]> git.proxmox.com Git - mirror_ubuntu-bionic-kernel.git/blob - drivers/net/tun.c
projects / mirror_ubuntu-bionic-kernel.git / blob
? search:


b79722d9f8a050872f36207e533deaf900504339
[mirror_ubuntu-bionic-kernel.git] / drivers / net / tun.c
1 /*
2 * TUN - Universal TUN/TAP device driver.
3 * Copyright (C) 1999-2002 Maxim Krasnyansky <maxk@qualcomm.com>
4 *
5 * This program is free software; you can redistribute it and/or modify
6 * it under the terms of the GNU General Public License as published by
7 * the Free Software Foundation; either version 2 of the License, or
8 * (at your option) any later version.
9 *
10 * This program is distributed in the hope that it will be useful,
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
13 * GNU General Public License for more details.
14 *
15 * $Id: tun.c,v 1.15 2002/03/01 02:44:24 maxk Exp $
16 */
17
18 /*
19 * Changes:
20 *
21 * Mike Kershaw <dragorn@kismetwireless.net> 2005/08/14
22 * Add TUNSETLINK ioctl to set the link encapsulation
23 *
24 * Mark Smith <markzzzsmith@yahoo.com.au>
25 * Use eth_random_addr() for tap MAC address.
26 *
27 * Harald Roelle <harald.roelle@ifi.lmu.de> 2004/04/20
28 * Fixes in packet dropping, queue length setting and queue wakeup.
29 * Increased default tx queue length.
30 * Added ethtool API.
31 * Minor cleanups
32 *
33 * Daniel Podlejski <underley@underley.eu.org>
34 * Modifications for 2.3.99-pre5 kernel.
35 */
36
37 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
38
39 #define DRV_NAME "tun"
40 #define DRV_VERSION "1.6"
41 #define DRV_DESCRIPTION "Universal TUN/TAP device driver"
42 #define DRV_COPYRIGHT "(C) 1999-2004 Max Krasnyansky <maxk@qualcomm.com>"
43
44 #include <linux/module.h>
45 #include <linux/errno.h>
46 #include <linux/kernel.h>
47 #include <linux/sched/signal.h>
48 #include <linux/major.h>
49 #include <linux/slab.h>
50 #include <linux/poll.h>
51 #include <linux/fcntl.h>
52 #include <linux/init.h>
53 #include <linux/skbuff.h>
54 #include <linux/netdevice.h>
55 #include <linux/etherdevice.h>
56 #include <linux/miscdevice.h>
57 #include <linux/ethtool.h>
58 #include <linux/rtnetlink.h>
59 #include <linux/compat.h>
60 #include <linux/if.h>
61 #include <linux/if_arp.h>
62 #include <linux/if_ether.h>
63 #include <linux/if_tun.h>
64 #include <linux/if_vlan.h>
65 #include <linux/crc32.h>
66 #include <linux/nsproxy.h>
67 #include <linux/virtio_net.h>
68 #include <linux/rcupdate.h>
69 #include <net/net_namespace.h>
70 #include <net/netns/generic.h>
71 #include <net/rtnetlink.h>
72 #include <net/sock.h>
73 #include <linux/seq_file.h>
74 #include <linux/uio.h>
75 #include <linux/skb_array.h>
76 #include <linux/bpf.h>
77 #include <linux/bpf_trace.h>
78 #include <linux/mutex.h>
79
80 #include <linux/uaccess.h>
81
82 /* Uncomment to enable debugging */
83 /* #define TUN_DEBUG 1 */
84
85 #ifdef TUN_DEBUG
86 static int debug;
87
88 #define tun_debug(level, tun, fmt, args...) \
89 do { \
90 if (tun->debug) \
91 netdev_printk(level, tun->dev, fmt, ##args); \
92 } while (0)
93 #define DBG1(level, fmt, args...) \
94 do { \
95 if (debug == 2) \
96 printk(level fmt, ##args); \
97 } while (0)
98 #else
99 #define tun_debug(level, tun, fmt, args...) \
100 do { \
101 if (0) \
102 netdev_printk(level, tun->dev, fmt, ##args); \
103 } while (0)
104 #define DBG1(level, fmt, args...) \
105 do { \
106 if (0) \
107 printk(level fmt, ##args); \
108 } while (0)
109 #endif
110
111 #define TUN_HEADROOM 256
112 #define TUN_RX_PAD (NET_IP_ALIGN + NET_SKB_PAD)
113
114 /* TUN device flags */
115
116 /* IFF_ATTACH_QUEUE is never stored in device flags,
117 * overload it to mean fasync when stored there.
118 */
119 #define TUN_FASYNC IFF_ATTACH_QUEUE
120 /* High bits in flags field are unused. */
121 #define TUN_VNET_LE 0x80000000
122 #define TUN_VNET_BE 0x40000000
123
124 #define TUN_FEATURES (IFF_NO_PI | IFF_ONE_QUEUE | IFF_VNET_HDR | \
125 IFF_MULTI_QUEUE | IFF_NAPI | IFF_NAPI_FRAGS)
126
127 #define GOODCOPY_LEN 128
128
129 #define FLT_EXACT_COUNT 8
130 struct tap_filter {
131 unsigned int count; /* Number of addrs. Zero means disabled */
132 u32 mask[2]; /* Mask of the hashed addrs */
133 unsigned char addr[FLT_EXACT_COUNT][ETH_ALEN];
134 };
135
136 /* MAX_TAP_QUEUES 256 is chosen to allow rx/tx queues to be equal
137 * to max number of VCPUs in guest. */
138 #define MAX_TAP_QUEUES 256
139 #define MAX_TAP_FLOWS 4096
140
141 #define TUN_FLOW_EXPIRE (3 * HZ)
142
143 struct tun_pcpu_stats {
144 u64 rx_packets;
145 u64 rx_bytes;
146 u64 tx_packets;
147 u64 tx_bytes;
148 struct u64_stats_sync syncp;
149 u32 rx_dropped;
150 u32 tx_dropped;
151 u32 rx_frame_errors;
152 };
153
154 /* A tun_file connects an open character device to a tuntap netdevice. It
155 * also contains all socket related structures (except sock_fprog and tap_filter)
156 * to serve as one transmit queue for tuntap device. The sock_fprog and
157 * tap_filter were kept in tun_struct since they were used for filtering for the
158 * netdevice not for a specific queue (at least I didn't see the requirement for
159 * this).
160 *
161 * RCU usage:
162 * The tun_file and tun_struct are loosely coupled, the pointer from one to the
163 * other can only be read while rcu_read_lock or rtnl_lock is held.
164 */
165 struct tun_file {
166 struct sock sk;
167 struct socket socket;
168 struct socket_wq wq;
169 struct tun_struct __rcu *tun;
170 struct fasync_struct *fasync;
171 /* only used for fasnyc */
172 unsigned int flags;
173 union {
174 u16 queue_index;
175 unsigned int ifindex;
176 };
177 struct napi_struct napi;
178 bool napi_enabled;
179 bool napi_frags_enabled;
180 struct mutex napi_mutex; /* Protects access to the above napi */
181 struct list_head next;
182 struct tun_struct *detached;
183 struct skb_array tx_array;
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;
195 };
196
197 #define TUN_NUM_FLOW_ENTRIES 1024
198
199 /* Since the socket were moved to tun_file, to preserve the behavior of persist
200 * device, socket filter, sndbuf and vnet header size were restore when the
201 * file were attached to a persist device.
202 */
203 struct tun_struct {
204 struct tun_file __rcu *tfiles[MAX_TAP_QUEUES];
205 unsigned int numqueues;
206 unsigned int flags;
207 kuid_t owner;
208 kgid_t group;
209
210 struct net_device *dev;
211 netdev_features_t set_features;
212 #define TUN_USER_FEATURES (NETIF_F_HW_CSUM|NETIF_F_TSO_ECN|NETIF_F_TSO| \
213 NETIF_F_TSO6)
214
215 int align;
216 int vnet_hdr_sz;
217 int sndbuf;
218 struct tap_filter txflt;
219 struct sock_fprog fprog;
220 /* protected by rtnl lock */
221 bool filter_attached;
222 #ifdef TUN_DEBUG
223 int debug;
224 #endif
225 spinlock_t lock;
226 struct hlist_head flows[TUN_NUM_FLOW_ENTRIES];
227 struct timer_list flow_gc_timer;
228 unsigned long ageing_time;
229 unsigned int numdisabled;
230 struct list_head disabled;
231 void *security;
232 u32 flow_count;
233 u32 rx_batched;
234 struct tun_pcpu_stats __percpu *pcpu_stats;
235 struct bpf_prog __rcu *xdp_prog;
236 };
237
238 static int tun_napi_receive(struct napi_struct *napi, int budget)
239 {
240 struct tun_file *tfile = container_of(napi, struct tun_file, napi);
241 struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
242 struct sk_buff_head process_queue;
243 struct sk_buff *skb;
244 int received = 0;
245
246 __skb_queue_head_init(&process_queue);
247
248 spin_lock(&queue->lock);
249 skb_queue_splice_tail_init(queue, &process_queue);
250 spin_unlock(&queue->lock);
251
252 while (received < budget && (skb = __skb_dequeue(&process_queue))) {
253 napi_gro_receive(napi, skb);
254 ++received;
255 }
256
257 if (!skb_queue_empty(&process_queue)) {
258 spin_lock(&queue->lock);
259 skb_queue_splice(&process_queue, queue);
260 spin_unlock(&queue->lock);
261 }
262
263 return received;
264 }
265
266 static int tun_napi_poll(struct napi_struct *napi, int budget)
267 {
268 unsigned int received;
269
270 received = tun_napi_receive(napi, budget);
271
272 if (received < budget)
273 napi_complete_done(napi, received);
274
275 return received;
276 }
277
278 static void tun_napi_init(struct tun_struct *tun, struct tun_file *tfile,
279 bool napi_en, bool napi_frags)
280 {
281 tfile->napi_enabled = napi_en;
282 tfile->napi_frags_enabled = napi_en && napi_frags;
283 if (napi_en) {
284 netif_napi_add(tun->dev, &tfile->napi, tun_napi_poll,
285 NAPI_POLL_WEIGHT);
286 napi_enable(&tfile->napi);
287 }
288 }
289
290 static void tun_napi_disable(struct tun_file *tfile)
291 {
292 if (tfile->napi_enabled)
293 napi_disable(&tfile->napi);
294 }
295
296 static void tun_napi_del(struct tun_file *tfile)
297 {
298 if (tfile->napi_enabled)
299 netif_napi_del(&tfile->napi);
300 }
301
302 static bool tun_napi_frags_enabled(const struct tun_file *tfile)
303 {
304 return tfile->napi_frags_enabled;
305 }
306
307 #ifdef CONFIG_TUN_VNET_CROSS_LE
308 static inline bool tun_legacy_is_little_endian(struct tun_struct *tun)
309 {
310 return tun->flags & TUN_VNET_BE ? false :
311 virtio_legacy_is_little_endian();
312 }
313
314 static long tun_get_vnet_be(struct tun_struct *tun, int __user *argp)
315 {
316 int be = !!(tun->flags & TUN_VNET_BE);
317
318 if (put_user(be, argp))
319 return -EFAULT;
320
321 return 0;
322 }
323
324 static long tun_set_vnet_be(struct tun_struct *tun, int __user *argp)
325 {
326 int be;
327
328 if (get_user(be, argp))
329 return -EFAULT;
330
331 if (be)
332 tun->flags |= TUN_VNET_BE;
333 else
334 tun->flags &= ~TUN_VNET_BE;
335
336 return 0;
337 }
338 #else
339 static inline bool tun_legacy_is_little_endian(struct tun_struct *tun)
340 {
341 return virtio_legacy_is_little_endian();
342 }
343
344 static long tun_get_vnet_be(struct tun_struct *tun, int __user *argp)
345 {
346 return -EINVAL;
347 }
348
349 static long tun_set_vnet_be(struct tun_struct *tun, int __user *argp)
350 {
351 return -EINVAL;
352 }
353 #endif /* CONFIG_TUN_VNET_CROSS_LE */
354
355 static inline bool tun_is_little_endian(struct tun_struct *tun)
356 {
357 return tun->flags & TUN_VNET_LE ||
358 tun_legacy_is_little_endian(tun);
359 }
360
361 static inline u16 tun16_to_cpu(struct tun_struct *tun, __virtio16 val)
362 {
363 return __virtio16_to_cpu(tun_is_little_endian(tun), val);
364 }
365
366 static inline __virtio16 cpu_to_tun16(struct tun_struct *tun, u16 val)
367 {
368 return __cpu_to_virtio16(tun_is_little_endian(tun), val);
369 }
370
371 static inline u32 tun_hashfn(u32 rxhash)
372 {
373 return rxhash & 0x3ff;
374 }
375
376 static struct tun_flow_entry *tun_flow_find(struct hlist_head *head, u32 rxhash)
377 {
378 struct tun_flow_entry *e;
379
380 hlist_for_each_entry_rcu(e, head, hash_link) {
381 if (e->rxhash == rxhash)
382 return e;
383 }
384 return NULL;
385 }
386
387 static struct tun_flow_entry *tun_flow_create(struct tun_struct *tun,
388 struct hlist_head *head,
389 u32 rxhash, u16 queue_index)
390 {
391 struct tun_flow_entry *e = kmalloc(sizeof(*e), GFP_ATOMIC);
392
393 if (e) {
394 tun_debug(KERN_INFO, tun, "create flow: hash %u index %u\n",
395 rxhash, queue_index);
396 e->updated = jiffies;
397 e->rxhash = rxhash;
398 e->rps_rxhash = 0;
399 e->queue_index = queue_index;
400 e->tun = tun;
401 hlist_add_head_rcu(&e->hash_link, head);
402 ++tun->flow_count;
403 }
404 return e;
405 }
406
407 static void tun_flow_delete(struct tun_struct *tun, struct tun_flow_entry *e)
408 {
409 tun_debug(KERN_INFO, tun, "delete flow: hash %u index %u\n",
410 e->rxhash, e->queue_index);
411 hlist_del_rcu(&e->hash_link);
412 kfree_rcu(e, rcu);
413 --tun->flow_count;
414 }
415
416 static void tun_flow_flush(struct tun_struct *tun)
417 {
418 int i;
419
420 spin_lock_bh(&tun->lock);
421 for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
422 struct tun_flow_entry *e;
423 struct hlist_node *n;
424
425 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link)
426 tun_flow_delete(tun, e);
427 }
428 spin_unlock_bh(&tun->lock);
429 }
430
431 static void tun_flow_delete_by_queue(struct tun_struct *tun, u16 queue_index)
432 {
433 int i;
434
435 spin_lock_bh(&tun->lock);
436 for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
437 struct tun_flow_entry *e;
438 struct hlist_node *n;
439
440 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
441 if (e->queue_index == queue_index)
442 tun_flow_delete(tun, e);
443 }
444 }
445 spin_unlock_bh(&tun->lock);
446 }
447
448 static void tun_flow_cleanup(struct timer_list *t)
449 {
450 struct tun_struct *tun = from_timer(tun, t, flow_gc_timer);
451 unsigned long delay = tun->ageing_time;
452 unsigned long next_timer = jiffies + delay;
453 unsigned long count = 0;
454 int i;
455
456 tun_debug(KERN_INFO, tun, "tun_flow_cleanup\n");
457
458 spin_lock(&tun->lock);
459 for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
460 struct tun_flow_entry *e;
461 struct hlist_node *n;
462
463 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
464 unsigned long this_timer;
465
466 this_timer = e->updated + delay;
467 if (time_before_eq(this_timer, jiffies)) {
468 tun_flow_delete(tun, e);
469 continue;
470 }
471 count++;
472 if (time_before(this_timer, next_timer))
473 next_timer = this_timer;
474 }
475 }
476
477 if (count)
478 mod_timer(&tun->flow_gc_timer, round_jiffies_up(next_timer));
479 spin_unlock(&tun->lock);
480 }
481
482 static void tun_flow_update(struct tun_struct *tun, u32 rxhash,
483 struct tun_file *tfile)
484 {
485 struct hlist_head *head;
486 struct tun_flow_entry *e;
487 unsigned long delay = tun->ageing_time;
488 u16 queue_index = tfile->queue_index;
489
490 if (!rxhash)
491 return;
492 else
493 head = &tun->flows[tun_hashfn(rxhash)];
494
495 rcu_read_lock();
496
497 /* We may get a very small possibility of OOO during switching, not
498 * worth to optimize.*/
499 if (tun->numqueues == 1 || tfile->detached)
500 goto unlock;
501
502 e = tun_flow_find(head, rxhash);
503 if (likely(e)) {
504 /* TODO: keep queueing to old queue until it's empty? */
505 e->queue_index = queue_index;
506 e->updated = jiffies;
507 sock_rps_record_flow_hash(e->rps_rxhash);
508 } else {
509 spin_lock_bh(&tun->lock);
510 if (!tun_flow_find(head, rxhash) &&
511 tun->flow_count < MAX_TAP_FLOWS)
512 tun_flow_create(tun, head, rxhash, queue_index);
513
514 if (!timer_pending(&tun->flow_gc_timer))
515 mod_timer(&tun->flow_gc_timer,
516 round_jiffies_up(jiffies + delay));
517 spin_unlock_bh(&tun->lock);
518 }
519
520 unlock:
521 rcu_read_unlock();
522 }
523
524 /**
525 * Save the hash received in the stack receive path and update the
526 * flow_hash table accordingly.
527 */
528 static inline void tun_flow_save_rps_rxhash(struct tun_flow_entry *e, u32 hash)
529 {
530 if (unlikely(e->rps_rxhash != hash))
531 e->rps_rxhash = hash;
532 }
533
534 /* We try to identify a flow through its rxhash first. The reason that
535 * we do not check rxq no. is because some cards(e.g 82599), chooses
536 * the rxq based on the txq where the last packet of the flow comes. As
537 * the userspace application move between processors, we may get a
538 * different rxq no. here. If we could not get rxhash, then we would
539 * hope the rxq no. may help here.
540 */
541 static u16 tun_select_queue(struct net_device *dev, struct sk_buff *skb,
542 void *accel_priv, select_queue_fallback_t fallback)
543 {
544 struct tun_struct *tun = netdev_priv(dev);
545 struct tun_flow_entry *e;
546 u32 txq = 0;
547 u32 numqueues = 0;
548
549 rcu_read_lock();
550 numqueues = READ_ONCE(tun->numqueues);
551
552 txq = __skb_get_hash_symmetric(skb);
553 if (txq) {
554 e = tun_flow_find(&tun->flows[tun_hashfn(txq)], txq);
555 if (e) {
556 tun_flow_save_rps_rxhash(e, txq);
557 txq = e->queue_index;
558 } else
559 /* use multiply and shift instead of expensive divide */
560 txq = ((u64)txq * numqueues) >> 32;
561 } else if (likely(skb_rx_queue_recorded(skb))) {
562 txq = skb_get_rx_queue(skb);
563 while (unlikely(txq >= numqueues))
564 txq -= numqueues;
565 }
566
567 rcu_read_unlock();
568 return txq;
569 }
570
571 static inline bool tun_not_capable(struct tun_struct *tun)
572 {
573 const struct cred *cred = current_cred();
574 struct net *net = dev_net(tun->dev);
575
576 return ((uid_valid(tun->owner) && !uid_eq(cred->euid, tun->owner)) ||
577 (gid_valid(tun->group) && !in_egroup_p(tun->group))) &&
578 !ns_capable(net->user_ns, CAP_NET_ADMIN);
579 }
580
581 static void tun_set_real_num_queues(struct tun_struct *tun)
582 {
583 netif_set_real_num_tx_queues(tun->dev, tun->numqueues);
584 netif_set_real_num_rx_queues(tun->dev, tun->numqueues);
585 }
586
587 static void tun_disable_queue(struct tun_struct *tun, struct tun_file *tfile)
588 {
589 tfile->detached = tun;
590 list_add_tail(&tfile->next, &tun->disabled);
591 ++tun->numdisabled;
592 }
593
594 static struct tun_struct *tun_enable_queue(struct tun_file *tfile)
595 {
596 struct tun_struct *tun = tfile->detached;
597
598 tfile->detached = NULL;
599 list_del_init(&tfile->next);
600 --tun->numdisabled;
601 return tun;
602 }
603
604 static void tun_queue_purge(struct tun_file *tfile)
605 {
606 struct sk_buff *skb;
607
608 while ((skb = skb_array_consume(&tfile->tx_array)) != NULL)
609 kfree_skb(skb);
610
611 skb_queue_purge(&tfile->sk.sk_write_queue);
612 skb_queue_purge(&tfile->sk.sk_error_queue);
613 }
614
615 static void __tun_detach(struct tun_file *tfile, bool clean)
616 {
617 struct tun_file *ntfile;
618 struct tun_struct *tun;
619
620 tun = rtnl_dereference(tfile->tun);
621
622 if (tun && clean) {
623 tun_napi_disable(tfile);
624 tun_napi_del(tfile);
625 }
626
627 if (tun && !tfile->detached) {
628 u16 index = tfile->queue_index;
629 BUG_ON(index >= tun->numqueues);
630
631 rcu_assign_pointer(tun->tfiles[index],
632 tun->tfiles[tun->numqueues - 1]);
633 ntfile = rtnl_dereference(tun->tfiles[index]);
634 ntfile->queue_index = index;
635
636 --tun->numqueues;
637 if (clean) {
638 RCU_INIT_POINTER(tfile->tun, NULL);
639 sock_put(&tfile->sk);
640 } else
641 tun_disable_queue(tun, tfile);
642
643 synchronize_net();
644 tun_flow_delete_by_queue(tun, tun->numqueues + 1);
645 /* Drop read queue */
646 tun_queue_purge(tfile);
647 tun_set_real_num_queues(tun);
648 } else if (tfile->detached && clean) {
649 tun = tun_enable_queue(tfile);
650 sock_put(&tfile->sk);
651 }
652
653 if (clean) {
654 if (tun && tun->numqueues == 0 && tun->numdisabled == 0) {
655 netif_carrier_off(tun->dev);
656
657 if (!(tun->flags & IFF_PERSIST) &&
658 tun->dev->reg_state == NETREG_REGISTERED)
659 unregister_netdevice(tun->dev);
660 }
661 skb_array_cleanup(&tfile->tx_array);
662 sock_put(&tfile->sk);
663 }
664 }
665
666 static void tun_detach(struct tun_file *tfile, bool clean)
667 {
668 rtnl_lock();
669 __tun_detach(tfile, clean);
670 rtnl_unlock();
671 }
672
673 static void tun_detach_all(struct net_device *dev)
674 {
675 struct tun_struct *tun = netdev_priv(dev);
676 struct bpf_prog *xdp_prog = rtnl_dereference(tun->xdp_prog);
677 struct tun_file *tfile, *tmp;
678 int i, n = tun->numqueues;
679
680 for (i = 0; i < n; i++) {
681 tfile = rtnl_dereference(tun->tfiles[i]);
682 BUG_ON(!tfile);
683 tun_napi_disable(tfile);
684 tfile->socket.sk->sk_shutdown = RCV_SHUTDOWN;
685 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
686 RCU_INIT_POINTER(tfile->tun, NULL);
687 --tun->numqueues;
688 }
689 list_for_each_entry(tfile, &tun->disabled, next) {
690 tfile->socket.sk->sk_shutdown = RCV_SHUTDOWN;
691 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
692 RCU_INIT_POINTER(tfile->tun, NULL);
693 }
694 BUG_ON(tun->numqueues != 0);
695
696 synchronize_net();
697 for (i = 0; i < n; i++) {
698 tfile = rtnl_dereference(tun->tfiles[i]);
699 tun_napi_del(tfile);
700 /* Drop read queue */
701 tun_queue_purge(tfile);
702 sock_put(&tfile->sk);
703 }
704 list_for_each_entry_safe(tfile, tmp, &tun->disabled, next) {
705 tun_enable_queue(tfile);
706 tun_queue_purge(tfile);
707 sock_put(&tfile->sk);
708 }
709 BUG_ON(tun->numdisabled != 0);
710
711 if (xdp_prog)
712 bpf_prog_put(xdp_prog);
713
714 if (tun->flags & IFF_PERSIST)
715 module_put(THIS_MODULE);
716 }
717
718 static int tun_attach(struct tun_struct *tun, struct file *file,
719 bool skip_filter, bool napi, bool napi_frags)
720 {
721 struct tun_file *tfile = file->private_data;
722 struct net_device *dev = tun->dev;
723 int err;
724
725 err = security_tun_dev_attach(tfile->socket.sk, tun->security);
726 if (err < 0)
727 goto out;
728
729 err = -EINVAL;
730 if (rtnl_dereference(tfile->tun) && !tfile->detached)
731 goto out;
732
733 err = -EBUSY;
734 if (!(tun->flags & IFF_MULTI_QUEUE) && tun->numqueues == 1)
735 goto out;
736
737 err = -E2BIG;
738 if (!tfile->detached &&
739 tun->numqueues + tun->numdisabled == MAX_TAP_QUEUES)
740 goto out;
741
742 err = 0;
743
744 /* Re-attach the filter to persist device */
745 if (!skip_filter && (tun->filter_attached == true)) {
746 lock_sock(tfile->socket.sk);
747 err = sk_attach_filter(&tun->fprog, tfile->socket.sk);
748 release_sock(tfile->socket.sk);
749 if (!err)
750 goto out;
751 }
752
753 if (!tfile->detached &&
754 skb_array_resize(&tfile->tx_array, dev->tx_queue_len, GFP_KERNEL)) {
755 err = -ENOMEM;
756 goto out;
757 }
758
759 tfile->queue_index = tun->numqueues;
760 tfile->socket.sk->sk_shutdown &= ~RCV_SHUTDOWN;
761 rcu_assign_pointer(tfile->tun, tun);
762 rcu_assign_pointer(tun->tfiles[tun->numqueues], tfile);
763 tun->numqueues++;
764
765 if (tfile->detached) {
766 tun_enable_queue(tfile);
767 } else {
768 sock_hold(&tfile->sk);
769 tun_napi_init(tun, tfile, napi, napi_frags);
770 }
771
772 tun_set_real_num_queues(tun);
773
774 /* device is allowed to go away first, so no need to hold extra
775 * refcnt.
776 */
777
778 out:
779 return err;
780 }
781
782 static struct tun_struct *tun_get(struct tun_file *tfile)
783 {
784 struct tun_struct *tun;
785
786 rcu_read_lock();
787 tun = rcu_dereference(tfile->tun);
788 if (tun)
789 dev_hold(tun->dev);
790 rcu_read_unlock();
791
792 return tun;
793 }
794
795 static void tun_put(struct tun_struct *tun)
796 {
797 dev_put(tun->dev);
798 }
799
800 /* TAP filtering */
801 static void addr_hash_set(u32 *mask, const u8 *addr)
802 {
803 int n = ether_crc(ETH_ALEN, addr) >> 26;
804 mask[n >> 5] |= (1 << (n & 31));
805 }
806
807 static unsigned int addr_hash_test(const u32 *mask, const u8 *addr)
808 {
809 int n = ether_crc(ETH_ALEN, addr) >> 26;
810 return mask[n >> 5] & (1 << (n & 31));
811 }
812
813 static int update_filter(struct tap_filter *filter, void __user *arg)
814 {
815 struct { u8 u[ETH_ALEN]; } *addr;
816 struct tun_filter uf;
817 int err, alen, n, nexact;
818
819 if (copy_from_user(&uf, arg, sizeof(uf)))
820 return -EFAULT;
821
822 if (!uf.count) {
823 /* Disabled */
824 filter->count = 0;
825 return 0;
826 }
827
828 alen = ETH_ALEN * uf.count;
829 addr = memdup_user(arg + sizeof(uf), alen);
830 if (IS_ERR(addr))
831 return PTR_ERR(addr);
832
833 /* The filter is updated without holding any locks. Which is
834 * perfectly safe. We disable it first and in the worst
835 * case we'll accept a few undesired packets. */
836 filter->count = 0;
837 wmb();
838
839 /* Use first set of addresses as an exact filter */
840 for (n = 0; n < uf.count && n < FLT_EXACT_COUNT; n++)
841 memcpy(filter->addr[n], addr[n].u, ETH_ALEN);
842
843 nexact = n;
844
845 /* Remaining multicast addresses are hashed,
846 * unicast will leave the filter disabled. */
847 memset(filter->mask, 0, sizeof(filter->mask));
848 for (; n < uf.count; n++) {
849 if (!is_multicast_ether_addr(addr[n].u)) {
850 err = 0; /* no filter */
851 goto free_addr;
852 }
853 addr_hash_set(filter->mask, addr[n].u);
854 }
855
856 /* For ALLMULTI just set the mask to all ones.
857 * This overrides the mask populated above. */
858 if ((uf.flags & TUN_FLT_ALLMULTI))
859 memset(filter->mask, ~0, sizeof(filter->mask));
860
861 /* Now enable the filter */
862 wmb();
863 filter->count = nexact;
864
865 /* Return the number of exact filters */
866 err = nexact;
867 free_addr:
868 kfree(addr);
869 return err;
870 }
871
872 /* Returns: 0 - drop, !=0 - accept */
873 static int run_filter(struct tap_filter *filter, const struct sk_buff *skb)
874 {
875 /* Cannot use eth_hdr(skb) here because skb_mac_hdr() is incorrect
876 * at this point. */
877 struct ethhdr *eh = (struct ethhdr *) skb->data;
878 int i;
879
880 /* Exact match */
881 for (i = 0; i < filter->count; i++)
882 if (ether_addr_equal(eh->h_dest, filter->addr[i]))
883 return 1;
884
885 /* Inexact match (multicast only) */
886 if (is_multicast_ether_addr(eh->h_dest))
887 return addr_hash_test(filter->mask, eh->h_dest);
888
889 return 0;
890 }
891
892 /*
893 * Checks whether the packet is accepted or not.
894 * Returns: 0 - drop, !=0 - accept
895 */
896 static int check_filter(struct tap_filter *filter, const struct sk_buff *skb)
897 {
898 if (!filter->count)
899 return 1;
900
901 return run_filter(filter, skb);
902 }
903
904 /* Network device part of the driver */
905
906 static const struct ethtool_ops tun_ethtool_ops;
907
908 /* Net device detach from fd. */
909 static void tun_net_uninit(struct net_device *dev)
910 {
911 tun_detach_all(dev);
912 }
913
914 /* Net device open. */
915 static int tun_net_open(struct net_device *dev)
916 {
917 struct tun_struct *tun = netdev_priv(dev);
918 int i;
919
920 netif_tx_start_all_queues(dev);
921
922 for (i = 0; i < tun->numqueues; i++) {
923 struct tun_file *tfile;
924
925 tfile = rtnl_dereference(tun->tfiles[i]);
926 tfile->socket.sk->sk_write_space(tfile->socket.sk);
927 }
928
929 return 0;
930 }
931
932 /* Net device close. */
933 static int tun_net_close(struct net_device *dev)
934 {
935 netif_tx_stop_all_queues(dev);
936 return 0;
937 }
938
939 /* Net device start xmit */
940 static netdev_tx_t tun_net_xmit(struct sk_buff *skb, struct net_device *dev)
941 {
942 struct tun_struct *tun = netdev_priv(dev);
943 int txq = skb->queue_mapping;
944 struct tun_file *tfile;
945 u32 numqueues = 0;
946
947 rcu_read_lock();
948 tfile = rcu_dereference(tun->tfiles[txq]);
949 numqueues = READ_ONCE(tun->numqueues);
950
951 /* Drop packet if interface is not attached */
952 if (txq >= numqueues)
953 goto drop;
954
955 #ifdef CONFIG_RPS
956 if (numqueues == 1 && static_key_false(&rps_needed)) {
957 /* Select queue was not called for the skbuff, so we extract the
958 * RPS hash and save it into the flow_table here.
959 */
960 __u32 rxhash;
961
962 rxhash = __skb_get_hash_symmetric(skb);
963 if (rxhash) {
964 struct tun_flow_entry *e;
965 e = tun_flow_find(&tun->flows[tun_hashfn(rxhash)],
966 rxhash);
967 if (e)
968 tun_flow_save_rps_rxhash(e, rxhash);
969 }
970 }
971 #endif
972
973 tun_debug(KERN_INFO, tun, "tun_net_xmit %d\n", skb->len);
974
975 BUG_ON(!tfile);
976
977 /* Drop if the filter does not like it.
978 * This is a noop if the filter is disabled.
979 * Filter can be enabled only for the TAP devices. */
980 if (!check_filter(&tun->txflt, skb))
981 goto drop;
982
983 if (tfile->socket.sk->sk_filter &&
984 sk_filter(tfile->socket.sk, skb))
985 goto drop;
986
987 if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC)))
988 goto drop;
989
990 skb_tx_timestamp(skb);
991
992 /* Orphan the skb - required as we might hang on to it
993 * for indefinite time.
994 */
995 skb_orphan(skb);
996
997 nf_reset(skb);
998
999 if (skb_array_produce(&tfile->tx_array, skb))
1000 goto drop;
1001
1002 /* Notify and wake up reader process */
1003 if (tfile->flags & TUN_FASYNC)
1004 kill_fasync(&tfile->fasync, SIGIO, POLL_IN);
1005 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
1006
1007 rcu_read_unlock();
1008 return NETDEV_TX_OK;
1009
1010 drop:
1011 this_cpu_inc(tun->pcpu_stats->tx_dropped);
1012 skb_tx_error(skb);
1013 kfree_skb(skb);
1014 rcu_read_unlock();
1015 return NET_XMIT_DROP;
1016 }
1017
1018 static void tun_net_mclist(struct net_device *dev)
1019 {
1020 /*
1021 * This callback is supposed to deal with mc filter in
1022 * _rx_ path and has nothing to do with the _tx_ path.
1023 * In rx path we always accept everything userspace gives us.
1024 */
1025 }
1026
1027 static netdev_features_t tun_net_fix_features(struct net_device *dev,
1028 netdev_features_t features)
1029 {
1030 struct tun_struct *tun = netdev_priv(dev);
1031
1032 return (features & tun->set_features) | (features & ~TUN_USER_FEATURES);
1033 }
1034 #ifdef CONFIG_NET_POLL_CONTROLLER
1035 static void tun_poll_controller(struct net_device *dev)
1036 {
1037 /*
1038 * Tun only receives frames when:
1039 * 1) the char device endpoint gets data from user space
1040 * 2) the tun socket gets a sendmsg call from user space
1041 * If NAPI is not enabled, since both of those are synchronous
1042 * operations, we are guaranteed never to have pending data when we poll
1043 * for it so there is nothing to do here but return.
1044 * We need this though so netpoll recognizes us as an interface that
1045 * supports polling, which enables bridge devices in virt setups to
1046 * still use netconsole
1047 * If NAPI is enabled, however, we need to schedule polling for all
1048 * queues unless we are using napi_gro_frags(), which we call in
1049 * process context and not in NAPI context.
1050 */
1051 struct tun_struct *tun = netdev_priv(dev);
1052
1053 if (tun->flags & IFF_NAPI) {
1054 struct tun_file *tfile;
1055 int i;
1056
1057 rcu_read_lock();
1058 for (i = 0; i < tun->numqueues; i++) {
1059 tfile = rcu_dereference(tun->tfiles[i]);
1060 if (!tun_napi_frags_enabled(tfile) &&
1061 tfile->napi_enabled)
1062 napi_schedule(&tfile->napi);
1063 }
1064 rcu_read_unlock();
1065 }
1066 return;
1067 }
1068 #endif
1069
1070 static void tun_set_headroom(struct net_device *dev, int new_hr)
1071 {
1072 struct tun_struct *tun = netdev_priv(dev);
1073
1074 if (new_hr < NET_SKB_PAD)
1075 new_hr = NET_SKB_PAD;
1076
1077 tun->align = new_hr;
1078 }
1079
1080 static void
1081 tun_net_get_stats64(struct net_device *dev, struct rtnl_link_stats64 *stats)
1082 {
1083 u32 rx_dropped = 0, tx_dropped = 0, rx_frame_errors = 0;
1084 struct tun_struct *tun = netdev_priv(dev);
1085 struct tun_pcpu_stats *p;
1086 int i;
1087
1088 for_each_possible_cpu(i) {
1089 u64 rxpackets, rxbytes, txpackets, txbytes;
1090 unsigned int start;
1091
1092 p = per_cpu_ptr(tun->pcpu_stats, i);
1093 do {
1094 start = u64_stats_fetch_begin(&p->syncp);
1095 rxpackets = p->rx_packets;
1096 rxbytes = p->rx_bytes;
1097 txpackets = p->tx_packets;
1098 txbytes = p->tx_bytes;
1099 } while (u64_stats_fetch_retry(&p->syncp, start));
1100
1101 stats->rx_packets += rxpackets;
1102 stats->rx_bytes += rxbytes;
1103 stats->tx_packets += txpackets;
1104 stats->tx_bytes += txbytes;
1105
1106 /* u32 counters */
1107 rx_dropped += p->rx_dropped;
1108 rx_frame_errors += p->rx_frame_errors;
1109 tx_dropped += p->tx_dropped;
1110 }
1111 stats->rx_dropped = rx_dropped;
1112 stats->rx_frame_errors = rx_frame_errors;
1113 stats->tx_dropped = tx_dropped;
1114 }
1115
1116 static int tun_xdp_set(struct net_device *dev, struct bpf_prog *prog,
1117 struct netlink_ext_ack *extack)
1118 {
1119 struct tun_struct *tun = netdev_priv(dev);
1120 struct bpf_prog *old_prog;
1121
1122 old_prog = rtnl_dereference(tun->xdp_prog);
1123 rcu_assign_pointer(tun->xdp_prog, prog);
1124 if (old_prog)
1125 bpf_prog_put(old_prog);
1126
1127 return 0;
1128 }
1129
1130 static u32 tun_xdp_query(struct net_device *dev)
1131 {
1132 struct tun_struct *tun = netdev_priv(dev);
1133 const struct bpf_prog *xdp_prog;
1134
1135 xdp_prog = rtnl_dereference(tun->xdp_prog);
1136 if (xdp_prog)
1137 return xdp_prog->aux->id;
1138
1139 return 0;
1140 }
1141
1142 static int tun_xdp(struct net_device *dev, struct netdev_bpf *xdp)
1143 {
1144 switch (xdp->command) {
1145 case XDP_SETUP_PROG:
1146 return tun_xdp_set(dev, xdp->prog, xdp->extack);
1147 case XDP_QUERY_PROG:
1148 xdp->prog_id = tun_xdp_query(dev);
1149 xdp->prog_attached = !!xdp->prog_id;
1150 return 0;
1151 default:
1152 return -EINVAL;
1153 }
1154 }
1155
1156 static int tun_net_change_carrier(struct net_device *dev, bool new_carrier)
1157 {
1158 if (new_carrier) {
1159 struct tun_struct *tun = netdev_priv(dev);
1160
1161 if (!tun->numqueues)
1162 return -EPERM;
1163
1164 netif_carrier_on(dev);
1165 } else {
1166 netif_carrier_off(dev);
1167 }
1168 return 0;
1169 }
1170
1171 static const struct net_device_ops tun_netdev_ops = {
1172 .ndo_uninit = tun_net_uninit,
1173 .ndo_open = tun_net_open,
1174 .ndo_stop = tun_net_close,
1175 .ndo_start_xmit = tun_net_xmit,
1176 .ndo_fix_features = tun_net_fix_features,
1177 .ndo_select_queue = tun_select_queue,
1178 #ifdef CONFIG_NET_POLL_CONTROLLER
1179 .ndo_poll_controller = tun_poll_controller,
1180 #endif
1181 .ndo_set_rx_headroom = tun_set_headroom,
1182 .ndo_get_stats64 = tun_net_get_stats64,
1183 .ndo_change_carrier = tun_net_change_carrier,
1184 };
1185
1186 static const struct net_device_ops tap_netdev_ops = {
1187 .ndo_uninit = tun_net_uninit,
1188 .ndo_open = tun_net_open,
1189 .ndo_stop = tun_net_close,
1190 .ndo_start_xmit = tun_net_xmit,
1191 .ndo_fix_features = tun_net_fix_features,
1192 .ndo_set_rx_mode = tun_net_mclist,
1193 .ndo_set_mac_address = eth_mac_addr,
1194 .ndo_validate_addr = eth_validate_addr,
1195 .ndo_select_queue = tun_select_queue,
1196 #ifdef CONFIG_NET_POLL_CONTROLLER
1197 .ndo_poll_controller = tun_poll_controller,
1198 #endif
1199 .ndo_features_check = passthru_features_check,
1200 .ndo_set_rx_headroom = tun_set_headroom,
1201 .ndo_get_stats64 = tun_net_get_stats64,
1202 .ndo_bpf = tun_xdp,
1203 .ndo_change_carrier = tun_net_change_carrier,
1204 };
1205
1206 static void tun_flow_init(struct tun_struct *tun)
1207 {
1208 int i;
1209
1210 for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++)
1211 INIT_HLIST_HEAD(&tun->flows[i]);
1212
1213 tun->ageing_time = TUN_FLOW_EXPIRE;
1214 timer_setup(&tun->flow_gc_timer, tun_flow_cleanup, 0);
1215 mod_timer(&tun->flow_gc_timer,
1216 round_jiffies_up(jiffies + tun->ageing_time));
1217 }
1218
1219 static void tun_flow_uninit(struct tun_struct *tun)
1220 {
1221 del_timer_sync(&tun->flow_gc_timer);
1222 tun_flow_flush(tun);
1223 }
1224
1225 #define MIN_MTU 68
1226 #define MAX_MTU 65535
1227
1228 /* Initialize net device. */
1229 static void tun_net_init(struct net_device *dev)
1230 {
1231 struct tun_struct *tun = netdev_priv(dev);
1232
1233 switch (tun->flags & TUN_TYPE_MASK) {
1234 case IFF_TUN:
1235 dev->netdev_ops = &tun_netdev_ops;
1236
1237 /* Point-to-Point TUN Device */
1238 dev->hard_header_len = 0;
1239 dev->addr_len = 0;
1240 dev->mtu = 1500;
1241
1242 /* Zero header length */
1243 dev->type = ARPHRD_NONE;
1244 dev->flags = IFF_POINTOPOINT | IFF_NOARP | IFF_MULTICAST;
1245 break;
1246
1247 case IFF_TAP:
1248 dev->netdev_ops = &tap_netdev_ops;
1249 /* Ethernet TAP Device */
1250 ether_setup(dev);
1251 dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1252 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
1253
1254 eth_hw_addr_random(dev);
1255
1256 break;
1257 }
1258
1259 dev->min_mtu = MIN_MTU;
1260 dev->max_mtu = MAX_MTU - dev->hard_header_len;
1261 }
1262
1263 /* Character device part */
1264
1265 /* Poll */
1266 static unsigned int tun_chr_poll(struct file *file, poll_table *wait)
1267 {
1268 struct tun_file *tfile = file->private_data;
1269 struct tun_struct *tun = tun_get(tfile);
1270 struct sock *sk;
1271 unsigned int mask = 0;
1272
1273 if (!tun)
1274 return POLLERR;
1275
1276 sk = tfile->socket.sk;
1277
1278 tun_debug(KERN_INFO, tun, "tun_chr_poll\n");
1279
1280 poll_wait(file, sk_sleep(sk), wait);
1281
1282 if (!skb_array_empty(&tfile->tx_array))
1283 mask |= POLLIN | POLLRDNORM;
1284
1285 if (tun->dev->flags & IFF_UP &&
1286 (sock_writeable(sk) ||
1287 (!test_and_set_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags) &&
1288 sock_writeable(sk))))
1289 mask |= POLLOUT | POLLWRNORM;
1290
1291 if (tun->dev->reg_state != NETREG_REGISTERED)
1292 mask = POLLERR;
1293
1294 tun_put(tun);
1295 return mask;
1296 }
1297
1298 static struct sk_buff *tun_napi_alloc_frags(struct tun_file *tfile,
1299 size_t len,
1300 const struct iov_iter *it)
1301 {
1302 struct sk_buff *skb;
1303 size_t linear;
1304 int err;
1305 int i;
1306
1307 if (it->nr_segs > MAX_SKB_FRAGS + 1)
1308 return ERR_PTR(-ENOMEM);
1309
1310 local_bh_disable();
1311 skb = napi_get_frags(&tfile->napi);
1312 local_bh_enable();
1313 if (!skb)
1314 return ERR_PTR(-ENOMEM);
1315
1316 linear = iov_iter_single_seg_count(it);
1317 err = __skb_grow(skb, linear);
1318 if (err)
1319 goto free;
1320
1321 skb->len = len;
1322 skb->data_len = len - linear;
1323 skb->truesize += skb->data_len;
1324
1325 for (i = 1; i < it->nr_segs; i++) {
1326 struct page_frag *pfrag = &current->task_frag;
1327 size_t fragsz = it->iov[i].iov_len;
1328
1329 if (fragsz == 0 || fragsz > PAGE_SIZE) {
1330 err = -EINVAL;
1331 goto free;
1332 }
1333
1334 if (!skb_page_frag_refill(fragsz, pfrag, GFP_KERNEL)) {
1335 err = -ENOMEM;
1336 goto free;
1337 }
1338
1339 skb_fill_page_desc(skb, i - 1, pfrag->page,
1340 pfrag->offset, fragsz);
1341 page_ref_inc(pfrag->page);
1342 pfrag->offset += fragsz;
1343 }
1344
1345 return skb;
1346 free:
1347 /* frees skb and all frags allocated with napi_alloc_frag() */
1348 napi_free_frags(&tfile->napi);
1349 return ERR_PTR(err);
1350 }
1351
1352 /* prepad is the amount to reserve at front. len is length after that.
1353 * linear is a hint as to how much to copy (usually headers). */
1354 static struct sk_buff *tun_alloc_skb(struct tun_file *tfile,
1355 size_t prepad, size_t len,
1356 size_t linear, int noblock)
1357 {
1358 struct sock *sk = tfile->socket.sk;
1359 struct sk_buff *skb;
1360 int err;
1361
1362 /* Under a page? Don't bother with paged skb. */
1363 if (prepad + len < PAGE_SIZE || !linear)
1364 linear = len;
1365
1366 skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
1367 &err, 0);
1368 if (!skb)
1369 return ERR_PTR(err);
1370
1371 skb_reserve(skb, prepad);
1372 skb_put(skb, linear);
1373 skb->data_len = len - linear;
1374 skb->len += len - linear;
1375
1376 return skb;
1377 }
1378
1379 static void tun_rx_batched(struct tun_struct *tun, struct tun_file *tfile,
1380 struct sk_buff *skb, int more)
1381 {
1382 struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
1383 struct sk_buff_head process_queue;
1384 u32 rx_batched = tun->rx_batched;
1385 bool rcv = false;
1386
1387 if (!rx_batched || (!more && skb_queue_empty(queue))) {
1388 local_bh_disable();
1389 netif_receive_skb(skb);
1390 local_bh_enable();
1391 return;
1392 }
1393
1394 spin_lock(&queue->lock);
1395 if (!more || skb_queue_len(queue) == rx_batched) {
1396 __skb_queue_head_init(&process_queue);
1397 skb_queue_splice_tail_init(queue, &process_queue);
1398 rcv = true;
1399 } else {
1400 __skb_queue_tail(queue, skb);
1401 }
1402 spin_unlock(&queue->lock);
1403
1404 if (rcv) {
1405 struct sk_buff *nskb;
1406
1407 local_bh_disable();
1408 while ((nskb = __skb_dequeue(&process_queue)))
1409 netif_receive_skb(nskb);
1410 netif_receive_skb(skb);
1411 local_bh_enable();
1412 }
1413 }
1414
1415 static bool tun_can_build_skb(struct tun_struct *tun, struct tun_file *tfile,
1416 int len, int noblock, bool zerocopy)
1417 {
1418 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
1419 return false;
1420
1421 if (tfile->socket.sk->sk_sndbuf != INT_MAX)
1422 return false;
1423
1424 if (!noblock)
1425 return false;
1426
1427 if (zerocopy)
1428 return false;
1429
1430 if (SKB_DATA_ALIGN(len + TUN_RX_PAD) +
1431 SKB_DATA_ALIGN(sizeof(struct skb_shared_info)) > PAGE_SIZE)
1432 return false;
1433
1434 return true;
1435 }
1436
1437 static struct sk_buff *tun_build_skb(struct tun_struct *tun,
1438 struct tun_file *tfile,
1439 struct iov_iter *from,
1440 struct virtio_net_hdr *hdr,
1441 int len, int *skb_xdp)
1442 {
1443 struct page_frag *alloc_frag = &current->task_frag;
1444 struct sk_buff *skb;
1445 struct bpf_prog *xdp_prog;
1446 int buflen = SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
1447 unsigned int delta = 0;
1448 char *buf;
1449 size_t copied;
1450 bool xdp_xmit = false;
1451 int err, pad = TUN_RX_PAD;
1452
1453 rcu_read_lock();
1454 xdp_prog = rcu_dereference(tun->xdp_prog);
1455 if (xdp_prog)
1456 pad += TUN_HEADROOM;
1457 buflen += SKB_DATA_ALIGN(len + pad);
1458 rcu_read_unlock();
1459
1460 alloc_frag->offset = ALIGN((u64)alloc_frag->offset, SMP_CACHE_BYTES);
1461 if (unlikely(!skb_page_frag_refill(buflen, alloc_frag, GFP_KERNEL)))
1462 return ERR_PTR(-ENOMEM);
1463
1464 buf = (char *)page_address(alloc_frag->page) + alloc_frag->offset;
1465 copied = copy_page_from_iter(alloc_frag->page,
1466 alloc_frag->offset + pad,
1467 len, from);
1468 if (copied != len)
1469 return ERR_PTR(-EFAULT);
1470
1471 /* There's a small window that XDP may be set after the check
1472 * of xdp_prog above, this should be rare and for simplicity
1473 * we do XDP on skb in case the headroom is not enough.
1474 */
1475 if (hdr->gso_type || !xdp_prog)
1476 *skb_xdp = 1;
1477 else
1478 *skb_xdp = 0;
1479
1480 local_bh_disable();
1481 rcu_read_lock();
1482 xdp_prog = rcu_dereference(tun->xdp_prog);
1483 if (xdp_prog && !*skb_xdp) {
1484 struct xdp_buff xdp;
1485 void *orig_data;
1486 u32 act;
1487
1488 xdp.data_hard_start = buf;
1489 xdp.data = buf + pad;
1490 xdp_set_data_meta_invalid(&xdp);
1491 xdp.data_end = xdp.data + len;
1492 orig_data = xdp.data;
1493 act = bpf_prog_run_xdp(xdp_prog, &xdp);
1494
1495 switch (act) {
1496 case XDP_REDIRECT:
1497 get_page(alloc_frag->page);
1498 alloc_frag->offset += buflen;
1499 err = xdp_do_redirect(tun->dev, &xdp, xdp_prog);
1500 xdp_do_flush_map();
1501 if (err)
1502 goto err_redirect;
1503 rcu_read_unlock();
1504 local_bh_enable();
1505 return NULL;
1506 case XDP_TX:
1507 xdp_xmit = true;
1508 /* fall through */
1509 case XDP_PASS:
1510 delta = orig_data - xdp.data;
1511 break;
1512 default:
1513 bpf_warn_invalid_xdp_action(act);
1514 /* fall through */
1515 case XDP_ABORTED:
1516 trace_xdp_exception(tun->dev, xdp_prog, act);
1517 /* fall through */
1518 case XDP_DROP:
1519 goto err_xdp;
1520 }
1521 }
1522
1523 skb = build_skb(buf, buflen);
1524 if (!skb) {
1525 rcu_read_unlock();
1526 local_bh_enable();
1527 return ERR_PTR(-ENOMEM);
1528 }
1529
1530 skb_reserve(skb, pad - delta);
1531 skb_put(skb, len + delta);
1532 get_page(alloc_frag->page);
1533 alloc_frag->offset += buflen;
1534
1535 if (xdp_xmit) {
1536 skb->dev = tun->dev;
1537 generic_xdp_tx(skb, xdp_prog);
1538 rcu_read_unlock();
1539 local_bh_enable();
1540 return NULL;
1541 }
1542
1543 rcu_read_unlock();
1544 local_bh_enable();
1545
1546 return skb;
1547
1548 err_redirect:
1549 put_page(alloc_frag->page);
1550 err_xdp:
1551 rcu_read_unlock();
1552 local_bh_enable();
1553 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1554 return NULL;
1555 }
1556
1557 /* Get packet from user space buffer */
1558 static ssize_t tun_get_user(struct tun_struct *tun, struct tun_file *tfile,
1559 void *msg_control, struct iov_iter *from,
1560 int noblock, bool more)
1561 {
1562 struct tun_pi pi = { 0, cpu_to_be16(ETH_P_IP) };
1563 struct sk_buff *skb;
1564 size_t total_len = iov_iter_count(from);
1565 size_t len = total_len, align = tun->align, linear;
1566 struct virtio_net_hdr gso = { 0 };
1567 struct tun_pcpu_stats *stats;
1568 int good_linear;
1569 int copylen;
1570 bool zerocopy = false;
1571 int err;
1572 u32 rxhash;
1573 int skb_xdp = 1;
1574 bool frags = tun_napi_frags_enabled(tfile);
1575
1576 if (!(tun->dev->flags & IFF_UP))
1577 return -EIO;
1578
1579 if (!(tun->flags & IFF_NO_PI)) {
1580 if (len < sizeof(pi))
1581 return -EINVAL;
1582 len -= sizeof(pi);
1583
1584 if (!copy_from_iter_full(&pi, sizeof(pi), from))
1585 return -EFAULT;
1586 }
1587
1588 if (tun->flags & IFF_VNET_HDR) {
1589 int vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
1590
1591 if (len < vnet_hdr_sz)
1592 return -EINVAL;
1593 len -= vnet_hdr_sz;
1594
1595 if (!copy_from_iter_full(&gso, sizeof(gso), from))
1596 return -EFAULT;
1597
1598 if ((gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
1599 tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2 > tun16_to_cpu(tun, gso.hdr_len))
1600 gso.hdr_len = cpu_to_tun16(tun, tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2);
1601
1602 if (tun16_to_cpu(tun, gso.hdr_len) > len)
1603 return -EINVAL;
1604 iov_iter_advance(from, vnet_hdr_sz - sizeof(gso));
1605 }
1606
1607 if ((tun->flags & TUN_TYPE_MASK) == IFF_TAP) {
1608 align += NET_IP_ALIGN;
1609 if (unlikely(len < ETH_HLEN ||
1610 (gso.hdr_len && tun16_to_cpu(tun, gso.hdr_len) < ETH_HLEN)))
1611 return -EINVAL;
1612 }
1613
1614 good_linear = SKB_MAX_HEAD(align);
1615
1616 if (msg_control) {
1617 struct iov_iter i = *from;
1618
1619 /* There are 256 bytes to be copied in skb, so there is
1620 * enough room for skb expand head in case it is used.
1621 * The rest of the buffer is mapped from userspace.
1622 */
1623 copylen = gso.hdr_len ? tun16_to_cpu(tun, gso.hdr_len) : GOODCOPY_LEN;
1624 if (copylen > good_linear)
1625 copylen = good_linear;
1626 linear = copylen;
1627 iov_iter_advance(&i, copylen);
1628 if (iov_iter_npages(&i, INT_MAX) <= MAX_SKB_FRAGS)
1629 zerocopy = true;
1630 }
1631
1632 if (!frags && tun_can_build_skb(tun, tfile, len, noblock, zerocopy)) {
1633 /* For the packet that is not easy to be processed
1634 * (e.g gso or jumbo packet), we will do it at after
1635 * skb was created with generic XDP routine.
1636 */
1637 skb = tun_build_skb(tun, tfile, from, &gso, len, &skb_xdp);
1638 if (IS_ERR(skb)) {
1639 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1640 return PTR_ERR(skb);
1641 }
1642 if (!skb)
1643 return total_len;
1644 } else {
1645 if (!zerocopy) {
1646 copylen = len;
1647 if (tun16_to_cpu(tun, gso.hdr_len) > good_linear)
1648 linear = good_linear;
1649 else
1650 linear = tun16_to_cpu(tun, gso.hdr_len);
1651 }
1652
1653 if (frags) {
1654 mutex_lock(&tfile->napi_mutex);
1655 skb = tun_napi_alloc_frags(tfile, copylen, from);
1656 /* tun_napi_alloc_frags() enforces a layout for the skb.
1657 * If zerocopy is enabled, then this layout will be
1658 * overwritten by zerocopy_sg_from_iter().
1659 */
1660 zerocopy = false;
1661 } else {
1662 skb = tun_alloc_skb(tfile, align, copylen, linear,
1663 noblock);
1664 }
1665
1666 if (IS_ERR(skb)) {
1667 if (PTR_ERR(skb) != -EAGAIN)
1668 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1669 if (frags)
1670 mutex_unlock(&tfile->napi_mutex);
1671 return PTR_ERR(skb);
1672 }
1673
1674 if (zerocopy)
1675 err = zerocopy_sg_from_iter(skb, from);
1676 else
1677 err = skb_copy_datagram_from_iter(skb, 0, from, len);
1678
1679 if (err) {
1680 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1681 kfree_skb(skb);
1682 if (frags) {
1683 tfile->napi.skb = NULL;
1684 mutex_unlock(&tfile->napi_mutex);
1685 }
1686
1687 return -EFAULT;
1688 }
1689 }
1690
1691 if (virtio_net_hdr_to_skb(skb, &gso, tun_is_little_endian(tun))) {
1692 this_cpu_inc(tun->pcpu_stats->rx_frame_errors);
1693 kfree_skb(skb);
1694 if (frags) {
1695 tfile->napi.skb = NULL;
1696 mutex_unlock(&tfile->napi_mutex);
1697 }
1698
1699 return -EINVAL;
1700 }
1701
1702 switch (tun->flags & TUN_TYPE_MASK) {
1703 case IFF_TUN:
1704 if (tun->flags & IFF_NO_PI) {
1705 u8 ip_version = skb->len ? (skb->data[0] >> 4) : 0;
1706
1707 switch (ip_version) {
1708 case 4:
1709 pi.proto = htons(ETH_P_IP);
1710 break;
1711 case 6:
1712 pi.proto = htons(ETH_P_IPV6);
1713 break;
1714 default:
1715 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1716 kfree_skb(skb);
1717 return -EINVAL;
1718 }
1719 }
1720
1721 skb_reset_mac_header(skb);
1722 skb->protocol = pi.proto;
1723 skb->dev = tun->dev;
1724 break;
1725 case IFF_TAP:
1726 if (!frags)
1727 skb->protocol = eth_type_trans(skb, tun->dev);
1728 break;
1729 }
1730
1731 /* copy skb_ubuf_info for callback when skb has no error */
1732 if (zerocopy) {
1733 skb_shinfo(skb)->destructor_arg = msg_control;
1734 skb_shinfo(skb)->tx_flags |= SKBTX_DEV_ZEROCOPY;
1735 skb_shinfo(skb)->tx_flags |= SKBTX_SHARED_FRAG;
1736 } else if (msg_control) {
1737 struct ubuf_info *uarg = msg_control;
1738 uarg->callback(uarg, false);
1739 }
1740
1741 skb_reset_network_header(skb);
1742 skb_probe_transport_header(skb, 0);
1743
1744 if (skb_xdp) {
1745 struct bpf_prog *xdp_prog;
1746 int ret;
1747
1748 local_bh_disable();
1749 rcu_read_lock();
1750 xdp_prog = rcu_dereference(tun->xdp_prog);
1751 if (xdp_prog) {
1752 ret = do_xdp_generic(xdp_prog, skb);
1753 if (ret != XDP_PASS) {
1754 rcu_read_unlock();
1755 local_bh_enable();
1756 return total_len;
1757 }
1758 }
1759 rcu_read_unlock();
1760 local_bh_enable();
1761 }
1762
1763 rxhash = __skb_get_hash_symmetric(skb);
1764
1765 if (frags) {
1766 /* Exercise flow dissector code path. */
1767 u32 headlen = eth_get_headlen(skb->data, skb_headlen(skb));
1768
1769 if (unlikely(headlen > skb_headlen(skb))) {
1770 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1771 napi_free_frags(&tfile->napi);
1772 mutex_unlock(&tfile->napi_mutex);
1773 WARN_ON(1);
1774 return -ENOMEM;
1775 }
1776
1777 local_bh_disable();
1778 napi_gro_frags(&tfile->napi);
1779 local_bh_enable();
1780 mutex_unlock(&tfile->napi_mutex);
1781 } else if (tfile->napi_enabled) {
1782 struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
1783 int queue_len;
1784
1785 spin_lock_bh(&queue->lock);
1786 __skb_queue_tail(queue, skb);
1787 queue_len = skb_queue_len(queue);
1788 spin_unlock(&queue->lock);
1789
1790 if (!more || queue_len > NAPI_POLL_WEIGHT)
1791 napi_schedule(&tfile->napi);
1792
1793 local_bh_enable();
1794 } else if (!IS_ENABLED(CONFIG_4KSTACKS)) {
1795 tun_rx_batched(tun, tfile, skb, more);
1796 } else {
1797 netif_rx_ni(skb);
1798 }
1799
1800 stats = get_cpu_ptr(tun->pcpu_stats);
1801 u64_stats_update_begin(&stats->syncp);
1802 stats->rx_packets++;
1803 stats->rx_bytes += len;
1804 u64_stats_update_end(&stats->syncp);
1805 put_cpu_ptr(stats);
1806
1807 tun_flow_update(tun, rxhash, tfile);
1808 return total_len;
1809 }
1810
1811 static ssize_t tun_chr_write_iter(struct kiocb *iocb, struct iov_iter *from)
1812 {
1813 struct file *file = iocb->ki_filp;
1814 struct tun_file *tfile = file->private_data;
1815 struct tun_struct *tun = tun_get(tfile);
1816 ssize_t result;
1817
1818 if (!tun)
1819 return -EBADFD;
1820
1821 result = tun_get_user(tun, tfile, NULL, from,
1822 file->f_flags & O_NONBLOCK, false);
1823
1824 tun_put(tun);
1825 return result;
1826 }
1827
1828 /* Put packet to the user space buffer */
1829 static ssize_t tun_put_user(struct tun_struct *tun,
1830 struct tun_file *tfile,
1831 struct sk_buff *skb,
1832 struct iov_iter *iter)
1833 {
1834 struct tun_pi pi = { 0, skb->protocol };
1835 struct tun_pcpu_stats *stats;
1836 ssize_t total;
1837 int vlan_offset = 0;
1838 int vlan_hlen = 0;
1839 int vnet_hdr_sz = 0;
1840
1841 if (skb_vlan_tag_present(skb))
1842 vlan_hlen = VLAN_HLEN;
1843
1844 if (tun->flags & IFF_VNET_HDR)
1845 vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
1846
1847 total = skb->len + vlan_hlen + vnet_hdr_sz;
1848
1849 if (!(tun->flags & IFF_NO_PI)) {
1850 if (iov_iter_count(iter) < sizeof(pi))
1851 return -EINVAL;
1852
1853 total += sizeof(pi);
1854 if (iov_iter_count(iter) < total) {
1855 /* Packet will be striped */
1856 pi.flags |= TUN_PKT_STRIP;
1857 }
1858
1859 if (copy_to_iter(&pi, sizeof(pi), iter) != sizeof(pi))
1860 return -EFAULT;
1861 }
1862
1863 if (vnet_hdr_sz) {
1864 struct virtio_net_hdr gso;
1865
1866 if (iov_iter_count(iter) < vnet_hdr_sz)
1867 return -EINVAL;
1868
1869 if (virtio_net_hdr_from_skb(skb, &gso,
1870 tun_is_little_endian(tun), true,
1871 vlan_hlen)) {
1872 struct skb_shared_info *sinfo = skb_shinfo(skb);
1873 pr_err("unexpected GSO type: "
1874 "0x%x, gso_size %d, hdr_len %d\n",
1875 sinfo->gso_type, tun16_to_cpu(tun, gso.gso_size),
1876 tun16_to_cpu(tun, gso.hdr_len));
1877 print_hex_dump(KERN_ERR, "tun: ",
1878 DUMP_PREFIX_NONE,
1879 16, 1, skb->head,
1880 min((int)tun16_to_cpu(tun, gso.hdr_len), 64), true);
1881 WARN_ON_ONCE(1);
1882 return -EINVAL;
1883 }
1884
1885 if (copy_to_iter(&gso, sizeof(gso), iter) != sizeof(gso))
1886 return -EFAULT;
1887
1888 iov_iter_advance(iter, vnet_hdr_sz - sizeof(gso));
1889 }
1890
1891 if (vlan_hlen) {
1892 int ret;
1893 struct {
1894 __be16 h_vlan_proto;
1895 __be16 h_vlan_TCI;
1896 } veth;
1897
1898 veth.h_vlan_proto = skb->vlan_proto;
1899 veth.h_vlan_TCI = htons(skb_vlan_tag_get(skb));
1900
1901 vlan_offset = offsetof(struct vlan_ethhdr, h_vlan_proto);
1902
1903 ret = skb_copy_datagram_iter(skb, 0, iter, vlan_offset);
1904 if (ret || !iov_iter_count(iter))
1905 goto done;
1906
1907 ret = copy_to_iter(&veth, sizeof(veth), iter);
1908 if (ret != sizeof(veth) || !iov_iter_count(iter))
1909 goto done;
1910 }
1911
1912 skb_copy_datagram_iter(skb, vlan_offset, iter, skb->len - vlan_offset);
1913
1914 done:
1915 /* caller is in process context, */
1916 stats = get_cpu_ptr(tun->pcpu_stats);
1917 u64_stats_update_begin(&stats->syncp);
1918 stats->tx_packets++;
1919 stats->tx_bytes += skb->len + vlan_hlen;
1920 u64_stats_update_end(&stats->syncp);
1921 put_cpu_ptr(tun->pcpu_stats);
1922
1923 return total;
1924 }
1925
1926 static struct sk_buff *tun_ring_recv(struct tun_file *tfile, int noblock,
1927 int *err)
1928 {
1929 DECLARE_WAITQUEUE(wait, current);
1930 struct sk_buff *skb = NULL;
1931 int error = 0;
1932
1933 skb = skb_array_consume(&tfile->tx_array);
1934 if (skb)
1935 goto out;
1936 if (noblock) {
1937 error = -EAGAIN;
1938 goto out;
1939 }
1940
1941 add_wait_queue(&tfile->wq.wait, &wait);
1942 current->state = TASK_INTERRUPTIBLE;
1943
1944 while (1) {
1945 skb = skb_array_consume(&tfile->tx_array);
1946 if (skb)
1947 break;
1948 if (signal_pending(current)) {
1949 error = -ERESTARTSYS;
1950 break;
1951 }
1952 if (tfile->socket.sk->sk_shutdown & RCV_SHUTDOWN) {
1953 error = -EFAULT;
1954 break;
1955 }
1956
1957 schedule();
1958 }
1959
1960 current->state = TASK_RUNNING;
1961 remove_wait_queue(&tfile->wq.wait, &wait);
1962
1963 out:
1964 *err = error;
1965 return skb;
1966 }
1967
1968 static ssize_t tun_do_read(struct tun_struct *tun, struct tun_file *tfile,
1969 struct iov_iter *to,
1970 int noblock, struct sk_buff *skb)
1971 {
1972 ssize_t ret;
1973 int err;
1974
1975 tun_debug(KERN_INFO, tun, "tun_do_read\n");
1976
1977 if (!iov_iter_count(to)) {
1978 if (skb)
1979 kfree_skb(skb);
1980 return 0;
1981 }
1982
1983 if (!skb) {
1984 /* Read frames from ring */
1985 skb = tun_ring_recv(tfile, noblock, &err);
1986 if (!skb)
1987 return err;
1988 }
1989
1990 ret = tun_put_user(tun, tfile, skb, to);
1991 if (unlikely(ret < 0))
1992 kfree_skb(skb);
1993 else
1994 consume_skb(skb);
1995
1996 return ret;
1997 }
1998
1999 static ssize_t tun_chr_read_iter(struct kiocb *iocb, struct iov_iter *to)
2000 {
2001 struct file *file = iocb->ki_filp;
2002 struct tun_file *tfile = file->private_data;
2003 struct tun_struct *tun = tun_get(tfile);
2004 ssize_t len = iov_iter_count(to), ret;
2005
2006 if (!tun)
2007 return -EBADFD;
2008 ret = tun_do_read(tun, tfile, to, file->f_flags & O_NONBLOCK, NULL);
2009 ret = min_t(ssize_t, ret, len);
2010 if (ret > 0)
2011 iocb->ki_pos = ret;
2012 tun_put(tun);
2013 return ret;
2014 }
2015
2016 static void tun_free_netdev(struct net_device *dev)
2017 {
2018 struct tun_struct *tun = netdev_priv(dev);
2019
2020 BUG_ON(!(list_empty(&tun->disabled)));
2021 free_percpu(tun->pcpu_stats);
2022 tun_flow_uninit(tun);
2023 security_tun_dev_free_security(tun->security);
2024 }
2025
2026 static void tun_setup(struct net_device *dev)
2027 {
2028 struct tun_struct *tun = netdev_priv(dev);
2029
2030 tun->owner = INVALID_UID;
2031 tun->group = INVALID_GID;
2032
2033 dev->ethtool_ops = &tun_ethtool_ops;
2034 dev->needs_free_netdev = true;
2035 dev->priv_destructor = tun_free_netdev;
2036 /* We prefer our own queue length */
2037 dev->tx_queue_len = TUN_READQ_SIZE;
2038 }
2039
2040 /* Trivial set of netlink ops to allow deleting tun or tap
2041 * device with netlink.
2042 */
2043 static int tun_validate(struct nlattr *tb[], struct nlattr *data[],
2044 struct netlink_ext_ack *extack)
2045 {
2046 return -EINVAL;
2047 }
2048
2049 static struct rtnl_link_ops tun_link_ops __read_mostly = {
2050 .kind = DRV_NAME,
2051 .priv_size = sizeof(struct tun_struct),
2052 .setup = tun_setup,
2053 .validate = tun_validate,
2054 };
2055
2056 static void tun_sock_write_space(struct sock *sk)
2057 {
2058 struct tun_file *tfile;
2059 wait_queue_head_t *wqueue;
2060
2061 if (!sock_writeable(sk))
2062 return;
2063
2064 if (!test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags))
2065 return;
2066
2067 wqueue = sk_sleep(sk);
2068 if (wqueue && waitqueue_active(wqueue))
2069 wake_up_interruptible_sync_poll(wqueue, POLLOUT |
2070 POLLWRNORM | POLLWRBAND);
2071
2072 tfile = container_of(sk, struct tun_file, sk);
2073 kill_fasync(&tfile->fasync, SIGIO, POLL_OUT);
2074 }
2075
2076 static int tun_sendmsg(struct socket *sock, struct msghdr *m, size_t total_len)
2077 {
2078 int ret;
2079 struct tun_file *tfile = container_of(sock, struct tun_file, socket);
2080 struct tun_struct *tun = tun_get(tfile);
2081
2082 if (!tun)
2083 return -EBADFD;
2084
2085 ret = tun_get_user(tun, tfile, m->msg_control, &m->msg_iter,
2086 m->msg_flags & MSG_DONTWAIT,
2087 m->msg_flags & MSG_MORE);
2088 tun_put(tun);
2089 return ret;
2090 }
2091
2092 static int tun_recvmsg(struct socket *sock, struct msghdr *m, size_t total_len,
2093 int flags)
2094 {
2095 struct tun_file *tfile = container_of(sock, struct tun_file, socket);
2096 struct tun_struct *tun = tun_get(tfile);
2097 struct sk_buff *skb = m->msg_control;
2098 int ret;
2099
2100 if (!tun) {
2101 ret = -EBADFD;
2102 goto out_free_skb;
2103 }
2104
2105 if (flags & ~(MSG_DONTWAIT|MSG_TRUNC|MSG_ERRQUEUE)) {
2106 ret = -EINVAL;
2107 goto out_put_tun;
2108 }
2109 if (flags & MSG_ERRQUEUE) {
2110 ret = sock_recv_errqueue(sock->sk, m, total_len,
2111 SOL_PACKET, TUN_TX_TIMESTAMP);
2112 goto out;
2113 }
2114 ret = tun_do_read(tun, tfile, &m->msg_iter, flags & MSG_DONTWAIT, skb);
2115 if (ret > (ssize_t)total_len) {
2116 m->msg_flags |= MSG_TRUNC;
2117 ret = flags & MSG_TRUNC ? ret : total_len;
2118 }
2119 out:
2120 tun_put(tun);
2121 return ret;
2122
2123 out_put_tun:
2124 tun_put(tun);
2125 out_free_skb:
2126 if (skb)
2127 kfree_skb(skb);
2128 return ret;
2129 }
2130
2131 static int tun_peek_len(struct socket *sock)
2132 {
2133 struct tun_file *tfile = container_of(sock, struct tun_file, socket);
2134 struct tun_struct *tun;
2135 int ret = 0;
2136
2137 tun = tun_get(tfile);
2138 if (!tun)
2139 return 0;
2140
2141 ret = skb_array_peek_len(&tfile->tx_array);
2142 tun_put(tun);
2143
2144 return ret;
2145 }
2146
2147 /* Ops structure to mimic raw sockets with tun */
2148 static const struct proto_ops tun_socket_ops = {
2149 .peek_len = tun_peek_len,
2150 .sendmsg = tun_sendmsg,
2151 .recvmsg = tun_recvmsg,
2152 };
2153
2154 static struct proto tun_proto = {
2155 .name = "tun",
2156 .owner = THIS_MODULE,
2157 .obj_size = sizeof(struct tun_file),
2158 };
2159
2160 static int tun_flags(struct tun_struct *tun)
2161 {
2162 return tun->flags & (TUN_FEATURES | IFF_PERSIST | IFF_TUN | IFF_TAP);
2163 }
2164
2165 static ssize_t tun_show_flags(struct device *dev, struct device_attribute *attr,
2166 char *buf)
2167 {
2168 struct tun_struct *tun = netdev_priv(to_net_dev(dev));
2169 return sprintf(buf, "0x%x\n", tun_flags(tun));
2170 }
2171
2172 static ssize_t tun_show_owner(struct device *dev, struct device_attribute *attr,
2173 char *buf)
2174 {
2175 struct tun_struct *tun = netdev_priv(to_net_dev(dev));
2176 return uid_valid(tun->owner)?
2177 sprintf(buf, "%u\n",
2178 from_kuid_munged(current_user_ns(), tun->owner)):
2179 sprintf(buf, "-1\n");
2180 }
2181
2182 static ssize_t tun_show_group(struct device *dev, struct device_attribute *attr,
2183 char *buf)
2184 {
2185 struct tun_struct *tun = netdev_priv(to_net_dev(dev));
2186 return gid_valid(tun->group) ?
2187 sprintf(buf, "%u\n",
2188 from_kgid_munged(current_user_ns(), tun->group)):
2189 sprintf(buf, "-1\n");
2190 }
2191
2192 static DEVICE_ATTR(tun_flags, 0444, tun_show_flags, NULL);
2193 static DEVICE_ATTR(owner, 0444, tun_show_owner, NULL);
2194 static DEVICE_ATTR(group, 0444, tun_show_group, NULL);
2195
2196 static struct attribute *tun_dev_attrs[] = {
2197 &dev_attr_tun_flags.attr,
2198 &dev_attr_owner.attr,
2199 &dev_attr_group.attr,
2200 NULL
2201 };
2202
2203 static const struct attribute_group tun_attr_group = {
2204 .attrs = tun_dev_attrs
2205 };
2206
2207 static int tun_set_iff(struct net *net, struct file *file, struct ifreq *ifr)
2208 {
2209 struct tun_struct *tun;
2210 struct tun_file *tfile = file->private_data;
2211 struct net_device *dev;
2212 int err;
2213
2214 if (tfile->detached)
2215 return -EINVAL;
2216
2217 if ((ifr->ifr_flags & IFF_NAPI_FRAGS)) {
2218 if (!capable(CAP_NET_ADMIN))
2219 return -EPERM;
2220
2221 if (!(ifr->ifr_flags & IFF_NAPI) ||
2222 (ifr->ifr_flags & TUN_TYPE_MASK) != IFF_TAP)
2223 return -EINVAL;
2224 }
2225
2226 dev = __dev_get_by_name(net, ifr->ifr_name);
2227 if (dev) {
2228 if (ifr->ifr_flags & IFF_TUN_EXCL)
2229 return -EBUSY;
2230 if ((ifr->ifr_flags & IFF_TUN) && dev->netdev_ops == &tun_netdev_ops)
2231 tun = netdev_priv(dev);
2232 else if ((ifr->ifr_flags & IFF_TAP) && dev->netdev_ops == &tap_netdev_ops)
2233 tun = netdev_priv(dev);
2234 else
2235 return -EINVAL;
2236
2237 if (!!(ifr->ifr_flags & IFF_MULTI_QUEUE) !=
2238 !!(tun->flags & IFF_MULTI_QUEUE))
2239 return -EINVAL;
2240
2241 if (tun_not_capable(tun))
2242 return -EPERM;
2243 err = security_tun_dev_open(tun->security);
2244 if (err < 0)
2245 return err;
2246
2247 err = tun_attach(tun, file, ifr->ifr_flags & IFF_NOFILTER,
2248 ifr->ifr_flags & IFF_NAPI,
2249 ifr->ifr_flags & IFF_NAPI_FRAGS);
2250 if (err < 0)
2251 return err;
2252
2253 if (tun->flags & IFF_MULTI_QUEUE &&
2254 (tun->numqueues + tun->numdisabled > 1)) {
2255 /* One or more queue has already been attached, no need
2256 * to initialize the device again.
2257 */
2258 return 0;
2259 }
2260 }
2261 else {
2262 char *name;
2263 unsigned long flags = 0;
2264 int queues = ifr->ifr_flags & IFF_MULTI_QUEUE ?
2265 MAX_TAP_QUEUES : 1;
2266
2267 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
2268 return -EPERM;
2269 err = security_tun_dev_create();
2270 if (err < 0)
2271 return err;
2272
2273 /* Set dev type */
2274 if (ifr->ifr_flags & IFF_TUN) {
2275 /* TUN device */
2276 flags |= IFF_TUN;
2277 name = "tun%d";
2278 } else if (ifr->ifr_flags & IFF_TAP) {
2279 /* TAP device */
2280 flags |= IFF_TAP;
2281 name = "tap%d";
2282 } else
2283 return -EINVAL;
2284
2285 if (*ifr->ifr_name)
2286 name = ifr->ifr_name;
2287
2288 dev = alloc_netdev_mqs(sizeof(struct tun_struct), name,
2289 NET_NAME_UNKNOWN, tun_setup, queues,
2290 queues);
2291
2292 if (!dev)
2293 return -ENOMEM;
2294 err = dev_get_valid_name(net, dev, name);
2295 if (err < 0)
2296 goto err_free_dev;
2297
2298 dev_net_set(dev, net);
2299 dev->rtnl_link_ops = &tun_link_ops;
2300 dev->ifindex = tfile->ifindex;
2301 dev->sysfs_groups[0] = &tun_attr_group;
2302
2303 tun = netdev_priv(dev);
2304 tun->dev = dev;
2305 tun->flags = flags;
2306 tun->txflt.count = 0;
2307 tun->vnet_hdr_sz = sizeof(struct virtio_net_hdr);
2308
2309 tun->align = NET_SKB_PAD;
2310 tun->filter_attached = false;
2311 tun->sndbuf = tfile->socket.sk->sk_sndbuf;
2312 tun->rx_batched = 0;
2313
2314 tun->pcpu_stats = netdev_alloc_pcpu_stats(struct tun_pcpu_stats);
2315 if (!tun->pcpu_stats) {
2316 err = -ENOMEM;
2317 goto err_free_dev;
2318 }
2319
2320 spin_lock_init(&tun->lock);
2321
2322 err = security_tun_dev_alloc_security(&tun->security);
2323 if (err < 0)
2324 goto err_free_stat;
2325
2326 tun_net_init(dev);
2327 tun_flow_init(tun);
2328
2329 dev->hw_features = NETIF_F_SG | NETIF_F_FRAGLIST |
2330 TUN_USER_FEATURES | NETIF_F_HW_VLAN_CTAG_TX |
2331 NETIF_F_HW_VLAN_STAG_TX;
2332 dev->features = dev->hw_features | NETIF_F_LLTX;
2333 dev->vlan_features = dev->features &
2334 ~(NETIF_F_HW_VLAN_CTAG_TX |
2335 NETIF_F_HW_VLAN_STAG_TX);
2336
2337 INIT_LIST_HEAD(&tun->disabled);
2338 err = tun_attach(tun, file, false, ifr->ifr_flags & IFF_NAPI,
2339 ifr->ifr_flags & IFF_NAPI_FRAGS);
2340 if (err < 0)
2341 goto err_free_flow;
2342
2343 err = register_netdevice(tun->dev);
2344 if (err < 0)
2345 goto err_detach;
2346 }
2347
2348 netif_carrier_on(tun->dev);
2349
2350 tun_debug(KERN_INFO, tun, "tun_set_iff\n");
2351
2352 tun->flags = (tun->flags & ~TUN_FEATURES) |
2353 (ifr->ifr_flags & TUN_FEATURES);
2354
2355 /* Make sure persistent devices do not get stuck in
2356 * xoff state.
2357 */
2358 if (netif_running(tun->dev))
2359 netif_tx_wake_all_queues(tun->dev);
2360
2361 strcpy(ifr->ifr_name, tun->dev->name);
2362 return 0;
2363
2364 err_detach:
2365 tun_detach_all(dev);
2366 /* register_netdevice() already called tun_free_netdev() */
2367 goto err_free_dev;
2368
2369 err_free_flow:
2370 tun_flow_uninit(tun);
2371 security_tun_dev_free_security(tun->security);
2372 err_free_stat:
2373 free_percpu(tun->pcpu_stats);
2374 err_free_dev:
2375 free_netdev(dev);
2376 return err;
2377 }
2378
2379 static void tun_get_iff(struct net *net, struct tun_struct *tun,
2380 struct ifreq *ifr)
2381 {
2382 tun_debug(KERN_INFO, tun, "tun_get_iff\n");
2383
2384 strcpy(ifr->ifr_name, tun->dev->name);
2385
2386 ifr->ifr_flags = tun_flags(tun);
2387
2388 }
2389
2390 /* This is like a cut-down ethtool ops, except done via tun fd so no
2391 * privs required. */
2392 static int set_offload(struct tun_struct *tun, unsigned long arg)
2393 {
2394 netdev_features_t features = 0;
2395
2396 if (arg & TUN_F_CSUM) {
2397 features |= NETIF_F_HW_CSUM;
2398 arg &= ~TUN_F_CSUM;
2399
2400 if (arg & (TUN_F_TSO4|TUN_F_TSO6)) {
2401 if (arg & TUN_F_TSO_ECN) {
2402 features |= NETIF_F_TSO_ECN;
2403 arg &= ~TUN_F_TSO_ECN;
2404 }
2405 if (arg & TUN_F_TSO4)
2406 features |= NETIF_F_TSO;
2407 if (arg & TUN_F_TSO6)
2408 features |= NETIF_F_TSO6;
2409 arg &= ~(TUN_F_TSO4|TUN_F_TSO6);
2410 }
2411
2412 arg &= ~TUN_F_UFO;
2413 }
2414
2415 /* This gives the user a way to test for new features in future by
2416 * trying to set them. */
2417 if (arg)
2418 return -EINVAL;
2419
2420 tun->set_features = features;
2421 tun->dev->wanted_features &= ~TUN_USER_FEATURES;
2422 tun->dev->wanted_features |= features;
2423 netdev_update_features(tun->dev);
2424
2425 return 0;
2426 }
2427
2428 static void tun_detach_filter(struct tun_struct *tun, int n)
2429 {
2430 int i;
2431 struct tun_file *tfile;
2432
2433 for (i = 0; i < n; i++) {
2434 tfile = rtnl_dereference(tun->tfiles[i]);
2435 lock_sock(tfile->socket.sk);
2436 sk_detach_filter(tfile->socket.sk);
2437 release_sock(tfile->socket.sk);
2438 }
2439
2440 tun->filter_attached = false;
2441 }
2442
2443 static int tun_attach_filter(struct tun_struct *tun)
2444 {
2445 int i, ret = 0;
2446 struct tun_file *tfile;
2447
2448 for (i = 0; i < tun->numqueues; i++) {
2449 tfile = rtnl_dereference(tun->tfiles[i]);
2450 lock_sock(tfile->socket.sk);
2451 ret = sk_attach_filter(&tun->fprog, tfile->socket.sk);
2452 release_sock(tfile->socket.sk);
2453 if (ret) {
2454 tun_detach_filter(tun, i);
2455 return ret;
2456 }
2457 }
2458
2459 tun->filter_attached = true;
2460 return ret;
2461 }
2462
2463 static void tun_set_sndbuf(struct tun_struct *tun)
2464 {
2465 struct tun_file *tfile;
2466 int i;
2467
2468 for (i = 0; i < tun->numqueues; i++) {
2469 tfile = rtnl_dereference(tun->tfiles[i]);
2470 tfile->socket.sk->sk_sndbuf = tun->sndbuf;
2471 }
2472 }
2473
2474 static int tun_set_queue(struct file *file, struct ifreq *ifr)
2475 {
2476 struct tun_file *tfile = file->private_data;
2477 struct tun_struct *tun;
2478 int ret = 0;
2479
2480 rtnl_lock();
2481
2482 if (ifr->ifr_flags & IFF_ATTACH_QUEUE) {
2483 tun = tfile->detached;
2484 if (!tun) {
2485 ret = -EINVAL;
2486 goto unlock;
2487 }
2488 ret = security_tun_dev_attach_queue(tun->security);
2489 if (ret < 0)
2490 goto unlock;
2491 ret = tun_attach(tun, file, false, tun->flags & IFF_NAPI,
2492 tun->flags & IFF_NAPI_FRAGS);
2493 } else if (ifr->ifr_flags & IFF_DETACH_QUEUE) {
2494 tun = rtnl_dereference(tfile->tun);
2495 if (!tun || !(tun->flags & IFF_MULTI_QUEUE) || tfile->detached)
2496 ret = -EINVAL;
2497 else
2498 __tun_detach(tfile, false);
2499 } else
2500 ret = -EINVAL;
2501
2502 unlock:
2503 rtnl_unlock();
2504 return ret;
2505 }
2506
2507 static long __tun_chr_ioctl(struct file *file, unsigned int cmd,
2508 unsigned long arg, int ifreq_len)
2509 {
2510 struct tun_file *tfile = file->private_data;
2511 struct tun_struct *tun;
2512 void __user* argp = (void __user*)arg;
2513 unsigned int ifindex, carrier;
2514 struct ifreq ifr;
2515 kuid_t owner;
2516 kgid_t group;
2517 int sndbuf;
2518 int vnet_hdr_sz;
2519 int le;
2520 int ret;
2521
2522 if (cmd == TUNSETIFF || cmd == TUNSETQUEUE || _IOC_TYPE(cmd) == SOCK_IOC_TYPE) {
2523 if (copy_from_user(&ifr, argp, ifreq_len))
2524 return -EFAULT;
2525 } else {
2526 memset(&ifr, 0, sizeof(ifr));
2527 }
2528 if (cmd == TUNGETFEATURES) {
2529 /* Currently this just means: "what IFF flags are valid?".
2530 * This is needed because we never checked for invalid flags on
2531 * TUNSETIFF.
2532 */
2533 return put_user(IFF_TUN | IFF_TAP | TUN_FEATURES,
2534 (unsigned int __user*)argp);
2535 } else if (cmd == TUNSETQUEUE)
2536 return tun_set_queue(file, &ifr);
2537
2538 ret = 0;
2539 rtnl_lock();
2540
2541 tun = tun_get(tfile);
2542 if (cmd == TUNSETIFF) {
2543 ret = -EEXIST;
2544 if (tun)
2545 goto unlock;
2546
2547 ifr.ifr_name[IFNAMSIZ-1] = '\0';
2548
2549 ret = tun_set_iff(sock_net(&tfile->sk), file, &ifr);
2550
2551 if (ret)
2552 goto unlock;
2553
2554 if (copy_to_user(argp, &ifr, ifreq_len))
2555 ret = -EFAULT;
2556 goto unlock;
2557 }
2558 if (cmd == TUNSETIFINDEX) {
2559 ret = -EPERM;
2560 if (tun)
2561 goto unlock;
2562
2563 ret = -EFAULT;
2564 if (copy_from_user(&ifindex, argp, sizeof(ifindex)))
2565 goto unlock;
2566
2567 ret = 0;
2568 tfile->ifindex = ifindex;
2569 goto unlock;
2570 }
2571
2572 ret = -EBADFD;
2573 if (!tun)
2574 goto unlock;
2575
2576 tun_debug(KERN_INFO, tun, "tun_chr_ioctl cmd %u\n", cmd);
2577
2578 ret = 0;
2579 switch (cmd) {
2580 case TUNGETIFF:
2581 tun_get_iff(current->nsproxy->net_ns, tun, &ifr);
2582
2583 if (tfile->detached)
2584 ifr.ifr_flags |= IFF_DETACH_QUEUE;
2585 if (!tfile->socket.sk->sk_filter)
2586 ifr.ifr_flags |= IFF_NOFILTER;
2587
2588 if (copy_to_user(argp, &ifr, ifreq_len))
2589 ret = -EFAULT;
2590 break;
2591
2592 case TUNSETNOCSUM:
2593 /* Disable/Enable checksum */
2594
2595 /* [unimplemented] */
2596 tun_debug(KERN_INFO, tun, "ignored: set checksum %s\n",
2597 arg ? "disabled" : "enabled");
2598 break;
2599
2600 case TUNSETPERSIST:
2601 /* Disable/Enable persist mode. Keep an extra reference to the
2602 * module to prevent the module being unprobed.
2603 */
2604 if (arg && !(tun->flags & IFF_PERSIST)) {
2605 tun->flags |= IFF_PERSIST;
2606 __module_get(THIS_MODULE);
2607 }
2608 if (!arg && (tun->flags & IFF_PERSIST)) {
2609 tun->flags &= ~IFF_PERSIST;
2610 module_put(THIS_MODULE);
2611 }
2612
2613 tun_debug(KERN_INFO, tun, "persist %s\n",
2614 arg ? "enabled" : "disabled");
2615 break;
2616
2617 case TUNSETOWNER:
2618 /* Set owner of the device */
2619 owner = make_kuid(current_user_ns(), arg);
2620 if (!uid_valid(owner)) {
2621 ret = -EINVAL;
2622 break;
2623 }
2624 tun->owner = owner;
2625 tun_debug(KERN_INFO, tun, "owner set to %u\n",
2626 from_kuid(&init_user_ns, tun->owner));
2627 break;
2628
2629 case TUNSETGROUP:
2630 /* Set group of the device */
2631 group = make_kgid(current_user_ns(), arg);
2632 if (!gid_valid(group)) {
2633 ret = -EINVAL;
2634 break;
2635 }
2636 tun->group = group;
2637 tun_debug(KERN_INFO, tun, "group set to %u\n",
2638 from_kgid(&init_user_ns, tun->group));
2639 break;
2640
2641 case TUNSETLINK:
2642 /* Only allow setting the type when the interface is down */
2643 if (tun->dev->flags & IFF_UP) {
2644 tun_debug(KERN_INFO, tun,
2645 "Linktype set failed because interface is up\n");
2646 ret = -EBUSY;
2647 } else {
2648 tun->dev->type = (int) arg;
2649 tun_debug(KERN_INFO, tun, "linktype set to %d\n",
2650 tun->dev->type);
2651 ret = 0;
2652 }
2653 break;
2654
2655 #ifdef TUN_DEBUG
2656 case TUNSETDEBUG:
2657 tun->debug = arg;
2658 break;
2659 #endif
2660 case TUNSETOFFLOAD:
2661 ret = set_offload(tun, arg);
2662 break;
2663
2664 case TUNSETTXFILTER:
2665 /* Can be set only for TAPs */
2666 ret = -EINVAL;
2667 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
2668 break;
2669 ret = update_filter(&tun->txflt, (void __user *)arg);
2670 break;
2671
2672 case SIOCGIFHWADDR:
2673 /* Get hw address */
2674 memcpy(ifr.ifr_hwaddr.sa_data, tun->dev->dev_addr, ETH_ALEN);
2675 ifr.ifr_hwaddr.sa_family = tun->dev->type;
2676 if (copy_to_user(argp, &ifr, ifreq_len))
2677 ret = -EFAULT;
2678 break;
2679
2680 case SIOCSIFHWADDR:
2681 /* Set hw address */
2682 tun_debug(KERN_DEBUG, tun, "set hw address: %pM\n",
2683 ifr.ifr_hwaddr.sa_data);
2684
2685 ret = dev_set_mac_address(tun->dev, &ifr.ifr_hwaddr);
2686 break;
2687
2688 case TUNGETSNDBUF:
2689 sndbuf = tfile->socket.sk->sk_sndbuf;
2690 if (copy_to_user(argp, &sndbuf, sizeof(sndbuf)))
2691 ret = -EFAULT;
2692 break;
2693
2694 case TUNSETSNDBUF:
2695 if (copy_from_user(&sndbuf, argp, sizeof(sndbuf))) {
2696 ret = -EFAULT;
2697 break;
2698 }
2699 if (sndbuf <= 0) {
2700 ret = -EINVAL;
2701 break;
2702 }
2703
2704 tun->sndbuf = sndbuf;
2705 tun_set_sndbuf(tun);
2706 break;
2707
2708 case TUNGETVNETHDRSZ:
2709 vnet_hdr_sz = tun->vnet_hdr_sz;
2710 if (copy_to_user(argp, &vnet_hdr_sz, sizeof(vnet_hdr_sz)))
2711 ret = -EFAULT;
2712 break;
2713
2714 case TUNSETVNETHDRSZ:
2715 if (copy_from_user(&vnet_hdr_sz, argp, sizeof(vnet_hdr_sz))) {
2716 ret = -EFAULT;
2717 break;
2718 }
2719 if (vnet_hdr_sz < (int)sizeof(struct virtio_net_hdr)) {
2720 ret = -EINVAL;
2721 break;
2722 }
2723
2724 tun->vnet_hdr_sz = vnet_hdr_sz;
2725 break;
2726
2727 case TUNGETVNETLE:
2728 le = !!(tun->flags & TUN_VNET_LE);
2729 if (put_user(le, (int __user *)argp))
2730 ret = -EFAULT;
2731 break;
2732
2733 case TUNSETVNETLE:
2734 if (get_user(le, (int __user *)argp)) {
2735 ret = -EFAULT;
2736 break;
2737 }
2738 if (le)
2739 tun->flags |= TUN_VNET_LE;
2740 else
2741 tun->flags &= ~TUN_VNET_LE;
2742 break;
2743
2744 case TUNGETVNETBE:
2745 ret = tun_get_vnet_be(tun, argp);
2746 break;
2747
2748 case TUNSETVNETBE:
2749 ret = tun_set_vnet_be(tun, argp);
2750 break;
2751
2752 case TUNATTACHFILTER:
2753 /* Can be set only for TAPs */
2754 ret = -EINVAL;
2755 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
2756 break;
2757 ret = -EFAULT;
2758 if (copy_from_user(&tun->fprog, argp, sizeof(tun->fprog)))
2759 break;
2760
2761 ret = tun_attach_filter(tun);
2762 break;
2763
2764 case TUNDETACHFILTER:
2765 /* Can be set only for TAPs */
2766 ret = -EINVAL;
2767 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
2768 break;
2769 ret = 0;
2770 tun_detach_filter(tun, tun->numqueues);
2771 break;
2772
2773 case TUNGETFILTER:
2774 ret = -EINVAL;
2775 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
2776 break;
2777 ret = -EFAULT;
2778 if (copy_to_user(argp, &tun->fprog, sizeof(tun->fprog)))
2779 break;
2780 ret = 0;
2781 break;
2782
2783 case TUNSETCARRIER:
2784 ret = -EFAULT;
2785 if (copy_from_user(&carrier, argp, sizeof(carrier)))
2786 goto unlock;
2787
2788 ret = tun_net_change_carrier(tun->dev, (bool)carrier);
2789 break;
2790
2791 default:
2792 ret = -EINVAL;
2793 break;
2794 }
2795
2796 unlock:
2797 rtnl_unlock();
2798 if (tun)
2799 tun_put(tun);
2800 return ret;
2801 }
2802
2803 static long tun_chr_ioctl(struct file *file,
2804 unsigned int cmd, unsigned long arg)
2805 {
2806 return __tun_chr_ioctl(file, cmd, arg, sizeof (struct ifreq));
2807 }
2808
2809 #ifdef CONFIG_COMPAT
2810 static long tun_chr_compat_ioctl(struct file *file,
2811 unsigned int cmd, unsigned long arg)
2812 {
2813 switch (cmd) {
2814 case TUNSETIFF:
2815 case TUNGETIFF:
2816 case TUNSETTXFILTER:
2817 case TUNGETSNDBUF:
2818 case TUNSETSNDBUF:
2819 case SIOCGIFHWADDR:
2820 case SIOCSIFHWADDR:
2821 arg = (unsigned long)compat_ptr(arg);
2822 break;
2823 default:
2824 arg = (compat_ulong_t)arg;
2825 break;
2826 }
2827
2828 /*
2829 * compat_ifreq is shorter than ifreq, so we must not access beyond
2830 * the end of that structure. All fields that are used in this
2831 * driver are compatible though, we don't need to convert the
2832 * contents.
2833 */
2834 return __tun_chr_ioctl(file, cmd, arg, sizeof(struct compat_ifreq));
2835 }
2836 #endif /* CONFIG_COMPAT */
2837
2838 static int tun_chr_fasync(int fd, struct file *file, int on)
2839 {
2840 struct tun_file *tfile = file->private_data;
2841 int ret;
2842
2843 if ((ret = fasync_helper(fd, file, on, &tfile->fasync)) < 0)
2844 goto out;
2845
2846 if (on) {
2847 __f_setown(file, task_pid(current), PIDTYPE_PID, 0);
2848 tfile->flags |= TUN_FASYNC;
2849 } else
2850 tfile->flags &= ~TUN_FASYNC;
2851 ret = 0;
2852 out:
2853 return ret;
2854 }
2855
2856 static int tun_chr_open(struct inode *inode, struct file * file)
2857 {
2858 struct net *net = current->nsproxy->net_ns;
2859 struct tun_file *tfile;
2860
2861 DBG1(KERN_INFO, "tunX: tun_chr_open\n");
2862
2863 tfile = (struct tun_file *)sk_alloc(net, AF_UNSPEC, GFP_KERNEL,
2864 &tun_proto, 0);
2865 if (!tfile)
2866 return -ENOMEM;
2867 if (skb_array_init(&tfile->tx_array, 0, GFP_KERNEL)) {
2868 sk_free(&tfile->sk);
2869 return -ENOMEM;
2870 }
2871
2872 mutex_init(&tfile->napi_mutex);
2873 RCU_INIT_POINTER(tfile->tun, NULL);
2874 tfile->flags = 0;
2875 tfile->ifindex = 0;
2876
2877 init_waitqueue_head(&tfile->wq.wait);
2878 RCU_INIT_POINTER(tfile->socket.wq, &tfile->wq);
2879
2880 tfile->socket.file = file;
2881 tfile->socket.ops = &tun_socket_ops;
2882
2883 sock_init_data(&tfile->socket, &tfile->sk);
2884
2885 tfile->sk.sk_write_space = tun_sock_write_space;
2886 tfile->sk.sk_sndbuf = INT_MAX;
2887
2888 file->private_data = tfile;
2889 INIT_LIST_HEAD(&tfile->next);
2890
2891 sock_set_flag(&tfile->sk, SOCK_ZEROCOPY);
2892
2893 return 0;
2894 }
2895
2896 static int tun_chr_close(struct inode *inode, struct file *file)
2897 {
2898 struct tun_file *tfile = file->private_data;
2899
2900 tun_detach(tfile, true);
2901
2902 return 0;
2903 }
2904
2905 #ifdef CONFIG_PROC_FS
2906 static void tun_chr_show_fdinfo(struct seq_file *m, struct file *file)
2907 {
2908 struct tun_file *tfile = file->private_data;
2909 struct tun_struct *tun;
2910 struct ifreq ifr;
2911
2912 memset(&ifr, 0, sizeof(ifr));
2913
2914 rtnl_lock();
2915 tun = tun_get(tfile);
2916 if (tun)
2917 tun_get_iff(current->nsproxy->net_ns, tun, &ifr);
2918 rtnl_unlock();
2919
2920 if (tun)
2921 tun_put(tun);
2922
2923 seq_printf(m, "iff:\t%s\n", ifr.ifr_name);
2924 }
2925 #endif
2926
2927 static const struct file_operations tun_fops = {
2928 .owner = THIS_MODULE,
2929 .llseek = no_llseek,
2930 .read_iter = tun_chr_read_iter,
2931 .write_iter = tun_chr_write_iter,
2932 .poll = tun_chr_poll,
2933 .unlocked_ioctl = tun_chr_ioctl,
2934 #ifdef CONFIG_COMPAT
2935 .compat_ioctl = tun_chr_compat_ioctl,
2936 #endif
2937 .open = tun_chr_open,
2938 .release = tun_chr_close,
2939 .fasync = tun_chr_fasync,
2940 #ifdef CONFIG_PROC_FS
2941 .show_fdinfo = tun_chr_show_fdinfo,
2942 #endif
2943 };
2944
2945 static struct miscdevice tun_miscdev = {
2946 .minor = TUN_MINOR,
2947 .name = "tun",
2948 .nodename = "net/tun",
2949 .fops = &tun_fops,
2950 };
2951
2952 /* ethtool interface */
2953
2954 static int tun_get_link_ksettings(struct net_device *dev,
2955 struct ethtool_link_ksettings *cmd)
2956 {
2957 ethtool_link_ksettings_zero_link_mode(cmd, supported);
2958 ethtool_link_ksettings_zero_link_mode(cmd, advertising);
2959 cmd->base.speed = SPEED_10;
2960 cmd->base.duplex = DUPLEX_FULL;
2961 cmd->base.port = PORT_TP;
2962 cmd->base.phy_address = 0;
2963 cmd->base.autoneg = AUTONEG_DISABLE;
2964 return 0;
2965 }
2966
2967 static void tun_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
2968 {
2969 struct tun_struct *tun = netdev_priv(dev);
2970
2971 strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
2972 strlcpy(info->version, DRV_VERSION, sizeof(info->version));
2973
2974 switch (tun->flags & TUN_TYPE_MASK) {
2975 case IFF_TUN:
2976 strlcpy(info->bus_info, "tun", sizeof(info->bus_info));
2977 break;
2978 case IFF_TAP:
2979 strlcpy(info->bus_info, "tap", sizeof(info->bus_info));
2980 break;
2981 }
2982 }
2983
2984 static u32 tun_get_msglevel(struct net_device *dev)
2985 {
2986 #ifdef TUN_DEBUG
2987 struct tun_struct *tun = netdev_priv(dev);
2988 return tun->debug;
2989 #else
2990 return -EOPNOTSUPP;
2991 #endif
2992 }
2993
2994 static void tun_set_msglevel(struct net_device *dev, u32 value)
2995 {
2996 #ifdef TUN_DEBUG
2997 struct tun_struct *tun = netdev_priv(dev);
2998 tun->debug = value;
2999 #endif
3000 }
3001
3002 static int tun_get_coalesce(struct net_device *dev,
3003 struct ethtool_coalesce *ec)
3004 {
3005 struct tun_struct *tun = netdev_priv(dev);
3006
3007 ec->rx_max_coalesced_frames = tun->rx_batched;
3008
3009 return 0;
3010 }
3011
3012 static int tun_set_coalesce(struct net_device *dev,
3013 struct ethtool_coalesce *ec)
3014 {
3015 struct tun_struct *tun = netdev_priv(dev);
3016
3017 if (ec->rx_max_coalesced_frames > NAPI_POLL_WEIGHT)
3018 tun->rx_batched = NAPI_POLL_WEIGHT;
3019 else
3020 tun->rx_batched = ec->rx_max_coalesced_frames;
3021
3022 return 0;
3023 }
3024
3025 static const struct ethtool_ops tun_ethtool_ops = {
3026 .get_drvinfo = tun_get_drvinfo,
3027 .get_msglevel = tun_get_msglevel,
3028 .set_msglevel = tun_set_msglevel,
3029 .get_link = ethtool_op_get_link,
3030 .get_ts_info = ethtool_op_get_ts_info,
3031 .get_coalesce = tun_get_coalesce,
3032 .set_coalesce = tun_set_coalesce,
3033 .get_link_ksettings = tun_get_link_ksettings,
3034 };
3035
3036 static int tun_queue_resize(struct tun_struct *tun)
3037 {
3038 struct net_device *dev = tun->dev;
3039 struct tun_file *tfile;
3040 struct skb_array **arrays;
3041 int n = tun->numqueues + tun->numdisabled;
3042 int ret, i;
3043
3044 arrays = kmalloc_array(n, sizeof(*arrays), GFP_KERNEL);
3045 if (!arrays)
3046 return -ENOMEM;
3047
3048 for (i = 0; i < tun->numqueues; i++) {
3049 tfile = rtnl_dereference(tun->tfiles[i]);
3050 arrays[i] = &tfile->tx_array;
3051 }
3052 list_for_each_entry(tfile, &tun->disabled, next)
3053 arrays[i++] = &tfile->tx_array;
3054
3055 ret = skb_array_resize_multiple(arrays, n,
3056 dev->tx_queue_len, GFP_KERNEL);
3057
3058 kfree(arrays);
3059 return ret;
3060 }
3061
3062 static int tun_device_event(struct notifier_block *unused,
3063 unsigned long event, void *ptr)
3064 {
3065 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
3066 struct tun_struct *tun = netdev_priv(dev);
3067
3068 if (dev->rtnl_link_ops != &tun_link_ops)
3069 return NOTIFY_DONE;
3070
3071 switch (event) {
3072 case NETDEV_CHANGE_TX_QUEUE_LEN:
3073 if (tun_queue_resize(tun))
3074 return NOTIFY_BAD;
3075 break;
3076 default:
3077 break;
3078 }
3079
3080 return NOTIFY_DONE;
3081 }
3082
3083 static struct notifier_block tun_notifier_block __read_mostly = {
3084 .notifier_call = tun_device_event,
3085 };
3086
3087 static int __init tun_init(void)
3088 {
3089 int ret = 0;
3090
3091 pr_info("%s, %s\n", DRV_DESCRIPTION, DRV_VERSION);
3092
3093 ret = rtnl_link_register(&tun_link_ops);
3094 if (ret) {
3095 pr_err("Can't register link_ops\n");
3096 goto err_linkops;
3097 }
3098
3099 ret = misc_register(&tun_miscdev);
3100 if (ret) {
3101 pr_err("Can't register misc device %d\n", TUN_MINOR);
3102 goto err_misc;
3103 }
3104
3105 ret = register_netdevice_notifier(&tun_notifier_block);
3106 if (ret) {
3107 pr_err("Can't register netdevice notifier\n");
3108 goto err_notifier;
3109 }
3110
3111 return 0;
3112
3113 err_notifier:
3114 misc_deregister(&tun_miscdev);
3115 err_misc:
3116 rtnl_link_unregister(&tun_link_ops);
3117 err_linkops:
3118 return ret;
3119 }
3120
3121 static void tun_cleanup(void)
3122 {
3123 misc_deregister(&tun_miscdev);
3124 rtnl_link_unregister(&tun_link_ops);
3125 unregister_netdevice_notifier(&tun_notifier_block);
3126 }
3127
3128 /* Get an underlying socket object from tun file. Returns error unless file is
3129 * attached to a device. The returned object works like a packet socket, it
3130 * can be used for sock_sendmsg/sock_recvmsg. The caller is responsible for
3131 * holding a reference to the file for as long as the socket is in use. */
3132 struct socket *tun_get_socket(struct file *file)
3133 {
3134 struct tun_file *tfile;
3135 if (file->f_op != &tun_fops)
3136 return ERR_PTR(-EINVAL);
3137 tfile = file->private_data;
3138 if (!tfile)
3139 return ERR_PTR(-EBADFD);
3140 return &tfile->socket;
3141 }
3142 EXPORT_SYMBOL_GPL(tun_get_socket);
3143
3144 struct skb_array *tun_get_skb_array(struct file *file)
3145 {
3146 struct tun_file *tfile;
3147
3148 if (file->f_op != &tun_fops)
3149 return ERR_PTR(-EINVAL);
3150 tfile = file->private_data;
3151 if (!tfile)
3152 return ERR_PTR(-EBADFD);
3153 return &tfile->tx_array;
3154 }
3155 EXPORT_SYMBOL_GPL(tun_get_skb_array);
3156
3157 module_init(tun_init);
3158 module_exit(tun_cleanup);
3159 MODULE_DESCRIPTION(DRV_DESCRIPTION);
3160 MODULE_AUTHOR(DRV_COPYRIGHT);
3161 MODULE_LICENSE("GPL");
3162 MODULE_ALIAS_MISCDEV(TUN_MINOR);
3163 MODULE_ALIAS("devname:net/tun");
ubuntu-bionic-kernel mirror