one transmit queue for tuntap device. The sock_fprog and
152  * tap_filter were kept in tun_struct since they were used for filtering for the
153  * netdevice not for a specific queue (at least I didn't see the requirement for
154  * this).
155  *
156  * RCU usage:
157  * The tun_file and tun_struct are loosely coupled, the pointer from one to the
158  * other can only be read while rcu_read_lock or rtnl_lock is held.
159  */
160 struct tun_file {
161 	struct sock sk;
162 	struct socket socket;
163 	struct tun_struct __rcu *tun;
164 	struct fasync_struct *fasync;
165 	/* only used for fasnyc */
166 	unsigned int flags;
167 	union {
168 		u16 queue_index;
169 		unsigned int ifindex;
170 	};
171 	struct napi_struct napi;
172 	bool napi_enabled;
173 	bool napi_frags_enabled;
174 	struct mutex napi_mutex;	/* Protects access to the above napi */
175 	struct list_head next;
176 	struct tun_struct *detached;
177 	struct ptr_ring tx_ring;
178 	struct xdp_rxq_info xdp_rxq;
179 };
180 
181 struct tun_page {
182 	struct page *page;
183 	int count;
184 };
185 
186 struct tun_flow_entry {
187 	struct hlist_node hash_link;
188 	struct rcu_head rcu;
189 	struct tun_struct *tun;
190 
191 	u32 rxhash;
192 	u32 rps_rxhash;
193 	int queue_index;
194 	unsigned long updated ____cacheline_aligned_in_smp;
195 };
196 
197 #define TUN_NUM_FLOW_ENTRIES 1024
198 #define TUN_MASK_FLOW_ENTRIES (TUN_NUM_FLOW_ENTRIES - 1)
199 
200 struct tun_prog {
201 	struct rcu_head rcu;
202 	struct bpf_prog *prog;
203 };
204 
205 /* Since the socket were moved to tun_file, to preserve the behavior of persist
206  * device, socket filter, sndbuf and vnet header size were restore when the
207  * file were attached to a persist device.
208  */
209 struct tun_struct {
210 	struct tun_file __rcu	*tfiles[MAX_TAP_QUEUES];
211 	unsigned int            numqueues;
212 	unsigned int 		flags;
213 	kuid_t			owner;
214 	kgid_t			group;
215 
216 	struct net_device	*dev;
217 	netdev_features_t	set_features;
218 #define TUN_USER_FEATURES (NETIF_F_HW_CSUM|NETIF_F_TSO_ECN|NETIF_F_TSO| \
219 			  NETIF_F_TSO6)
220 
221 	int			align;
222 	int			vnet_hdr_sz;
223 	int			sndbuf;
224 	struct tap_filter	txflt;
225 	struct sock_fprog	fprog;
226 	/* protected by rtnl lock */
227 	bool			filter_attached;
228 #ifdef TUN_DEBUG
229 	int debug;
230 #endif
231 	spinlock_t lock;
232 	struct hlist_head flows[TUN_NUM_FLOW_ENTRIES];
233 	struct timer_list flow_gc_timer;
234 	unsigned long ageing_time;
235 	unsigned int numdisabled;
236 	struct list_head disabled;
237 	void *security;
238 	u32 flow_count;
239 	u32 rx_batched;
240 	struct tun_pcpu_stats __percpu *pcpu_stats;
241 	struct bpf_prog __rcu *xdp_prog;
242 	struct tun_prog __rcu *steering_prog;
243 	struct tun_prog __rcu *filter_prog;
244 	struct ethtool_link_ksettings link_ksettings;
245 };
246 
247 struct veth {
248 	__be16 h_vlan_proto;
249 	__be16 h_vlan_TCI;
250 };
251 
252 bool tun_is_xdp_frame(void *ptr)
253 {
254 	return (unsigned long)ptr & TUN_XDP_FLAG;
255 }
256 EXPORT_SYMBOL(tun_is_xdp_frame);
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, bool napi_frags)
312 {
313 	tfile->napi_enabled = napi_en;
314 	tfile->napi_frags_enabled = napi_en && napi_frags;
315 	if (napi_en) {
316 		netif_napi_add(tun->dev, &tfile->napi, tun_napi_poll,
317 			       NAPI_POLL_WEIGHT);
318 		napi_enable(&tfile->napi);
319 	}
320 }
321 
322 static void tun_napi_disable(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_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_file *tfile)
335 {
336 	return tfile->napi_frags_enabled;
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 & TUN_MASK_FLOW_ENTRIES;
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 	head = &tun->flows[tun_hashfn(rxhash)];
523 
524 	rcu_read_lock();
525 
526 	e = tun_flow_find(head, rxhash);
527 	if (likely(e)) {
528 		/* TODO: keep queueing to old queue until it's empty? */
529 		if (e->queue_index != queue_index)
530 			e->queue_index = queue_index;
531 		if (e->updated != jiffies)
532 			e->updated = jiffies;
533 		sock_rps_record_flow_hash(e->rps_rxhash);
534 	} else {
535 		spin_lock_bh(&tun->lock);
536 		if (!tun_flow_find(head, rxhash) &&
537 		    tun->flow_count < MAX_TAP_FLOWS)
538 			tun_flow_create(tun, head, rxhash, queue_index);
539 
540 		if (!timer_pending(&tun->flow_gc_timer))
541 			mod_timer(&tun->flow_gc_timer,
542 				  round_jiffies_up(jiffies + delay));
543 		spin_unlock_bh(&tun->lock);
544 	}
545 
546 	rcu_read_unlock();
547 }
548 
549 /**
550  * Save the hash received in the stack receive path and update the
551  * flow_hash table accordingly.
552  */
553 static inline void tun_flow_save_rps_rxhash(struct tun_flow_entry *e, u32 hash)
554 {
555 	if (unlikely(e->rps_rxhash != hash))
556 		e->rps_rxhash = hash;
557 }
558 
559 /* We try to identify a flow through its rxhash. The reason that
560  * we do not check rxq no. is because some cards(e.g 82599), chooses
561  * the rxq based on the txq where the last packet of the flow comes. As
562  * the userspace application move between processors, we may get a
563  * different rxq no. here.
564  */
565 static u16 tun_automq_select_queue(struct tun_struct *tun, struct sk_buff *skb)
566 {
567 	struct tun_flow_entry *e;
568 	u32 txq = 0;
569 	u32 numqueues = 0;
570 
571 	numqueues = READ_ONCE(tun->numqueues);
572 
573 	txq = __skb_get_hash_symmetric(skb);
574 	e = tun_flow_find(&tun->flows[tun_hashfn(txq)], txq);
575 	if (e) {
576 		tun_flow_save_rps_rxhash(e, txq);
577 		txq = e->queue_index;
578 	} else {
579 		/* use multiply and shift instead of expensive divide */
580 		txq = ((u64)txq * numqueues) >> 32;
581 	}
582 
583 	return txq;
584 }
585 
586 static u16 tun_ebpf_select_queue(struct tun_struct *tun, struct sk_buff *skb)
587 {
588 	struct tun_prog *prog;
589 	u32 numqueues;
590 	u16 ret = 0;
591 
592 	numqueues = READ_ONCE(tun->numqueues);
593 	if (!numqueues)
594 		return 0;
595 
596 	prog = rcu_dereference(tun->steering_prog);
597 	if (prog)
598 		ret = bpf_prog_run_clear_cb(prog->prog, skb);
599 
600 	return ret % numqueues;
601 }
602 
603 static u16 tun_select_queue(struct net_device *dev, struct sk_buff *skb,
604 			    struct net_device *sb_dev)
605 {
606 	struct tun_struct *tun = netdev_priv(dev);
607 	u16 ret;
608 
609 	rcu_read_lock();
610 	if (rcu_dereference(tun->steering_prog))
611 		ret = tun_ebpf_select_queue(tun, skb);
612 	else
613 		ret = tun_automq_select_queue(tun, skb);
614 	rcu_read_unlock();
615 
616 	return ret;
617 }
618 
619 static inline bool tun_not_capable(struct tun_struct *tun)
620 {
621 	const struct cred *cred = current_cred();
622 	struct net *net = dev_net(tun->dev);
623 
624 	return ((uid_valid(tun->owner) && !uid_eq(cred->euid, tun->owner)) ||
625 		  (gid_valid(tun->group) && !in_egroup_p(tun->group))) &&
626 		!ns_capable(net->user_ns, CAP_NET_ADMIN);
627 }
628 
629 static void tun_set_real_num_queues(struct tun_struct *tun)
630 {
631 	netif_set_real_num_tx_queues(tun->dev, tun->numqueues);
632 	netif_set_real_num_rx_queues(tun->dev, tun->numqueues);
633 }
634 
635 static void tun_disable_queue(struct tun_struct *tun, struct tun_file *tfile)
636 {
637 	tfile->detached = tun;
638 	list_add_tail(&tfile->next, &tun->disabled);
639 	++tun->numdisabled;
640 }
641 
642 static struct tun_struct *tun_enable_queue(struct tun_file *tfile)
643 {
644 	struct tun_struct *tun = tfile->detached;
645 
646 	tfile->detached = NULL;
647 	list_del_init(&tfile->next);
648 	--tun->numdisabled;
649 	return tun;
650 }
651 
652 void tun_ptr_free(void *ptr)
653 {
654 	if (!ptr)
655 		return;
656 	if (tun_is_xdp_frame(ptr)) {
657 		struct xdp_frame *xdpf = tun_ptr_to_xdp(ptr);
658 
659 		xdp_return_frame(xdpf);
660 	} else {
661 		__skb_array_destroy_skb(ptr);
662 	}
663 }
664 EXPORT_SYMBOL_GPL(tun_ptr_free);
665 
666 static void tun_queue_purge(struct tun_file *tfile)
667 {
668 	void *ptr;
669 
670 	while ((ptr = ptr_ring_consume(&tfile->tx_ring)) != NULL)
671 		tun_ptr_free(ptr);
672 
673 	skb_queue_purge(&tfile->sk.sk_write_queue);
674 	skb_queue_purge(&tfile->sk.sk_error_queue);
675 }
676 
677 static void __tun_detach(struct tun_file *tfile, bool clean)
678 {
679 	struct tun_file *ntfile;
680 	struct tun_struct *tun;
681 
682 	tun = rtnl_dereference(tfile->tun);
683 
684 	if (tun && clean) {
685 		tun_napi_disable(tfile);
686 		tun_napi_del(tfile);
687 	}
688 
689 	if (tun && !tfile->detached) {
690 		u16 index = tfile->queue_index;
691 		BUG_ON(index >= tun->numqueues);
692 
693 		rcu_assign_pointer(tun->tfiles[index],
694 				   tun->tfiles[tun->numqueues - 1]);
695 		ntfile = rtnl_dereference(tun->tfiles[index]);
696 		ntfile->queue_index = index;
697 		rcu_assign_pointer(tun->tfiles[tun->numqueues - 1],
698 				   NULL);
699 
700 		--tun->numqueues;
701 		if (clean) {
702 			RCU_INIT_POINTER(tfile->tun, NULL);
703 			sock_put(&tfile->sk);
704 		} else
705 			tun_disable_queue(tun, tfile);
706 
707 		synchronize_net();
708 		tun_flow_delete_by_queue(tun, tun->numqueues + 1);
709 		/* Drop read queue */
710 		tun_queue_purge(tfile);
711 		tun_set_real_num_queues(tun);
712 	} else if (tfile->detached && clean) {
713 		tun = tun_enable_queue(tfile);
714 		sock_put(&tfile->sk);
715 	}
716 
717 	if (clean) {
718 		if (tun && tun->numqueues == 0 && tun->numdisabled == 0) {
719 			netif_carrier_off(tun->dev);
720 
721 			if (!(tun->flags & IFF_PERSIST) &&
722 			    tun->dev->reg_state == NETREG_REGISTERED)
723 				unregister_netdevice(tun->dev);
724 		}
725 		if (tun)
726 			xdp_rxq_info_unreg(&tfile->xdp_rxq);
727 		ptr_ring_cleanup(&tfile->tx_ring, tun_ptr_free);
728 		sock_put(&tfile->sk);
729 	}
730 }
731 
732 static void tun_detach(struct tun_file *tfile, bool clean)
733 {
734 	struct tun_struct *tun;
735 	struct net_device *dev;
736 
737 	rtnl_lock();
738 	tun = rtnl_dereference(tfile->tun);
739 	dev = tun ? tun->dev : NULL;
740 	__tun_detach(tfile, clean);
741 	if (dev)
742 		netdev_state_change(dev);
743 	rtnl_unlock();
744 }
745 
746 static void tun_detach_all(struct net_device *dev)
747 {
748 	struct tun_struct *tun = netdev_priv(dev);
749 	struct tun_file *tfile, *tmp;
750 	int i, n = tun->numqueues;
751 
752 	for (i = 0; i < n; i++) {
753 		tfile = rtnl_dereference(tun->tfiles[i]);
754 		BUG_ON(!tfile);
755 		tun_napi_disable(tfile);
756 		tfile->socket.sk->sk_shutdown = RCV_SHUTDOWN;
757 		tfile->socket.sk->sk_data_ready(tfile->socket.sk);
758 		RCU_INIT_POINTER(tfile->tun, NULL);
759 		--tun->numqueues;
760 	}
761 	list_for_each_entry(tfile, &tun->disabled, next) {
762 		tfile->socket.sk->sk_shutdown = RCV_SHUTDOWN;
763 		tfile->socket.sk->sk_data_ready(tfile->socket.sk);
764 		RCU_INIT_POINTER(tfile->tun, NULL);
765 	}
766 	BUG_ON(tun->numqueues != 0);
767 
768 	synchronize_net();
769 	for (i = 0; i < n; i++) {
770 		tfile = rtnl_dereference(tun->tfiles[i]);
771 		tun_napi_del(tfile);
772 		/* Drop read queue */
773 		tun_queue_purge(tfile);
774 		xdp_rxq_info_unreg(&tfile->xdp_rxq);
775 		sock_put(&tfile->sk);
776 	}
777 	list_for_each_entry_safe(tfile, tmp, &tun->disabled, next) {
778 		tun_enable_queue(tfile);
779 		tun_queue_purge(tfile);
780 		xdp_rxq_info_unreg(&tfile->xdp_rxq);
781 		sock_put(&tfile->sk);
782 	}
783 	BUG_ON(tun->numdisabled != 0);
784 
785 	if (tun->flags & IFF_PERSIST)
786 		module_put(THIS_MODULE);
787 }
788 
789 static int tun_attach(struct tun_struct *tun, struct file *file,
790 		      bool skip_filter, bool napi, bool napi_frags,
791 		      bool publish_tun)
792 {
793 	struct tun_file *tfile = file->private_data;
794 	struct net_device *dev = tun->dev;
795 	int err;
796 
797 	err = security_tun_dev_attach(tfile->socket.sk, tun->security);
798 	if (err < 0)
799 		goto out;
800 
801 	err = -EINVAL;
802 	if (rtnl_dereference(tfile->tun) && !tfile->detached)
803 		goto out;
804 
805 	err = -EBUSY;
806 	if (!(tun->flags & IFF_MULTI_QUEUE) && tun->numqueues == 1)
807 		goto out;
808 
809 	err = -E2BIG;
810 	if (!tfile->detached &&
811 	    tun->numqueues + tun->numdisabled == MAX_TAP_QUEUES)
812 		goto out;
813 
814 	err = 0;
815 
816 	/* Re-attach the filter to persist device */
817 	if (!skip_filter && (tun->filter_attached == true)) {
818 		lock_sock(tfile->socket.sk);
819 		err = sk_attach_filter(&tun->fprog, tfile->socket.sk);
820 		release_sock(tfile->socket.sk);
821 		if (!err)
822 			goto out;
823 	}
824 
825 	if (!tfile->detached &&
826 	    ptr_ring_resize(&tfile->tx_ring, dev->tx_queue_len,
827 			    GFP_KERNEL, tun_ptr_free)) {
828 		err = -ENOMEM;
829 		goto out;
830 	}
831 
832 	tfile->queue_index = tun->numqueues;
833 	tfile->socket.sk->sk_shutdown &= ~RCV_SHUTDOWN;
834 
835 	if (tfile->detached) {
836 		/* Re-attach detached tfile, updating XDP queue_index */
837 		WARN_ON(!xdp_rxq_info_is_reg(&tfile->xdp_rxq));
838 
839 		if (tfile->xdp_rxq.queue_index    != tfile->queue_index)
840 			tfile->xdp_rxq.queue_index = tfile->queue_index;
841 	} else {
842 		/* Setup XDP RX-queue info, for new tfile getting attached */
843 		err = xdp_rxq_info_reg(&tfile->xdp_rxq,
844 				       tun->dev, tfile->queue_index);
845 		if (err < 0)
846 			goto out;
847 		err = xdp_rxq_info_reg_mem_model(&tfile->xdp_rxq,
848 						 MEM_TYPE_PAGE_SHARED, NULL);
849 		if (err < 0) {
850 			xdp_rxq_info_unreg(&tfile->xdp_rxq);
851 			goto out;
852 		}
853 		err = 0;
854 	}
855 
856 	if (tfile->detached) {
857 		tun_enable_queue(tfile);
858 	} else {
859 		sock_hold(&tfile->sk);
860 		tun_napi_init(tun, tfile, napi, napi_frags);
861 	}
862 
863 	if (rtnl_dereference(tun->xdp_prog))
864 		sock_set_flag(&tfile->sk, SOCK_XDP);
865 
866 	/* device is allowed to go away first, so no need to hold extra
867 	 * refcnt.
868 	 */
869 
870 	/* Publish tfile->tun and tun->tfiles only after we've fully
871 	 * initialized tfile; otherwise we risk using half-initialized
872 	 * object.
873 	 */
874 	if (publish_tun)
875 		rcu_assign_pointer(tfile->tun, tun);
876 	rcu_assign_pointer(tun->tfiles[tun->numqueues], tfile);
877 	tun->numqueues++;
878 	tun_set_real_num_queues(tun);
879 out:
880 	return err;
881 }
882 
883 static struct tun_struct *tun_get(struct tun_file *tfile)
884 {
885 	struct tun_struct *tun;
886 
887 	rcu_read_lock();
888 	tun = rcu_dereference(tfile->tun);
889 	if (tun)
890 		dev_hold(tun->dev);
891 	rcu_read_unlock();
892 
893 	return tun;
894 }
895 
896 static void tun_put(struct tun_struct *tun)
897 {
898 	dev_put(tun->dev);
899 }
900 
901 /* TAP filtering */
902 static void addr_hash_set(u32 *mask, const u8 *addr)
903 {
904 	int n = ether_crc(ETH_ALEN, addr) >> 26;
905 	mask[n >> 5] |= (1 << (n & 31));
906 }
907 
908 static unsigned int addr_hash_test(const u32 *mask, const u8 *addr)
909 {
910 	int n = ether_crc(ETH_ALEN, addr) >> 26;
911 	return mask[n >> 5] & (1 << (n & 31));
912 }
913 
914 static int update_filter(struct tap_filter *filter, void __user *arg)
915 {
916 	struct { u8 u[ETH_ALEN]; } *addr;
917 	struct tun_filter uf;
918 	int err, alen, n, nexact;
919 
920 	if (copy_from_user(&uf, arg, sizeof(uf)))
921 		return -EFAULT;
922 
923 	if (!uf.count) {
924 		/* Disabled */
925 		filter->count = 0;
926 		return 0;
927 	}
928 
929 	alen = ETH_ALEN * uf.count;
930 	addr = memdup_user(arg + sizeof(uf), alen);
931 	if (IS_ERR(addr))
932 		return PTR_ERR(addr);
933 
934 	/* The filter is updated without holding any locks. Which is
935 	 * perfectly safe. We disable it first and in the worst
936 	 * case we'll accept a few undesired packets. */
937 	filter->count = 0;
938 	wmb();
939 
940 	/* Use first set of addresses as an exact filter */
941 	for (n = 0; n < uf.count && n < FLT_EXACT_COUNT; n++)
942 		memcpy(filter->addr[n], addr[n].u, ETH_ALEN);
943 
944 	nexact = n;
945 
946 	/* Remaining multicast addresses are hashed,
947 	 * unicast will leave the filter disabled. */
948 	memset(filter->mask, 0, sizeof(filter->mask));
949 	for (; n < uf.count; n++) {
950 		if (!is_multicast_ether_addr(addr[n].u)) {
951 			err = 0; /* no filter */
952 			goto free_addr;
953 		}
954 		addr_hash_set(filter->mask, addr[n].u);
955 	}
956 
957 	/* For ALLMULTI just set the mask to all ones.
958 	 * This overrides the mask populated above. */
959 	if ((uf.flags & TUN_FLT_ALLMULTI))
960 		memset(filter->mask, ~0, sizeof(filter->mask));
961 
962 	/* Now enable the filter */
963 	wmb();
964 	filter->count = nexact;
965 
966 	/* Return the number of exact filters */
967 	err = nexact;
968 free_addr:
969 	kfree(addr);
970 	return err;
971 }
972 
973 /* Returns: 0 - drop, !=0 - accept */
974 static int run_filter(struct tap_filter *filter, const struct sk_buff *skb)
975 {
976 	/* Cannot use eth_hdr(skb) here because skb_mac_hdr() is incorrect
977 	 * at this point. */
978 	struct ethhdr *eh = (struct ethhdr *) skb->data;
979 	int i;
980 
981 	/* Exact match */
982 	for (i = 0; i < filter->count; i++)
983 		if (ether_addr_equal(eh->h_dest, filter->addr[i]))
984 			return 1;
985 
986 	/* Inexact match (multicast only) */
987 	if (is_multicast_ether_addr(eh->h_dest))
988 		return addr_hash_test(filter->mask, eh->h_dest);
989 
990 	return 0;
991 }
992 
993 /*
994  * Checks whether the packet is accepted or not.
995  * Returns: 0 - drop, !=0 - accept
996  */
997 static int check_filter(struct tap_filter *filter, const struct sk_buff *skb)
998 {
999 	if (!filter->count)
1000 		return 1;
1001 
1002 	return run_filter(filter, skb);
1003 }
1004 
1005 /* Network device part of the driver */
1006 
1007 static const struct ethtool_ops tun_ethtool_ops;
1008 
1009 /* Net device detach from fd. */
1010 static void tun_net_uninit(struct net_device *dev)
1011 {
1012 	tun_detach_all(dev);
1013 }
1014 
1015 /* Net device open. */
1016 static int tun_net_open(struct net_device *dev)
1017 {
1018 	netif_tx_start_all_queues(dev);
1019 
1020 	return 0;
1021 }
1022 
1023 /* Net device close. */
1024 static int tun_net_close(struct net_device *dev)
1025 {
1026 	netif_tx_stop_all_queues(dev);
1027 	return 0;
1028 }
1029 
1030 /* Net device start xmit */
1031 static void tun_automq_xmit(struct tun_struct *tun, struct sk_buff *skb)
1032 {
1033 #ifdef CONFIG_RPS
1034 	if (tun->numqueues == 1 && static_branch_unlikely(&rps_needed)) {
1035 		/* Select queue was not called for the skbuff, so we extract the
1036 		 * RPS hash and save it into the flow_table here.
1037 		 */
1038 		struct tun_flow_entry *e;
1039 		__u32 rxhash;
1040 
1041 		rxhash = __skb_get_hash_symmetric(skb);
1042 		e = tun_flow_find(&tun->flows[tun_hashfn(rxhash)], rxhash);
1043 		if (e)
1044 			tun_flow_save_rps_rxhash(e, rxhash);
1045 	}
1046 #endif
1047 }
1048 
1049 static unsigned int run_ebpf_filter(struct tun_struct *tun,
1050 				    struct sk_buff *skb,
1051 				    int len)
1052 {
1053 	struct tun_prog *prog = rcu_dereference(tun->filter_prog);
1054 
1055 	if (prog)
1056 		len = bpf_prog_run_clear_cb(prog->prog, skb);
1057 
1058 	return len;
1059 }
1060 
1061 /* Net device start xmit */
1062 static netdev_tx_t tun_net_xmit(struct sk_buff *skb, struct net_device *dev)
1063 {
1064 	struct tun_struct *tun = netdev_priv(dev);
1065 	int txq = skb->queue_mapping;
1066 	struct tun_file *tfile;
1067 	int len = skb->len;
1068 
1069 	rcu_read_lock();
1070 	tfile = rcu_dereference(tun->tfiles[txq]);
1071 
1072 	/* Drop packet if interface is not attached */
1073 	if (!tfile)
1074 		goto drop;
1075 
1076 	if (!rcu_dereference(tun->steering_prog))
1077 		tun_automq_xmit(tun, skb);
1078 
1079 	tun_debug(KERN_INFO, tun, "tun_net_xmit %d\n", skb->len);
1080 
1081 	BUG_ON(!tfile);
1082 
1083 	/* Drop if the filter does not like it.
1084 	 * This is a noop if the filter is disabled.
1085 	 * Filter can be enabled only for the TAP devices. */
1086 	if (!check_filter(&tun->txflt, skb))
1087 		goto drop;
1088 
1089 	if (tfile->socket.sk->sk_filter &&
1090 	    sk_filter(tfile->socket.sk, skb))
1091 		goto drop;
1092 
1093 	len = run_ebpf_filter(tun, skb, len);
1094 	if (len == 0 || pskb_trim(skb, len))
1095 		goto drop;
1096 
1097 	if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC)))
1098 		goto drop;
1099 
1100 	skb_tx_timestamp(skb);
1101 
1102 	/* Orphan the skb - required as we might hang on to it
1103 	 * for indefinite time.
1104 	 */
1105 	skb_orphan(skb);
1106 
1107 	nf_reset_ct(skb);
1108 
1109 	if (ptr_ring_produce(&tfile->tx_ring, skb))
1110 		goto drop;
1111 
1112 	/* Notify and wake up reader process */
1113 	if (tfile->flags & TUN_FASYNC)
1114 		kill_fasync(&tfile->fasync, SIGIO, POLL_IN);
1115 	tfile->socket.sk->sk_data_ready(tfile->socket.sk);
1116 
1117 	rcu_read_unlock();
1118 	return NETDEV_TX_OK;
1119 
1120 drop:
1121 	this_cpu_inc(tun->pcpu_stats->tx_dropped);
1122 	skb_tx_error(skb);
1123 	kfree_skb(skb);
1124 	rcu_read_unlock();
1125 	return NET_XMIT_DROP;
1126 }
1127 
1128 static void tun_net_mclist(struct net_device *dev)
1129 {
1130 	/*
1131 	 * This callback is supposed to deal with mc filter in
1132 	 * _rx_ path and has nothing to do with the _tx_ path.
1133 	 * In rx path we always accept everything userspace gives us.
1134 	 */
1135 }
1136 
1137 static netdev_features_t tun_net_fix_features(struct net_device *dev,
1138 	netdev_features_t features)
1139 {
1140 	struct tun_struct *tun = netdev_priv(dev);
1141 
1142 	return (features & tun->set_features) | (features & ~TUN_USER_FEATURES);
1143 }
1144 
1145 static void tun_set_headroom(struct net_device *dev, int new_hr)
1146 {
1147 	struct tun_struct *tun = netdev_priv(dev);
1148 
1149 	if (new_hr < NET_SKB_PAD)
1150 		new_hr = NET_SKB_PAD;
1151 
1152 	tun->align = new_hr;
1153 }
1154 
1155 static void
1156 tun_net_get_stats64(struct net_device *dev, struct rtnl_link_stats64 *stats)
1157 {
1158 	u32 rx_dropped = 0, tx_dropped = 0, rx_frame_errors = 0;
1159 	struct tun_struct *tun = netdev_priv(dev);
1160 	struct tun_pcpu_stats *p;
1161 	int i;
1162 
1163 	for_each_possible_cpu(i) {
1164 		u64 rxpackets, rxbytes, txpackets, txbytes;
1165 		unsigned int start;
1166 
1167 		p = per_cpu_ptr(tun->pcpu_stats, i);
1168 		do {
1169 			start = u64_stats_fetch_begin(&p->syncp);
1170 			rxpackets	= p->rx_packets;
1171 			rxbytes		= p->rx_bytes;
1172 			txpackets	= p->tx_packets;
1173 			txbytes		= p->tx_bytes;
1174 		} while (u64_stats_fetch_retry(&p->syncp, start));
1175 
1176 		stats->rx_packets	+= rxpackets;
1177 		stats->rx_bytes		+= rxbytes;
1178 		stats->tx_packets	+= txpackets;
1179 		stats->tx_bytes		+= txbytes;
1180 
1181 		/* u32 counters */
1182 		rx_dropped	+= p->rx_dropped;
1183 		rx_frame_errors	+= p->rx_frame_errors;
1184 		tx_dropped	+= p->tx_dropped;
1185 	}
1186 	stats->rx_dropped  = rx_dropped;
1187 	stats->rx_frame_errors = rx_frame_errors;
1188 	stats->tx_dropped = tx_dropped;
1189 }
1190 
1191 static int tun_xdp_set(struct net_device *dev, struct bpf_prog *prog,
1192 		       struct netlink_ext_ack *extack)
1193 {
1194 	struct tun_struct *tun = netdev_priv(dev);
1195 	struct tun_file *tfile;
1196 	struct bpf_prog *old_prog;
1197 	int i;
1198 
1199 	old_prog = rtnl_dereference(tun->xdp_prog);
1200 	rcu_assign_pointer(tun->xdp_prog, prog);
1201 	if (old_prog)
1202 		bpf_prog_put(old_prog);
1203 
1204 	for (i = 0; i < tun->numqueues; i++) {
1205 		tfile = rtnl_dereference(tun->tfiles[i]);
1206 		if (prog)
1207 			sock_set_flag(&tfile->sk, SOCK_XDP);
1208 		else
1209 			sock_reset_flag(&tfile->sk, SOCK_XDP);
1210 	}
1211 	list_for_each_entry(tfile, &tun->disabled, next) {
1212 		if (prog)
1213 			sock_set_flag(&tfile->sk, SOCK_XDP);
1214 		else
1215 			sock_reset_flag(&tfile->sk, SOCK_XDP);
1216 	}
1217 
1218 	return 0;
1219 }
1220 
1221 static u32 tun_xdp_query(struct net_device *dev)
1222 {
1223 	struct tun_struct *tun = netdev_priv(dev);
1224 	const struct bpf_prog *xdp_prog;
1225 
1226 	xdp_prog = rtnl_dereference(tun->xdp_prog);
1227 	if (xdp_prog)
1228 		return xdp_prog->aux->id;
1229 
1230 	return 0;
1231 }
1232 
1233 static int tun_xdp(struct net_device *dev, struct netdev_bpf *xdp)
1234 {
1235 	switch (xdp->command) {
1236 	case XDP_SETUP_PROG:
1237 		return tun_xdp_set(dev, xdp->prog, xdp->extack);
1238 	case XDP_QUERY_PROG:
1239 		xdp->prog_id = tun_xdp_query(dev);
1240 		return 0;
1241 	default:
1242 		return -EINVAL;
1243 	}
1244 }
1245 
1246 static int tun_net_change_carrier(struct net_device *dev, bool new_carrier)
1247 {
1248 	if (new_carrier) {
1249 		struct tun_struct *tun = netdev_priv(dev);
1250 
1251 		if (!tun->numqueues)
1252 			return -EPERM;
1253 
1254 		netif_carrier_on(dev);
1255 	} else {
1256 		netif_carrier_off(dev);
1257 	}
1258 	return 0;
1259 }
1260 
1261 static const struct net_device_ops tun_netdev_ops = {
1262 	.ndo_uninit		= tun_net_uninit,
1263 	.ndo_open		= tun_net_open,
1264 	.ndo_stop		= tun_net_close,
1265 	.ndo_start_xmit		= tun_net_xmit,
1266 	.ndo_fix_features	= tun_net_fix_features,
1267 	.ndo_select_queue	= tun_select_queue,
1268 	.ndo_set_rx_headroom	= tun_set_headroom,
1269 	.ndo_get_stats64	= tun_net_get_stats64,
1270 	.ndo_change_carrier	= tun_net_change_carrier,
1271 };
1272 
1273 static void __tun_xdp_flush_tfile(struct tun_file *tfile)
1274 {
1275 	/* Notify and wake up reader process */
1276 	if (tfile->flags & TUN_FASYNC)
1277 		kill_fasync(&tfile->fasync, SIGIO, POLL_IN);
1278 	tfile->socket.sk->sk_data_ready(tfile->socket.sk);
1279 }
1280 
1281 static int tun_xdp_xmit(struct net_device *dev, int n,
1282 			struct xdp_frame **frames, u32 flags)
1283 {
1284 	struct tun_struct *tun = netdev_priv(dev);
1285 	struct tun_file *tfile;
1286 	u32 numqueues;
1287 	int drops = 0;
1288 	int cnt = n;
1289 	int i;
1290 
1291 	if (unlikely(flags & ~XDP_XMIT_FLAGS_MASK))
1292 		return -EINVAL;
1293 
1294 	rcu_read_lock();
1295 
1296 resample:
1297 	numqueues = READ_ONCE(tun->numqueues);
1298 	if (!numqueues) {
1299 		rcu_read_unlock();
1300 		return -ENXIO; /* Caller will free/return all frames */
1301 	}
1302 
1303 	tfile = rcu_dereference(tun->tfiles[smp_processor_id() %
1304 					    numqueues]);
1305 	if (unlikely(!tfile))
1306 		goto resample;
1307 
1308 	spin_lock(&tfile->tx_ring.producer_lock);
1309 	for (i = 0; i < n; i++) {
1310 		struct xdp_frame *xdp = frames[i];
1311 		/* Encode the XDP flag into lowest bit for consumer to differ
1312 		 * XDP buffer from sk_buff.
1313 		 */
1314 		void *frame = tun_xdp_to_ptr(xdp);
1315 
1316 		if (__ptr_ring_produce(&tfile->tx_ring, frame)) {
1317 			this_cpu_inc(tun->pcpu_stats->tx_dropped);
1318 			xdp_return_frame_rx_napi(xdp);
1319 			drops++;
1320 		}
1321 	}
1322 	spin_unlock(&tfile->tx_ring.producer_lock);
1323 
1324 	if (flags & XDP_XMIT_FLUSH)
1325 		__tun_xdp_flush_tfile(tfile);
1326 
1327 	rcu_read_unlock();
1328 	return cnt - drops;
1329 }
1330 
1331 static int tun_xdp_tx(struct net_device *dev, struct xdp_buff *xdp)
1332 {
1333 	struct xdp_frame *frame = convert_to_xdp_frame(xdp);
1334 
1335 	if (unlikely(!frame))
1336 		return -EOVERFLOW;
1337 
1338 	return tun_xdp_xmit(dev, 1, &frame, XDP_XMIT_FLUSH);
1339 }
1340 
1341 static const struct net_device_ops tap_netdev_ops = {
1342 	.ndo_uninit		= tun_net_uninit,
1343 	.ndo_open		= tun_net_open,
1344 	.ndo_stop		= tun_net_close,
1345 	.ndo_start_xmit		= tun_net_xmit,
1346 	.ndo_fix_features	= tun_net_fix_features,
1347 	.ndo_set_rx_mode	= tun_net_mclist,
1348 	.ndo_set_mac_address	= eth_mac_addr,
1349 	.ndo_validate_addr	= eth_validate_addr,
1350 	.ndo_select_queue	= tun_select_queue,
1351 	.ndo_features_check	= passthru_features_check,
1352 	.ndo_set_rx_headroom	= tun_set_headroom,
1353 	.ndo_get_stats64	= tun_net_get_stats64,
1354 	.ndo_bpf		= tun_xdp,
1355 	.ndo_xdp_xmit		= tun_xdp_xmit,
1356 	.ndo_change_carrier	= tun_net_change_carrier,
1357 };
1358 
1359 static void tun_flow_init(struct tun_struct *tun)
1360 {
1361 	int i;
1362 
1363 	for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++)
1364 		INIT_HLIST_HEAD(&tun->flows[i]);
1365 
1366 	tun->ageing_time = TUN_FLOW_EXPIRE;
1367 	timer_setup(&tun->flow_gc_timer, tun_flow_cleanup, 0);
1368 	mod_timer(&tun->flow_gc_timer,
1369 		  round_jiffies_up(jiffies + tun->ageing_time));
1370 }
1371 
1372 static void tun_flow_uninit(struct tun_struct *tun)
1373 {
1374 	del_timer_sync(&tun->flow_gc_timer);
1375 	tun_flow_flush(tun);
1376 }
1377 
1378 #define MIN_MTU 68
1379 #define MAX_MTU 65535
1380 
1381 /* Initialize net device. */
1382 static void tun_net_init(struct net_device *dev)
1383 {
1384 	struct tun_struct *tun = netdev_priv(dev);
1385 
1386 	switch (tun->flags & TUN_TYPE_MASK) {
1387 	case IFF_TUN:
1388 		dev->netdev_ops = &tun_netdev_ops;
1389 
1390 		/* Point-to-Point TUN Device */
1391 		dev->hard_header_len = 0;
1392 		dev->addr_len = 0;
1393 		dev->mtu = 1500;
1394 
1395 		/* Zero header length */
1396 		dev->type = ARPHRD_NONE;
1397 		dev->flags = IFF_POINTOPOINT | IFF_NOARP | IFF_MULTICAST;
1398 		break;
1399 
1400 	case IFF_TAP:
1401 		dev->netdev_ops = &tap_netdev_ops;
1402 		/* Ethernet TAP Device */
1403 		ether_setup(dev);
1404 		dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1405 		dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
1406 
1407 		eth_hw_addr_random(dev);
1408 
1409 		break;
1410 	}
1411 
1412 	dev->min_mtu = MIN_MTU;
1413 	dev->max_mtu = MAX_MTU - dev->hard_header_len;
1414 }
1415 
1416 static bool tun_sock_writeable(struct tun_struct *tun, struct tun_file *tfile)
1417 {
1418 	struct sock *sk = tfile->socket.sk;
1419 
1420 	return (tun->dev->flags & IFF_UP) && sock_writeable(sk);
1421 }
1422 
1423 /* Character device part */
1424 
1425 /* Poll */
1426 static __poll_t tun_chr_poll(struct file *file, poll_table *wait)
1427 {
1428 	struct tun_file *tfile = file->private_data;
1429 	struct tun_struct *tun = tun_get(tfile);
1430 	struct sock *sk;
1431 	__poll_t mask = 0;
1432 
1433 	if (!tun)
1434 		return EPOLLERR;
1435 
1436 	sk = tfile->socket.sk;
1437 
1438 	tun_debug(KERN_INFO, tun, "tun_chr_poll\n");
1439 
1440 	poll_wait(file, sk_sleep(sk), wait);
1441 
1442 	if (!ptr_ring_empty(&tfile->tx_ring))
1443 		mask |= EPOLLIN | EPOLLRDNORM;
1444 
1445 	/* Make sure SOCKWQ_ASYNC_NOSPACE is set if not writable to
1446 	 * guarantee EPOLLOUT to be raised by either here or
1447 	 * tun_sock_write_space(). Then process could get notification
1448 	 * after it writes to a down device and meets -EIO.
1449 	 */
1450 	if (tun_sock_writeable(tun, tfile) ||
1451 	    (!test_and_set_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags) &&
1452 	     tun_sock_writeable(tun, tfile)))
1453 		mask |= EPOLLOUT | EPOLLWRNORM;
1454 
1455 	if (tun->dev->reg_state != NETREG_REGISTERED)
1456 		mask = EPOLLERR;
1457 
1458 	tun_put(tun);
1459 	return mask;
1460 }
1461 
1462 static struct sk_buff *tun_napi_alloc_frags(struct tun_file *tfile,
1463 					    size_t len,
1464 					    const struct iov_iter *it)
1465 {
1466 	struct sk_buff *skb;
1467 	size_t linear;
1468 	int err;
1469 	int i;
1470 
1471 	if (it->nr_segs > MAX_SKB_FRAGS + 1)
1472 		return ERR_PTR(-ENOMEM);
1473 
1474 	local_bh_disable();
1475 	skb = napi_get_frags(&tfile->napi);
1476 	local_bh_enable();
1477 	if (!skb)
1478 		return ERR_PTR(-ENOMEM);
1479 
1480 	linear = iov_iter_single_seg_count(it);
1481 	err = __skb_grow(skb, linear);
1482 	if (err)
1483 		goto free;
1484 
1485 	skb->len = len;
1486 	skb->data_len = len - linear;
1487 	skb->truesize += skb->data_len;
1488 
1489 	for (i = 1; i < it->nr_segs; i++) {
1490 		size_t fragsz = it->iov[i].iov_len;
1491 		struct page *page;
1492 		void *frag;
1493 
1494 		if (fragsz == 0 || fragsz > PAGE_SIZE) {
1495 			err = -EINVAL;
1496 			goto free;
1497 		}
1498 		frag = netdev_alloc_frag(fragsz);
1499 		if (!frag) {
1500 			err = -ENOMEM;
1501 			goto free;
1502 		}
1503 		page = virt_to_head_page(frag);
1504 		skb_fill_page_desc(skb, i - 1, page,
1505 				   frag - page_address(page), fragsz);
1506 	}
1507 
1508 	return skb;
1509 free:
1510 	/* frees skb and all frags allocated with napi_alloc_frag() */
1511 	napi_free_frags(&tfile->napi);
1512 	return ERR_PTR(err);
1513 }
1514 
1515 /* prepad is the amount to reserve at front.  len is length after that.
1516  * linear is a hint as to how much to copy (usually headers). */
1517 static struct sk_buff *tun_alloc_skb(struct tun_file *tfile,
1518 				     size_t prepad, size_t len,
1519 				     size_t linear, int noblock)
1520 {
1521 	struct sock *sk = tfile->socket.sk;
1522 	struct sk_buff *skb;
1523 	int err;
1524 
1525 	/* Under a page?  Don't bother with paged skb. */
1526 	if (prepad + len < PAGE_SIZE || !linear)
1527 		linear = len;
1528 
1529 	skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
1530 				   &err, 0);
1531 	if (!skb)
1532 		return ERR_PTR(err);
1533 
1534 	skb_reserve(skb, prepad);
1535 	skb_put(skb, linear);
1536 	skb->data_len = len - linear;
1537 	skb->len += len - linear;
1538 
1539 	return skb;
1540 }
1541 
1542 static void tun_rx_batched(struct tun_struct *tun, struct tun_file *tfile,
1543 			   struct sk_buff *skb, int more)
1544 {
1545 	struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
1546 	struct sk_buff_head process_queue;
1547 	u32 rx_batched = tun->rx_batched;
1548 	bool rcv = false;
1549 
1550 	if (!rx_batched || (!more && skb_queue_empty(queue))) {
1551 		local_bh_disable();
1552 		skb_record_rx_queue(skb, tfile->queue_index);
1553 		netif_receive_skb(skb);
1554 		local_bh_enable();
1555 		return;
1556 	}
1557 
1558 	spin_lock(&queue->lock);
1559 	if (!more || skb_queue_len(queue) == rx_batched) {
1560 		__skb_queue_head_init(&process_queue);
1561 		skb_queue_splice_tail_init(queue, &process_queue);
1562 		rcv = true;
1563 	} else {
1564 		__skb_queue_tail(queue, skb);
1565 	}
1566 	spin_unlock(&queue->lock);
1567 
1568 	if (rcv) {
1569 		struct sk_buff *nskb;
1570 
1571 		local_bh_disable();
1572 		while ((nskb = __skb_dequeue(&process_queue))) {
1573 			skb_record_rx_queue(nskb, tfile->queue_index);
1574 			netif_receive_skb(nskb);
1575 		}
1576 		skb_record_rx_queue(skb, tfile->queue_index);
1577 		netif_receive_skb(skb);
1578 		local_bh_enable();
1579 	}
1580 }
1581 
1582 static bool tun_can_build_skb(struct tun_struct *tun, struct tun_file *tfile,
1583 			      int len, int noblock, bool zerocopy)
1584 {
1585 	if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
1586 		return false;
1587 
1588 	if (tfile->socket.sk->sk_sndbuf != INT_MAX)
1589 		return false;
1590 
1591 	if (!noblock)
1592 		return false;
1593 
1594 	if (zerocopy)
1595 		return false;
1596 
1597 	if (SKB_DATA_ALIGN(len + TUN_RX_PAD) +
1598 	    SKB_DATA_ALIGN(sizeof(struct skb_shared_info)) > PAGE_SIZE)
1599 		return false;
1600 
1601 	return true;
1602 }
1603 
1604 static struct sk_buff *__tun_build_skb(struct tun_file *tfile,
1605 				       struct page_frag *alloc_frag, char *buf,
1606 				       int buflen, int len, int pad)
1607 {
1608 	struct sk_buff *skb = build_skb(buf, buflen);
1609 
1610 	if (!skb)
1611 		return ERR_PTR(-ENOMEM);
1612 
1613 	skb_reserve(skb, pad);
1614 	skb_put(skb, len);
1615 	skb_set_owner_w(skb, tfile->socket.sk);
1616 
1617 	get_page(alloc_frag->page);
1618 	alloc_frag->offset += buflen;
1619 
1620 	return skb;
1621 }
1622 
1623 static int tun_xdp_act(struct tun_struct *tun, struct bpf_prog *xdp_prog,
1624 		       struct xdp_buff *xdp, u32 act)
1625 {
1626 	int err;
1627 
1628 	switch (act) {
1629 	case XDP_REDIRECT:
1630 		err = xdp_do_redirect(tun->dev, xdp, xdp_prog);
1631 		if (err)
1632 			return err;
1633 		break;
1634 	case XDP_TX:
1635 		err = tun_xdp_tx(tun->dev, xdp);
1636 		if (err < 0)
1637 			return err;
1638 		break;
1639 	case XDP_PASS:
1640 		break;
1641 	default:
1642 		bpf_warn_invalid_xdp_action(act);
1643 		/* fall through */
1644 	case XDP_ABORTED:
1645 		trace_xdp_exception(tun->dev, xdp_prog, act);
1646 		/* fall through */
1647 	case XDP_DROP:
1648 		this_cpu_inc(tun->pcpu_stats->rx_dropped);
1649 		break;
1650 	}
1651 
1652 	return act;
1653 }
1654 
1655 static struct sk_buff *tun_build_skb(struct tun_struct *tun,
1656 				     struct tun_file *tfile,
1657 				     struct iov_iter *from,
1658 				     struct virtio_net_hdr *hdr,
1659 				     int len, int *skb_xdp)
1660 {
1661 	struct page_frag *alloc_frag = &current->task_frag;
1662 	struct bpf_prog *xdp_prog;
1663 	int buflen = SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
1664 	char *buf;
1665 	size_t copied;
1666 	int pad = TUN_RX_PAD;
1667 	int err = 0;
1668 
1669 	rcu_read_lock();
1670 	xdp_prog = rcu_dereference(tun->xdp_prog);
1671 	if (xdp_prog)
1672 		pad += XDP_PACKET_HEADROOM;
1673 	buflen += SKB_DATA_ALIGN(len + pad);
1674 	rcu_read_unlock();
1675 
1676 	alloc_frag->offset = ALIGN((u64)alloc_frag->offset, SMP_CACHE_BYTES);
1677 	if (unlikely(!skb_page_frag_refill(buflen, alloc_frag, GFP_KERNEL)))
1678 		return ERR_PTR(-ENOMEM);
1679 
1680 	buf = (char *)page_address(alloc_frag->page) + alloc_frag->offset;
1681 	copied = copy_page_from_iter(alloc_frag->page,
1682 				     alloc_frag->offset + pad,
1683 				     len, from);
1684 	if (copied != len)
1685 		return ERR_PTR(-EFAULT);
1686 
1687 	/* There's a small window that XDP may be set after the check
1688 	 * of xdp_prog above, this should be rare and for simplicity
1689 	 * we do XDP on skb in case the headroom is not enough.
1690 	 */
1691 	if (hdr->gso_type || !xdp_prog) {
1692 		*skb_xdp = 1;
1693 		return __tun_build_skb(tfile, alloc_frag, buf, buflen, len,
1694 				       pad);
1695 	}
1696 
1697 	*skb_xdp = 0;
1698 
1699 	local_bh_disable();
1700 	rcu_read_lock();
1701 	xdp_prog = rcu_dereference(tun->xdp_prog);
1702 	if (xdp_prog) {
1703 		struct xdp_buff xdp;
1704 		u32 act;
1705 
1706 		xdp.data_hard_start = buf;
1707 		xdp.data = buf + pad;
1708 		xdp_set_data_meta_invalid(&xdp);
1709 		xdp.data_end = xdp.data + len;
1710 		xdp.rxq = &tfile->xdp_rxq;
1711 
1712 		act = bpf_prog_run_xdp(xdp_prog, &xdp);
1713 		if (act == XDP_REDIRECT || act == XDP_TX) {
1714 			get_page(alloc_frag->page);
1715 			alloc_frag->offset += buflen;
1716 		}
1717 		err = tun_xdp_act(tun, xdp_prog, &xdp, act);
1718 		if (err < 0)
1719 			goto err_xdp;
1720 		if (err == XDP_REDIRECT)
1721 			xdp_do_flush_map();
1722 		if (err != XDP_PASS)
1723 			goto out;
1724 
1725 		pad = xdp.data - xdp.data_hard_start;
1726 		len = xdp.data_end - xdp.data;
1727 	}
1728 	rcu_read_unlock();
1729 	local_bh_enable();
1730 
1731 	return __tun_build_skb(tfile, alloc_frag, buf, buflen, len, pad);
1732 
1733 err_xdp:
1734 	put_page(alloc_frag->page);
1735 out:
1736 	rcu_read_unlock();
1737 	local_bh_enable();
1738 	return NULL;
1739 }
1740 
1741 /* Get packet from user space buffer */
1742 static ssize_t tun_get_user(struct tun_struct *tun, struct tun_file *tfile,
1743 			    void *msg_control, struct iov_iter *from,
1744 			    int noblock, bool more)
1745 {
1746 	struct tun_pi pi = { 0, cpu_to_be16(ETH_P_IP) };
1747 	struct sk_buff *skb;
1748 	size_t total_len = iov_iter_count(from);
1749 	size_t len = total_len, align = tun->align, linear;
1750 	struct virtio_net_hdr gso = { 0 };
1751 	struct tun_pcpu_stats *stats;
1752 	int good_linear;
1753 	int copylen;
1754 	bool zerocopy = false;
1755 	int err;
1756 	u32 rxhash = 0;
1757 	int skb_xdp = 1;
1758 	bool frags = tun_napi_frags_enabled(tfile);
1759 
1760 	if (!(tun->flags & IFF_NO_PI)) {
1761 		if (len < sizeof(pi))
1762 			return -EINVAL;
1763 		len -= sizeof(pi);
1764 
1765 		if (!copy_from_iter_full(&pi, sizeof(pi), from))
1766 			return -EFAULT;
1767 	}
1768 
1769 	if (tun->flags & IFF_VNET_HDR) {
1770 		int vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
1771 
1772 		if (len < vnet_hdr_sz)
1773 			return -EINVAL;
1774 		len -= vnet_hdr_sz;
1775 
1776 		if (!copy_from_iter_full(&gso, sizeof(gso), from))
1777 			return -EFAULT;
1778 
1779 		if ((gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
1780 		    tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2 > tun16_to_cpu(tun, gso.hdr_len))
1781 			gso.hdr_len = cpu_to_tun16(tun, tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2);
1782 
1783 		if (tun16_to_cpu(tun, gso.hdr_len) > len)
1784 			return -EINVAL;
1785 		iov_iter_advance(from, vnet_hdr_sz - sizeof(gso));
1786 	}
1787 
1788 	if ((tun->flags & TUN_TYPE_MASK) == IFF_TAP) {
1789 		align += NET_IP_ALIGN;
1790 		if (unlikely(len < ETH_HLEN ||
1791 			     (gso.hdr_len && tun16_to_cpu(tun, gso.hdr_len) < ETH_HLEN)))
1792 			return -EINVAL;
1793 	}
1794 
1795 	good_linear = SKB_MAX_HEAD(align);
1796 
1797 	if (msg_control) {
1798 		struct iov_iter i = *from;
1799 
1800 		/* There are 256 bytes to be copied in skb, so there is
1801 		 * enough room for skb expand head in case it is used.
1802 		 * The rest of the buffer is mapped from userspace.
1803 		 */
1804 		copylen = gso.hdr_len ? tun16_to_cpu(tun, gso.hdr_len) : GOODCOPY_LEN;
1805 		if (copylen > good_linear)
1806 			copylen = good_linear;
1807 		linear = copylen;
1808 		iov_iter_advance(&i, copylen);
1809 		if (iov_iter_npages(&i, INT_MAX) <= MAX_SKB_FRAGS)
1810 			zerocopy = true;
1811 	}
1812 
1813 	if (!frags && tun_can_build_skb(tun, tfile, len, noblock, zerocopy)) {
1814 		/* For the packet that is not easy to be processed
1815 		 * (e.g gso or jumbo packet), we will do it at after
1816 		 * skb was created with generic XDP routine.
1817 		 */
1818 		skb = tun_build_skb(tun, tfile, from, &gso, len, &skb_xdp);
1819 		if (IS_ERR(skb)) {
1820 			this_cpu_inc(tun->pcpu_stats->rx_dropped);
1821 			return PTR_ERR(skb);
1822 		}
1823 		if (!skb)
1824 			return total_len;
1825 	} else {
1826 		if (!zerocopy) {
1827 			copylen = len;
1828 			if (tun16_to_cpu(tun, gso.hdr_len) > good_linear)
1829 				linear = good_linear;
1830 			else
1831 				linear = tun16_to_cpu(tun, gso.hdr_len);
1832 		}
1833 
1834 		if (frags) {
1835 			mutex_lock(&tfile->napi_mutex);
1836 			skb = tun_napi_alloc_frags(tfile, copylen, from);
1837 			/* tun_napi_alloc_frags() enforces a layout for the skb.
1838 			 * If zerocopy is enabled, then this layout will be
1839 			 * overwritten by zerocopy_sg_from_iter().
1840 			 */
1841 			zerocopy = false;
1842 		} else {
1843 			skb = tun_alloc_skb(tfile, align, copylen, linear,
1844 					    noblock);
1845 		}
1846 
1847 		if (IS_ERR(skb)) {
1848 			if (PTR_ERR(skb) != -EAGAIN)
1849 				this_cpu_inc(tun->pcpu_stats->rx_dropped);
1850 			if (frags)
1851 				mutex_unlock(&tfile->napi_mutex);
1852 			return PTR_ERR(skb);
1853 		}
1854 
1855 		if (zerocopy)
1856 			err = zerocopy_sg_from_iter(skb, from);
1857 		else
1858 			err = skb_copy_datagram_from_iter(skb, 0, from, len);
1859 
1860 		if (err) {
1861 			err = -EFAULT;
1862 drop:
1863 			this_cpu_inc(tun->pcpu_stats->rx_dropped);
1864 			kfree_skb(skb);
1865 			if (frags) {
1866 				tfile->napi.skb = NULL;
1867 				mutex_unlock(&tfile->napi_mutex);
1868 			}
1869 
1870 			return err;
1871 		}
1872 	}
1873 
1874 	if (virtio_net_hdr_to_skb(skb, &gso, tun_is_little_endian(tun))) {
1875 		this_cpu_inc(tun->pcpu_stats->rx_frame_errors);
1876 		kfree_skb(skb);
1877 		if (frags) {
1878 			tfile->napi.skb = NULL;
1879 			mutex_unlock(&tfile->napi_mutex);
1880 		}
1881 
1882 		return -EINVAL;
1883 	}
1884 
1885 	switch (tun->flags & TUN_TYPE_MASK) {
1886 	case IFF_TUN:
1887 		if (tun->flags & IFF_NO_PI) {
1888 			u8 ip_version = skb->len ? (skb->data[0] >> 4) : 0;
1889 
1890 			switch (ip_version) {
1891 			case 4:
1892 				pi.proto = htons(ETH_P_IP);
1893 				break;
1894 			case 6:
1895 				pi.proto = htons(ETH_P_IPV6);
1896 				break;
1897 			default:
1898 				this_cpu_inc(tun->pcpu_stats->rx_dropped);
1899 				kfree_skb(skb);
1900 				return -EINVAL;
1901 			}
1902 		}
1903 
1904 		skb_reset_mac_header(skb);
1905 		skb->protocol = pi.proto;
1906 		skb->dev = tun->dev;
1907 		break;
1908 	case IFF_TAP:
1909 		if (!frags)
1910 			skb->protocol = eth_type_trans(skb, tun->dev);
1911 		break;
1912 	}
1913 
1914 	/* copy skb_ubuf_info for callback when skb has no error */
1915 	if (zerocopy) {
1916 		skb_shinfo(skb)->destructor_arg = msg_control;
1917 		skb_shinfo(skb)->tx_flags |= SKBTX_DEV_ZEROCOPY;
1918 		skb_shinfo(skb)->tx_flags |= SKBTX_SHARED_FRAG;
1919 	} else if (msg_control) {
1920 		struct ubuf_info *uarg = msg_control;
1921 		uarg->callback(uarg, false);
1922 	}
1923 
1924 	skb_reset_network_header(skb);
1925 	skb_probe_transport_header(skb);
1926 
1927 	if (skb_xdp) {
1928 		struct bpf_prog *xdp_prog;
1929 		int ret;
1930 
1931 		local_bh_disable();
1932 		rcu_read_lock();
1933 		xdp_prog = rcu_dereference(tun->xdp_prog);
1934 		if (xdp_prog) {
1935 			ret = do_xdp_generic(xdp_prog, skb);
1936 			if (ret != XDP_PASS) {
1937 				rcu_read_unlock();
1938 				local_bh_enable();
1939 				return total_len;
1940 			}
1941 		}
1942 		rcu_read_unlock();
1943 		local_bh_enable();
1944 	}
1945 
1946 	/* Compute the costly rx hash only if needed for flow updates.
1947 	 * We may get a very small possibility of OOO during switching, not
1948 	 * worth to optimize.
1949 	 */
1950 	if (!rcu_access_pointer(tun->steering_prog) && tun->numqueues > 1 &&
1951 	    !tfile->detached)
1952 		rxhash = __skb_get_hash_symmetric(skb);
1953 
1954 	rcu_read_lock();
1955 	if (unlikely(!(tun->dev->flags & IFF_UP))) {
1956 		err = -EIO;
1957 		rcu_read_unlock();
1958 		goto drop;
1959 	}
1960 
1961 	if (frags) {
1962 		/* Exercise flow dissector code path. */
1963 		u32 headlen = eth_get_headlen(tun->dev, skb->data,
1964 					      skb_headlen(skb));
1965 
1966 		if (unlikely(headlen > skb_headlen(skb))) {
1967 			this_cpu_inc(tun->pcpu_stats->rx_dropped);
1968 			napi_free_frags(&tfile->napi);
1969 			rcu_read_unlock();
1970 			mutex_unlock(&tfile->napi_mutex);
1971 			WARN_ON(1);
1972 			return -ENOMEM;
1973 		}
1974 
1975 		local_bh_disable();
1976 		napi_gro_frags(&tfile->napi);
1977 		local_bh_enable();
1978 		mutex_unlock(&tfile->napi_mutex);
1979 	} else if (tfile->napi_enabled) {
1980 		struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
1981 		int queue_len;
1982 
1983 		spin_lock_bh(&queue->lock);
1984 		__skb_queue_tail(queue, skb);
1985 		queue_len = skb_queue_len(queue);
1986 		spin_unlock(&queue->lock);
1987 
1988 		if (!more || queue_len > NAPI_POLL_WEIGHT)
1989 			napi_schedule(&tfile->napi);
1990 
1991 		local_bh_enable();
1992 	} else if (!IS_ENABLED(CONFIG_4KSTACKS)) {
1993 		tun_rx_batched(tun, tfile, skb, more);
1994 	} else {
1995 		netif_rx_ni(skb);
1996 	}
1997 	rcu_read_unlock();
1998 
1999 	stats = get_cpu_ptr(tun->pcpu_stats);
2000 	u64_stats_update_begin(&stats->syncp);
2001 	stats->rx_packets++;
2002 	stats->rx_bytes += len;
2003 	u64_stats_update_end(&stats->syncp);
2004 	put_cpu_ptr(stats);
2005 
2006 	if (rxhash)
2007 		tun_flow_update(tun, rxhash, tfile);
2008 
2009 	return total_len;
2010 }
2011 
2012 static ssize_t tun_chr_write_iter(struct kiocb *iocb, struct iov_iter *from)
2013 {
2014 	struct file *file = iocb->ki_filp;
2015 	struct tun_file *tfile = file->private_data;
2016 	struct tun_struct *tun = tun_get(tfile);
2017 	ssize_t result;
2018 
2019 	if (!tun)
2020 		return -EBADFD;
2021 
2022 	result = tun_get_user(tun, tfile, NULL, from,
2023 			      file->f_flags & O_NONBLOCK, false);
2024 
2025 	tun_put(tun);
2026 	return result;
2027 }
2028 
2029 static ssize_t tun_put_user_xdp(struct tun_struct *tun,
2030 				struct tun_file *tfile,
2031 				struct xdp_frame *xdp_frame,
2032 				struct iov_iter *iter)
2033 {
2034 	int vnet_hdr_sz = 0;
2035 	size_t size = xdp_frame->len;
2036 	struct tun_pcpu_stats *stats;
2037 	size_t ret;
2038 
2039 	if (tun->flags & IFF_VNET_HDR) {
2040 		struct virtio_net_hdr gso = { 0 };
2041 
2042 		vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
2043 		if (unlikely(iov_iter_count(iter) < vnet_hdr_sz))
2044 			return -EINVAL;
2045 		if (unlikely(copy_to_iter(&gso, sizeof(gso), iter) !=
2046 			     sizeof(gso)))
2047 			return -EFAULT;
2048 		iov_iter_advance(iter, vnet_hdr_sz - sizeof(gso));
2049 	}
2050 
2051 	ret = copy_to_iter(xdp_frame->data, size, iter) + vnet_hdr_sz;
2052 
2053 	stats = get_cpu_ptr(tun->pcpu_stats);
2054 	u64_stats_update_begin(&stats->syncp);
2055 	stats->tx_packets++;
2056 	stats->tx_bytes += ret;
2057 	u64_stats_update_end(&stats->syncp);
2058 	put_cpu_ptr(tun->pcpu_stats);
2059 
2060 	return ret;
2061 }
2062 
2063 /* Put packet to the user space buffer */
2064 static ssize_t tun_put_user(struct tun_struct *tun,
2065 			    struct tun_file *tfile,
2066 			    struct sk_buff *skb,
2067 			    struct iov_iter *iter)
2068 {
2069 	struct tun_pi pi = { 0, skb->protocol };
2070 	struct tun_pcpu_stats *stats;
2071 	ssize_t total;
2072 	int vlan_offset = 0;
2073 	int vlan_hlen = 0;
2074 	int vnet_hdr_sz = 0;
2075 
2076 	if (skb_vlan_tag_present(skb))
2077 		vlan_hlen = VLAN_HLEN;
2078 
2079 	if (tun->flags & IFF_VNET_HDR)
2080 		vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
2081 
2082 	total = skb->len + vlan_hlen + vnet_hdr_sz;
2083 
2084 	if (!(tun->flags & IFF_NO_PI)) {
2085 		if (iov_iter_count(iter) < sizeof(pi))
2086 			return -EINVAL;
2087 
2088 		total += sizeof(pi);
2089 		if (iov_iter_count(iter) < total) {
2090 			/* Packet will be striped */
2091 			pi.flags |= TUN_PKT_STRIP;
2092 		}
2093 
2094 		if (copy_to_iter(&pi, sizeof(pi), iter) != sizeof(pi))
2095 			return -EFAULT;
2096 	}
2097 
2098 	if (vnet_hdr_sz) {
2099 		struct virtio_net_hdr gso;
2100 
2101 		if (iov_iter_count(iter) < vnet_hdr_sz)
2102 			return -EINVAL;
2103 
2104 		if (virtio_net_hdr_from_skb(skb, &gso,
2105 					    tun_is_little_endian(tun), true,
2106 					    vlan_hlen)) {
2107 			struct skb_shared_info *sinfo = skb_shinfo(skb);
2108 			pr_err("unexpected GSO type: "
2109 			       "0x%x, gso_size %d, hdr_len %d\n",
2110 			       sinfo->gso_type, tun16_to_cpu(tun, gso.gso_size),
2111 			       tun16_to_cpu(tun, gso.hdr_len));
2112 			print_hex_dump(KERN_ERR, "tun: ",
2113 				       DUMP_PREFIX_NONE,
2114 				       16, 1, skb->head,
2115 				       min((int)tun16_to_cpu(tun, gso.hdr_len), 64), true);
2116 			WARN_ON_ONCE(1);
2117 			return -EINVAL;
2118 		}
2119 
2120 		if (copy_to_iter(&gso, sizeof(gso), iter) != sizeof(gso))
2121 			return -EFAULT;
2122 
2123 		iov_iter_advance(iter, vnet_hdr_sz - sizeof(gso));
2124 	}
2125 
2126 	if (vlan_hlen) {
2127 		int ret;
2128 		struct veth veth;
2129 
2130 		veth.h_vlan_proto = skb->vlan_proto;
2131 		veth.h_vlan_TCI = htons(skb_vlan_tag_get(skb));
2132 
2133 		vlan_offset = offsetof(struct vlan_ethhdr, h_vlan_proto);
2134 
2135 		ret = skb_copy_datagram_iter(skb, 0, iter, vlan_offset);
2136 		if (ret || !iov_iter_count(iter))
2137 			goto done;
2138 
2139 		ret = copy_to_iter(&veth, sizeof(veth), iter);
2140 		if (ret != sizeof(veth) || !iov_iter_count(iter))
2141 			goto done;
2142 	}
2143 
2144 	skb_copy_datagram_iter(skb, vlan_offset, iter, skb->len - vlan_offset);
2145 
2146 done:
2147 	/* caller is in process context, */
2148 	stats = get_cpu_ptr(tun->pcpu_stats);
2149 	u64_stats_update_begin(&stats->syncp);
2150 	stats->tx_packets++;
2151 	stats->tx_bytes += skb->len + vlan_hlen;
2152 	u64_stats_update_end(&stats->syncp);
2153 	put_cpu_ptr(tun->pcpu_stats);
2154 
2155 	return total;
2156 }
2157 
2158 static void *tun_ring_recv(struct tun_file *tfile, int noblock, int *err)
2159 {
2160 	DECLARE_WAITQUEUE(wait, current);
2161 	void *ptr = NULL;
2162 	int error = 0;
2163 
2164 	ptr = ptr_ring_consume(&tfile->tx_ring);
2165 	if (ptr)
2166 		goto out;
2167 	if (noblock) {
2168 		error = -EAGAIN;
2169 		goto out;
2170 	}
2171 
2172 	add_wait_queue(&tfile->socket.wq.wait, &wait);
2173 
2174 	while (1) {
2175 		set_current_state(TASK_INTERRUPTIBLE);
2176 		ptr = ptr_ring_consume(&tfile->tx_ring);
2177 		if (ptr)
2178 			break;
2179 		if (signal_pending(current)) {
2180 			error = -ERESTARTSYS;
2181 			break;
2182 		}
2183 		if (tfile->socket.sk->sk_shutdown & RCV_SHUTDOWN) {
2184 			error = -EFAULT;
2185 			break;
2186 		}
2187 
2188 		schedule();
2189 	}
2190 
2191 	__set_current_state(TASK_RUNNING);
2192 	remove_wait_queue(&tfile->socket.wq.wait, &wait);
2193 
2194 out:
2195 	*err = error;
2196 	return ptr;
2197 }
2198 
2199 static ssize_t tun_do_read(struct tun_struct *tun, struct tun_file *tfile,
2200 			   struct iov_iter *to,
2201 			   int noblock, void *ptr)
2202 {
2203 	ssize_t ret;
2204 	int err;
2205 
2206 	tun_debug(KERN_INFO, tun, "tun_do_read\n");
2207 
2208 	if (!iov_iter_count(to)) {
2209 		tun_ptr_free(ptr);
2210 		return 0;
2211 	}
2212 
2213 	if (!ptr) {
2214 		/* Read frames from ring */
2215 		ptr = tun_ring_recv(tfile, noblock, &err);
2216 		if (!ptr)
2217 			return err;
2218 	}
2219 
2220 	if (tun_is_xdp_frame(ptr)) {
2221 		struct xdp_frame *xdpf = tun_ptr_to_xdp(ptr);
2222 
2223 		ret = tun_put_user_xdp(tun, tfile, xdpf, to);
2224 		xdp_return_frame(xdpf);
2225 	} else {
2226 		struct sk_buff *skb = ptr;
2227 
2228 		ret = tun_put_user(tun, tfile, skb, to);
2229 		if (unlikely(ret < 0))
2230 			kfree_skb(skb);
2231 		else
2232 			consume_skb(skb);
2233 	}
2234 
2235 	return ret;
2236 }
2237 
2238 static ssize_t tun_chr_read_iter(struct kiocb *iocb, struct iov_iter *to)
2239 {
2240 	struct file *file = iocb->ki_filp;
2241 	struct tun_file *tfile = file->private_data;
2242 	struct tun_struct *tun = tun_get(tfile);
2243 	ssize_t len = iov_iter_count(to), ret;
2244 
2245 	if (!tun)
2246 		return -EBADFD;
2247 	ret = tun_do_read(tun, tfile, to, file->f_flags & O_NONBLOCK, NULL);
2248 	ret = min_t(ssize_t, ret, len);
2249 	if (ret > 0)
2250 		iocb->ki_pos = ret;
2251 	tun_put(tun);
2252 	return ret;
2253 }
2254 
2255 static void tun_prog_free(struct rcu_head *rcu)
2256 {
2257 	struct tun_prog *prog = container_of(rcu, struct tun_prog, rcu);
2258 
2259 	bpf_prog_destroy(prog->prog);
2260 	kfree(prog);
2261 }
2262 
2263 static int __tun_set_ebpf(struct tun_struct *tun,
2264 			  struct tun_prog __rcu **prog_p,
2265 			  struct bpf_prog *prog)
2266 {
2267 	struct tun_prog *old, *new = NULL;
2268 
2269 	if (prog) {
2270 		new = kmalloc(sizeof(*new), GFP_KERNEL);
2271 		if (!new)
2272 			return -ENOMEM;
2273 		new->prog = prog;
2274 	}
2275 
2276 	spin_lock_bh(&tun->lock);
2277 	old = rcu_dereference_protected(*prog_p,
2278 					lockdep_is_held(&tun->lock));
2279 	rcu_assign_pointer(*prog_p, new);
2280 	spin_unlock_bh(&tun->lock);
2281 
2282 	if (old)
2283 		call_rcu(&old->rcu, tun_prog_free);
2284 
2285 	return 0;
2286 }
2287 
2288 static void tun_free_netdev(struct net_device *dev)
2289 {
2290 	struct tun_struct *tun = netdev_priv(dev);
2291 
2292 	BUG_ON(!(list_empty(&tun->disabled)));
2293 	free_percpu(tun->pcpu_stats);
2294 	tun_flow_uninit(tun);
2295 	security_tun_dev_free_security(tun->security);
2296 	__tun_set_ebpf(tun, &tun->steering_prog, NULL);
2297 	__tun_set_ebpf(tun, &tun->filter_prog, NULL);
2298 }
2299 
2300 static void tun_setup(struct net_device *dev)
2301 {
2302 	struct tun_struct *tun = netdev_priv(dev);
2303 
2304 	tun->owner = INVALID_UID;
2305 	tun->group = INVALID_GID;
2306 	tun_default_link_ksettings(dev, &tun->link_ksettings);
2307 
2308 	dev->ethtool_ops = &tun_ethtool_ops;
2309 	dev->needs_free_netdev = true;
2310 	dev->priv_destructor = tun_free_netdev;
2311 	/* We prefer our own queue length */
2312 	dev->tx_queue_len = TUN_READQ_SIZE;
2313 }
2314 
2315 /* Trivial set of netlink ops to allow deleting tun or tap
2316  * device with netlink.
2317  */
2318 static int tun_validate(struct nlattr *tb[], struct nlattr *data[],
2319 			struct netlink_ext_ack *extack)
2320 {
2321 	NL_SET_ERR_MSG(extack,
2322 		       "tun/tap creation via rtnetlink is not supported.");
2323 	return -EOPNOTSUPP;
2324 }
2325 
2326 static size_t tun_get_size(const struct net_device *dev)
2327 {
2328 	BUILD_BUG_ON(sizeof(u32) != sizeof(uid_t));
2329 	BUILD_BUG_ON(sizeof(u32) != sizeof(gid_t));
2330 
2331 	return nla_total_size(sizeof(uid_t)) + /* OWNER */
2332 	       nla_total_size(sizeof(gid_t)) + /* GROUP */
2333 	       nla_total_size(sizeof(u8)) + /* TYPE */
2334 	       nla_total_size(sizeof(u8)) + /* PI */
2335 	       nla_total_size(sizeof(u8)) + /* VNET_HDR */
2336 	       nla_total_size(sizeof(u8)) + /* PERSIST */
2337 	       nla_total_size(sizeof(u8)) + /* MULTI_QUEUE */
2338 	       nla_total_size(sizeof(u32)) + /* NUM_QUEUES */
2339 	       nla_total_size(sizeof(u32)) + /* NUM_DISABLED_QUEUES */
2340 	       0;
2341 }
2342 
2343 static int tun_fill_info(struct sk_buff *skb, const struct net_device *dev)
2344 {
2345 	struct tun_struct *tun = netdev_priv(dev);
2346 
2347 	if (nla_put_u8(skb, IFLA_TUN_TYPE, tun->flags & TUN_TYPE_MASK))
2348 		goto nla_put_failure;
2349 	if (uid_valid(tun->owner) &&
2350 	    nla_put_u32(skb, IFLA_TUN_OWNER,
2351 			from_kuid_munged(current_user_ns(), tun->owner)))
2352 		goto nla_put_failure;
2353 	if (gid_valid(tun->group) &&
2354 	    nla_put_u32(skb, IFLA_TUN_GROUP,
2355 			from_kgid_munged(current_user_ns(), tun->group)))
2356 		goto nla_put_failure;
2357 	if (nla_put_u8(skb, IFLA_TUN_PI, !(tun->flags & IFF_NO_PI)))
2358 		goto nla_put_failure;
2359 	if (nla_put_u8(skb, IFLA_TUN_VNET_HDR, !!(tun->flags & IFF_VNET_HDR)))
2360 		goto nla_put_failure;
2361 	if (nla_put_u8(skb, IFLA_TUN_PERSIST, !!(tun->flags & IFF_PERSIST)))
2362 		goto nla_put_failure;
2363 	if (nla_put_u8(skb, IFLA_TUN_MULTI_QUEUE,
2364 		       !!(tun->flags & IFF_MULTI_QUEUE)))
2365 		goto nla_put_failure;
2366 	if (tun->flags & IFF_MULTI_QUEUE) {
2367 		if (nla_put_u32(skb, IFLA_TUN_NUM_QUEUES, tun->numqueues))
2368 			goto nla_put_failure;
2369 		if (nla_put_u32(skb, IFLA_TUN_NUM_DISABLED_QUEUES,
2370 				tun->numdisabled))
2371 			goto nla_put_failure;
2372 	}
2373 
2374 	return 0;
2375 
2376 nla_put_failure:
2377 	return -EMSGSIZE;
2378 }
2379 
2380 static struct rtnl_link_ops tun_link_ops __read_mostly = {
2381 	.kind		= DRV_NAME,
2382 	.priv_size	= sizeof(struct tun_struct),
2383 	.setup		= tun_setup,
2384 	.validate	= tun_validate,
2385 	.get_size       = tun_get_size,
2386 	.fill_info      = tun_fill_info,
2387 };
2388 
2389 static void tun_sock_write_space(struct sock *sk)
2390 {
2391 	struct tun_file *tfile;
2392 	wait_queue_head_t *wqueue;
2393 
2394 	if (!sock_writeable(sk))
2395 		return;
2396 
2397 	if (!test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags))
2398 		return;
2399 
2400 	wqueue = sk_sleep(sk);
2401 	if (wqueue && waitqueue_active(wqueue))
2402 		wake_up_interruptible_sync_poll(wqueue, EPOLLOUT |
2403 						EPOLLWRNORM | EPOLLWRBAND);
2404 
2405 	tfile = container_of(sk, struct tun_file, sk);
2406 	kill_fasync(&tfile->fasync, SIGIO, POLL_OUT);
2407 }
2408 
2409 static void tun_put_page(struct tun_page *tpage)
2410 {
2411 	if (tpage->page)
2412 		__page_frag_cache_drain(tpage->page, tpage->count);
2413 }
2414 
2415 static int tun_xdp_one(struct tun_struct *tun,
2416 		       struct tun_file *tfile,
2417 		       struct xdp_buff *xdp, int *flush,
2418 		       struct tun_page *tpage)
2419 {
2420 	unsigned int datasize = xdp->data_end - xdp->data;
2421 	struct tun_xdp_hdr *hdr = xdp->data_hard_start;
2422 	struct virtio_net_hdr *gso = &hdr->gso;
2423 	struct tun_pcpu_stats *stats;
2424 	struct bpf_prog *xdp_prog;
2425 	struct sk_buff *skb = NULL;
2426 	u32 rxhash = 0, act;
2427 	int buflen = hdr->buflen;
2428 	int err = 0;
2429 	bool skb_xdp = false;
2430 	struct page *page;
2431 
2432 	xdp_prog = rcu_dereference(tun->xdp_prog);
2433 	if (xdp_prog) {
2434 		if (gso->gso_type) {
2435 			skb_xdp = true;
2436 			goto build;
2437 		}
2438 		xdp_set_data_meta_invalid(xdp);
2439 		xdp->rxq = &tfile->xdp_rxq;
2440 
2441 		act = bpf_prog_run_xdp(xdp_prog, xdp);
2442 		err = tun_xdp_act(tun, xdp_prog, xdp, act);
2443 		if (err < 0) {
2444 			put_page(virt_to_head_page(xdp->data));
2445 			return err;
2446 		}
2447 
2448 		switch (err) {
2449 		case XDP_REDIRECT:
2450 			*flush = true;
2451 			/* fall through */
2452 		case XDP_TX:
2453 			return 0;
2454 		case XDP_PASS:
2455 			break;
2456 		default:
2457 			page = virt_to_head_page(xdp->data);
2458 			if (tpage->page == page) {
2459 				++tpage->count;
2460 			} else {
2461 				tun_put_page(tpage);
2462 				tpage->page = page;
2463 				tpage->count = 1;
2464 			}
2465 			return 0;
2466 		}
2467 	}
2468 
2469 build:
2470 	skb = build_skb(xdp->data_hard_start, buflen);
2471 	if (!skb) {
2472 		err = -ENOMEM;
2473 		goto out;
2474 	}
2475 
2476 	skb_reserve(skb, xdp->data - xdp->data_hard_start);
2477 	skb_put(skb, xdp->data_end - xdp->data);
2478 
2479 	if (virtio_net_hdr_to_skb(skb, gso, tun_is_little_endian(tun))) {
2480 		this_cpu_inc(tun->pcpu_stats->rx_frame_errors);
2481 		kfree_skb(skb);
2482 		err = -EINVAL;
2483 		goto out;
2484 	}
2485 
2486 	skb->protocol = eth_type_trans(skb, tun->dev);
2487 	skb_reset_network_header(skb);
2488 	skb_probe_transport_header(skb);
2489 
2490 	if (skb_xdp) {
2491 		err = do_xdp_generic(xdp_prog, skb);
2492 		if (err != XDP_PASS)
2493 			goto out;
2494 	}
2495 
2496 	if (!rcu_dereference(tun->steering_prog) && tun->numqueues > 1 &&
2497 	    !tfile->detached)
2498 		rxhash = __skb_get_hash_symmetric(skb);
2499 
2500 	skb_record_rx_queue(skb, tfile->queue_index);
2501 	netif_receive_skb(skb);
2502 
2503 	/* No need for get_cpu_ptr() here since this function is
2504 	 * always called with bh disabled
2505 	 */
2506 	stats = this_cpu_ptr(tun->pcpu_stats);
2507 	u64_stats_update_begin(&stats->syncp);
2508 	stats->rx_packets++;
2509 	stats->rx_bytes += datasize;
2510 	u64_stats_update_end(&stats->syncp);
2511 
2512 	if (rxhash)
2513 		tun_flow_update(tun, rxhash, tfile);
2514 
2515 out:
2516 	return err;
2517 }
2518 
2519 static int tun_sendmsg(struct socket *sock, struct msghdr *m, size_t total_len)
2520 {
2521 	int ret, i;
2522 	struct tun_file *tfile = container_of(sock, struct tun_file, socket);
2523 	struct tun_struct *tun = tun_get(tfile);
2524 	struct tun_msg_ctl *ctl = m->msg_control;
2525 	struct xdp_buff *xdp;
2526 
2527 	if (!tun)
2528 		return -EBADFD;
2529 
2530 	if (ctl && (ctl->type == TUN_MSG_PTR)) {
2531 		struct tun_page tpage;
2532 		int n = ctl->num;
2533 		int flush = 0;
2534 
2535 		memset(&tpage, 0, sizeof(tpage));
2536 
2537 		local_bh_disable();
2538 		rcu_read_lock();
2539 
2540 		for (i = 0; i < n; i++) {
2541 			xdp = &((struct xdp_buff *)ctl->ptr)[i];
2542 			tun_xdp_one(tun, tfile, xdp, &flush, &tpage);
2543 		}
2544 
2545 		if (flush)
2546 			xdp_do_flush_map();
2547 
2548 		rcu_read_unlock();
2549 		local_bh_enable();
2550 
2551 		tun_put_page(&tpage);
2552 
2553 		ret = total_len;
2554 		goto out;
2555 	}
2556 
2557 	ret = tun_get_user(tun, tfile, ctl ? ctl->ptr : NULL, &m->msg_iter,
2558 			   m->msg_flags & MSG_DONTWAIT,
2559 			   m->msg_flags & MSG_MORE);
2560 out:
2561 	tun_put(tun);
2562 	return ret;
2563 }
2564 
2565 static int tun_recvmsg(struct socket *sock, struct msghdr *m, size_t total_len,
2566 		       int flags)
2567 {
2568 	struct tun_file *tfile = container_of(sock, struct tun_file, socket);
2569 	struct tun_struct *tun = tun_get(tfile);
2570 	void *ptr = m->msg_control;
2571 	int ret;
2572 
2573 	if (!tun) {
2574 		ret = -EBADFD;
2575 		goto out_free;
2576 	}
2577 
2578 	if (flags & ~(MSG_DONTWAIT|MSG_TRUNC|MSG_ERRQUEUE)) {
2579 		ret = -EINVAL;
2580 		goto out_put_tun;
2581 	}
2582 	if (flags & MSG_ERRQUEUE) {
2583 		ret = sock_recv_errqueue(sock->sk, m, total_len,
2584 					 SOL_PACKET, TUN_TX_TIMESTAMP);
2585 		goto out;
2586 	}
2587 	ret = tun_do_read(tun, tfile, &m->msg_iter, flags & MSG_DONTWAIT, ptr);
2588 	if (ret > (ssize_t)total_len) {
2589 		m->msg_flags |= MSG_TRUNC;
2590 		ret = flags & MSG_TRUNC ? ret : total_len;
2591 	}
2592 out:
2593 	tun_put(tun);
2594 	return ret;
2595 
2596 out_put_tun:
2597 	tun_put(tun);
2598 out_free:
2599 	tun_ptr_free(ptr);
2600 	return ret;
2601 }
2602 
2603 static int tun_ptr_peek_len(void *ptr)
2604 {
2605 	if (likely(ptr)) {
2606 		if (tun_is_xdp_frame(ptr)) {
2607 			struct xdp_frame *xdpf = tun_ptr_to_xdp(ptr);
2608 
2609 			return xdpf->len;
2610 		}
2611 		return __skb_array_len_with_tag(ptr);
2612 	} else {
2613 		return 0;
2614 	}
2615 }
2616 
2617 static int tun_peek_len(struct socket *sock)
2618 {
2619 	struct tun_file *tfile = container_of(sock, struct tun_file, socket);
2620 	struct tun_struct *tun;
2621 	int ret = 0;
2622 
2623 	tun = tun_get(tfile);
2624 	if (!tun)
2625 		return 0;
2626 
2627 	ret = PTR_RING_PEEK_CALL(&tfile->tx_ring, tun_ptr_peek_len);
2628 	tun_put(tun);
2629 
2630 	return ret;
2631 }
2632 
2633 /* Ops structure to mimic raw sockets with tun */
2634 static const struct proto_ops tun_socket_ops = {
2635 	.peek_len = tun_peek_len,
2636 	.sendmsg = tun_sendmsg,
2637 	.recvmsg = tun_recvmsg,
2638 };
2639 
2640 static struct proto tun_proto = {
2641 	.name		= "tun",
2642 	.owner		= THIS_MODULE,
2643 	.obj_size	= sizeof(struct tun_file),
2644 };
2645 
2646 static int tun_flags(struct tun_struct *tun)
2647 {
2648 	return tun->flags & (TUN_FEATURES | IFF_PERSIST | IFF_TUN | IFF_TAP);
2649 }
2650 
2651 static ssize_t tun_show_flags(struct device *dev, struct device_attribute *attr,
2652 			      char *buf)
2653 {
2654 	struct tun_struct *tun = netdev_priv(to_net_dev(dev));
2655 	return sprintf(buf, "0x%x\n", tun_flags(tun));
2656 }
2657 
2658 static ssize_t tun_show_owner(struct device *dev, struct device_attribute *attr,
2659 			      char *buf)
2660 {
2661 	struct tun_struct *tun = netdev_priv(to_net_dev(dev));
2662 	return uid_valid(tun->owner)?
2663 		sprintf(buf, "%u\n",
2664 			from_kuid_munged(current_user_ns(), tun->owner)):
2665 		sprintf(buf, "-1\n");
2666 }
2667 
2668 static ssize_t tun_show_group(struct device *dev, struct device_attribute *attr,
2669 			      char *buf)
2670 {
2671 	struct tun_struct *tun = netdev_priv(to_net_dev(dev));
2672 	return gid_valid(tun->group) ?
2673 		sprintf(buf, "%u\n",
2674 			from_kgid_munged(current_user_ns(), tun->group)):
2675 		sprintf(buf, "-1\n");
2676 }
2677 
2678 static DEVICE_ATTR(tun_flags, 0444, tun_show_flags, NULL);
2679 static DEVICE_ATTR(owner, 0444, tun_show_owner, NULL);
2680 static DEVICE_ATTR(group, 0444, tun_show_group, NULL);
2681 
2682 static struct attribute *tun_dev_attrs[] = {
2683 	&dev_attr_tun_flags.attr,
2684 	&dev_attr_owner.attr,
2685 	&dev_attr_group.attr,
2686 	NULL
2687 };
2688 
2689 static const struct attribute_group tun_attr_group = {
2690 	.attrs = tun_dev_attrs
2691 };
2692 
2693 static int tun_set_iff(struct net *net, struct file *file, struct ifreq *ifr)
2694 {
2695 	struct tun_struct *tun;
2696 	struct tun_file *tfile = file->private_data;
2697 	struct net_device *dev;
2698 	int err;
2699 
2700 	if (tfile->detached)
2701 		return -EINVAL;
2702 
2703 	if ((ifr->ifr_flags & IFF_NAPI_FRAGS)) {
2704 		if (!capable(CAP_NET_ADMIN))
2705 			return -EPERM;
2706 
2707 		if (!(ifr->ifr_flags & IFF_NAPI) ||
2708 		    (ifr->ifr_flags & TUN_TYPE_MASK) != IFF_TAP)
2709 			return -EINVAL;
2710 	}
2711 
2712 	dev = __dev_get_by_name(net, ifr->ifr_name);
2713 	if (dev) {
2714 		if (ifr->ifr_flags & IFF_TUN_EXCL)
2715 			return -EBUSY;
2716 		if ((ifr->ifr_flags & IFF_TUN) && dev->netdev_ops == &tun_netdev_ops)
2717 			tun = netdev_priv(dev);
2718 		else if ((ifr->ifr_flags & IFF_TAP) && dev->netdev_ops == &tap_netdev_ops)
2719 			tun = netdev_priv(dev);
2720 		else
2721 			return -EINVAL;
2722 
2723 		if (!!(ifr->ifr_flags & IFF_MULTI_QUEUE) !=
2724 		    !!(tun->flags & IFF_MULTI_QUEUE))
2725 			return -EINVAL;
2726 
2727 		if (tun_not_capable(tun))
2728 			return -EPERM;
2729 		err = security_tun_dev_open(tun->security);
2730 		if (err < 0)
2731 			return err;
2732 
2733 		err = tun_attach(tun, file, ifr->ifr_flags & IFF_NOFILTER,
2734 				 ifr->ifr_flags & IFF_NAPI,
2735 				 ifr->ifr_flags & IFF_NAPI_FRAGS, true);
2736 		if (err < 0)
2737 			return err;
2738 
2739 		if (tun->flags & IFF_MULTI_QUEUE &&
2740 		    (tun->numqueues + tun->numdisabled > 1)) {
2741 			/* One or more queue has already been attached, no need
2742 			 * to initialize the device again.
2743 			 */
2744 			netdev_state_change(dev);
2745 			return 0;
2746 		}
2747 
2748 		tun->flags = (tun->flags & ~TUN_FEATURES) |
2749 			      (ifr->ifr_flags & TUN_FEATURES);
2750 
2751 		netdev_state_change(dev);
2752 	} else {
2753 		char *name;
2754 		unsigned long flags = 0;
2755 		int queues = ifr->ifr_flags & IFF_MULTI_QUEUE ?
2756 			     MAX_TAP_QUEUES : 1;
2757 
2758 		if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
2759 			return -EPERM;
2760 		err = security_tun_dev_create();
2761 		if (err < 0)
2762 			return err;
2763 
2764 		/* Set dev type */
2765 		if (ifr->ifr_flags & IFF_TUN) {
2766 			/* TUN device */
2767 			flags |= IFF_TUN;
2768 			name = "tun%d";
2769 		} else if (ifr->ifr_flags & IFF_TAP) {
2770 			/* TAP device */
2771 			flags |= IFF_TAP;
2772 			name = "tap%d";
2773 		} else
2774 			return -EINVAL;
2775 
2776 		if (*ifr->ifr_name)
2777 			name = ifr->ifr_name;
2778 
2779 		dev = alloc_netdev_mqs(sizeof(struct tun_struct), name,
2780 				       NET_NAME_UNKNOWN, tun_setup, queues,
2781 				       queues);
2782 
2783 		if (!dev)
2784 			return -ENOMEM;
2785 		err = dev_get_valid_name(net, dev, name);
2786 		if (err < 0)
2787 			goto err_free_dev;
2788 
2789 		dev_net_set(dev, net);
2790 		dev->rtnl_link_ops = &tun_link_ops;
2791 		dev->ifindex = tfile->ifindex;
2792 		dev->sysfs_groups[0] = &tun_attr_group;
2793 
2794 		tun = netdev_priv(dev);
2795 		tun->dev = dev;
2796 		tun->flags = flags;
2797 		tun->txflt.count = 0;
2798 		tun->vnet_hdr_sz = sizeof(struct virtio_net_hdr);
2799 
2800 		tun->align = NET_SKB_PAD;
2801 		tun->filter_attached = false;
2802 		tun->sndbuf = tfile->socket.sk->sk_sndbuf;
2803 		tun->rx_batched = 0;
2804 		RCU_INIT_POINTER(tun->steering_prog, NULL);
2805 
2806 		tun->pcpu_stats = netdev_alloc_pcpu_stats(struct tun_pcpu_stats);
2807 		if (!tun->pcpu_stats) {
2808 			err = -ENOMEM;
2809 			goto err_free_dev;
2810 		}
2811 
2812 		spin_lock_init(&tun->lock);
2813 
2814 		err = security_tun_dev_alloc_security(&tun->security);
2815 		if (err < 0)
2816 			goto err_free_stat;
2817 
2818 		tun_net_init(dev);
2819 		tun_flow_init(tun);
2820 
2821 		dev->hw_features = NETIF_F_SG | NETIF_F_FRAGLIST |
2822 				   TUN_USER_FEATURES | NETIF_F_HW_VLAN_CTAG_TX |
2823 				   NETIF_F_HW_VLAN_STAG_TX;
2824 		dev->features = dev->hw_features | NETIF_F_LLTX;
2825 		dev->vlan_features = dev->features &
2826 				     ~(NETIF_F_HW_VLAN_CTAG_TX |
2827 				       NETIF_F_HW_VLAN_STAG_TX);
2828 
2829 		tun->flags = (tun->flags & ~TUN_FEATURES) |
2830 			      (ifr->ifr_flags & TUN_FEATURES);
2831 
2832 		INIT_LIST_HEAD(&tun->disabled);
2833 		err = tun_attach(tun, file, false, ifr->ifr_flags & IFF_NAPI,
2834 				 ifr->ifr_flags & IFF_NAPI_FRAGS, false);
2835 		if (err < 0)
2836 			goto err_free_flow;
2837 
2838 		err = register_netdevice(tun->dev);
2839 		if (err < 0)
2840 			goto err_detach;
2841 		/* free_netdev() won't check refcnt, to aovid race
2842 		 * with dev_put() we need publish tun after registration.
2843 		 */
2844 		rcu_assign_pointer(tfile->tun, tun);
2845 	}
2846 
2847 	netif_carrier_on(tun->dev);
2848 
2849 	tun_debug(KERN_INFO, tun, "tun_set_iff\n");
2850 
2851 	/* Make sure persistent devices do not get stuck in
2852 	 * xoff state.
2853 	 */
2854 	if (netif_running(tun->dev))
2855 		netif_tx_wake_all_queues(tun->dev);
2856 
2857 	strcpy(ifr->ifr_name, tun->dev->name);
2858 	return 0;
2859 
2860 err_detach:
2861 	tun_detach_all(dev);
2862 	/* register_netdevice() already called tun_free_netdev() */
2863 	goto err_free_dev;
2864 
2865 err_free_flow:
2866 	tun_flow_uninit(tun);
2867 	security_tun_dev_free_security(tun->security);
2868 err_free_stat:
2869 	free_percpu(tun->pcpu_stats);
2870 err_free_dev:
2871 	free_netdev(dev);
2872 	return err;
2873 }
2874 
2875 static void tun_get_iff(struct tun_struct *tun, struct ifreq *ifr)
2876 {
2877 	tun_debug(KERN_INFO, tun, "tun_get_iff\n");
2878 
2879 	strcpy(ifr->ifr_name, tun->dev->name);
2880 
2881 	ifr->ifr_flags = tun_flags(tun);
2882 
2883 }
2884 
2885 /* This is like a cut-down ethtool ops, except done via tun fd so no
2886  * privs required. */
2887 static int set_offload(struct tun_struct *tun, unsigned long arg)
2888 {
2889 	netdev_features_t features = 0;
2890 
2891 	if (arg & TUN_F_CSUM) {
2892 		features |= NETIF_F_HW_CSUM;
2893 		arg &= ~TUN_F_CSUM;
2894 
2895 		if (arg & (TUN_F_TSO4|TUN_F_TSO6)) {
2896 			if (arg & TUN_F_TSO_ECN) {
2897 				features |= NETIF_F_TSO_ECN;
2898 				arg &= ~TUN_F_TSO_ECN;
2899 			}
2900 			if (arg & TUN_F_TSO4)
2901 				features |= NETIF_F_TSO;
2902 			if (arg & TUN_F_TSO6)
2903 				features |= NETIF_F_TSO6;
2904 			arg &= ~(TUN_F_TSO4|TUN_F_TSO6);
2905 		}
2906 
2907 		arg &= ~TUN_F_UFO;
2908 	}
2909 
2910 	/* This gives the user a way to test for new features in future by
2911 	 * trying to set them. */
2912 	if (arg)
2913 		return -EINVAL;
2914 
2915 	tun->set_features = features;
2916 	tun->dev->wanted_features &= ~TUN_USER_FEATURES;
2917 	tun->dev->wanted_features |= features;
2918 	netdev_update_features(tun->dev);
2919 
2920 	return 0;
2921 }
2922 
2923 static void tun_detach_filter(struct tun_struct *tun, int n)
2924 {
2925 	int i;
2926 	struct tun_file *tfile;
2927 
2928 	for (i = 0; i < n; i++) {
2929 		tfile = rtnl_dereference(tun->tfiles[i]);
2930 		lock_sock(tfile->socket.sk);
2931 		sk_detach_filter(tfile->socket.sk);
2932 		release_sock(tfile->socket.sk);
2933 	}
2934 
2935 	tun->filter_attached = false;
2936 }
2937 
2938 static int tun_attach_filter(struct tun_struct *tun)
2939 {
2940 	int i, ret = 0;
2941 	struct tun_file *tfile;
2942 
2943 	for (i = 0; i < tun->numqueues; i++) {
2944 		tfile = rtnl_dereference(tun->tfiles[i]);
2945 		lock_sock(tfile->socket.sk);
2946 		ret = sk_attach_filter(&tun->fprog, tfile->socket.sk);
2947 		release_sock(tfile->socket.sk);
2948 		if (ret) {
2949 			tun_detach_filter(tun, i);
2950 			return ret;
2951 		}
2952 	}
2953 
2954 	tun->filter_attached = true;
2955 	return ret;
2956 }
2957 
2958 static void tun_set_sndbuf(struct tun_struct *tun)
2959 {
2960 	struct tun_file *tfile;
2961 	int i;
2962 
2963 	for (i = 0; i < tun->numqueues; i++) {
2964 		tfile = rtnl_dereference(tun->tfiles[i]);
2965 		tfile->socket.sk->sk_sndbuf = tun->sndbuf;
2966 	}
2967 }
2968 
2969 static int tun_set_queue(struct file *file, struct ifreq *ifr)
2970 {
2971 	struct tun_file *tfile = file->private_data;
2972 	struct tun_struct *tun;
2973 	int ret = 0;
2974 
2975 	rtnl_lock();
2976 
2977 	if (ifr->ifr_flags & IFF_ATTACH_QUEUE) {
2978 		tun = tfile->detached;
2979 		if (!tun) {
2980 			ret = -EINVAL;
2981 			goto unlock;
2982 		}
2983 		ret = security_tun_dev_attach_queue(tun->security);
2984 		if (ret < 0)
2985 			goto unlock;
2986 		ret = tun_attach(tun, file, false, tun->flags & IFF_NAPI,
2987 				 tun->flags & IFF_NAPI_FRAGS, true);
2988 	} else if (ifr->ifr_flags & IFF_DETACH_QUEUE) {
2989 		tun = rtnl_dereference(tfile->tun);
2990 		if (!tun || !(tun->flags & IFF_MULTI_QUEUE) || tfile->detached)
2991 			ret = -EINVAL;
2992 		else
2993 			__tun_detach(tfile, false);
2994 	} else
2995 		ret = -EINVAL;
2996 
2997 	if (ret >= 0)
2998 		netdev_state_change(tun->dev);
2999 
3000 unlock:
3001 	rtnl_unlock();
3002 	return ret;
3003 }
3004 
3005 static int tun_set_ebpf(struct tun_struct *tun, struct tun_prog **prog_p,
3006 			void __user *data)
3007 {
3008 	struct bpf_prog *prog;
3009 	int fd;
3010 
3011 	if (copy_from_user(&fd, data, sizeof(fd)))
3012 		return -EFAULT;
3013 
3014 	if (fd == -1) {
3015 		prog = NULL;
3016 	} else {
3017 		prog = bpf_prog_get_type(fd, BPF_PROG_TYPE_SOCKET_FILTER);
3018 		if (IS_ERR(prog))
3019 			return PTR_ERR(prog);
3020 	}
3021 
3022 	return __tun_set_ebpf(tun, prog_p, prog);
3023 }
3024 
3025 static long __tun_chr_ioctl(struct file *file, unsigned int cmd,
3026 			    unsigned long arg, int ifreq_len)
3027 {
3028 	struct tun_file *tfile = file->private_data;
3029 	struct net *net = sock_net(&tfile->sk);
3030 	struct tun_struct *tun;
3031 	void __user* argp = (void __user*)arg;
3032 	unsigned int ifindex, carrier;
3033 	struct ifreq ifr;
3034 	kuid_t owner;
3035 	kgid_t group;
3036 	int sndbuf;
3037 	int vnet_hdr_sz;
3038 	int le;
3039 	int ret;
3040 	bool do_notify = false;
3041 
3042 	if (cmd == TUNSETIFF || cmd == TUNSETQUEUE ||
3043 	    (_IOC_TYPE(cmd) == SOCK_IOC_TYPE && cmd != SIOCGSKNS)) {
3044 		if (copy_from_user(&ifr, argp, ifreq_len))
3045 			return -EFAULT;
3046 	} else {
3047 		memset(&ifr, 0, sizeof(ifr));
3048 	}
3049 	if (cmd == TUNGETFEATURES) {
3050 		/* Currently this just means: "what IFF flags are valid?".
3051 		 * This is needed because we never checked for invalid flags on
3052 		 * TUNSETIFF.
3053 		 */
3054 		return put_user(IFF_TUN | IFF_TAP | TUN_FEATURES,
3055 				(unsigned int __user*)argp);
3056 	} else if (cmd == TUNSETQUEUE) {
3057 		return tun_set_queue(file, &ifr);
3058 	} else if (cmd == SIOCGSKNS) {
3059 		if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
3060 			return -EPERM;
3061 		return open_related_ns(&net->ns, get_net_ns);
3062 	}
3063 
3064 	ret = 0;
3065 	rtnl_lock();
3066 
3067 	tun = tun_get(tfile);
3068 	if (cmd == TUNSETIFF) {
3069 		ret = -EEXIST;
3070 		if (tun)
3071 			goto unlock;
3072 
3073 		ifr.ifr_name[IFNAMSIZ-1] = '\0';
3074 
3075 		ret = tun_set_iff(net, file, &ifr);
3076 
3077 		if (ret)
3078 			goto unlock;
3079 
3080 		if (copy_to_user(argp, &ifr, ifreq_len))
3081 			ret = -EFAULT;
3082 		goto unlock;
3083 	}
3084 	if (cmd == TUNSETIFINDEX) {
3085 		ret = -EPERM;
3086 		if (tun)
3087 			goto unlock;
3088 
3089 		ret = -EFAULT;
3090 		if (copy_from_user(&ifindex, argp, sizeof(ifindex)))
3091 			goto unlock;
3092 
3093 		ret = 0;
3094 		tfile->ifindex = ifindex;
3095 		goto unlock;
3096 	}
3097 
3098 	ret = -EBADFD;
3099 	if (!tun)
3100 		goto unlock;
3101 
3102 	tun_debug(KERN_INFO, tun, "tun_chr_ioctl cmd %u\n", cmd);
3103 
3104 	net = dev_net(tun->dev);
3105 	ret = 0;
3106 	switch (cmd) {
3107 	case TUNGETIFF:
3108 		tun_get_iff(tun, &ifr);
3109 
3110 		if (tfile->detached)
3111 			ifr.ifr_flags |= IFF_DETACH_QUEUE;
3112 		if (!tfile->socket.sk->sk_filter)
3113 			ifr.ifr_flags |= IFF_NOFILTER;
3114 
3115 		if (copy_to_user(argp, &ifr, ifreq_len))
3116 			ret = -EFAULT;
3117 		break;
3118 
3119 	case TUNSETNOCSUM:
3120 		/* Disable/Enable checksum */
3121 
3122 		/* [unimplemented] */
3123 		tun_debug(KERN_INFO, tun, "ignored: set checksum %s\n",
3124 			  arg ? "disabled" : "enabled");
3125 		break;
3126 
3127 	case TUNSETPERSIST:
3128 		/* Disable/Enable persist mode. Keep an extra reference to the
3129 		 * module to prevent the module being unprobed.
3130 		 */
3131 		if (arg && !(tun->flags & IFF_PERSIST)) {
3132 			tun->flags |= IFF_PERSIST;
3133 			__module_get(THIS_MODULE);
3134 			do_notify = true;
3135 		}
3136 		if (!arg && (tun->flags & IFF_PERSIST)) {
3137 			tun->flags &= ~IFF_PERSIST;
3138 			module_put(THIS_MODULE);
3139 			do_notify = true;
3140 		}
3141 
3142 		tun_debug(KERN_INFO, tun, "persist %s\n",
3143 			  arg ? "enabled" : "disabled");
3144 		break;
3145 
3146 	case TUNSETOWNER:
3147 		/* Set owner of the device */
3148 		owner = make_kuid(current_user_ns(), arg);
3149 		if (!uid_valid(owner)) {
3150 			ret = -EINVAL;
3151 			break;
3152 		}
3153 		tun->owner = owner;
3154 		do_notify = true;
3155 		tun_debug(KERN_INFO, tun, "owner set to %u\n",
3156 			  from_kuid(&init_user_ns, tun->owner));
3157 		break;
3158 
3159 	case TUNSETGROUP:
3160 		/* Set group of the device */
3161 		group = make_kgid(current_user_ns(), arg);
3162 		if (!gid_valid(group)) {
3163 			ret = -EINVAL;
3164 			break;
3165 		}
3166 		tun->group = group;
3167 		do_notify = true;
3168 		tun_debug(KERN_INFO, tun, "group set to %u\n",
3169 			  from_kgid(&init_user_ns, tun->group));
3170 		break;
3171 
3172 	case TUNSETLINK:
3173 		/* Only allow setting the type when the interface is down */
3174 		if (tun->dev->flags & IFF_UP) {
3175 			tun_debug(KERN_INFO, tun,
3176 				  "Linktype set failed because interface is up\n");
3177 			ret = -EBUSY;
3178 		} else {
3179 			tun->dev->type = (int) arg;
3180 			tun_debug(KERN_INFO, tun, "linktype set to %d\n",
3181 				  tun->dev->type);
3182 			ret = 0;
3183 		}
3184 		break;
3185 
3186 #ifdef TUN_DEBUG
3187 	case TUNSETDEBUG:
3188 		tun->debug = arg;
3189 		break;
3190 #endif
3191 	case TUNSETOFFLOAD:
3192 		ret = set_offload(tun, arg);
3193 		break;
3194 
3195 	case TUNSETTXFILTER:
3196 		/* Can be set only for TAPs */
3197 		ret = -EINVAL;
3198 		if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3199 			break;
3200 		ret = update_filter(&tun->txflt, (void __user *)arg);
3201 		break;
3202 
3203 	case SIOCGIFHWADDR:
3204 		/* Get hw address */
3205 		memcpy(ifr.ifr_hwaddr.sa_data, tun->dev->dev_addr, ETH_ALEN);
3206 		ifr.ifr_hwaddr.sa_family = tun->dev->type;
3207 		if (copy_to_user(argp, &ifr, ifreq_len))
3208 			ret = -EFAULT;
3209 		break;
3210 
3211 	case SIOCSIFHWADDR:
3212 		/* Set hw address */
3213 		tun_debug(KERN_DEBUG, tun, "set hw address: %pM\n",
3214 			  ifr.ifr_hwaddr.sa_data);
3215 
3216 		ret = dev_set_mac_address(tun->dev, &ifr.ifr_hwaddr, NULL);
3217 		break;
3218 
3219 	case TUNGETSNDBUF:
3220 		sndbuf = tfile->socket.sk->sk_sndbuf;
3221 		if (copy_to_user(argp, &sndbuf, sizeof(sndbuf)))
3222 			ret = -EFAULT;
3223 		break;
3224 
3225 	case TUNSETSNDBUF:
3226 		if (copy_from_user(&sndbuf, argp, sizeof(sndbuf))) {
3227 			ret = -EFAULT;
3228 			break;
3229 		}
3230 		if (sndbuf <= 0) {
3231 			ret = -EINVAL;
3232 			break;
3233 		}
3234 
3235 		tun->sndbuf = sndbuf;
3236 		tun_set_sndbuf(tun);
3237 		break;
3238 
3239 	case TUNGETVNETHDRSZ:
3240 		vnet_hdr_sz = tun->vnet_hdr_sz;
3241 		if (copy_to_user(argp, &vnet_hdr_sz, sizeof(vnet_hdr_sz)))
3242 			ret = -EFAULT;
3243 		break;
3244 
3245 	case TUNSETVNETHDRSZ:
3246 		if (copy_from_user(&vnet_hdr_sz, argp, sizeof(vnet_hdr_sz))) {
3247 			ret = -EFAULT;
3248 			break;
3249 		}
3250 		if (vnet_hdr_sz < (int)sizeof(struct virtio_net_hdr)) {
3251 			ret = -EINVAL;
3252 			break;
3253 		}
3254 
3255 		tun->vnet_hdr_sz = vnet_hdr_sz;
3256 		break;
3257 
3258 	case TUNGETVNETLE:
3259 		le = !!(tun->flags & TUN_VNET_LE);
3260 		if (put_user(le, (int __user *)argp))
3261 			ret = -EFAULT;
3262 		break;
3263 
3264 	case TUNSETVNETLE:
3265 		if (get_user(le, (int __user *)argp)) {
3266 			ret = -EFAULT;
3267 			break;
3268 		}
3269 		if (le)
3270 			tun->flags |= TUN_VNET_LE;
3271 		else
3272 			tun->flags &= ~TUN_VNET_LE;
3273 		break;
3274 
3275 	case TUNGETVNETBE:
3276 		ret = tun_get_vnet_be(tun, argp);
3277 		break;
3278 
3279 	case TUNSETVNETBE:
3280 		ret = tun_set_vnet_be(tun, argp);
3281 		break;
3282 
3283 	case TUNATTACHFILTER:
3284 		/* Can be set only for TAPs */
3285 		ret = -EINVAL;
3286 		if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3287 			break;
3288 		ret = -EFAULT;
3289 		if (copy_from_user(&tun->fprog, argp, sizeof(tun->fprog)))
3290 			break;
3291 
3292 		ret = tun_attach_filter(tun);
3293 		break;
3294 
3295 	case TUNDETACHFILTER:
3296 		/* Can be set only for TAPs */
3297 		ret = -EINVAL;
3298 		if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3299 			break;
3300 		ret = 0;
3301 		tun_detach_filter(tun, tun->numqueues);
3302 		break;
3303 
3304 	case TUNGETFILTER:
3305 		ret = -EINVAL;
3306 		if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3307 			break;
3308 		ret = -EFAULT;
3309 		if (copy_to_user(argp, &tun->fprog, sizeof(tun->fprog)))
3310 			break;
3311 		ret = 0;
3312 		break;
3313 
3314 	case TUNSETSTEERINGEBPF:
3315 		ret = tun_set_ebpf(tun, &tun->steering_prog, argp);
3316 		break;
3317 
3318 	case TUNSETFILTEREBPF:
3319 		ret = tun_set_ebpf(tun, &tun->filter_prog, argp);
3320 		break;
3321 
3322 	case TUNSETCARRIER:
3323 		ret = -EFAULT;
3324 		if (copy_from_user(&carrier, argp, sizeof(carrier)))
3325 			goto unlock;
3326 
3327 		ret = tun_net_change_carrier(tun->dev, (bool)carrier);
3328 		break;
3329 
3330 	case TUNGETDEVNETNS:
3331 		ret = -EPERM;
3332 		if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
3333 			goto unlock;
3334 		ret = open_related_ns(&net->ns, get_net_ns);
3335 		break;
3336 
3337 	default:
3338 		ret = -EINVAL;
3339 		break;
3340 	}
3341 
3342 	if (do_notify)
3343 		netdev_state_change(tun->dev);
3344 
3345 unlock:
3346 	rtnl_unlock();
3347 	if (tun)
3348 		tun_put(tun);
3349 	return ret;
3350 }
3351 
3352 static long tun_chr_ioctl(struct file *file,
3353 			  unsigned int cmd, unsigned long arg)
3354 {
3355 	return __tun_chr_ioctl(file, cmd, arg, sizeof (struct ifreq));
3356 }
3357 
3358 #ifdef CONFIG_COMPAT
3359 static long tun_chr_compat_ioctl(struct file *file,
3360 			 unsigned int cmd, unsigned long arg)
3361 {
3362 	switch (cmd) {
3363 	case TUNSETIFF:
3364 	case TUNGETIFF:
3365 	case TUNSETTXFILTER:
3366 	case TUNGETSNDBUF:
3367 	case TUNSETSNDBUF:
3368 	case SIOCGIFHWADDR:
3369 	case SIOCSIFHWADDR:
3370 		arg = (unsigned long)compat_ptr(arg);
3371 		break;
3372 	default:
3373 		arg = (compat_ulong_t)arg;
3374 		break;
3375 	}
3376 
3377 	/*
3378 	 * compat_ifreq is shorter than ifreq, so we must not access beyond
3379 	 * the end of that structure. All fields that are used in this
3380 	 * driver are compatible though, we don't need to convert the
3381 	 * contents.
3382 	 */
3383 	return __tun_chr_ioctl(file, cmd, arg, sizeof(struct compat_ifreq));
3384 }
3385 #endif /* CONFIG_COMPAT */
3386 
3387 static int tun_chr_fasync(int fd, struct file *file, int on)
3388 {
3389 	struct tun_file *tfile = file->private_data;
3390 	int ret;
3391 
3392 	if ((ret = fasync_helper(fd, file, on, &tfile->fasync)) < 0)
3393 		goto out;
3394 
3395 	if (on) {
3396 		__f_setown(file, task_pid(current), PIDTYPE_TGID, 0);
3397 		tfile->flags |= TUN_FASYNC;
3398 	} else
3399 		tfile->flags &= ~TUN_FASYNC;
3400 	ret = 0;
3401 out:
3402 	return ret;
3403 }
3404 
3405 static int tun_chr_open(struct inode *inode, struct file * file)
3406 {
3407 	struct net *net = current->nsproxy->net_ns;
3408 	struct tun_file *tfile;
3409 
3410 	DBG1(KERN_INFO, "tunX: tun_chr_open\n");
3411 
3412 	tfile = (struct tun_file *)sk_alloc(net, AF_UNSPEC, GFP_KERNEL,
3413 					    &tun_proto, 0);
3414 	if (!tfile)
3415 		return -ENOMEM;
3416 	if (ptr_ring_init(&tfile->tx_ring, 0, GFP_KERNEL)) {
3417 		sk_free(&tfile->sk);
3418 		return -ENOMEM;
3419 	}
3420 
3421 	mutex_init(&tfile->napi_mutex);
3422 	RCU_INIT_POINTER(tfile->tun, NULL);
3423 	tfile->flags = 0;
3424 	tfile->ifindex = 0;
3425 
3426 	init_waitqueue_head(&tfile->socket.wq.wait);
3427 
3428 	tfile->socket.file = file;
3429 	tfile->socket.ops = &tun_socket_ops;
3430 
3431 	sock_init_data(&tfile->socket, &tfile->sk);
3432 
3433 	tfile->sk.sk_write_space = tun_sock_write_space;
3434 	tfile->sk.sk_sndbuf = INT_MAX;
3435 
3436 	file->private_data = tfile;
3437 	INIT_LIST_HEAD(&tfile->next);
3438 
3439 	sock_set_flag(&tfile->sk, SOCK_ZEROCOPY);
3440 
3441 	return 0;
3442 }
3443 
3444 static int tun_chr_close(struct inode *inode, struct file *file)
3445 {
3446 	struct tun_file *tfile = file->private_data;
3447 
3448 	tun_detach(tfile, true);
3449 
3450 	return 0;
3451 }
3452 
3453 #ifdef CONFIG_PROC_FS
3454 static void tun_chr_show_fdinfo(struct seq_file *m, struct file *file)
3455 {
3456 	struct tun_file *tfile = file->private_data;
3457 	struct tun_struct *tun;
3458 	struct ifreq ifr;
3459 
3460 	memset(&ifr, 0, sizeof(ifr));
3461 
3462 	rtnl_lock();
3463 	tun = tun_get(tfile);
3464 	if (tun)
3465 		tun_get_iff(tun, &ifr);
3466 	rtnl_unlock();
3467 
3468 	if (tun)
3469 		tun_put(tun);
3470 
3471 	seq_printf(m, "iff:\t%s\n", ifr.ifr_name);
3472 }
3473 #endif
3474 
3475 static const struct file_operations tun_fops = {
3476 	.owner	= THIS_MODULE,
3477 	.llseek = no_llseek,
3478 	.read_iter  = tun_chr_read_iter,
3479 	.write_iter = tun_chr_write_iter,
3480 	.poll	= tun_chr_poll,
3481 	.unlocked_ioctl	= tun_chr_ioctl,
3482 #ifdef CONFIG_COMPAT
3483 	.compat_ioctl = tun_chr_compat_ioctl,
3484 #endif
3485 	.open	= tun_chr_open,
3486 	.release = tun_chr_close,
3487 	.fasync = tun_chr_fasync,
3488 #ifdef CONFIG_PROC_FS
3489 	.show_fdinfo = tun_chr_show_fdinfo,
3490 #endif
3491 };
3492 
3493 static struct miscdevice tun_miscdev = {
3494 	.minor = TUN_MINOR,
3495 	.name = "tun",
3496 	.nodename = "net/tun",
3497 	.fops = &tun_fops,
3498 };
3499 
3500 /* ethtool interface */
3501 
3502 static void tun_default_link_ksettings(struct net_device *dev,
3503 				       struct ethtool_link_ksettings *cmd)
3504 {
3505 	ethtool_link_ksettings_zero_link_mode(cmd, supported);
3506 	ethtool_link_ksettings_zero_link_mode(cmd, advertising);
3507 	cmd->base.speed		= SPEED_10;
3508 	cmd->base.duplex	= DUPLEX_FULL;
3509 	cmd->base.port		= PORT_TP;
3510 	cmd->base.phy_address	= 0;
3511 	cmd->base.autoneg	= AUTONEG_DISABLE;
3512 }
3513 
3514 static int tun_get_link_ksettings(struct net_device *dev,
3515 				  struct ethtool_link_ksettings *cmd)
3516 {
3517 	struct tun_struct *tun = netdev_priv(dev);
3518 
3519 	memcpy(cmd, &tun->link_ksettings, sizeof(*cmd));
3520 	return 0;
3521 }
3522 
3523 static int tun_set_link_ksettings(struct net_device *dev,
3524 				  const struct ethtool_link_ksettings *cmd)
3525 {
3526 	struct tun_struct *tun = netdev_priv(dev);
3527 
3528 	memcpy(&tun->link_ksettings, cmd, sizeof(*cmd));
3529 	return 0;
3530 }
3531 
3532 static void tun_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
3533 {
3534 	struct tun_struct *tun = netdev_priv(dev);
3535 
3536 	strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
3537 	strlcpy(info->version, DRV_VERSION, sizeof(info->version));
3538 
3539 	switch (tun->flags & TUN_TYPE_MASK) {
3540 	case IFF_TUN:
3541 		strlcpy(info->bus_info, "tun", sizeof(info->bus_info));
3542 		break;
3543 	case IFF_TAP:
3544 		strlcpy(info->bus_info, "tap", sizeof(info->bus_info));
3545 		break;
3546 	}
3547 }
3548 
3549 static u32 tun_get_msglevel(struct net_device *dev)
3550 {
3551 #ifdef TUN_DEBUG
3552 	struct tun_struct *tun = netdev_priv(dev);
3553 	return tun->debug;
3554 #else
3555 	return -EOPNOTSUPP;
3556 #endif
3557 }
3558 
3559 static void tun_set_msglevel(struct net_device *dev, u32 value)
3560 {
3561 #ifdef TUN_DEBUG
3562 	struct tun_struct *tun = netdev_priv(dev);
3563 	tun->debug = value;
3564 #endif
3565 }
3566 
3567 static int tun_get_coalesce(struct net_device *dev,
3568 			    struct ethtool_coalesce *ec)
3569 {
3570 	struct tun_struct *tun = netdev_priv(dev);
3571 
3572 	ec->rx_max_coalesced_frames = tun->rx_batched;
3573 
3574 	return 0;
3575 }
3576 
3577 static int tun_set_coalesce(struct net_device *dev,
3578 			    struct ethtool_coalesce *ec)
3579 {
3580 	struct tun_struct *tun = netdev_priv(dev);
3581 
3582 	if (ec->rx_max_coalesced_frames > NAPI_POLL_WEIGHT)
3583 		tun->rx_batched = NAPI_POLL_WEIGHT;
3584 	else
3585 		tun->rx_batched = ec->rx_max_coalesced_frames;
3586 
3587 	return 0;
3588 }
3589 
3590 static const struct ethtool_ops tun_ethtool_ops = {
3591 	.get_drvinfo	= tun_get_drvinfo,
3592 	.get_msglevel	= tun_get_msglevel,
3593 	.set_msglevel	= tun_set_msglevel,
3594 	.get_link	= ethtool_op_get_link,
3595 	.get_ts_info	= ethtool_op_get_ts_info,
3596 	.get_coalesce   = tun_get_coalesce,
3597 	.set_coalesce   = tun_set_coalesce,
3598 	.get_link_ksettings = tun_get_link_ksettings,
3599 	.set_link_ksettings = tun_set_link_ksettings,
3600 };
3601 
3602 static int tun_queue_resize(struct tun_struct *tun)
3603 {
3604 	struct net_device *dev = tun->dev;
3605 	struct tun_file *tfile;
3606 	struct ptr_ring **rings;
3607 	int n = tun->numqueues + tun->numdisabled;
3608 	int ret, i;
3609 
3610 	rings = kmalloc_array(n, sizeof(*rings), GFP_KERNEL);
3611 	if (!rings)
3612 		return -ENOMEM;
3613 
3614 	for (i = 0; i < tun->numqueues; i++) {
3615 		tfile = rtnl_dereference(tun->tfiles[i]);
3616 		rings[i] = &tfile->tx_ring;
3617 	}
3618 	list_for_each_entry(tfile, &tun->disabled, next)
3619 		rings[i++] = &tfile->tx_ring;
3620 
3621 	ret = ptr_ring_resize_multiple(rings, n,
3622 				       dev->tx_queue_len, GFP_KERNEL,
3623 				       tun_ptr_free);
3624 
3625 	kfree(rings);
3626 	return ret;
3627 }
3628 
3629 static int tun_device_event(struct notifier_block *unused,
3630 			    unsigned long event, void *ptr)
3631 {
3632 	struct net_device *dev = netdev_notifier_info_to_dev(ptr);
3633 	struct tun_struct *tun = netdev_priv(dev);
3634 	int i;
3635 
3636 	if (dev->rtnl_link_ops != &tun_link_ops)
3637 		return NOTIFY_DONE;
3638 
3639 	switch (event) {
3640 	case NETDEV_CHANGE_TX_QUEUE_LEN:
3641 		if (tun_queue_resize(tun))
3642 			return NOTIFY_BAD;
3643 		break;
3644 	case NETDEV_UP:
3645 		for (i = 0; i < tun->numqueues; i++) {
3646 			struct tun_file *tfile;
3647 
3648 			tfile = rtnl_dereference(tun->tfiles[i]);
3649 			tfile->socket.sk->sk_write_space(tfile->socket.sk);
3650 		}
3651 		break;
3652 	default:
3653 		break;
3654 	}
3655 
3656 	return NOTIFY_DONE;
3657 }
3658 
3659 static struct notifier_block tun_notifier_block __read_mostly = {
3660 	.notifier_call	= tun_device_event,
3661 };
3662 
3663 static int __init tun_init(void)
3664 {
3665 	int ret = 0;
3666 
3667 	pr_info("%s, %s\n", DRV_DESCRIPTION, DRV_VERSION);
3668 
3669 	ret = rtnl_link_register(&tun_link_ops);
3670 	if (ret) {
3671 		pr_err("Can't register link_ops\n");
3672 		goto err_linkops;
3673 	}
3674 
3675 	ret = misc_register(&tun_miscdev);
3676 	if (ret) {
3677 		pr_err("Can't register misc device %d\n", TUN_MINOR);
3678 		goto err_misc;
3679 	}
3680 
3681 	ret = register_netdevice_notifier(&tun_notifier_block);
3682 	if (ret) {
3683 		pr_err("Can't register netdevice notifier\n");
3684 		goto err_notifier;
3685 	}
3686 
3687 	return  0;
3688 
3689 err_notifier:
3690 	misc_deregister(&tun_miscdev);
3691 err_misc:
3692 	rtnl_link_unregister(&tun_link_ops);
3693 err_linkops:
3694 	return ret;
3695 }
3696 
3697 static void tun_cleanup(void)
3698 {
3699 	misc_deregister(&tun_miscdev);
3700 	rtnl_link_unregister(&tun_link_ops);
3701 	unregister_netdevice_notifier(&tun_notifier_block);
3702 }
3703 
3704 /* Get an underlying socket object from tun file.  Returns error unless file is
3705  * attached to a device.  The returned object works like a packet socket, it
3706  * can be used for sock_sendmsg/sock_recvmsg.  The caller is responsible for
3707  * holding a reference to the file for as long as the socket is in use. */
3708 struct socket *tun_get_socket(struct file *file)
3709 {
3710 	struct tun_file *tfile;
3711 	if (file->f_op != &tun_fops)
3712 		return ERR_PTR(-EINVAL);
3713 	tfile = file->private_data;
3714 	if (!tfile)
3715 		return ERR_PTR(-EBADFD);
3716 	return &tfile->socket;
3717 }
3718 EXPORT_SYMBOL_GPL(tun_get_socket);
3719 
3720 struct ptr_ring *tun_get_tx_ring(struct file *file)
3721 {
3722 	struct tun_file *tfile;
3723 
3724 	if (file->f_op != &tun_fops)
3725 		return ERR_PTR(-EINVAL);
3726 	tfile = file->private_data;
3727 	if (!tfile)
3728 		return ERR_PTR(-EBADFD);
3729 	return &tfile->tx_ring;
3730 }
3731 EXPORT_SYMBOL_GPL(tun_get_tx_ring);
3732 
3733 module_init(tun_init);
3734 module_exit(tun_cleanup);
3735 MODULE_DESCRIPTION(DRV_DESCRIPTION);
3736 MODULE_AUTHOR(DRV_COPYRIGHT);
3737 MODULE_LICENSE("GPL");
3738 MODULE_ALIAS_MISCDEV(TUN_MINOR);
3739 MODULE_ALIAS("devname:net/tun");
3740