]> git.proxmox.com Git - mirror_ubuntu-bionic-kernel.git/blob - drivers/net/tun.c
projects / mirror_ubuntu-bionic-kernel.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
tun: add mutex_unlock() call and napi.skb clearing in tun_get_user()
[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_tx_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 bool publish_tun)
721 {
722 struct tun_file *tfile = file->private_data;
723 struct net_device *dev = tun->dev;
724 int err;
725
726 err = security_tun_dev_attach(tfile->socket.sk, tun->security);
727 if (err < 0)
728 goto out;
729
730 err = -EINVAL;
731 if (rtnl_dereference(tfile->tun) && !tfile->detached)
732 goto out;
733
734 err = -EBUSY;
735 if (!(tun->flags & IFF_MULTI_QUEUE) && tun->numqueues == 1)
736 goto out;
737
738 err = -E2BIG;
739 if (!tfile->detached &&
740 tun->numqueues + tun->numdisabled == MAX_TAP_QUEUES)
741 goto out;
742
743 err = 0;
744
745 /* Re-attach the filter to persist device */
746 if (!skip_filter && (tun->filter_attached == true)) {
747 lock_sock(tfile->socket.sk);
748 err = sk_attach_filter(&tun->fprog, tfile->socket.sk);
749 release_sock(tfile->socket.sk);
750 if (!err)
751 goto out;
752 }
753
754 if (!tfile->detached &&
755 skb_array_resize(&tfile->tx_array, dev->tx_queue_len, GFP_KERNEL)) {
756 err = -ENOMEM;
757 goto out;
758 }
759
760 tfile->queue_index = tun->numqueues;
761 tfile->socket.sk->sk_shutdown &= ~RCV_SHUTDOWN;
762 if (tfile->detached) {
763 tun_enable_queue(tfile);
764 } else {
765 sock_hold(&tfile->sk);
766 tun_napi_init(tun, tfile, napi, napi_frags);
767 }
768
769 /* device is allowed to go away first, so no need to hold extra
770 * refcnt.
771 */
772
773 /* Publish tfile->tun and tun->tfiles only after we've fully
774 * initialized tfile; otherwise we risk using half-initialized
775 * object.
776 */
777 if (publish_tun)
778 rcu_assign_pointer(tfile->tun, tun);
779 rcu_assign_pointer(tun->tfiles[tun->numqueues], tfile);
780 tun->numqueues++;
781 tun_set_real_num_queues(tun);
782 out:
783 return err;
784 }
785
786 static struct tun_struct *tun_get(struct tun_file *tfile)
787 {
788 struct tun_struct *tun;
789
790 rcu_read_lock();
791 tun = rcu_dereference(tfile->tun);
792 if (tun)
793 dev_hold(tun->dev);
794 rcu_read_unlock();
795
796 return tun;
797 }
798
799 static void tun_put(struct tun_struct *tun)
800 {
801 dev_put(tun->dev);
802 }
803
804 /* TAP filtering */
805 static void addr_hash_set(u32 *mask, const u8 *addr)
806 {
807 int n = ether_crc(ETH_ALEN, addr) >> 26;
808 mask[n >> 5] |= (1 << (n & 31));
809 }
810
811 static unsigned int addr_hash_test(const u32 *mask, const u8 *addr)
812 {
813 int n = ether_crc(ETH_ALEN, addr) >> 26;
814 return mask[n >> 5] & (1 << (n & 31));
815 }
816
817 static int update_filter(struct tap_filter *filter, void __user *arg)
818 {
819 struct { u8 u[ETH_ALEN]; } *addr;
820 struct tun_filter uf;
821 int err, alen, n, nexact;
822
823 if (copy_from_user(&uf, arg, sizeof(uf)))
824 return -EFAULT;
825
826 if (!uf.count) {
827 /* Disabled */
828 filter->count = 0;
829 return 0;
830 }
831
832 alen = ETH_ALEN * uf.count;
833 addr = memdup_user(arg + sizeof(uf), alen);
834 if (IS_ERR(addr))
835 return PTR_ERR(addr);
836
837 /* The filter is updated without holding any locks. Which is
838 * perfectly safe. We disable it first and in the worst
839 * case we'll accept a few undesired packets. */
840 filter->count = 0;
841 wmb();
842
843 /* Use first set of addresses as an exact filter */
844 for (n = 0; n < uf.count && n < FLT_EXACT_COUNT; n++)
845 memcpy(filter->addr[n], addr[n].u, ETH_ALEN);
846
847 nexact = n;
848
849 /* Remaining multicast addresses are hashed,
850 * unicast will leave the filter disabled. */
851 memset(filter->mask, 0, sizeof(filter->mask));
852 for (; n < uf.count; n++) {
853 if (!is_multicast_ether_addr(addr[n].u)) {
854 err = 0; /* no filter */
855 goto free_addr;
856 }
857 addr_hash_set(filter->mask, addr[n].u);
858 }
859
860 /* For ALLMULTI just set the mask to all ones.
861 * This overrides the mask populated above. */
862 if ((uf.flags & TUN_FLT_ALLMULTI))
863 memset(filter->mask, ~0, sizeof(filter->mask));
864
865 /* Now enable the filter */
866 wmb();
867 filter->count = nexact;
868
869 /* Return the number of exact filters */
870 err = nexact;
871 free_addr:
872 kfree(addr);
873 return err;
874 }
875
876 /* Returns: 0 - drop, !=0 - accept */
877 static int run_filter(struct tap_filter *filter, const struct sk_buff *skb)
878 {
879 /* Cannot use eth_hdr(skb) here because skb_mac_hdr() is incorrect
880 * at this point. */
881 struct ethhdr *eh = (struct ethhdr *) skb->data;
882 int i;
883
884 /* Exact match */
885 for (i = 0; i < filter->count; i++)
886 if (ether_addr_equal(eh->h_dest, filter->addr[i]))
887 return 1;
888
889 /* Inexact match (multicast only) */
890 if (is_multicast_ether_addr(eh->h_dest))
891 return addr_hash_test(filter->mask, eh->h_dest);
892
893 return 0;
894 }
895
896 /*
897 * Checks whether the packet is accepted or not.
898 * Returns: 0 - drop, !=0 - accept
899 */
900 static int check_filter(struct tap_filter *filter, const struct sk_buff *skb)
901 {
902 if (!filter->count)
903 return 1;
904
905 return run_filter(filter, skb);
906 }
907
908 /* Network device part of the driver */
909
910 static const struct ethtool_ops tun_ethtool_ops;
911
912 /* Net device detach from fd. */
913 static void tun_net_uninit(struct net_device *dev)
914 {
915 tun_detach_all(dev);
916 }
917
918 /* Net device open. */
919 static int tun_net_open(struct net_device *dev)
920 {
921 netif_tx_start_all_queues(dev);
922
923 return 0;
924 }
925
926 /* Net device close. */
927 static int tun_net_close(struct net_device *dev)
928 {
929 netif_tx_stop_all_queues(dev);
930 return 0;
931 }
932
933 /* Net device start xmit */
934 static netdev_tx_t tun_net_xmit(struct sk_buff *skb, struct net_device *dev)
935 {
936 struct tun_struct *tun = netdev_priv(dev);
937 int txq = skb->queue_mapping;
938 struct tun_file *tfile;
939 u32 numqueues = 0;
940
941 rcu_read_lock();
942 tfile = rcu_dereference(tun->tfiles[txq]);
943 numqueues = READ_ONCE(tun->numqueues);
944
945 /* Drop packet if interface is not attached */
946 if (txq >= numqueues)
947 goto drop;
948
949 #ifdef CONFIG_RPS
950 if (numqueues == 1 && static_key_false(&rps_needed)) {
951 /* Select queue was not called for the skbuff, so we extract the
952 * RPS hash and save it into the flow_table here.
953 */
954 __u32 rxhash;
955
956 rxhash = __skb_get_hash_symmetric(skb);
957 if (rxhash) {
958 struct tun_flow_entry *e;
959 e = tun_flow_find(&tun->flows[tun_hashfn(rxhash)],
960 rxhash);
961 if (e)
962 tun_flow_save_rps_rxhash(e, rxhash);
963 }
964 }
965 #endif
966
967 tun_debug(KERN_INFO, tun, "tun_net_xmit %d\n", skb->len);
968
969 BUG_ON(!tfile);
970
971 /* Drop if the filter does not like it.
972 * This is a noop if the filter is disabled.
973 * Filter can be enabled only for the TAP devices. */
974 if (!check_filter(&tun->txflt, skb))
975 goto drop;
976
977 if (tfile->socket.sk->sk_filter &&
978 sk_filter(tfile->socket.sk, skb))
979 goto drop;
980
981 if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC)))
982 goto drop;
983
984 skb_tx_timestamp(skb);
985
986 /* Orphan the skb - required as we might hang on to it
987 * for indefinite time.
988 */
989 skb_orphan(skb);
990
991 nf_reset(skb);
992
993 if (skb_array_produce(&tfile->tx_array, skb))
994 goto drop;
995
996 /* Notify and wake up reader process */
997 if (tfile->flags & TUN_FASYNC)
998 kill_fasync(&tfile->fasync, SIGIO, POLL_IN);
999 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
1000
1001 rcu_read_unlock();
1002 return NETDEV_TX_OK;
1003
1004 drop:
1005 this_cpu_inc(tun->pcpu_stats->tx_dropped);
1006 skb_tx_error(skb);
1007 kfree_skb(skb);
1008 rcu_read_unlock();
1009 return NET_XMIT_DROP;
1010 }
1011
1012 static void tun_net_mclist(struct net_device *dev)
1013 {
1014 /*
1015 * This callback is supposed to deal with mc filter in
1016 * _rx_ path and has nothing to do with the _tx_ path.
1017 * In rx path we always accept everything userspace gives us.
1018 */
1019 }
1020
1021 static netdev_features_t tun_net_fix_features(struct net_device *dev,
1022 netdev_features_t features)
1023 {
1024 struct tun_struct *tun = netdev_priv(dev);
1025
1026 return (features & tun->set_features) | (features & ~TUN_USER_FEATURES);
1027 }
1028 #ifdef CONFIG_NET_POLL_CONTROLLER
1029 static void tun_poll_controller(struct net_device *dev)
1030 {
1031 /*
1032 * Tun only receives frames when:
1033 * 1) the char device endpoint gets data from user space
1034 * 2) the tun socket gets a sendmsg call from user space
1035 * If NAPI is not enabled, since both of those are synchronous
1036 * operations, we are guaranteed never to have pending data when we poll
1037 * for it so there is nothing to do here but return.
1038 * We need this though so netpoll recognizes us as an interface that
1039 * supports polling, which enables bridge devices in virt setups to
1040 * still use netconsole
1041 * If NAPI is enabled, however, we need to schedule polling for all
1042 * queues unless we are using napi_gro_frags(), which we call in
1043 * process context and not in NAPI context.
1044 */
1045 struct tun_struct *tun = netdev_priv(dev);
1046
1047 if (tun->flags & IFF_NAPI) {
1048 struct tun_file *tfile;
1049 int i;
1050
1051 rcu_read_lock();
1052 for (i = 0; i < tun->numqueues; i++) {
1053 tfile = rcu_dereference(tun->tfiles[i]);
1054 if (!tun_napi_frags_enabled(tfile) &&
1055 tfile->napi_enabled)
1056 napi_schedule(&tfile->napi);
1057 }
1058 rcu_read_unlock();
1059 }
1060 return;
1061 }
1062 #endif
1063
1064 static void tun_set_headroom(struct net_device *dev, int new_hr)
1065 {
1066 struct tun_struct *tun = netdev_priv(dev);
1067
1068 if (new_hr < NET_SKB_PAD)
1069 new_hr = NET_SKB_PAD;
1070
1071 tun->align = new_hr;
1072 }
1073
1074 static void
1075 tun_net_get_stats64(struct net_device *dev, struct rtnl_link_stats64 *stats)
1076 {
1077 u32 rx_dropped = 0, tx_dropped = 0, rx_frame_errors = 0;
1078 struct tun_struct *tun = netdev_priv(dev);
1079 struct tun_pcpu_stats *p;
1080 int i;
1081
1082 for_each_possible_cpu(i) {
1083 u64 rxpackets, rxbytes, txpackets, txbytes;
1084 unsigned int start;
1085
1086 p = per_cpu_ptr(tun->pcpu_stats, i);
1087 do {
1088 start = u64_stats_fetch_begin(&p->syncp);
1089 rxpackets = p->rx_packets;
1090 rxbytes = p->rx_bytes;
1091 txpackets = p->tx_packets;
1092 txbytes = p->tx_bytes;
1093 } while (u64_stats_fetch_retry(&p->syncp, start));
1094
1095 stats->rx_packets += rxpackets;
1096 stats->rx_bytes += rxbytes;
1097 stats->tx_packets += txpackets;
1098 stats->tx_bytes += txbytes;
1099
1100 /* u32 counters */
1101 rx_dropped += p->rx_dropped;
1102 rx_frame_errors += p->rx_frame_errors;
1103 tx_dropped += p->tx_dropped;
1104 }
1105 stats->rx_dropped = rx_dropped;
1106 stats->rx_frame_errors = rx_frame_errors;
1107 stats->tx_dropped = tx_dropped;
1108 }
1109
1110 static int tun_xdp_set(struct net_device *dev, struct bpf_prog *prog,
1111 struct netlink_ext_ack *extack)
1112 {
1113 struct tun_struct *tun = netdev_priv(dev);
1114 struct bpf_prog *old_prog;
1115
1116 old_prog = rtnl_dereference(tun->xdp_prog);
1117 rcu_assign_pointer(tun->xdp_prog, prog);
1118 if (old_prog)
1119 bpf_prog_put(old_prog);
1120
1121 return 0;
1122 }
1123
1124 static u32 tun_xdp_query(struct net_device *dev)
1125 {
1126 struct tun_struct *tun = netdev_priv(dev);
1127 const struct bpf_prog *xdp_prog;
1128
1129 xdp_prog = rtnl_dereference(tun->xdp_prog);
1130 if (xdp_prog)
1131 return xdp_prog->aux->id;
1132
1133 return 0;
1134 }
1135
1136 static int tun_xdp(struct net_device *dev, struct netdev_bpf *xdp)
1137 {
1138 switch (xdp->command) {
1139 case XDP_SETUP_PROG:
1140 return tun_xdp_set(dev, xdp->prog, xdp->extack);
1141 case XDP_QUERY_PROG:
1142 xdp->prog_id = tun_xdp_query(dev);
1143 xdp->prog_attached = !!xdp->prog_id;
1144 return 0;
1145 default:
1146 return -EINVAL;
1147 }
1148 }
1149
1150 static int tun_net_change_carrier(struct net_device *dev, bool new_carrier)
1151 {
1152 if (new_carrier) {
1153 struct tun_struct *tun = netdev_priv(dev);
1154
1155 if (!tun->numqueues)
1156 return -EPERM;
1157
1158 netif_carrier_on(dev);
1159 } else {
1160 netif_carrier_off(dev);
1161 }
1162 return 0;
1163 }
1164
1165 static const struct net_device_ops tun_netdev_ops = {
1166 .ndo_uninit = tun_net_uninit,
1167 .ndo_open = tun_net_open,
1168 .ndo_stop = tun_net_close,
1169 .ndo_start_xmit = tun_net_xmit,
1170 .ndo_fix_features = tun_net_fix_features,
1171 .ndo_select_queue = tun_select_queue,
1172 #ifdef CONFIG_NET_POLL_CONTROLLER
1173 .ndo_poll_controller = tun_poll_controller,
1174 #endif
1175 .ndo_set_rx_headroom = tun_set_headroom,
1176 .ndo_get_stats64 = tun_net_get_stats64,
1177 .ndo_change_carrier = tun_net_change_carrier,
1178 };
1179
1180 static const struct net_device_ops tap_netdev_ops = {
1181 .ndo_uninit = tun_net_uninit,
1182 .ndo_open = tun_net_open,
1183 .ndo_stop = tun_net_close,
1184 .ndo_start_xmit = tun_net_xmit,
1185 .ndo_fix_features = tun_net_fix_features,
1186 .ndo_set_rx_mode = tun_net_mclist,
1187 .ndo_set_mac_address = eth_mac_addr,
1188 .ndo_validate_addr = eth_validate_addr,
1189 .ndo_select_queue = tun_select_queue,
1190 #ifdef CONFIG_NET_POLL_CONTROLLER
1191 .ndo_poll_controller = tun_poll_controller,
1192 #endif
1193 .ndo_features_check = passthru_features_check,
1194 .ndo_set_rx_headroom = tun_set_headroom,
1195 .ndo_get_stats64 = tun_net_get_stats64,
1196 .ndo_bpf = tun_xdp,
1197 .ndo_change_carrier = tun_net_change_carrier,
1198 };
1199
1200 static void tun_flow_init(struct tun_struct *tun)
1201 {
1202 int i;
1203
1204 for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++)
1205 INIT_HLIST_HEAD(&tun->flows[i]);
1206
1207 tun->ageing_time = TUN_FLOW_EXPIRE;
1208 timer_setup(&tun->flow_gc_timer, tun_flow_cleanup, 0);
1209 mod_timer(&tun->flow_gc_timer,
1210 round_jiffies_up(jiffies + tun->ageing_time));
1211 }
1212
1213 static void tun_flow_uninit(struct tun_struct *tun)
1214 {
1215 del_timer_sync(&tun->flow_gc_timer);
1216 tun_flow_flush(tun);
1217 }
1218
1219 #define MIN_MTU 68
1220 #define MAX_MTU 65535
1221
1222 /* Initialize net device. */
1223 static void tun_net_init(struct net_device *dev)
1224 {
1225 struct tun_struct *tun = netdev_priv(dev);
1226
1227 switch (tun->flags & TUN_TYPE_MASK) {
1228 case IFF_TUN:
1229 dev->netdev_ops = &tun_netdev_ops;
1230
1231 /* Point-to-Point TUN Device */
1232 dev->hard_header_len = 0;
1233 dev->addr_len = 0;
1234 dev->mtu = 1500;
1235
1236 /* Zero header length */
1237 dev->type = ARPHRD_NONE;
1238 dev->flags = IFF_POINTOPOINT | IFF_NOARP | IFF_MULTICAST;
1239 break;
1240
1241 case IFF_TAP:
1242 dev->netdev_ops = &tap_netdev_ops;
1243 /* Ethernet TAP Device */
1244 ether_setup(dev);
1245 dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1246 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
1247
1248 eth_hw_addr_random(dev);
1249
1250 break;
1251 }
1252
1253 dev->min_mtu = MIN_MTU;
1254 dev->max_mtu = MAX_MTU - dev->hard_header_len;
1255 }
1256
1257 static bool tun_sock_writeable(struct tun_struct *tun, struct tun_file *tfile)
1258 {
1259 struct sock *sk = tfile->socket.sk;
1260
1261 return (tun->dev->flags & IFF_UP) && sock_writeable(sk);
1262 }
1263
1264 /* Character device part */
1265
1266 /* Poll */
1267 static unsigned int tun_chr_poll(struct file *file, poll_table *wait)
1268 {
1269 struct tun_file *tfile = file->private_data;
1270 struct tun_struct *tun = tun_get(tfile);
1271 struct sock *sk;
1272 unsigned int mask = 0;
1273
1274 if (!tun)
1275 return POLLERR;
1276
1277 sk = tfile->socket.sk;
1278
1279 tun_debug(KERN_INFO, tun, "tun_chr_poll\n");
1280
1281 poll_wait(file, sk_sleep(sk), wait);
1282
1283 if (!skb_array_empty(&tfile->tx_array))
1284 mask |= POLLIN | POLLRDNORM;
1285
1286 /* Make sure SOCKWQ_ASYNC_NOSPACE is set if not writable to
1287 * guarantee EPOLLOUT to be raised by either here or
1288 * tun_sock_write_space(). Then process could get notification
1289 * after it writes to a down device and meets -EIO.
1290 */
1291 if (tun_sock_writeable(tun, tfile) ||
1292 (!test_and_set_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags) &&
1293 tun_sock_writeable(tun, tfile)))
1294 mask |= POLLOUT | POLLWRNORM;
1295
1296 if (tun->dev->reg_state != NETREG_REGISTERED)
1297 mask = POLLERR;
1298
1299 tun_put(tun);
1300 return mask;
1301 }
1302
1303 static struct sk_buff *tun_napi_alloc_frags(struct tun_file *tfile,
1304 size_t len,
1305 const struct iov_iter *it)
1306 {
1307 struct sk_buff *skb;
1308 size_t linear;
1309 int err;
1310 int i;
1311
1312 if (it->nr_segs > MAX_SKB_FRAGS + 1)
1313 return ERR_PTR(-ENOMEM);
1314
1315 local_bh_disable();
1316 skb = napi_get_frags(&tfile->napi);
1317 local_bh_enable();
1318 if (!skb)
1319 return ERR_PTR(-ENOMEM);
1320
1321 linear = iov_iter_single_seg_count(it);
1322 err = __skb_grow(skb, linear);
1323 if (err)
1324 goto free;
1325
1326 skb->len = len;
1327 skb->data_len = len - linear;
1328 skb->truesize += skb->data_len;
1329
1330 for (i = 1; i < it->nr_segs; i++) {
1331 struct page_frag *pfrag = &current->task_frag;
1332 size_t fragsz = it->iov[i].iov_len;
1333
1334 if (fragsz == 0 || fragsz > PAGE_SIZE) {
1335 err = -EINVAL;
1336 goto free;
1337 }
1338
1339 if (!skb_page_frag_refill(fragsz, pfrag, GFP_KERNEL)) {
1340 err = -ENOMEM;
1341 goto free;
1342 }
1343
1344 skb_fill_page_desc(skb, i - 1, pfrag->page,
1345 pfrag->offset, fragsz);
1346 page_ref_inc(pfrag->page);
1347 pfrag->offset += fragsz;
1348 }
1349
1350 return skb;
1351 free:
1352 /* frees skb and all frags allocated with napi_alloc_frag() */
1353 napi_free_frags(&tfile->napi);
1354 return ERR_PTR(err);
1355 }
1356
1357 /* prepad is the amount to reserve at front. len is length after that.
1358 * linear is a hint as to how much to copy (usually headers). */
1359 static struct sk_buff *tun_alloc_skb(struct tun_file *tfile,
1360 size_t prepad, size_t len,
1361 size_t linear, int noblock)
1362 {
1363 struct sock *sk = tfile->socket.sk;
1364 struct sk_buff *skb;
1365 int err;
1366
1367 /* Under a page? Don't bother with paged skb. */
1368 if (prepad + len < PAGE_SIZE || !linear)
1369 linear = len;
1370
1371 skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
1372 &err, 0);
1373 if (!skb)
1374 return ERR_PTR(err);
1375
1376 skb_reserve(skb, prepad);
1377 skb_put(skb, linear);
1378 skb->data_len = len - linear;
1379 skb->len += len - linear;
1380
1381 return skb;
1382 }
1383
1384 static void tun_rx_batched(struct tun_struct *tun, struct tun_file *tfile,
1385 struct sk_buff *skb, int more)
1386 {
1387 struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
1388 struct sk_buff_head process_queue;
1389 u32 rx_batched = tun->rx_batched;
1390 bool rcv = false;
1391
1392 if (!rx_batched || (!more && skb_queue_empty(queue))) {
1393 local_bh_disable();
1394 skb_record_rx_queue(skb, tfile->queue_index);
1395 netif_receive_skb(skb);
1396 local_bh_enable();
1397 return;
1398 }
1399
1400 spin_lock(&queue->lock);
1401 if (!more || skb_queue_len(queue) == rx_batched) {
1402 __skb_queue_head_init(&process_queue);
1403 skb_queue_splice_tail_init(queue, &process_queue);
1404 rcv = true;
1405 } else {
1406 __skb_queue_tail(queue, skb);
1407 }
1408 spin_unlock(&queue->lock);
1409
1410 if (rcv) {
1411 struct sk_buff *nskb;
1412
1413 local_bh_disable();
1414 while ((nskb = __skb_dequeue(&process_queue))) {
1415 skb_record_rx_queue(nskb, tfile->queue_index);
1416 netif_receive_skb(nskb);
1417 }
1418 skb_record_rx_queue(skb, tfile->queue_index);
1419 netif_receive_skb(skb);
1420 local_bh_enable();
1421 }
1422 }
1423
1424 static bool tun_can_build_skb(struct tun_struct *tun, struct tun_file *tfile,
1425 int len, int noblock, bool zerocopy)
1426 {
1427 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
1428 return false;
1429
1430 if (tfile->socket.sk->sk_sndbuf != INT_MAX)
1431 return false;
1432
1433 if (!noblock)
1434 return false;
1435
1436 if (zerocopy)
1437 return false;
1438
1439 if (SKB_DATA_ALIGN(len + TUN_RX_PAD) +
1440 SKB_DATA_ALIGN(sizeof(struct skb_shared_info)) > PAGE_SIZE)
1441 return false;
1442
1443 return true;
1444 }
1445
1446 static struct sk_buff *tun_build_skb(struct tun_struct *tun,
1447 struct tun_file *tfile,
1448 struct iov_iter *from,
1449 struct virtio_net_hdr *hdr,
1450 int len, int *skb_xdp)
1451 {
1452 struct page_frag *alloc_frag = &current->task_frag;
1453 struct sk_buff *skb;
1454 struct bpf_prog *xdp_prog;
1455 int buflen = SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
1456 unsigned int delta = 0;
1457 char *buf;
1458 size_t copied;
1459 bool xdp_xmit = false;
1460 int err, pad = TUN_RX_PAD;
1461
1462 rcu_read_lock();
1463 xdp_prog = rcu_dereference(tun->xdp_prog);
1464 if (xdp_prog)
1465 pad += TUN_HEADROOM;
1466 buflen += SKB_DATA_ALIGN(len + pad);
1467 rcu_read_unlock();
1468
1469 alloc_frag->offset = ALIGN((u64)alloc_frag->offset, SMP_CACHE_BYTES);
1470 if (unlikely(!skb_page_frag_refill(buflen, alloc_frag, GFP_KERNEL)))
1471 return ERR_PTR(-ENOMEM);
1472
1473 buf = (char *)page_address(alloc_frag->page) + alloc_frag->offset;
1474 copied = copy_page_from_iter(alloc_frag->page,
1475 alloc_frag->offset + pad,
1476 len, from);
1477 if (copied != len)
1478 return ERR_PTR(-EFAULT);
1479
1480 /* There's a small window that XDP may be set after the check
1481 * of xdp_prog above, this should be rare and for simplicity
1482 * we do XDP on skb in case the headroom is not enough.
1483 */
1484 if (hdr->gso_type || !xdp_prog)
1485 *skb_xdp = 1;
1486 else
1487 *skb_xdp = 0;
1488
1489 local_bh_disable();
1490 rcu_read_lock();
1491 xdp_prog = rcu_dereference(tun->xdp_prog);
1492 if (xdp_prog && !*skb_xdp) {
1493 struct xdp_buff xdp;
1494 void *orig_data;
1495 u32 act;
1496
1497 xdp.data_hard_start = buf;
1498 xdp.data = buf + pad;
1499 xdp_set_data_meta_invalid(&xdp);
1500 xdp.data_end = xdp.data + len;
1501 orig_data = xdp.data;
1502 act = bpf_prog_run_xdp(xdp_prog, &xdp);
1503
1504 switch (act) {
1505 case XDP_REDIRECT:
1506 get_page(alloc_frag->page);
1507 alloc_frag->offset += buflen;
1508 err = xdp_do_redirect(tun->dev, &xdp, xdp_prog);
1509 xdp_do_flush_map();
1510 if (err)
1511 goto err_redirect;
1512 rcu_read_unlock();
1513 local_bh_enable();
1514 return NULL;
1515 case XDP_TX:
1516 xdp_xmit = true;
1517 /* fall through */
1518 case XDP_PASS:
1519 delta = orig_data - xdp.data;
1520 break;
1521 default:
1522 bpf_warn_invalid_xdp_action(act);
1523 /* fall through */
1524 case XDP_ABORTED:
1525 trace_xdp_exception(tun->dev, xdp_prog, act);
1526 /* fall through */
1527 case XDP_DROP:
1528 goto err_xdp;
1529 }
1530 }
1531
1532 skb = build_skb(buf, buflen);
1533 if (!skb) {
1534 rcu_read_unlock();
1535 local_bh_enable();
1536 return ERR_PTR(-ENOMEM);
1537 }
1538
1539 skb_reserve(skb, pad - delta);
1540 skb_put(skb, len + delta);
1541 skb_set_owner_w(skb, tfile->socket.sk);
1542 get_page(alloc_frag->page);
1543 alloc_frag->offset += buflen;
1544
1545 if (xdp_xmit) {
1546 skb->dev = tun->dev;
1547 generic_xdp_tx(skb, xdp_prog);
1548 rcu_read_unlock();
1549 local_bh_enable();
1550 return NULL;
1551 }
1552
1553 rcu_read_unlock();
1554 local_bh_enable();
1555
1556 return skb;
1557
1558 err_redirect:
1559 put_page(alloc_frag->page);
1560 err_xdp:
1561 rcu_read_unlock();
1562 local_bh_enable();
1563 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1564 return NULL;
1565 }
1566
1567 /* Get packet from user space buffer */
1568 static ssize_t tun_get_user(struct tun_struct *tun, struct tun_file *tfile,
1569 void *msg_control, struct iov_iter *from,
1570 int noblock, bool more)
1571 {
1572 struct tun_pi pi = { 0, cpu_to_be16(ETH_P_IP) };
1573 struct sk_buff *skb;
1574 size_t total_len = iov_iter_count(from);
1575 size_t len = total_len, align = tun->align, linear;
1576 struct virtio_net_hdr gso = { 0 };
1577 struct tun_pcpu_stats *stats;
1578 int good_linear;
1579 int copylen;
1580 bool zerocopy = false;
1581 int err;
1582 u32 rxhash;
1583 int skb_xdp = 1;
1584 bool frags = tun_napi_frags_enabled(tfile);
1585
1586 if (!(tun->flags & IFF_NO_PI)) {
1587 if (len < sizeof(pi))
1588 return -EINVAL;
1589 len -= sizeof(pi);
1590
1591 if (!copy_from_iter_full(&pi, sizeof(pi), from))
1592 return -EFAULT;
1593 }
1594
1595 if (tun->flags & IFF_VNET_HDR) {
1596 int vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
1597
1598 if (len < vnet_hdr_sz)
1599 return -EINVAL;
1600 len -= vnet_hdr_sz;
1601
1602 if (!copy_from_iter_full(&gso, sizeof(gso), from))
1603 return -EFAULT;
1604
1605 if ((gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
1606 tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2 > tun16_to_cpu(tun, gso.hdr_len))
1607 gso.hdr_len = cpu_to_tun16(tun, tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2);
1608
1609 if (tun16_to_cpu(tun, gso.hdr_len) > len)
1610 return -EINVAL;
1611 iov_iter_advance(from, vnet_hdr_sz - sizeof(gso));
1612 }
1613
1614 if ((tun->flags & TUN_TYPE_MASK) == IFF_TAP) {
1615 align += NET_IP_ALIGN;
1616 if (unlikely(len < ETH_HLEN ||
1617 (gso.hdr_len && tun16_to_cpu(tun, gso.hdr_len) < ETH_HLEN)))
1618 return -EINVAL;
1619 }
1620
1621 good_linear = SKB_MAX_HEAD(align);
1622
1623 if (msg_control) {
1624 struct iov_iter i = *from;
1625
1626 /* There are 256 bytes to be copied in skb, so there is
1627 * enough room for skb expand head in case it is used.
1628 * The rest of the buffer is mapped from userspace.
1629 */
1630 copylen = gso.hdr_len ? tun16_to_cpu(tun, gso.hdr_len) : GOODCOPY_LEN;
1631 if (copylen > good_linear)
1632 copylen = good_linear;
1633 linear = copylen;
1634 iov_iter_advance(&i, copylen);
1635 if (iov_iter_npages(&i, INT_MAX) <= MAX_SKB_FRAGS)
1636 zerocopy = true;
1637 }
1638
1639 if (!frags && tun_can_build_skb(tun, tfile, len, noblock, zerocopy)) {
1640 /* For the packet that is not easy to be processed
1641 * (e.g gso or jumbo packet), we will do it at after
1642 * skb was created with generic XDP routine.
1643 */
1644 skb = tun_build_skb(tun, tfile, from, &gso, len, &skb_xdp);
1645 if (IS_ERR(skb)) {
1646 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1647 return PTR_ERR(skb);
1648 }
1649 if (!skb)
1650 return total_len;
1651 } else {
1652 if (!zerocopy) {
1653 copylen = len;
1654 if (tun16_to_cpu(tun, gso.hdr_len) > good_linear)
1655 linear = good_linear;
1656 else
1657 linear = tun16_to_cpu(tun, gso.hdr_len);
1658 }
1659
1660 if (frags) {
1661 mutex_lock(&tfile->napi_mutex);
1662 skb = tun_napi_alloc_frags(tfile, copylen, from);
1663 /* tun_napi_alloc_frags() enforces a layout for the skb.
1664 * If zerocopy is enabled, then this layout will be
1665 * overwritten by zerocopy_sg_from_iter().
1666 */
1667 zerocopy = false;
1668 } else {
1669 skb = tun_alloc_skb(tfile, align, copylen, linear,
1670 noblock);
1671 }
1672
1673 if (IS_ERR(skb)) {
1674 if (PTR_ERR(skb) != -EAGAIN)
1675 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1676 if (frags)
1677 mutex_unlock(&tfile->napi_mutex);
1678 return PTR_ERR(skb);
1679 }
1680
1681 if (zerocopy)
1682 err = zerocopy_sg_from_iter(skb, from);
1683 else
1684 err = skb_copy_datagram_from_iter(skb, 0, from, len);
1685
1686 if (err) {
1687 err = -EFAULT;
1688 drop:
1689 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1690 kfree_skb(skb);
1691 if (frags) {
1692 tfile->napi.skb = NULL;
1693 mutex_unlock(&tfile->napi_mutex);
1694 }
1695
1696 return err;
1697 }
1698 }
1699
1700 if (virtio_net_hdr_to_skb(skb, &gso, tun_is_little_endian(tun))) {
1701 this_cpu_inc(tun->pcpu_stats->rx_frame_errors);
1702 kfree_skb(skb);
1703 if (frags) {
1704 tfile->napi.skb = NULL;
1705 mutex_unlock(&tfile->napi_mutex);
1706 }
1707
1708 return -EINVAL;
1709 }
1710
1711 switch (tun->flags & TUN_TYPE_MASK) {
1712 case IFF_TUN:
1713 if (tun->flags & IFF_NO_PI) {
1714 u8 ip_version = skb->len ? (skb->data[0] >> 4) : 0;
1715
1716 switch (ip_version) {
1717 case 4:
1718 pi.proto = htons(ETH_P_IP);
1719 break;
1720 case 6:
1721 pi.proto = htons(ETH_P_IPV6);
1722 break;
1723 default:
1724 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1725 kfree_skb(skb);
1726 return -EINVAL;
1727 }
1728 }
1729
1730 skb_reset_mac_header(skb);
1731 skb->protocol = pi.proto;
1732 skb->dev = tun->dev;
1733 break;
1734 case IFF_TAP:
1735 if (!frags)
1736 skb->protocol = eth_type_trans(skb, tun->dev);
1737 break;
1738 }
1739
1740 /* copy skb_ubuf_info for callback when skb has no error */
1741 if (zerocopy) {
1742 skb_shinfo(skb)->destructor_arg = msg_control;
1743 skb_shinfo(skb)->tx_flags |= SKBTX_DEV_ZEROCOPY;
1744 skb_shinfo(skb)->tx_flags |= SKBTX_SHARED_FRAG;
1745 } else if (msg_control) {
1746 struct ubuf_info *uarg = msg_control;
1747 uarg->callback(uarg, false);
1748 }
1749
1750 skb_reset_network_header(skb);
1751 skb_probe_transport_header(skb, 0);
1752
1753 if (skb_xdp) {
1754 struct bpf_prog *xdp_prog;
1755 int ret;
1756
1757 local_bh_disable();
1758 rcu_read_lock();
1759 xdp_prog = rcu_dereference(tun->xdp_prog);
1760 if (xdp_prog) {
1761 ret = do_xdp_generic(xdp_prog, skb);
1762 if (ret != XDP_PASS) {
1763 rcu_read_unlock();
1764 local_bh_enable();
1765 if (frags) {
1766 tfile->napi.skb = NULL;
1767 mutex_unlock(&tfile->napi_mutex);
1768 }
1769 return total_len;
1770 }
1771 }
1772 rcu_read_unlock();
1773 local_bh_enable();
1774 }
1775
1776 rxhash = __skb_get_hash_symmetric(skb);
1777
1778 rcu_read_lock();
1779 if (unlikely(!(tun->dev->flags & IFF_UP))) {
1780 err = -EIO;
1781 rcu_read_unlock();
1782 goto drop;
1783 }
1784
1785 if (frags) {
1786 /* Exercise flow dissector code path. */
1787 u32 headlen = eth_get_headlen(skb->data, skb_headlen(skb));
1788
1789 if (unlikely(headlen > skb_headlen(skb))) {
1790 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1791 napi_free_frags(&tfile->napi);
1792 rcu_read_unlock();
1793 mutex_unlock(&tfile->napi_mutex);
1794 WARN_ON(1);
1795 return -ENOMEM;
1796 }
1797
1798 local_bh_disable();
1799 napi_gro_frags(&tfile->napi);
1800 local_bh_enable();
1801 mutex_unlock(&tfile->napi_mutex);
1802 } else if (tfile->napi_enabled) {
1803 struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
1804 int queue_len;
1805
1806 spin_lock_bh(&queue->lock);
1807 __skb_queue_tail(queue, skb);
1808 queue_len = skb_queue_len(queue);
1809 spin_unlock(&queue->lock);
1810
1811 if (!more || queue_len > NAPI_POLL_WEIGHT)
1812 napi_schedule(&tfile->napi);
1813
1814 local_bh_enable();
1815 } else if (!IS_ENABLED(CONFIG_4KSTACKS)) {
1816 tun_rx_batched(tun, tfile, skb, more);
1817 } else {
1818 netif_rx_ni(skb);
1819 }
1820 rcu_read_unlock();
1821
1822 stats = get_cpu_ptr(tun->pcpu_stats);
1823 u64_stats_update_begin(&stats->syncp);
1824 stats->rx_packets++;
1825 stats->rx_bytes += len;
1826 u64_stats_update_end(&stats->syncp);
1827 put_cpu_ptr(stats);
1828
1829 tun_flow_update(tun, rxhash, tfile);
1830 return total_len;
1831 }
1832
1833 static ssize_t tun_chr_write_iter(struct kiocb *iocb, struct iov_iter *from)
1834 {
1835 struct file *file = iocb->ki_filp;
1836 struct tun_file *tfile = file->private_data;
1837 struct tun_struct *tun = tun_get(tfile);
1838 ssize_t result;
1839
1840 if (!tun)
1841 return -EBADFD;
1842
1843 result = tun_get_user(tun, tfile, NULL, from,
1844 file->f_flags & O_NONBLOCK, false);
1845
1846 tun_put(tun);
1847 return result;
1848 }
1849
1850 /* Put packet to the user space buffer */
1851 static ssize_t tun_put_user(struct tun_struct *tun,
1852 struct tun_file *tfile,
1853 struct sk_buff *skb,
1854 struct iov_iter *iter)
1855 {
1856 struct tun_pi pi = { 0, skb->protocol };
1857 struct tun_pcpu_stats *stats;
1858 ssize_t total;
1859 int vlan_offset = 0;
1860 int vlan_hlen = 0;
1861 int vnet_hdr_sz = 0;
1862
1863 if (skb_vlan_tag_present(skb))
1864 vlan_hlen = VLAN_HLEN;
1865
1866 if (tun->flags & IFF_VNET_HDR)
1867 vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
1868
1869 total = skb->len + vlan_hlen + vnet_hdr_sz;
1870
1871 if (!(tun->flags & IFF_NO_PI)) {
1872 if (iov_iter_count(iter) < sizeof(pi))
1873 return -EINVAL;
1874
1875 total += sizeof(pi);
1876 if (iov_iter_count(iter) < total) {
1877 /* Packet will be striped */
1878 pi.flags |= TUN_PKT_STRIP;
1879 }
1880
1881 if (copy_to_iter(&pi, sizeof(pi), iter) != sizeof(pi))
1882 return -EFAULT;
1883 }
1884
1885 if (vnet_hdr_sz) {
1886 struct virtio_net_hdr gso;
1887
1888 if (iov_iter_count(iter) < vnet_hdr_sz)
1889 return -EINVAL;
1890
1891 if (virtio_net_hdr_from_skb(skb, &gso,
1892 tun_is_little_endian(tun), true,
1893 vlan_hlen)) {
1894 struct skb_shared_info *sinfo = skb_shinfo(skb);
1895 pr_err("unexpected GSO type: "
1896 "0x%x, gso_size %d, hdr_len %d\n",
1897 sinfo->gso_type, tun16_to_cpu(tun, gso.gso_size),
1898 tun16_to_cpu(tun, gso.hdr_len));
1899 print_hex_dump(KERN_ERR, "tun: ",
1900 DUMP_PREFIX_NONE,
1901 16, 1, skb->head,
1902 min((int)tun16_to_cpu(tun, gso.hdr_len), 64), true);
1903 WARN_ON_ONCE(1);
1904 return -EINVAL;
1905 }
1906
1907 if (copy_to_iter(&gso, sizeof(gso), iter) != sizeof(gso))
1908 return -EFAULT;
1909
1910 iov_iter_advance(iter, vnet_hdr_sz - sizeof(gso));
1911 }
1912
1913 if (vlan_hlen) {
1914 int ret;
1915 struct {
1916 __be16 h_vlan_proto;
1917 __be16 h_vlan_TCI;
1918 } veth;
1919
1920 veth.h_vlan_proto = skb->vlan_proto;
1921 veth.h_vlan_TCI = htons(skb_vlan_tag_get(skb));
1922
1923 vlan_offset = offsetof(struct vlan_ethhdr, h_vlan_proto);
1924
1925 ret = skb_copy_datagram_iter(skb, 0, iter, vlan_offset);
1926 if (ret || !iov_iter_count(iter))
1927 goto done;
1928
1929 ret = copy_to_iter(&veth, sizeof(veth), iter);
1930 if (ret != sizeof(veth) || !iov_iter_count(iter))
1931 goto done;
1932 }
1933
1934 skb_copy_datagram_iter(skb, vlan_offset, iter, skb->len - vlan_offset);
1935
1936 done:
1937 /* caller is in process context, */
1938 stats = get_cpu_ptr(tun->pcpu_stats);
1939 u64_stats_update_begin(&stats->syncp);
1940 stats->tx_packets++;
1941 stats->tx_bytes += skb->len + vlan_hlen;
1942 u64_stats_update_end(&stats->syncp);
1943 put_cpu_ptr(tun->pcpu_stats);
1944
1945 return total;
1946 }
1947
1948 static struct sk_buff *tun_ring_recv(struct tun_file *tfile, int noblock,
1949 int *err)
1950 {
1951 DECLARE_WAITQUEUE(wait, current);
1952 struct sk_buff *skb = NULL;
1953 int error = 0;
1954
1955 skb = skb_array_consume(&tfile->tx_array);
1956 if (skb)
1957 goto out;
1958 if (noblock) {
1959 error = -EAGAIN;
1960 goto out;
1961 }
1962
1963 add_wait_queue(&tfile->wq.wait, &wait);
1964
1965 while (1) {
1966 set_current_state(TASK_INTERRUPTIBLE);
1967 skb = skb_array_consume(&tfile->tx_array);
1968 if (skb)
1969 break;
1970 if (signal_pending(current)) {
1971 error = -ERESTARTSYS;
1972 break;
1973 }
1974 if (tfile->socket.sk->sk_shutdown & RCV_SHUTDOWN) {
1975 error = -EFAULT;
1976 break;
1977 }
1978
1979 schedule();
1980 }
1981
1982 __set_current_state(TASK_RUNNING);
1983 remove_wait_queue(&tfile->wq.wait, &wait);
1984
1985 out:
1986 *err = error;
1987 return skb;
1988 }
1989
1990 static ssize_t tun_do_read(struct tun_struct *tun, struct tun_file *tfile,
1991 struct iov_iter *to,
1992 int noblock, struct sk_buff *skb)
1993 {
1994 ssize_t ret;
1995 int err;
1996
1997 tun_debug(KERN_INFO, tun, "tun_do_read\n");
1998
1999 if (!iov_iter_count(to)) {
2000 if (skb)
2001 kfree_skb(skb);
2002 return 0;
2003 }
2004
2005 if (!skb) {
2006 /* Read frames from ring */
2007 skb = tun_ring_recv(tfile, noblock, &err);
2008 if (!skb)
2009 return err;
2010 }
2011
2012 ret = tun_put_user(tun, tfile, skb, to);
2013 if (unlikely(ret < 0))
2014 kfree_skb(skb);
2015 else
2016 consume_skb(skb);
2017
2018 return ret;
2019 }
2020
2021 static ssize_t tun_chr_read_iter(struct kiocb *iocb, struct iov_iter *to)
2022 {
2023 struct file *file = iocb->ki_filp;
2024 struct tun_file *tfile = file->private_data;
2025 struct tun_struct *tun = tun_get(tfile);
2026 ssize_t len = iov_iter_count(to), ret;
2027
2028 if (!tun)
2029 return -EBADFD;
2030 ret = tun_do_read(tun, tfile, to, file->f_flags & O_NONBLOCK, NULL);
2031 ret = min_t(ssize_t, ret, len);
2032 if (ret > 0)
2033 iocb->ki_pos = ret;
2034 tun_put(tun);
2035 return ret;
2036 }
2037
2038 static void tun_free_netdev(struct net_device *dev)
2039 {
2040 struct tun_struct *tun = netdev_priv(dev);
2041
2042 BUG_ON(!(list_empty(&tun->disabled)));
2043 free_percpu(tun->pcpu_stats);
2044 tun_flow_uninit(tun);
2045 security_tun_dev_free_security(tun->security);
2046 }
2047
2048 static void tun_setup(struct net_device *dev)
2049 {
2050 struct tun_struct *tun = netdev_priv(dev);
2051
2052 tun->owner = INVALID_UID;
2053 tun->group = INVALID_GID;
2054
2055 dev->ethtool_ops = &tun_ethtool_ops;
2056 dev->needs_free_netdev = true;
2057 dev->priv_destructor = tun_free_netdev;
2058 /* We prefer our own queue length */
2059 dev->tx_queue_len = TUN_READQ_SIZE;
2060 }
2061
2062 /* Trivial set of netlink ops to allow deleting tun or tap
2063 * device with netlink.
2064 */
2065 static int tun_validate(struct nlattr *tb[], struct nlattr *data[],
2066 struct netlink_ext_ack *extack)
2067 {
2068 NL_SET_ERR_MSG(extack,
2069 "tun/tap creation via rtnetlink is not supported.");
2070 return -EOPNOTSUPP;
2071 }
2072
2073 static struct rtnl_link_ops tun_link_ops __read_mostly = {
2074 .kind = DRV_NAME,
2075 .priv_size = sizeof(struct tun_struct),
2076 .setup = tun_setup,
2077 .validate = tun_validate,
2078 };
2079
2080 static void tun_sock_write_space(struct sock *sk)
2081 {
2082 struct tun_file *tfile;
2083 wait_queue_head_t *wqueue;
2084
2085 if (!sock_writeable(sk))
2086 return;
2087
2088 if (!test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags))
2089 return;
2090
2091 wqueue = sk_sleep(sk);
2092 if (wqueue && waitqueue_active(wqueue))
2093 wake_up_interruptible_sync_poll(wqueue, POLLOUT |
2094 POLLWRNORM | POLLWRBAND);
2095
2096 tfile = container_of(sk, struct tun_file, sk);
2097 kill_fasync(&tfile->fasync, SIGIO, POLL_OUT);
2098 }
2099
2100 static int tun_sendmsg(struct socket *sock, struct msghdr *m, size_t total_len)
2101 {
2102 int ret;
2103 struct tun_file *tfile = container_of(sock, struct tun_file, socket);
2104 struct tun_struct *tun = tun_get(tfile);
2105
2106 if (!tun)
2107 return -EBADFD;
2108
2109 ret = tun_get_user(tun, tfile, m->msg_control, &m->msg_iter,
2110 m->msg_flags & MSG_DONTWAIT,
2111 m->msg_flags & MSG_MORE);
2112 tun_put(tun);
2113 return ret;
2114 }
2115
2116 static int tun_recvmsg(struct socket *sock, struct msghdr *m, size_t total_len,
2117 int flags)
2118 {
2119 struct tun_file *tfile = container_of(sock, struct tun_file, socket);
2120 struct tun_struct *tun = tun_get(tfile);
2121 struct sk_buff *skb = m->msg_control;
2122 int ret;
2123
2124 if (!tun) {
2125 ret = -EBADFD;
2126 goto out_free_skb;
2127 }
2128
2129 if (flags & ~(MSG_DONTWAIT|MSG_TRUNC|MSG_ERRQUEUE)) {
2130 ret = -EINVAL;
2131 goto out_put_tun;
2132 }
2133 if (flags & MSG_ERRQUEUE) {
2134 ret = sock_recv_errqueue(sock->sk, m, total_len,
2135 SOL_PACKET, TUN_TX_TIMESTAMP);
2136 goto out;
2137 }
2138 ret = tun_do_read(tun, tfile, &m->msg_iter, flags & MSG_DONTWAIT, skb);
2139 if (ret > (ssize_t)total_len) {
2140 m->msg_flags |= MSG_TRUNC;
2141 ret = flags & MSG_TRUNC ? ret : total_len;
2142 }
2143 out:
2144 tun_put(tun);
2145 return ret;
2146
2147 out_put_tun:
2148 tun_put(tun);
2149 out_free_skb:
2150 if (skb)
2151 kfree_skb(skb);
2152 return ret;
2153 }
2154
2155 static int tun_peek_len(struct socket *sock)
2156 {
2157 struct tun_file *tfile = container_of(sock, struct tun_file, socket);
2158 struct tun_struct *tun;
2159 int ret = 0;
2160
2161 tun = tun_get(tfile);
2162 if (!tun)
2163 return 0;
2164
2165 ret = skb_array_peek_len(&tfile->tx_array);
2166 tun_put(tun);
2167
2168 return ret;
2169 }
2170
2171 /* Ops structure to mimic raw sockets with tun */
2172 static const struct proto_ops tun_socket_ops = {
2173 .peek_len = tun_peek_len,
2174 .sendmsg = tun_sendmsg,
2175 .recvmsg = tun_recvmsg,
2176 };
2177
2178 static struct proto tun_proto = {
2179 .name = "tun",
2180 .owner = THIS_MODULE,
2181 .obj_size = sizeof(struct tun_file),
2182 };
2183
2184 static int tun_flags(struct tun_struct *tun)
2185 {
2186 return tun->flags & (TUN_FEATURES | IFF_PERSIST | IFF_TUN | IFF_TAP);
2187 }
2188
2189 static ssize_t tun_show_flags(struct device *dev, struct device_attribute *attr,
2190 char *buf)
2191 {
2192 struct tun_struct *tun = netdev_priv(to_net_dev(dev));
2193 return sprintf(buf, "0x%x\n", tun_flags(tun));
2194 }
2195
2196 static ssize_t tun_show_owner(struct device *dev, struct device_attribute *attr,
2197 char *buf)
2198 {
2199 struct tun_struct *tun = netdev_priv(to_net_dev(dev));
2200 return uid_valid(tun->owner)?
2201 sprintf(buf, "%u\n",
2202 from_kuid_munged(current_user_ns(), tun->owner)):
2203 sprintf(buf, "-1\n");
2204 }
2205
2206 static ssize_t tun_show_group(struct device *dev, struct device_attribute *attr,
2207 char *buf)
2208 {
2209 struct tun_struct *tun = netdev_priv(to_net_dev(dev));
2210 return gid_valid(tun->group) ?
2211 sprintf(buf, "%u\n",
2212 from_kgid_munged(current_user_ns(), tun->group)):
2213 sprintf(buf, "-1\n");
2214 }
2215
2216 static DEVICE_ATTR(tun_flags, 0444, tun_show_flags, NULL);
2217 static DEVICE_ATTR(owner, 0444, tun_show_owner, NULL);
2218 static DEVICE_ATTR(group, 0444, tun_show_group, NULL);
2219
2220 static struct attribute *tun_dev_attrs[] = {
2221 &dev_attr_tun_flags.attr,
2222 &dev_attr_owner.attr,
2223 &dev_attr_group.attr,
2224 NULL
2225 };
2226
2227 static const struct attribute_group tun_attr_group = {
2228 .attrs = tun_dev_attrs
2229 };
2230
2231 static int tun_set_iff(struct net *net, struct file *file, struct ifreq *ifr)
2232 {
2233 struct tun_struct *tun;
2234 struct tun_file *tfile = file->private_data;
2235 struct net_device *dev;
2236 int err;
2237
2238 if (tfile->detached)
2239 return -EINVAL;
2240
2241 if ((ifr->ifr_flags & IFF_NAPI_FRAGS)) {
2242 if (!capable(CAP_NET_ADMIN))
2243 return -EPERM;
2244
2245 if (!(ifr->ifr_flags & IFF_NAPI) ||
2246 (ifr->ifr_flags & TUN_TYPE_MASK) != IFF_TAP)
2247 return -EINVAL;
2248 }
2249
2250 dev = __dev_get_by_name(net, ifr->ifr_name);
2251 if (dev) {
2252 if (ifr->ifr_flags & IFF_TUN_EXCL)
2253 return -EBUSY;
2254 if ((ifr->ifr_flags & IFF_TUN) && dev->netdev_ops == &tun_netdev_ops)
2255 tun = netdev_priv(dev);
2256 else if ((ifr->ifr_flags & IFF_TAP) && dev->netdev_ops == &tap_netdev_ops)
2257 tun = netdev_priv(dev);
2258 else
2259 return -EINVAL;
2260
2261 if (!!(ifr->ifr_flags & IFF_MULTI_QUEUE) !=
2262 !!(tun->flags & IFF_MULTI_QUEUE))
2263 return -EINVAL;
2264
2265 if (tun_not_capable(tun))
2266 return -EPERM;
2267 err = security_tun_dev_open(tun->security);
2268 if (err < 0)
2269 return err;
2270
2271 err = tun_attach(tun, file, ifr->ifr_flags & IFF_NOFILTER,
2272 ifr->ifr_flags & IFF_NAPI,
2273 ifr->ifr_flags & IFF_NAPI_FRAGS, true);
2274 if (err < 0)
2275 return err;
2276
2277 if (tun->flags & IFF_MULTI_QUEUE &&
2278 (tun->numqueues + tun->numdisabled > 1)) {
2279 /* One or more queue has already been attached, no need
2280 * to initialize the device again.
2281 */
2282 return 0;
2283 }
2284 }
2285 else {
2286 char *name;
2287 unsigned long flags = 0;
2288 int queues = ifr->ifr_flags & IFF_MULTI_QUEUE ?
2289 MAX_TAP_QUEUES : 1;
2290
2291 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
2292 return -EPERM;
2293 err = security_tun_dev_create();
2294 if (err < 0)
2295 return err;
2296
2297 /* Set dev type */
2298 if (ifr->ifr_flags & IFF_TUN) {
2299 /* TUN device */
2300 flags |= IFF_TUN;
2301 name = "tun%d";
2302 } else if (ifr->ifr_flags & IFF_TAP) {
2303 /* TAP device */
2304 flags |= IFF_TAP;
2305 name = "tap%d";
2306 } else
2307 return -EINVAL;
2308
2309 if (*ifr->ifr_name)
2310 name = ifr->ifr_name;
2311
2312 dev = alloc_netdev_mqs(sizeof(struct tun_struct), name,
2313 NET_NAME_UNKNOWN, tun_setup, queues,
2314 queues);
2315
2316 if (!dev)
2317 return -ENOMEM;
2318 err = dev_get_valid_name(net, dev, name);
2319 if (err < 0)
2320 goto err_free_dev;
2321
2322 dev_net_set(dev, net);
2323 dev->rtnl_link_ops = &tun_link_ops;
2324 dev->ifindex = tfile->ifindex;
2325 dev->sysfs_groups[0] = &tun_attr_group;
2326
2327 tun = netdev_priv(dev);
2328 tun->dev = dev;
2329 tun->flags = flags;
2330 tun->txflt.count = 0;
2331 tun->vnet_hdr_sz = sizeof(struct virtio_net_hdr);
2332
2333 tun->align = NET_SKB_PAD;
2334 tun->filter_attached = false;
2335 tun->sndbuf = tfile->socket.sk->sk_sndbuf;
2336 tun->rx_batched = 0;
2337
2338 tun->pcpu_stats = netdev_alloc_pcpu_stats(struct tun_pcpu_stats);
2339 if (!tun->pcpu_stats) {
2340 err = -ENOMEM;
2341 goto err_free_dev;
2342 }
2343
2344 spin_lock_init(&tun->lock);
2345
2346 err = security_tun_dev_alloc_security(&tun->security);
2347 if (err < 0)
2348 goto err_free_stat;
2349
2350 tun_net_init(dev);
2351 tun_flow_init(tun);
2352
2353 dev->hw_features = NETIF_F_SG | NETIF_F_FRAGLIST |
2354 TUN_USER_FEATURES | NETIF_F_HW_VLAN_CTAG_TX |
2355 NETIF_F_HW_VLAN_STAG_TX;
2356 dev->features = dev->hw_features | NETIF_F_LLTX;
2357 dev->vlan_features = dev->features &
2358 ~(NETIF_F_HW_VLAN_CTAG_TX |
2359 NETIF_F_HW_VLAN_STAG_TX);
2360
2361 INIT_LIST_HEAD(&tun->disabled);
2362 err = tun_attach(tun, file, false, ifr->ifr_flags & IFF_NAPI,
2363 ifr->ifr_flags & IFF_NAPI_FRAGS, false);
2364 if (err < 0)
2365 goto err_free_flow;
2366
2367 err = register_netdevice(tun->dev);
2368 if (err < 0)
2369 goto err_detach;
2370 /* free_netdev() won't check refcnt, to aovid race
2371 * with dev_put() we need publish tun after registration.
2372 */
2373 rcu_assign_pointer(tfile->tun, tun);
2374 }
2375
2376 netif_carrier_on(tun->dev);
2377
2378 tun_debug(KERN_INFO, tun, "tun_set_iff\n");
2379
2380 tun->flags = (tun->flags & ~TUN_FEATURES) |
2381 (ifr->ifr_flags & TUN_FEATURES);
2382
2383 /* Make sure persistent devices do not get stuck in
2384 * xoff state.
2385 */
2386 if (netif_running(tun->dev))
2387 netif_tx_wake_all_queues(tun->dev);
2388
2389 strcpy(ifr->ifr_name, tun->dev->name);
2390 return 0;
2391
2392 err_detach:
2393 tun_detach_all(dev);
2394 /* register_netdevice() already called tun_free_netdev() */
2395 goto err_free_dev;
2396
2397 err_free_flow:
2398 tun_flow_uninit(tun);
2399 security_tun_dev_free_security(tun->security);
2400 err_free_stat:
2401 free_percpu(tun->pcpu_stats);
2402 err_free_dev:
2403 free_netdev(dev);
2404 return err;
2405 }
2406
2407 static void tun_get_iff(struct net *net, struct tun_struct *tun,
2408 struct ifreq *ifr)
2409 {
2410 tun_debug(KERN_INFO, tun, "tun_get_iff\n");
2411
2412 strcpy(ifr->ifr_name, tun->dev->name);
2413
2414 ifr->ifr_flags = tun_flags(tun);
2415
2416 }
2417
2418 /* This is like a cut-down ethtool ops, except done via tun fd so no
2419 * privs required. */
2420 static int set_offload(struct tun_struct *tun, unsigned long arg)
2421 {
2422 netdev_features_t features = 0;
2423
2424 if (arg & TUN_F_CSUM) {
2425 features |= NETIF_F_HW_CSUM;
2426 arg &= ~TUN_F_CSUM;
2427
2428 if (arg & (TUN_F_TSO4|TUN_F_TSO6)) {
2429 if (arg & TUN_F_TSO_ECN) {
2430 features |= NETIF_F_TSO_ECN;
2431 arg &= ~TUN_F_TSO_ECN;
2432 }
2433 if (arg & TUN_F_TSO4)
2434 features |= NETIF_F_TSO;
2435 if (arg & TUN_F_TSO6)
2436 features |= NETIF_F_TSO6;
2437 arg &= ~(TUN_F_TSO4|TUN_F_TSO6);
2438 }
2439
2440 arg &= ~TUN_F_UFO;
2441 }
2442
2443 /* This gives the user a way to test for new features in future by
2444 * trying to set them. */
2445 if (arg)
2446 return -EINVAL;
2447
2448 tun->set_features = features;
2449 tun->dev->wanted_features &= ~TUN_USER_FEATURES;
2450 tun->dev->wanted_features |= features;
2451 netdev_update_features(tun->dev);
2452
2453 return 0;
2454 }
2455
2456 static void tun_detach_filter(struct tun_struct *tun, int n)
2457 {
2458 int i;
2459 struct tun_file *tfile;
2460
2461 for (i = 0; i < n; i++) {
2462 tfile = rtnl_dereference(tun->tfiles[i]);
2463 lock_sock(tfile->socket.sk);
2464 sk_detach_filter(tfile->socket.sk);
2465 release_sock(tfile->socket.sk);
2466 }
2467
2468 tun->filter_attached = false;
2469 }
2470
2471 static int tun_attach_filter(struct tun_struct *tun)
2472 {
2473 int i, ret = 0;
2474 struct tun_file *tfile;
2475
2476 for (i = 0; i < tun->numqueues; i++) {
2477 tfile = rtnl_dereference(tun->tfiles[i]);
2478 lock_sock(tfile->socket.sk);
2479 ret = sk_attach_filter(&tun->fprog, tfile->socket.sk);
2480 release_sock(tfile->socket.sk);
2481 if (ret) {
2482 tun_detach_filter(tun, i);
2483 return ret;
2484 }
2485 }
2486
2487 tun->filter_attached = true;
2488 return ret;
2489 }
2490
2491 static void tun_set_sndbuf(struct tun_struct *tun)
2492 {
2493 struct tun_file *tfile;
2494 int i;
2495
2496 for (i = 0; i < tun->numqueues; i++) {
2497 tfile = rtnl_dereference(tun->tfiles[i]);
2498 tfile->socket.sk->sk_sndbuf = tun->sndbuf;
2499 }
2500 }
2501
2502 static int tun_set_queue(struct file *file, struct ifreq *ifr)
2503 {
2504 struct tun_file *tfile = file->private_data;
2505 struct tun_struct *tun;
2506 int ret = 0;
2507
2508 rtnl_lock();
2509
2510 if (ifr->ifr_flags & IFF_ATTACH_QUEUE) {
2511 tun = tfile->detached;
2512 if (!tun) {
2513 ret = -EINVAL;
2514 goto unlock;
2515 }
2516 ret = security_tun_dev_attach_queue(tun->security);
2517 if (ret < 0)
2518 goto unlock;
2519 ret = tun_attach(tun, file, false, tun->flags & IFF_NAPI,
2520 tun->flags & IFF_NAPI_FRAGS, true);
2521 } else if (ifr->ifr_flags & IFF_DETACH_QUEUE) {
2522 tun = rtnl_dereference(tfile->tun);
2523 if (!tun || !(tun->flags & IFF_MULTI_QUEUE) || tfile->detached)
2524 ret = -EINVAL;
2525 else
2526 __tun_detach(tfile, false);
2527 } else
2528 ret = -EINVAL;
2529
2530 unlock:
2531 rtnl_unlock();
2532 return ret;
2533 }
2534
2535 static long __tun_chr_ioctl(struct file *file, unsigned int cmd,
2536 unsigned long arg, int ifreq_len)
2537 {
2538 struct tun_file *tfile = file->private_data;
2539 struct tun_struct *tun;
2540 void __user* argp = (void __user*)arg;
2541 unsigned int ifindex, carrier;
2542 struct ifreq ifr;
2543 kuid_t owner;
2544 kgid_t group;
2545 int sndbuf;
2546 int vnet_hdr_sz;
2547 int le;
2548 int ret;
2549
2550 if (cmd == TUNSETIFF || cmd == TUNSETQUEUE || _IOC_TYPE(cmd) == SOCK_IOC_TYPE) {
2551 if (copy_from_user(&ifr, argp, ifreq_len))
2552 return -EFAULT;
2553 } else {
2554 memset(&ifr, 0, sizeof(ifr));
2555 }
2556 if (cmd == TUNGETFEATURES) {
2557 /* Currently this just means: "what IFF flags are valid?".
2558 * This is needed because we never checked for invalid flags on
2559 * TUNSETIFF.
2560 */
2561 return put_user(IFF_TUN | IFF_TAP | TUN_FEATURES,
2562 (unsigned int __user*)argp);
2563 } else if (cmd == TUNSETQUEUE)
2564 return tun_set_queue(file, &ifr);
2565
2566 ret = 0;
2567 rtnl_lock();
2568
2569 tun = tun_get(tfile);
2570 if (cmd == TUNSETIFF) {
2571 ret = -EEXIST;
2572 if (tun)
2573 goto unlock;
2574
2575 ifr.ifr_name[IFNAMSIZ-1] = '\0';
2576
2577 ret = tun_set_iff(sock_net(&tfile->sk), file, &ifr);
2578
2579 if (ret)
2580 goto unlock;
2581
2582 if (copy_to_user(argp, &ifr, ifreq_len))
2583 ret = -EFAULT;
2584 goto unlock;
2585 }
2586 if (cmd == TUNSETIFINDEX) {
2587 ret = -EPERM;
2588 if (tun)
2589 goto unlock;
2590
2591 ret = -EFAULT;
2592 if (copy_from_user(&ifindex, argp, sizeof(ifindex)))
2593 goto unlock;
2594
2595 ret = 0;
2596 tfile->ifindex = ifindex;
2597 goto unlock;
2598 }
2599
2600 ret = -EBADFD;
2601 if (!tun)
2602 goto unlock;
2603
2604 tun_debug(KERN_INFO, tun, "tun_chr_ioctl cmd %u\n", cmd);
2605
2606 ret = 0;
2607 switch (cmd) {
2608 case TUNGETIFF:
2609 tun_get_iff(current->nsproxy->net_ns, tun, &ifr);
2610
2611 if (tfile->detached)
2612 ifr.ifr_flags |= IFF_DETACH_QUEUE;
2613 if (!tfile->socket.sk->sk_filter)
2614 ifr.ifr_flags |= IFF_NOFILTER;
2615
2616 if (copy_to_user(argp, &ifr, ifreq_len))
2617 ret = -EFAULT;
2618 break;
2619
2620 case TUNSETNOCSUM:
2621 /* Disable/Enable checksum */
2622
2623 /* [unimplemented] */
2624 tun_debug(KERN_INFO, tun, "ignored: set checksum %s\n",
2625 arg ? "disabled" : "enabled");
2626 break;
2627
2628 case TUNSETPERSIST:
2629 /* Disable/Enable persist mode. Keep an extra reference to the
2630 * module to prevent the module being unprobed.
2631 */
2632 if (arg && !(tun->flags & IFF_PERSIST)) {
2633 tun->flags |= IFF_PERSIST;
2634 __module_get(THIS_MODULE);
2635 }
2636 if (!arg && (tun->flags & IFF_PERSIST)) {
2637 tun->flags &= ~IFF_PERSIST;
2638 module_put(THIS_MODULE);
2639 }
2640
2641 tun_debug(KERN_INFO, tun, "persist %s\n",
2642 arg ? "enabled" : "disabled");
2643 break;
2644
2645 case TUNSETOWNER:
2646 /* Set owner of the device */
2647 owner = make_kuid(current_user_ns(), arg);
2648 if (!uid_valid(owner)) {
2649 ret = -EINVAL;
2650 break;
2651 }
2652 tun->owner = owner;
2653 tun_debug(KERN_INFO, tun, "owner set to %u\n",
2654 from_kuid(&init_user_ns, tun->owner));
2655 break;
2656
2657 case TUNSETGROUP:
2658 /* Set group of the device */
2659 group = make_kgid(current_user_ns(), arg);
2660 if (!gid_valid(group)) {
2661 ret = -EINVAL;
2662 break;
2663 }
2664 tun->group = group;
2665 tun_debug(KERN_INFO, tun, "group set to %u\n",
2666 from_kgid(&init_user_ns, tun->group));
2667 break;
2668
2669 case TUNSETLINK:
2670 /* Only allow setting the type when the interface is down */
2671 if (tun->dev->flags & IFF_UP) {
2672 tun_debug(KERN_INFO, tun,
2673 "Linktype set failed because interface is up\n");
2674 ret = -EBUSY;
2675 } else {
2676 tun->dev->type = (int) arg;
2677 tun_debug(KERN_INFO, tun, "linktype set to %d\n",
2678 tun->dev->type);
2679 ret = 0;
2680 }
2681 break;
2682
2683 #ifdef TUN_DEBUG
2684 case TUNSETDEBUG:
2685 tun->debug = arg;
2686 break;
2687 #endif
2688 case TUNSETOFFLOAD:
2689 ret = set_offload(tun, arg);
2690 break;
2691
2692 case TUNSETTXFILTER:
2693 /* Can be set only for TAPs */
2694 ret = -EINVAL;
2695 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
2696 break;
2697 ret = update_filter(&tun->txflt, (void __user *)arg);
2698 break;
2699
2700 case SIOCGIFHWADDR:
2701 /* Get hw address */
2702 memcpy(ifr.ifr_hwaddr.sa_data, tun->dev->dev_addr, ETH_ALEN);
2703 ifr.ifr_hwaddr.sa_family = tun->dev->type;
2704 if (copy_to_user(argp, &ifr, ifreq_len))
2705 ret = -EFAULT;
2706 break;
2707
2708 case SIOCSIFHWADDR:
2709 /* Set hw address */
2710 tun_debug(KERN_DEBUG, tun, "set hw address: %pM\n",
2711 ifr.ifr_hwaddr.sa_data);
2712
2713 ret = dev_set_mac_address(tun->dev, &ifr.ifr_hwaddr);
2714 break;
2715
2716 case TUNGETSNDBUF:
2717 sndbuf = tfile->socket.sk->sk_sndbuf;
2718 if (copy_to_user(argp, &sndbuf, sizeof(sndbuf)))
2719 ret = -EFAULT;
2720 break;
2721
2722 case TUNSETSNDBUF:
2723 if (copy_from_user(&sndbuf, argp, sizeof(sndbuf))) {
2724 ret = -EFAULT;
2725 break;
2726 }
2727 if (sndbuf <= 0) {
2728 ret = -EINVAL;
2729 break;
2730 }
2731
2732 tun->sndbuf = sndbuf;
2733 tun_set_sndbuf(tun);
2734 break;
2735
2736 case TUNGETVNETHDRSZ:
2737 vnet_hdr_sz = tun->vnet_hdr_sz;
2738 if (copy_to_user(argp, &vnet_hdr_sz, sizeof(vnet_hdr_sz)))
2739 ret = -EFAULT;
2740 break;
2741
2742 case TUNSETVNETHDRSZ:
2743 if (copy_from_user(&vnet_hdr_sz, argp, sizeof(vnet_hdr_sz))) {
2744 ret = -EFAULT;
2745 break;
2746 }
2747 if (vnet_hdr_sz < (int)sizeof(struct virtio_net_hdr)) {
2748 ret = -EINVAL;
2749 break;
2750 }
2751
2752 tun->vnet_hdr_sz = vnet_hdr_sz;
2753 break;
2754
2755 case TUNGETVNETLE:
2756 le = !!(tun->flags & TUN_VNET_LE);
2757 if (put_user(le, (int __user *)argp))
2758 ret = -EFAULT;
2759 break;
2760
2761 case TUNSETVNETLE:
2762 if (get_user(le, (int __user *)argp)) {
2763 ret = -EFAULT;
2764 break;
2765 }
2766 if (le)
2767 tun->flags |= TUN_VNET_LE;
2768 else
2769 tun->flags &= ~TUN_VNET_LE;
2770 break;
2771
2772 case TUNGETVNETBE:
2773 ret = tun_get_vnet_be(tun, argp);
2774 break;
2775
2776 case TUNSETVNETBE:
2777 ret = tun_set_vnet_be(tun, argp);
2778 break;
2779
2780 case TUNATTACHFILTER:
2781 /* Can be set only for TAPs */
2782 ret = -EINVAL;
2783 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
2784 break;
2785 ret = -EFAULT;
2786 if (copy_from_user(&tun->fprog, argp, sizeof(tun->fprog)))
2787 break;
2788
2789 ret = tun_attach_filter(tun);
2790 break;
2791
2792 case TUNDETACHFILTER:
2793 /* Can be set only for TAPs */
2794 ret = -EINVAL;
2795 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
2796 break;
2797 ret = 0;
2798 tun_detach_filter(tun, tun->numqueues);
2799 break;
2800
2801 case TUNGETFILTER:
2802 ret = -EINVAL;
2803 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
2804 break;
2805 ret = -EFAULT;
2806 if (copy_to_user(argp, &tun->fprog, sizeof(tun->fprog)))
2807 break;
2808 ret = 0;
2809 break;
2810
2811 case TUNSETCARRIER:
2812 ret = -EFAULT;
2813 if (copy_from_user(&carrier, argp, sizeof(carrier)))
2814 goto unlock;
2815
2816 ret = tun_net_change_carrier(tun->dev, (bool)carrier);
2817 break;
2818
2819 default:
2820 ret = -EINVAL;
2821 break;
2822 }
2823
2824 unlock:
2825 rtnl_unlock();
2826 if (tun)
2827 tun_put(tun);
2828 return ret;
2829 }
2830
2831 static long tun_chr_ioctl(struct file *file,
2832 unsigned int cmd, unsigned long arg)
2833 {
2834 return __tun_chr_ioctl(file, cmd, arg, sizeof (struct ifreq));
2835 }
2836
2837 #ifdef CONFIG_COMPAT
2838 static long tun_chr_compat_ioctl(struct file *file,
2839 unsigned int cmd, unsigned long arg)
2840 {
2841 switch (cmd) {
2842 case TUNSETIFF:
2843 case TUNGETIFF:
2844 case TUNSETTXFILTER:
2845 case TUNGETSNDBUF:
2846 case TUNSETSNDBUF:
2847 case SIOCGIFHWADDR:
2848 case SIOCSIFHWADDR:
2849 arg = (unsigned long)compat_ptr(arg);
2850 break;
2851 default:
2852 arg = (compat_ulong_t)arg;
2853 break;
2854 }
2855
2856 /*
2857 * compat_ifreq is shorter than ifreq, so we must not access beyond
2858 * the end of that structure. All fields that are used in this
2859 * driver are compatible though, we don't need to convert the
2860 * contents.
2861 */
2862 return __tun_chr_ioctl(file, cmd, arg, sizeof(struct compat_ifreq));
2863 }
2864 #endif /* CONFIG_COMPAT */
2865
2866 static int tun_chr_fasync(int fd, struct file *file, int on)
2867 {
2868 struct tun_file *tfile = file->private_data;
2869 int ret;
2870
2871 if ((ret = fasync_helper(fd, file, on, &tfile->fasync)) < 0)
2872 goto out;
2873
2874 if (on) {
2875 __f_setown(file, task_pid(current), PIDTYPE_PID, 0);
2876 tfile->flags |= TUN_FASYNC;
2877 } else
2878 tfile->flags &= ~TUN_FASYNC;
2879 ret = 0;
2880 out:
2881 return ret;
2882 }
2883
2884 static int tun_chr_open(struct inode *inode, struct file * file)
2885 {
2886 struct net *net = current->nsproxy->net_ns;
2887 struct tun_file *tfile;
2888
2889 DBG1(KERN_INFO, "tunX: tun_chr_open\n");
2890
2891 tfile = (struct tun_file *)sk_alloc(net, AF_UNSPEC, GFP_KERNEL,
2892 &tun_proto, 0);
2893 if (!tfile)
2894 return -ENOMEM;
2895 if (skb_array_init(&tfile->tx_array, 0, GFP_KERNEL)) {
2896 sk_free(&tfile->sk);
2897 return -ENOMEM;
2898 }
2899
2900 mutex_init(&tfile->napi_mutex);
2901 RCU_INIT_POINTER(tfile->tun, NULL);
2902 tfile->flags = 0;
2903 tfile->ifindex = 0;
2904
2905 init_waitqueue_head(&tfile->wq.wait);
2906 RCU_INIT_POINTER(tfile->socket.wq, &tfile->wq);
2907
2908 tfile->socket.file = file;
2909 tfile->socket.ops = &tun_socket_ops;
2910
2911 sock_init_data(&tfile->socket, &tfile->sk);
2912
2913 tfile->sk.sk_write_space = tun_sock_write_space;
2914 tfile->sk.sk_sndbuf = INT_MAX;
2915
2916 file->private_data = tfile;
2917 INIT_LIST_HEAD(&tfile->next);
2918
2919 sock_set_flag(&tfile->sk, SOCK_ZEROCOPY);
2920
2921 return 0;
2922 }
2923
2924 static int tun_chr_close(struct inode *inode, struct file *file)
2925 {
2926 struct tun_file *tfile = file->private_data;
2927
2928 tun_detach(tfile, true);
2929
2930 return 0;
2931 }
2932
2933 #ifdef CONFIG_PROC_FS
2934 static void tun_chr_show_fdinfo(struct seq_file *m, struct file *file)
2935 {
2936 struct tun_file *tfile = file->private_data;
2937 struct tun_struct *tun;
2938 struct ifreq ifr;
2939
2940 memset(&ifr, 0, sizeof(ifr));
2941
2942 rtnl_lock();
2943 tun = tun_get(tfile);
2944 if (tun)
2945 tun_get_iff(current->nsproxy->net_ns, tun, &ifr);
2946 rtnl_unlock();
2947
2948 if (tun)
2949 tun_put(tun);
2950
2951 seq_printf(m, "iff:\t%s\n", ifr.ifr_name);
2952 }
2953 #endif
2954
2955 static const struct file_operations tun_fops = {
2956 .owner = THIS_MODULE,
2957 .llseek = no_llseek,
2958 .read_iter = tun_chr_read_iter,
2959 .write_iter = tun_chr_write_iter,
2960 .poll = tun_chr_poll,
2961 .unlocked_ioctl = tun_chr_ioctl,
2962 #ifdef CONFIG_COMPAT
2963 .compat_ioctl = tun_chr_compat_ioctl,
2964 #endif
2965 .open = tun_chr_open,
2966 .release = tun_chr_close,
2967 .fasync = tun_chr_fasync,
2968 #ifdef CONFIG_PROC_FS
2969 .show_fdinfo = tun_chr_show_fdinfo,
2970 #endif
2971 };
2972
2973 static struct miscdevice tun_miscdev = {
2974 .minor = TUN_MINOR,
2975 .name = "tun",
2976 .nodename = "net/tun",
2977 .fops = &tun_fops,
2978 };
2979
2980 /* ethtool interface */
2981
2982 static int tun_get_link_ksettings(struct net_device *dev,
2983 struct ethtool_link_ksettings *cmd)
2984 {
2985 ethtool_link_ksettings_zero_link_mode(cmd, supported);
2986 ethtool_link_ksettings_zero_link_mode(cmd, advertising);
2987 cmd->base.speed = SPEED_10;
2988 cmd->base.duplex = DUPLEX_FULL;
2989 cmd->base.port = PORT_TP;
2990 cmd->base.phy_address = 0;
2991 cmd->base.autoneg = AUTONEG_DISABLE;
2992 return 0;
2993 }
2994
2995 static void tun_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
2996 {
2997 struct tun_struct *tun = netdev_priv(dev);
2998
2999 strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
3000 strlcpy(info->version, DRV_VERSION, sizeof(info->version));
3001
3002 switch (tun->flags & TUN_TYPE_MASK) {
3003 case IFF_TUN:
3004 strlcpy(info->bus_info, "tun", sizeof(info->bus_info));
3005 break;
3006 case IFF_TAP:
3007 strlcpy(info->bus_info, "tap", sizeof(info->bus_info));
3008 break;
3009 }
3010 }
3011
3012 static u32 tun_get_msglevel(struct net_device *dev)
3013 {
3014 #ifdef TUN_DEBUG
3015 struct tun_struct *tun = netdev_priv(dev);
3016 return tun->debug;
3017 #else
3018 return -EOPNOTSUPP;
3019 #endif
3020 }
3021
3022 static void tun_set_msglevel(struct net_device *dev, u32 value)
3023 {
3024 #ifdef TUN_DEBUG
3025 struct tun_struct *tun = netdev_priv(dev);
3026 tun->debug = value;
3027 #endif
3028 }
3029
3030 static int tun_get_coalesce(struct net_device *dev,
3031 struct ethtool_coalesce *ec)
3032 {
3033 struct tun_struct *tun = netdev_priv(dev);
3034
3035 ec->rx_max_coalesced_frames = tun->rx_batched;
3036
3037 return 0;
3038 }
3039
3040 static int tun_set_coalesce(struct net_device *dev,
3041 struct ethtool_coalesce *ec)
3042 {
3043 struct tun_struct *tun = netdev_priv(dev);
3044
3045 if (ec->rx_max_coalesced_frames > NAPI_POLL_WEIGHT)
3046 tun->rx_batched = NAPI_POLL_WEIGHT;
3047 else
3048 tun->rx_batched = ec->rx_max_coalesced_frames;
3049
3050 return 0;
3051 }
3052
3053 static const struct ethtool_ops tun_ethtool_ops = {
3054 .get_drvinfo = tun_get_drvinfo,
3055 .get_msglevel = tun_get_msglevel,
3056 .set_msglevel = tun_set_msglevel,
3057 .get_link = ethtool_op_get_link,
3058 .get_ts_info = ethtool_op_get_ts_info,
3059 .get_coalesce = tun_get_coalesce,
3060 .set_coalesce = tun_set_coalesce,
3061 .get_link_ksettings = tun_get_link_ksettings,
3062 };
3063
3064 static int tun_queue_resize(struct tun_struct *tun)
3065 {
3066 struct net_device *dev = tun->dev;
3067 struct tun_file *tfile;
3068 struct skb_array **arrays;
3069 int n = tun->numqueues + tun->numdisabled;
3070 int ret, i;
3071
3072 arrays = kmalloc_array(n, sizeof(*arrays), GFP_KERNEL);
3073 if (!arrays)
3074 return -ENOMEM;
3075
3076 for (i = 0; i < tun->numqueues; i++) {
3077 tfile = rtnl_dereference(tun->tfiles[i]);
3078 arrays[i] = &tfile->tx_array;
3079 }
3080 list_for_each_entry(tfile, &tun->disabled, next)
3081 arrays[i++] = &tfile->tx_array;
3082
3083 ret = skb_array_resize_multiple(arrays, n,
3084 dev->tx_queue_len, GFP_KERNEL);
3085
3086 kfree(arrays);
3087 return ret;
3088 }
3089
3090 static int tun_device_event(struct notifier_block *unused,
3091 unsigned long event, void *ptr)
3092 {
3093 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
3094 struct tun_struct *tun = netdev_priv(dev);
3095 int i;
3096
3097 if (dev->rtnl_link_ops != &tun_link_ops)
3098 return NOTIFY_DONE;
3099
3100 switch (event) {
3101 case NETDEV_CHANGE_TX_QUEUE_LEN:
3102 if (tun_queue_resize(tun))
3103 return NOTIFY_BAD;
3104 break;
3105 case NETDEV_UP:
3106 for (i = 0; i < tun->numqueues; i++) {
3107 struct tun_file *tfile;
3108
3109 tfile = rtnl_dereference(tun->tfiles[i]);
3110 tfile->socket.sk->sk_write_space(tfile->socket.sk);
3111 }
3112 break;
3113 default:
3114 break;
3115 }
3116
3117 return NOTIFY_DONE;
3118 }
3119
3120 static struct notifier_block tun_notifier_block __read_mostly = {
3121 .notifier_call = tun_device_event,
3122 };
3123
3124 static int __init tun_init(void)
3125 {
3126 int ret = 0;
3127
3128 pr_info("%s, %s\n", DRV_DESCRIPTION, DRV_VERSION);
3129
3130 ret = rtnl_link_register(&tun_link_ops);
3131 if (ret) {
3132 pr_err("Can't register link_ops\n");
3133 goto err_linkops;
3134 }
3135
3136 ret = misc_register(&tun_miscdev);
3137 if (ret) {
3138 pr_err("Can't register misc device %d\n", TUN_MINOR);
3139 goto err_misc;
3140 }
3141
3142 ret = register_netdevice_notifier(&tun_notifier_block);
3143 if (ret) {
3144 pr_err("Can't register netdevice notifier\n");
3145 goto err_notifier;
3146 }
3147
3148 return 0;
3149
3150 err_notifier:
3151 misc_deregister(&tun_miscdev);
3152 err_misc:
3153 rtnl_link_unregister(&tun_link_ops);
3154 err_linkops:
3155 return ret;
3156 }
3157
3158 static void tun_cleanup(void)
3159 {
3160 misc_deregister(&tun_miscdev);
3161 rtnl_link_unregister(&tun_link_ops);
3162 unregister_netdevice_notifier(&tun_notifier_block);
3163 }
3164
3165 /* Get an underlying socket object from tun file. Returns error unless file is
3166 * attached to a device. The returned object works like a packet socket, it
3167 * can be used for sock_sendmsg/sock_recvmsg. The caller is responsible for
3168 * holding a reference to the file for as long as the socket is in use. */
3169 struct socket *tun_get_socket(struct file *file)
3170 {
3171 struct tun_file *tfile;
3172 if (file->f_op != &tun_fops)
3173 return ERR_PTR(-EINVAL);
3174 tfile = file->private_data;
3175 if (!tfile)
3176 return ERR_PTR(-EBADFD);
3177 return &tfile->socket;
3178 }
3179 EXPORT_SYMBOL_GPL(tun_get_socket);
3180
3181 struct skb_array *tun_get_skb_array(struct file *file)
3182 {
3183 struct tun_file *tfile;
3184
3185 if (file->f_op != &tun_fops)
3186 return ERR_PTR(-EINVAL);
3187 tfile = file->private_data;
3188 if (!tfile)
3189 return ERR_PTR(-EBADFD);
3190 return &tfile->tx_array;
3191 }
3192 EXPORT_SYMBOL_GPL(tun_get_skb_array);
3193
3194 module_init(tun_init);
3195 module_exit(tun_cleanup);
3196 MODULE_DESCRIPTION(DRV_DESCRIPTION);
3197 MODULE_AUTHOR(DRV_COPYRIGHT);
3198 MODULE_LICENSE("GPL");
3199 MODULE_ALIAS_MISCDEV(TUN_MINOR);
3200 MODULE_ALIAS("devname:net/tun");
ubuntu-bionic-kernel mirror