?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Merge tag 'block-5.11-2021-01-10' of git://git.kernel.dk/linux-block
[linux/fpc-iii.git] / drivers / net / tun.c
blob978ac0981d16017d648ebd04f7746fb498db3620
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3 * TUN - Universal TUN/TAP device driver.
4 * Copyright (C) 1999-2002 Maxim Krasnyansky <maxk@qualcomm.com>
5 *
6 * $Id: tun.c,v 1.15 2002/03/01 02:44:24 maxk Exp $
7 */
8
9 /*
10 * Changes:
11 *
12 * Mike Kershaw <dragorn@kismetwireless.net> 2005/08/14
13 * Add TUNSETLINK ioctl to set the link encapsulation
14 *
15 * Mark Smith <markzzzsmith@yahoo.com.au>
16 * Use eth_random_addr() for tap MAC address.
17 *
18 * Harald Roelle <harald.roelle@ifi.lmu.de> 2004/04/20
19 * Fixes in packet dropping, queue length setting and queue wakeup.
20 * Increased default tx queue length.
21 * Added ethtool API.
22 * Minor cleanups
23 *
24 * Daniel Podlejski <underley@underley.eu.org>
25 * Modifications for 2.3.99-pre5 kernel.
26 */
27
28 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
29
30 #define DRV_NAME "tun"
31 #define DRV_VERSION "1.6"
32 #define DRV_DESCRIPTION "Universal TUN/TAP device driver"
33 #define DRV_COPYRIGHT "(C) 1999-2004 Max Krasnyansky <maxk@qualcomm.com>"
34
35 #include <linux/module.h>
36 #include <linux/errno.h>
37 #include <linux/kernel.h>
38 #include <linux/sched/signal.h>
39 #include <linux/major.h>
40 #include <linux/slab.h>
41 #include <linux/poll.h>
42 #include <linux/fcntl.h>
43 #include <linux/init.h>
44 #include <linux/skbuff.h>
45 #include <linux/netdevice.h>
46 #include <linux/etherdevice.h>
47 #include <linux/miscdevice.h>
48 #include <linux/ethtool.h>
49 #include <linux/rtnetlink.h>
50 #include <linux/compat.h>
51 #include <linux/if.h>
52 #include <linux/if_arp.h>
53 #include <linux/if_ether.h>
54 #include <linux/if_tun.h>
55 #include <linux/if_vlan.h>
56 #include <linux/crc32.h>
57 #include <linux/nsproxy.h>
58 #include <linux/virtio_net.h>
59 #include <linux/rcupdate.h>
60 #include <net/net_namespace.h>
61 #include <net/netns/generic.h>
62 #include <net/rtnetlink.h>
63 #include <net/sock.h>
64 #include <net/xdp.h>
65 #include <net/ip_tunnels.h>
66 #include <linux/seq_file.h>
67 #include <linux/uio.h>
68 #include <linux/skb_array.h>
69 #include <linux/bpf.h>
70 #include <linux/bpf_trace.h>
71 #include <linux/mutex.h>
72
73 #include <linux/uaccess.h>
74 #include <linux/proc_fs.h>
75
76 static void tun_default_link_ksettings(struct net_device *dev,
77 struct ethtool_link_ksettings *cmd);
78
79 #define TUN_RX_PAD (NET_IP_ALIGN + NET_SKB_PAD)
80
81 /* TUN device flags */
82
83 /* IFF_ATTACH_QUEUE is never stored in device flags,
84 * overload it to mean fasync when stored there.
85 */
86 #define TUN_FASYNC IFF_ATTACH_QUEUE
87 /* High bits in flags field are unused. */
88 #define TUN_VNET_LE 0x80000000
89 #define TUN_VNET_BE 0x40000000
90
91 #define TUN_FEATURES (IFF_NO_PI | IFF_ONE_QUEUE | IFF_VNET_HDR | \
92 IFF_MULTI_QUEUE | IFF_NAPI | IFF_NAPI_FRAGS)
93
94 #define GOODCOPY_LEN 128
95
96 #define FLT_EXACT_COUNT 8
97 struct tap_filter {
98 unsigned int count; /* Number of addrs. Zero means disabled */
99 u32 mask[2]; /* Mask of the hashed addrs */
100 unsigned char addr[FLT_EXACT_COUNT][ETH_ALEN];
101 };
102
103 /* MAX_TAP_QUEUES 256 is chosen to allow rx/tx queues to be equal
104 * to max number of VCPUs in guest. */
105 #define MAX_TAP_QUEUES 256
106 #define MAX_TAP_FLOWS 4096
107
108 #define TUN_FLOW_EXPIRE (3 * HZ)
109
110 /* A tun_file connects an open character device to a tuntap netdevice. It
111 * also contains all socket related structures (except sock_fprog and tap_filter)
112 * to serve as one transmit queue for tuntap device. The sock_fprog and
113 * tap_filter were kept in tun_struct since they were used for filtering for the
114 * netdevice not for a specific queue (at least I didn't see the requirement for
115 * this).
116 *
117 * RCU usage:
118 * The tun_file and tun_struct are loosely coupled, the pointer from one to the
119 * other can only be read while rcu_read_lock or rtnl_lock is held.
120 */
121 struct tun_file {
122 struct sock sk;
123 struct socket socket;
124 struct tun_struct __rcu *tun;
125 struct fasync_struct *fasync;
126 /* only used for fasnyc */
127 unsigned int flags;
128 union {
129 u16 queue_index;
130 unsigned int ifindex;
131 };
132 struct napi_struct napi;
133 bool napi_enabled;
134 bool napi_frags_enabled;
135 struct mutex napi_mutex; /* Protects access to the above napi */
136 struct list_head next;
137 struct tun_struct *detached;
138 struct ptr_ring tx_ring;
139 struct xdp_rxq_info xdp_rxq;
140 };
141
142 struct tun_page {
143 struct page *page;
144 int count;
145 };
146
147 struct tun_flow_entry {
148 struct hlist_node hash_link;
149 struct rcu_head rcu;
150 struct tun_struct *tun;
151
152 u32 rxhash;
153 u32 rps_rxhash;
154 int queue_index;
155 unsigned long updated ____cacheline_aligned_in_smp;
156 };
157
158 #define TUN_NUM_FLOW_ENTRIES 1024
159 #define TUN_MASK_FLOW_ENTRIES (TUN_NUM_FLOW_ENTRIES - 1)
160
161 struct tun_prog {
162 struct rcu_head rcu;
163 struct bpf_prog *prog;
164 };
165
166 /* Since the socket were moved to tun_file, to preserve the behavior of persist
167 * device, socket filter, sndbuf and vnet header size were restore when the
168 * file were attached to a persist device.
169 */
170 struct tun_struct {
171 struct tun_file __rcu *tfiles[MAX_TAP_QUEUES];
172 unsigned int numqueues;
173 unsigned int flags;
174 kuid_t owner;
175 kgid_t group;
176
177 struct net_device *dev;
178 netdev_features_t set_features;
179 #define TUN_USER_FEATURES (NETIF_F_HW_CSUM|NETIF_F_TSO_ECN|NETIF_F_TSO| \
180 NETIF_F_TSO6)
181
182 int align;
183 int vnet_hdr_sz;
184 int sndbuf;
185 struct tap_filter txflt;
186 struct sock_fprog fprog;
187 /* protected by rtnl lock */
188 bool filter_attached;
189 u32 msg_enable;
190 spinlock_t lock;
191 struct hlist_head flows[TUN_NUM_FLOW_ENTRIES];
192 struct timer_list flow_gc_timer;
193 unsigned long ageing_time;
194 unsigned int numdisabled;
195 struct list_head disabled;
196 void *security;
197 u32 flow_count;
198 u32 rx_batched;
199 atomic_long_t rx_frame_errors;
200 struct bpf_prog __rcu *xdp_prog;
201 struct tun_prog __rcu *steering_prog;
202 struct tun_prog __rcu *filter_prog;
203 struct ethtool_link_ksettings link_ksettings;
204 };
205
206 struct veth {
207 __be16 h_vlan_proto;
208 __be16 h_vlan_TCI;
209 };
210
211 static int tun_napi_receive(struct napi_struct *napi, int budget)
212 {
213 struct tun_file *tfile = container_of(napi, struct tun_file, napi);
214 struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
215 struct sk_buff_head process_queue;
216 struct sk_buff *skb;
217 int received = 0;
218
219 __skb_queue_head_init(&process_queue);
220
221 spin_lock(&queue->lock);
222 skb_queue_splice_tail_init(queue, &process_queue);
223 spin_unlock(&queue->lock);
224
225 while (received < budget && (skb = __skb_dequeue(&process_queue))) {
226 napi_gro_receive(napi, skb);
227 ++received;
228 }
229
230 if (!skb_queue_empty(&process_queue)) {
231 spin_lock(&queue->lock);
232 skb_queue_splice(&process_queue, queue);
233 spin_unlock(&queue->lock);
234 }
235
236 return received;
237 }
238
239 static int tun_napi_poll(struct napi_struct *napi, int budget)
240 {
241 unsigned int received;
242
243 received = tun_napi_receive(napi, budget);
244
245 if (received < budget)
246 napi_complete_done(napi, received);
247
248 return received;
249 }
250
251 static void tun_napi_init(struct tun_struct *tun, struct tun_file *tfile,
252 bool napi_en, bool napi_frags)
253 {
254 tfile->napi_enabled = napi_en;
255 tfile->napi_frags_enabled = napi_en && napi_frags;
256 if (napi_en) {
257 netif_tx_napi_add(tun->dev, &tfile->napi, tun_napi_poll,
258 NAPI_POLL_WEIGHT);
259 napi_enable(&tfile->napi);
260 }
261 }
262
263 static void tun_napi_disable(struct tun_file *tfile)
264 {
265 if (tfile->napi_enabled)
266 napi_disable(&tfile->napi);
267 }
268
269 static void tun_napi_del(struct tun_file *tfile)
270 {
271 if (tfile->napi_enabled)
272 netif_napi_del(&tfile->napi);
273 }
274
275 static bool tun_napi_frags_enabled(const struct tun_file *tfile)
276 {
277 return tfile->napi_frags_enabled;
278 }
279
280 #ifdef CONFIG_TUN_VNET_CROSS_LE
281 static inline bool tun_legacy_is_little_endian(struct tun_struct *tun)
282 {
283 return tun->flags & TUN_VNET_BE ? false :
284 virtio_legacy_is_little_endian();
285 }
286
287 static long tun_get_vnet_be(struct tun_struct *tun, int __user *argp)
288 {
289 int be = !!(tun->flags & TUN_VNET_BE);
290
291 if (put_user(be, argp))
292 return -EFAULT;
293
294 return 0;
295 }
296
297 static long tun_set_vnet_be(struct tun_struct *tun, int __user *argp)
298 {
299 int be;
300
301 if (get_user(be, argp))
302 return -EFAULT;
303
304 if (be)
305 tun->flags |= TUN_VNET_BE;
306 else
307 tun->flags &= ~TUN_VNET_BE;
308
309 return 0;
310 }
311 #else
312 static inline bool tun_legacy_is_little_endian(struct tun_struct *tun)
313 {
314 return virtio_legacy_is_little_endian();
315 }
316
317 static long tun_get_vnet_be(struct tun_struct *tun, int __user *argp)
318 {
319 return -EINVAL;
320 }
321
322 static long tun_set_vnet_be(struct tun_struct *tun, int __user *argp)
323 {
324 return -EINVAL;
325 }
326 #endif /* CONFIG_TUN_VNET_CROSS_LE */
327
328 static inline bool tun_is_little_endian(struct tun_struct *tun)
329 {
330 return tun->flags & TUN_VNET_LE ||
331 tun_legacy_is_little_endian(tun);
332 }
333
334 static inline u16 tun16_to_cpu(struct tun_struct *tun, __virtio16 val)
335 {
336 return __virtio16_to_cpu(tun_is_little_endian(tun), val);
337 }
338
339 static inline __virtio16 cpu_to_tun16(struct tun_struct *tun, u16 val)
340 {
341 return __cpu_to_virtio16(tun_is_little_endian(tun), val);
342 }
343
344 static inline u32 tun_hashfn(u32 rxhash)
345 {
346 return rxhash & TUN_MASK_FLOW_ENTRIES;
347 }
348
349 static struct tun_flow_entry *tun_flow_find(struct hlist_head *head, u32 rxhash)
350 {
351 struct tun_flow_entry *e;
352
353 hlist_for_each_entry_rcu(e, head, hash_link) {
354 if (e->rxhash == rxhash)
355 return e;
356 }
357 return NULL;
358 }
359
360 static struct tun_flow_entry *tun_flow_create(struct tun_struct *tun,
361 struct hlist_head *head,
362 u32 rxhash, u16 queue_index)
363 {
364 struct tun_flow_entry *e = kmalloc(sizeof(*e), GFP_ATOMIC);
365
366 if (e) {
367 netif_info(tun, tx_queued, tun->dev,
368 "create flow: hash %u index %u\n",
369 rxhash, queue_index);
370 e->updated = jiffies;
371 e->rxhash = rxhash;
372 e->rps_rxhash = 0;
373 e->queue_index = queue_index;
374 e->tun = tun;
375 hlist_add_head_rcu(&e->hash_link, head);
376 ++tun->flow_count;
377 }
378 return e;
379 }
380
381 static void tun_flow_delete(struct tun_struct *tun, struct tun_flow_entry *e)
382 {
383 netif_info(tun, tx_queued, tun->dev, "delete flow: hash %u index %u\n",
384 e->rxhash, e->queue_index);
385 hlist_del_rcu(&e->hash_link);
386 kfree_rcu(e, rcu);
387 --tun->flow_count;
388 }
389
390 static void tun_flow_flush(struct tun_struct *tun)
391 {
392 int i;
393
394 spin_lock_bh(&tun->lock);
395 for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
396 struct tun_flow_entry *e;
397 struct hlist_node *n;
398
399 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link)
400 tun_flow_delete(tun, e);
401 }
402 spin_unlock_bh(&tun->lock);
403 }
404
405 static void tun_flow_delete_by_queue(struct tun_struct *tun, u16 queue_index)
406 {
407 int i;
408
409 spin_lock_bh(&tun->lock);
410 for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
411 struct tun_flow_entry *e;
412 struct hlist_node *n;
413
414 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
415 if (e->queue_index == queue_index)
416 tun_flow_delete(tun, e);
417 }
418 }
419 spin_unlock_bh(&tun->lock);
420 }
421
422 static void tun_flow_cleanup(struct timer_list *t)
423 {
424 struct tun_struct *tun = from_timer(tun, t, flow_gc_timer);
425 unsigned long delay = tun->ageing_time;
426 unsigned long next_timer = jiffies + delay;
427 unsigned long count = 0;
428 int i;
429
430 spin_lock(&tun->lock);
431 for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
432 struct tun_flow_entry *e;
433 struct hlist_node *n;
434
435 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
436 unsigned long this_timer;
437
438 this_timer = e->updated + delay;
439 if (time_before_eq(this_timer, jiffies)) {
440 tun_flow_delete(tun, e);
441 continue;
442 }
443 count++;
444 if (time_before(this_timer, next_timer))
445 next_timer = this_timer;
446 }
447 }
448
449 if (count)
450 mod_timer(&tun->flow_gc_timer, round_jiffies_up(next_timer));
451 spin_unlock(&tun->lock);
452 }
453
454 static void tun_flow_update(struct tun_struct *tun, u32 rxhash,
455 struct tun_file *tfile)
456 {
457 struct hlist_head *head;
458 struct tun_flow_entry *e;
459 unsigned long delay = tun->ageing_time;
460 u16 queue_index = tfile->queue_index;
461
462 head = &tun->flows[tun_hashfn(rxhash)];
463
464 rcu_read_lock();
465
466 e = tun_flow_find(head, rxhash);
467 if (likely(e)) {
468 /* TODO: keep queueing to old queue until it's empty? */
469 if (READ_ONCE(e->queue_index) != queue_index)
470 WRITE_ONCE(e->queue_index, queue_index);
471 if (e->updated != jiffies)
472 e->updated = jiffies;
473 sock_rps_record_flow_hash(e->rps_rxhash);
474 } else {
475 spin_lock_bh(&tun->lock);
476 if (!tun_flow_find(head, rxhash) &&
477 tun->flow_count < MAX_TAP_FLOWS)
478 tun_flow_create(tun, head, rxhash, queue_index);
479
480 if (!timer_pending(&tun->flow_gc_timer))
481 mod_timer(&tun->flow_gc_timer,
482 round_jiffies_up(jiffies + delay));
483 spin_unlock_bh(&tun->lock);
484 }
485
486 rcu_read_unlock();
487 }
488
489 /* Save the hash received in the stack receive path and update the
490 * flow_hash table accordingly.
491 */
492 static inline void tun_flow_save_rps_rxhash(struct tun_flow_entry *e, u32 hash)
493 {
494 if (unlikely(e->rps_rxhash != hash))
495 e->rps_rxhash = hash;
496 }
497
498 /* We try to identify a flow through its rxhash. The reason that
499 * we do not check rxq no. is because some cards(e.g 82599), chooses
500 * the rxq based on the txq where the last packet of the flow comes. As
501 * the userspace application move between processors, we may get a
502 * different rxq no. here.
503 */
504 static u16 tun_automq_select_queue(struct tun_struct *tun, struct sk_buff *skb)
505 {
506 struct tun_flow_entry *e;
507 u32 txq = 0;
508 u32 numqueues = 0;
509
510 numqueues = READ_ONCE(tun->numqueues);
511
512 txq = __skb_get_hash_symmetric(skb);
513 e = tun_flow_find(&tun->flows[tun_hashfn(txq)], txq);
514 if (e) {
515 tun_flow_save_rps_rxhash(e, txq);
516 txq = e->queue_index;
517 } else {
518 /* use multiply and shift instead of expensive divide */
519 txq = ((u64)txq * numqueues) >> 32;
520 }
521
522 return txq;
523 }
524
525 static u16 tun_ebpf_select_queue(struct tun_struct *tun, struct sk_buff *skb)
526 {
527 struct tun_prog *prog;
528 u32 numqueues;
529 u16 ret = 0;
530
531 numqueues = READ_ONCE(tun->numqueues);
532 if (!numqueues)
533 return 0;
534
535 prog = rcu_dereference(tun->steering_prog);
536 if (prog)
537 ret = bpf_prog_run_clear_cb(prog->prog, skb);
538
539 return ret % numqueues;
540 }
541
542 static u16 tun_select_queue(struct net_device *dev, struct sk_buff *skb,
543 struct net_device *sb_dev)
544 {
545 struct tun_struct *tun = netdev_priv(dev);
546 u16 ret;
547
548 rcu_read_lock();
549 if (rcu_dereference(tun->steering_prog))
550 ret = tun_ebpf_select_queue(tun, skb);
551 else
552 ret = tun_automq_select_queue(tun, skb);
553 rcu_read_unlock();
554
555 return ret;
556 }
557
558 static inline bool tun_not_capable(struct tun_struct *tun)
559 {
560 const struct cred *cred = current_cred();
561 struct net *net = dev_net(tun->dev);
562
563 return ((uid_valid(tun->owner) && !uid_eq(cred->euid, tun->owner)) ||
564 (gid_valid(tun->group) && !in_egroup_p(tun->group))) &&
565 !ns_capable(net->user_ns, CAP_NET_ADMIN);
566 }
567
568 static void tun_set_real_num_queues(struct tun_struct *tun)
569 {
570 netif_set_real_num_tx_queues(tun->dev, tun->numqueues);
571 netif_set_real_num_rx_queues(tun->dev, tun->numqueues);
572 }
573
574 static void tun_disable_queue(struct tun_struct *tun, struct tun_file *tfile)
575 {
576 tfile->detached = tun;
577 list_add_tail(&tfile->next, &tun->disabled);
578 ++tun->numdisabled;
579 }
580
581 static struct tun_struct *tun_enable_queue(struct tun_file *tfile)
582 {
583 struct tun_struct *tun = tfile->detached;
584
585 tfile->detached = NULL;
586 list_del_init(&tfile->next);
587 --tun->numdisabled;
588 return tun;
589 }
590
591 void tun_ptr_free(void *ptr)
592 {
593 if (!ptr)
594 return;
595 if (tun_is_xdp_frame(ptr)) {
596 struct xdp_frame *xdpf = tun_ptr_to_xdp(ptr);
597
598 xdp_return_frame(xdpf);
599 } else {
600 __skb_array_destroy_skb(ptr);
601 }
602 }
603 EXPORT_SYMBOL_GPL(tun_ptr_free);
604
605 static void tun_queue_purge(struct tun_file *tfile)
606 {
607 void *ptr;
608
609 while ((ptr = ptr_ring_consume(&tfile->tx_ring)) != NULL)
610 tun_ptr_free(ptr);
611
612 skb_queue_purge(&tfile->sk.sk_write_queue);
613 skb_queue_purge(&tfile->sk.sk_error_queue);
614 }
615
616 static void __tun_detach(struct tun_file *tfile, bool clean)
617 {
618 struct tun_file *ntfile;
619 struct tun_struct *tun;
620
621 tun = rtnl_dereference(tfile->tun);
622
623 if (tun && clean) {
624 tun_napi_disable(tfile);
625 tun_napi_del(tfile);
626 }
627
628 if (tun && !tfile->detached) {
629 u16 index = tfile->queue_index;
630 BUG_ON(index >= tun->numqueues);
631
632 rcu_assign_pointer(tun->tfiles[index],
633 tun->tfiles[tun->numqueues - 1]);
634 ntfile = rtnl_dereference(tun->tfiles[index]);
635 ntfile->queue_index = index;
636 rcu_assign_pointer(tun->tfiles[tun->numqueues - 1],
637 NULL);
638
639 --tun->numqueues;
640 if (clean) {
641 RCU_INIT_POINTER(tfile->tun, NULL);
642 sock_put(&tfile->sk);
643 } else
644 tun_disable_queue(tun, tfile);
645
646 synchronize_net();
647 tun_flow_delete_by_queue(tun, tun->numqueues + 1);
648 /* Drop read queue */
649 tun_queue_purge(tfile);
650 tun_set_real_num_queues(tun);
651 } else if (tfile->detached && clean) {
652 tun = tun_enable_queue(tfile);
653 sock_put(&tfile->sk);
654 }
655
656 if (clean) {
657 if (tun && tun->numqueues == 0 && tun->numdisabled == 0) {
658 netif_carrier_off(tun->dev);
659
660 if (!(tun->flags & IFF_PERSIST) &&
661 tun->dev->reg_state == NETREG_REGISTERED)
662 unregister_netdevice(tun->dev);
663 }
664 if (tun)
665 xdp_rxq_info_unreg(&tfile->xdp_rxq);
666 ptr_ring_cleanup(&tfile->tx_ring, tun_ptr_free);
667 sock_put(&tfile->sk);
668 }
669 }
670
671 static void tun_detach(struct tun_file *tfile, bool clean)
672 {
673 struct tun_struct *tun;
674 struct net_device *dev;
675
676 rtnl_lock();
677 tun = rtnl_dereference(tfile->tun);
678 dev = tun ? tun->dev : NULL;
679 __tun_detach(tfile, clean);
680 if (dev)
681 netdev_state_change(dev);
682 rtnl_unlock();
683 }
684
685 static void tun_detach_all(struct net_device *dev)
686 {
687 struct tun_struct *tun = netdev_priv(dev);
688 struct tun_file *tfile, *tmp;
689 int i, n = tun->numqueues;
690
691 for (i = 0; i < n; i++) {
692 tfile = rtnl_dereference(tun->tfiles[i]);
693 BUG_ON(!tfile);
694 tun_napi_disable(tfile);
695 tfile->socket.sk->sk_shutdown = RCV_SHUTDOWN;
696 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
697 RCU_INIT_POINTER(tfile->tun, NULL);
698 --tun->numqueues;
699 }
700 list_for_each_entry(tfile, &tun->disabled, next) {
701 tfile->socket.sk->sk_shutdown = RCV_SHUTDOWN;
702 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
703 RCU_INIT_POINTER(tfile->tun, NULL);
704 }
705 BUG_ON(tun->numqueues != 0);
706
707 synchronize_net();
708 for (i = 0; i < n; i++) {
709 tfile = rtnl_dereference(tun->tfiles[i]);
710 tun_napi_del(tfile);
711 /* Drop read queue */
712 tun_queue_purge(tfile);
713 xdp_rxq_info_unreg(&tfile->xdp_rxq);
714 sock_put(&tfile->sk);
715 }
716 list_for_each_entry_safe(tfile, tmp, &tun->disabled, next) {
717 tun_enable_queue(tfile);
718 tun_queue_purge(tfile);
719 xdp_rxq_info_unreg(&tfile->xdp_rxq);
720 sock_put(&tfile->sk);
721 }
722 BUG_ON(tun->numdisabled != 0);
723
724 if (tun->flags & IFF_PERSIST)
725 module_put(THIS_MODULE);
726 }
727
728 static int tun_attach(struct tun_struct *tun, struct file *file,
729 bool skip_filter, bool napi, bool napi_frags,
730 bool publish_tun)
731 {
732 struct tun_file *tfile = file->private_data;
733 struct net_device *dev = tun->dev;
734 int err;
735
736 err = security_tun_dev_attach(tfile->socket.sk, tun->security);
737 if (err < 0)
738 goto out;
739
740 err = -EINVAL;
741 if (rtnl_dereference(tfile->tun) && !tfile->detached)
742 goto out;
743
744 err = -EBUSY;
745 if (!(tun->flags & IFF_MULTI_QUEUE) && tun->numqueues == 1)
746 goto out;
747
748 err = -E2BIG;
749 if (!tfile->detached &&
750 tun->numqueues + tun->numdisabled == MAX_TAP_QUEUES)
751 goto out;
752
753 err = 0;
754
755 /* Re-attach the filter to persist device */
756 if (!skip_filter && (tun->filter_attached == true)) {
757 lock_sock(tfile->socket.sk);
758 err = sk_attach_filter(&tun->fprog, tfile->socket.sk);
759 release_sock(tfile->socket.sk);
760 if (!err)
761 goto out;
762 }
763
764 if (!tfile->detached &&
765 ptr_ring_resize(&tfile->tx_ring, dev->tx_queue_len,
766 GFP_KERNEL, tun_ptr_free)) {
767 err = -ENOMEM;
768 goto out;
769 }
770
771 tfile->queue_index = tun->numqueues;
772 tfile->socket.sk->sk_shutdown &= ~RCV_SHUTDOWN;
773
774 if (tfile->detached) {
775 /* Re-attach detached tfile, updating XDP queue_index */
776 WARN_ON(!xdp_rxq_info_is_reg(&tfile->xdp_rxq));
777
778 if (tfile->xdp_rxq.queue_index != tfile->queue_index)
779 tfile->xdp_rxq.queue_index = tfile->queue_index;
780 } else {
781 /* Setup XDP RX-queue info, for new tfile getting attached */
782 err = xdp_rxq_info_reg(&tfile->xdp_rxq,
783 tun->dev, tfile->queue_index, 0);
784 if (err < 0)
785 goto out;
786 err = xdp_rxq_info_reg_mem_model(&tfile->xdp_rxq,
787 MEM_TYPE_PAGE_SHARED, NULL);
788 if (err < 0) {
789 xdp_rxq_info_unreg(&tfile->xdp_rxq);
790 goto out;
791 }
792 err = 0;
793 }
794
795 if (tfile->detached) {
796 tun_enable_queue(tfile);
797 } else {
798 sock_hold(&tfile->sk);
799 tun_napi_init(tun, tfile, napi, napi_frags);
800 }
801
802 if (rtnl_dereference(tun->xdp_prog))
803 sock_set_flag(&tfile->sk, SOCK_XDP);
804
805 /* device is allowed to go away first, so no need to hold extra
806 * refcnt.
807 */
808
809 /* Publish tfile->tun and tun->tfiles only after we've fully
810 * initialized tfile; otherwise we risk using half-initialized
811 * object.
812 */
813 if (publish_tun)
814 rcu_assign_pointer(tfile->tun, tun);
815 rcu_assign_pointer(tun->tfiles[tun->numqueues], tfile);
816 tun->numqueues++;
817 tun_set_real_num_queues(tun);
818 out:
819 return err;
820 }
821
822 static struct tun_struct *tun_get(struct tun_file *tfile)
823 {
824 struct tun_struct *tun;
825
826 rcu_read_lock();
827 tun = rcu_dereference(tfile->tun);
828 if (tun)
829 dev_hold(tun->dev);
830 rcu_read_unlock();
831
832 return tun;
833 }
834
835 static void tun_put(struct tun_struct *tun)
836 {
837 dev_put(tun->dev);
838 }
839
840 /* TAP filtering */
841 static void addr_hash_set(u32 *mask, const u8 *addr)
842 {
843 int n = ether_crc(ETH_ALEN, addr) >> 26;
844 mask[n >> 5] |= (1 << (n & 31));
845 }
846
847 static unsigned int addr_hash_test(const u32 *mask, const u8 *addr)
848 {
849 int n = ether_crc(ETH_ALEN, addr) >> 26;
850 return mask[n >> 5] & (1 << (n & 31));
851 }
852
853 static int update_filter(struct tap_filter *filter, void __user *arg)
854 {
855 struct { u8 u[ETH_ALEN]; } *addr;
856 struct tun_filter uf;
857 int err, alen, n, nexact;
858
859 if (copy_from_user(&uf, arg, sizeof(uf)))
860 return -EFAULT;
861
862 if (!uf.count) {
863 /* Disabled */
864 filter->count = 0;
865 return 0;
866 }
867
868 alen = ETH_ALEN * uf.count;
869 addr = memdup_user(arg + sizeof(uf), alen);
870 if (IS_ERR(addr))
871 return PTR_ERR(addr);
872
873 /* The filter is updated without holding any locks. Which is
874 * perfectly safe. We disable it first and in the worst
875 * case we'll accept a few undesired packets. */
876 filter->count = 0;
877 wmb();
878
879 /* Use first set of addresses as an exact filter */
880 for (n = 0; n < uf.count && n < FLT_EXACT_COUNT; n++)
881 memcpy(filter->addr[n], addr[n].u, ETH_ALEN);
882
883 nexact = n;
884
885 /* Remaining multicast addresses are hashed,
886 * unicast will leave the filter disabled. */
887 memset(filter->mask, 0, sizeof(filter->mask));
888 for (; n < uf.count; n++) {
889 if (!is_multicast_ether_addr(addr[n].u)) {
890 err = 0; /* no filter */
891 goto free_addr;
892 }
893 addr_hash_set(filter->mask, addr[n].u);
894 }
895
896 /* For ALLMULTI just set the mask to all ones.
897 * This overrides the mask populated above. */
898 if ((uf.flags & TUN_FLT_ALLMULTI))
899 memset(filter->mask, ~0, sizeof(filter->mask));
900
901 /* Now enable the filter */
902 wmb();
903 filter->count = nexact;
904
905 /* Return the number of exact filters */
906 err = nexact;
907 free_addr:
908 kfree(addr);
909 return err;
910 }
911
912 /* Returns: 0 - drop, !=0 - accept */
913 static int run_filter(struct tap_filter *filter, const struct sk_buff *skb)
914 {
915 /* Cannot use eth_hdr(skb) here because skb_mac_hdr() is incorrect
916 * at this point. */
917 struct ethhdr *eh = (struct ethhdr *) skb->data;
918 int i;
919
920 /* Exact match */
921 for (i = 0; i < filter->count; i++)
922 if (ether_addr_equal(eh->h_dest, filter->addr[i]))
923 return 1;
924
925 /* Inexact match (multicast only) */
926 if (is_multicast_ether_addr(eh->h_dest))
927 return addr_hash_test(filter->mask, eh->h_dest);
928
929 return 0;
930 }
931
932 /*
933 * Checks whether the packet is accepted or not.
934 * Returns: 0 - drop, !=0 - accept
935 */
936 static int check_filter(struct tap_filter *filter, const struct sk_buff *skb)
937 {
938 if (!filter->count)
939 return 1;
940
941 return run_filter(filter, skb);
942 }
943
944 /* Network device part of the driver */
945
946 static const struct ethtool_ops tun_ethtool_ops;
947
948 /* Net device detach from fd. */
949 static void tun_net_uninit(struct net_device *dev)
950 {
951 tun_detach_all(dev);
952 }
953
954 /* Net device open. */
955 static int tun_net_open(struct net_device *dev)
956 {
957 netif_tx_start_all_queues(dev);
958
959 return 0;
960 }
961
962 /* Net device close. */
963 static int tun_net_close(struct net_device *dev)
964 {
965 netif_tx_stop_all_queues(dev);
966 return 0;
967 }
968
969 /* Net device start xmit */
970 static void tun_automq_xmit(struct tun_struct *tun, struct sk_buff *skb)
971 {
972 #ifdef CONFIG_RPS
973 if (tun->numqueues == 1 && static_branch_unlikely(&rps_needed)) {
974 /* Select queue was not called for the skbuff, so we extract the
975 * RPS hash and save it into the flow_table here.
976 */
977 struct tun_flow_entry *e;
978 __u32 rxhash;
979
980 rxhash = __skb_get_hash_symmetric(skb);
981 e = tun_flow_find(&tun->flows[tun_hashfn(rxhash)], rxhash);
982 if (e)
983 tun_flow_save_rps_rxhash(e, rxhash);
984 }
985 #endif
986 }
987
988 static unsigned int run_ebpf_filter(struct tun_struct *tun,
989 struct sk_buff *skb,
990 int len)
991 {
992 struct tun_prog *prog = rcu_dereference(tun->filter_prog);
993
994 if (prog)
995 len = bpf_prog_run_clear_cb(prog->prog, skb);
996
997 return len;
998 }
999
1000 /* Net device start xmit */
1001 static netdev_tx_t tun_net_xmit(struct sk_buff *skb, struct net_device *dev)
1002 {
1003 struct tun_struct *tun = netdev_priv(dev);
1004 int txq = skb->queue_mapping;
1005 struct tun_file *tfile;
1006 int len = skb->len;
1007
1008 rcu_read_lock();
1009 tfile = rcu_dereference(tun->tfiles[txq]);
1010
1011 /* Drop packet if interface is not attached */
1012 if (!tfile)
1013 goto drop;
1014
1015 if (!rcu_dereference(tun->steering_prog))
1016 tun_automq_xmit(tun, skb);
1017
1018 netif_info(tun, tx_queued, tun->dev, "%s %d\n", __func__, skb->len);
1019
1020 /* Drop if the filter does not like it.
1021 * This is a noop if the filter is disabled.
1022 * Filter can be enabled only for the TAP devices. */
1023 if (!check_filter(&tun->txflt, skb))
1024 goto drop;
1025
1026 if (tfile->socket.sk->sk_filter &&
1027 sk_filter(tfile->socket.sk, skb))
1028 goto drop;
1029
1030 len = run_ebpf_filter(tun, skb, len);
1031 if (len == 0 || pskb_trim(skb, len))
1032 goto drop;
1033
1034 if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC)))
1035 goto drop;
1036
1037 skb_tx_timestamp(skb);
1038
1039 /* Orphan the skb - required as we might hang on to it
1040 * for indefinite time.
1041 */
1042 skb_orphan(skb);
1043
1044 nf_reset_ct(skb);
1045
1046 if (ptr_ring_produce(&tfile->tx_ring, skb))
1047 goto drop;
1048
1049 /* Notify and wake up reader process */
1050 if (tfile->flags & TUN_FASYNC)
1051 kill_fasync(&tfile->fasync, SIGIO, POLL_IN);
1052 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
1053
1054 rcu_read_unlock();
1055 return NETDEV_TX_OK;
1056
1057 drop:
1058 atomic_long_inc(&dev->tx_dropped);
1059 skb_tx_error(skb);
1060 kfree_skb(skb);
1061 rcu_read_unlock();
1062 return NET_XMIT_DROP;
1063 }
1064
1065 static void tun_net_mclist(struct net_device *dev)
1066 {
1067 /*
1068 * This callback is supposed to deal with mc filter in
1069 * _rx_ path and has nothing to do with the _tx_ path.
1070 * In rx path we always accept everything userspace gives us.
1071 */
1072 }
1073
1074 static netdev_features_t tun_net_fix_features(struct net_device *dev,
1075 netdev_features_t features)
1076 {
1077 struct tun_struct *tun = netdev_priv(dev);
1078
1079 return (features & tun->set_features) | (features & ~TUN_USER_FEATURES);
1080 }
1081
1082 static void tun_set_headroom(struct net_device *dev, int new_hr)
1083 {
1084 struct tun_struct *tun = netdev_priv(dev);
1085
1086 if (new_hr < NET_SKB_PAD)
1087 new_hr = NET_SKB_PAD;
1088
1089 tun->align = new_hr;
1090 }
1091
1092 static void
1093 tun_net_get_stats64(struct net_device *dev, struct rtnl_link_stats64 *stats)
1094 {
1095 struct tun_struct *tun = netdev_priv(dev);
1096
1097 dev_get_tstats64(dev, stats);
1098
1099 stats->rx_frame_errors +=
1100 (unsigned long)atomic_long_read(&tun->rx_frame_errors);
1101 }
1102
1103 static int tun_xdp_set(struct net_device *dev, struct bpf_prog *prog,
1104 struct netlink_ext_ack *extack)
1105 {
1106 struct tun_struct *tun = netdev_priv(dev);
1107 struct tun_file *tfile;
1108 struct bpf_prog *old_prog;
1109 int i;
1110
1111 old_prog = rtnl_dereference(tun->xdp_prog);
1112 rcu_assign_pointer(tun->xdp_prog, prog);
1113 if (old_prog)
1114 bpf_prog_put(old_prog);
1115
1116 for (i = 0; i < tun->numqueues; i++) {
1117 tfile = rtnl_dereference(tun->tfiles[i]);
1118 if (prog)
1119 sock_set_flag(&tfile->sk, SOCK_XDP);
1120 else
1121 sock_reset_flag(&tfile->sk, SOCK_XDP);
1122 }
1123 list_for_each_entry(tfile, &tun->disabled, next) {
1124 if (prog)
1125 sock_set_flag(&tfile->sk, SOCK_XDP);
1126 else
1127 sock_reset_flag(&tfile->sk, SOCK_XDP);
1128 }
1129
1130 return 0;
1131 }
1132
1133 static int tun_xdp(struct net_device *dev, struct netdev_bpf *xdp)
1134 {
1135 switch (xdp->command) {
1136 case XDP_SETUP_PROG:
1137 return tun_xdp_set(dev, xdp->prog, xdp->extack);
1138 default:
1139 return -EINVAL;
1140 }
1141 }
1142
1143 static int tun_net_change_carrier(struct net_device *dev, bool new_carrier)
1144 {
1145 if (new_carrier) {
1146 struct tun_struct *tun = netdev_priv(dev);
1147
1148 if (!tun->numqueues)
1149 return -EPERM;
1150
1151 netif_carrier_on(dev);
1152 } else {
1153 netif_carrier_off(dev);
1154 }
1155 return 0;
1156 }
1157
1158 static const struct net_device_ops tun_netdev_ops = {
1159 .ndo_uninit = tun_net_uninit,
1160 .ndo_open = tun_net_open,
1161 .ndo_stop = tun_net_close,
1162 .ndo_start_xmit = tun_net_xmit,
1163 .ndo_fix_features = tun_net_fix_features,
1164 .ndo_select_queue = tun_select_queue,
1165 .ndo_set_rx_headroom = tun_set_headroom,
1166 .ndo_get_stats64 = tun_net_get_stats64,
1167 .ndo_change_carrier = tun_net_change_carrier,
1168 };
1169
1170 static void __tun_xdp_flush_tfile(struct tun_file *tfile)
1171 {
1172 /* Notify and wake up reader process */
1173 if (tfile->flags & TUN_FASYNC)
1174 kill_fasync(&tfile->fasync, SIGIO, POLL_IN);
1175 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
1176 }
1177
1178 static int tun_xdp_xmit(struct net_device *dev, int n,
1179 struct xdp_frame **frames, u32 flags)
1180 {
1181 struct tun_struct *tun = netdev_priv(dev);
1182 struct tun_file *tfile;
1183 u32 numqueues;
1184 int drops = 0;
1185 int cnt = n;
1186 int i;
1187
1188 if (unlikely(flags & ~XDP_XMIT_FLAGS_MASK))
1189 return -EINVAL;
1190
1191 rcu_read_lock();
1192
1193 resample:
1194 numqueues = READ_ONCE(tun->numqueues);
1195 if (!numqueues) {
1196 rcu_read_unlock();
1197 return -ENXIO; /* Caller will free/return all frames */
1198 }
1199
1200 tfile = rcu_dereference(tun->tfiles[smp_processor_id() %
1201 numqueues]);
1202 if (unlikely(!tfile))
1203 goto resample;
1204
1205 spin_lock(&tfile->tx_ring.producer_lock);
1206 for (i = 0; i < n; i++) {
1207 struct xdp_frame *xdp = frames[i];
1208 /* Encode the XDP flag into lowest bit for consumer to differ
1209 * XDP buffer from sk_buff.
1210 */
1211 void *frame = tun_xdp_to_ptr(xdp);
1212
1213 if (__ptr_ring_produce(&tfile->tx_ring, frame)) {
1214 atomic_long_inc(&dev->tx_dropped);
1215 xdp_return_frame_rx_napi(xdp);
1216 drops++;
1217 }
1218 }
1219 spin_unlock(&tfile->tx_ring.producer_lock);
1220
1221 if (flags & XDP_XMIT_FLUSH)
1222 __tun_xdp_flush_tfile(tfile);
1223
1224 rcu_read_unlock();
1225 return cnt - drops;
1226 }
1227
1228 static int tun_xdp_tx(struct net_device *dev, struct xdp_buff *xdp)
1229 {
1230 struct xdp_frame *frame = xdp_convert_buff_to_frame(xdp);
1231
1232 if (unlikely(!frame))
1233 return -EOVERFLOW;
1234
1235 return tun_xdp_xmit(dev, 1, &frame, XDP_XMIT_FLUSH);
1236 }
1237
1238 static const struct net_device_ops tap_netdev_ops = {
1239 .ndo_uninit = tun_net_uninit,
1240 .ndo_open = tun_net_open,
1241 .ndo_stop = tun_net_close,
1242 .ndo_start_xmit = tun_net_xmit,
1243 .ndo_fix_features = tun_net_fix_features,
1244 .ndo_set_rx_mode = tun_net_mclist,
1245 .ndo_set_mac_address = eth_mac_addr,
1246 .ndo_validate_addr = eth_validate_addr,
1247 .ndo_select_queue = tun_select_queue,
1248 .ndo_features_check = passthru_features_check,
1249 .ndo_set_rx_headroom = tun_set_headroom,
1250 .ndo_get_stats64 = dev_get_tstats64,
1251 .ndo_bpf = tun_xdp,
1252 .ndo_xdp_xmit = tun_xdp_xmit,
1253 .ndo_change_carrier = tun_net_change_carrier,
1254 };
1255
1256 static void tun_flow_init(struct tun_struct *tun)
1257 {
1258 int i;
1259
1260 for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++)
1261 INIT_HLIST_HEAD(&tun->flows[i]);
1262
1263 tun->ageing_time = TUN_FLOW_EXPIRE;
1264 timer_setup(&tun->flow_gc_timer, tun_flow_cleanup, 0);
1265 mod_timer(&tun->flow_gc_timer,
1266 round_jiffies_up(jiffies + tun->ageing_time));
1267 }
1268
1269 static void tun_flow_uninit(struct tun_struct *tun)
1270 {
1271 del_timer_sync(&tun->flow_gc_timer);
1272 tun_flow_flush(tun);
1273 }
1274
1275 #define MIN_MTU 68
1276 #define MAX_MTU 65535
1277
1278 /* Initialize net device. */
1279 static void tun_net_init(struct net_device *dev)
1280 {
1281 struct tun_struct *tun = netdev_priv(dev);
1282
1283 switch (tun->flags & TUN_TYPE_MASK) {
1284 case IFF_TUN:
1285 dev->netdev_ops = &tun_netdev_ops;
1286 dev->header_ops = &ip_tunnel_header_ops;
1287
1288 /* Point-to-Point TUN Device */
1289 dev->hard_header_len = 0;
1290 dev->addr_len = 0;
1291 dev->mtu = 1500;
1292
1293 /* Zero header length */
1294 dev->type = ARPHRD_NONE;
1295 dev->flags = IFF_POINTOPOINT | IFF_NOARP | IFF_MULTICAST;
1296 break;
1297
1298 case IFF_TAP:
1299 dev->netdev_ops = &tap_netdev_ops;
1300 /* Ethernet TAP Device */
1301 ether_setup(dev);
1302 dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1303 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
1304
1305 eth_hw_addr_random(dev);
1306
1307 break;
1308 }
1309
1310 dev->min_mtu = MIN_MTU;
1311 dev->max_mtu = MAX_MTU - dev->hard_header_len;
1312 }
1313
1314 static bool tun_sock_writeable(struct tun_struct *tun, struct tun_file *tfile)
1315 {
1316 struct sock *sk = tfile->socket.sk;
1317
1318 return (tun->dev->flags & IFF_UP) && sock_writeable(sk);
1319 }
1320
1321 /* Character device part */
1322
1323 /* Poll */
1324 static __poll_t tun_chr_poll(struct file *file, poll_table *wait)
1325 {
1326 struct tun_file *tfile = file->private_data;
1327 struct tun_struct *tun = tun_get(tfile);
1328 struct sock *sk;
1329 __poll_t mask = 0;
1330
1331 if (!tun)
1332 return EPOLLERR;
1333
1334 sk = tfile->socket.sk;
1335
1336 poll_wait(file, sk_sleep(sk), wait);
1337
1338 if (!ptr_ring_empty(&tfile->tx_ring))
1339 mask |= EPOLLIN | EPOLLRDNORM;
1340
1341 /* Make sure SOCKWQ_ASYNC_NOSPACE is set if not writable to
1342 * guarantee EPOLLOUT to be raised by either here or
1343 * tun_sock_write_space(). Then process could get notification
1344 * after it writes to a down device and meets -EIO.
1345 */
1346 if (tun_sock_writeable(tun, tfile) ||
1347 (!test_and_set_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags) &&
1348 tun_sock_writeable(tun, tfile)))
1349 mask |= EPOLLOUT | EPOLLWRNORM;
1350
1351 if (tun->dev->reg_state != NETREG_REGISTERED)
1352 mask = EPOLLERR;
1353
1354 tun_put(tun);
1355 return mask;
1356 }
1357
1358 static struct sk_buff *tun_napi_alloc_frags(struct tun_file *tfile,
1359 size_t len,
1360 const struct iov_iter *it)
1361 {
1362 struct sk_buff *skb;
1363 size_t linear;
1364 int err;
1365 int i;
1366
1367 if (it->nr_segs > MAX_SKB_FRAGS + 1)
1368 return ERR_PTR(-EMSGSIZE);
1369
1370 local_bh_disable();
1371 skb = napi_get_frags(&tfile->napi);
1372 local_bh_enable();
1373 if (!skb)
1374 return ERR_PTR(-ENOMEM);
1375
1376 linear = iov_iter_single_seg_count(it);
1377 err = __skb_grow(skb, linear);
1378 if (err)
1379 goto free;
1380
1381 skb->len = len;
1382 skb->data_len = len - linear;
1383 skb->truesize += skb->data_len;
1384
1385 for (i = 1; i < it->nr_segs; i++) {
1386 size_t fragsz = it->iov[i].iov_len;
1387 struct page *page;
1388 void *frag;
1389
1390 if (fragsz == 0 || fragsz > PAGE_SIZE) {
1391 err = -EINVAL;
1392 goto free;
1393 }
1394 frag = netdev_alloc_frag(fragsz);
1395 if (!frag) {
1396 err = -ENOMEM;
1397 goto free;
1398 }
1399 page = virt_to_head_page(frag);
1400 skb_fill_page_desc(skb, i - 1, page,
1401 frag - page_address(page), fragsz);
1402 }
1403
1404 return skb;
1405 free:
1406 /* frees skb and all frags allocated with napi_alloc_frag() */
1407 napi_free_frags(&tfile->napi);
1408 return ERR_PTR(err);
1409 }
1410
1411 /* prepad is the amount to reserve at front. len is length after that.
1412 * linear is a hint as to how much to copy (usually headers). */
1413 static struct sk_buff *tun_alloc_skb(struct tun_file *tfile,
1414 size_t prepad, size_t len,
1415 size_t linear, int noblock)
1416 {
1417 struct sock *sk = tfile->socket.sk;
1418 struct sk_buff *skb;
1419 int err;
1420
1421 /* Under a page? Don't bother with paged skb. */
1422 if (prepad + len < PAGE_SIZE || !linear)
1423 linear = len;
1424
1425 skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
1426 &err, 0);
1427 if (!skb)
1428 return ERR_PTR(err);
1429
1430 skb_reserve(skb, prepad);
1431 skb_put(skb, linear);
1432 skb->data_len = len - linear;
1433 skb->len += len - linear;
1434
1435 return skb;
1436 }
1437
1438 static void tun_rx_batched(struct tun_struct *tun, struct tun_file *tfile,
1439 struct sk_buff *skb, int more)
1440 {
1441 struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
1442 struct sk_buff_head process_queue;
1443 u32 rx_batched = tun->rx_batched;
1444 bool rcv = false;
1445
1446 if (!rx_batched || (!more && skb_queue_empty(queue))) {
1447 local_bh_disable();
1448 skb_record_rx_queue(skb, tfile->queue_index);
1449 netif_receive_skb(skb);
1450 local_bh_enable();
1451 return;
1452 }
1453
1454 spin_lock(&queue->lock);
1455 if (!more || skb_queue_len(queue) == rx_batched) {
1456 __skb_queue_head_init(&process_queue);
1457 skb_queue_splice_tail_init(queue, &process_queue);
1458 rcv = true;
1459 } else {
1460 __skb_queue_tail(queue, skb);
1461 }
1462 spin_unlock(&queue->lock);
1463
1464 if (rcv) {
1465 struct sk_buff *nskb;
1466
1467 local_bh_disable();
1468 while ((nskb = __skb_dequeue(&process_queue))) {
1469 skb_record_rx_queue(nskb, tfile->queue_index);
1470 netif_receive_skb(nskb);
1471 }
1472 skb_record_rx_queue(skb, tfile->queue_index);
1473 netif_receive_skb(skb);
1474 local_bh_enable();
1475 }
1476 }
1477
1478 static bool tun_can_build_skb(struct tun_struct *tun, struct tun_file *tfile,
1479 int len, int noblock, bool zerocopy)
1480 {
1481 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
1482 return false;
1483
1484 if (tfile->socket.sk->sk_sndbuf != INT_MAX)
1485 return false;
1486
1487 if (!noblock)
1488 return false;
1489
1490 if (zerocopy)
1491 return false;
1492
1493 if (SKB_DATA_ALIGN(len + TUN_RX_PAD) +
1494 SKB_DATA_ALIGN(sizeof(struct skb_shared_info)) > PAGE_SIZE)
1495 return false;
1496
1497 return true;
1498 }
1499
1500 static struct sk_buff *__tun_build_skb(struct tun_file *tfile,
1501 struct page_frag *alloc_frag, char *buf,
1502 int buflen, int len, int pad)
1503 {
1504 struct sk_buff *skb = build_skb(buf, buflen);
1505
1506 if (!skb)
1507 return ERR_PTR(-ENOMEM);
1508
1509 skb_reserve(skb, pad);
1510 skb_put(skb, len);
1511 skb_set_owner_w(skb, tfile->socket.sk);
1512
1513 get_page(alloc_frag->page);
1514 alloc_frag->offset += buflen;
1515
1516 return skb;
1517 }
1518
1519 static int tun_xdp_act(struct tun_struct *tun, struct bpf_prog *xdp_prog,
1520 struct xdp_buff *xdp, u32 act)
1521 {
1522 int err;
1523
1524 switch (act) {
1525 case XDP_REDIRECT:
1526 err = xdp_do_redirect(tun->dev, xdp, xdp_prog);
1527 if (err)
1528 return err;
1529 break;
1530 case XDP_TX:
1531 err = tun_xdp_tx(tun->dev, xdp);
1532 if (err < 0)
1533 return err;
1534 break;
1535 case XDP_PASS:
1536 break;
1537 default:
1538 bpf_warn_invalid_xdp_action(act);
1539 fallthrough;
1540 case XDP_ABORTED:
1541 trace_xdp_exception(tun->dev, xdp_prog, act);
1542 fallthrough;
1543 case XDP_DROP:
1544 atomic_long_inc(&tun->dev->rx_dropped);
1545 break;
1546 }
1547
1548 return act;
1549 }
1550
1551 static struct sk_buff *tun_build_skb(struct tun_struct *tun,
1552 struct tun_file *tfile,
1553 struct iov_iter *from,
1554 struct virtio_net_hdr *hdr,
1555 int len, int *skb_xdp)
1556 {
1557 struct page_frag *alloc_frag = &current->task_frag;
1558 struct bpf_prog *xdp_prog;
1559 int buflen = SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
1560 char *buf;
1561 size_t copied;
1562 int pad = TUN_RX_PAD;
1563 int err = 0;
1564
1565 rcu_read_lock();
1566 xdp_prog = rcu_dereference(tun->xdp_prog);
1567 if (xdp_prog)
1568 pad += XDP_PACKET_HEADROOM;
1569 buflen += SKB_DATA_ALIGN(len + pad);
1570 rcu_read_unlock();
1571
1572 alloc_frag->offset = ALIGN((u64)alloc_frag->offset, SMP_CACHE_BYTES);
1573 if (unlikely(!skb_page_frag_refill(buflen, alloc_frag, GFP_KERNEL)))
1574 return ERR_PTR(-ENOMEM);
1575
1576 buf = (char *)page_address(alloc_frag->page) + alloc_frag->offset;
1577 copied = copy_page_from_iter(alloc_frag->page,
1578 alloc_frag->offset + pad,
1579 len, from);
1580 if (copied != len)
1581 return ERR_PTR(-EFAULT);
1582
1583 /* There's a small window that XDP may be set after the check
1584 * of xdp_prog above, this should be rare and for simplicity
1585 * we do XDP on skb in case the headroom is not enough.
1586 */
1587 if (hdr->gso_type || !xdp_prog) {
1588 *skb_xdp = 1;
1589 return __tun_build_skb(tfile, alloc_frag, buf, buflen, len,
1590 pad);
1591 }
1592
1593 *skb_xdp = 0;
1594
1595 local_bh_disable();
1596 rcu_read_lock();
1597 xdp_prog = rcu_dereference(tun->xdp_prog);
1598 if (xdp_prog) {
1599 struct xdp_buff xdp;
1600 u32 act;
1601
1602 xdp.data_hard_start = buf;
1603 xdp.data = buf + pad;
1604 xdp_set_data_meta_invalid(&xdp);
1605 xdp.data_end = xdp.data + len;
1606 xdp.rxq = &tfile->xdp_rxq;
1607 xdp.frame_sz = buflen;
1608
1609 act = bpf_prog_run_xdp(xdp_prog, &xdp);
1610 if (act == XDP_REDIRECT || act == XDP_TX) {
1611 get_page(alloc_frag->page);
1612 alloc_frag->offset += buflen;
1613 }
1614 err = tun_xdp_act(tun, xdp_prog, &xdp, act);
1615 if (err < 0) {
1616 if (act == XDP_REDIRECT || act == XDP_TX)
1617 put_page(alloc_frag->page);
1618 goto out;
1619 }
1620
1621 if (err == XDP_REDIRECT)
1622 xdp_do_flush();
1623 if (err != XDP_PASS)
1624 goto out;
1625
1626 pad = xdp.data - xdp.data_hard_start;
1627 len = xdp.data_end - xdp.data;
1628 }
1629 rcu_read_unlock();
1630 local_bh_enable();
1631
1632 return __tun_build_skb(tfile, alloc_frag, buf, buflen, len, pad);
1633
1634 out:
1635 rcu_read_unlock();
1636 local_bh_enable();
1637 return NULL;
1638 }
1639
1640 /* Get packet from user space buffer */
1641 static ssize_t tun_get_user(struct tun_struct *tun, struct tun_file *tfile,
1642 void *msg_control, struct iov_iter *from,
1643 int noblock, bool more)
1644 {
1645 struct tun_pi pi = { 0, cpu_to_be16(ETH_P_IP) };
1646 struct sk_buff *skb;
1647 size_t total_len = iov_iter_count(from);
1648 size_t len = total_len, align = tun->align, linear;
1649 struct virtio_net_hdr gso = { 0 };
1650 int good_linear;
1651 int copylen;
1652 bool zerocopy = false;
1653 int err;
1654 u32 rxhash = 0;
1655 int skb_xdp = 1;
1656 bool frags = tun_napi_frags_enabled(tfile);
1657
1658 if (!(tun->flags & IFF_NO_PI)) {
1659 if (len < sizeof(pi))
1660 return -EINVAL;
1661 len -= sizeof(pi);
1662
1663 if (!copy_from_iter_full(&pi, sizeof(pi), from))
1664 return -EFAULT;
1665 }
1666
1667 if (tun->flags & IFF_VNET_HDR) {
1668 int vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
1669
1670 if (len < vnet_hdr_sz)
1671 return -EINVAL;
1672 len -= vnet_hdr_sz;
1673
1674 if (!copy_from_iter_full(&gso, sizeof(gso), from))
1675 return -EFAULT;
1676
1677 if ((gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
1678 tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2 > tun16_to_cpu(tun, gso.hdr_len))
1679 gso.hdr_len = cpu_to_tun16(tun, tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2);
1680
1681 if (tun16_to_cpu(tun, gso.hdr_len) > len)
1682 return -EINVAL;
1683 iov_iter_advance(from, vnet_hdr_sz - sizeof(gso));
1684 }
1685
1686 if ((tun->flags & TUN_TYPE_MASK) == IFF_TAP) {
1687 align += NET_IP_ALIGN;
1688 if (unlikely(len < ETH_HLEN ||
1689 (gso.hdr_len && tun16_to_cpu(tun, gso.hdr_len) < ETH_HLEN)))
1690 return -EINVAL;
1691 }
1692
1693 good_linear = SKB_MAX_HEAD(align);
1694
1695 if (msg_control) {
1696 struct iov_iter i = *from;
1697
1698 /* There are 256 bytes to be copied in skb, so there is
1699 * enough room for skb expand head in case it is used.
1700 * The rest of the buffer is mapped from userspace.
1701 */
1702 copylen = gso.hdr_len ? tun16_to_cpu(tun, gso.hdr_len) : GOODCOPY_LEN;
1703 if (copylen > good_linear)
1704 copylen = good_linear;
1705 linear = copylen;
1706 iov_iter_advance(&i, copylen);
1707 if (iov_iter_npages(&i, INT_MAX) <= MAX_SKB_FRAGS)
1708 zerocopy = true;
1709 }
1710
1711 if (!frags && tun_can_build_skb(tun, tfile, len, noblock, zerocopy)) {
1712 /* For the packet that is not easy to be processed
1713 * (e.g gso or jumbo packet), we will do it at after
1714 * skb was created with generic XDP routine.
1715 */
1716 skb = tun_build_skb(tun, tfile, from, &gso, len, &skb_xdp);
1717 if (IS_ERR(skb)) {
1718 atomic_long_inc(&tun->dev->rx_dropped);
1719 return PTR_ERR(skb);
1720 }
1721 if (!skb)
1722 return total_len;
1723 } else {
1724 if (!zerocopy) {
1725 copylen = len;
1726 if (tun16_to_cpu(tun, gso.hdr_len) > good_linear)
1727 linear = good_linear;
1728 else
1729 linear = tun16_to_cpu(tun, gso.hdr_len);
1730 }
1731
1732 if (frags) {
1733 mutex_lock(&tfile->napi_mutex);
1734 skb = tun_napi_alloc_frags(tfile, copylen, from);
1735 /* tun_napi_alloc_frags() enforces a layout for the skb.
1736 * If zerocopy is enabled, then this layout will be
1737 * overwritten by zerocopy_sg_from_iter().
1738 */
1739 zerocopy = false;
1740 } else {
1741 skb = tun_alloc_skb(tfile, align, copylen, linear,
1742 noblock);
1743 }
1744
1745 if (IS_ERR(skb)) {
1746 if (PTR_ERR(skb) != -EAGAIN)
1747 atomic_long_inc(&tun->dev->rx_dropped);
1748 if (frags)
1749 mutex_unlock(&tfile->napi_mutex);
1750 return PTR_ERR(skb);
1751 }
1752
1753 if (zerocopy)
1754 err = zerocopy_sg_from_iter(skb, from);
1755 else
1756 err = skb_copy_datagram_from_iter(skb, 0, from, len);
1757
1758 if (err) {
1759 err = -EFAULT;
1760 drop:
1761 atomic_long_inc(&tun->dev->rx_dropped);
1762 kfree_skb(skb);
1763 if (frags) {
1764 tfile->napi.skb = NULL;
1765 mutex_unlock(&tfile->napi_mutex);
1766 }
1767
1768 return err;
1769 }
1770 }
1771
1772 if (virtio_net_hdr_to_skb(skb, &gso, tun_is_little_endian(tun))) {
1773 atomic_long_inc(&tun->rx_frame_errors);
1774 kfree_skb(skb);
1775 if (frags) {
1776 tfile->napi.skb = NULL;
1777 mutex_unlock(&tfile->napi_mutex);
1778 }
1779
1780 return -EINVAL;
1781 }
1782
1783 switch (tun->flags & TUN_TYPE_MASK) {
1784 case IFF_TUN:
1785 if (tun->flags & IFF_NO_PI) {
1786 u8 ip_version = skb->len ? (skb->data[0] >> 4) : 0;
1787
1788 switch (ip_version) {
1789 case 4:
1790 pi.proto = htons(ETH_P_IP);
1791 break;
1792 case 6:
1793 pi.proto = htons(ETH_P_IPV6);
1794 break;
1795 default:
1796 atomic_long_inc(&tun->dev->rx_dropped);
1797 kfree_skb(skb);
1798 return -EINVAL;
1799 }
1800 }
1801
1802 skb_reset_mac_header(skb);
1803 skb->protocol = pi.proto;
1804 skb->dev = tun->dev;
1805 break;
1806 case IFF_TAP:
1807 if (frags && !pskb_may_pull(skb, ETH_HLEN)) {
1808 err = -ENOMEM;
1809 goto drop;
1810 }
1811 skb->protocol = eth_type_trans(skb, tun->dev);
1812 break;
1813 }
1814
1815 /* copy skb_ubuf_info for callback when skb has no error */
1816 if (zerocopy) {
1817 skb_shinfo(skb)->destructor_arg = msg_control;
1818 skb_shinfo(skb)->tx_flags |= SKBTX_DEV_ZEROCOPY;
1819 skb_shinfo(skb)->tx_flags |= SKBTX_SHARED_FRAG;
1820 } else if (msg_control) {
1821 struct ubuf_info *uarg = msg_control;
1822 uarg->callback(uarg, false);
1823 }
1824
1825 skb_reset_network_header(skb);
1826 skb_probe_transport_header(skb);
1827 skb_record_rx_queue(skb, tfile->queue_index);
1828
1829 if (skb_xdp) {
1830 struct bpf_prog *xdp_prog;
1831 int ret;
1832
1833 local_bh_disable();
1834 rcu_read_lock();
1835 xdp_prog = rcu_dereference(tun->xdp_prog);
1836 if (xdp_prog) {
1837 ret = do_xdp_generic(xdp_prog, skb);
1838 if (ret != XDP_PASS) {
1839 rcu_read_unlock();
1840 local_bh_enable();
1841 if (frags) {
1842 tfile->napi.skb = NULL;
1843 mutex_unlock(&tfile->napi_mutex);
1844 }
1845 return total_len;
1846 }
1847 }
1848 rcu_read_unlock();
1849 local_bh_enable();
1850 }
1851
1852 /* Compute the costly rx hash only if needed for flow updates.
1853 * We may get a very small possibility of OOO during switching, not
1854 * worth to optimize.
1855 */
1856 if (!rcu_access_pointer(tun->steering_prog) && tun->numqueues > 1 &&
1857 !tfile->detached)
1858 rxhash = __skb_get_hash_symmetric(skb);
1859
1860 rcu_read_lock();
1861 if (unlikely(!(tun->dev->flags & IFF_UP))) {
1862 err = -EIO;
1863 rcu_read_unlock();
1864 goto drop;
1865 }
1866
1867 if (frags) {
1868 u32 headlen;
1869
1870 /* Exercise flow dissector code path. */
1871 skb_push(skb, ETH_HLEN);
1872 headlen = eth_get_headlen(tun->dev, skb->data,
1873 skb_headlen(skb));
1874
1875 if (unlikely(headlen > skb_headlen(skb))) {
1876 atomic_long_inc(&tun->dev->rx_dropped);
1877 napi_free_frags(&tfile->napi);
1878 rcu_read_unlock();
1879 mutex_unlock(&tfile->napi_mutex);
1880 WARN_ON(1);
1881 return -ENOMEM;
1882 }
1883
1884 local_bh_disable();
1885 napi_gro_frags(&tfile->napi);
1886 local_bh_enable();
1887 mutex_unlock(&tfile->napi_mutex);
1888 } else if (tfile->napi_enabled) {
1889 struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
1890 int queue_len;
1891
1892 spin_lock_bh(&queue->lock);
1893 __skb_queue_tail(queue, skb);
1894 queue_len = skb_queue_len(queue);
1895 spin_unlock(&queue->lock);
1896
1897 if (!more || queue_len > NAPI_POLL_WEIGHT)
1898 napi_schedule(&tfile->napi);
1899
1900 local_bh_enable();
1901 } else if (!IS_ENABLED(CONFIG_4KSTACKS)) {
1902 tun_rx_batched(tun, tfile, skb, more);
1903 } else {
1904 netif_rx_ni(skb);
1905 }
1906 rcu_read_unlock();
1907
1908 preempt_disable();
1909 dev_sw_netstats_rx_add(tun->dev, len);
1910 preempt_enable();
1911
1912 if (rxhash)
1913 tun_flow_update(tun, rxhash, tfile);
1914
1915 return total_len;
1916 }
1917
1918 static ssize_t tun_chr_write_iter(struct kiocb *iocb, struct iov_iter *from)
1919 {
1920 struct file *file = iocb->ki_filp;
1921 struct tun_file *tfile = file->private_data;
1922 struct tun_struct *tun = tun_get(tfile);
1923 ssize_t result;
1924 int noblock = 0;
1925
1926 if (!tun)
1927 return -EBADFD;
1928
1929 if ((file->f_flags & O_NONBLOCK) || (iocb->ki_flags & IOCB_NOWAIT))
1930 noblock = 1;
1931
1932 result = tun_get_user(tun, tfile, NULL, from, noblock, false);
1933
1934 tun_put(tun);
1935 return result;
1936 }
1937
1938 static ssize_t tun_put_user_xdp(struct tun_struct *tun,
1939 struct tun_file *tfile,
1940 struct xdp_frame *xdp_frame,
1941 struct iov_iter *iter)
1942 {
1943 int vnet_hdr_sz = 0;
1944 size_t size = xdp_frame->len;
1945 size_t ret;
1946
1947 if (tun->flags & IFF_VNET_HDR) {
1948 struct virtio_net_hdr gso = { 0 };
1949
1950 vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
1951 if (unlikely(iov_iter_count(iter) < vnet_hdr_sz))
1952 return -EINVAL;
1953 if (unlikely(copy_to_iter(&gso, sizeof(gso), iter) !=
1954 sizeof(gso)))
1955 return -EFAULT;
1956 iov_iter_advance(iter, vnet_hdr_sz - sizeof(gso));
1957 }
1958
1959 ret = copy_to_iter(xdp_frame->data, size, iter) + vnet_hdr_sz;
1960
1961 preempt_disable();
1962 dev_sw_netstats_tx_add(tun->dev, 1, ret);
1963 preempt_enable();
1964
1965 return ret;
1966 }
1967
1968 /* Put packet to the user space buffer */
1969 static ssize_t tun_put_user(struct tun_struct *tun,
1970 struct tun_file *tfile,
1971 struct sk_buff *skb,
1972 struct iov_iter *iter)
1973 {
1974 struct tun_pi pi = { 0, skb->protocol };
1975 ssize_t total;
1976 int vlan_offset = 0;
1977 int vlan_hlen = 0;
1978 int vnet_hdr_sz = 0;
1979
1980 if (skb_vlan_tag_present(skb))
1981 vlan_hlen = VLAN_HLEN;
1982
1983 if (tun->flags & IFF_VNET_HDR)
1984 vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
1985
1986 total = skb->len + vlan_hlen + vnet_hdr_sz;
1987
1988 if (!(tun->flags & IFF_NO_PI)) {
1989 if (iov_iter_count(iter) < sizeof(pi))
1990 return -EINVAL;
1991
1992 total += sizeof(pi);
1993 if (iov_iter_count(iter) < total) {
1994 /* Packet will be striped */
1995 pi.flags |= TUN_PKT_STRIP;
1996 }
1997
1998 if (copy_to_iter(&pi, sizeof(pi), iter) != sizeof(pi))
1999 return -EFAULT;
2000 }
2001
2002 if (vnet_hdr_sz) {
2003 struct virtio_net_hdr gso;
2004
2005 if (iov_iter_count(iter) < vnet_hdr_sz)
2006 return -EINVAL;
2007
2008 if (virtio_net_hdr_from_skb(skb, &gso,
2009 tun_is_little_endian(tun), true,
2010 vlan_hlen)) {
2011 struct skb_shared_info *sinfo = skb_shinfo(skb);
2012 pr_err("unexpected GSO type: "
2013 "0x%x, gso_size %d, hdr_len %d\n",
2014 sinfo->gso_type, tun16_to_cpu(tun, gso.gso_size),
2015 tun16_to_cpu(tun, gso.hdr_len));
2016 print_hex_dump(KERN_ERR, "tun: ",
2017 DUMP_PREFIX_NONE,
2018 16, 1, skb->head,
2019 min((int)tun16_to_cpu(tun, gso.hdr_len), 64), true);
2020 WARN_ON_ONCE(1);
2021 return -EINVAL;
2022 }
2023
2024 if (copy_to_iter(&gso, sizeof(gso), iter) != sizeof(gso))
2025 return -EFAULT;
2026
2027 iov_iter_advance(iter, vnet_hdr_sz - sizeof(gso));
2028 }
2029
2030 if (vlan_hlen) {
2031 int ret;
2032 struct veth veth;
2033
2034 veth.h_vlan_proto = skb->vlan_proto;
2035 veth.h_vlan_TCI = htons(skb_vlan_tag_get(skb));
2036
2037 vlan_offset = offsetof(struct vlan_ethhdr, h_vlan_proto);
2038
2039 ret = skb_copy_datagram_iter(skb, 0, iter, vlan_offset);
2040 if (ret || !iov_iter_count(iter))
2041 goto done;
2042
2043 ret = copy_to_iter(&veth, sizeof(veth), iter);
2044 if (ret != sizeof(veth) || !iov_iter_count(iter))
2045 goto done;
2046 }
2047
2048 skb_copy_datagram_iter(skb, vlan_offset, iter, skb->len - vlan_offset);
2049
2050 done:
2051 /* caller is in process context, */
2052 preempt_disable();
2053 dev_sw_netstats_tx_add(tun->dev, 1, skb->len + vlan_hlen);
2054 preempt_enable();
2055
2056 return total;
2057 }
2058
2059 static void *tun_ring_recv(struct tun_file *tfile, int noblock, int *err)
2060 {
2061 DECLARE_WAITQUEUE(wait, current);
2062 void *ptr = NULL;
2063 int error = 0;
2064
2065 ptr = ptr_ring_consume(&tfile->tx_ring);
2066 if (ptr)
2067 goto out;
2068 if (noblock) {
2069 error = -EAGAIN;
2070 goto out;
2071 }
2072
2073 add_wait_queue(&tfile->socket.wq.wait, &wait);
2074
2075 while (1) {
2076 set_current_state(TASK_INTERRUPTIBLE);
2077 ptr = ptr_ring_consume(&tfile->tx_ring);
2078 if (ptr)
2079 break;
2080 if (signal_pending(current)) {
2081 error = -ERESTARTSYS;
2082 break;
2083 }
2084 if (tfile->socket.sk->sk_shutdown & RCV_SHUTDOWN) {
2085 error = -EFAULT;
2086 break;
2087 }
2088
2089 schedule();
2090 }
2091
2092 __set_current_state(TASK_RUNNING);
2093 remove_wait_queue(&tfile->socket.wq.wait, &wait);
2094
2095 out:
2096 *err = error;
2097 return ptr;
2098 }
2099
2100 static ssize_t tun_do_read(struct tun_struct *tun, struct tun_file *tfile,
2101 struct iov_iter *to,
2102 int noblock, void *ptr)
2103 {
2104 ssize_t ret;
2105 int err;
2106
2107 if (!iov_iter_count(to)) {
2108 tun_ptr_free(ptr);
2109 return 0;
2110 }
2111
2112 if (!ptr) {
2113 /* Read frames from ring */
2114 ptr = tun_ring_recv(tfile, noblock, &err);
2115 if (!ptr)
2116 return err;
2117 }
2118
2119 if (tun_is_xdp_frame(ptr)) {
2120 struct xdp_frame *xdpf = tun_ptr_to_xdp(ptr);
2121
2122 ret = tun_put_user_xdp(tun, tfile, xdpf, to);
2123 xdp_return_frame(xdpf);
2124 } else {
2125 struct sk_buff *skb = ptr;
2126
2127 ret = tun_put_user(tun, tfile, skb, to);
2128 if (unlikely(ret < 0))
2129 kfree_skb(skb);
2130 else
2131 consume_skb(skb);
2132 }
2133
2134 return ret;
2135 }
2136
2137 static ssize_t tun_chr_read_iter(struct kiocb *iocb, struct iov_iter *to)
2138 {
2139 struct file *file = iocb->ki_filp;
2140 struct tun_file *tfile = file->private_data;
2141 struct tun_struct *tun = tun_get(tfile);
2142 ssize_t len = iov_iter_count(to), ret;
2143 int noblock = 0;
2144
2145 if (!tun)
2146 return -EBADFD;
2147
2148 if ((file->f_flags & O_NONBLOCK) || (iocb->ki_flags & IOCB_NOWAIT))
2149 noblock = 1;
2150
2151 ret = tun_do_read(tun, tfile, to, noblock, NULL);
2152 ret = min_t(ssize_t, ret, len);
2153 if (ret > 0)
2154 iocb->ki_pos = ret;
2155 tun_put(tun);
2156 return ret;
2157 }
2158
2159 static void tun_prog_free(struct rcu_head *rcu)
2160 {
2161 struct tun_prog *prog = container_of(rcu, struct tun_prog, rcu);
2162
2163 bpf_prog_destroy(prog->prog);
2164 kfree(prog);
2165 }
2166
2167 static int __tun_set_ebpf(struct tun_struct *tun,
2168 struct tun_prog __rcu **prog_p,
2169 struct bpf_prog *prog)
2170 {
2171 struct tun_prog *old, *new = NULL;
2172
2173 if (prog) {
2174 new = kmalloc(sizeof(*new), GFP_KERNEL);
2175 if (!new)
2176 return -ENOMEM;
2177 new->prog = prog;
2178 }
2179
2180 spin_lock_bh(&tun->lock);
2181 old = rcu_dereference_protected(*prog_p,
2182 lockdep_is_held(&tun->lock));
2183 rcu_assign_pointer(*prog_p, new);
2184 spin_unlock_bh(&tun->lock);
2185
2186 if (old)
2187 call_rcu(&old->rcu, tun_prog_free);
2188
2189 return 0;
2190 }
2191
2192 static void tun_free_netdev(struct net_device *dev)
2193 {
2194 struct tun_struct *tun = netdev_priv(dev);
2195
2196 BUG_ON(!(list_empty(&tun->disabled)));
2197
2198 free_percpu(dev->tstats);
2199 /* We clear tstats so that tun_set_iff() can tell if
2200 * tun_free_netdev() has been called from register_netdevice().
2201 */
2202 dev->tstats = NULL;
2203
2204 tun_flow_uninit(tun);
2205 security_tun_dev_free_security(tun->security);
2206 __tun_set_ebpf(tun, &tun->steering_prog, NULL);
2207 __tun_set_ebpf(tun, &tun->filter_prog, NULL);
2208 }
2209
2210 static void tun_setup(struct net_device *dev)
2211 {
2212 struct tun_struct *tun = netdev_priv(dev);
2213
2214 tun->owner = INVALID_UID;
2215 tun->group = INVALID_GID;
2216 tun_default_link_ksettings(dev, &tun->link_ksettings);
2217
2218 dev->ethtool_ops = &tun_ethtool_ops;
2219 dev->needs_free_netdev = true;
2220 dev->priv_destructor = tun_free_netdev;
2221 /* We prefer our own queue length */
2222 dev->tx_queue_len = TUN_READQ_SIZE;
2223 }
2224
2225 /* Trivial set of netlink ops to allow deleting tun or tap
2226 * device with netlink.
2227 */
2228 static int tun_validate(struct nlattr *tb[], struct nlattr *data[],
2229 struct netlink_ext_ack *extack)
2230 {
2231 NL_SET_ERR_MSG(extack,
2232 "tun/tap creation via rtnetlink is not supported.");
2233 return -EOPNOTSUPP;
2234 }
2235
2236 static size_t tun_get_size(const struct net_device *dev)
2237 {
2238 BUILD_BUG_ON(sizeof(u32) != sizeof(uid_t));
2239 BUILD_BUG_ON(sizeof(u32) != sizeof(gid_t));
2240
2241 return nla_total_size(sizeof(uid_t)) + /* OWNER */
2242 nla_total_size(sizeof(gid_t)) + /* GROUP */
2243 nla_total_size(sizeof(u8)) + /* TYPE */
2244 nla_total_size(sizeof(u8)) + /* PI */
2245 nla_total_size(sizeof(u8)) + /* VNET_HDR */
2246 nla_total_size(sizeof(u8)) + /* PERSIST */
2247 nla_total_size(sizeof(u8)) + /* MULTI_QUEUE */
2248 nla_total_size(sizeof(u32)) + /* NUM_QUEUES */
2249 nla_total_size(sizeof(u32)) + /* NUM_DISABLED_QUEUES */
2250 0;
2251 }
2252
2253 static int tun_fill_info(struct sk_buff *skb, const struct net_device *dev)
2254 {
2255 struct tun_struct *tun = netdev_priv(dev);
2256
2257 if (nla_put_u8(skb, IFLA_TUN_TYPE, tun->flags & TUN_TYPE_MASK))
2258 goto nla_put_failure;
2259 if (uid_valid(tun->owner) &&
2260 nla_put_u32(skb, IFLA_TUN_OWNER,
2261 from_kuid_munged(current_user_ns(), tun->owner)))
2262 goto nla_put_failure;
2263 if (gid_valid(tun->group) &&
2264 nla_put_u32(skb, IFLA_TUN_GROUP,
2265 from_kgid_munged(current_user_ns(), tun->group)))
2266 goto nla_put_failure;
2267 if (nla_put_u8(skb, IFLA_TUN_PI, !(tun->flags & IFF_NO_PI)))
2268 goto nla_put_failure;
2269 if (nla_put_u8(skb, IFLA_TUN_VNET_HDR, !!(tun->flags & IFF_VNET_HDR)))
2270 goto nla_put_failure;
2271 if (nla_put_u8(skb, IFLA_TUN_PERSIST, !!(tun->flags & IFF_PERSIST)))
2272 goto nla_put_failure;
2273 if (nla_put_u8(skb, IFLA_TUN_MULTI_QUEUE,
2274 !!(tun->flags & IFF_MULTI_QUEUE)))
2275 goto nla_put_failure;
2276 if (tun->flags & IFF_MULTI_QUEUE) {
2277 if (nla_put_u32(skb, IFLA_TUN_NUM_QUEUES, tun->numqueues))
2278 goto nla_put_failure;
2279 if (nla_put_u32(skb, IFLA_TUN_NUM_DISABLED_QUEUES,
2280 tun->numdisabled))
2281 goto nla_put_failure;
2282 }
2283
2284 return 0;
2285
2286 nla_put_failure:
2287 return -EMSGSIZE;
2288 }
2289
2290 static struct rtnl_link_ops tun_link_ops __read_mostly = {
2291 .kind = DRV_NAME,
2292 .priv_size = sizeof(struct tun_struct),
2293 .setup = tun_setup,
2294 .validate = tun_validate,
2295 .get_size = tun_get_size,
2296 .fill_info = tun_fill_info,
2297 };
2298
2299 static void tun_sock_write_space(struct sock *sk)
2300 {
2301 struct tun_file *tfile;
2302 wait_queue_head_t *wqueue;
2303
2304 if (!sock_writeable(sk))
2305 return;
2306
2307 if (!test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags))
2308 return;
2309
2310 wqueue = sk_sleep(sk);
2311 if (wqueue && waitqueue_active(wqueue))
2312 wake_up_interruptible_sync_poll(wqueue, EPOLLOUT |
2313 EPOLLWRNORM | EPOLLWRBAND);
2314
2315 tfile = container_of(sk, struct tun_file, sk);
2316 kill_fasync(&tfile->fasync, SIGIO, POLL_OUT);
2317 }
2318
2319 static void tun_put_page(struct tun_page *tpage)
2320 {
2321 if (tpage->page)
2322 __page_frag_cache_drain(tpage->page, tpage->count);
2323 }
2324
2325 static int tun_xdp_one(struct tun_struct *tun,
2326 struct tun_file *tfile,
2327 struct xdp_buff *xdp, int *flush,
2328 struct tun_page *tpage)
2329 {
2330 unsigned int datasize = xdp->data_end - xdp->data;
2331 struct tun_xdp_hdr *hdr = xdp->data_hard_start;
2332 struct virtio_net_hdr *gso = &hdr->gso;
2333 struct bpf_prog *xdp_prog;
2334 struct sk_buff *skb = NULL;
2335 u32 rxhash = 0, act;
2336 int buflen = hdr->buflen;
2337 int err = 0;
2338 bool skb_xdp = false;
2339 struct page *page;
2340
2341 xdp_prog = rcu_dereference(tun->xdp_prog);
2342 if (xdp_prog) {
2343 if (gso->gso_type) {
2344 skb_xdp = true;
2345 goto build;
2346 }
2347 xdp_set_data_meta_invalid(xdp);
2348 xdp->rxq = &tfile->xdp_rxq;
2349 xdp->frame_sz = buflen;
2350
2351 act = bpf_prog_run_xdp(xdp_prog, xdp);
2352 err = tun_xdp_act(tun, xdp_prog, xdp, act);
2353 if (err < 0) {
2354 put_page(virt_to_head_page(xdp->data));
2355 return err;
2356 }
2357
2358 switch (err) {
2359 case XDP_REDIRECT:
2360 *flush = true;
2361 fallthrough;
2362 case XDP_TX:
2363 return 0;
2364 case XDP_PASS:
2365 break;
2366 default:
2367 page = virt_to_head_page(xdp->data);
2368 if (tpage->page == page) {
2369 ++tpage->count;
2370 } else {
2371 tun_put_page(tpage);
2372 tpage->page = page;
2373 tpage->count = 1;
2374 }
2375 return 0;
2376 }
2377 }
2378
2379 build:
2380 skb = build_skb(xdp->data_hard_start, buflen);
2381 if (!skb) {
2382 err = -ENOMEM;
2383 goto out;
2384 }
2385
2386 skb_reserve(skb, xdp->data - xdp->data_hard_start);
2387 skb_put(skb, xdp->data_end - xdp->data);
2388
2389 if (virtio_net_hdr_to_skb(skb, gso, tun_is_little_endian(tun))) {
2390 atomic_long_inc(&tun->rx_frame_errors);
2391 kfree_skb(skb);
2392 err = -EINVAL;
2393 goto out;
2394 }
2395
2396 skb->protocol = eth_type_trans(skb, tun->dev);
2397 skb_reset_network_header(skb);
2398 skb_probe_transport_header(skb);
2399 skb_record_rx_queue(skb, tfile->queue_index);
2400
2401 if (skb_xdp) {
2402 err = do_xdp_generic(xdp_prog, skb);
2403 if (err != XDP_PASS)
2404 goto out;
2405 }
2406
2407 if (!rcu_dereference(tun->steering_prog) && tun->numqueues > 1 &&
2408 !tfile->detached)
2409 rxhash = __skb_get_hash_symmetric(skb);
2410
2411 netif_receive_skb(skb);
2412
2413 /* No need to disable preemption here since this function is
2414 * always called with bh disabled
2415 */
2416 dev_sw_netstats_rx_add(tun->dev, datasize);
2417
2418 if (rxhash)
2419 tun_flow_update(tun, rxhash, tfile);
2420
2421 out:
2422 return err;
2423 }
2424
2425 static int tun_sendmsg(struct socket *sock, struct msghdr *m, size_t total_len)
2426 {
2427 int ret, i;
2428 struct tun_file *tfile = container_of(sock, struct tun_file, socket);
2429 struct tun_struct *tun = tun_get(tfile);
2430 struct tun_msg_ctl *ctl = m->msg_control;
2431 struct xdp_buff *xdp;
2432
2433 if (!tun)
2434 return -EBADFD;
2435
2436 if (ctl && (ctl->type == TUN_MSG_PTR)) {
2437 struct tun_page tpage;
2438 int n = ctl->num;
2439 int flush = 0;
2440
2441 memset(&tpage, 0, sizeof(tpage));
2442
2443 local_bh_disable();
2444 rcu_read_lock();
2445
2446 for (i = 0; i < n; i++) {
2447 xdp = &((struct xdp_buff *)ctl->ptr)[i];
2448 tun_xdp_one(tun, tfile, xdp, &flush, &tpage);
2449 }
2450
2451 if (flush)
2452 xdp_do_flush();
2453
2454 rcu_read_unlock();
2455 local_bh_enable();
2456
2457 tun_put_page(&tpage);
2458
2459 ret = total_len;
2460 goto out;
2461 }
2462
2463 ret = tun_get_user(tun, tfile, ctl ? ctl->ptr : NULL, &m->msg_iter,
2464 m->msg_flags & MSG_DONTWAIT,
2465 m->msg_flags & MSG_MORE);
2466 out:
2467 tun_put(tun);
2468 return ret;
2469 }
2470
2471 static int tun_recvmsg(struct socket *sock, struct msghdr *m, size_t total_len,
2472 int flags)
2473 {
2474 struct tun_file *tfile = container_of(sock, struct tun_file, socket);
2475 struct tun_struct *tun = tun_get(tfile);
2476 void *ptr = m->msg_control;
2477 int ret;
2478
2479 if (!tun) {
2480 ret = -EBADFD;
2481 goto out_free;
2482 }
2483
2484 if (flags & ~(MSG_DONTWAIT|MSG_TRUNC|MSG_ERRQUEUE)) {
2485 ret = -EINVAL;
2486 goto out_put_tun;
2487 }
2488 if (flags & MSG_ERRQUEUE) {
2489 ret = sock_recv_errqueue(sock->sk, m, total_len,
2490 SOL_PACKET, TUN_TX_TIMESTAMP);
2491 goto out;
2492 }
2493 ret = tun_do_read(tun, tfile, &m->msg_iter, flags & MSG_DONTWAIT, ptr);
2494 if (ret > (ssize_t)total_len) {
2495 m->msg_flags |= MSG_TRUNC;
2496 ret = flags & MSG_TRUNC ? ret : total_len;
2497 }
2498 out:
2499 tun_put(tun);
2500 return ret;
2501
2502 out_put_tun:
2503 tun_put(tun);
2504 out_free:
2505 tun_ptr_free(ptr);
2506 return ret;
2507 }
2508
2509 static int tun_ptr_peek_len(void *ptr)
2510 {
2511 if (likely(ptr)) {
2512 if (tun_is_xdp_frame(ptr)) {
2513 struct xdp_frame *xdpf = tun_ptr_to_xdp(ptr);
2514
2515 return xdpf->len;
2516 }
2517 return __skb_array_len_with_tag(ptr);
2518 } else {
2519 return 0;
2520 }
2521 }
2522
2523 static int tun_peek_len(struct socket *sock)
2524 {
2525 struct tun_file *tfile = container_of(sock, struct tun_file, socket);
2526 struct tun_struct *tun;
2527 int ret = 0;
2528
2529 tun = tun_get(tfile);
2530 if (!tun)
2531 return 0;
2532
2533 ret = PTR_RING_PEEK_CALL(&tfile->tx_ring, tun_ptr_peek_len);
2534 tun_put(tun);
2535
2536 return ret;
2537 }
2538
2539 /* Ops structure to mimic raw sockets with tun */
2540 static const struct proto_ops tun_socket_ops = {
2541 .peek_len = tun_peek_len,
2542 .sendmsg = tun_sendmsg,
2543 .recvmsg = tun_recvmsg,
2544 };
2545
2546 static struct proto tun_proto = {
2547 .name = "tun",
2548 .owner = THIS_MODULE,
2549 .obj_size = sizeof(struct tun_file),
2550 };
2551
2552 static int tun_flags(struct tun_struct *tun)
2553 {
2554 return tun->flags & (TUN_FEATURES | IFF_PERSIST | IFF_TUN | IFF_TAP);
2555 }
2556
2557 static ssize_t tun_show_flags(struct device *dev, struct device_attribute *attr,
2558 char *buf)
2559 {
2560 struct tun_struct *tun = netdev_priv(to_net_dev(dev));
2561 return sprintf(buf, "0x%x\n", tun_flags(tun));
2562 }
2563
2564 static ssize_t tun_show_owner(struct device *dev, struct device_attribute *attr,
2565 char *buf)
2566 {
2567 struct tun_struct *tun = netdev_priv(to_net_dev(dev));
2568 return uid_valid(tun->owner)?
2569 sprintf(buf, "%u\n",
2570 from_kuid_munged(current_user_ns(), tun->owner)):
2571 sprintf(buf, "-1\n");
2572 }
2573
2574 static ssize_t tun_show_group(struct device *dev, struct device_attribute *attr,
2575 char *buf)
2576 {
2577 struct tun_struct *tun = netdev_priv(to_net_dev(dev));
2578 return gid_valid(tun->group) ?
2579 sprintf(buf, "%u\n",
2580 from_kgid_munged(current_user_ns(), tun->group)):
2581 sprintf(buf, "-1\n");
2582 }
2583
2584 static DEVICE_ATTR(tun_flags, 0444, tun_show_flags, NULL);
2585 static DEVICE_ATTR(owner, 0444, tun_show_owner, NULL);
2586 static DEVICE_ATTR(group, 0444, tun_show_group, NULL);
2587
2588 static struct attribute *tun_dev_attrs[] = {
2589 &dev_attr_tun_flags.attr,
2590 &dev_attr_owner.attr,
2591 &dev_attr_group.attr,
2592 NULL
2593 };
2594
2595 static const struct attribute_group tun_attr_group = {
2596 .attrs = tun_dev_attrs
2597 };
2598
2599 static int tun_set_iff(struct net *net, struct file *file, struct ifreq *ifr)
2600 {
2601 struct tun_struct *tun;
2602 struct tun_file *tfile = file->private_data;
2603 struct net_device *dev;
2604 int err;
2605
2606 if (tfile->detached)
2607 return -EINVAL;
2608
2609 if ((ifr->ifr_flags & IFF_NAPI_FRAGS)) {
2610 if (!capable(CAP_NET_ADMIN))
2611 return -EPERM;
2612
2613 if (!(ifr->ifr_flags & IFF_NAPI) ||
2614 (ifr->ifr_flags & TUN_TYPE_MASK) != IFF_TAP)
2615 return -EINVAL;
2616 }
2617
2618 dev = __dev_get_by_name(net, ifr->ifr_name);
2619 if (dev) {
2620 if (ifr->ifr_flags & IFF_TUN_EXCL)
2621 return -EBUSY;
2622 if ((ifr->ifr_flags & IFF_TUN) && dev->netdev_ops == &tun_netdev_ops)
2623 tun = netdev_priv(dev);
2624 else if ((ifr->ifr_flags & IFF_TAP) && dev->netdev_ops == &tap_netdev_ops)
2625 tun = netdev_priv(dev);
2626 else
2627 return -EINVAL;
2628
2629 if (!!(ifr->ifr_flags & IFF_MULTI_QUEUE) !=
2630 !!(tun->flags & IFF_MULTI_QUEUE))
2631 return -EINVAL;
2632
2633 if (tun_not_capable(tun))
2634 return -EPERM;
2635 err = security_tun_dev_open(tun->security);
2636 if (err < 0)
2637 return err;
2638
2639 err = tun_attach(tun, file, ifr->ifr_flags & IFF_NOFILTER,
2640 ifr->ifr_flags & IFF_NAPI,
2641 ifr->ifr_flags & IFF_NAPI_FRAGS, true);
2642 if (err < 0)
2643 return err;
2644
2645 if (tun->flags & IFF_MULTI_QUEUE &&
2646 (tun->numqueues + tun->numdisabled > 1)) {
2647 /* One or more queue has already been attached, no need
2648 * to initialize the device again.
2649 */
2650 netdev_state_change(dev);
2651 return 0;
2652 }
2653
2654 tun->flags = (tun->flags & ~TUN_FEATURES) |
2655 (ifr->ifr_flags & TUN_FEATURES);
2656
2657 netdev_state_change(dev);
2658 } else {
2659 char *name;
2660 unsigned long flags = 0;
2661 int queues = ifr->ifr_flags & IFF_MULTI_QUEUE ?
2662 MAX_TAP_QUEUES : 1;
2663
2664 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
2665 return -EPERM;
2666 err = security_tun_dev_create();
2667 if (err < 0)
2668 return err;
2669
2670 /* Set dev type */
2671 if (ifr->ifr_flags & IFF_TUN) {
2672 /* TUN device */
2673 flags |= IFF_TUN;
2674 name = "tun%d";
2675 } else if (ifr->ifr_flags & IFF_TAP) {
2676 /* TAP device */
2677 flags |= IFF_TAP;
2678 name = "tap%d";
2679 } else
2680 return -EINVAL;
2681
2682 if (*ifr->ifr_name)
2683 name = ifr->ifr_name;
2684
2685 dev = alloc_netdev_mqs(sizeof(struct tun_struct), name,
2686 NET_NAME_UNKNOWN, tun_setup, queues,
2687 queues);
2688
2689 if (!dev)
2690 return -ENOMEM;
2691
2692 dev_net_set(dev, net);
2693 dev->rtnl_link_ops = &tun_link_ops;
2694 dev->ifindex = tfile->ifindex;
2695 dev->sysfs_groups[0] = &tun_attr_group;
2696
2697 tun = netdev_priv(dev);
2698 tun->dev = dev;
2699 tun->flags = flags;
2700 tun->txflt.count = 0;
2701 tun->vnet_hdr_sz = sizeof(struct virtio_net_hdr);
2702
2703 tun->align = NET_SKB_PAD;
2704 tun->filter_attached = false;
2705 tun->sndbuf = tfile->socket.sk->sk_sndbuf;
2706 tun->rx_batched = 0;
2707 RCU_INIT_POINTER(tun->steering_prog, NULL);
2708
2709 dev->tstats = netdev_alloc_pcpu_stats(struct pcpu_sw_netstats);
2710 if (!dev->tstats) {
2711 err = -ENOMEM;
2712 goto err_free_dev;
2713 }
2714
2715 spin_lock_init(&tun->lock);
2716
2717 err = security_tun_dev_alloc_security(&tun->security);
2718 if (err < 0)
2719 goto err_free_stat;
2720
2721 tun_net_init(dev);
2722 tun_flow_init(tun);
2723
2724 dev->hw_features = NETIF_F_SG | NETIF_F_FRAGLIST |
2725 TUN_USER_FEATURES | NETIF_F_HW_VLAN_CTAG_TX |
2726 NETIF_F_HW_VLAN_STAG_TX;
2727 dev->features = dev->hw_features | NETIF_F_LLTX;
2728 dev->vlan_features = dev->features &
2729 ~(NETIF_F_HW_VLAN_CTAG_TX |
2730 NETIF_F_HW_VLAN_STAG_TX);
2731
2732 tun->flags = (tun->flags & ~TUN_FEATURES) |
2733 (ifr->ifr_flags & TUN_FEATURES);
2734
2735 INIT_LIST_HEAD(&tun->disabled);
2736 err = tun_attach(tun, file, false, ifr->ifr_flags & IFF_NAPI,
2737 ifr->ifr_flags & IFF_NAPI_FRAGS, false);
2738 if (err < 0)
2739 goto err_free_flow;
2740
2741 err = register_netdevice(tun->dev);
2742 if (err < 0)
2743 goto err_detach;
2744 /* free_netdev() won't check refcnt, to aovid race
2745 * with dev_put() we need publish tun after registration.
2746 */
2747 rcu_assign_pointer(tfile->tun, tun);
2748 }
2749
2750 netif_carrier_on(tun->dev);
2751
2752 /* Make sure persistent devices do not get stuck in
2753 * xoff state.
2754 */
2755 if (netif_running(tun->dev))
2756 netif_tx_wake_all_queues(tun->dev);
2757
2758 strcpy(ifr->ifr_name, tun->dev->name);
2759 return 0;
2760
2761 err_detach:
2762 tun_detach_all(dev);
2763 /* We are here because register_netdevice() has failed.
2764 * If register_netdevice() already called tun_free_netdev()
2765 * while dealing with the error, dev->stats has been cleared.
2766 */
2767 if (!dev->tstats)
2768 goto err_free_dev;
2769
2770 err_free_flow:
2771 tun_flow_uninit(tun);
2772 security_tun_dev_free_security(tun->security);
2773 err_free_stat:
2774 free_percpu(dev->tstats);
2775 err_free_dev:
2776 free_netdev(dev);
2777 return err;
2778 }
2779
2780 static void tun_get_iff(struct tun_struct *tun, struct ifreq *ifr)
2781 {
2782 strcpy(ifr->ifr_name, tun->dev->name);
2783
2784 ifr->ifr_flags = tun_flags(tun);
2785
2786 }
2787
2788 /* This is like a cut-down ethtool ops, except done via tun fd so no
2789 * privs required. */
2790 static int set_offload(struct tun_struct *tun, unsigned long arg)
2791 {
2792 netdev_features_t features = 0;
2793
2794 if (arg & TUN_F_CSUM) {
2795 features |= NETIF_F_HW_CSUM;
2796 arg &= ~TUN_F_CSUM;
2797
2798 if (arg & (TUN_F_TSO4|TUN_F_TSO6)) {
2799 if (arg & TUN_F_TSO_ECN) {
2800 features |= NETIF_F_TSO_ECN;
2801 arg &= ~TUN_F_TSO_ECN;
2802 }
2803 if (arg & TUN_F_TSO4)
2804 features |= NETIF_F_TSO;
2805 if (arg & TUN_F_TSO6)
2806 features |= NETIF_F_TSO6;
2807 arg &= ~(TUN_F_TSO4|TUN_F_TSO6);
2808 }
2809
2810 arg &= ~TUN_F_UFO;
2811 }
2812
2813 /* This gives the user a way to test for new features in future by
2814 * trying to set them. */
2815 if (arg)
2816 return -EINVAL;
2817
2818 tun->set_features = features;
2819 tun->dev->wanted_features &= ~TUN_USER_FEATURES;
2820 tun->dev->wanted_features |= features;
2821 netdev_update_features(tun->dev);
2822
2823 return 0;
2824 }
2825
2826 static void tun_detach_filter(struct tun_struct *tun, int n)
2827 {
2828 int i;
2829 struct tun_file *tfile;
2830
2831 for (i = 0; i < n; i++) {
2832 tfile = rtnl_dereference(tun->tfiles[i]);
2833 lock_sock(tfile->socket.sk);
2834 sk_detach_filter(tfile->socket.sk);
2835 release_sock(tfile->socket.sk);
2836 }
2837
2838 tun->filter_attached = false;
2839 }
2840
2841 static int tun_attach_filter(struct tun_struct *tun)
2842 {
2843 int i, ret = 0;
2844 struct tun_file *tfile;
2845
2846 for (i = 0; i < tun->numqueues; i++) {
2847 tfile = rtnl_dereference(tun->tfiles[i]);
2848 lock_sock(tfile->socket.sk);
2849 ret = sk_attach_filter(&tun->fprog, tfile->socket.sk);
2850 release_sock(tfile->socket.sk);
2851 if (ret) {
2852 tun_detach_filter(tun, i);
2853 return ret;
2854 }
2855 }
2856
2857 tun->filter_attached = true;
2858 return ret;
2859 }
2860
2861 static void tun_set_sndbuf(struct tun_struct *tun)
2862 {
2863 struct tun_file *tfile;
2864 int i;
2865
2866 for (i = 0; i < tun->numqueues; i++) {
2867 tfile = rtnl_dereference(tun->tfiles[i]);
2868 tfile->socket.sk->sk_sndbuf = tun->sndbuf;
2869 }
2870 }
2871
2872 static int tun_set_queue(struct file *file, struct ifreq *ifr)
2873 {
2874 struct tun_file *tfile = file->private_data;
2875 struct tun_struct *tun;
2876 int ret = 0;
2877
2878 rtnl_lock();
2879
2880 if (ifr->ifr_flags & IFF_ATTACH_QUEUE) {
2881 tun = tfile->detached;
2882 if (!tun) {
2883 ret = -EINVAL;
2884 goto unlock;
2885 }
2886 ret = security_tun_dev_attach_queue(tun->security);
2887 if (ret < 0)
2888 goto unlock;
2889 ret = tun_attach(tun, file, false, tun->flags & IFF_NAPI,
2890 tun->flags & IFF_NAPI_FRAGS, true);
2891 } else if (ifr->ifr_flags & IFF_DETACH_QUEUE) {
2892 tun = rtnl_dereference(tfile->tun);
2893 if (!tun || !(tun->flags & IFF_MULTI_QUEUE) || tfile->detached)
2894 ret = -EINVAL;
2895 else
2896 __tun_detach(tfile, false);
2897 } else
2898 ret = -EINVAL;
2899
2900 if (ret >= 0)
2901 netdev_state_change(tun->dev);
2902
2903 unlock:
2904 rtnl_unlock();
2905 return ret;
2906 }
2907
2908 static int tun_set_ebpf(struct tun_struct *tun, struct tun_prog __rcu **prog_p,
2909 void __user *data)
2910 {
2911 struct bpf_prog *prog;
2912 int fd;
2913
2914 if (copy_from_user(&fd, data, sizeof(fd)))
2915 return -EFAULT;
2916
2917 if (fd == -1) {
2918 prog = NULL;
2919 } else {
2920 prog = bpf_prog_get_type(fd, BPF_PROG_TYPE_SOCKET_FILTER);
2921 if (IS_ERR(prog))
2922 return PTR_ERR(prog);
2923 }
2924
2925 return __tun_set_ebpf(tun, prog_p, prog);
2926 }
2927
2928 static long __tun_chr_ioctl(struct file *file, unsigned int cmd,
2929 unsigned long arg, int ifreq_len)
2930 {
2931 struct tun_file *tfile = file->private_data;
2932 struct net *net = sock_net(&tfile->sk);
2933 struct tun_struct *tun;
2934 void __user* argp = (void __user*)arg;
2935 unsigned int ifindex, carrier;
2936 struct ifreq ifr;
2937 kuid_t owner;
2938 kgid_t group;
2939 int sndbuf;
2940 int vnet_hdr_sz;
2941 int le;
2942 int ret;
2943 bool do_notify = false;
2944
2945 if (cmd == TUNSETIFF || cmd == TUNSETQUEUE ||
2946 (_IOC_TYPE(cmd) == SOCK_IOC_TYPE && cmd != SIOCGSKNS)) {
2947 if (copy_from_user(&ifr, argp, ifreq_len))
2948 return -EFAULT;
2949 } else {
2950 memset(&ifr, 0, sizeof(ifr));
2951 }
2952 if (cmd == TUNGETFEATURES) {
2953 /* Currently this just means: "what IFF flags are valid?".
2954 * This is needed because we never checked for invalid flags on
2955 * TUNSETIFF.
2956 */
2957 return put_user(IFF_TUN | IFF_TAP | TUN_FEATURES,
2958 (unsigned int __user*)argp);
2959 } else if (cmd == TUNSETQUEUE) {
2960 return tun_set_queue(file, &ifr);
2961 } else if (cmd == SIOCGSKNS) {
2962 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
2963 return -EPERM;
2964 return open_related_ns(&net->ns, get_net_ns);
2965 }
2966
2967 ret = 0;
2968 rtnl_lock();
2969
2970 tun = tun_get(tfile);
2971 if (cmd == TUNSETIFF) {
2972 ret = -EEXIST;
2973 if (tun)
2974 goto unlock;
2975
2976 ifr.ifr_name[IFNAMSIZ-1] = '\0';
2977
2978 ret = tun_set_iff(net, file, &ifr);
2979
2980 if (ret)
2981 goto unlock;
2982
2983 if (copy_to_user(argp, &ifr, ifreq_len))
2984 ret = -EFAULT;
2985 goto unlock;
2986 }
2987 if (cmd == TUNSETIFINDEX) {
2988 ret = -EPERM;
2989 if (tun)
2990 goto unlock;
2991
2992 ret = -EFAULT;
2993 if (copy_from_user(&ifindex, argp, sizeof(ifindex)))
2994 goto unlock;
2995
2996 ret = 0;
2997 tfile->ifindex = ifindex;
2998 goto unlock;
2999 }
3000
3001 ret = -EBADFD;
3002 if (!tun)
3003 goto unlock;
3004
3005 netif_info(tun, drv, tun->dev, "tun_chr_ioctl cmd %u\n", cmd);
3006
3007 net = dev_net(tun->dev);
3008 ret = 0;
3009 switch (cmd) {
3010 case TUNGETIFF:
3011 tun_get_iff(tun, &ifr);
3012
3013 if (tfile->detached)
3014 ifr.ifr_flags |= IFF_DETACH_QUEUE;
3015 if (!tfile->socket.sk->sk_filter)
3016 ifr.ifr_flags |= IFF_NOFILTER;
3017
3018 if (copy_to_user(argp, &ifr, ifreq_len))
3019 ret = -EFAULT;
3020 break;
3021
3022 case TUNSETNOCSUM:
3023 /* Disable/Enable checksum */
3024
3025 /* [unimplemented] */
3026 netif_info(tun, drv, tun->dev, "ignored: set checksum %s\n",
3027 arg ? "disabled" : "enabled");
3028 break;
3029
3030 case TUNSETPERSIST:
3031 /* Disable/Enable persist mode. Keep an extra reference to the
3032 * module to prevent the module being unprobed.
3033 */
3034 if (arg && !(tun->flags & IFF_PERSIST)) {
3035 tun->flags |= IFF_PERSIST;
3036 __module_get(THIS_MODULE);
3037 do_notify = true;
3038 }
3039 if (!arg && (tun->flags & IFF_PERSIST)) {
3040 tun->flags &= ~IFF_PERSIST;
3041 module_put(THIS_MODULE);
3042 do_notify = true;
3043 }
3044
3045 netif_info(tun, drv, tun->dev, "persist %s\n",
3046 arg ? "enabled" : "disabled");
3047 break;
3048
3049 case TUNSETOWNER:
3050 /* Set owner of the device */
3051 owner = make_kuid(current_user_ns(), arg);
3052 if (!uid_valid(owner)) {
3053 ret = -EINVAL;
3054 break;
3055 }
3056 tun->owner = owner;
3057 do_notify = true;
3058 netif_info(tun, drv, tun->dev, "owner set to %u\n",
3059 from_kuid(&init_user_ns, tun->owner));
3060 break;
3061
3062 case TUNSETGROUP:
3063 /* Set group of the device */
3064 group = make_kgid(current_user_ns(), arg);
3065 if (!gid_valid(group)) {
3066 ret = -EINVAL;
3067 break;
3068 }
3069 tun->group = group;
3070 do_notify = true;
3071 netif_info(tun, drv, tun->dev, "group set to %u\n",
3072 from_kgid(&init_user_ns, tun->group));
3073 break;
3074
3075 case TUNSETLINK:
3076 /* Only allow setting the type when the interface is down */
3077 if (tun->dev->flags & IFF_UP) {
3078 netif_info(tun, drv, tun->dev,
3079 "Linktype set failed because interface is up\n");
3080 ret = -EBUSY;
3081 } else {
3082 ret = call_netdevice_notifiers(NETDEV_PRE_TYPE_CHANGE,
3083 tun->dev);
3084 ret = notifier_to_errno(ret);
3085 if (ret) {
3086 netif_info(tun, drv, tun->dev,
3087 "Refused to change device type\n");
3088 break;
3089 }
3090 tun->dev->type = (int) arg;
3091 netif_info(tun, drv, tun->dev, "linktype set to %d\n",
3092 tun->dev->type);
3093 call_netdevice_notifiers(NETDEV_POST_TYPE_CHANGE,
3094 tun->dev);
3095 }
3096 break;
3097
3098 case TUNSETDEBUG:
3099 tun->msg_enable = (u32)arg;
3100 break;
3101
3102 case TUNSETOFFLOAD:
3103 ret = set_offload(tun, arg);
3104 break;
3105
3106 case TUNSETTXFILTER:
3107 /* Can be set only for TAPs */
3108 ret = -EINVAL;
3109 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3110 break;
3111 ret = update_filter(&tun->txflt, (void __user *)arg);
3112 break;
3113
3114 case SIOCGIFHWADDR:
3115 /* Get hw address */
3116 memcpy(ifr.ifr_hwaddr.sa_data, tun->dev->dev_addr, ETH_ALEN);
3117 ifr.ifr_hwaddr.sa_family = tun->dev->type;
3118 if (copy_to_user(argp, &ifr, ifreq_len))
3119 ret = -EFAULT;
3120 break;
3121
3122 case SIOCSIFHWADDR:
3123 /* Set hw address */
3124 ret = dev_set_mac_address(tun->dev, &ifr.ifr_hwaddr, NULL);
3125 break;
3126
3127 case TUNGETSNDBUF:
3128 sndbuf = tfile->socket.sk->sk_sndbuf;
3129 if (copy_to_user(argp, &sndbuf, sizeof(sndbuf)))
3130 ret = -EFAULT;
3131 break;
3132
3133 case TUNSETSNDBUF:
3134 if (copy_from_user(&sndbuf, argp, sizeof(sndbuf))) {
3135 ret = -EFAULT;
3136 break;
3137 }
3138 if (sndbuf <= 0) {
3139 ret = -EINVAL;
3140 break;
3141 }
3142
3143 tun->sndbuf = sndbuf;
3144 tun_set_sndbuf(tun);
3145 break;
3146
3147 case TUNGETVNETHDRSZ:
3148 vnet_hdr_sz = tun->vnet_hdr_sz;
3149 if (copy_to_user(argp, &vnet_hdr_sz, sizeof(vnet_hdr_sz)))
3150 ret = -EFAULT;
3151 break;
3152
3153 case TUNSETVNETHDRSZ:
3154 if (copy_from_user(&vnet_hdr_sz, argp, sizeof(vnet_hdr_sz))) {
3155 ret = -EFAULT;
3156 break;
3157 }
3158 if (vnet_hdr_sz < (int)sizeof(struct virtio_net_hdr)) {
3159 ret = -EINVAL;
3160 break;
3161 }
3162
3163 tun->vnet_hdr_sz = vnet_hdr_sz;
3164 break;
3165
3166 case TUNGETVNETLE:
3167 le = !!(tun->flags & TUN_VNET_LE);
3168 if (put_user(le, (int __user *)argp))
3169 ret = -EFAULT;
3170 break;
3171
3172 case TUNSETVNETLE:
3173 if (get_user(le, (int __user *)argp)) {
3174 ret = -EFAULT;
3175 break;
3176 }
3177 if (le)
3178 tun->flags |= TUN_VNET_LE;
3179 else
3180 tun->flags &= ~TUN_VNET_LE;
3181 break;
3182
3183 case TUNGETVNETBE:
3184 ret = tun_get_vnet_be(tun, argp);
3185 break;
3186
3187 case TUNSETVNETBE:
3188 ret = tun_set_vnet_be(tun, argp);
3189 break;
3190
3191 case TUNATTACHFILTER:
3192 /* Can be set only for TAPs */
3193 ret = -EINVAL;
3194 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3195 break;
3196 ret = -EFAULT;
3197 if (copy_from_user(&tun->fprog, argp, sizeof(tun->fprog)))
3198 break;
3199
3200 ret = tun_attach_filter(tun);
3201 break;
3202
3203 case TUNDETACHFILTER:
3204 /* Can be set only for TAPs */
3205 ret = -EINVAL;
3206 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3207 break;
3208 ret = 0;
3209 tun_detach_filter(tun, tun->numqueues);
3210 break;
3211
3212 case TUNGETFILTER:
3213 ret = -EINVAL;
3214 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3215 break;
3216 ret = -EFAULT;
3217 if (copy_to_user(argp, &tun->fprog, sizeof(tun->fprog)))
3218 break;
3219 ret = 0;
3220 break;
3221
3222 case TUNSETSTEERINGEBPF:
3223 ret = tun_set_ebpf(tun, &tun->steering_prog, argp);
3224 break;
3225
3226 case TUNSETFILTEREBPF:
3227 ret = tun_set_ebpf(tun, &tun->filter_prog, argp);
3228 break;
3229
3230 case TUNSETCARRIER:
3231 ret = -EFAULT;
3232 if (copy_from_user(&carrier, argp, sizeof(carrier)))
3233 goto unlock;
3234
3235 ret = tun_net_change_carrier(tun->dev, (bool)carrier);
3236 break;
3237
3238 case TUNGETDEVNETNS:
3239 ret = -EPERM;
3240 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
3241 goto unlock;
3242 ret = open_related_ns(&net->ns, get_net_ns);
3243 break;
3244
3245 default:
3246 ret = -EINVAL;
3247 break;
3248 }
3249
3250 if (do_notify)
3251 netdev_state_change(tun->dev);
3252
3253 unlock:
3254 rtnl_unlock();
3255 if (tun)
3256 tun_put(tun);
3257 return ret;
3258 }
3259
3260 static long tun_chr_ioctl(struct file *file,
3261 unsigned int cmd, unsigned long arg)
3262 {
3263 return __tun_chr_ioctl(file, cmd, arg, sizeof (struct ifreq));
3264 }
3265
3266 #ifdef CONFIG_COMPAT
3267 static long tun_chr_compat_ioctl(struct file *file,
3268 unsigned int cmd, unsigned long arg)
3269 {
3270 switch (cmd) {
3271 case TUNSETIFF:
3272 case TUNGETIFF:
3273 case TUNSETTXFILTER:
3274 case TUNGETSNDBUF:
3275 case TUNSETSNDBUF:
3276 case SIOCGIFHWADDR:
3277 case SIOCSIFHWADDR:
3278 arg = (unsigned long)compat_ptr(arg);
3279 break;
3280 default:
3281 arg = (compat_ulong_t)arg;
3282 break;
3283 }
3284
3285 /*
3286 * compat_ifreq is shorter than ifreq, so we must not access beyond
3287 * the end of that structure. All fields that are used in this
3288 * driver are compatible though, we don't need to convert the
3289 * contents.
3290 */
3291 return __tun_chr_ioctl(file, cmd, arg, sizeof(struct compat_ifreq));
3292 }
3293 #endif /* CONFIG_COMPAT */
3294
3295 static int tun_chr_fasync(int fd, struct file *file, int on)
3296 {
3297 struct tun_file *tfile = file->private_data;
3298 int ret;
3299
3300 if ((ret = fasync_helper(fd, file, on, &tfile->fasync)) < 0)
3301 goto out;
3302
3303 if (on) {
3304 __f_setown(file, task_pid(current), PIDTYPE_TGID, 0);
3305 tfile->flags |= TUN_FASYNC;
3306 } else
3307 tfile->flags &= ~TUN_FASYNC;
3308 ret = 0;
3309 out:
3310 return ret;
3311 }
3312
3313 static int tun_chr_open(struct inode *inode, struct file * file)
3314 {
3315 struct net *net = current->nsproxy->net_ns;
3316 struct tun_file *tfile;
3317
3318 tfile = (struct tun_file *)sk_alloc(net, AF_UNSPEC, GFP_KERNEL,
3319 &tun_proto, 0);
3320 if (!tfile)
3321 return -ENOMEM;
3322 if (ptr_ring_init(&tfile->tx_ring, 0, GFP_KERNEL)) {
3323 sk_free(&tfile->sk);
3324 return -ENOMEM;
3325 }
3326
3327 mutex_init(&tfile->napi_mutex);
3328 RCU_INIT_POINTER(tfile->tun, NULL);
3329 tfile->flags = 0;
3330 tfile->ifindex = 0;
3331
3332 init_waitqueue_head(&tfile->socket.wq.wait);
3333
3334 tfile->socket.file = file;
3335 tfile->socket.ops = &tun_socket_ops;
3336
3337 sock_init_data(&tfile->socket, &tfile->sk);
3338
3339 tfile->sk.sk_write_space = tun_sock_write_space;
3340 tfile->sk.sk_sndbuf = INT_MAX;
3341
3342 file->private_data = tfile;
3343 INIT_LIST_HEAD(&tfile->next);
3344
3345 sock_set_flag(&tfile->sk, SOCK_ZEROCOPY);
3346
3347 return 0;
3348 }
3349
3350 static int tun_chr_close(struct inode *inode, struct file *file)
3351 {
3352 struct tun_file *tfile = file->private_data;
3353
3354 tun_detach(tfile, true);
3355
3356 return 0;
3357 }
3358
3359 #ifdef CONFIG_PROC_FS
3360 static void tun_chr_show_fdinfo(struct seq_file *m, struct file *file)
3361 {
3362 struct tun_file *tfile = file->private_data;
3363 struct tun_struct *tun;
3364 struct ifreq ifr;
3365
3366 memset(&ifr, 0, sizeof(ifr));
3367
3368 rtnl_lock();
3369 tun = tun_get(tfile);
3370 if (tun)
3371 tun_get_iff(tun, &ifr);
3372 rtnl_unlock();
3373
3374 if (tun)
3375 tun_put(tun);
3376
3377 seq_printf(m, "iff:\t%s\n", ifr.ifr_name);
3378 }
3379 #endif
3380
3381 static const struct file_operations tun_fops = {
3382 .owner = THIS_MODULE,
3383 .llseek = no_llseek,
3384 .read_iter = tun_chr_read_iter,
3385 .write_iter = tun_chr_write_iter,
3386 .poll = tun_chr_poll,
3387 .unlocked_ioctl = tun_chr_ioctl,
3388 #ifdef CONFIG_COMPAT
3389 .compat_ioctl = tun_chr_compat_ioctl,
3390 #endif
3391 .open = tun_chr_open,
3392 .release = tun_chr_close,
3393 .fasync = tun_chr_fasync,
3394 #ifdef CONFIG_PROC_FS
3395 .show_fdinfo = tun_chr_show_fdinfo,
3396 #endif
3397 };
3398
3399 static struct miscdevice tun_miscdev = {
3400 .minor = TUN_MINOR,
3401 .name = "tun",
3402 .nodename = "net/tun",
3403 .fops = &tun_fops,
3404 };
3405
3406 /* ethtool interface */
3407
3408 static void tun_default_link_ksettings(struct net_device *dev,
3409 struct ethtool_link_ksettings *cmd)
3410 {
3411 ethtool_link_ksettings_zero_link_mode(cmd, supported);
3412 ethtool_link_ksettings_zero_link_mode(cmd, advertising);
3413 cmd->base.speed = SPEED_10;
3414 cmd->base.duplex = DUPLEX_FULL;
3415 cmd->base.port = PORT_TP;
3416 cmd->base.phy_address = 0;
3417 cmd->base.autoneg = AUTONEG_DISABLE;
3418 }
3419
3420 static int tun_get_link_ksettings(struct net_device *dev,
3421 struct ethtool_link_ksettings *cmd)
3422 {
3423 struct tun_struct *tun = netdev_priv(dev);
3424
3425 memcpy(cmd, &tun->link_ksettings, sizeof(*cmd));
3426 return 0;
3427 }
3428
3429 static int tun_set_link_ksettings(struct net_device *dev,
3430 const struct ethtool_link_ksettings *cmd)
3431 {
3432 struct tun_struct *tun = netdev_priv(dev);
3433
3434 memcpy(&tun->link_ksettings, cmd, sizeof(*cmd));
3435 return 0;
3436 }
3437
3438 static void tun_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
3439 {
3440 struct tun_struct *tun = netdev_priv(dev);
3441
3442 strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
3443 strlcpy(info->version, DRV_VERSION, sizeof(info->version));
3444
3445 switch (tun->flags & TUN_TYPE_MASK) {
3446 case IFF_TUN:
3447 strlcpy(info->bus_info, "tun", sizeof(info->bus_info));
3448 break;
3449 case IFF_TAP:
3450 strlcpy(info->bus_info, "tap", sizeof(info->bus_info));
3451 break;
3452 }
3453 }
3454
3455 static u32 tun_get_msglevel(struct net_device *dev)
3456 {
3457 struct tun_struct *tun = netdev_priv(dev);
3458
3459 return tun->msg_enable;
3460 }
3461
3462 static void tun_set_msglevel(struct net_device *dev, u32 value)
3463 {
3464 struct tun_struct *tun = netdev_priv(dev);
3465
3466 tun->msg_enable = value;
3467 }
3468
3469 static int tun_get_coalesce(struct net_device *dev,
3470 struct ethtool_coalesce *ec)
3471 {
3472 struct tun_struct *tun = netdev_priv(dev);
3473
3474 ec->rx_max_coalesced_frames = tun->rx_batched;
3475
3476 return 0;
3477 }
3478
3479 static int tun_set_coalesce(struct net_device *dev,
3480 struct ethtool_coalesce *ec)
3481 {
3482 struct tun_struct *tun = netdev_priv(dev);
3483
3484 if (ec->rx_max_coalesced_frames > NAPI_POLL_WEIGHT)
3485 tun->rx_batched = NAPI_POLL_WEIGHT;
3486 else
3487 tun->rx_batched = ec->rx_max_coalesced_frames;
3488
3489 return 0;
3490 }
3491
3492 static const struct ethtool_ops tun_ethtool_ops = {
3493 .supported_coalesce_params = ETHTOOL_COALESCE_RX_MAX_FRAMES,
3494 .get_drvinfo = tun_get_drvinfo,
3495 .get_msglevel = tun_get_msglevel,
3496 .set_msglevel = tun_set_msglevel,
3497 .get_link = ethtool_op_get_link,
3498 .get_ts_info = ethtool_op_get_ts_info,
3499 .get_coalesce = tun_get_coalesce,
3500 .set_coalesce = tun_set_coalesce,
3501 .get_link_ksettings = tun_get_link_ksettings,
3502 .set_link_ksettings = tun_set_link_ksettings,
3503 };
3504
3505 static int tun_queue_resize(struct tun_struct *tun)
3506 {
3507 struct net_device *dev = tun->dev;
3508 struct tun_file *tfile;
3509 struct ptr_ring **rings;
3510 int n = tun->numqueues + tun->numdisabled;
3511 int ret, i;
3512
3513 rings = kmalloc_array(n, sizeof(*rings), GFP_KERNEL);
3514 if (!rings)
3515 return -ENOMEM;
3516
3517 for (i = 0; i < tun->numqueues; i++) {
3518 tfile = rtnl_dereference(tun->tfiles[i]);
3519 rings[i] = &tfile->tx_ring;
3520 }
3521 list_for_each_entry(tfile, &tun->disabled, next)
3522 rings[i++] = &tfile->tx_ring;
3523
3524 ret = ptr_ring_resize_multiple(rings, n,
3525 dev->tx_queue_len, GFP_KERNEL,
3526 tun_ptr_free);
3527
3528 kfree(rings);
3529 return ret;
3530 }
3531
3532 static int tun_device_event(struct notifier_block *unused,
3533 unsigned long event, void *ptr)
3534 {
3535 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
3536 struct tun_struct *tun = netdev_priv(dev);
3537 int i;
3538
3539 if (dev->rtnl_link_ops != &tun_link_ops)
3540 return NOTIFY_DONE;
3541
3542 switch (event) {
3543 case NETDEV_CHANGE_TX_QUEUE_LEN:
3544 if (tun_queue_resize(tun))
3545 return NOTIFY_BAD;
3546 break;
3547 case NETDEV_UP:
3548 for (i = 0; i < tun->numqueues; i++) {
3549 struct tun_file *tfile;
3550
3551 tfile = rtnl_dereference(tun->tfiles[i]);
3552 tfile->socket.sk->sk_write_space(tfile->socket.sk);
3553 }
3554 break;
3555 default:
3556 break;
3557 }
3558
3559 return NOTIFY_DONE;
3560 }
3561
3562 static struct notifier_block tun_notifier_block __read_mostly = {
3563 .notifier_call = tun_device_event,
3564 };
3565
3566 static int __init tun_init(void)
3567 {
3568 int ret = 0;
3569
3570 pr_info("%s, %s\n", DRV_DESCRIPTION, DRV_VERSION);
3571
3572 ret = rtnl_link_register(&tun_link_ops);
3573 if (ret) {
3574 pr_err("Can't register link_ops\n");
3575 goto err_linkops;
3576 }
3577
3578 ret = misc_register(&tun_miscdev);
3579 if (ret) {
3580 pr_err("Can't register misc device %d\n", TUN_MINOR);
3581 goto err_misc;
3582 }
3583
3584 ret = register_netdevice_notifier(&tun_notifier_block);
3585 if (ret) {
3586 pr_err("Can't register netdevice notifier\n");
3587 goto err_notifier;
3588 }
3589
3590 return 0;
3591
3592 err_notifier:
3593 misc_deregister(&tun_miscdev);
3594 err_misc:
3595 rtnl_link_unregister(&tun_link_ops);
3596 err_linkops:
3597 return ret;
3598 }
3599
3600 static void tun_cleanup(void)
3601 {
3602 misc_deregister(&tun_miscdev);
3603 rtnl_link_unregister(&tun_link_ops);
3604 unregister_netdevice_notifier(&tun_notifier_block);
3605 }
3606
3607 /* Get an underlying socket object from tun file. Returns error unless file is
3608 * attached to a device. The returned object works like a packet socket, it
3609 * can be used for sock_sendmsg/sock_recvmsg. The caller is responsible for
3610 * holding a reference to the file for as long as the socket is in use. */
3611 struct socket *tun_get_socket(struct file *file)
3612 {
3613 struct tun_file *tfile;
3614 if (file->f_op != &tun_fops)
3615 return ERR_PTR(-EINVAL);
3616 tfile = file->private_data;
3617 if (!tfile)
3618 return ERR_PTR(-EBADFD);
3619 return &tfile->socket;
3620 }
3621 EXPORT_SYMBOL_GPL(tun_get_socket);
3622
3623 struct ptr_ring *tun_get_tx_ring(struct file *file)
3624 {
3625 struct tun_file *tfile;
3626
3627 if (file->f_op != &tun_fops)
3628 return ERR_PTR(-EINVAL);
3629 tfile = file->private_data;
3630 if (!tfile)
3631 return ERR_PTR(-EBADFD);
3632 return &tfile->tx_ring;
3633 }
3634 EXPORT_SYMBOL_GPL(tun_get_tx_ring);
3635
3636 module_init(tun_init);
3637 module_exit(tun_cleanup);
3638 MODULE_DESCRIPTION(DRV_DESCRIPTION);
3639 MODULE_AUTHOR(DRV_COPYRIGHT);
3640 MODULE_LICENSE("GPL");
3641 MODULE_ALIAS_MISCDEV(TUN_MINOR);
3642 MODULE_ALIAS("devname:net/tun");
Linux with added FPC-III support