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


01c706f8fb74e586c7ac1312216a91e255020585
[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 if (tfile->detached) {
762 tun_enable_queue(tfile);
763 } else {
764 sock_hold(&tfile->sk);
765 tun_napi_init(tun, tfile, napi, napi_frags);
766 }
767
768 /* device is allowed to go away first, so no need to hold extra
769 * refcnt.
770 */
771
772 /* Publish tfile->tun and tun->tfiles only after we've fully
773 * initialized tfile; otherwise we risk using half-initialized
774 * object.
775 */
776 rcu_assign_pointer(tfile->tun, tun);
777 rcu_assign_pointer(tun->tfiles[tun->numqueues], tfile);
778 tun->numqueues++;
779 tun_set_real_num_queues(tun);
780 out:
781 return err;
782 }
783
784 static struct tun_struct *tun_get(struct tun_file *tfile)
785 {
786 struct tun_struct *tun;
787
788 rcu_read_lock();
789 tun = rcu_dereference(tfile->tun);
790 if (tun)
791 dev_hold(tun->dev);
792 rcu_read_unlock();
793
794 return tun;
795 }
796
797 static void tun_put(struct tun_struct *tun)
798 {
799 dev_put(tun->dev);
800 }
801
802 /* TAP filtering */
803 static void addr_hash_set(u32 *mask, const u8 *addr)
804 {
805 int n = ether_crc(ETH_ALEN, addr) >> 26;
806 mask[n >> 5] |= (1 << (n & 31));
807 }
808
809 static unsigned int addr_hash_test(const u32 *mask, const u8 *addr)
810 {
811 int n = ether_crc(ETH_ALEN, addr) >> 26;
812 return mask[n >> 5] & (1 << (n & 31));
813 }
814
815 static int update_filter(struct tap_filter *filter, void __user *arg)
816 {
817 struct { u8 u[ETH_ALEN]; } *addr;
818 struct tun_filter uf;
819 int err, alen, n, nexact;
820
821 if (copy_from_user(&uf, arg, sizeof(uf)))
822 return -EFAULT;
823
824 if (!uf.count) {
825 /* Disabled */
826 filter->count = 0;
827 return 0;
828 }
829
830 alen = ETH_ALEN * uf.count;
831 addr = memdup_user(arg + sizeof(uf), alen);
832 if (IS_ERR(addr))
833 return PTR_ERR(addr);
834
835 /* The filter is updated without holding any locks. Which is
836 * perfectly safe. We disable it first and in the worst
837 * case we'll accept a few undesired packets. */
838 filter->count = 0;
839 wmb();
840
841 /* Use first set of addresses as an exact filter */
842 for (n = 0; n < uf.count && n < FLT_EXACT_COUNT; n++)
843 memcpy(filter->addr[n], addr[n].u, ETH_ALEN);
844
845 nexact = n;
846
847 /* Remaining multicast addresses are hashed,
848 * unicast will leave the filter disabled. */
849 memset(filter->mask, 0, sizeof(filter->mask));
850 for (; n < uf.count; n++) {
851 if (!is_multicast_ether_addr(addr[n].u)) {
852 err = 0; /* no filter */
853 goto free_addr;
854 }
855 addr_hash_set(filter->mask, addr[n].u);
856 }
857
858 /* For ALLMULTI just set the mask to all ones.
859 * This overrides the mask populated above. */
860 if ((uf.flags & TUN_FLT_ALLMULTI))
861 memset(filter->mask, ~0, sizeof(filter->mask));
862
863 /* Now enable the filter */
864 wmb();
865 filter->count = nexact;
866
867 /* Return the number of exact filters */
868 err = nexact;
869 free_addr:
870 kfree(addr);
871 return err;
872 }
873
874 /* Returns: 0 - drop, !=0 - accept */
875 static int run_filter(struct tap_filter *filter, const struct sk_buff *skb)
876 {
877 /* Cannot use eth_hdr(skb) here because skb_mac_hdr() is incorrect
878 * at this point. */
879 struct ethhdr *eh = (struct ethhdr *) skb->data;
880 int i;
881
882 /* Exact match */
883 for (i = 0; i < filter->count; i++)
884 if (ether_addr_equal(eh->h_dest, filter->addr[i]))
885 return 1;
886
887 /* Inexact match (multicast only) */
888 if (is_multicast_ether_addr(eh->h_dest))
889 return addr_hash_test(filter->mask, eh->h_dest);
890
891 return 0;
892 }
893
894 /*
895 * Checks whether the packet is accepted or not.
896 * Returns: 0 - drop, !=0 - accept
897 */
898 static int check_filter(struct tap_filter *filter, const struct sk_buff *skb)
899 {
900 if (!filter->count)
901 return 1;
902
903 return run_filter(filter, skb);
904 }
905
906 /* Network device part of the driver */
907
908 static const struct ethtool_ops tun_ethtool_ops;
909
910 /* Net device detach from fd. */
911 static void tun_net_uninit(struct net_device *dev)
912 {
913 tun_detach_all(dev);
914 }
915
916 /* Net device open. */
917 static int tun_net_open(struct net_device *dev)
918 {
919 netif_tx_start_all_queues(dev);
920
921 return 0;
922 }
923
924 /* Net device close. */
925 static int tun_net_close(struct net_device *dev)
926 {
927 netif_tx_stop_all_queues(dev);
928 return 0;
929 }
930
931 /* Net device start xmit */
932 static netdev_tx_t tun_net_xmit(struct sk_buff *skb, struct net_device *dev)
933 {
934 struct tun_struct *tun = netdev_priv(dev);
935 int txq = skb->queue_mapping;
936 struct tun_file *tfile;
937 u32 numqueues = 0;
938
939 rcu_read_lock();
940 tfile = rcu_dereference(tun->tfiles[txq]);
941 numqueues = READ_ONCE(tun->numqueues);
942
943 /* Drop packet if interface is not attached */
944 if (txq >= numqueues)
945 goto drop;
946
947 #ifdef CONFIG_RPS
948 if (numqueues == 1 && static_key_false(&rps_needed)) {
949 /* Select queue was not called for the skbuff, so we extract the
950 * RPS hash and save it into the flow_table here.
951 */
952 __u32 rxhash;
953
954 rxhash = __skb_get_hash_symmetric(skb);
955 if (rxhash) {
956 struct tun_flow_entry *e;
957 e = tun_flow_find(&tun->flows[tun_hashfn(rxhash)],
958 rxhash);
959 if (e)
960 tun_flow_save_rps_rxhash(e, rxhash);
961 }
962 }
963 #endif
964
965 tun_debug(KERN_INFO, tun, "tun_net_xmit %d\n", skb->len);
966
967 BUG_ON(!tfile);
968
969 /* Drop if the filter does not like it.
970 * This is a noop if the filter is disabled.
971 * Filter can be enabled only for the TAP devices. */
972 if (!check_filter(&tun->txflt, skb))
973 goto drop;
974
975 if (tfile->socket.sk->sk_filter &&
976 sk_filter(tfile->socket.sk, skb))
977 goto drop;
978
979 if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC)))
980 goto drop;
981
982 skb_tx_timestamp(skb);
983
984 /* Orphan the skb - required as we might hang on to it
985 * for indefinite time.
986 */
987 skb_orphan(skb);
988
989 nf_reset(skb);
990
991 if (skb_array_produce(&tfile->tx_array, skb))
992 goto drop;
993
994 /* Notify and wake up reader process */
995 if (tfile->flags & TUN_FASYNC)
996 kill_fasync(&tfile->fasync, SIGIO, POLL_IN);
997 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
998
999 rcu_read_unlock();
1000 return NETDEV_TX_OK;
1001
1002 drop:
1003 this_cpu_inc(tun->pcpu_stats->tx_dropped);
1004 skb_tx_error(skb);
1005 kfree_skb(skb);
1006 rcu_read_unlock();
1007 return NET_XMIT_DROP;
1008 }
1009
1010 static void tun_net_mclist(struct net_device *dev)
1011 {
1012 /*
1013 * This callback is supposed to deal with mc filter in
1014 * _rx_ path and has nothing to do with the _tx_ path.
1015 * In rx path we always accept everything userspace gives us.
1016 */
1017 }
1018
1019 static netdev_features_t tun_net_fix_features(struct net_device *dev,
1020 netdev_features_t features)
1021 {
1022 struct tun_struct *tun = netdev_priv(dev);
1023
1024 return (features & tun->set_features) | (features & ~TUN_USER_FEATURES);
1025 }
1026 #ifdef CONFIG_NET_POLL_CONTROLLER
1027 static void tun_poll_controller(struct net_device *dev)
1028 {
1029 /*
1030 * Tun only receives frames when:
1031 * 1) the char device endpoint gets data from user space
1032 * 2) the tun socket gets a sendmsg call from user space
1033 * If NAPI is not enabled, since both of those are synchronous
1034 * operations, we are guaranteed never to have pending data when we poll
1035 * for it so there is nothing to do here but return.
1036 * We need this though so netpoll recognizes us as an interface that
1037 * supports polling, which enables bridge devices in virt setups to
1038 * still use netconsole
1039 * If NAPI is enabled, however, we need to schedule polling for all
1040 * queues unless we are using napi_gro_frags(), which we call in
1041 * process context and not in NAPI context.
1042 */
1043 struct tun_struct *tun = netdev_priv(dev);
1044
1045 if (tun->flags & IFF_NAPI) {
1046 struct tun_file *tfile;
1047 int i;
1048
1049 rcu_read_lock();
1050 for (i = 0; i < tun->numqueues; i++) {
1051 tfile = rcu_dereference(tun->tfiles[i]);
1052 if (!tun_napi_frags_enabled(tfile) &&
1053 tfile->napi_enabled)
1054 napi_schedule(&tfile->napi);
1055 }
1056 rcu_read_unlock();
1057 }
1058 return;
1059 }
1060 #endif
1061
1062 static void tun_set_headroom(struct net_device *dev, int new_hr)
1063 {
1064 struct tun_struct *tun = netdev_priv(dev);
1065
1066 if (new_hr < NET_SKB_PAD)
1067 new_hr = NET_SKB_PAD;
1068
1069 tun->align = new_hr;
1070 }
1071
1072 static void
1073 tun_net_get_stats64(struct net_device *dev, struct rtnl_link_stats64 *stats)
1074 {
1075 u32 rx_dropped = 0, tx_dropped = 0, rx_frame_errors = 0;
1076 struct tun_struct *tun = netdev_priv(dev);
1077 struct tun_pcpu_stats *p;
1078 int i;
1079
1080 for_each_possible_cpu(i) {
1081 u64 rxpackets, rxbytes, txpackets, txbytes;
1082 unsigned int start;
1083
1084 p = per_cpu_ptr(tun->pcpu_stats, i);
1085 do {
1086 start = u64_stats_fetch_begin(&p->syncp);
1087 rxpackets = p->rx_packets;
1088 rxbytes = p->rx_bytes;
1089 txpackets = p->tx_packets;
1090 txbytes = p->tx_bytes;
1091 } while (u64_stats_fetch_retry(&p->syncp, start));
1092
1093 stats->rx_packets += rxpackets;
1094 stats->rx_bytes += rxbytes;
1095 stats->tx_packets += txpackets;
1096 stats->tx_bytes += txbytes;
1097
1098 /* u32 counters */
1099 rx_dropped += p->rx_dropped;
1100 rx_frame_errors += p->rx_frame_errors;
1101 tx_dropped += p->tx_dropped;
1102 }
1103 stats->rx_dropped = rx_dropped;
1104 stats->rx_frame_errors = rx_frame_errors;
1105 stats->tx_dropped = tx_dropped;
1106 }
1107
1108 static int tun_xdp_set(struct net_device *dev, struct bpf_prog *prog,
1109 struct netlink_ext_ack *extack)
1110 {
1111 struct tun_struct *tun = netdev_priv(dev);
1112 struct bpf_prog *old_prog;
1113
1114 old_prog = rtnl_dereference(tun->xdp_prog);
1115 rcu_assign_pointer(tun->xdp_prog, prog);
1116 if (old_prog)
1117 bpf_prog_put(old_prog);
1118
1119 return 0;
1120 }
1121
1122 static u32 tun_xdp_query(struct net_device *dev)
1123 {
1124 struct tun_struct *tun = netdev_priv(dev);
1125 const struct bpf_prog *xdp_prog;
1126
1127 xdp_prog = rtnl_dereference(tun->xdp_prog);
1128 if (xdp_prog)
1129 return xdp_prog->aux->id;
1130
1131 return 0;
1132 }
1133
1134 static int tun_xdp(struct net_device *dev, struct netdev_bpf *xdp)
1135 {
1136 switch (xdp->command) {
1137 case XDP_SETUP_PROG:
1138 return tun_xdp_set(dev, xdp->prog, xdp->extack);
1139 case XDP_QUERY_PROG:
1140 xdp->prog_id = tun_xdp_query(dev);
1141 xdp->prog_attached = !!xdp->prog_id;
1142 return 0;
1143 default:
1144 return -EINVAL;
1145 }
1146 }
1147
1148 static int tun_net_change_carrier(struct net_device *dev, bool new_carrier)
1149 {
1150 if (new_carrier) {
1151 struct tun_struct *tun = netdev_priv(dev);
1152
1153 if (!tun->numqueues)
1154 return -EPERM;
1155
1156 netif_carrier_on(dev);
1157 } else {
1158 netif_carrier_off(dev);
1159 }
1160 return 0;
1161 }
1162
1163 static const struct net_device_ops tun_netdev_ops = {
1164 .ndo_uninit = tun_net_uninit,
1165 .ndo_open = tun_net_open,
1166 .ndo_stop = tun_net_close,
1167 .ndo_start_xmit = tun_net_xmit,
1168 .ndo_fix_features = tun_net_fix_features,
1169 .ndo_select_queue = tun_select_queue,
1170 #ifdef CONFIG_NET_POLL_CONTROLLER
1171 .ndo_poll_controller = tun_poll_controller,
1172 #endif
1173 .ndo_set_rx_headroom = tun_set_headroom,
1174 .ndo_get_stats64 = tun_net_get_stats64,
1175 .ndo_change_carrier = tun_net_change_carrier,
1176 };
1177
1178 static const struct net_device_ops tap_netdev_ops = {
1179 .ndo_uninit = tun_net_uninit,
1180 .ndo_open = tun_net_open,
1181 .ndo_stop = tun_net_close,
1182 .ndo_start_xmit = tun_net_xmit,
1183 .ndo_fix_features = tun_net_fix_features,
1184 .ndo_set_rx_mode = tun_net_mclist,
1185 .ndo_set_mac_address = eth_mac_addr,
1186 .ndo_validate_addr = eth_validate_addr,
1187 .ndo_select_queue = tun_select_queue,
1188 #ifdef CONFIG_NET_POLL_CONTROLLER
1189 .ndo_poll_controller = tun_poll_controller,
1190 #endif
1191 .ndo_features_check = passthru_features_check,
1192 .ndo_set_rx_headroom = tun_set_headroom,
1193 .ndo_get_stats64 = tun_net_get_stats64,
1194 .ndo_bpf = tun_xdp,
1195 .ndo_change_carrier = tun_net_change_carrier,
1196 };
1197
1198 static void tun_flow_init(struct tun_struct *tun)
1199 {
1200 int i;
1201
1202 for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++)
1203 INIT_HLIST_HEAD(&tun->flows[i]);
1204
1205 tun->ageing_time = TUN_FLOW_EXPIRE;
1206 timer_setup(&tun->flow_gc_timer, tun_flow_cleanup, 0);
1207 mod_timer(&tun->flow_gc_timer,
1208 round_jiffies_up(jiffies + tun->ageing_time));
1209 }
1210
1211 static void tun_flow_uninit(struct tun_struct *tun)
1212 {
1213 del_timer_sync(&tun->flow_gc_timer);
1214 tun_flow_flush(tun);
1215 }
1216
1217 #define MIN_MTU 68
1218 #define MAX_MTU 65535
1219
1220 /* Initialize net device. */
1221 static void tun_net_init(struct net_device *dev)
1222 {
1223 struct tun_struct *tun = netdev_priv(dev);
1224
1225 switch (tun->flags & TUN_TYPE_MASK) {
1226 case IFF_TUN:
1227 dev->netdev_ops = &tun_netdev_ops;
1228
1229 /* Point-to-Point TUN Device */
1230 dev->hard_header_len = 0;
1231 dev->addr_len = 0;
1232 dev->mtu = 1500;
1233
1234 /* Zero header length */
1235 dev->type = ARPHRD_NONE;
1236 dev->flags = IFF_POINTOPOINT | IFF_NOARP | IFF_MULTICAST;
1237 break;
1238
1239 case IFF_TAP:
1240 dev->netdev_ops = &tap_netdev_ops;
1241 /* Ethernet TAP Device */
1242 ether_setup(dev);
1243 dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1244 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
1245
1246 eth_hw_addr_random(dev);
1247
1248 break;
1249 }
1250
1251 dev->min_mtu = MIN_MTU;
1252 dev->max_mtu = MAX_MTU - dev->hard_header_len;
1253 }
1254
1255 /* Character device part */
1256
1257 /* Poll */
1258 static unsigned int tun_chr_poll(struct file *file, poll_table *wait)
1259 {
1260 struct tun_file *tfile = file->private_data;
1261 struct tun_struct *tun = tun_get(tfile);
1262 struct sock *sk;
1263 unsigned int mask = 0;
1264
1265 if (!tun)
1266 return POLLERR;
1267
1268 sk = tfile->socket.sk;
1269
1270 tun_debug(KERN_INFO, tun, "tun_chr_poll\n");
1271
1272 poll_wait(file, sk_sleep(sk), wait);
1273
1274 if (!skb_array_empty(&tfile->tx_array))
1275 mask |= POLLIN | POLLRDNORM;
1276
1277 if (tun->dev->flags & IFF_UP &&
1278 (sock_writeable(sk) ||
1279 (!test_and_set_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags) &&
1280 sock_writeable(sk))))
1281 mask |= POLLOUT | POLLWRNORM;
1282
1283 if (tun->dev->reg_state != NETREG_REGISTERED)
1284 mask = POLLERR;
1285
1286 tun_put(tun);
1287 return mask;
1288 }
1289
1290 static struct sk_buff *tun_napi_alloc_frags(struct tun_file *tfile,
1291 size_t len,
1292 const struct iov_iter *it)
1293 {
1294 struct sk_buff *skb;
1295 size_t linear;
1296 int err;
1297 int i;
1298
1299 if (it->nr_segs > MAX_SKB_FRAGS + 1)
1300 return ERR_PTR(-ENOMEM);
1301
1302 local_bh_disable();
1303 skb = napi_get_frags(&tfile->napi);
1304 local_bh_enable();
1305 if (!skb)
1306 return ERR_PTR(-ENOMEM);
1307
1308 linear = iov_iter_single_seg_count(it);
1309 err = __skb_grow(skb, linear);
1310 if (err)
1311 goto free;
1312
1313 skb->len = len;
1314 skb->data_len = len - linear;
1315 skb->truesize += skb->data_len;
1316
1317 for (i = 1; i < it->nr_segs; i++) {
1318 struct page_frag *pfrag = &current->task_frag;
1319 size_t fragsz = it->iov[i].iov_len;
1320
1321 if (fragsz == 0 || fragsz > PAGE_SIZE) {
1322 err = -EINVAL;
1323 goto free;
1324 }
1325
1326 if (!skb_page_frag_refill(fragsz, pfrag, GFP_KERNEL)) {
1327 err = -ENOMEM;
1328 goto free;
1329 }
1330
1331 skb_fill_page_desc(skb, i - 1, pfrag->page,
1332 pfrag->offset, fragsz);
1333 page_ref_inc(pfrag->page);
1334 pfrag->offset += fragsz;
1335 }
1336
1337 return skb;
1338 free:
1339 /* frees skb and all frags allocated with napi_alloc_frag() */
1340 napi_free_frags(&tfile->napi);
1341 return ERR_PTR(err);
1342 }
1343
1344 /* prepad is the amount to reserve at front. len is length after that.
1345 * linear is a hint as to how much to copy (usually headers). */
1346 static struct sk_buff *tun_alloc_skb(struct tun_file *tfile,
1347 size_t prepad, size_t len,
1348 size_t linear, int noblock)
1349 {
1350 struct sock *sk = tfile->socket.sk;
1351 struct sk_buff *skb;
1352 int err;
1353
1354 /* Under a page? Don't bother with paged skb. */
1355 if (prepad + len < PAGE_SIZE || !linear)
1356 linear = len;
1357
1358 skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
1359 &err, 0);
1360 if (!skb)
1361 return ERR_PTR(err);
1362
1363 skb_reserve(skb, prepad);
1364 skb_put(skb, linear);
1365 skb->data_len = len - linear;
1366 skb->len += len - linear;
1367
1368 return skb;
1369 }
1370
1371 static void tun_rx_batched(struct tun_struct *tun, struct tun_file *tfile,
1372 struct sk_buff *skb, int more)
1373 {
1374 struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
1375 struct sk_buff_head process_queue;
1376 u32 rx_batched = tun->rx_batched;
1377 bool rcv = false;
1378
1379 if (!rx_batched || (!more && skb_queue_empty(queue))) {
1380 local_bh_disable();
1381 skb_record_rx_queue(skb, tfile->queue_index);
1382 netif_receive_skb(skb);
1383 local_bh_enable();
1384 return;
1385 }
1386
1387 spin_lock(&queue->lock);
1388 if (!more || skb_queue_len(queue) == rx_batched) {
1389 __skb_queue_head_init(&process_queue);
1390 skb_queue_splice_tail_init(queue, &process_queue);
1391 rcv = true;
1392 } else {
1393 __skb_queue_tail(queue, skb);
1394 }
1395 spin_unlock(&queue->lock);
1396
1397 if (rcv) {
1398 struct sk_buff *nskb;
1399
1400 local_bh_disable();
1401 while ((nskb = __skb_dequeue(&process_queue))) {
1402 skb_record_rx_queue(nskb, tfile->queue_index);
1403 netif_receive_skb(nskb);
1404 }
1405 skb_record_rx_queue(skb, tfile->queue_index);
1406 netif_receive_skb(skb);
1407 local_bh_enable();
1408 }
1409 }
1410
1411 static bool tun_can_build_skb(struct tun_struct *tun, struct tun_file *tfile,
1412 int len, int noblock, bool zerocopy)
1413 {
1414 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
1415 return false;
1416
1417 if (tfile->socket.sk->sk_sndbuf != INT_MAX)
1418 return false;
1419
1420 if (!noblock)
1421 return false;
1422
1423 if (zerocopy)
1424 return false;
1425
1426 if (SKB_DATA_ALIGN(len + TUN_RX_PAD) +
1427 SKB_DATA_ALIGN(sizeof(struct skb_shared_info)) > PAGE_SIZE)
1428 return false;
1429
1430 return true;
1431 }
1432
1433 static struct sk_buff *tun_build_skb(struct tun_struct *tun,
1434 struct tun_file *tfile,
1435 struct iov_iter *from,
1436 struct virtio_net_hdr *hdr,
1437 int len, int *skb_xdp)
1438 {
1439 struct page_frag *alloc_frag = &current->task_frag;
1440 struct sk_buff *skb;
1441 struct bpf_prog *xdp_prog;
1442 int buflen = SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
1443 unsigned int delta = 0;
1444 char *buf;
1445 size_t copied;
1446 bool xdp_xmit = false;
1447 int err, pad = TUN_RX_PAD;
1448
1449 rcu_read_lock();
1450 xdp_prog = rcu_dereference(tun->xdp_prog);
1451 if (xdp_prog)
1452 pad += TUN_HEADROOM;
1453 buflen += SKB_DATA_ALIGN(len + pad);
1454 rcu_read_unlock();
1455
1456 alloc_frag->offset = ALIGN((u64)alloc_frag->offset, SMP_CACHE_BYTES);
1457 if (unlikely(!skb_page_frag_refill(buflen, alloc_frag, GFP_KERNEL)))
1458 return ERR_PTR(-ENOMEM);
1459
1460 buf = (char *)page_address(alloc_frag->page) + alloc_frag->offset;
1461 copied = copy_page_from_iter(alloc_frag->page,
1462 alloc_frag->offset + pad,
1463 len, from);
1464 if (copied != len)
1465 return ERR_PTR(-EFAULT);
1466
1467 /* There's a small window that XDP may be set after the check
1468 * of xdp_prog above, this should be rare and for simplicity
1469 * we do XDP on skb in case the headroom is not enough.
1470 */
1471 if (hdr->gso_type || !xdp_prog)
1472 *skb_xdp = 1;
1473 else
1474 *skb_xdp = 0;
1475
1476 local_bh_disable();
1477 rcu_read_lock();
1478 xdp_prog = rcu_dereference(tun->xdp_prog);
1479 if (xdp_prog && !*skb_xdp) {
1480 struct xdp_buff xdp;
1481 void *orig_data;
1482 u32 act;
1483
1484 xdp.data_hard_start = buf;
1485 xdp.data = buf + pad;
1486 xdp_set_data_meta_invalid(&xdp);
1487 xdp.data_end = xdp.data + len;
1488 orig_data = xdp.data;
1489 act = bpf_prog_run_xdp(xdp_prog, &xdp);
1490
1491 switch (act) {
1492 case XDP_REDIRECT:
1493 get_page(alloc_frag->page);
1494 alloc_frag->offset += buflen;
1495 err = xdp_do_redirect(tun->dev, &xdp, xdp_prog);
1496 xdp_do_flush_map();
1497 if (err)
1498 goto err_redirect;
1499 rcu_read_unlock();
1500 local_bh_enable();
1501 return NULL;
1502 case XDP_TX:
1503 xdp_xmit = true;
1504 /* fall through */
1505 case XDP_PASS:
1506 delta = orig_data - xdp.data;
1507 break;
1508 default:
1509 bpf_warn_invalid_xdp_action(act);
1510 /* fall through */
1511 case XDP_ABORTED:
1512 trace_xdp_exception(tun->dev, xdp_prog, act);
1513 /* fall through */
1514 case XDP_DROP:
1515 goto err_xdp;
1516 }
1517 }
1518
1519 skb = build_skb(buf, buflen);
1520 if (!skb) {
1521 rcu_read_unlock();
1522 local_bh_enable();
1523 return ERR_PTR(-ENOMEM);
1524 }
1525
1526 skb_reserve(skb, pad - delta);
1527 skb_put(skb, len + delta);
1528 skb_set_owner_w(skb, tfile->socket.sk);
1529 get_page(alloc_frag->page);
1530 alloc_frag->offset += buflen;
1531
1532 if (xdp_xmit) {
1533 skb->dev = tun->dev;
1534 generic_xdp_tx(skb, xdp_prog);
1535 rcu_read_unlock();
1536 local_bh_enable();
1537 return NULL;
1538 }
1539
1540 rcu_read_unlock();
1541 local_bh_enable();
1542
1543 return skb;
1544
1545 err_redirect:
1546 put_page(alloc_frag->page);
1547 err_xdp:
1548 rcu_read_unlock();
1549 local_bh_enable();
1550 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1551 return NULL;
1552 }
1553
1554 /* Get packet from user space buffer */
1555 static ssize_t tun_get_user(struct tun_struct *tun, struct tun_file *tfile,
1556 void *msg_control, struct iov_iter *from,
1557 int noblock, bool more)
1558 {
1559 struct tun_pi pi = { 0, cpu_to_be16(ETH_P_IP) };
1560 struct sk_buff *skb;
1561 size_t total_len = iov_iter_count(from);
1562 size_t len = total_len, align = tun->align, linear;
1563 struct virtio_net_hdr gso = { 0 };
1564 struct tun_pcpu_stats *stats;
1565 int good_linear;
1566 int copylen;
1567 bool zerocopy = false;
1568 int err;
1569 u32 rxhash;
1570 int skb_xdp = 1;
1571 bool frags = tun_napi_frags_enabled(tfile);
1572
1573 if (!(tun->flags & IFF_NO_PI)) {
1574 if (len < sizeof(pi))
1575 return -EINVAL;
1576 len -= sizeof(pi);
1577
1578 if (!copy_from_iter_full(&pi, sizeof(pi), from))
1579 return -EFAULT;
1580 }
1581
1582 if (tun->flags & IFF_VNET_HDR) {
1583 int vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
1584
1585 if (len < vnet_hdr_sz)
1586 return -EINVAL;
1587 len -= vnet_hdr_sz;
1588
1589 if (!copy_from_iter_full(&gso, sizeof(gso), from))
1590 return -EFAULT;
1591
1592 if ((gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
1593 tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2 > tun16_to_cpu(tun, gso.hdr_len))
1594 gso.hdr_len = cpu_to_tun16(tun, tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2);
1595
1596 if (tun16_to_cpu(tun, gso.hdr_len) > len)
1597 return -EINVAL;
1598 iov_iter_advance(from, vnet_hdr_sz - sizeof(gso));
1599 }
1600
1601 if ((tun->flags & TUN_TYPE_MASK) == IFF_TAP) {
1602 align += NET_IP_ALIGN;
1603 if (unlikely(len < ETH_HLEN ||
1604 (gso.hdr_len && tun16_to_cpu(tun, gso.hdr_len) < ETH_HLEN)))
1605 return -EINVAL;
1606 }
1607
1608 good_linear = SKB_MAX_HEAD(align);
1609
1610 if (msg_control) {
1611 struct iov_iter i = *from;
1612
1613 /* There are 256 bytes to be copied in skb, so there is
1614 * enough room for skb expand head in case it is used.
1615 * The rest of the buffer is mapped from userspace.
1616 */
1617 copylen = gso.hdr_len ? tun16_to_cpu(tun, gso.hdr_len) : GOODCOPY_LEN;
1618 if (copylen > good_linear)
1619 copylen = good_linear;
1620 linear = copylen;
1621 iov_iter_advance(&i, copylen);
1622 if (iov_iter_npages(&i, INT_MAX) <= MAX_SKB_FRAGS)
1623 zerocopy = true;
1624 }
1625
1626 if (!frags && tun_can_build_skb(tun, tfile, len, noblock, zerocopy)) {
1627 /* For the packet that is not easy to be processed
1628 * (e.g gso or jumbo packet), we will do it at after
1629 * skb was created with generic XDP routine.
1630 */
1631 skb = tun_build_skb(tun, tfile, from, &gso, len, &skb_xdp);
1632 if (IS_ERR(skb)) {
1633 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1634 return PTR_ERR(skb);
1635 }
1636 if (!skb)
1637 return total_len;
1638 } else {
1639 if (!zerocopy) {
1640 copylen = len;
1641 if (tun16_to_cpu(tun, gso.hdr_len) > good_linear)
1642 linear = good_linear;
1643 else
1644 linear = tun16_to_cpu(tun, gso.hdr_len);
1645 }
1646
1647 if (frags) {
1648 mutex_lock(&tfile->napi_mutex);
1649 skb = tun_napi_alloc_frags(tfile, copylen, from);
1650 /* tun_napi_alloc_frags() enforces a layout for the skb.
1651 * If zerocopy is enabled, then this layout will be
1652 * overwritten by zerocopy_sg_from_iter().
1653 */
1654 zerocopy = false;
1655 } else {
1656 skb = tun_alloc_skb(tfile, align, copylen, linear,
1657 noblock);
1658 }
1659
1660 if (IS_ERR(skb)) {
1661 if (PTR_ERR(skb) != -EAGAIN)
1662 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1663 if (frags)
1664 mutex_unlock(&tfile->napi_mutex);
1665 return PTR_ERR(skb);
1666 }
1667
1668 if (zerocopy)
1669 err = zerocopy_sg_from_iter(skb, from);
1670 else
1671 err = skb_copy_datagram_from_iter(skb, 0, from, len);
1672
1673 if (err) {
1674 err = -EFAULT;
1675 drop:
1676 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1677 kfree_skb(skb);
1678 if (frags) {
1679 tfile->napi.skb = NULL;
1680 mutex_unlock(&tfile->napi_mutex);
1681 }
1682
1683 return err;
1684 }
1685 }
1686
1687 if (virtio_net_hdr_to_skb(skb, &gso, tun_is_little_endian(tun))) {
1688 this_cpu_inc(tun->pcpu_stats->rx_frame_errors);
1689 kfree_skb(skb);
1690 if (frags) {
1691 tfile->napi.skb = NULL;
1692 mutex_unlock(&tfile->napi_mutex);
1693 }
1694
1695 return -EINVAL;
1696 }
1697
1698 switch (tun->flags & TUN_TYPE_MASK) {
1699 case IFF_TUN:
1700 if (tun->flags & IFF_NO_PI) {
1701 u8 ip_version = skb->len ? (skb->data[0] >> 4) : 0;
1702
1703 switch (ip_version) {
1704 case 4:
1705 pi.proto = htons(ETH_P_IP);
1706 break;
1707 case 6:
1708 pi.proto = htons(ETH_P_IPV6);
1709 break;
1710 default:
1711 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1712 kfree_skb(skb);
1713 return -EINVAL;
1714 }
1715 }
1716
1717 skb_reset_mac_header(skb);
1718 skb->protocol = pi.proto;
1719 skb->dev = tun->dev;
1720 break;
1721 case IFF_TAP:
1722 if (!frags)
1723 skb->protocol = eth_type_trans(skb, tun->dev);
1724 break;
1725 }
1726
1727 /* copy skb_ubuf_info for callback when skb has no error */
1728 if (zerocopy) {
1729 skb_shinfo(skb)->destructor_arg = msg_control;
1730 skb_shinfo(skb)->tx_flags |= SKBTX_DEV_ZEROCOPY;
1731 skb_shinfo(skb)->tx_flags |= SKBTX_SHARED_FRAG;
1732 } else if (msg_control) {
1733 struct ubuf_info *uarg = msg_control;
1734 uarg->callback(uarg, false);
1735 }
1736
1737 skb_reset_network_header(skb);
1738 skb_probe_transport_header(skb, 0);
1739
1740 if (skb_xdp) {
1741 struct bpf_prog *xdp_prog;
1742 int ret;
1743
1744 local_bh_disable();
1745 rcu_read_lock();
1746 xdp_prog = rcu_dereference(tun->xdp_prog);
1747 if (xdp_prog) {
1748 ret = do_xdp_generic(xdp_prog, skb);
1749 if (ret != XDP_PASS) {
1750 rcu_read_unlock();
1751 local_bh_enable();
1752 return total_len;
1753 }
1754 }
1755 rcu_read_unlock();
1756 local_bh_enable();
1757 }
1758
1759 rxhash = __skb_get_hash_symmetric(skb);
1760
1761 rcu_read_lock();
1762 if (unlikely(!(tun->dev->flags & IFF_UP))) {
1763 err = -EIO;
1764 rcu_read_unlock();
1765 goto drop;
1766 }
1767
1768 if (frags) {
1769 /* Exercise flow dissector code path. */
1770 u32 headlen = eth_get_headlen(skb->data, skb_headlen(skb));
1771
1772 if (unlikely(headlen > skb_headlen(skb))) {
1773 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1774 napi_free_frags(&tfile->napi);
1775 rcu_read_unlock();
1776 mutex_unlock(&tfile->napi_mutex);
1777 WARN_ON(1);
1778 return -ENOMEM;
1779 }
1780
1781 local_bh_disable();
1782 napi_gro_frags(&tfile->napi);
1783 local_bh_enable();
1784 mutex_unlock(&tfile->napi_mutex);
1785 } else if (tfile->napi_enabled) {
1786 struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
1787 int queue_len;
1788
1789 spin_lock_bh(&queue->lock);
1790 __skb_queue_tail(queue, skb);
1791 queue_len = skb_queue_len(queue);
1792 spin_unlock(&queue->lock);
1793
1794 if (!more || queue_len > NAPI_POLL_WEIGHT)
1795 napi_schedule(&tfile->napi);
1796
1797 local_bh_enable();
1798 } else if (!IS_ENABLED(CONFIG_4KSTACKS)) {
1799 tun_rx_batched(tun, tfile, skb, more);
1800 } else {
1801 netif_rx_ni(skb);
1802 }
1803 rcu_read_unlock();
1804
1805 stats = get_cpu_ptr(tun->pcpu_stats);
1806 u64_stats_update_begin(&stats->syncp);
1807 stats->rx_packets++;
1808 stats->rx_bytes += len;
1809 u64_stats_update_end(&stats->syncp);
1810 put_cpu_ptr(stats);
1811
1812 tun_flow_update(tun, rxhash, tfile);
1813 return total_len;
1814 }
1815
1816 static ssize_t tun_chr_write_iter(struct kiocb *iocb, struct iov_iter *from)
1817 {
1818 struct file *file = iocb->ki_filp;
1819 struct tun_file *tfile = file->private_data;
1820 struct tun_struct *tun = tun_get(tfile);
1821 ssize_t result;
1822
1823 if (!tun)
1824 return -EBADFD;
1825
1826 result = tun_get_user(tun, tfile, NULL, from,
1827 file->f_flags & O_NONBLOCK, false);
1828
1829 tun_put(tun);
1830 return result;
1831 }
1832
1833 /* Put packet to the user space buffer */
1834 static ssize_t tun_put_user(struct tun_struct *tun,
1835 struct tun_file *tfile,
1836 struct sk_buff *skb,
1837 struct iov_iter *iter)
1838 {
1839 struct tun_pi pi = { 0, skb->protocol };
1840 struct tun_pcpu_stats *stats;
1841 ssize_t total;
1842 int vlan_offset = 0;
1843 int vlan_hlen = 0;
1844 int vnet_hdr_sz = 0;
1845
1846 if (skb_vlan_tag_present(skb))
1847 vlan_hlen = VLAN_HLEN;
1848
1849 if (tun->flags & IFF_VNET_HDR)
1850 vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
1851
1852 total = skb->len + vlan_hlen + vnet_hdr_sz;
1853
1854 if (!(tun->flags & IFF_NO_PI)) {
1855 if (iov_iter_count(iter) < sizeof(pi))
1856 return -EINVAL;
1857
1858 total += sizeof(pi);
1859 if (iov_iter_count(iter) < total) {
1860 /* Packet will be striped */
1861 pi.flags |= TUN_PKT_STRIP;
1862 }
1863
1864 if (copy_to_iter(&pi, sizeof(pi), iter) != sizeof(pi))
1865 return -EFAULT;
1866 }
1867
1868 if (vnet_hdr_sz) {
1869 struct virtio_net_hdr gso;
1870
1871 if (iov_iter_count(iter) < vnet_hdr_sz)
1872 return -EINVAL;
1873
1874 if (virtio_net_hdr_from_skb(skb, &gso,
1875 tun_is_little_endian(tun), true,
1876 vlan_hlen)) {
1877 struct skb_shared_info *sinfo = skb_shinfo(skb);
1878 pr_err("unexpected GSO type: "
1879 "0x%x, gso_size %d, hdr_len %d\n",
1880 sinfo->gso_type, tun16_to_cpu(tun, gso.gso_size),
1881 tun16_to_cpu(tun, gso.hdr_len));
1882 print_hex_dump(KERN_ERR, "tun: ",
1883 DUMP_PREFIX_NONE,
1884 16, 1, skb->head,
1885 min((int)tun16_to_cpu(tun, gso.hdr_len), 64), true);
1886 WARN_ON_ONCE(1);
1887 return -EINVAL;
1888 }
1889
1890 if (copy_to_iter(&gso, sizeof(gso), iter) != sizeof(gso))
1891 return -EFAULT;
1892
1893 iov_iter_advance(iter, vnet_hdr_sz - sizeof(gso));
1894 }
1895
1896 if (vlan_hlen) {
1897 int ret;
1898 struct {
1899 __be16 h_vlan_proto;
1900 __be16 h_vlan_TCI;
1901 } veth;
1902
1903 veth.h_vlan_proto = skb->vlan_proto;
1904 veth.h_vlan_TCI = htons(skb_vlan_tag_get(skb));
1905
1906 vlan_offset = offsetof(struct vlan_ethhdr, h_vlan_proto);
1907
1908 ret = skb_copy_datagram_iter(skb, 0, iter, vlan_offset);
1909 if (ret || !iov_iter_count(iter))
1910 goto done;
1911
1912 ret = copy_to_iter(&veth, sizeof(veth), iter);
1913 if (ret != sizeof(veth) || !iov_iter_count(iter))
1914 goto done;
1915 }
1916
1917 skb_copy_datagram_iter(skb, vlan_offset, iter, skb->len - vlan_offset);
1918
1919 done:
1920 /* caller is in process context, */
1921 stats = get_cpu_ptr(tun->pcpu_stats);
1922 u64_stats_update_begin(&stats->syncp);
1923 stats->tx_packets++;
1924 stats->tx_bytes += skb->len + vlan_hlen;
1925 u64_stats_update_end(&stats->syncp);
1926 put_cpu_ptr(tun->pcpu_stats);
1927
1928 return total;
1929 }
1930
1931 static struct sk_buff *tun_ring_recv(struct tun_file *tfile, int noblock,
1932 int *err)
1933 {
1934 DECLARE_WAITQUEUE(wait, current);
1935 struct sk_buff *skb = NULL;
1936 int error = 0;
1937
1938 skb = skb_array_consume(&tfile->tx_array);
1939 if (skb)
1940 goto out;
1941 if (noblock) {
1942 error = -EAGAIN;
1943 goto out;
1944 }
1945
1946 add_wait_queue(&tfile->wq.wait, &wait);
1947
1948 while (1) {
1949 set_current_state(TASK_INTERRUPTIBLE);
1950 skb = skb_array_consume(&tfile->tx_array);
1951 if (skb)
1952 break;
1953 if (signal_pending(current)) {
1954 error = -ERESTARTSYS;
1955 break;
1956 }
1957 if (tfile->socket.sk->sk_shutdown & RCV_SHUTDOWN) {
1958 error = -EFAULT;
1959 break;
1960 }
1961
1962 schedule();
1963 }
1964
1965 __set_current_state(TASK_RUNNING);
1966 remove_wait_queue(&tfile->wq.wait, &wait);
1967
1968 out:
1969 *err = error;
1970 return skb;
1971 }
1972
1973 static ssize_t tun_do_read(struct tun_struct *tun, struct tun_file *tfile,
1974 struct iov_iter *to,
1975 int noblock, struct sk_buff *skb)
1976 {
1977 ssize_t ret;
1978 int err;
1979
1980 tun_debug(KERN_INFO, tun, "tun_do_read\n");
1981
1982 if (!iov_iter_count(to)) {
1983 if (skb)
1984 kfree_skb(skb);
1985 return 0;
1986 }
1987
1988 if (!skb) {
1989 /* Read frames from ring */
1990 skb = tun_ring_recv(tfile, noblock, &err);
1991 if (!skb)
1992 return err;
1993 }
1994
1995 ret = tun_put_user(tun, tfile, skb, to);
1996 if (unlikely(ret < 0))
1997 kfree_skb(skb);
1998 else
1999 consume_skb(skb);
2000
2001 return ret;
2002 }
2003
2004 static ssize_t tun_chr_read_iter(struct kiocb *iocb, struct iov_iter *to)
2005 {
2006 struct file *file = iocb->ki_filp;
2007 struct tun_file *tfile = file->private_data;
2008 struct tun_struct *tun = tun_get(tfile);
2009 ssize_t len = iov_iter_count(to), ret;
2010
2011 if (!tun)
2012 return -EBADFD;
2013 ret = tun_do_read(tun, tfile, to, file->f_flags & O_NONBLOCK, NULL);
2014 ret = min_t(ssize_t, ret, len);
2015 if (ret > 0)
2016 iocb->ki_pos = ret;
2017 tun_put(tun);
2018 return ret;
2019 }
2020
2021 static void tun_free_netdev(struct net_device *dev)
2022 {
2023 struct tun_struct *tun = netdev_priv(dev);
2024
2025 BUG_ON(!(list_empty(&tun->disabled)));
2026 free_percpu(tun->pcpu_stats);
2027 tun_flow_uninit(tun);
2028 security_tun_dev_free_security(tun->security);
2029 }
2030
2031 static void tun_setup(struct net_device *dev)
2032 {
2033 struct tun_struct *tun = netdev_priv(dev);
2034
2035 tun->owner = INVALID_UID;
2036 tun->group = INVALID_GID;
2037
2038 dev->ethtool_ops = &tun_ethtool_ops;
2039 dev->needs_free_netdev = true;
2040 dev->priv_destructor = tun_free_netdev;
2041 /* We prefer our own queue length */
2042 dev->tx_queue_len = TUN_READQ_SIZE;
2043 }
2044
2045 /* Trivial set of netlink ops to allow deleting tun or tap
2046 * device with netlink.
2047 */
2048 static int tun_validate(struct nlattr *tb[], struct nlattr *data[],
2049 struct netlink_ext_ack *extack)
2050 {
2051 NL_SET_ERR_MSG(extack,
2052 "tun/tap creation via rtnetlink is not supported.");
2053 return -EOPNOTSUPP;
2054 }
2055
2056 static struct rtnl_link_ops tun_link_ops __read_mostly = {
2057 .kind = DRV_NAME,
2058 .priv_size = sizeof(struct tun_struct),
2059 .setup = tun_setup,
2060 .validate = tun_validate,
2061 };
2062
2063 static void tun_sock_write_space(struct sock *sk)
2064 {
2065 struct tun_file *tfile;
2066 wait_queue_head_t *wqueue;
2067
2068 if (!sock_writeable(sk))
2069 return;
2070
2071 if (!test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags))
2072 return;
2073
2074 wqueue = sk_sleep(sk);
2075 if (wqueue && waitqueue_active(wqueue))
2076 wake_up_interruptible_sync_poll(wqueue, POLLOUT |
2077 POLLWRNORM | POLLWRBAND);
2078
2079 tfile = container_of(sk, struct tun_file, sk);
2080 kill_fasync(&tfile->fasync, SIGIO, POLL_OUT);
2081 }
2082
2083 static int tun_sendmsg(struct socket *sock, struct msghdr *m, size_t total_len)
2084 {
2085 int ret;
2086 struct tun_file *tfile = container_of(sock, struct tun_file, socket);
2087 struct tun_struct *tun = tun_get(tfile);
2088
2089 if (!tun)
2090 return -EBADFD;
2091
2092 ret = tun_get_user(tun, tfile, m->msg_control, &m->msg_iter,
2093 m->msg_flags & MSG_DONTWAIT,
2094 m->msg_flags & MSG_MORE);
2095 tun_put(tun);
2096 return ret;
2097 }
2098
2099 static int tun_recvmsg(struct socket *sock, struct msghdr *m, size_t total_len,
2100 int flags)
2101 {
2102 struct tun_file *tfile = container_of(sock, struct tun_file, socket);
2103 struct tun_struct *tun = tun_get(tfile);
2104 struct sk_buff *skb = m->msg_control;
2105 int ret;
2106
2107 if (!tun) {
2108 ret = -EBADFD;
2109 goto out_free_skb;
2110 }
2111
2112 if (flags & ~(MSG_DONTWAIT|MSG_TRUNC|MSG_ERRQUEUE)) {
2113 ret = -EINVAL;
2114 goto out_put_tun;
2115 }
2116 if (flags & MSG_ERRQUEUE) {
2117 ret = sock_recv_errqueue(sock->sk, m, total_len,
2118 SOL_PACKET, TUN_TX_TIMESTAMP);
2119 goto out;
2120 }
2121 ret = tun_do_read(tun, tfile, &m->msg_iter, flags & MSG_DONTWAIT, skb);
2122 if (ret > (ssize_t)total_len) {
2123 m->msg_flags |= MSG_TRUNC;
2124 ret = flags & MSG_TRUNC ? ret : total_len;
2125 }
2126 out:
2127 tun_put(tun);
2128 return ret;
2129
2130 out_put_tun:
2131 tun_put(tun);
2132 out_free_skb:
2133 if (skb)
2134 kfree_skb(skb);
2135 return ret;
2136 }
2137
2138 static int tun_peek_len(struct socket *sock)
2139 {
2140 struct tun_file *tfile = container_of(sock, struct tun_file, socket);
2141 struct tun_struct *tun;
2142 int ret = 0;
2143
2144 tun = tun_get(tfile);
2145 if (!tun)
2146 return 0;
2147
2148 ret = skb_array_peek_len(&tfile->tx_array);
2149 tun_put(tun);
2150
2151 return ret;
2152 }
2153
2154 /* Ops structure to mimic raw sockets with tun */
2155 static const struct proto_ops tun_socket_ops = {
2156 .peek_len = tun_peek_len,
2157 .sendmsg = tun_sendmsg,
2158 .recvmsg = tun_recvmsg,
2159 };
2160
2161 static struct proto tun_proto = {
2162 .name = "tun",
2163 .owner = THIS_MODULE,
2164 .obj_size = sizeof(struct tun_file),
2165 };
2166
2167 static int tun_flags(struct tun_struct *tun)
2168 {
2169 return tun->flags & (TUN_FEATURES | IFF_PERSIST | IFF_TUN | IFF_TAP);
2170 }
2171
2172 static ssize_t tun_show_flags(struct device *dev, struct device_attribute *attr,
2173 char *buf)
2174 {
2175 struct tun_struct *tun = netdev_priv(to_net_dev(dev));
2176 return sprintf(buf, "0x%x\n", tun_flags(tun));
2177 }
2178
2179 static ssize_t tun_show_owner(struct device *dev, struct device_attribute *attr,
2180 char *buf)
2181 {
2182 struct tun_struct *tun = netdev_priv(to_net_dev(dev));
2183 return uid_valid(tun->owner)?
2184 sprintf(buf, "%u\n",
2185 from_kuid_munged(current_user_ns(), tun->owner)):
2186 sprintf(buf, "-1\n");
2187 }
2188
2189 static ssize_t tun_show_group(struct device *dev, struct device_attribute *attr,
2190 char *buf)
2191 {
2192 struct tun_struct *tun = netdev_priv(to_net_dev(dev));
2193 return gid_valid(tun->group) ?
2194 sprintf(buf, "%u\n",
2195 from_kgid_munged(current_user_ns(), tun->group)):
2196 sprintf(buf, "-1\n");
2197 }
2198
2199 static DEVICE_ATTR(tun_flags, 0444, tun_show_flags, NULL);
2200 static DEVICE_ATTR(owner, 0444, tun_show_owner, NULL);
2201 static DEVICE_ATTR(group, 0444, tun_show_group, NULL);
2202
2203 static struct attribute *tun_dev_attrs[] = {
2204 &dev_attr_tun_flags.attr,
2205 &dev_attr_owner.attr,
2206 &dev_attr_group.attr,
2207 NULL
2208 };
2209
2210 static const struct attribute_group tun_attr_group = {
2211 .attrs = tun_dev_attrs
2212 };
2213
2214 static int tun_set_iff(struct net *net, struct file *file, struct ifreq *ifr)
2215 {
2216 struct tun_struct *tun;
2217 struct tun_file *tfile = file->private_data;
2218 struct net_device *dev;
2219 int err;
2220
2221 if (tfile->detached)
2222 return -EINVAL;
2223
2224 if ((ifr->ifr_flags & IFF_NAPI_FRAGS)) {
2225 if (!capable(CAP_NET_ADMIN))
2226 return -EPERM;
2227
2228 if (!(ifr->ifr_flags & IFF_NAPI) ||
2229 (ifr->ifr_flags & TUN_TYPE_MASK) != IFF_TAP)
2230 return -EINVAL;
2231 }
2232
2233 dev = __dev_get_by_name(net, ifr->ifr_name);
2234 if (dev) {
2235 if (ifr->ifr_flags & IFF_TUN_EXCL)
2236 return -EBUSY;
2237 if ((ifr->ifr_flags & IFF_TUN) && dev->netdev_ops == &tun_netdev_ops)
2238 tun = netdev_priv(dev);
2239 else if ((ifr->ifr_flags & IFF_TAP) && dev->netdev_ops == &tap_netdev_ops)
2240 tun = netdev_priv(dev);
2241 else
2242 return -EINVAL;
2243
2244 if (!!(ifr->ifr_flags & IFF_MULTI_QUEUE) !=
2245 !!(tun->flags & IFF_MULTI_QUEUE))
2246 return -EINVAL;
2247
2248 if (tun_not_capable(tun))
2249 return -EPERM;
2250 err = security_tun_dev_open(tun->security);
2251 if (err < 0)
2252 return err;
2253
2254 err = tun_attach(tun, file, ifr->ifr_flags & IFF_NOFILTER,
2255 ifr->ifr_flags & IFF_NAPI,
2256 ifr->ifr_flags & IFF_NAPI_FRAGS);
2257 if (err < 0)
2258 return err;
2259
2260 if (tun->flags & IFF_MULTI_QUEUE &&
2261 (tun->numqueues + tun->numdisabled > 1)) {
2262 /* One or more queue has already been attached, no need
2263 * to initialize the device again.
2264 */
2265 return 0;
2266 }
2267 }
2268 else {
2269 char *name;
2270 unsigned long flags = 0;
2271 int queues = ifr->ifr_flags & IFF_MULTI_QUEUE ?
2272 MAX_TAP_QUEUES : 1;
2273
2274 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
2275 return -EPERM;
2276 err = security_tun_dev_create();
2277 if (err < 0)
2278 return err;
2279
2280 /* Set dev type */
2281 if (ifr->ifr_flags & IFF_TUN) {
2282 /* TUN device */
2283 flags |= IFF_TUN;
2284 name = "tun%d";
2285 } else if (ifr->ifr_flags & IFF_TAP) {
2286 /* TAP device */
2287 flags |= IFF_TAP;
2288 name = "tap%d";
2289 } else
2290 return -EINVAL;
2291
2292 if (*ifr->ifr_name)
2293 name = ifr->ifr_name;
2294
2295 dev = alloc_netdev_mqs(sizeof(struct tun_struct), name,
2296 NET_NAME_UNKNOWN, tun_setup, queues,
2297 queues);
2298
2299 if (!dev)
2300 return -ENOMEM;
2301 err = dev_get_valid_name(net, dev, name);
2302 if (err < 0)
2303 goto err_free_dev;
2304
2305 dev_net_set(dev, net);
2306 dev->rtnl_link_ops = &tun_link_ops;
2307 dev->ifindex = tfile->ifindex;
2308 dev->sysfs_groups[0] = &tun_attr_group;
2309
2310 tun = netdev_priv(dev);
2311 tun->dev = dev;
2312 tun->flags = flags;
2313 tun->txflt.count = 0;
2314 tun->vnet_hdr_sz = sizeof(struct virtio_net_hdr);
2315
2316 tun->align = NET_SKB_PAD;
2317 tun->filter_attached = false;
2318 tun->sndbuf = tfile->socket.sk->sk_sndbuf;
2319 tun->rx_batched = 0;
2320
2321 tun->pcpu_stats = netdev_alloc_pcpu_stats(struct tun_pcpu_stats);
2322 if (!tun->pcpu_stats) {
2323 err = -ENOMEM;
2324 goto err_free_dev;
2325 }
2326
2327 spin_lock_init(&tun->lock);
2328
2329 err = security_tun_dev_alloc_security(&tun->security);
2330 if (err < 0)
2331 goto err_free_stat;
2332
2333 tun_net_init(dev);
2334 tun_flow_init(tun);
2335
2336 dev->hw_features = NETIF_F_SG | NETIF_F_FRAGLIST |
2337 TUN_USER_FEATURES | NETIF_F_HW_VLAN_CTAG_TX |
2338 NETIF_F_HW_VLAN_STAG_TX;
2339 dev->features = dev->hw_features | NETIF_F_LLTX;
2340 dev->vlan_features = dev->features &
2341 ~(NETIF_F_HW_VLAN_CTAG_TX |
2342 NETIF_F_HW_VLAN_STAG_TX);
2343
2344 INIT_LIST_HEAD(&tun->disabled);
2345 err = tun_attach(tun, file, false, ifr->ifr_flags & IFF_NAPI,
2346 ifr->ifr_flags & IFF_NAPI_FRAGS);
2347 if (err < 0)
2348 goto err_free_flow;
2349
2350 err = register_netdevice(tun->dev);
2351 if (err < 0)
2352 goto err_detach;
2353 }
2354
2355 netif_carrier_on(tun->dev);
2356
2357 tun_debug(KERN_INFO, tun, "tun_set_iff\n");
2358
2359 tun->flags = (tun->flags & ~TUN_FEATURES) |
2360 (ifr->ifr_flags & TUN_FEATURES);
2361
2362 /* Make sure persistent devices do not get stuck in
2363 * xoff state.
2364 */
2365 if (netif_running(tun->dev))
2366 netif_tx_wake_all_queues(tun->dev);
2367
2368 strcpy(ifr->ifr_name, tun->dev->name);
2369 return 0;
2370
2371 err_detach:
2372 tun_detach_all(dev);
2373 /* register_netdevice() already called tun_free_netdev() */
2374 goto err_free_dev;
2375
2376 err_free_flow:
2377 tun_flow_uninit(tun);
2378 security_tun_dev_free_security(tun->security);
2379 err_free_stat:
2380 free_percpu(tun->pcpu_stats);
2381 err_free_dev:
2382 free_netdev(dev);
2383 return err;
2384 }
2385
2386 static void tun_get_iff(struct net *net, struct tun_struct *tun,
2387 struct ifreq *ifr)
2388 {
2389 tun_debug(KERN_INFO, tun, "tun_get_iff\n");
2390
2391 strcpy(ifr->ifr_name, tun->dev->name);
2392
2393 ifr->ifr_flags = tun_flags(tun);
2394
2395 }
2396
2397 /* This is like a cut-down ethtool ops, except done via tun fd so no
2398 * privs required. */
2399 static int set_offload(struct tun_struct *tun, unsigned long arg)
2400 {
2401 netdev_features_t features = 0;
2402
2403 if (arg & TUN_F_CSUM) {
2404 features |= NETIF_F_HW_CSUM;
2405 arg &= ~TUN_F_CSUM;
2406
2407 if (arg & (TUN_F_TSO4|TUN_F_TSO6)) {
2408 if (arg & TUN_F_TSO_ECN) {
2409 features |= NETIF_F_TSO_ECN;
2410 arg &= ~TUN_F_TSO_ECN;
2411 }
2412 if (arg & TUN_F_TSO4)
2413 features |= NETIF_F_TSO;
2414 if (arg & TUN_F_TSO6)
2415 features |= NETIF_F_TSO6;
2416 arg &= ~(TUN_F_TSO4|TUN_F_TSO6);
2417 }
2418
2419 arg &= ~TUN_F_UFO;
2420 }
2421
2422 /* This gives the user a way to test for new features in future by
2423 * trying to set them. */
2424 if (arg)
2425 return -EINVAL;
2426
2427 tun->set_features = features;
2428 tun->dev->wanted_features &= ~TUN_USER_FEATURES;
2429 tun->dev->wanted_features |= features;
2430 netdev_update_features(tun->dev);
2431
2432 return 0;
2433 }
2434
2435 static void tun_detach_filter(struct tun_struct *tun, int n)
2436 {
2437 int i;
2438 struct tun_file *tfile;
2439
2440 for (i = 0; i < n; i++) {
2441 tfile = rtnl_dereference(tun->tfiles[i]);
2442 lock_sock(tfile->socket.sk);
2443 sk_detach_filter(tfile->socket.sk);
2444 release_sock(tfile->socket.sk);
2445 }
2446
2447 tun->filter_attached = false;
2448 }
2449
2450 static int tun_attach_filter(struct tun_struct *tun)
2451 {
2452 int i, ret = 0;
2453 struct tun_file *tfile;
2454
2455 for (i = 0; i < tun->numqueues; i++) {
2456 tfile = rtnl_dereference(tun->tfiles[i]);
2457 lock_sock(tfile->socket.sk);
2458 ret = sk_attach_filter(&tun->fprog, tfile->socket.sk);
2459 release_sock(tfile->socket.sk);
2460 if (ret) {
2461 tun_detach_filter(tun, i);
2462 return ret;
2463 }
2464 }
2465
2466 tun->filter_attached = true;
2467 return ret;
2468 }
2469
2470 static void tun_set_sndbuf(struct tun_struct *tun)
2471 {
2472 struct tun_file *tfile;
2473 int i;
2474
2475 for (i = 0; i < tun->numqueues; i++) {
2476 tfile = rtnl_dereference(tun->tfiles[i]);
2477 tfile->socket.sk->sk_sndbuf = tun->sndbuf;
2478 }
2479 }
2480
2481 static int tun_set_queue(struct file *file, struct ifreq *ifr)
2482 {
2483 struct tun_file *tfile = file->private_data;
2484 struct tun_struct *tun;
2485 int ret = 0;
2486
2487 rtnl_lock();
2488
2489 if (ifr->ifr_flags & IFF_ATTACH_QUEUE) {
2490 tun = tfile->detached;
2491 if (!tun) {
2492 ret = -EINVAL;
2493 goto unlock;
2494 }
2495 ret = security_tun_dev_attach_queue(tun->security);
2496 if (ret < 0)
2497 goto unlock;
2498 ret = tun_attach(tun, file, false, tun->flags & IFF_NAPI,
2499 tun->flags & IFF_NAPI_FRAGS);
2500 } else if (ifr->ifr_flags & IFF_DETACH_QUEUE) {
2501 tun = rtnl_dereference(tfile->tun);
2502 if (!tun || !(tun->flags & IFF_MULTI_QUEUE) || tfile->detached)
2503 ret = -EINVAL;
2504 else
2505 __tun_detach(tfile, false);
2506 } else
2507 ret = -EINVAL;
2508
2509 unlock:
2510 rtnl_unlock();
2511 return ret;
2512 }
2513
2514 static long __tun_chr_ioctl(struct file *file, unsigned int cmd,
2515 unsigned long arg, int ifreq_len)
2516 {
2517 struct tun_file *tfile = file->private_data;
2518 struct tun_struct *tun;
2519 void __user* argp = (void __user*)arg;
2520 unsigned int ifindex, carrier;
2521 struct ifreq ifr;
2522 kuid_t owner;
2523 kgid_t group;
2524 int sndbuf;
2525 int vnet_hdr_sz;
2526 int le;
2527 int ret;
2528
2529 if (cmd == TUNSETIFF || cmd == TUNSETQUEUE || _IOC_TYPE(cmd) == SOCK_IOC_TYPE) {
2530 if (copy_from_user(&ifr, argp, ifreq_len))
2531 return -EFAULT;
2532 } else {
2533 memset(&ifr, 0, sizeof(ifr));
2534 }
2535 if (cmd == TUNGETFEATURES) {
2536 /* Currently this just means: "what IFF flags are valid?".
2537 * This is needed because we never checked for invalid flags on
2538 * TUNSETIFF.
2539 */
2540 return put_user(IFF_TUN | IFF_TAP | TUN_FEATURES,
2541 (unsigned int __user*)argp);
2542 } else if (cmd == TUNSETQUEUE)
2543 return tun_set_queue(file, &ifr);
2544
2545 ret = 0;
2546 rtnl_lock();
2547
2548 tun = tun_get(tfile);
2549 if (cmd == TUNSETIFF) {
2550 ret = -EEXIST;
2551 if (tun)
2552 goto unlock;
2553
2554 ifr.ifr_name[IFNAMSIZ-1] = '\0';
2555
2556 ret = tun_set_iff(sock_net(&tfile->sk), file, &ifr);
2557
2558 if (ret)
2559 goto unlock;
2560
2561 if (copy_to_user(argp, &ifr, ifreq_len))
2562 ret = -EFAULT;
2563 goto unlock;
2564 }
2565 if (cmd == TUNSETIFINDEX) {
2566 ret = -EPERM;
2567 if (tun)
2568 goto unlock;
2569
2570 ret = -EFAULT;
2571 if (copy_from_user(&ifindex, argp, sizeof(ifindex)))
2572 goto unlock;
2573
2574 ret = 0;
2575 tfile->ifindex = ifindex;
2576 goto unlock;
2577 }
2578
2579 ret = -EBADFD;
2580 if (!tun)
2581 goto unlock;
2582
2583 tun_debug(KERN_INFO, tun, "tun_chr_ioctl cmd %u\n", cmd);
2584
2585 ret = 0;
2586 switch (cmd) {
2587 case TUNGETIFF:
2588 tun_get_iff(current->nsproxy->net_ns, tun, &ifr);
2589
2590 if (tfile->detached)
2591 ifr.ifr_flags |= IFF_DETACH_QUEUE;
2592 if (!tfile->socket.sk->sk_filter)
2593 ifr.ifr_flags |= IFF_NOFILTER;
2594
2595 if (copy_to_user(argp, &ifr, ifreq_len))
2596 ret = -EFAULT;
2597 break;
2598
2599 case TUNSETNOCSUM:
2600 /* Disable/Enable checksum */
2601
2602 /* [unimplemented] */
2603 tun_debug(KERN_INFO, tun, "ignored: set checksum %s\n",
2604 arg ? "disabled" : "enabled");
2605 break;
2606
2607 case TUNSETPERSIST:
2608 /* Disable/Enable persist mode. Keep an extra reference to the
2609 * module to prevent the module being unprobed.
2610 */
2611 if (arg && !(tun->flags & IFF_PERSIST)) {
2612 tun->flags |= IFF_PERSIST;
2613 __module_get(THIS_MODULE);
2614 }
2615 if (!arg && (tun->flags & IFF_PERSIST)) {
2616 tun->flags &= ~IFF_PERSIST;
2617 module_put(THIS_MODULE);
2618 }
2619
2620 tun_debug(KERN_INFO, tun, "persist %s\n",
2621 arg ? "enabled" : "disabled");
2622 break;
2623
2624 case TUNSETOWNER:
2625 /* Set owner of the device */
2626 owner = make_kuid(current_user_ns(), arg);
2627 if (!uid_valid(owner)) {
2628 ret = -EINVAL;
2629 break;
2630 }
2631 tun->owner = owner;
2632 tun_debug(KERN_INFO, tun, "owner set to %u\n",
2633 from_kuid(&init_user_ns, tun->owner));
2634 break;
2635
2636 case TUNSETGROUP:
2637 /* Set group of the device */
2638 group = make_kgid(current_user_ns(), arg);
2639 if (!gid_valid(group)) {
2640 ret = -EINVAL;
2641 break;
2642 }
2643 tun->group = group;
2644 tun_debug(KERN_INFO, tun, "group set to %u\n",
2645 from_kgid(&init_user_ns, tun->group));
2646 break;
2647
2648 case TUNSETLINK:
2649 /* Only allow setting the type when the interface is down */
2650 if (tun->dev->flags & IFF_UP) {
2651 tun_debug(KERN_INFO, tun,
2652 "Linktype set failed because interface is up\n");
2653 ret = -EBUSY;
2654 } else {
2655 tun->dev->type = (int) arg;
2656 tun_debug(KERN_INFO, tun, "linktype set to %d\n",
2657 tun->dev->type);
2658 ret = 0;
2659 }
2660 break;
2661
2662 #ifdef TUN_DEBUG
2663 case TUNSETDEBUG:
2664 tun->debug = arg;
2665 break;
2666 #endif
2667 case TUNSETOFFLOAD:
2668 ret = set_offload(tun, arg);
2669 break;
2670
2671 case TUNSETTXFILTER:
2672 /* Can be set only for TAPs */
2673 ret = -EINVAL;
2674 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
2675 break;
2676 ret = update_filter(&tun->txflt, (void __user *)arg);
2677 break;
2678
2679 case SIOCGIFHWADDR:
2680 /* Get hw address */
2681 memcpy(ifr.ifr_hwaddr.sa_data, tun->dev->dev_addr, ETH_ALEN);
2682 ifr.ifr_hwaddr.sa_family = tun->dev->type;
2683 if (copy_to_user(argp, &ifr, ifreq_len))
2684 ret = -EFAULT;
2685 break;
2686
2687 case SIOCSIFHWADDR:
2688 /* Set hw address */
2689 tun_debug(KERN_DEBUG, tun, "set hw address: %pM\n",
2690 ifr.ifr_hwaddr.sa_data);
2691
2692 ret = dev_set_mac_address(tun->dev, &ifr.ifr_hwaddr);
2693 break;
2694
2695 case TUNGETSNDBUF:
2696 sndbuf = tfile->socket.sk->sk_sndbuf;
2697 if (copy_to_user(argp, &sndbuf, sizeof(sndbuf)))
2698 ret = -EFAULT;
2699 break;
2700
2701 case TUNSETSNDBUF:
2702 if (copy_from_user(&sndbuf, argp, sizeof(sndbuf))) {
2703 ret = -EFAULT;
2704 break;
2705 }
2706 if (sndbuf <= 0) {
2707 ret = -EINVAL;
2708 break;
2709 }
2710
2711 tun->sndbuf = sndbuf;
2712 tun_set_sndbuf(tun);
2713 break;
2714
2715 case TUNGETVNETHDRSZ:
2716 vnet_hdr_sz = tun->vnet_hdr_sz;
2717 if (copy_to_user(argp, &vnet_hdr_sz, sizeof(vnet_hdr_sz)))
2718 ret = -EFAULT;
2719 break;
2720
2721 case TUNSETVNETHDRSZ:
2722 if (copy_from_user(&vnet_hdr_sz, argp, sizeof(vnet_hdr_sz))) {
2723 ret = -EFAULT;
2724 break;
2725 }
2726 if (vnet_hdr_sz < (int)sizeof(struct virtio_net_hdr)) {
2727 ret = -EINVAL;
2728 break;
2729 }
2730
2731 tun->vnet_hdr_sz = vnet_hdr_sz;
2732 break;
2733
2734 case TUNGETVNETLE:
2735 le = !!(tun->flags & TUN_VNET_LE);
2736 if (put_user(le, (int __user *)argp))
2737 ret = -EFAULT;
2738 break;
2739
2740 case TUNSETVNETLE:
2741 if (get_user(le, (int __user *)argp)) {
2742 ret = -EFAULT;
2743 break;
2744 }
2745 if (le)
2746 tun->flags |= TUN_VNET_LE;
2747 else
2748 tun->flags &= ~TUN_VNET_LE;
2749 break;
2750
2751 case TUNGETVNETBE:
2752 ret = tun_get_vnet_be(tun, argp);
2753 break;
2754
2755 case TUNSETVNETBE:
2756 ret = tun_set_vnet_be(tun, argp);
2757 break;
2758
2759 case TUNATTACHFILTER:
2760 /* Can be set only for TAPs */
2761 ret = -EINVAL;
2762 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
2763 break;
2764 ret = -EFAULT;
2765 if (copy_from_user(&tun->fprog, argp, sizeof(tun->fprog)))
2766 break;
2767
2768 ret = tun_attach_filter(tun);
2769 break;
2770
2771 case TUNDETACHFILTER:
2772 /* Can be set only for TAPs */
2773 ret = -EINVAL;
2774 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
2775 break;
2776 ret = 0;
2777 tun_detach_filter(tun, tun->numqueues);
2778 break;
2779
2780 case TUNGETFILTER:
2781 ret = -EINVAL;
2782 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
2783 break;
2784 ret = -EFAULT;
2785 if (copy_to_user(argp, &tun->fprog, sizeof(tun->fprog)))
2786 break;
2787 ret = 0;
2788 break;
2789
2790 case TUNSETCARRIER:
2791 ret = -EFAULT;
2792 if (copy_from_user(&carrier, argp, sizeof(carrier)))
2793 goto unlock;
2794
2795 ret = tun_net_change_carrier(tun->dev, (bool)carrier);
2796 break;
2797
2798 default:
2799 ret = -EINVAL;
2800 break;
2801 }
2802
2803 unlock:
2804 rtnl_unlock();
2805 if (tun)
2806 tun_put(tun);
2807 return ret;
2808 }
2809
2810 static long tun_chr_ioctl(struct file *file,
2811 unsigned int cmd, unsigned long arg)
2812 {
2813 return __tun_chr_ioctl(file, cmd, arg, sizeof (struct ifreq));
2814 }
2815
2816 #ifdef CONFIG_COMPAT
2817 static long tun_chr_compat_ioctl(struct file *file,
2818 unsigned int cmd, unsigned long arg)
2819 {
2820 switch (cmd) {
2821 case TUNSETIFF:
2822 case TUNGETIFF:
2823 case TUNSETTXFILTER:
2824 case TUNGETSNDBUF:
2825 case TUNSETSNDBUF:
2826 case SIOCGIFHWADDR:
2827 case SIOCSIFHWADDR:
2828 arg = (unsigned long)compat_ptr(arg);
2829 break;
2830 default:
2831 arg = (compat_ulong_t)arg;
2832 break;
2833 }
2834
2835 /*
2836 * compat_ifreq is shorter than ifreq, so we must not access beyond
2837 * the end of that structure. All fields that are used in this
2838 * driver are compatible though, we don't need to convert the
2839 * contents.
2840 */
2841 return __tun_chr_ioctl(file, cmd, arg, sizeof(struct compat_ifreq));
2842 }
2843 #endif /* CONFIG_COMPAT */
2844
2845 static int tun_chr_fasync(int fd, struct file *file, int on)
2846 {
2847 struct tun_file *tfile = file->private_data;
2848 int ret;
2849
2850 if ((ret = fasync_helper(fd, file, on, &tfile->fasync)) < 0)
2851 goto out;
2852
2853 if (on) {
2854 __f_setown(file, task_pid(current), PIDTYPE_PID, 0);
2855 tfile->flags |= TUN_FASYNC;
2856 } else
2857 tfile->flags &= ~TUN_FASYNC;
2858 ret = 0;
2859 out:
2860 return ret;
2861 }
2862
2863 static int tun_chr_open(struct inode *inode, struct file * file)
2864 {
2865 struct net *net = current->nsproxy->net_ns;
2866 struct tun_file *tfile;
2867
2868 DBG1(KERN_INFO, "tunX: tun_chr_open\n");
2869
2870 tfile = (struct tun_file *)sk_alloc(net, AF_UNSPEC, GFP_KERNEL,
2871 &tun_proto, 0);
2872 if (!tfile)
2873 return -ENOMEM;
2874 if (skb_array_init(&tfile->tx_array, 0, GFP_KERNEL)) {
2875 sk_free(&tfile->sk);
2876 return -ENOMEM;
2877 }
2878
2879 mutex_init(&tfile->napi_mutex);
2880 RCU_INIT_POINTER(tfile->tun, NULL);
2881 tfile->flags = 0;
2882 tfile->ifindex = 0;
2883
2884 init_waitqueue_head(&tfile->wq.wait);
2885 RCU_INIT_POINTER(tfile->socket.wq, &tfile->wq);
2886
2887 tfile->socket.file = file;
2888 tfile->socket.ops = &tun_socket_ops;
2889
2890 sock_init_data(&tfile->socket, &tfile->sk);
2891
2892 tfile->sk.sk_write_space = tun_sock_write_space;
2893 tfile->sk.sk_sndbuf = INT_MAX;
2894
2895 file->private_data = tfile;
2896 INIT_LIST_HEAD(&tfile->next);
2897
2898 sock_set_flag(&tfile->sk, SOCK_ZEROCOPY);
2899
2900 return 0;
2901 }
2902
2903 static int tun_chr_close(struct inode *inode, struct file *file)
2904 {
2905 struct tun_file *tfile = file->private_data;
2906
2907 tun_detach(tfile, true);
2908
2909 return 0;
2910 }
2911
2912 #ifdef CONFIG_PROC_FS
2913 static void tun_chr_show_fdinfo(struct seq_file *m, struct file *file)
2914 {
2915 struct tun_file *tfile = file->private_data;
2916 struct tun_struct *tun;
2917 struct ifreq ifr;
2918
2919 memset(&ifr, 0, sizeof(ifr));
2920
2921 rtnl_lock();
2922 tun = tun_get(tfile);
2923 if (tun)
2924 tun_get_iff(current->nsproxy->net_ns, tun, &ifr);
2925 rtnl_unlock();
2926
2927 if (tun)
2928 tun_put(tun);
2929
2930 seq_printf(m, "iff:\t%s\n", ifr.ifr_name);
2931 }
2932 #endif
2933
2934 static const struct file_operations tun_fops = {
2935 .owner = THIS_MODULE,
2936 .llseek = no_llseek,
2937 .read_iter = tun_chr_read_iter,
2938 .write_iter = tun_chr_write_iter,
2939 .poll = tun_chr_poll,
2940 .unlocked_ioctl = tun_chr_ioctl,
2941 #ifdef CONFIG_COMPAT
2942 .compat_ioctl = tun_chr_compat_ioctl,
2943 #endif
2944 .open = tun_chr_open,
2945 .release = tun_chr_close,
2946 .fasync = tun_chr_fasync,
2947 #ifdef CONFIG_PROC_FS
2948 .show_fdinfo = tun_chr_show_fdinfo,
2949 #endif
2950 };
2951
2952 static struct miscdevice tun_miscdev = {
2953 .minor = TUN_MINOR,
2954 .name = "tun",
2955 .nodename = "net/tun",
2956 .fops = &tun_fops,
2957 };
2958
2959 /* ethtool interface */
2960
2961 static int tun_get_link_ksettings(struct net_device *dev,
2962 struct ethtool_link_ksettings *cmd)
2963 {
2964 ethtool_link_ksettings_zero_link_mode(cmd, supported);
2965 ethtool_link_ksettings_zero_link_mode(cmd, advertising);
2966 cmd->base.speed = SPEED_10;
2967 cmd->base.duplex = DUPLEX_FULL;
2968 cmd->base.port = PORT_TP;
2969 cmd->base.phy_address = 0;
2970 cmd->base.autoneg = AUTONEG_DISABLE;
2971 return 0;
2972 }
2973
2974 static void tun_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
2975 {
2976 struct tun_struct *tun = netdev_priv(dev);
2977
2978 strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
2979 strlcpy(info->version, DRV_VERSION, sizeof(info->version));
2980
2981 switch (tun->flags & TUN_TYPE_MASK) {
2982 case IFF_TUN:
2983 strlcpy(info->bus_info, "tun", sizeof(info->bus_info));
2984 break;
2985 case IFF_TAP:
2986 strlcpy(info->bus_info, "tap", sizeof(info->bus_info));
2987 break;
2988 }
2989 }
2990
2991 static u32 tun_get_msglevel(struct net_device *dev)
2992 {
2993 #ifdef TUN_DEBUG
2994 struct tun_struct *tun = netdev_priv(dev);
2995 return tun->debug;
2996 #else
2997 return -EOPNOTSUPP;
2998 #endif
2999 }
3000
3001 static void tun_set_msglevel(struct net_device *dev, u32 value)
3002 {
3003 #ifdef TUN_DEBUG
3004 struct tun_struct *tun = netdev_priv(dev);
3005 tun->debug = value;
3006 #endif
3007 }
3008
3009 static int tun_get_coalesce(struct net_device *dev,
3010 struct ethtool_coalesce *ec)
3011 {
3012 struct tun_struct *tun = netdev_priv(dev);
3013
3014 ec->rx_max_coalesced_frames = tun->rx_batched;
3015
3016 return 0;
3017 }
3018
3019 static int tun_set_coalesce(struct net_device *dev,
3020 struct ethtool_coalesce *ec)
3021 {
3022 struct tun_struct *tun = netdev_priv(dev);
3023
3024 if (ec->rx_max_coalesced_frames > NAPI_POLL_WEIGHT)
3025 tun->rx_batched = NAPI_POLL_WEIGHT;
3026 else
3027 tun->rx_batched = ec->rx_max_coalesced_frames;
3028
3029 return 0;
3030 }
3031
3032 static const struct ethtool_ops tun_ethtool_ops = {
3033 .get_drvinfo = tun_get_drvinfo,
3034 .get_msglevel = tun_get_msglevel,
3035 .set_msglevel = tun_set_msglevel,
3036 .get_link = ethtool_op_get_link,
3037 .get_ts_info = ethtool_op_get_ts_info,
3038 .get_coalesce = tun_get_coalesce,
3039 .set_coalesce = tun_set_coalesce,
3040 .get_link_ksettings = tun_get_link_ksettings,
3041 };
3042
3043 static int tun_queue_resize(struct tun_struct *tun)
3044 {
3045 struct net_device *dev = tun->dev;
3046 struct tun_file *tfile;
3047 struct skb_array **arrays;
3048 int n = tun->numqueues + tun->numdisabled;
3049 int ret, i;
3050
3051 arrays = kmalloc_array(n, sizeof(*arrays), GFP_KERNEL);
3052 if (!arrays)
3053 return -ENOMEM;
3054
3055 for (i = 0; i < tun->numqueues; i++) {
3056 tfile = rtnl_dereference(tun->tfiles[i]);
3057 arrays[i] = &tfile->tx_array;
3058 }
3059 list_for_each_entry(tfile, &tun->disabled, next)
3060 arrays[i++] = &tfile->tx_array;
3061
3062 ret = skb_array_resize_multiple(arrays, n,
3063 dev->tx_queue_len, GFP_KERNEL);
3064
3065 kfree(arrays);
3066 return ret;
3067 }
3068
3069 static int tun_device_event(struct notifier_block *unused,
3070 unsigned long event, void *ptr)
3071 {
3072 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
3073 struct tun_struct *tun = netdev_priv(dev);
3074 int i;
3075
3076 if (dev->rtnl_link_ops != &tun_link_ops)
3077 return NOTIFY_DONE;
3078
3079 switch (event) {
3080 case NETDEV_CHANGE_TX_QUEUE_LEN:
3081 if (tun_queue_resize(tun))
3082 return NOTIFY_BAD;
3083 break;
3084 case NETDEV_UP:
3085 for (i = 0; i < tun->numqueues; i++) {
3086 struct tun_file *tfile;
3087
3088 tfile = rtnl_dereference(tun->tfiles[i]);
3089 tfile->socket.sk->sk_write_space(tfile->socket.sk);
3090 }
3091 break;
3092 default:
3093 break;
3094 }
3095
3096 return NOTIFY_DONE;
3097 }
3098
3099 static struct notifier_block tun_notifier_block __read_mostly = {
3100 .notifier_call = tun_device_event,
3101 };
3102
3103 static int __init tun_init(void)
3104 {
3105 int ret = 0;
3106
3107 pr_info("%s, %s\n", DRV_DESCRIPTION, DRV_VERSION);
3108
3109 ret = rtnl_link_register(&tun_link_ops);
3110 if (ret) {
3111 pr_err("Can't register link_ops\n");
3112 goto err_linkops;
3113 }
3114
3115 ret = misc_register(&tun_miscdev);
3116 if (ret) {
3117 pr_err("Can't register misc device %d\n", TUN_MINOR);
3118 goto err_misc;
3119 }
3120
3121 ret = register_netdevice_notifier(&tun_notifier_block);
3122 if (ret) {
3123 pr_err("Can't register netdevice notifier\n");
3124 goto err_notifier;
3125 }
3126
3127 return 0;
3128
3129 err_notifier:
3130 misc_deregister(&tun_miscdev);
3131 err_misc:
3132 rtnl_link_unregister(&tun_link_ops);
3133 err_linkops:
3134 return ret;
3135 }
3136
3137 static void tun_cleanup(void)
3138 {
3139 misc_deregister(&tun_miscdev);
3140 rtnl_link_unregister(&tun_link_ops);
3141 unregister_netdevice_notifier(&tun_notifier_block);
3142 }
3143
3144 /* Get an underlying socket object from tun file. Returns error unless file is
3145 * attached to a device. The returned object works like a packet socket, it
3146 * can be used for sock_sendmsg/sock_recvmsg. The caller is responsible for
3147 * holding a reference to the file for as long as the socket is in use. */
3148 struct socket *tun_get_socket(struct file *file)
3149 {
3150 struct tun_file *tfile;
3151 if (file->f_op != &tun_fops)
3152 return ERR_PTR(-EINVAL);
3153 tfile = file->private_data;
3154 if (!tfile)
3155 return ERR_PTR(-EBADFD);
3156 return &tfile->socket;
3157 }
3158 EXPORT_SYMBOL_GPL(tun_get_socket);
3159
3160 struct skb_array *tun_get_skb_array(struct file *file)
3161 {
3162 struct tun_file *tfile;
3163
3164 if (file->f_op != &tun_fops)
3165 return ERR_PTR(-EINVAL);
3166 tfile = file->private_data;
3167 if (!tfile)
3168 return ERR_PTR(-EBADFD);
3169 return &tfile->tx_array;
3170 }
3171 EXPORT_SYMBOL_GPL(tun_get_skb_array);
3172
3173 module_init(tun_init);
3174 module_exit(tun_cleanup);
3175 MODULE_DESCRIPTION(DRV_DESCRIPTION);
3176 MODULE_AUTHOR(DRV_COPYRIGHT);
3177 MODULE_LICENSE("GPL");
3178 MODULE_ALIAS_MISCDEV(TUN_MINOR);
3179 MODULE_ALIAS("devname:net/tun");
ubuntu-bionic-kernel mirror