This source file includes following definitions.
- tun_is_xdp_frame
- tun_xdp_to_ptr
- tun_ptr_to_xdp
- tun_napi_receive
- tun_napi_poll
- tun_napi_init
- tun_napi_disable
- tun_napi_del
- tun_napi_frags_enabled
- tun_legacy_is_little_endian
- tun_get_vnet_be
- tun_set_vnet_be
- tun_legacy_is_little_endian
- tun_get_vnet_be
- tun_set_vnet_be
- tun_is_little_endian
- tun16_to_cpu
- cpu_to_tun16
- tun_hashfn
- tun_flow_find
- tun_flow_create
- tun_flow_delete
- tun_flow_flush
- tun_flow_delete_by_queue
- tun_flow_cleanup
- tun_flow_update
- tun_flow_save_rps_rxhash
- tun_automq_select_queue
- tun_ebpf_select_queue
- tun_select_queue
- tun_not_capable
- tun_set_real_num_queues
- tun_disable_queue
- tun_enable_queue
- tun_ptr_free
- tun_queue_purge
- __tun_detach
- tun_detach
- tun_detach_all
- tun_attach
- tun_get
- tun_put
- addr_hash_set
- addr_hash_test
- update_filter
- run_filter
- check_filter
- tun_net_uninit
- tun_net_open
- tun_net_close
- tun_automq_xmit
- run_ebpf_filter
- tun_net_xmit
- tun_net_mclist
- tun_net_fix_features
- tun_set_headroom
- tun_net_get_stats64
- tun_xdp_set
- tun_xdp_query
- tun_xdp
- tun_net_change_carrier
- __tun_xdp_flush_tfile
- tun_xdp_xmit
- tun_xdp_tx
- tun_flow_init
- tun_flow_uninit
- tun_net_init
- tun_sock_writeable
- tun_chr_poll
- tun_napi_alloc_frags
- tun_alloc_skb
- tun_rx_batched
- tun_can_build_skb
- __tun_build_skb
- tun_xdp_act
- tun_build_skb
- tun_get_user
- tun_chr_write_iter
- tun_put_user_xdp
- tun_put_user
- tun_ring_recv
- tun_do_read
- tun_chr_read_iter
- tun_prog_free
- __tun_set_ebpf
- tun_free_netdev
- tun_setup
- tun_validate
- tun_get_size
- tun_fill_info
- tun_sock_write_space
- tun_put_page
- tun_xdp_one
- tun_sendmsg
- tun_recvmsg
- tun_ptr_peek_len
- tun_peek_len
- tun_flags
- tun_show_flags
- tun_show_owner
- tun_show_group
- tun_set_iff
- tun_get_iff
- set_offload
- tun_detach_filter
- tun_attach_filter
- tun_set_sndbuf
- tun_set_queue
- tun_set_ebpf
- __tun_chr_ioctl
- tun_chr_ioctl
- tun_chr_compat_ioctl
- tun_chr_fasync
- tun_chr_open
- tun_chr_close
- tun_chr_show_fdinfo
- tun_default_link_ksettings
- tun_get_link_ksettings
- tun_set_link_ksettings
- tun_get_drvinfo
- tun_get_msglevel
- tun_set_msglevel
- tun_get_coalesce
- tun_set_coalesce
- tun_queue_resize
- tun_device_event
- tun_init
- tun_cleanup
- tun_get_socket
- tun_get_tx_ring
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
29
30 #define DRV_NAME "tun"
31 #define DRV_VERSION "1.6"
32 #define DRV_DESCRIPTION "Universal TUN/TAP device driver"
33 #define DRV_COPYRIGHT "(C) 1999-2004 Max Krasnyansky <maxk@qualcomm.com>"
34
35 #include <linux/module.h>
36 #include <linux/errno.h>
37 #include <linux/kernel.h>
38 #include <linux/sched/signal.h>
39 #include <linux/major.h>
40 #include <linux/slab.h>
41 #include <linux/poll.h>
42 #include <linux/fcntl.h>
43 #include <linux/init.h>
44 #include <linux/skbuff.h>
45 #include <linux/netdevice.h>
46 #include <linux/etherdevice.h>
47 #include <linux/miscdevice.h>
48 #include <linux/ethtool.h>
49 #include <linux/rtnetlink.h>
50 #include <linux/compat.h>
51 #include <linux/if.h>
52 #include <linux/if_arp.h>
53 #include <linux/if_ether.h>
54 #include <linux/if_tun.h>
55 #include <linux/if_vlan.h>
56 #include <linux/crc32.h>
57 #include <linux/nsproxy.h>
58 #include <linux/virtio_net.h>
59 #include <linux/rcupdate.h>
60 #include <net/net_namespace.h>
61 #include <net/netns/generic.h>
62 #include <net/rtnetlink.h>
63 #include <net/sock.h>
64 #include <net/xdp.h>
65 #include <linux/seq_file.h>
66 #include <linux/uio.h>
67 #include <linux/skb_array.h>
68 #include <linux/bpf.h>
69 #include <linux/bpf_trace.h>
70 #include <linux/mutex.h>
71
72 #include <linux/uaccess.h>
73 #include <linux/proc_fs.h>
74
75 static void tun_default_link_ksettings(struct net_device *dev,
76 struct ethtool_link_ksettings *cmd);
77
78
79
80
81 #ifdef TUN_DEBUG
82 static int debug;
83
84 #define tun_debug(level, tun, fmt, args...) \
85 do { \
86 if (tun->debug) \
87 netdev_printk(level, tun->dev, fmt, ##args); \
88 } while (0)
89 #define DBG1(level, fmt, args...) \
90 do { \
91 if (debug == 2) \
92 printk(level fmt, ##args); \
93 } while (0)
94 #else
95 #define tun_debug(level, tun, fmt, args...) \
96 do { \
97 if (0) \
98 netdev_printk(level, tun->dev, fmt, ##args); \
99 } while (0)
100 #define DBG1(level, fmt, args...) \
101 do { \
102 if (0) \
103 printk(level fmt, ##args); \
104 } while (0)
105 #endif
106
107 #define TUN_RX_PAD (NET_IP_ALIGN + NET_SKB_PAD)
108
109
110
111
112
113
114 #define TUN_FASYNC IFF_ATTACH_QUEUE
115
116 #define TUN_VNET_LE 0x80000000
117 #define TUN_VNET_BE 0x40000000
118
119 #define TUN_FEATURES (IFF_NO_PI | IFF_ONE_QUEUE | IFF_VNET_HDR | \
120 IFF_MULTI_QUEUE | IFF_NAPI | IFF_NAPI_FRAGS)
121
122 #define GOODCOPY_LEN 128
123
124 #define FLT_EXACT_COUNT 8
125 struct tap_filter {
126 unsigned int count;
127 u32 mask[2];
128 unsigned char addr[FLT_EXACT_COUNT][ETH_ALEN];
129 };
130
131
132
133 #define MAX_TAP_QUEUES 256
134 #define MAX_TAP_FLOWS 4096
135
136 #define TUN_FLOW_EXPIRE (3 * HZ)
137
138 struct tun_pcpu_stats {
139 u64 rx_packets;
140 u64 rx_bytes;
141 u64 tx_packets;
142 u64 tx_bytes;
143 struct u64_stats_sync syncp;
144 u32 rx_dropped;
145 u32 tx_dropped;
146 u32 rx_frame_errors;
147 };
148
149
150
151
152
153
154
155
156
157
158
159
160 struct tun_file {
161 struct sock sk;
162 struct socket socket;
163 struct tun_struct __rcu *tun;
164 struct fasync_struct *fasync;
165
166 unsigned int flags;
167 union {
168 u16 queue_index;
169 unsigned int ifindex;
170 };
171 struct napi_struct napi;
172 bool napi_enabled;
173 bool napi_frags_enabled;
174 struct mutex napi_mutex;
175 struct list_head next;
176 struct tun_struct *detached;
177 struct ptr_ring tx_ring;
178 struct xdp_rxq_info xdp_rxq;
179 };
180
181 struct tun_page {
182 struct page *page;
183 int count;
184 };
185
186 struct tun_flow_entry {
187 struct hlist_node hash_link;
188 struct rcu_head rcu;
189 struct tun_struct *tun;
190
191 u32 rxhash;
192 u32 rps_rxhash;
193 int queue_index;
194 unsigned long updated ____cacheline_aligned_in_smp;
195 };
196
197 #define TUN_NUM_FLOW_ENTRIES 1024
198 #define TUN_MASK_FLOW_ENTRIES (TUN_NUM_FLOW_ENTRIES - 1)
199
200 struct tun_prog {
201 struct rcu_head rcu;
202 struct bpf_prog *prog;
203 };
204
205
206
207
208
209 struct tun_struct {
210 struct tun_file __rcu *tfiles[MAX_TAP_QUEUES];
211 unsigned int numqueues;
212 unsigned int flags;
213 kuid_t owner;
214 kgid_t group;
215
216 struct net_device *dev;
217 netdev_features_t set_features;
218 #define TUN_USER_FEATURES (NETIF_F_HW_CSUM|NETIF_F_TSO_ECN|NETIF_F_TSO| \
219 NETIF_F_TSO6)
220
221 int align;
222 int vnet_hdr_sz;
223 int sndbuf;
224 struct tap_filter txflt;
225 struct sock_fprog fprog;
226
227 bool filter_attached;
228 #ifdef TUN_DEBUG
229 int debug;
230 #endif
231 spinlock_t lock;
232 struct hlist_head flows[TUN_NUM_FLOW_ENTRIES];
233 struct timer_list flow_gc_timer;
234 unsigned long ageing_time;
235 unsigned int numdisabled;
236 struct list_head disabled;
237 void *security;
238 u32 flow_count;
239 u32 rx_batched;
240 struct tun_pcpu_stats __percpu *pcpu_stats;
241 struct bpf_prog __rcu *xdp_prog;
242 struct tun_prog __rcu *steering_prog;
243 struct tun_prog __rcu *filter_prog;
244 struct ethtool_link_ksettings link_ksettings;
245 };
246
247 struct veth {
248 __be16 h_vlan_proto;
249 __be16 h_vlan_TCI;
250 };
251
252 bool tun_is_xdp_frame(void *ptr)
253 {
254 return (unsigned long)ptr & TUN_XDP_FLAG;
255 }
256 EXPORT_SYMBOL(tun_is_xdp_frame);
257
258 void *tun_xdp_to_ptr(void *ptr)
259 {
260 return (void *)((unsigned long)ptr | TUN_XDP_FLAG);
261 }
262 EXPORT_SYMBOL(tun_xdp_to_ptr);
263
264 void *tun_ptr_to_xdp(void *ptr)
265 {
266 return (void *)((unsigned long)ptr & ~TUN_XDP_FLAG);
267 }
268 EXPORT_SYMBOL(tun_ptr_to_xdp);
269
270 static int tun_napi_receive(struct napi_struct *napi, int budget)
271 {
272 struct tun_file *tfile = container_of(napi, struct tun_file, napi);
273 struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
274 struct sk_buff_head process_queue;
275 struct sk_buff *skb;
276 int received = 0;
277
278 __skb_queue_head_init(&process_queue);
279
280 spin_lock(&queue->lock);
281 skb_queue_splice_tail_init(queue, &process_queue);
282 spin_unlock(&queue->lock);
283
284 while (received < budget && (skb = __skb_dequeue(&process_queue))) {
285 napi_gro_receive(napi, skb);
286 ++received;
287 }
288
289 if (!skb_queue_empty(&process_queue)) {
290 spin_lock(&queue->lock);
291 skb_queue_splice(&process_queue, queue);
292 spin_unlock(&queue->lock);
293 }
294
295 return received;
296 }
297
298 static int tun_napi_poll(struct napi_struct *napi, int budget)
299 {
300 unsigned int received;
301
302 received = tun_napi_receive(napi, budget);
303
304 if (received < budget)
305 napi_complete_done(napi, received);
306
307 return received;
308 }
309
310 static void tun_napi_init(struct tun_struct *tun, struct tun_file *tfile,
311 bool napi_en, bool napi_frags)
312 {
313 tfile->napi_enabled = napi_en;
314 tfile->napi_frags_enabled = napi_en && napi_frags;
315 if (napi_en) {
316 netif_tx_napi_add(tun->dev, &tfile->napi, tun_napi_poll,
317 NAPI_POLL_WEIGHT);
318 napi_enable(&tfile->napi);
319 }
320 }
321
322 static void tun_napi_disable(struct tun_file *tfile)
323 {
324 if (tfile->napi_enabled)
325 napi_disable(&tfile->napi);
326 }
327
328 static void tun_napi_del(struct tun_file *tfile)
329 {
330 if (tfile->napi_enabled)
331 netif_napi_del(&tfile->napi);
332 }
333
334 static bool tun_napi_frags_enabled(const struct tun_file *tfile)
335 {
336 return tfile->napi_frags_enabled;
337 }
338
339 #ifdef CONFIG_TUN_VNET_CROSS_LE
340 static inline bool tun_legacy_is_little_endian(struct tun_struct *tun)
341 {
342 return tun->flags & TUN_VNET_BE ? false :
343 virtio_legacy_is_little_endian();
344 }
345
346 static long tun_get_vnet_be(struct tun_struct *tun, int __user *argp)
347 {
348 int be = !!(tun->flags & TUN_VNET_BE);
349
350 if (put_user(be, argp))
351 return -EFAULT;
352
353 return 0;
354 }
355
356 static long tun_set_vnet_be(struct tun_struct *tun, int __user *argp)
357 {
358 int be;
359
360 if (get_user(be, argp))
361 return -EFAULT;
362
363 if (be)
364 tun->flags |= TUN_VNET_BE;
365 else
366 tun->flags &= ~TUN_VNET_BE;
367
368 return 0;
369 }
370 #else
371 static inline bool tun_legacy_is_little_endian(struct tun_struct *tun)
372 {
373 return virtio_legacy_is_little_endian();
374 }
375
376 static long tun_get_vnet_be(struct tun_struct *tun, int __user *argp)
377 {
378 return -EINVAL;
379 }
380
381 static long tun_set_vnet_be(struct tun_struct *tun, int __user *argp)
382 {
383 return -EINVAL;
384 }
385 #endif
386
387 static inline bool tun_is_little_endian(struct tun_struct *tun)
388 {
389 return tun->flags & TUN_VNET_LE ||
390 tun_legacy_is_little_endian(tun);
391 }
392
393 static inline u16 tun16_to_cpu(struct tun_struct *tun, __virtio16 val)
394 {
395 return __virtio16_to_cpu(tun_is_little_endian(tun), val);
396 }
397
398 static inline __virtio16 cpu_to_tun16(struct tun_struct *tun, u16 val)
399 {
400 return __cpu_to_virtio16(tun_is_little_endian(tun), val);
401 }
402
403 static inline u32 tun_hashfn(u32 rxhash)
404 {
405 return rxhash & TUN_MASK_FLOW_ENTRIES;
406 }
407
408 static struct tun_flow_entry *tun_flow_find(struct hlist_head *head, u32 rxhash)
409 {
410 struct tun_flow_entry *e;
411
412 hlist_for_each_entry_rcu(e, head, hash_link) {
413 if (e->rxhash == rxhash)
414 return e;
415 }
416 return NULL;
417 }
418
419 static struct tun_flow_entry *tun_flow_create(struct tun_struct *tun,
420 struct hlist_head *head,
421 u32 rxhash, u16 queue_index)
422 {
423 struct tun_flow_entry *e = kmalloc(sizeof(*e), GFP_ATOMIC);
424
425 if (e) {
426 tun_debug(KERN_INFO, tun, "create flow: hash %u index %u\n",
427 rxhash, queue_index);
428 e->updated = jiffies;
429 e->rxhash = rxhash;
430 e->rps_rxhash = 0;
431 e->queue_index = queue_index;
432 e->tun = tun;
433 hlist_add_head_rcu(&e->hash_link, head);
434 ++tun->flow_count;
435 }
436 return e;
437 }
438
439 static void tun_flow_delete(struct tun_struct *tun, struct tun_flow_entry *e)
440 {
441 tun_debug(KERN_INFO, tun, "delete flow: hash %u index %u\n",
442 e->rxhash, e->queue_index);
443 hlist_del_rcu(&e->hash_link);
444 kfree_rcu(e, rcu);
445 --tun->flow_count;
446 }
447
448 static void tun_flow_flush(struct tun_struct *tun)
449 {
450 int i;
451
452 spin_lock_bh(&tun->lock);
453 for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
454 struct tun_flow_entry *e;
455 struct hlist_node *n;
456
457 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link)
458 tun_flow_delete(tun, e);
459 }
460 spin_unlock_bh(&tun->lock);
461 }
462
463 static void tun_flow_delete_by_queue(struct tun_struct *tun, u16 queue_index)
464 {
465 int i;
466
467 spin_lock_bh(&tun->lock);
468 for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
469 struct tun_flow_entry *e;
470 struct hlist_node *n;
471
472 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
473 if (e->queue_index == queue_index)
474 tun_flow_delete(tun, e);
475 }
476 }
477 spin_unlock_bh(&tun->lock);
478 }
479
480 static void tun_flow_cleanup(struct timer_list *t)
481 {
482 struct tun_struct *tun = from_timer(tun, t, flow_gc_timer);
483 unsigned long delay = tun->ageing_time;
484 unsigned long next_timer = jiffies + delay;
485 unsigned long count = 0;
486 int i;
487
488 tun_debug(KERN_INFO, tun, "tun_flow_cleanup\n");
489
490 spin_lock(&tun->lock);
491 for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
492 struct tun_flow_entry *e;
493 struct hlist_node *n;
494
495 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
496 unsigned long this_timer;
497
498 this_timer = e->updated + delay;
499 if (time_before_eq(this_timer, jiffies)) {
500 tun_flow_delete(tun, e);
501 continue;
502 }
503 count++;
504 if (time_before(this_timer, next_timer))
505 next_timer = this_timer;
506 }
507 }
508
509 if (count)
510 mod_timer(&tun->flow_gc_timer, round_jiffies_up(next_timer));
511 spin_unlock(&tun->lock);
512 }
513
514 static void tun_flow_update(struct tun_struct *tun, u32 rxhash,
515 struct tun_file *tfile)
516 {
517 struct hlist_head *head;
518 struct tun_flow_entry *e;
519 unsigned long delay = tun->ageing_time;
520 u16 queue_index = tfile->queue_index;
521
522 head = &tun->flows[tun_hashfn(rxhash)];
523
524 rcu_read_lock();
525
526 e = tun_flow_find(head, rxhash);
527 if (likely(e)) {
528
529 if (READ_ONCE(e->queue_index) != queue_index)
530 WRITE_ONCE(e->queue_index, queue_index);
531 if (e->updated != jiffies)
532 e->updated = jiffies;
533 sock_rps_record_flow_hash(e->rps_rxhash);
534 } else {
535 spin_lock_bh(&tun->lock);
536 if (!tun_flow_find(head, rxhash) &&
537 tun->flow_count < MAX_TAP_FLOWS)
538 tun_flow_create(tun, head, rxhash, queue_index);
539
540 if (!timer_pending(&tun->flow_gc_timer))
541 mod_timer(&tun->flow_gc_timer,
542 round_jiffies_up(jiffies + delay));
543 spin_unlock_bh(&tun->lock);
544 }
545
546 rcu_read_unlock();
547 }
548
549
550
551
552
553 static inline void tun_flow_save_rps_rxhash(struct tun_flow_entry *e, u32 hash)
554 {
555 if (unlikely(e->rps_rxhash != hash))
556 e->rps_rxhash = hash;
557 }
558
559
560
561
562
563
564
565 static u16 tun_automq_select_queue(struct tun_struct *tun, struct sk_buff *skb)
566 {
567 struct tun_flow_entry *e;
568 u32 txq = 0;
569 u32 numqueues = 0;
570
571 numqueues = READ_ONCE(tun->numqueues);
572
573 txq = __skb_get_hash_symmetric(skb);
574 e = tun_flow_find(&tun->flows[tun_hashfn(txq)], txq);
575 if (e) {
576 tun_flow_save_rps_rxhash(e, txq);
577 txq = e->queue_index;
578 } else {
579
580 txq = ((u64)txq * numqueues) >> 32;
581 }
582
583 return txq;
584 }
585
586 static u16 tun_ebpf_select_queue(struct tun_struct *tun, struct sk_buff *skb)
587 {
588 struct tun_prog *prog;
589 u32 numqueues;
590 u16 ret = 0;
591
592 numqueues = READ_ONCE(tun->numqueues);
593 if (!numqueues)
594 return 0;
595
596 prog = rcu_dereference(tun->steering_prog);
597 if (prog)
598 ret = bpf_prog_run_clear_cb(prog->prog, skb);
599
600 return ret % numqueues;
601 }
602
603 static u16 tun_select_queue(struct net_device *dev, struct sk_buff *skb,
604 struct net_device *sb_dev)
605 {
606 struct tun_struct *tun = netdev_priv(dev);
607 u16 ret;
608
609 rcu_read_lock();
610 if (rcu_dereference(tun->steering_prog))
611 ret = tun_ebpf_select_queue(tun, skb);
612 else
613 ret = tun_automq_select_queue(tun, skb);
614 rcu_read_unlock();
615
616 return ret;
617 }
618
619 static inline bool tun_not_capable(struct tun_struct *tun)
620 {
621 const struct cred *cred = current_cred();
622 struct net *net = dev_net(tun->dev);
623
624 return ((uid_valid(tun->owner) && !uid_eq(cred->euid, tun->owner)) ||
625 (gid_valid(tun->group) && !in_egroup_p(tun->group))) &&
626 !ns_capable(net->user_ns, CAP_NET_ADMIN);
627 }
628
629 static void tun_set_real_num_queues(struct tun_struct *tun)
630 {
631 netif_set_real_num_tx_queues(tun->dev, tun->numqueues);
632 netif_set_real_num_rx_queues(tun->dev, tun->numqueues);
633 }
634
635 static void tun_disable_queue(struct tun_struct *tun, struct tun_file *tfile)
636 {
637 tfile->detached = tun;
638 list_add_tail(&tfile->next, &tun->disabled);
639 ++tun->numdisabled;
640 }
641
642 static struct tun_struct *tun_enable_queue(struct tun_file *tfile)
643 {
644 struct tun_struct *tun = tfile->detached;
645
646 tfile->detached = NULL;
647 list_del_init(&tfile->next);
648 --tun->numdisabled;
649 return tun;
650 }
651
652 void tun_ptr_free(void *ptr)
653 {
654 if (!ptr)
655 return;
656 if (tun_is_xdp_frame(ptr)) {
657 struct xdp_frame *xdpf = tun_ptr_to_xdp(ptr);
658
659 xdp_return_frame(xdpf);
660 } else {
661 __skb_array_destroy_skb(ptr);
662 }
663 }
664 EXPORT_SYMBOL_GPL(tun_ptr_free);
665
666 static void tun_queue_purge(struct tun_file *tfile)
667 {
668 void *ptr;
669
670 while ((ptr = ptr_ring_consume(&tfile->tx_ring)) != NULL)
671 tun_ptr_free(ptr);
672
673 skb_queue_purge(&tfile->sk.sk_write_queue);
674 skb_queue_purge(&tfile->sk.sk_error_queue);
675 }
676
677 static void __tun_detach(struct tun_file *tfile, bool clean)
678 {
679 struct tun_file *ntfile;
680 struct tun_struct *tun;
681
682 tun = rtnl_dereference(tfile->tun);
683
684 if (tun && clean) {
685 tun_napi_disable(tfile);
686 tun_napi_del(tfile);
687 }
688
689 if (tun && !tfile->detached) {
690 u16 index = tfile->queue_index;
691 BUG_ON(index >= tun->numqueues);
692
693 rcu_assign_pointer(tun->tfiles[index],
694 tun->tfiles[tun->numqueues - 1]);
695 ntfile = rtnl_dereference(tun->tfiles[index]);
696 ntfile->queue_index = index;
697 rcu_assign_pointer(tun->tfiles[tun->numqueues - 1],
698 NULL);
699
700 --tun->numqueues;
701 if (clean) {
702 RCU_INIT_POINTER(tfile->tun, NULL);
703 sock_put(&tfile->sk);
704 } else
705 tun_disable_queue(tun, tfile);
706
707 synchronize_net();
708 tun_flow_delete_by_queue(tun, tun->numqueues + 1);
709
710 tun_queue_purge(tfile);
711 tun_set_real_num_queues(tun);
712 } else if (tfile->detached && clean) {
713 tun = tun_enable_queue(tfile);
714 sock_put(&tfile->sk);
715 }
716
717 if (clean) {
718 if (tun && tun->numqueues == 0 && tun->numdisabled == 0) {
719 netif_carrier_off(tun->dev);
720
721 if (!(tun->flags & IFF_PERSIST) &&
722 tun->dev->reg_state == NETREG_REGISTERED)
723 unregister_netdevice(tun->dev);
724 }
725 if (tun)
726 xdp_rxq_info_unreg(&tfile->xdp_rxq);
727 ptr_ring_cleanup(&tfile->tx_ring, tun_ptr_free);
728 sock_put(&tfile->sk);
729 }
730 }
731
732 static void tun_detach(struct tun_file *tfile, bool clean)
733 {
734 struct tun_struct *tun;
735 struct net_device *dev;
736
737 rtnl_lock();
738 tun = rtnl_dereference(tfile->tun);
739 dev = tun ? tun->dev : NULL;
740 __tun_detach(tfile, clean);
741 if (dev)
742 netdev_state_change(dev);
743 rtnl_unlock();
744 }
745
746 static void tun_detach_all(struct net_device *dev)
747 {
748 struct tun_struct *tun = netdev_priv(dev);
749 struct tun_file *tfile, *tmp;
750 int i, n = tun->numqueues;
751
752 for (i = 0; i < n; i++) {
753 tfile = rtnl_dereference(tun->tfiles[i]);
754 BUG_ON(!tfile);
755 tun_napi_disable(tfile);
756 tfile->socket.sk->sk_shutdown = RCV_SHUTDOWN;
757 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
758 RCU_INIT_POINTER(tfile->tun, NULL);
759 --tun->numqueues;
760 }
761 list_for_each_entry(tfile, &tun->disabled, next) {
762 tfile->socket.sk->sk_shutdown = RCV_SHUTDOWN;
763 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
764 RCU_INIT_POINTER(tfile->tun, NULL);
765 }
766 BUG_ON(tun->numqueues != 0);
767
768 synchronize_net();
769 for (i = 0; i < n; i++) {
770 tfile = rtnl_dereference(tun->tfiles[i]);
771 tun_napi_del(tfile);
772
773 tun_queue_purge(tfile);
774 xdp_rxq_info_unreg(&tfile->xdp_rxq);
775 sock_put(&tfile->sk);
776 }
777 list_for_each_entry_safe(tfile, tmp, &tun->disabled, next) {
778 tun_enable_queue(tfile);
779 tun_queue_purge(tfile);
780 xdp_rxq_info_unreg(&tfile->xdp_rxq);
781 sock_put(&tfile->sk);
782 }
783 BUG_ON(tun->numdisabled != 0);
784
785 if (tun->flags & IFF_PERSIST)
786 module_put(THIS_MODULE);
787 }
788
789 static int tun_attach(struct tun_struct *tun, struct file *file,
790 bool skip_filter, bool napi, bool napi_frags,
791 bool publish_tun)
792 {
793 struct tun_file *tfile = file->private_data;
794 struct net_device *dev = tun->dev;
795 int err;
796
797 err = security_tun_dev_attach(tfile->socket.sk, tun->security);
798 if (err < 0)
799 goto out;
800
801 err = -EINVAL;
802 if (rtnl_dereference(tfile->tun) && !tfile->detached)
803 goto out;
804
805 err = -EBUSY;
806 if (!(tun->flags & IFF_MULTI_QUEUE) && tun->numqueues == 1)
807 goto out;
808
809 err = -E2BIG;
810 if (!tfile->detached &&
811 tun->numqueues + tun->numdisabled == MAX_TAP_QUEUES)
812 goto out;
813
814 err = 0;
815
816
817 if (!skip_filter && (tun->filter_attached == true)) {
818 lock_sock(tfile->socket.sk);
819 err = sk_attach_filter(&tun->fprog, tfile->socket.sk);
820 release_sock(tfile->socket.sk);
821 if (!err)
822 goto out;
823 }
824
825 if (!tfile->detached &&
826 ptr_ring_resize(&tfile->tx_ring, dev->tx_queue_len,
827 GFP_KERNEL, tun_ptr_free)) {
828 err = -ENOMEM;
829 goto out;
830 }
831
832 tfile->queue_index = tun->numqueues;
833 tfile->socket.sk->sk_shutdown &= ~RCV_SHUTDOWN;
834
835 if (tfile->detached) {
836
837 WARN_ON(!xdp_rxq_info_is_reg(&tfile->xdp_rxq));
838
839 if (tfile->xdp_rxq.queue_index != tfile->queue_index)
840 tfile->xdp_rxq.queue_index = tfile->queue_index;
841 } else {
842
843 err = xdp_rxq_info_reg(&tfile->xdp_rxq,
844 tun->dev, tfile->queue_index);
845 if (err < 0)
846 goto out;
847 err = xdp_rxq_info_reg_mem_model(&tfile->xdp_rxq,
848 MEM_TYPE_PAGE_SHARED, NULL);
849 if (err < 0) {
850 xdp_rxq_info_unreg(&tfile->xdp_rxq);
851 goto out;
852 }
853 err = 0;
854 }
855
856 if (tfile->detached) {
857 tun_enable_queue(tfile);
858 } else {
859 sock_hold(&tfile->sk);
860 tun_napi_init(tun, tfile, napi, napi_frags);
861 }
862
863 if (rtnl_dereference(tun->xdp_prog))
864 sock_set_flag(&tfile->sk, SOCK_XDP);
865
866
867
868
869
870
871
872
873
874 if (publish_tun)
875 rcu_assign_pointer(tfile->tun, tun);
876 rcu_assign_pointer(tun->tfiles[tun->numqueues], tfile);
877 tun->numqueues++;
878 tun_set_real_num_queues(tun);
879 out:
880 return err;
881 }
882
883 static struct tun_struct *tun_get(struct tun_file *tfile)
884 {
885 struct tun_struct *tun;
886
887 rcu_read_lock();
888 tun = rcu_dereference(tfile->tun);
889 if (tun)
890 dev_hold(tun->dev);
891 rcu_read_unlock();
892
893 return tun;
894 }
895
896 static void tun_put(struct tun_struct *tun)
897 {
898 dev_put(tun->dev);
899 }
900
901
902 static void addr_hash_set(u32 *mask, const u8 *addr)
903 {
904 int n = ether_crc(ETH_ALEN, addr) >> 26;
905 mask[n >> 5] |= (1 << (n & 31));
906 }
907
908 static unsigned int addr_hash_test(const u32 *mask, const u8 *addr)
909 {
910 int n = ether_crc(ETH_ALEN, addr) >> 26;
911 return mask[n >> 5] & (1 << (n & 31));
912 }
913
914 static int update_filter(struct tap_filter *filter, void __user *arg)
915 {
916 struct { u8 u[ETH_ALEN]; } *addr;
917 struct tun_filter uf;
918 int err, alen, n, nexact;
919
920 if (copy_from_user(&uf, arg, sizeof(uf)))
921 return -EFAULT;
922
923 if (!uf.count) {
924
925 filter->count = 0;
926 return 0;
927 }
928
929 alen = ETH_ALEN * uf.count;
930 addr = memdup_user(arg + sizeof(uf), alen);
931 if (IS_ERR(addr))
932 return PTR_ERR(addr);
933
934
935
936
937 filter->count = 0;
938 wmb();
939
940
941 for (n = 0; n < uf.count && n < FLT_EXACT_COUNT; n++)
942 memcpy(filter->addr[n], addr[n].u, ETH_ALEN);
943
944 nexact = n;
945
946
947
948 memset(filter->mask, 0, sizeof(filter->mask));
949 for (; n < uf.count; n++) {
950 if (!is_multicast_ether_addr(addr[n].u)) {
951 err = 0;
952 goto free_addr;
953 }
954 addr_hash_set(filter->mask, addr[n].u);
955 }
956
957
958
959 if ((uf.flags & TUN_FLT_ALLMULTI))
960 memset(filter->mask, ~0, sizeof(filter->mask));
961
962
963 wmb();
964 filter->count = nexact;
965
966
967 err = nexact;
968 free_addr:
969 kfree(addr);
970 return err;
971 }
972
973
974 static int run_filter(struct tap_filter *filter, const struct sk_buff *skb)
975 {
976
977
978 struct ethhdr *eh = (struct ethhdr *) skb->data;
979 int i;
980
981
982 for (i = 0; i < filter->count; i++)
983 if (ether_addr_equal(eh->h_dest, filter->addr[i]))
984 return 1;
985
986
987 if (is_multicast_ether_addr(eh->h_dest))
988 return addr_hash_test(filter->mask, eh->h_dest);
989
990 return 0;
991 }
992
993
994
995
996
997 static int check_filter(struct tap_filter *filter, const struct sk_buff *skb)
998 {
999 if (!filter->count)
1000 return 1;
1001
1002 return run_filter(filter, skb);
1003 }
1004
1005
1006
1007 static const struct ethtool_ops tun_ethtool_ops;
1008
1009
1010 static void tun_net_uninit(struct net_device *dev)
1011 {
1012 tun_detach_all(dev);
1013 }
1014
1015
1016 static int tun_net_open(struct net_device *dev)
1017 {
1018 netif_tx_start_all_queues(dev);
1019
1020 return 0;
1021 }
1022
1023
1024 static int tun_net_close(struct net_device *dev)
1025 {
1026 netif_tx_stop_all_queues(dev);
1027 return 0;
1028 }
1029
1030
1031 static void tun_automq_xmit(struct tun_struct *tun, struct sk_buff *skb)
1032 {
1033 #ifdef CONFIG_RPS
1034 if (tun->numqueues == 1 && static_branch_unlikely(&rps_needed)) {
1035
1036
1037
1038 struct tun_flow_entry *e;
1039 __u32 rxhash;
1040
1041 rxhash = __skb_get_hash_symmetric(skb);
1042 e = tun_flow_find(&tun->flows[tun_hashfn(rxhash)], rxhash);
1043 if (e)
1044 tun_flow_save_rps_rxhash(e, rxhash);
1045 }
1046 #endif
1047 }
1048
1049 static unsigned int run_ebpf_filter(struct tun_struct *tun,
1050 struct sk_buff *skb,
1051 int len)
1052 {
1053 struct tun_prog *prog = rcu_dereference(tun->filter_prog);
1054
1055 if (prog)
1056 len = bpf_prog_run_clear_cb(prog->prog, skb);
1057
1058 return len;
1059 }
1060
1061
1062 static netdev_tx_t tun_net_xmit(struct sk_buff *skb, struct net_device *dev)
1063 {
1064 struct tun_struct *tun = netdev_priv(dev);
1065 int txq = skb->queue_mapping;
1066 struct tun_file *tfile;
1067 int len = skb->len;
1068
1069 rcu_read_lock();
1070 tfile = rcu_dereference(tun->tfiles[txq]);
1071
1072
1073 if (!tfile)
1074 goto drop;
1075
1076 if (!rcu_dereference(tun->steering_prog))
1077 tun_automq_xmit(tun, skb);
1078
1079 tun_debug(KERN_INFO, tun, "tun_net_xmit %d\n", skb->len);
1080
1081 BUG_ON(!tfile);
1082
1083
1084
1085
1086 if (!check_filter(&tun->txflt, skb))
1087 goto drop;
1088
1089 if (tfile->socket.sk->sk_filter &&
1090 sk_filter(tfile->socket.sk, skb))
1091 goto drop;
1092
1093 len = run_ebpf_filter(tun, skb, len);
1094 if (len == 0 || pskb_trim(skb, len))
1095 goto drop;
1096
1097 if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC)))
1098 goto drop;
1099
1100 skb_tx_timestamp(skb);
1101
1102
1103
1104
1105 skb_orphan(skb);
1106
1107 nf_reset_ct(skb);
1108
1109 if (ptr_ring_produce(&tfile->tx_ring, skb))
1110 goto drop;
1111
1112
1113 if (tfile->flags & TUN_FASYNC)
1114 kill_fasync(&tfile->fasync, SIGIO, POLL_IN);
1115 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
1116
1117 rcu_read_unlock();
1118 return NETDEV_TX_OK;
1119
1120 drop:
1121 this_cpu_inc(tun->pcpu_stats->tx_dropped);
1122 skb_tx_error(skb);
1123 kfree_skb(skb);
1124 rcu_read_unlock();
1125 return NET_XMIT_DROP;
1126 }
1127
1128 static void tun_net_mclist(struct net_device *dev)
1129 {
1130
1131
1132
1133
1134
1135 }
1136
1137 static netdev_features_t tun_net_fix_features(struct net_device *dev,
1138 netdev_features_t features)
1139 {
1140 struct tun_struct *tun = netdev_priv(dev);
1141
1142 return (features & tun->set_features) | (features & ~TUN_USER_FEATURES);
1143 }
1144
1145 static void tun_set_headroom(struct net_device *dev, int new_hr)
1146 {
1147 struct tun_struct *tun = netdev_priv(dev);
1148
1149 if (new_hr < NET_SKB_PAD)
1150 new_hr = NET_SKB_PAD;
1151
1152 tun->align = new_hr;
1153 }
1154
1155 static void
1156 tun_net_get_stats64(struct net_device *dev, struct rtnl_link_stats64 *stats)
1157 {
1158 u32 rx_dropped = 0, tx_dropped = 0, rx_frame_errors = 0;
1159 struct tun_struct *tun = netdev_priv(dev);
1160 struct tun_pcpu_stats *p;
1161 int i;
1162
1163 for_each_possible_cpu(i) {
1164 u64 rxpackets, rxbytes, txpackets, txbytes;
1165 unsigned int start;
1166
1167 p = per_cpu_ptr(tun->pcpu_stats, i);
1168 do {
1169 start = u64_stats_fetch_begin(&p->syncp);
1170 rxpackets = p->rx_packets;
1171 rxbytes = p->rx_bytes;
1172 txpackets = p->tx_packets;
1173 txbytes = p->tx_bytes;
1174 } while (u64_stats_fetch_retry(&p->syncp, start));
1175
1176 stats->rx_packets += rxpackets;
1177 stats->rx_bytes += rxbytes;
1178 stats->tx_packets += txpackets;
1179 stats->tx_bytes += txbytes;
1180
1181
1182 rx_dropped += p->rx_dropped;
1183 rx_frame_errors += p->rx_frame_errors;
1184 tx_dropped += p->tx_dropped;
1185 }
1186 stats->rx_dropped = rx_dropped;
1187 stats->rx_frame_errors = rx_frame_errors;
1188 stats->tx_dropped = tx_dropped;
1189 }
1190
1191 static int tun_xdp_set(struct net_device *dev, struct bpf_prog *prog,
1192 struct netlink_ext_ack *extack)
1193 {
1194 struct tun_struct *tun = netdev_priv(dev);
1195 struct tun_file *tfile;
1196 struct bpf_prog *old_prog;
1197 int i;
1198
1199 old_prog = rtnl_dereference(tun->xdp_prog);
1200 rcu_assign_pointer(tun->xdp_prog, prog);
1201 if (old_prog)
1202 bpf_prog_put(old_prog);
1203
1204 for (i = 0; i < tun->numqueues; i++) {
1205 tfile = rtnl_dereference(tun->tfiles[i]);
1206 if (prog)
1207 sock_set_flag(&tfile->sk, SOCK_XDP);
1208 else
1209 sock_reset_flag(&tfile->sk, SOCK_XDP);
1210 }
1211 list_for_each_entry(tfile, &tun->disabled, next) {
1212 if (prog)
1213 sock_set_flag(&tfile->sk, SOCK_XDP);
1214 else
1215 sock_reset_flag(&tfile->sk, SOCK_XDP);
1216 }
1217
1218 return 0;
1219 }
1220
1221 static u32 tun_xdp_query(struct net_device *dev)
1222 {
1223 struct tun_struct *tun = netdev_priv(dev);
1224 const struct bpf_prog *xdp_prog;
1225
1226 xdp_prog = rtnl_dereference(tun->xdp_prog);
1227 if (xdp_prog)
1228 return xdp_prog->aux->id;
1229
1230 return 0;
1231 }
1232
1233 static int tun_xdp(struct net_device *dev, struct netdev_bpf *xdp)
1234 {
1235 switch (xdp->command) {
1236 case XDP_SETUP_PROG:
1237 return tun_xdp_set(dev, xdp->prog, xdp->extack);
1238 case XDP_QUERY_PROG:
1239 xdp->prog_id = tun_xdp_query(dev);
1240 return 0;
1241 default:
1242 return -EINVAL;
1243 }
1244 }
1245
1246 static int tun_net_change_carrier(struct net_device *dev, bool new_carrier)
1247 {
1248 if (new_carrier) {
1249 struct tun_struct *tun = netdev_priv(dev);
1250
1251 if (!tun->numqueues)
1252 return -EPERM;
1253
1254 netif_carrier_on(dev);
1255 } else {
1256 netif_carrier_off(dev);
1257 }
1258 return 0;
1259 }
1260
1261 static const struct net_device_ops tun_netdev_ops = {
1262 .ndo_uninit = tun_net_uninit,
1263 .ndo_open = tun_net_open,
1264 .ndo_stop = tun_net_close,
1265 .ndo_start_xmit = tun_net_xmit,
1266 .ndo_fix_features = tun_net_fix_features,
1267 .ndo_select_queue = tun_select_queue,
1268 .ndo_set_rx_headroom = tun_set_headroom,
1269 .ndo_get_stats64 = tun_net_get_stats64,
1270 .ndo_change_carrier = tun_net_change_carrier,
1271 };
1272
1273 static void __tun_xdp_flush_tfile(struct tun_file *tfile)
1274 {
1275
1276 if (tfile->flags & TUN_FASYNC)
1277 kill_fasync(&tfile->fasync, SIGIO, POLL_IN);
1278 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
1279 }
1280
1281 static int tun_xdp_xmit(struct net_device *dev, int n,
1282 struct xdp_frame **frames, u32 flags)
1283 {
1284 struct tun_struct *tun = netdev_priv(dev);
1285 struct tun_file *tfile;
1286 u32 numqueues;
1287 int drops = 0;
1288 int cnt = n;
1289 int i;
1290
1291 if (unlikely(flags & ~XDP_XMIT_FLAGS_MASK))
1292 return -EINVAL;
1293
1294 rcu_read_lock();
1295
1296 resample:
1297 numqueues = READ_ONCE(tun->numqueues);
1298 if (!numqueues) {
1299 rcu_read_unlock();
1300 return -ENXIO;
1301 }
1302
1303 tfile = rcu_dereference(tun->tfiles[smp_processor_id() %
1304 numqueues]);
1305 if (unlikely(!tfile))
1306 goto resample;
1307
1308 spin_lock(&tfile->tx_ring.producer_lock);
1309 for (i = 0; i < n; i++) {
1310 struct xdp_frame *xdp = frames[i];
1311
1312
1313
1314 void *frame = tun_xdp_to_ptr(xdp);
1315
1316 if (__ptr_ring_produce(&tfile->tx_ring, frame)) {
1317 this_cpu_inc(tun->pcpu_stats->tx_dropped);
1318 xdp_return_frame_rx_napi(xdp);
1319 drops++;
1320 }
1321 }
1322 spin_unlock(&tfile->tx_ring.producer_lock);
1323
1324 if (flags & XDP_XMIT_FLUSH)
1325 __tun_xdp_flush_tfile(tfile);
1326
1327 rcu_read_unlock();
1328 return cnt - drops;
1329 }
1330
1331 static int tun_xdp_tx(struct net_device *dev, struct xdp_buff *xdp)
1332 {
1333 struct xdp_frame *frame = convert_to_xdp_frame(xdp);
1334
1335 if (unlikely(!frame))
1336 return -EOVERFLOW;
1337
1338 return tun_xdp_xmit(dev, 1, &frame, XDP_XMIT_FLUSH);
1339 }
1340
1341 static const struct net_device_ops tap_netdev_ops = {
1342 .ndo_uninit = tun_net_uninit,
1343 .ndo_open = tun_net_open,
1344 .ndo_stop = tun_net_close,
1345 .ndo_start_xmit = tun_net_xmit,
1346 .ndo_fix_features = tun_net_fix_features,
1347 .ndo_set_rx_mode = tun_net_mclist,
1348 .ndo_set_mac_address = eth_mac_addr,
1349 .ndo_validate_addr = eth_validate_addr,
1350 .ndo_select_queue = tun_select_queue,
1351 .ndo_features_check = passthru_features_check,
1352 .ndo_set_rx_headroom = tun_set_headroom,
1353 .ndo_get_stats64 = tun_net_get_stats64,
1354 .ndo_bpf = tun_xdp,
1355 .ndo_xdp_xmit = tun_xdp_xmit,
1356 .ndo_change_carrier = tun_net_change_carrier,
1357 };
1358
1359 static void tun_flow_init(struct tun_struct *tun)
1360 {
1361 int i;
1362
1363 for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++)
1364 INIT_HLIST_HEAD(&tun->flows[i]);
1365
1366 tun->ageing_time = TUN_FLOW_EXPIRE;
1367 timer_setup(&tun->flow_gc_timer, tun_flow_cleanup, 0);
1368 mod_timer(&tun->flow_gc_timer,
1369 round_jiffies_up(jiffies + tun->ageing_time));
1370 }
1371
1372 static void tun_flow_uninit(struct tun_struct *tun)
1373 {
1374 del_timer_sync(&tun->flow_gc_timer);
1375 tun_flow_flush(tun);
1376 }
1377
1378 #define MIN_MTU 68
1379 #define MAX_MTU 65535
1380
1381
1382 static void tun_net_init(struct net_device *dev)
1383 {
1384 struct tun_struct *tun = netdev_priv(dev);
1385
1386 switch (tun->flags & TUN_TYPE_MASK) {
1387 case IFF_TUN:
1388 dev->netdev_ops = &tun_netdev_ops;
1389
1390
1391 dev->hard_header_len = 0;
1392 dev->addr_len = 0;
1393 dev->mtu = 1500;
1394
1395
1396 dev->type = ARPHRD_NONE;
1397 dev->flags = IFF_POINTOPOINT | IFF_NOARP | IFF_MULTICAST;
1398 break;
1399
1400 case IFF_TAP:
1401 dev->netdev_ops = &tap_netdev_ops;
1402
1403 ether_setup(dev);
1404 dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1405 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
1406
1407 eth_hw_addr_random(dev);
1408
1409 break;
1410 }
1411
1412 dev->min_mtu = MIN_MTU;
1413 dev->max_mtu = MAX_MTU - dev->hard_header_len;
1414 }
1415
1416 static bool tun_sock_writeable(struct tun_struct *tun, struct tun_file *tfile)
1417 {
1418 struct sock *sk = tfile->socket.sk;
1419
1420 return (tun->dev->flags & IFF_UP) && sock_writeable(sk);
1421 }
1422
1423
1424
1425
1426 static __poll_t tun_chr_poll(struct file *file, poll_table *wait)
1427 {
1428 struct tun_file *tfile = file->private_data;
1429 struct tun_struct *tun = tun_get(tfile);
1430 struct sock *sk;
1431 __poll_t mask = 0;
1432
1433 if (!tun)
1434 return EPOLLERR;
1435
1436 sk = tfile->socket.sk;
1437
1438 tun_debug(KERN_INFO, tun, "tun_chr_poll\n");
1439
1440 poll_wait(file, sk_sleep(sk), wait);
1441
1442 if (!ptr_ring_empty(&tfile->tx_ring))
1443 mask |= EPOLLIN | EPOLLRDNORM;
1444
1445
1446
1447
1448
1449
1450 if (tun_sock_writeable(tun, tfile) ||
1451 (!test_and_set_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags) &&
1452 tun_sock_writeable(tun, tfile)))
1453 mask |= EPOLLOUT | EPOLLWRNORM;
1454
1455 if (tun->dev->reg_state != NETREG_REGISTERED)
1456 mask = EPOLLERR;
1457
1458 tun_put(tun);
1459 return mask;
1460 }
1461
1462 static struct sk_buff *tun_napi_alloc_frags(struct tun_file *tfile,
1463 size_t len,
1464 const struct iov_iter *it)
1465 {
1466 struct sk_buff *skb;
1467 size_t linear;
1468 int err;
1469 int i;
1470
1471 if (it->nr_segs > MAX_SKB_FRAGS + 1)
1472 return ERR_PTR(-ENOMEM);
1473
1474 local_bh_disable();
1475 skb = napi_get_frags(&tfile->napi);
1476 local_bh_enable();
1477 if (!skb)
1478 return ERR_PTR(-ENOMEM);
1479
1480 linear = iov_iter_single_seg_count(it);
1481 err = __skb_grow(skb, linear);
1482 if (err)
1483 goto free;
1484
1485 skb->len = len;
1486 skb->data_len = len - linear;
1487 skb->truesize += skb->data_len;
1488
1489 for (i = 1; i < it->nr_segs; i++) {
1490 size_t fragsz = it->iov[i].iov_len;
1491 struct page *page;
1492 void *frag;
1493
1494 if (fragsz == 0 || fragsz > PAGE_SIZE) {
1495 err = -EINVAL;
1496 goto free;
1497 }
1498 frag = netdev_alloc_frag(fragsz);
1499 if (!frag) {
1500 err = -ENOMEM;
1501 goto free;
1502 }
1503 page = virt_to_head_page(frag);
1504 skb_fill_page_desc(skb, i - 1, page,
1505 frag - page_address(page), fragsz);
1506 }
1507
1508 return skb;
1509 free:
1510
1511 napi_free_frags(&tfile->napi);
1512 return ERR_PTR(err);
1513 }
1514
1515
1516
1517 static struct sk_buff *tun_alloc_skb(struct tun_file *tfile,
1518 size_t prepad, size_t len,
1519 size_t linear, int noblock)
1520 {
1521 struct sock *sk = tfile->socket.sk;
1522 struct sk_buff *skb;
1523 int err;
1524
1525
1526 if (prepad + len < PAGE_SIZE || !linear)
1527 linear = len;
1528
1529 skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
1530 &err, 0);
1531 if (!skb)
1532 return ERR_PTR(err);
1533
1534 skb_reserve(skb, prepad);
1535 skb_put(skb, linear);
1536 skb->data_len = len - linear;
1537 skb->len += len - linear;
1538
1539 return skb;
1540 }
1541
1542 static void tun_rx_batched(struct tun_struct *tun, struct tun_file *tfile,
1543 struct sk_buff *skb, int more)
1544 {
1545 struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
1546 struct sk_buff_head process_queue;
1547 u32 rx_batched = tun->rx_batched;
1548 bool rcv = false;
1549
1550 if (!rx_batched || (!more && skb_queue_empty(queue))) {
1551 local_bh_disable();
1552 skb_record_rx_queue(skb, tfile->queue_index);
1553 netif_receive_skb(skb);
1554 local_bh_enable();
1555 return;
1556 }
1557
1558 spin_lock(&queue->lock);
1559 if (!more || skb_queue_len(queue) == rx_batched) {
1560 __skb_queue_head_init(&process_queue);
1561 skb_queue_splice_tail_init(queue, &process_queue);
1562 rcv = true;
1563 } else {
1564 __skb_queue_tail(queue, skb);
1565 }
1566 spin_unlock(&queue->lock);
1567
1568 if (rcv) {
1569 struct sk_buff *nskb;
1570
1571 local_bh_disable();
1572 while ((nskb = __skb_dequeue(&process_queue))) {
1573 skb_record_rx_queue(nskb, tfile->queue_index);
1574 netif_receive_skb(nskb);
1575 }
1576 skb_record_rx_queue(skb, tfile->queue_index);
1577 netif_receive_skb(skb);
1578 local_bh_enable();
1579 }
1580 }
1581
1582 static bool tun_can_build_skb(struct tun_struct *tun, struct tun_file *tfile,
1583 int len, int noblock, bool zerocopy)
1584 {
1585 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
1586 return false;
1587
1588 if (tfile->socket.sk->sk_sndbuf != INT_MAX)
1589 return false;
1590
1591 if (!noblock)
1592 return false;
1593
1594 if (zerocopy)
1595 return false;
1596
1597 if (SKB_DATA_ALIGN(len + TUN_RX_PAD) +
1598 SKB_DATA_ALIGN(sizeof(struct skb_shared_info)) > PAGE_SIZE)
1599 return false;
1600
1601 return true;
1602 }
1603
1604 static struct sk_buff *__tun_build_skb(struct tun_file *tfile,
1605 struct page_frag *alloc_frag, char *buf,
1606 int buflen, int len, int pad)
1607 {
1608 struct sk_buff *skb = build_skb(buf, buflen);
1609
1610 if (!skb)
1611 return ERR_PTR(-ENOMEM);
1612
1613 skb_reserve(skb, pad);
1614 skb_put(skb, len);
1615 skb_set_owner_w(skb, tfile->socket.sk);
1616
1617 get_page(alloc_frag->page);
1618 alloc_frag->offset += buflen;
1619
1620 return skb;
1621 }
1622
1623 static int tun_xdp_act(struct tun_struct *tun, struct bpf_prog *xdp_prog,
1624 struct xdp_buff *xdp, u32 act)
1625 {
1626 int err;
1627
1628 switch (act) {
1629 case XDP_REDIRECT:
1630 err = xdp_do_redirect(tun->dev, xdp, xdp_prog);
1631 if (err)
1632 return err;
1633 break;
1634 case XDP_TX:
1635 err = tun_xdp_tx(tun->dev, xdp);
1636 if (err < 0)
1637 return err;
1638 break;
1639 case XDP_PASS:
1640 break;
1641 default:
1642 bpf_warn_invalid_xdp_action(act);
1643
1644 case XDP_ABORTED:
1645 trace_xdp_exception(tun->dev, xdp_prog, act);
1646
1647 case XDP_DROP:
1648 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1649 break;
1650 }
1651
1652 return act;
1653 }
1654
1655 static struct sk_buff *tun_build_skb(struct tun_struct *tun,
1656 struct tun_file *tfile,
1657 struct iov_iter *from,
1658 struct virtio_net_hdr *hdr,
1659 int len, int *skb_xdp)
1660 {
1661 struct page_frag *alloc_frag = ¤t->task_frag;
1662 struct bpf_prog *xdp_prog;
1663 int buflen = SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
1664 char *buf;
1665 size_t copied;
1666 int pad = TUN_RX_PAD;
1667 int err = 0;
1668
1669 rcu_read_lock();
1670 xdp_prog = rcu_dereference(tun->xdp_prog);
1671 if (xdp_prog)
1672 pad += XDP_PACKET_HEADROOM;
1673 buflen += SKB_DATA_ALIGN(len + pad);
1674 rcu_read_unlock();
1675
1676 alloc_frag->offset = ALIGN((u64)alloc_frag->offset, SMP_CACHE_BYTES);
1677 if (unlikely(!skb_page_frag_refill(buflen, alloc_frag, GFP_KERNEL)))
1678 return ERR_PTR(-ENOMEM);
1679
1680 buf = (char *)page_address(alloc_frag->page) + alloc_frag->offset;
1681 copied = copy_page_from_iter(alloc_frag->page,
1682 alloc_frag->offset + pad,
1683 len, from);
1684 if (copied != len)
1685 return ERR_PTR(-EFAULT);
1686
1687
1688
1689
1690
1691 if (hdr->gso_type || !xdp_prog) {
1692 *skb_xdp = 1;
1693 return __tun_build_skb(tfile, alloc_frag, buf, buflen, len,
1694 pad);
1695 }
1696
1697 *skb_xdp = 0;
1698
1699 local_bh_disable();
1700 rcu_read_lock();
1701 xdp_prog = rcu_dereference(tun->xdp_prog);
1702 if (xdp_prog) {
1703 struct xdp_buff xdp;
1704 u32 act;
1705
1706 xdp.data_hard_start = buf;
1707 xdp.data = buf + pad;
1708 xdp_set_data_meta_invalid(&xdp);
1709 xdp.data_end = xdp.data + len;
1710 xdp.rxq = &tfile->xdp_rxq;
1711
1712 act = bpf_prog_run_xdp(xdp_prog, &xdp);
1713 if (act == XDP_REDIRECT || act == XDP_TX) {
1714 get_page(alloc_frag->page);
1715 alloc_frag->offset += buflen;
1716 }
1717 err = tun_xdp_act(tun, xdp_prog, &xdp, act);
1718 if (err < 0) {
1719 if (act == XDP_REDIRECT || act == XDP_TX)
1720 put_page(alloc_frag->page);
1721 goto out;
1722 }
1723
1724 if (err == XDP_REDIRECT)
1725 xdp_do_flush_map();
1726 if (err != XDP_PASS)
1727 goto out;
1728
1729 pad = xdp.data - xdp.data_hard_start;
1730 len = xdp.data_end - xdp.data;
1731 }
1732 rcu_read_unlock();
1733 local_bh_enable();
1734
1735 return __tun_build_skb(tfile, alloc_frag, buf, buflen, len, pad);
1736
1737 out:
1738 rcu_read_unlock();
1739 local_bh_enable();
1740 return NULL;
1741 }
1742
1743
1744 static ssize_t tun_get_user(struct tun_struct *tun, struct tun_file *tfile,
1745 void *msg_control, struct iov_iter *from,
1746 int noblock, bool more)
1747 {
1748 struct tun_pi pi = { 0, cpu_to_be16(ETH_P_IP) };
1749 struct sk_buff *skb;
1750 size_t total_len = iov_iter_count(from);
1751 size_t len = total_len, align = tun->align, linear;
1752 struct virtio_net_hdr gso = { 0 };
1753 struct tun_pcpu_stats *stats;
1754 int good_linear;
1755 int copylen;
1756 bool zerocopy = false;
1757 int err;
1758 u32 rxhash = 0;
1759 int skb_xdp = 1;
1760 bool frags = tun_napi_frags_enabled(tfile);
1761
1762 if (!(tun->flags & IFF_NO_PI)) {
1763 if (len < sizeof(pi))
1764 return -EINVAL;
1765 len -= sizeof(pi);
1766
1767 if (!copy_from_iter_full(&pi, sizeof(pi), from))
1768 return -EFAULT;
1769 }
1770
1771 if (tun->flags & IFF_VNET_HDR) {
1772 int vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
1773
1774 if (len < vnet_hdr_sz)
1775 return -EINVAL;
1776 len -= vnet_hdr_sz;
1777
1778 if (!copy_from_iter_full(&gso, sizeof(gso), from))
1779 return -EFAULT;
1780
1781 if ((gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
1782 tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2 > tun16_to_cpu(tun, gso.hdr_len))
1783 gso.hdr_len = cpu_to_tun16(tun, tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2);
1784
1785 if (tun16_to_cpu(tun, gso.hdr_len) > len)
1786 return -EINVAL;
1787 iov_iter_advance(from, vnet_hdr_sz - sizeof(gso));
1788 }
1789
1790 if ((tun->flags & TUN_TYPE_MASK) == IFF_TAP) {
1791 align += NET_IP_ALIGN;
1792 if (unlikely(len < ETH_HLEN ||
1793 (gso.hdr_len && tun16_to_cpu(tun, gso.hdr_len) < ETH_HLEN)))
1794 return -EINVAL;
1795 }
1796
1797 good_linear = SKB_MAX_HEAD(align);
1798
1799 if (msg_control) {
1800 struct iov_iter i = *from;
1801
1802
1803
1804
1805
1806 copylen = gso.hdr_len ? tun16_to_cpu(tun, gso.hdr_len) : GOODCOPY_LEN;
1807 if (copylen > good_linear)
1808 copylen = good_linear;
1809 linear = copylen;
1810 iov_iter_advance(&i, copylen);
1811 if (iov_iter_npages(&i, INT_MAX) <= MAX_SKB_FRAGS)
1812 zerocopy = true;
1813 }
1814
1815 if (!frags && tun_can_build_skb(tun, tfile, len, noblock, zerocopy)) {
1816
1817
1818
1819
1820 skb = tun_build_skb(tun, tfile, from, &gso, len, &skb_xdp);
1821 if (IS_ERR(skb)) {
1822 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1823 return PTR_ERR(skb);
1824 }
1825 if (!skb)
1826 return total_len;
1827 } else {
1828 if (!zerocopy) {
1829 copylen = len;
1830 if (tun16_to_cpu(tun, gso.hdr_len) > good_linear)
1831 linear = good_linear;
1832 else
1833 linear = tun16_to_cpu(tun, gso.hdr_len);
1834 }
1835
1836 if (frags) {
1837 mutex_lock(&tfile->napi_mutex);
1838 skb = tun_napi_alloc_frags(tfile, copylen, from);
1839
1840
1841
1842
1843 zerocopy = false;
1844 } else {
1845 skb = tun_alloc_skb(tfile, align, copylen, linear,
1846 noblock);
1847 }
1848
1849 if (IS_ERR(skb)) {
1850 if (PTR_ERR(skb) != -EAGAIN)
1851 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1852 if (frags)
1853 mutex_unlock(&tfile->napi_mutex);
1854 return PTR_ERR(skb);
1855 }
1856
1857 if (zerocopy)
1858 err = zerocopy_sg_from_iter(skb, from);
1859 else
1860 err = skb_copy_datagram_from_iter(skb, 0, from, len);
1861
1862 if (err) {
1863 err = -EFAULT;
1864 drop:
1865 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1866 kfree_skb(skb);
1867 if (frags) {
1868 tfile->napi.skb = NULL;
1869 mutex_unlock(&tfile->napi_mutex);
1870 }
1871
1872 return err;
1873 }
1874 }
1875
1876 if (virtio_net_hdr_to_skb(skb, &gso, tun_is_little_endian(tun))) {
1877 this_cpu_inc(tun->pcpu_stats->rx_frame_errors);
1878 kfree_skb(skb);
1879 if (frags) {
1880 tfile->napi.skb = NULL;
1881 mutex_unlock(&tfile->napi_mutex);
1882 }
1883
1884 return -EINVAL;
1885 }
1886
1887 switch (tun->flags & TUN_TYPE_MASK) {
1888 case IFF_TUN:
1889 if (tun->flags & IFF_NO_PI) {
1890 u8 ip_version = skb->len ? (skb->data[0] >> 4) : 0;
1891
1892 switch (ip_version) {
1893 case 4:
1894 pi.proto = htons(ETH_P_IP);
1895 break;
1896 case 6:
1897 pi.proto = htons(ETH_P_IPV6);
1898 break;
1899 default:
1900 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1901 kfree_skb(skb);
1902 return -EINVAL;
1903 }
1904 }
1905
1906 skb_reset_mac_header(skb);
1907 skb->protocol = pi.proto;
1908 skb->dev = tun->dev;
1909 break;
1910 case IFF_TAP:
1911 if (frags && !pskb_may_pull(skb, ETH_HLEN)) {
1912 err = -ENOMEM;
1913 goto drop;
1914 }
1915 skb->protocol = eth_type_trans(skb, tun->dev);
1916 break;
1917 }
1918
1919
1920 if (zerocopy) {
1921 skb_shinfo(skb)->destructor_arg = msg_control;
1922 skb_shinfo(skb)->tx_flags |= SKBTX_DEV_ZEROCOPY;
1923 skb_shinfo(skb)->tx_flags |= SKBTX_SHARED_FRAG;
1924 } else if (msg_control) {
1925 struct ubuf_info *uarg = msg_control;
1926 uarg->callback(uarg, false);
1927 }
1928
1929 skb_reset_network_header(skb);
1930 skb_probe_transport_header(skb);
1931 skb_record_rx_queue(skb, tfile->queue_index);
1932
1933 if (skb_xdp) {
1934 struct bpf_prog *xdp_prog;
1935 int ret;
1936
1937 local_bh_disable();
1938 rcu_read_lock();
1939 xdp_prog = rcu_dereference(tun->xdp_prog);
1940 if (xdp_prog) {
1941 ret = do_xdp_generic(xdp_prog, skb);
1942 if (ret != XDP_PASS) {
1943 rcu_read_unlock();
1944 local_bh_enable();
1945 if (frags) {
1946 tfile->napi.skb = NULL;
1947 mutex_unlock(&tfile->napi_mutex);
1948 }
1949 return total_len;
1950 }
1951 }
1952 rcu_read_unlock();
1953 local_bh_enable();
1954 }
1955
1956
1957
1958
1959
1960 if (!rcu_access_pointer(tun->steering_prog) && tun->numqueues > 1 &&
1961 !tfile->detached)
1962 rxhash = __skb_get_hash_symmetric(skb);
1963
1964 rcu_read_lock();
1965 if (unlikely(!(tun->dev->flags & IFF_UP))) {
1966 err = -EIO;
1967 rcu_read_unlock();
1968 goto drop;
1969 }
1970
1971 if (frags) {
1972 u32 headlen;
1973
1974
1975 skb_push(skb, ETH_HLEN);
1976 headlen = eth_get_headlen(tun->dev, skb->data,
1977 skb_headlen(skb));
1978
1979 if (unlikely(headlen > skb_headlen(skb))) {
1980 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1981 napi_free_frags(&tfile->napi);
1982 rcu_read_unlock();
1983 mutex_unlock(&tfile->napi_mutex);
1984 WARN_ON(1);
1985 return -ENOMEM;
1986 }
1987
1988 local_bh_disable();
1989 napi_gro_frags(&tfile->napi);
1990 local_bh_enable();
1991 mutex_unlock(&tfile->napi_mutex);
1992 } else if (tfile->napi_enabled) {
1993 struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
1994 int queue_len;
1995
1996 spin_lock_bh(&queue->lock);
1997 __skb_queue_tail(queue, skb);
1998 queue_len = skb_queue_len(queue);
1999 spin_unlock(&queue->lock);
2000
2001 if (!more || queue_len > NAPI_POLL_WEIGHT)
2002 napi_schedule(&tfile->napi);
2003
2004 local_bh_enable();
2005 } else if (!IS_ENABLED(CONFIG_4KSTACKS)) {
2006 tun_rx_batched(tun, tfile, skb, more);
2007 } else {
2008 netif_rx_ni(skb);
2009 }
2010 rcu_read_unlock();
2011
2012 stats = get_cpu_ptr(tun->pcpu_stats);
2013 u64_stats_update_begin(&stats->syncp);
2014 stats->rx_packets++;
2015 stats->rx_bytes += len;
2016 u64_stats_update_end(&stats->syncp);
2017 put_cpu_ptr(stats);
2018
2019 if (rxhash)
2020 tun_flow_update(tun, rxhash, tfile);
2021
2022 return total_len;
2023 }
2024
2025 static ssize_t tun_chr_write_iter(struct kiocb *iocb, struct iov_iter *from)
2026 {
2027 struct file *file = iocb->ki_filp;
2028 struct tun_file *tfile = file->private_data;
2029 struct tun_struct *tun = tun_get(tfile);
2030 ssize_t result;
2031
2032 if (!tun)
2033 return -EBADFD;
2034
2035 result = tun_get_user(tun, tfile, NULL, from,
2036 file->f_flags & O_NONBLOCK, false);
2037
2038 tun_put(tun);
2039 return result;
2040 }
2041
2042 static ssize_t tun_put_user_xdp(struct tun_struct *tun,
2043 struct tun_file *tfile,
2044 struct xdp_frame *xdp_frame,
2045 struct iov_iter *iter)
2046 {
2047 int vnet_hdr_sz = 0;
2048 size_t size = xdp_frame->len;
2049 struct tun_pcpu_stats *stats;
2050 size_t ret;
2051
2052 if (tun->flags & IFF_VNET_HDR) {
2053 struct virtio_net_hdr gso = { 0 };
2054
2055 vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
2056 if (unlikely(iov_iter_count(iter) < vnet_hdr_sz))
2057 return -EINVAL;
2058 if (unlikely(copy_to_iter(&gso, sizeof(gso), iter) !=
2059 sizeof(gso)))
2060 return -EFAULT;
2061 iov_iter_advance(iter, vnet_hdr_sz - sizeof(gso));
2062 }
2063
2064 ret = copy_to_iter(xdp_frame->data, size, iter) + vnet_hdr_sz;
2065
2066 stats = get_cpu_ptr(tun->pcpu_stats);
2067 u64_stats_update_begin(&stats->syncp);
2068 stats->tx_packets++;
2069 stats->tx_bytes += ret;
2070 u64_stats_update_end(&stats->syncp);
2071 put_cpu_ptr(tun->pcpu_stats);
2072
2073 return ret;
2074 }
2075
2076
2077 static ssize_t tun_put_user(struct tun_struct *tun,
2078 struct tun_file *tfile,
2079 struct sk_buff *skb,
2080 struct iov_iter *iter)
2081 {
2082 struct tun_pi pi = { 0, skb->protocol };
2083 struct tun_pcpu_stats *stats;
2084 ssize_t total;
2085 int vlan_offset = 0;
2086 int vlan_hlen = 0;
2087 int vnet_hdr_sz = 0;
2088
2089 if (skb_vlan_tag_present(skb))
2090 vlan_hlen = VLAN_HLEN;
2091
2092 if (tun->flags & IFF_VNET_HDR)
2093 vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
2094
2095 total = skb->len + vlan_hlen + vnet_hdr_sz;
2096
2097 if (!(tun->flags & IFF_NO_PI)) {
2098 if (iov_iter_count(iter) < sizeof(pi))
2099 return -EINVAL;
2100
2101 total += sizeof(pi);
2102 if (iov_iter_count(iter) < total) {
2103
2104 pi.flags |= TUN_PKT_STRIP;
2105 }
2106
2107 if (copy_to_iter(&pi, sizeof(pi), iter) != sizeof(pi))
2108 return -EFAULT;
2109 }
2110
2111 if (vnet_hdr_sz) {
2112 struct virtio_net_hdr gso;
2113
2114 if (iov_iter_count(iter) < vnet_hdr_sz)
2115 return -EINVAL;
2116
2117 if (virtio_net_hdr_from_skb(skb, &gso,
2118 tun_is_little_endian(tun), true,
2119 vlan_hlen)) {
2120 struct skb_shared_info *sinfo = skb_shinfo(skb);
2121 pr_err("unexpected GSO type: "
2122 "0x%x, gso_size %d, hdr_len %d\n",
2123 sinfo->gso_type, tun16_to_cpu(tun, gso.gso_size),
2124 tun16_to_cpu(tun, gso.hdr_len));
2125 print_hex_dump(KERN_ERR, "tun: ",
2126 DUMP_PREFIX_NONE,
2127 16, 1, skb->head,
2128 min((int)tun16_to_cpu(tun, gso.hdr_len), 64), true);
2129 WARN_ON_ONCE(1);
2130 return -EINVAL;
2131 }
2132
2133 if (copy_to_iter(&gso, sizeof(gso), iter) != sizeof(gso))
2134 return -EFAULT;
2135
2136 iov_iter_advance(iter, vnet_hdr_sz - sizeof(gso));
2137 }
2138
2139 if (vlan_hlen) {
2140 int ret;
2141 struct veth veth;
2142
2143 veth.h_vlan_proto = skb->vlan_proto;
2144 veth.h_vlan_TCI = htons(skb_vlan_tag_get(skb));
2145
2146 vlan_offset = offsetof(struct vlan_ethhdr, h_vlan_proto);
2147
2148 ret = skb_copy_datagram_iter(skb, 0, iter, vlan_offset);
2149 if (ret || !iov_iter_count(iter))
2150 goto done;
2151
2152 ret = copy_to_iter(&veth, sizeof(veth), iter);
2153 if (ret != sizeof(veth) || !iov_iter_count(iter))
2154 goto done;
2155 }
2156
2157 skb_copy_datagram_iter(skb, vlan_offset, iter, skb->len - vlan_offset);
2158
2159 done:
2160
2161 stats = get_cpu_ptr(tun->pcpu_stats);
2162 u64_stats_update_begin(&stats->syncp);
2163 stats->tx_packets++;
2164 stats->tx_bytes += skb->len + vlan_hlen;
2165 u64_stats_update_end(&stats->syncp);
2166 put_cpu_ptr(tun->pcpu_stats);
2167
2168 return total;
2169 }
2170
2171 static void *tun_ring_recv(struct tun_file *tfile, int noblock, int *err)
2172 {
2173 DECLARE_WAITQUEUE(wait, current);
2174 void *ptr = NULL;
2175 int error = 0;
2176
2177 ptr = ptr_ring_consume(&tfile->tx_ring);
2178 if (ptr)
2179 goto out;
2180 if (noblock) {
2181 error = -EAGAIN;
2182 goto out;
2183 }
2184
2185 add_wait_queue(&tfile->socket.wq.wait, &wait);
2186
2187 while (1) {
2188 set_current_state(TASK_INTERRUPTIBLE);
2189 ptr = ptr_ring_consume(&tfile->tx_ring);
2190 if (ptr)
2191 break;
2192 if (signal_pending(current)) {
2193 error = -ERESTARTSYS;
2194 break;
2195 }
2196 if (tfile->socket.sk->sk_shutdown & RCV_SHUTDOWN) {
2197 error = -EFAULT;
2198 break;
2199 }
2200
2201 schedule();
2202 }
2203
2204 __set_current_state(TASK_RUNNING);
2205 remove_wait_queue(&tfile->socket.wq.wait, &wait);
2206
2207 out:
2208 *err = error;
2209 return ptr;
2210 }
2211
2212 static ssize_t tun_do_read(struct tun_struct *tun, struct tun_file *tfile,
2213 struct iov_iter *to,
2214 int noblock, void *ptr)
2215 {
2216 ssize_t ret;
2217 int err;
2218
2219 tun_debug(KERN_INFO, tun, "tun_do_read\n");
2220
2221 if (!iov_iter_count(to)) {
2222 tun_ptr_free(ptr);
2223 return 0;
2224 }
2225
2226 if (!ptr) {
2227
2228 ptr = tun_ring_recv(tfile, noblock, &err);
2229 if (!ptr)
2230 return err;
2231 }
2232
2233 if (tun_is_xdp_frame(ptr)) {
2234 struct xdp_frame *xdpf = tun_ptr_to_xdp(ptr);
2235
2236 ret = tun_put_user_xdp(tun, tfile, xdpf, to);
2237 xdp_return_frame(xdpf);
2238 } else {
2239 struct sk_buff *skb = ptr;
2240
2241 ret = tun_put_user(tun, tfile, skb, to);
2242 if (unlikely(ret < 0))
2243 kfree_skb(skb);
2244 else
2245 consume_skb(skb);
2246 }
2247
2248 return ret;
2249 }
2250
2251 static ssize_t tun_chr_read_iter(struct kiocb *iocb, struct iov_iter *to)
2252 {
2253 struct file *file = iocb->ki_filp;
2254 struct tun_file *tfile = file->private_data;
2255 struct tun_struct *tun = tun_get(tfile);
2256 ssize_t len = iov_iter_count(to), ret;
2257
2258 if (!tun)
2259 return -EBADFD;
2260 ret = tun_do_read(tun, tfile, to, file->f_flags & O_NONBLOCK, NULL);
2261 ret = min_t(ssize_t, ret, len);
2262 if (ret > 0)
2263 iocb->ki_pos = ret;
2264 tun_put(tun);
2265 return ret;
2266 }
2267
2268 static void tun_prog_free(struct rcu_head *rcu)
2269 {
2270 struct tun_prog *prog = container_of(rcu, struct tun_prog, rcu);
2271
2272 bpf_prog_destroy(prog->prog);
2273 kfree(prog);
2274 }
2275
2276 static int __tun_set_ebpf(struct tun_struct *tun,
2277 struct tun_prog __rcu **prog_p,
2278 struct bpf_prog *prog)
2279 {
2280 struct tun_prog *old, *new = NULL;
2281
2282 if (prog) {
2283 new = kmalloc(sizeof(*new), GFP_KERNEL);
2284 if (!new)
2285 return -ENOMEM;
2286 new->prog = prog;
2287 }
2288
2289 spin_lock_bh(&tun->lock);
2290 old = rcu_dereference_protected(*prog_p,
2291 lockdep_is_held(&tun->lock));
2292 rcu_assign_pointer(*prog_p, new);
2293 spin_unlock_bh(&tun->lock);
2294
2295 if (old)
2296 call_rcu(&old->rcu, tun_prog_free);
2297
2298 return 0;
2299 }
2300
2301 static void tun_free_netdev(struct net_device *dev)
2302 {
2303 struct tun_struct *tun = netdev_priv(dev);
2304
2305 BUG_ON(!(list_empty(&tun->disabled)));
2306 free_percpu(tun->pcpu_stats);
2307 tun_flow_uninit(tun);
2308 security_tun_dev_free_security(tun->security);
2309 __tun_set_ebpf(tun, &tun->steering_prog, NULL);
2310 __tun_set_ebpf(tun, &tun->filter_prog, NULL);
2311 }
2312
2313 static void tun_setup(struct net_device *dev)
2314 {
2315 struct tun_struct *tun = netdev_priv(dev);
2316
2317 tun->owner = INVALID_UID;
2318 tun->group = INVALID_GID;
2319 tun_default_link_ksettings(dev, &tun->link_ksettings);
2320
2321 dev->ethtool_ops = &tun_ethtool_ops;
2322 dev->needs_free_netdev = true;
2323 dev->priv_destructor = tun_free_netdev;
2324
2325 dev->tx_queue_len = TUN_READQ_SIZE;
2326 }
2327
2328
2329
2330
2331 static int tun_validate(struct nlattr *tb[], struct nlattr *data[],
2332 struct netlink_ext_ack *extack)
2333 {
2334 NL_SET_ERR_MSG(extack,
2335 "tun/tap creation via rtnetlink is not supported.");
2336 return -EOPNOTSUPP;
2337 }
2338
2339 static size_t tun_get_size(const struct net_device *dev)
2340 {
2341 BUILD_BUG_ON(sizeof(u32) != sizeof(uid_t));
2342 BUILD_BUG_ON(sizeof(u32) != sizeof(gid_t));
2343
2344 return nla_total_size(sizeof(uid_t)) +
2345 nla_total_size(sizeof(gid_t)) +
2346 nla_total_size(sizeof(u8)) +
2347 nla_total_size(sizeof(u8)) +
2348 nla_total_size(sizeof(u8)) +
2349 nla_total_size(sizeof(u8)) +
2350 nla_total_size(sizeof(u8)) +
2351 nla_total_size(sizeof(u32)) +
2352 nla_total_size(sizeof(u32)) +
2353 0;
2354 }
2355
2356 static int tun_fill_info(struct sk_buff *skb, const struct net_device *dev)
2357 {
2358 struct tun_struct *tun = netdev_priv(dev);
2359
2360 if (nla_put_u8(skb, IFLA_TUN_TYPE, tun->flags & TUN_TYPE_MASK))
2361 goto nla_put_failure;
2362 if (uid_valid(tun->owner) &&
2363 nla_put_u32(skb, IFLA_TUN_OWNER,
2364 from_kuid_munged(current_user_ns(), tun->owner)))
2365 goto nla_put_failure;
2366 if (gid_valid(tun->group) &&
2367 nla_put_u32(skb, IFLA_TUN_GROUP,
2368 from_kgid_munged(current_user_ns(), tun->group)))
2369 goto nla_put_failure;
2370 if (nla_put_u8(skb, IFLA_TUN_PI, !(tun->flags & IFF_NO_PI)))
2371 goto nla_put_failure;
2372 if (nla_put_u8(skb, IFLA_TUN_VNET_HDR, !!(tun->flags & IFF_VNET_HDR)))
2373 goto nla_put_failure;
2374 if (nla_put_u8(skb, IFLA_TUN_PERSIST, !!(tun->flags & IFF_PERSIST)))
2375 goto nla_put_failure;
2376 if (nla_put_u8(skb, IFLA_TUN_MULTI_QUEUE,
2377 !!(tun->flags & IFF_MULTI_QUEUE)))
2378 goto nla_put_failure;
2379 if (tun->flags & IFF_MULTI_QUEUE) {
2380 if (nla_put_u32(skb, IFLA_TUN_NUM_QUEUES, tun->numqueues))
2381 goto nla_put_failure;
2382 if (nla_put_u32(skb, IFLA_TUN_NUM_DISABLED_QUEUES,
2383 tun->numdisabled))
2384 goto nla_put_failure;
2385 }
2386
2387 return 0;
2388
2389 nla_put_failure:
2390 return -EMSGSIZE;
2391 }
2392
2393 static struct rtnl_link_ops tun_link_ops __read_mostly = {
2394 .kind = DRV_NAME,
2395 .priv_size = sizeof(struct tun_struct),
2396 .setup = tun_setup,
2397 .validate = tun_validate,
2398 .get_size = tun_get_size,
2399 .fill_info = tun_fill_info,
2400 };
2401
2402 static void tun_sock_write_space(struct sock *sk)
2403 {
2404 struct tun_file *tfile;
2405 wait_queue_head_t *wqueue;
2406
2407 if (!sock_writeable(sk))
2408 return;
2409
2410 if (!test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags))
2411 return;
2412
2413 wqueue = sk_sleep(sk);
2414 if (wqueue && waitqueue_active(wqueue))
2415 wake_up_interruptible_sync_poll(wqueue, EPOLLOUT |
2416 EPOLLWRNORM | EPOLLWRBAND);
2417
2418 tfile = container_of(sk, struct tun_file, sk);
2419 kill_fasync(&tfile->fasync, SIGIO, POLL_OUT);
2420 }
2421
2422 static void tun_put_page(struct tun_page *tpage)
2423 {
2424 if (tpage->page)
2425 __page_frag_cache_drain(tpage->page, tpage->count);
2426 }
2427
2428 static int tun_xdp_one(struct tun_struct *tun,
2429 struct tun_file *tfile,
2430 struct xdp_buff *xdp, int *flush,
2431 struct tun_page *tpage)
2432 {
2433 unsigned int datasize = xdp->data_end - xdp->data;
2434 struct tun_xdp_hdr *hdr = xdp->data_hard_start;
2435 struct virtio_net_hdr *gso = &hdr->gso;
2436 struct tun_pcpu_stats *stats;
2437 struct bpf_prog *xdp_prog;
2438 struct sk_buff *skb = NULL;
2439 u32 rxhash = 0, act;
2440 int buflen = hdr->buflen;
2441 int err = 0;
2442 bool skb_xdp = false;
2443 struct page *page;
2444
2445 xdp_prog = rcu_dereference(tun->xdp_prog);
2446 if (xdp_prog) {
2447 if (gso->gso_type) {
2448 skb_xdp = true;
2449 goto build;
2450 }
2451 xdp_set_data_meta_invalid(xdp);
2452 xdp->rxq = &tfile->xdp_rxq;
2453
2454 act = bpf_prog_run_xdp(xdp_prog, xdp);
2455 err = tun_xdp_act(tun, xdp_prog, xdp, act);
2456 if (err < 0) {
2457 put_page(virt_to_head_page(xdp->data));
2458 return err;
2459 }
2460
2461 switch (err) {
2462 case XDP_REDIRECT:
2463 *flush = true;
2464
2465 case XDP_TX:
2466 return 0;
2467 case XDP_PASS:
2468 break;
2469 default:
2470 page = virt_to_head_page(xdp->data);
2471 if (tpage->page == page) {
2472 ++tpage->count;
2473 } else {
2474 tun_put_page(tpage);
2475 tpage->page = page;
2476 tpage->count = 1;
2477 }
2478 return 0;
2479 }
2480 }
2481
2482 build:
2483 skb = build_skb(xdp->data_hard_start, buflen);
2484 if (!skb) {
2485 err = -ENOMEM;
2486 goto out;
2487 }
2488
2489 skb_reserve(skb, xdp->data - xdp->data_hard_start);
2490 skb_put(skb, xdp->data_end - xdp->data);
2491
2492 if (virtio_net_hdr_to_skb(skb, gso, tun_is_little_endian(tun))) {
2493 this_cpu_inc(tun->pcpu_stats->rx_frame_errors);
2494 kfree_skb(skb);
2495 err = -EINVAL;
2496 goto out;
2497 }
2498
2499 skb->protocol = eth_type_trans(skb, tun->dev);
2500 skb_reset_network_header(skb);
2501 skb_probe_transport_header(skb);
2502 skb_record_rx_queue(skb, tfile->queue_index);
2503
2504 if (skb_xdp) {
2505 err = do_xdp_generic(xdp_prog, skb);
2506 if (err != XDP_PASS)
2507 goto out;
2508 }
2509
2510 if (!rcu_dereference(tun->steering_prog) && tun->numqueues > 1 &&
2511 !tfile->detached)
2512 rxhash = __skb_get_hash_symmetric(skb);
2513
2514 netif_receive_skb(skb);
2515
2516
2517
2518
2519 stats = this_cpu_ptr(tun->pcpu_stats);
2520 u64_stats_update_begin(&stats->syncp);
2521 stats->rx_packets++;
2522 stats->rx_bytes += datasize;
2523 u64_stats_update_end(&stats->syncp);
2524
2525 if (rxhash)
2526 tun_flow_update(tun, rxhash, tfile);
2527
2528 out:
2529 return err;
2530 }
2531
2532 static int tun_sendmsg(struct socket *sock, struct msghdr *m, size_t total_len)
2533 {
2534 int ret, i;
2535 struct tun_file *tfile = container_of(sock, struct tun_file, socket);
2536 struct tun_struct *tun = tun_get(tfile);
2537 struct tun_msg_ctl *ctl = m->msg_control;
2538 struct xdp_buff *xdp;
2539
2540 if (!tun)
2541 return -EBADFD;
2542
2543 if (ctl && (ctl->type == TUN_MSG_PTR)) {
2544 struct tun_page tpage;
2545 int n = ctl->num;
2546 int flush = 0;
2547
2548 memset(&tpage, 0, sizeof(tpage));
2549
2550 local_bh_disable();
2551 rcu_read_lock();
2552
2553 for (i = 0; i < n; i++) {
2554 xdp = &((struct xdp_buff *)ctl->ptr)[i];
2555 tun_xdp_one(tun, tfile, xdp, &flush, &tpage);
2556 }
2557
2558 if (flush)
2559 xdp_do_flush_map();
2560
2561 rcu_read_unlock();
2562 local_bh_enable();
2563
2564 tun_put_page(&tpage);
2565
2566 ret = total_len;
2567 goto out;
2568 }
2569
2570 ret = tun_get_user(tun, tfile, ctl ? ctl->ptr : NULL, &m->msg_iter,
2571 m->msg_flags & MSG_DONTWAIT,
2572 m->msg_flags & MSG_MORE);
2573 out:
2574 tun_put(tun);
2575 return ret;
2576 }
2577
2578 static int tun_recvmsg(struct socket *sock, struct msghdr *m, size_t total_len,
2579 int flags)
2580 {
2581 struct tun_file *tfile = container_of(sock, struct tun_file, socket);
2582 struct tun_struct *tun = tun_get(tfile);
2583 void *ptr = m->msg_control;
2584 int ret;
2585
2586 if (!tun) {
2587 ret = -EBADFD;
2588 goto out_free;
2589 }
2590
2591 if (flags & ~(MSG_DONTWAIT|MSG_TRUNC|MSG_ERRQUEUE)) {
2592 ret = -EINVAL;
2593 goto out_put_tun;
2594 }
2595 if (flags & MSG_ERRQUEUE) {
2596 ret = sock_recv_errqueue(sock->sk, m, total_len,
2597 SOL_PACKET, TUN_TX_TIMESTAMP);
2598 goto out;
2599 }
2600 ret = tun_do_read(tun, tfile, &m->msg_iter, flags & MSG_DONTWAIT, ptr);
2601 if (ret > (ssize_t)total_len) {
2602 m->msg_flags |= MSG_TRUNC;
2603 ret = flags & MSG_TRUNC ? ret : total_len;
2604 }
2605 out:
2606 tun_put(tun);
2607 return ret;
2608
2609 out_put_tun:
2610 tun_put(tun);
2611 out_free:
2612 tun_ptr_free(ptr);
2613 return ret;
2614 }
2615
2616 static int tun_ptr_peek_len(void *ptr)
2617 {
2618 if (likely(ptr)) {
2619 if (tun_is_xdp_frame(ptr)) {
2620 struct xdp_frame *xdpf = tun_ptr_to_xdp(ptr);
2621
2622 return xdpf->len;
2623 }
2624 return __skb_array_len_with_tag(ptr);
2625 } else {
2626 return 0;
2627 }
2628 }
2629
2630 static int tun_peek_len(struct socket *sock)
2631 {
2632 struct tun_file *tfile = container_of(sock, struct tun_file, socket);
2633 struct tun_struct *tun;
2634 int ret = 0;
2635
2636 tun = tun_get(tfile);
2637 if (!tun)
2638 return 0;
2639
2640 ret = PTR_RING_PEEK_CALL(&tfile->tx_ring, tun_ptr_peek_len);
2641 tun_put(tun);
2642
2643 return ret;
2644 }
2645
2646
2647 static const struct proto_ops tun_socket_ops = {
2648 .peek_len = tun_peek_len,
2649 .sendmsg = tun_sendmsg,
2650 .recvmsg = tun_recvmsg,
2651 };
2652
2653 static struct proto tun_proto = {
2654 .name = "tun",
2655 .owner = THIS_MODULE,
2656 .obj_size = sizeof(struct tun_file),
2657 };
2658
2659 static int tun_flags(struct tun_struct *tun)
2660 {
2661 return tun->flags & (TUN_FEATURES | IFF_PERSIST | IFF_TUN | IFF_TAP);
2662 }
2663
2664 static ssize_t tun_show_flags(struct device *dev, struct device_attribute *attr,
2665 char *buf)
2666 {
2667 struct tun_struct *tun = netdev_priv(to_net_dev(dev));
2668 return sprintf(buf, "0x%x\n", tun_flags(tun));
2669 }
2670
2671 static ssize_t tun_show_owner(struct device *dev, struct device_attribute *attr,
2672 char *buf)
2673 {
2674 struct tun_struct *tun = netdev_priv(to_net_dev(dev));
2675 return uid_valid(tun->owner)?
2676 sprintf(buf, "%u\n",
2677 from_kuid_munged(current_user_ns(), tun->owner)):
2678 sprintf(buf, "-1\n");
2679 }
2680
2681 static ssize_t tun_show_group(struct device *dev, struct device_attribute *attr,
2682 char *buf)
2683 {
2684 struct tun_struct *tun = netdev_priv(to_net_dev(dev));
2685 return gid_valid(tun->group) ?
2686 sprintf(buf, "%u\n",
2687 from_kgid_munged(current_user_ns(), tun->group)):
2688 sprintf(buf, "-1\n");
2689 }
2690
2691 static DEVICE_ATTR(tun_flags, 0444, tun_show_flags, NULL);
2692 static DEVICE_ATTR(owner, 0444, tun_show_owner, NULL);
2693 static DEVICE_ATTR(group, 0444, tun_show_group, NULL);
2694
2695 static struct attribute *tun_dev_attrs[] = {
2696 &dev_attr_tun_flags.attr,
2697 &dev_attr_owner.attr,
2698 &dev_attr_group.attr,
2699 NULL
2700 };
2701
2702 static const struct attribute_group tun_attr_group = {
2703 .attrs = tun_dev_attrs
2704 };
2705
2706 static int tun_set_iff(struct net *net, struct file *file, struct ifreq *ifr)
2707 {
2708 struct tun_struct *tun;
2709 struct tun_file *tfile = file->private_data;
2710 struct net_device *dev;
2711 int err;
2712
2713 if (tfile->detached)
2714 return -EINVAL;
2715
2716 if ((ifr->ifr_flags & IFF_NAPI_FRAGS)) {
2717 if (!capable(CAP_NET_ADMIN))
2718 return -EPERM;
2719
2720 if (!(ifr->ifr_flags & IFF_NAPI) ||
2721 (ifr->ifr_flags & TUN_TYPE_MASK) != IFF_TAP)
2722 return -EINVAL;
2723 }
2724
2725 dev = __dev_get_by_name(net, ifr->ifr_name);
2726 if (dev) {
2727 if (ifr->ifr_flags & IFF_TUN_EXCL)
2728 return -EBUSY;
2729 if ((ifr->ifr_flags & IFF_TUN) && dev->netdev_ops == &tun_netdev_ops)
2730 tun = netdev_priv(dev);
2731 else if ((ifr->ifr_flags & IFF_TAP) && dev->netdev_ops == &tap_netdev_ops)
2732 tun = netdev_priv(dev);
2733 else
2734 return -EINVAL;
2735
2736 if (!!(ifr->ifr_flags & IFF_MULTI_QUEUE) !=
2737 !!(tun->flags & IFF_MULTI_QUEUE))
2738 return -EINVAL;
2739
2740 if (tun_not_capable(tun))
2741 return -EPERM;
2742 err = security_tun_dev_open(tun->security);
2743 if (err < 0)
2744 return err;
2745
2746 err = tun_attach(tun, file, ifr->ifr_flags & IFF_NOFILTER,
2747 ifr->ifr_flags & IFF_NAPI,
2748 ifr->ifr_flags & IFF_NAPI_FRAGS, true);
2749 if (err < 0)
2750 return err;
2751
2752 if (tun->flags & IFF_MULTI_QUEUE &&
2753 (tun->numqueues + tun->numdisabled > 1)) {
2754
2755
2756
2757 netdev_state_change(dev);
2758 return 0;
2759 }
2760
2761 tun->flags = (tun->flags & ~TUN_FEATURES) |
2762 (ifr->ifr_flags & TUN_FEATURES);
2763
2764 netdev_state_change(dev);
2765 } else {
2766 char *name;
2767 unsigned long flags = 0;
2768 int queues = ifr->ifr_flags & IFF_MULTI_QUEUE ?
2769 MAX_TAP_QUEUES : 1;
2770
2771 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
2772 return -EPERM;
2773 err = security_tun_dev_create();
2774 if (err < 0)
2775 return err;
2776
2777
2778 if (ifr->ifr_flags & IFF_TUN) {
2779
2780 flags |= IFF_TUN;
2781 name = "tun%d";
2782 } else if (ifr->ifr_flags & IFF_TAP) {
2783
2784 flags |= IFF_TAP;
2785 name = "tap%d";
2786 } else
2787 return -EINVAL;
2788
2789 if (*ifr->ifr_name)
2790 name = ifr->ifr_name;
2791
2792 dev = alloc_netdev_mqs(sizeof(struct tun_struct), name,
2793 NET_NAME_UNKNOWN, tun_setup, queues,
2794 queues);
2795
2796 if (!dev)
2797 return -ENOMEM;
2798 err = dev_get_valid_name(net, dev, name);
2799 if (err < 0)
2800 goto err_free_dev;
2801
2802 dev_net_set(dev, net);
2803 dev->rtnl_link_ops = &tun_link_ops;
2804 dev->ifindex = tfile->ifindex;
2805 dev->sysfs_groups[0] = &tun_attr_group;
2806
2807 tun = netdev_priv(dev);
2808 tun->dev = dev;
2809 tun->flags = flags;
2810 tun->txflt.count = 0;
2811 tun->vnet_hdr_sz = sizeof(struct virtio_net_hdr);
2812
2813 tun->align = NET_SKB_PAD;
2814 tun->filter_attached = false;
2815 tun->sndbuf = tfile->socket.sk->sk_sndbuf;
2816 tun->rx_batched = 0;
2817 RCU_INIT_POINTER(tun->steering_prog, NULL);
2818
2819 tun->pcpu_stats = netdev_alloc_pcpu_stats(struct tun_pcpu_stats);
2820 if (!tun->pcpu_stats) {
2821 err = -ENOMEM;
2822 goto err_free_dev;
2823 }
2824
2825 spin_lock_init(&tun->lock);
2826
2827 err = security_tun_dev_alloc_security(&tun->security);
2828 if (err < 0)
2829 goto err_free_stat;
2830
2831 tun_net_init(dev);
2832 tun_flow_init(tun);
2833
2834 dev->hw_features = NETIF_F_SG | NETIF_F_FRAGLIST |
2835 TUN_USER_FEATURES | NETIF_F_HW_VLAN_CTAG_TX |
2836 NETIF_F_HW_VLAN_STAG_TX;
2837 dev->features = dev->hw_features | NETIF_F_LLTX;
2838 dev->vlan_features = dev->features &
2839 ~(NETIF_F_HW_VLAN_CTAG_TX |
2840 NETIF_F_HW_VLAN_STAG_TX);
2841
2842 tun->flags = (tun->flags & ~TUN_FEATURES) |
2843 (ifr->ifr_flags & TUN_FEATURES);
2844
2845 INIT_LIST_HEAD(&tun->disabled);
2846 err = tun_attach(tun, file, false, ifr->ifr_flags & IFF_NAPI,
2847 ifr->ifr_flags & IFF_NAPI_FRAGS, false);
2848 if (err < 0)
2849 goto err_free_flow;
2850
2851 err = register_netdevice(tun->dev);
2852 if (err < 0)
2853 goto err_detach;
2854
2855
2856
2857 rcu_assign_pointer(tfile->tun, tun);
2858 }
2859
2860 netif_carrier_on(tun->dev);
2861
2862 tun_debug(KERN_INFO, tun, "tun_set_iff\n");
2863
2864
2865
2866
2867 if (netif_running(tun->dev))
2868 netif_tx_wake_all_queues(tun->dev);
2869
2870 strcpy(ifr->ifr_name, tun->dev->name);
2871 return 0;
2872
2873 err_detach:
2874 tun_detach_all(dev);
2875
2876 goto err_free_dev;
2877
2878 err_free_flow:
2879 tun_flow_uninit(tun);
2880 security_tun_dev_free_security(tun->security);
2881 err_free_stat:
2882 free_percpu(tun->pcpu_stats);
2883 err_free_dev:
2884 free_netdev(dev);
2885 return err;
2886 }
2887
2888 static void tun_get_iff(struct tun_struct *tun, struct ifreq *ifr)
2889 {
2890 tun_debug(KERN_INFO, tun, "tun_get_iff\n");
2891
2892 strcpy(ifr->ifr_name, tun->dev->name);
2893
2894 ifr->ifr_flags = tun_flags(tun);
2895
2896 }
2897
2898
2899
2900 static int set_offload(struct tun_struct *tun, unsigned long arg)
2901 {
2902 netdev_features_t features = 0;
2903
2904 if (arg & TUN_F_CSUM) {
2905 features |= NETIF_F_HW_CSUM;
2906 arg &= ~TUN_F_CSUM;
2907
2908 if (arg & (TUN_F_TSO4|TUN_F_TSO6)) {
2909 if (arg & TUN_F_TSO_ECN) {
2910 features |= NETIF_F_TSO_ECN;
2911 arg &= ~TUN_F_TSO_ECN;
2912 }
2913 if (arg & TUN_F_TSO4)
2914 features |= NETIF_F_TSO;
2915 if (arg & TUN_F_TSO6)
2916 features |= NETIF_F_TSO6;
2917 arg &= ~(TUN_F_TSO4|TUN_F_TSO6);
2918 }
2919
2920 arg &= ~TUN_F_UFO;
2921 }
2922
2923
2924
2925 if (arg)
2926 return -EINVAL;
2927
2928 tun->set_features = features;
2929 tun->dev->wanted_features &= ~TUN_USER_FEATURES;
2930 tun->dev->wanted_features |= features;
2931 netdev_update_features(tun->dev);
2932
2933 return 0;
2934 }
2935
2936 static void tun_detach_filter(struct tun_struct *tun, int n)
2937 {
2938 int i;
2939 struct tun_file *tfile;
2940
2941 for (i = 0; i < n; i++) {
2942 tfile = rtnl_dereference(tun->tfiles[i]);
2943 lock_sock(tfile->socket.sk);
2944 sk_detach_filter(tfile->socket.sk);
2945 release_sock(tfile->socket.sk);
2946 }
2947
2948 tun->filter_attached = false;
2949 }
2950
2951 static int tun_attach_filter(struct tun_struct *tun)
2952 {
2953 int i, ret = 0;
2954 struct tun_file *tfile;
2955
2956 for (i = 0; i < tun->numqueues; i++) {
2957 tfile = rtnl_dereference(tun->tfiles[i]);
2958 lock_sock(tfile->socket.sk);
2959 ret = sk_attach_filter(&tun->fprog, tfile->socket.sk);
2960 release_sock(tfile->socket.sk);
2961 if (ret) {
2962 tun_detach_filter(tun, i);
2963 return ret;
2964 }
2965 }
2966
2967 tun->filter_attached = true;
2968 return ret;
2969 }
2970
2971 static void tun_set_sndbuf(struct tun_struct *tun)
2972 {
2973 struct tun_file *tfile;
2974 int i;
2975
2976 for (i = 0; i < tun->numqueues; i++) {
2977 tfile = rtnl_dereference(tun->tfiles[i]);
2978 tfile->socket.sk->sk_sndbuf = tun->sndbuf;
2979 }
2980 }
2981
2982 static int tun_set_queue(struct file *file, struct ifreq *ifr)
2983 {
2984 struct tun_file *tfile = file->private_data;
2985 struct tun_struct *tun;
2986 int ret = 0;
2987
2988 rtnl_lock();
2989
2990 if (ifr->ifr_flags & IFF_ATTACH_QUEUE) {
2991 tun = tfile->detached;
2992 if (!tun) {
2993 ret = -EINVAL;
2994 goto unlock;
2995 }
2996 ret = security_tun_dev_attach_queue(tun->security);
2997 if (ret < 0)
2998 goto unlock;
2999 ret = tun_attach(tun, file, false, tun->flags & IFF_NAPI,
3000 tun->flags & IFF_NAPI_FRAGS, true);
3001 } else if (ifr->ifr_flags & IFF_DETACH_QUEUE) {
3002 tun = rtnl_dereference(tfile->tun);
3003 if (!tun || !(tun->flags & IFF_MULTI_QUEUE) || tfile->detached)
3004 ret = -EINVAL;
3005 else
3006 __tun_detach(tfile, false);
3007 } else
3008 ret = -EINVAL;
3009
3010 if (ret >= 0)
3011 netdev_state_change(tun->dev);
3012
3013 unlock:
3014 rtnl_unlock();
3015 return ret;
3016 }
3017
3018 static int tun_set_ebpf(struct tun_struct *tun, struct tun_prog **prog_p,
3019 void __user *data)
3020 {
3021 struct bpf_prog *prog;
3022 int fd;
3023
3024 if (copy_from_user(&fd, data, sizeof(fd)))
3025 return -EFAULT;
3026
3027 if (fd == -1) {
3028 prog = NULL;
3029 } else {
3030 prog = bpf_prog_get_type(fd, BPF_PROG_TYPE_SOCKET_FILTER);
3031 if (IS_ERR(prog))
3032 return PTR_ERR(prog);
3033 }
3034
3035 return __tun_set_ebpf(tun, prog_p, prog);
3036 }
3037
3038 static long __tun_chr_ioctl(struct file *file, unsigned int cmd,
3039 unsigned long arg, int ifreq_len)
3040 {
3041 struct tun_file *tfile = file->private_data;
3042 struct net *net = sock_net(&tfile->sk);
3043 struct tun_struct *tun;
3044 void __user* argp = (void __user*)arg;
3045 unsigned int ifindex, carrier;
3046 struct ifreq ifr;
3047 kuid_t owner;
3048 kgid_t group;
3049 int sndbuf;
3050 int vnet_hdr_sz;
3051 int le;
3052 int ret;
3053 bool do_notify = false;
3054
3055 if (cmd == TUNSETIFF || cmd == TUNSETQUEUE ||
3056 (_IOC_TYPE(cmd) == SOCK_IOC_TYPE && cmd != SIOCGSKNS)) {
3057 if (copy_from_user(&ifr, argp, ifreq_len))
3058 return -EFAULT;
3059 } else {
3060 memset(&ifr, 0, sizeof(ifr));
3061 }
3062 if (cmd == TUNGETFEATURES) {
3063
3064
3065
3066
3067 return put_user(IFF_TUN | IFF_TAP | TUN_FEATURES,
3068 (unsigned int __user*)argp);
3069 } else if (cmd == TUNSETQUEUE) {
3070 return tun_set_queue(file, &ifr);
3071 } else if (cmd == SIOCGSKNS) {
3072 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
3073 return -EPERM;
3074 return open_related_ns(&net->ns, get_net_ns);
3075 }
3076
3077 ret = 0;
3078 rtnl_lock();
3079
3080 tun = tun_get(tfile);
3081 if (cmd == TUNSETIFF) {
3082 ret = -EEXIST;
3083 if (tun)
3084 goto unlock;
3085
3086 ifr.ifr_name[IFNAMSIZ-1] = '\0';
3087
3088 ret = tun_set_iff(net, file, &ifr);
3089
3090 if (ret)
3091 goto unlock;
3092
3093 if (copy_to_user(argp, &ifr, ifreq_len))
3094 ret = -EFAULT;
3095 goto unlock;
3096 }
3097 if (cmd == TUNSETIFINDEX) {
3098 ret = -EPERM;
3099 if (tun)
3100 goto unlock;
3101
3102 ret = -EFAULT;
3103 if (copy_from_user(&ifindex, argp, sizeof(ifindex)))
3104 goto unlock;
3105
3106 ret = 0;
3107 tfile->ifindex = ifindex;
3108 goto unlock;
3109 }
3110
3111 ret = -EBADFD;
3112 if (!tun)
3113 goto unlock;
3114
3115 tun_debug(KERN_INFO, tun, "tun_chr_ioctl cmd %u\n", cmd);
3116
3117 net = dev_net(tun->dev);
3118 ret = 0;
3119 switch (cmd) {
3120 case TUNGETIFF:
3121 tun_get_iff(tun, &ifr);
3122
3123 if (tfile->detached)
3124 ifr.ifr_flags |= IFF_DETACH_QUEUE;
3125 if (!tfile->socket.sk->sk_filter)
3126 ifr.ifr_flags |= IFF_NOFILTER;
3127
3128 if (copy_to_user(argp, &ifr, ifreq_len))
3129 ret = -EFAULT;
3130 break;
3131
3132 case TUNSETNOCSUM:
3133
3134
3135
3136 tun_debug(KERN_INFO, tun, "ignored: set checksum %s\n",
3137 arg ? "disabled" : "enabled");
3138 break;
3139
3140 case TUNSETPERSIST:
3141
3142
3143
3144 if (arg && !(tun->flags & IFF_PERSIST)) {
3145 tun->flags |= IFF_PERSIST;
3146 __module_get(THIS_MODULE);
3147 do_notify = true;
3148 }
3149 if (!arg && (tun->flags & IFF_PERSIST)) {
3150 tun->flags &= ~IFF_PERSIST;
3151 module_put(THIS_MODULE);
3152 do_notify = true;
3153 }
3154
3155 tun_debug(KERN_INFO, tun, "persist %s\n",
3156 arg ? "enabled" : "disabled");
3157 break;
3158
3159 case TUNSETOWNER:
3160
3161 owner = make_kuid(current_user_ns(), arg);
3162 if (!uid_valid(owner)) {
3163 ret = -EINVAL;
3164 break;
3165 }
3166 tun->owner = owner;
3167 do_notify = true;
3168 tun_debug(KERN_INFO, tun, "owner set to %u\n",
3169 from_kuid(&init_user_ns, tun->owner));
3170 break;
3171
3172 case TUNSETGROUP:
3173
3174 group = make_kgid(current_user_ns(), arg);
3175 if (!gid_valid(group)) {
3176 ret = -EINVAL;
3177 break;
3178 }
3179 tun->group = group;
3180 do_notify = true;
3181 tun_debug(KERN_INFO, tun, "group set to %u\n",
3182 from_kgid(&init_user_ns, tun->group));
3183 break;
3184
3185 case TUNSETLINK:
3186
3187 if (tun->dev->flags & IFF_UP) {
3188 tun_debug(KERN_INFO, tun,
3189 "Linktype set failed because interface is up\n");
3190 ret = -EBUSY;
3191 } else {
3192 tun->dev->type = (int) arg;
3193 tun_debug(KERN_INFO, tun, "linktype set to %d\n",
3194 tun->dev->type);
3195 ret = 0;
3196 }
3197 break;
3198
3199 #ifdef TUN_DEBUG
3200 case TUNSETDEBUG:
3201 tun->debug = arg;
3202 break;
3203 #endif
3204 case TUNSETOFFLOAD:
3205 ret = set_offload(tun, arg);
3206 break;
3207
3208 case TUNSETTXFILTER:
3209
3210 ret = -EINVAL;
3211 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3212 break;
3213 ret = update_filter(&tun->txflt, (void __user *)arg);
3214 break;
3215
3216 case SIOCGIFHWADDR:
3217
3218 memcpy(ifr.ifr_hwaddr.sa_data, tun->dev->dev_addr, ETH_ALEN);
3219 ifr.ifr_hwaddr.sa_family = tun->dev->type;
3220 if (copy_to_user(argp, &ifr, ifreq_len))
3221 ret = -EFAULT;
3222 break;
3223
3224 case SIOCSIFHWADDR:
3225
3226 tun_debug(KERN_DEBUG, tun, "set hw address: %pM\n",
3227 ifr.ifr_hwaddr.sa_data);
3228
3229 ret = dev_set_mac_address(tun->dev, &ifr.ifr_hwaddr, NULL);
3230 break;
3231
3232 case TUNGETSNDBUF:
3233 sndbuf = tfile->socket.sk->sk_sndbuf;
3234 if (copy_to_user(argp, &sndbuf, sizeof(sndbuf)))
3235 ret = -EFAULT;
3236 break;
3237
3238 case TUNSETSNDBUF:
3239 if (copy_from_user(&sndbuf, argp, sizeof(sndbuf))) {
3240 ret = -EFAULT;
3241 break;
3242 }
3243 if (sndbuf <= 0) {
3244 ret = -EINVAL;
3245 break;
3246 }
3247
3248 tun->sndbuf = sndbuf;
3249 tun_set_sndbuf(tun);
3250 break;
3251
3252 case TUNGETVNETHDRSZ:
3253 vnet_hdr_sz = tun->vnet_hdr_sz;
3254 if (copy_to_user(argp, &vnet_hdr_sz, sizeof(vnet_hdr_sz)))
3255 ret = -EFAULT;
3256 break;
3257
3258 case TUNSETVNETHDRSZ:
3259 if (copy_from_user(&vnet_hdr_sz, argp, sizeof(vnet_hdr_sz))) {
3260 ret = -EFAULT;
3261 break;
3262 }
3263 if (vnet_hdr_sz < (int)sizeof(struct virtio_net_hdr)) {
3264 ret = -EINVAL;
3265 break;
3266 }
3267
3268 tun->vnet_hdr_sz = vnet_hdr_sz;
3269 break;
3270
3271 case TUNGETVNETLE:
3272 le = !!(tun->flags & TUN_VNET_LE);
3273 if (put_user(le, (int __user *)argp))
3274 ret = -EFAULT;
3275 break;
3276
3277 case TUNSETVNETLE:
3278 if (get_user(le, (int __user *)argp)) {
3279 ret = -EFAULT;
3280 break;
3281 }
3282 if (le)
3283 tun->flags |= TUN_VNET_LE;
3284 else
3285 tun->flags &= ~TUN_VNET_LE;
3286 break;
3287
3288 case TUNGETVNETBE:
3289 ret = tun_get_vnet_be(tun, argp);
3290 break;
3291
3292 case TUNSETVNETBE:
3293 ret = tun_set_vnet_be(tun, argp);
3294 break;
3295
3296 case TUNATTACHFILTER:
3297
3298 ret = -EINVAL;
3299 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3300 break;
3301 ret = -EFAULT;
3302 if (copy_from_user(&tun->fprog, argp, sizeof(tun->fprog)))
3303 break;
3304
3305 ret = tun_attach_filter(tun);
3306 break;
3307
3308 case TUNDETACHFILTER:
3309
3310 ret = -EINVAL;
3311 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3312 break;
3313 ret = 0;
3314 tun_detach_filter(tun, tun->numqueues);
3315 break;
3316
3317 case TUNGETFILTER:
3318 ret = -EINVAL;
3319 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3320 break;
3321 ret = -EFAULT;
3322 if (copy_to_user(argp, &tun->fprog, sizeof(tun->fprog)))
3323 break;
3324 ret = 0;
3325 break;
3326
3327 case TUNSETSTEERINGEBPF:
3328 ret = tun_set_ebpf(tun, &tun->steering_prog, argp);
3329 break;
3330
3331 case TUNSETFILTEREBPF:
3332 ret = tun_set_ebpf(tun, &tun->filter_prog, argp);
3333 break;
3334
3335 case TUNSETCARRIER:
3336 ret = -EFAULT;
3337 if (copy_from_user(&carrier, argp, sizeof(carrier)))
3338 goto unlock;
3339
3340 ret = tun_net_change_carrier(tun->dev, (bool)carrier);
3341 break;
3342
3343 case TUNGETDEVNETNS:
3344 ret = -EPERM;
3345 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
3346 goto unlock;
3347 ret = open_related_ns(&net->ns, get_net_ns);
3348 break;
3349
3350 default:
3351 ret = -EINVAL;
3352 break;
3353 }
3354
3355 if (do_notify)
3356 netdev_state_change(tun->dev);
3357
3358 unlock:
3359 rtnl_unlock();
3360 if (tun)
3361 tun_put(tun);
3362 return ret;
3363 }
3364
3365 static long tun_chr_ioctl(struct file *file,
3366 unsigned int cmd, unsigned long arg)
3367 {
3368 return __tun_chr_ioctl(file, cmd, arg, sizeof (struct ifreq));
3369 }
3370
3371 #ifdef CONFIG_COMPAT
3372 static long tun_chr_compat_ioctl(struct file *file,
3373 unsigned int cmd, unsigned long arg)
3374 {
3375 switch (cmd) {
3376 case TUNSETIFF:
3377 case TUNGETIFF:
3378 case TUNSETTXFILTER:
3379 case TUNGETSNDBUF:
3380 case TUNSETSNDBUF:
3381 case SIOCGIFHWADDR:
3382 case SIOCSIFHWADDR:
3383 arg = (unsigned long)compat_ptr(arg);
3384 break;
3385 default:
3386 arg = (compat_ulong_t)arg;
3387 break;
3388 }
3389
3390
3391
3392
3393
3394
3395
3396 return __tun_chr_ioctl(file, cmd, arg, sizeof(struct compat_ifreq));
3397 }
3398 #endif
3399
3400 static int tun_chr_fasync(int fd, struct file *file, int on)
3401 {
3402 struct tun_file *tfile = file->private_data;
3403 int ret;
3404
3405 if ((ret = fasync_helper(fd, file, on, &tfile->fasync)) < 0)
3406 goto out;
3407
3408 if (on) {
3409 __f_setown(file, task_pid(current), PIDTYPE_TGID, 0);
3410 tfile->flags |= TUN_FASYNC;
3411 } else
3412 tfile->flags &= ~TUN_FASYNC;
3413 ret = 0;
3414 out:
3415 return ret;
3416 }
3417
3418 static int tun_chr_open(struct inode *inode, struct file * file)
3419 {
3420 struct net *net = current->nsproxy->net_ns;
3421 struct tun_file *tfile;
3422
3423 DBG1(KERN_INFO, "tunX: tun_chr_open\n");
3424
3425 tfile = (struct tun_file *)sk_alloc(net, AF_UNSPEC, GFP_KERNEL,
3426 &tun_proto, 0);
3427 if (!tfile)
3428 return -ENOMEM;
3429 if (ptr_ring_init(&tfile->tx_ring, 0, GFP_KERNEL)) {
3430 sk_free(&tfile->sk);
3431 return -ENOMEM;
3432 }
3433
3434 mutex_init(&tfile->napi_mutex);
3435 RCU_INIT_POINTER(tfile->tun, NULL);
3436 tfile->flags = 0;
3437 tfile->ifindex = 0;
3438
3439 init_waitqueue_head(&tfile->socket.wq.wait);
3440
3441 tfile->socket.file = file;
3442 tfile->socket.ops = &tun_socket_ops;
3443
3444 sock_init_data(&tfile->socket, &tfile->sk);
3445
3446 tfile->sk.sk_write_space = tun_sock_write_space;
3447 tfile->sk.sk_sndbuf = INT_MAX;
3448
3449 file->private_data = tfile;
3450 INIT_LIST_HEAD(&tfile->next);
3451
3452 sock_set_flag(&tfile->sk, SOCK_ZEROCOPY);
3453
3454 return 0;
3455 }
3456
3457 static int tun_chr_close(struct inode *inode, struct file *file)
3458 {
3459 struct tun_file *tfile = file->private_data;
3460
3461 tun_detach(tfile, true);
3462
3463 return 0;
3464 }
3465
3466 #ifdef CONFIG_PROC_FS
3467 static void tun_chr_show_fdinfo(struct seq_file *m, struct file *file)
3468 {
3469 struct tun_file *tfile = file->private_data;
3470 struct tun_struct *tun;
3471 struct ifreq ifr;
3472
3473 memset(&ifr, 0, sizeof(ifr));
3474
3475 rtnl_lock();
3476 tun = tun_get(tfile);
3477 if (tun)
3478 tun_get_iff(tun, &ifr);
3479 rtnl_unlock();
3480
3481 if (tun)
3482 tun_put(tun);
3483
3484 seq_printf(m, "iff:\t%s\n", ifr.ifr_name);
3485 }
3486 #endif
3487
3488 static const struct file_operations tun_fops = {
3489 .owner = THIS_MODULE,
3490 .llseek = no_llseek,
3491 .read_iter = tun_chr_read_iter,
3492 .write_iter = tun_chr_write_iter,
3493 .poll = tun_chr_poll,
3494 .unlocked_ioctl = tun_chr_ioctl,
3495 #ifdef CONFIG_COMPAT
3496 .compat_ioctl = tun_chr_compat_ioctl,
3497 #endif
3498 .open = tun_chr_open,
3499 .release = tun_chr_close,
3500 .fasync = tun_chr_fasync,
3501 #ifdef CONFIG_PROC_FS
3502 .show_fdinfo = tun_chr_show_fdinfo,
3503 #endif
3504 };
3505
3506 static struct miscdevice tun_miscdev = {
3507 .minor = TUN_MINOR,
3508 .name = "tun",
3509 .nodename = "net/tun",
3510 .fops = &tun_fops,
3511 };
3512
3513
3514
3515 static void tun_default_link_ksettings(struct net_device *dev,
3516 struct ethtool_link_ksettings *cmd)
3517 {
3518 ethtool_link_ksettings_zero_link_mode(cmd, supported);
3519 ethtool_link_ksettings_zero_link_mode(cmd, advertising);
3520 cmd->base.speed = SPEED_10;
3521 cmd->base.duplex = DUPLEX_FULL;
3522 cmd->base.port = PORT_TP;
3523 cmd->base.phy_address = 0;
3524 cmd->base.autoneg = AUTONEG_DISABLE;
3525 }
3526
3527 static int tun_get_link_ksettings(struct net_device *dev,
3528 struct ethtool_link_ksettings *cmd)
3529 {
3530 struct tun_struct *tun = netdev_priv(dev);
3531
3532 memcpy(cmd, &tun->link_ksettings, sizeof(*cmd));
3533 return 0;
3534 }
3535
3536 static int tun_set_link_ksettings(struct net_device *dev,
3537 const struct ethtool_link_ksettings *cmd)
3538 {
3539 struct tun_struct *tun = netdev_priv(dev);
3540
3541 memcpy(&tun->link_ksettings, cmd, sizeof(*cmd));
3542 return 0;
3543 }
3544
3545 static void tun_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
3546 {
3547 struct tun_struct *tun = netdev_priv(dev);
3548
3549 strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
3550 strlcpy(info->version, DRV_VERSION, sizeof(info->version));
3551
3552 switch (tun->flags & TUN_TYPE_MASK) {
3553 case IFF_TUN:
3554 strlcpy(info->bus_info, "tun", sizeof(info->bus_info));
3555 break;
3556 case IFF_TAP:
3557 strlcpy(info->bus_info, "tap", sizeof(info->bus_info));
3558 break;
3559 }
3560 }
3561
3562 static u32 tun_get_msglevel(struct net_device *dev)
3563 {
3564 #ifdef TUN_DEBUG
3565 struct tun_struct *tun = netdev_priv(dev);
3566 return tun->debug;
3567 #else
3568 return -EOPNOTSUPP;
3569 #endif
3570 }
3571
3572 static void tun_set_msglevel(struct net_device *dev, u32 value)
3573 {
3574 #ifdef TUN_DEBUG
3575 struct tun_struct *tun = netdev_priv(dev);
3576 tun->debug = value;
3577 #endif
3578 }
3579
3580 static int tun_get_coalesce(struct net_device *dev,
3581 struct ethtool_coalesce *ec)
3582 {
3583 struct tun_struct *tun = netdev_priv(dev);
3584
3585 ec->rx_max_coalesced_frames = tun->rx_batched;
3586
3587 return 0;
3588 }
3589
3590 static int tun_set_coalesce(struct net_device *dev,
3591 struct ethtool_coalesce *ec)
3592 {
3593 struct tun_struct *tun = netdev_priv(dev);
3594
3595 if (ec->rx_max_coalesced_frames > NAPI_POLL_WEIGHT)
3596 tun->rx_batched = NAPI_POLL_WEIGHT;
3597 else
3598 tun->rx_batched = ec->rx_max_coalesced_frames;
3599
3600 return 0;
3601 }
3602
3603 static const struct ethtool_ops tun_ethtool_ops = {
3604 .get_drvinfo = tun_get_drvinfo,
3605 .get_msglevel = tun_get_msglevel,
3606 .set_msglevel = tun_set_msglevel,
3607 .get_link = ethtool_op_get_link,
3608 .get_ts_info = ethtool_op_get_ts_info,
3609 .get_coalesce = tun_get_coalesce,
3610 .set_coalesce = tun_set_coalesce,
3611 .get_link_ksettings = tun_get_link_ksettings,
3612 .set_link_ksettings = tun_set_link_ksettings,
3613 };
3614
3615 static int tun_queue_resize(struct tun_struct *tun)
3616 {
3617 struct net_device *dev = tun->dev;
3618 struct tun_file *tfile;
3619 struct ptr_ring **rings;
3620 int n = tun->numqueues + tun->numdisabled;
3621 int ret, i;
3622
3623 rings = kmalloc_array(n, sizeof(*rings), GFP_KERNEL);
3624 if (!rings)
3625 return -ENOMEM;
3626
3627 for (i = 0; i < tun->numqueues; i++) {
3628 tfile = rtnl_dereference(tun->tfiles[i]);
3629 rings[i] = &tfile->tx_ring;
3630 }
3631 list_for_each_entry(tfile, &tun->disabled, next)
3632 rings[i++] = &tfile->tx_ring;
3633
3634 ret = ptr_ring_resize_multiple(rings, n,
3635 dev->tx_queue_len, GFP_KERNEL,
3636 tun_ptr_free);
3637
3638 kfree(rings);
3639 return ret;
3640 }
3641
3642 static int tun_device_event(struct notifier_block *unused,
3643 unsigned long event, void *ptr)
3644 {
3645 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
3646 struct tun_struct *tun = netdev_priv(dev);
3647 int i;
3648
3649 if (dev->rtnl_link_ops != &tun_link_ops)
3650 return NOTIFY_DONE;
3651
3652 switch (event) {
3653 case NETDEV_CHANGE_TX_QUEUE_LEN:
3654 if (tun_queue_resize(tun))
3655 return NOTIFY_BAD;
3656 break;
3657 case NETDEV_UP:
3658 for (i = 0; i < tun->numqueues; i++) {
3659 struct tun_file *tfile;
3660
3661 tfile = rtnl_dereference(tun->tfiles[i]);
3662 tfile->socket.sk->sk_write_space(tfile->socket.sk);
3663 }
3664 break;
3665 default:
3666 break;
3667 }
3668
3669 return NOTIFY_DONE;
3670 }
3671
3672 static struct notifier_block tun_notifier_block __read_mostly = {
3673 .notifier_call = tun_device_event,
3674 };
3675
3676 static int __init tun_init(void)
3677 {
3678 int ret = 0;
3679
3680 pr_info("%s, %s\n", DRV_DESCRIPTION, DRV_VERSION);
3681
3682 ret = rtnl_link_register(&tun_link_ops);
3683 if (ret) {
3684 pr_err("Can't register link_ops\n");
3685 goto err_linkops;
3686 }
3687
3688 ret = misc_register(&tun_miscdev);
3689 if (ret) {
3690 pr_err("Can't register misc device %d\n", TUN_MINOR);
3691 goto err_misc;
3692 }
3693
3694 ret = register_netdevice_notifier(&tun_notifier_block);
3695 if (ret) {
3696 pr_err("Can't register netdevice notifier\n");
3697 goto err_notifier;
3698 }
3699
3700 return 0;
3701
3702 err_notifier:
3703 misc_deregister(&tun_miscdev);
3704 err_misc:
3705 rtnl_link_unregister(&tun_link_ops);
3706 err_linkops:
3707 return ret;
3708 }
3709
3710 static void tun_cleanup(void)
3711 {
3712 misc_deregister(&tun_miscdev);
3713 rtnl_link_unregister(&tun_link_ops);
3714 unregister_netdevice_notifier(&tun_notifier_block);
3715 }
3716
3717
3718
3719
3720
3721 struct socket *tun_get_socket(struct file *file)
3722 {
3723 struct tun_file *tfile;
3724 if (file->f_op != &tun_fops)
3725 return ERR_PTR(-EINVAL);
3726 tfile = file->private_data;
3727 if (!tfile)
3728 return ERR_PTR(-EBADFD);
3729 return &tfile->socket;
3730 }
3731 EXPORT_SYMBOL_GPL(tun_get_socket);
3732
3733 struct ptr_ring *tun_get_tx_ring(struct file *file)
3734 {
3735 struct tun_file *tfile;
3736
3737 if (file->f_op != &tun_fops)
3738 return ERR_PTR(-EINVAL);
3739 tfile = file->private_data;
3740 if (!tfile)
3741 return ERR_PTR(-EBADFD);
3742 return &tfile->tx_ring;
3743 }
3744 EXPORT_SYMBOL_GPL(tun_get_tx_ring);
3745
3746 module_init(tun_init);
3747 module_exit(tun_cleanup);
3748 MODULE_DESCRIPTION(DRV_DESCRIPTION);
3749 MODULE_AUTHOR(DRV_COPYRIGHT);
3750 MODULE_LICENSE("GPL");
3751 MODULE_ALIAS_MISCDEV(TUN_MINOR);
3752 MODULE_ALIAS("devname:net/tun");