?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Linux 4.19.133
[linux/fpc-iii.git] / drivers / net / tun.c
blobd84a09172ddbf23c3d600c941331090fa3fd29c1
1 /*
2 * TUN - Universal TUN/TAP device driver.
3 * Copyright (C) 1999-2002 Maxim Krasnyansky <maxk@qualcomm.com>
4 *
5 * This program is free software; you can redistribute it and/or modify
6 * it under the terms of the GNU General Public License as published by
7 * the Free Software Foundation; either version 2 of the License, or
8 * (at your option) any later version.
9 *
10 * This program is distributed in the hope that it will be useful,
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
13 * GNU General Public License for more details.
14 *
15 * $Id: tun.c,v 1.15 2002/03/01 02:44:24 maxk Exp $
16 */
17
18 /*
19 * Changes:
20 *
21 * Mike Kershaw <dragorn@kismetwireless.net> 2005/08/14
22 * Add TUNSETLINK ioctl to set the link encapsulation
23 *
24 * Mark Smith <markzzzsmith@yahoo.com.au>
25 * Use eth_random_addr() for tap MAC address.
26 *
27 * Harald Roelle <harald.roelle@ifi.lmu.de> 2004/04/20
28 * Fixes in packet dropping, queue length setting and queue wakeup.
29 * Increased default tx queue length.
30 * Added ethtool API.
31 * Minor cleanups
32 *
33 * Daniel Podlejski <underley@underley.eu.org>
34 * Modifications for 2.3.99-pre5 kernel.
35 */
36
37 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
38
39 #define DRV_NAME "tun"
40 #define DRV_VERSION "1.6"
41 #define DRV_DESCRIPTION "Universal TUN/TAP device driver"
42 #define DRV_COPYRIGHT "(C) 1999-2004 Max Krasnyansky <maxk@qualcomm.com>"
43
44 #include <linux/module.h>
45 #include <linux/errno.h>
46 #include <linux/kernel.h>
47 #include <linux/sched/signal.h>
48 #include <linux/major.h>
49 #include <linux/slab.h>
50 #include <linux/poll.h>
51 #include <linux/fcntl.h>
52 #include <linux/init.h>
53 #include <linux/skbuff.h>
54 #include <linux/netdevice.h>
55 #include <linux/etherdevice.h>
56 #include <linux/miscdevice.h>
57 #include <linux/ethtool.h>
58 #include <linux/rtnetlink.h>
59 #include <linux/compat.h>
60 #include <linux/if.h>
61 #include <linux/if_arp.h>
62 #include <linux/if_ether.h>
63 #include <linux/if_tun.h>
64 #include <linux/if_vlan.h>
65 #include <linux/crc32.h>
66 #include <linux/nsproxy.h>
67 #include <linux/virtio_net.h>
68 #include <linux/rcupdate.h>
69 #include <net/net_namespace.h>
70 #include <net/netns/generic.h>
71 #include <net/rtnetlink.h>
72 #include <net/sock.h>
73 #include <net/xdp.h>
74 #include <linux/seq_file.h>
75 #include <linux/uio.h>
76 #include <linux/skb_array.h>
77 #include <linux/bpf.h>
78 #include <linux/bpf_trace.h>
79 #include <linux/mutex.h>
80
81 #include <linux/uaccess.h>
82 #include <linux/proc_fs.h>
83
84 static void tun_default_link_ksettings(struct net_device *dev,
85 struct ethtool_link_ksettings *cmd);
86
87 /* Uncomment to enable debugging */
88 /* #define TUN_DEBUG 1 */
89
90 #ifdef TUN_DEBUG
91 static int debug;
92
93 #define tun_debug(level, tun, fmt, args...) \
94 do { \
95 if (tun->debug) \
96 netdev_printk(level, tun->dev, fmt, ##args); \
97 } while (0)
98 #define DBG1(level, fmt, args...) \
99 do { \
100 if (debug == 2) \
101 printk(level fmt, ##args); \
102 } while (0)
103 #else
104 #define tun_debug(level, tun, fmt, args...) \
105 do { \
106 if (0) \
107 netdev_printk(level, tun->dev, fmt, ##args); \
108 } while (0)
109 #define DBG1(level, fmt, args...) \
110 do { \
111 if (0) \
112 printk(level fmt, ##args); \
113 } while (0)
114 #endif
115
116 #define TUN_HEADROOM 256
117 #define TUN_RX_PAD (NET_IP_ALIGN + NET_SKB_PAD)
118
119 /* TUN device flags */
120
121 /* IFF_ATTACH_QUEUE is never stored in device flags,
122 * overload it to mean fasync when stored there.
123 */
124 #define TUN_FASYNC IFF_ATTACH_QUEUE
125 /* High bits in flags field are unused. */
126 #define TUN_VNET_LE 0x80000000
127 #define TUN_VNET_BE 0x40000000
128
129 #define TUN_FEATURES (IFF_NO_PI | IFF_ONE_QUEUE | IFF_VNET_HDR | \
130 IFF_MULTI_QUEUE | IFF_NAPI | IFF_NAPI_FRAGS)
131
132 #define GOODCOPY_LEN 128
133
134 #define FLT_EXACT_COUNT 8
135 struct tap_filter {
136 unsigned int count; /* Number of addrs. Zero means disabled */
137 u32 mask[2]; /* Mask of the hashed addrs */
138 unsigned char addr[FLT_EXACT_COUNT][ETH_ALEN];
139 };
140
141 /* MAX_TAP_QUEUES 256 is chosen to allow rx/tx queues to be equal
142 * to max number of VCPUs in guest. */
143 #define MAX_TAP_QUEUES 256
144 #define MAX_TAP_FLOWS 4096
145
146 #define TUN_FLOW_EXPIRE (3 * HZ)
147
148 struct tun_pcpu_stats {
149 u64 rx_packets;
150 u64 rx_bytes;
151 u64 tx_packets;
152 u64 tx_bytes;
153 struct u64_stats_sync syncp;
154 u32 rx_dropped;
155 u32 tx_dropped;
156 u32 rx_frame_errors;
157 };
158
159 /* A tun_file connects an open character device to a tuntap netdevice. It
160 * also contains all socket related structures (except sock_fprog and tap_filter)
161 * to serve as one transmit queue for tuntap device. The sock_fprog and
162 * tap_filter were kept in tun_struct since they were used for filtering for the
163 * netdevice not for a specific queue (at least I didn't see the requirement for
164 * this).
165 *
166 * RCU usage:
167 * The tun_file and tun_struct are loosely coupled, the pointer from one to the
168 * other can only be read while rcu_read_lock or rtnl_lock is held.
169 */
170 struct tun_file {
171 struct sock sk;
172 struct socket socket;
173 struct socket_wq wq;
174 struct tun_struct __rcu *tun;
175 struct fasync_struct *fasync;
176 /* only used for fasnyc */
177 unsigned int flags;
178 union {
179 u16 queue_index;
180 unsigned int ifindex;
181 };
182 struct napi_struct napi;
183 bool napi_enabled;
184 bool napi_frags_enabled;
185 struct mutex napi_mutex; /* Protects access to the above napi */
186 struct list_head next;
187 struct tun_struct *detached;
188 struct ptr_ring tx_ring;
189 struct xdp_rxq_info xdp_rxq;
190 };
191
192 struct tun_flow_entry {
193 struct hlist_node hash_link;
194 struct rcu_head rcu;
195 struct tun_struct *tun;
196
197 u32 rxhash;
198 u32 rps_rxhash;
199 int queue_index;
200 unsigned long updated;
201 };
202
203 #define TUN_NUM_FLOW_ENTRIES 1024
204 #define TUN_MASK_FLOW_ENTRIES (TUN_NUM_FLOW_ENTRIES - 1)
205
206 struct tun_prog {
207 struct rcu_head rcu;
208 struct bpf_prog *prog;
209 };
210
211 /* Since the socket were moved to tun_file, to preserve the behavior of persist
212 * device, socket filter, sndbuf and vnet header size were restore when the
213 * file were attached to a persist device.
214 */
215 struct tun_struct {
216 struct tun_file __rcu *tfiles[MAX_TAP_QUEUES];
217 unsigned int numqueues;
218 unsigned int flags;
219 kuid_t owner;
220 kgid_t group;
221
222 struct net_device *dev;
223 netdev_features_t set_features;
224 #define TUN_USER_FEATURES (NETIF_F_HW_CSUM|NETIF_F_TSO_ECN|NETIF_F_TSO| \
225 NETIF_F_TSO6)
226
227 int align;
228 int vnet_hdr_sz;
229 int sndbuf;
230 struct tap_filter txflt;
231 struct sock_fprog fprog;
232 /* protected by rtnl lock */
233 bool filter_attached;
234 #ifdef TUN_DEBUG
235 int debug;
236 #endif
237 spinlock_t lock;
238 struct hlist_head flows[TUN_NUM_FLOW_ENTRIES];
239 struct timer_list flow_gc_timer;
240 unsigned long ageing_time;
241 unsigned int numdisabled;
242 struct list_head disabled;
243 void *security;
244 u32 flow_count;
245 u32 rx_batched;
246 struct tun_pcpu_stats __percpu *pcpu_stats;
247 struct bpf_prog __rcu *xdp_prog;
248 struct tun_prog __rcu *steering_prog;
249 struct tun_prog __rcu *filter_prog;
250 struct ethtool_link_ksettings link_ksettings;
251 };
252
253 struct veth {
254 __be16 h_vlan_proto;
255 __be16 h_vlan_TCI;
256 };
257
258 bool tun_is_xdp_frame(void *ptr)
259 {
260 return (unsigned long)ptr & TUN_XDP_FLAG;
261 }
262 EXPORT_SYMBOL(tun_is_xdp_frame);
263
264 void *tun_xdp_to_ptr(void *ptr)
265 {
266 return (void *)((unsigned long)ptr | TUN_XDP_FLAG);
267 }
268 EXPORT_SYMBOL(tun_xdp_to_ptr);
269
270 void *tun_ptr_to_xdp(void *ptr)
271 {
272 return (void *)((unsigned long)ptr & ~TUN_XDP_FLAG);
273 }
274 EXPORT_SYMBOL(tun_ptr_to_xdp);
275
276 static int tun_napi_receive(struct napi_struct *napi, int budget)
277 {
278 struct tun_file *tfile = container_of(napi, struct tun_file, napi);
279 struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
280 struct sk_buff_head process_queue;
281 struct sk_buff *skb;
282 int received = 0;
283
284 __skb_queue_head_init(&process_queue);
285
286 spin_lock(&queue->lock);
287 skb_queue_splice_tail_init(queue, &process_queue);
288 spin_unlock(&queue->lock);
289
290 while (received < budget && (skb = __skb_dequeue(&process_queue))) {
291 napi_gro_receive(napi, skb);
292 ++received;
293 }
294
295 if (!skb_queue_empty(&process_queue)) {
296 spin_lock(&queue->lock);
297 skb_queue_splice(&process_queue, queue);
298 spin_unlock(&queue->lock);
299 }
300
301 return received;
302 }
303
304 static int tun_napi_poll(struct napi_struct *napi, int budget)
305 {
306 unsigned int received;
307
308 received = tun_napi_receive(napi, budget);
309
310 if (received < budget)
311 napi_complete_done(napi, received);
312
313 return received;
314 }
315
316 static void tun_napi_init(struct tun_struct *tun, struct tun_file *tfile,
317 bool napi_en, bool napi_frags)
318 {
319 tfile->napi_enabled = napi_en;
320 tfile->napi_frags_enabled = napi_en && napi_frags;
321 if (napi_en) {
322 netif_tx_napi_add(tun->dev, &tfile->napi, tun_napi_poll,
323 NAPI_POLL_WEIGHT);
324 napi_enable(&tfile->napi);
325 }
326 }
327
328 static void tun_napi_disable(struct tun_file *tfile)
329 {
330 if (tfile->napi_enabled)
331 napi_disable(&tfile->napi);
332 }
333
334 static void tun_napi_del(struct tun_file *tfile)
335 {
336 if (tfile->napi_enabled)
337 netif_napi_del(&tfile->napi);
338 }
339
340 static bool tun_napi_frags_enabled(const struct tun_file *tfile)
341 {
342 return tfile->napi_frags_enabled;
343 }
344
345 #ifdef CONFIG_TUN_VNET_CROSS_LE
346 static inline bool tun_legacy_is_little_endian(struct tun_struct *tun)
347 {
348 return tun->flags & TUN_VNET_BE ? false :
349 virtio_legacy_is_little_endian();
350 }
351
352 static long tun_get_vnet_be(struct tun_struct *tun, int __user *argp)
353 {
354 int be = !!(tun->flags & TUN_VNET_BE);
355
356 if (put_user(be, argp))
357 return -EFAULT;
358
359 return 0;
360 }
361
362 static long tun_set_vnet_be(struct tun_struct *tun, int __user *argp)
363 {
364 int be;
365
366 if (get_user(be, argp))
367 return -EFAULT;
368
369 if (be)
370 tun->flags |= TUN_VNET_BE;
371 else
372 tun->flags &= ~TUN_VNET_BE;
373
374 return 0;
375 }
376 #else
377 static inline bool tun_legacy_is_little_endian(struct tun_struct *tun)
378 {
379 return virtio_legacy_is_little_endian();
380 }
381
382 static long tun_get_vnet_be(struct tun_struct *tun, int __user *argp)
383 {
384 return -EINVAL;
385 }
386
387 static long tun_set_vnet_be(struct tun_struct *tun, int __user *argp)
388 {
389 return -EINVAL;
390 }
391 #endif /* CONFIG_TUN_VNET_CROSS_LE */
392
393 static inline bool tun_is_little_endian(struct tun_struct *tun)
394 {
395 return tun->flags & TUN_VNET_LE ||
396 tun_legacy_is_little_endian(tun);
397 }
398
399 static inline u16 tun16_to_cpu(struct tun_struct *tun, __virtio16 val)
400 {
401 return __virtio16_to_cpu(tun_is_little_endian(tun), val);
402 }
403
404 static inline __virtio16 cpu_to_tun16(struct tun_struct *tun, u16 val)
405 {
406 return __cpu_to_virtio16(tun_is_little_endian(tun), val);
407 }
408
409 static inline u32 tun_hashfn(u32 rxhash)
410 {
411 return rxhash & TUN_MASK_FLOW_ENTRIES;
412 }
413
414 static struct tun_flow_entry *tun_flow_find(struct hlist_head *head, u32 rxhash)
415 {
416 struct tun_flow_entry *e;
417
418 hlist_for_each_entry_rcu(e, head, hash_link) {
419 if (e->rxhash == rxhash)
420 return e;
421 }
422 return NULL;
423 }
424
425 static struct tun_flow_entry *tun_flow_create(struct tun_struct *tun,
426 struct hlist_head *head,
427 u32 rxhash, u16 queue_index)
428 {
429 struct tun_flow_entry *e = kmalloc(sizeof(*e), GFP_ATOMIC);
430
431 if (e) {
432 tun_debug(KERN_INFO, tun, "create flow: hash %u index %u\n",
433 rxhash, queue_index);
434 e->updated = jiffies;
435 e->rxhash = rxhash;
436 e->rps_rxhash = 0;
437 e->queue_index = queue_index;
438 e->tun = tun;
439 hlist_add_head_rcu(&e->hash_link, head);
440 ++tun->flow_count;
441 }
442 return e;
443 }
444
445 static void tun_flow_delete(struct tun_struct *tun, struct tun_flow_entry *e)
446 {
447 tun_debug(KERN_INFO, tun, "delete flow: hash %u index %u\n",
448 e->rxhash, e->queue_index);
449 hlist_del_rcu(&e->hash_link);
450 kfree_rcu(e, rcu);
451 --tun->flow_count;
452 }
453
454 static void tun_flow_flush(struct tun_struct *tun)
455 {
456 int i;
457
458 spin_lock_bh(&tun->lock);
459 for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
460 struct tun_flow_entry *e;
461 struct hlist_node *n;
462
463 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link)
464 tun_flow_delete(tun, e);
465 }
466 spin_unlock_bh(&tun->lock);
467 }
468
469 static void tun_flow_delete_by_queue(struct tun_struct *tun, u16 queue_index)
470 {
471 int i;
472
473 spin_lock_bh(&tun->lock);
474 for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
475 struct tun_flow_entry *e;
476 struct hlist_node *n;
477
478 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
479 if (e->queue_index == queue_index)
480 tun_flow_delete(tun, e);
481 }
482 }
483 spin_unlock_bh(&tun->lock);
484 }
485
486 static void tun_flow_cleanup(struct timer_list *t)
487 {
488 struct tun_struct *tun = from_timer(tun, t, flow_gc_timer);
489 unsigned long delay = tun->ageing_time;
490 unsigned long next_timer = jiffies + delay;
491 unsigned long count = 0;
492 int i;
493
494 tun_debug(KERN_INFO, tun, "tun_flow_cleanup\n");
495
496 spin_lock(&tun->lock);
497 for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
498 struct tun_flow_entry *e;
499 struct hlist_node *n;
500
501 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
502 unsigned long this_timer;
503
504 this_timer = e->updated + delay;
505 if (time_before_eq(this_timer, jiffies)) {
506 tun_flow_delete(tun, e);
507 continue;
508 }
509 count++;
510 if (time_before(this_timer, next_timer))
511 next_timer = this_timer;
512 }
513 }
514
515 if (count)
516 mod_timer(&tun->flow_gc_timer, round_jiffies_up(next_timer));
517 spin_unlock(&tun->lock);
518 }
519
520 static void tun_flow_update(struct tun_struct *tun, u32 rxhash,
521 struct tun_file *tfile)
522 {
523 struct hlist_head *head;
524 struct tun_flow_entry *e;
525 unsigned long delay = tun->ageing_time;
526 u16 queue_index = tfile->queue_index;
527
528 if (!rxhash)
529 return;
530 else
531 head = &tun->flows[tun_hashfn(rxhash)];
532
533 rcu_read_lock();
534
535 e = tun_flow_find(head, rxhash);
536 if (likely(e)) {
537 /* TODO: keep queueing to old queue until it's empty? */
538 e->queue_index = queue_index;
539 e->updated = jiffies;
540 sock_rps_record_flow_hash(e->rps_rxhash);
541 } else {
542 spin_lock_bh(&tun->lock);
543 if (!tun_flow_find(head, rxhash) &&
544 tun->flow_count < MAX_TAP_FLOWS)
545 tun_flow_create(tun, head, rxhash, queue_index);
546
547 if (!timer_pending(&tun->flow_gc_timer))
548 mod_timer(&tun->flow_gc_timer,
549 round_jiffies_up(jiffies + delay));
550 spin_unlock_bh(&tun->lock);
551 }
552
553 rcu_read_unlock();
554 }
555
556 /**
557 * Save the hash received in the stack receive path and update the
558 * flow_hash table accordingly.
559 */
560 static inline void tun_flow_save_rps_rxhash(struct tun_flow_entry *e, u32 hash)
561 {
562 if (unlikely(e->rps_rxhash != hash))
563 e->rps_rxhash = hash;
564 }
565
566 /* We try to identify a flow through its rxhash first. The reason that
567 * we do not check rxq no. is because some cards(e.g 82599), chooses
568 * the rxq based on the txq where the last packet of the flow comes. As
569 * the userspace application move between processors, we may get a
570 * different rxq no. here. If we could not get rxhash, then we would
571 * hope the rxq no. may help here.
572 */
573 static u16 tun_automq_select_queue(struct tun_struct *tun, struct sk_buff *skb)
574 {
575 struct tun_flow_entry *e;
576 u32 txq = 0;
577 u32 numqueues = 0;
578
579 numqueues = READ_ONCE(tun->numqueues);
580
581 txq = __skb_get_hash_symmetric(skb);
582 if (txq) {
583 e = tun_flow_find(&tun->flows[tun_hashfn(txq)], txq);
584 if (e) {
585 tun_flow_save_rps_rxhash(e, txq);
586 txq = e->queue_index;
587 } else
588 /* use multiply and shift instead of expensive divide */
589 txq = ((u64)txq * numqueues) >> 32;
590 } else if (likely(skb_rx_queue_recorded(skb))) {
591 txq = skb_get_rx_queue(skb);
592 while (unlikely(txq >= numqueues))
593 txq -= numqueues;
594 }
595
596 return txq;
597 }
598
599 static u16 tun_ebpf_select_queue(struct tun_struct *tun, struct sk_buff *skb)
600 {
601 struct tun_prog *prog;
602 u32 numqueues;
603 u16 ret = 0;
604
605 numqueues = READ_ONCE(tun->numqueues);
606 if (!numqueues)
607 return 0;
608
609 prog = rcu_dereference(tun->steering_prog);
610 if (prog)
611 ret = bpf_prog_run_clear_cb(prog->prog, skb);
612
613 return ret % numqueues;
614 }
615
616 static u16 tun_select_queue(struct net_device *dev, struct sk_buff *skb,
617 struct net_device *sb_dev,
618 select_queue_fallback_t fallback)
619 {
620 struct tun_struct *tun = netdev_priv(dev);
621 u16 ret;
622
623 rcu_read_lock();
624 if (rcu_dereference(tun->steering_prog))
625 ret = tun_ebpf_select_queue(tun, skb);
626 else
627 ret = tun_automq_select_queue(tun, skb);
628 rcu_read_unlock();
629
630 return ret;
631 }
632
633 static inline bool tun_not_capable(struct tun_struct *tun)
634 {
635 const struct cred *cred = current_cred();
636 struct net *net = dev_net(tun->dev);
637
638 return ((uid_valid(tun->owner) && !uid_eq(cred->euid, tun->owner)) ||
639 (gid_valid(tun->group) && !in_egroup_p(tun->group))) &&
640 !ns_capable(net->user_ns, CAP_NET_ADMIN);
641 }
642
643 static void tun_set_real_num_queues(struct tun_struct *tun)
644 {
645 netif_set_real_num_tx_queues(tun->dev, tun->numqueues);
646 netif_set_real_num_rx_queues(tun->dev, tun->numqueues);
647 }
648
649 static void tun_disable_queue(struct tun_struct *tun, struct tun_file *tfile)
650 {
651 tfile->detached = tun;
652 list_add_tail(&tfile->next, &tun->disabled);
653 ++tun->numdisabled;
654 }
655
656 static struct tun_struct *tun_enable_queue(struct tun_file *tfile)
657 {
658 struct tun_struct *tun = tfile->detached;
659
660 tfile->detached = NULL;
661 list_del_init(&tfile->next);
662 --tun->numdisabled;
663 return tun;
664 }
665
666 void tun_ptr_free(void *ptr)
667 {
668 if (!ptr)
669 return;
670 if (tun_is_xdp_frame(ptr)) {
671 struct xdp_frame *xdpf = tun_ptr_to_xdp(ptr);
672
673 xdp_return_frame(xdpf);
674 } else {
675 __skb_array_destroy_skb(ptr);
676 }
677 }
678 EXPORT_SYMBOL_GPL(tun_ptr_free);
679
680 static void tun_queue_purge(struct tun_file *tfile)
681 {
682 void *ptr;
683
684 while ((ptr = ptr_ring_consume(&tfile->tx_ring)) != NULL)
685 tun_ptr_free(ptr);
686
687 skb_queue_purge(&tfile->sk.sk_write_queue);
688 skb_queue_purge(&tfile->sk.sk_error_queue);
689 }
690
691 static void __tun_detach(struct tun_file *tfile, bool clean)
692 {
693 struct tun_file *ntfile;
694 struct tun_struct *tun;
695
696 tun = rtnl_dereference(tfile->tun);
697
698 if (tun && clean) {
699 tun_napi_disable(tfile);
700 tun_napi_del(tfile);
701 }
702
703 if (tun && !tfile->detached) {
704 u16 index = tfile->queue_index;
705 BUG_ON(index >= tun->numqueues);
706
707 rcu_assign_pointer(tun->tfiles[index],
708 tun->tfiles[tun->numqueues - 1]);
709 ntfile = rtnl_dereference(tun->tfiles[index]);
710 ntfile->queue_index = index;
711 rcu_assign_pointer(tun->tfiles[tun->numqueues - 1],
712 NULL);
713
714 --tun->numqueues;
715 if (clean) {
716 RCU_INIT_POINTER(tfile->tun, NULL);
717 sock_put(&tfile->sk);
718 } else
719 tun_disable_queue(tun, tfile);
720
721 synchronize_net();
722 tun_flow_delete_by_queue(tun, tun->numqueues + 1);
723 /* Drop read queue */
724 tun_queue_purge(tfile);
725 tun_set_real_num_queues(tun);
726 } else if (tfile->detached && clean) {
727 tun = tun_enable_queue(tfile);
728 sock_put(&tfile->sk);
729 }
730
731 if (clean) {
732 if (tun && tun->numqueues == 0 && tun->numdisabled == 0) {
733 netif_carrier_off(tun->dev);
734
735 if (!(tun->flags & IFF_PERSIST) &&
736 tun->dev->reg_state == NETREG_REGISTERED)
737 unregister_netdevice(tun->dev);
738 }
739 if (tun)
740 xdp_rxq_info_unreg(&tfile->xdp_rxq);
741 ptr_ring_cleanup(&tfile->tx_ring, tun_ptr_free);
742 sock_put(&tfile->sk);
743 }
744 }
745
746 static void tun_detach(struct tun_file *tfile, bool clean)
747 {
748 struct tun_struct *tun;
749 struct net_device *dev;
750
751 rtnl_lock();
752 tun = rtnl_dereference(tfile->tun);
753 dev = tun ? tun->dev : NULL;
754 __tun_detach(tfile, clean);
755 if (dev)
756 netdev_state_change(dev);
757 rtnl_unlock();
758 }
759
760 static void tun_detach_all(struct net_device *dev)
761 {
762 struct tun_struct *tun = netdev_priv(dev);
763 struct tun_file *tfile, *tmp;
764 int i, n = tun->numqueues;
765
766 for (i = 0; i < n; i++) {
767 tfile = rtnl_dereference(tun->tfiles[i]);
768 BUG_ON(!tfile);
769 tun_napi_disable(tfile);
770 tfile->socket.sk->sk_shutdown = RCV_SHUTDOWN;
771 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
772 RCU_INIT_POINTER(tfile->tun, NULL);
773 --tun->numqueues;
774 }
775 list_for_each_entry(tfile, &tun->disabled, next) {
776 tfile->socket.sk->sk_shutdown = RCV_SHUTDOWN;
777 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
778 RCU_INIT_POINTER(tfile->tun, NULL);
779 }
780 BUG_ON(tun->numqueues != 0);
781
782 synchronize_net();
783 for (i = 0; i < n; i++) {
784 tfile = rtnl_dereference(tun->tfiles[i]);
785 tun_napi_del(tfile);
786 /* Drop read queue */
787 tun_queue_purge(tfile);
788 xdp_rxq_info_unreg(&tfile->xdp_rxq);
789 sock_put(&tfile->sk);
790 }
791 list_for_each_entry_safe(tfile, tmp, &tun->disabled, next) {
792 tun_enable_queue(tfile);
793 tun_queue_purge(tfile);
794 xdp_rxq_info_unreg(&tfile->xdp_rxq);
795 sock_put(&tfile->sk);
796 }
797 BUG_ON(tun->numdisabled != 0);
798
799 if (tun->flags & IFF_PERSIST)
800 module_put(THIS_MODULE);
801 }
802
803 static int tun_attach(struct tun_struct *tun, struct file *file,
804 bool skip_filter, bool napi, bool napi_frags,
805 bool publish_tun)
806 {
807 struct tun_file *tfile = file->private_data;
808 struct net_device *dev = tun->dev;
809 int err;
810
811 err = security_tun_dev_attach(tfile->socket.sk, tun->security);
812 if (err < 0)
813 goto out;
814
815 err = -EINVAL;
816 if (rtnl_dereference(tfile->tun) && !tfile->detached)
817 goto out;
818
819 err = -EBUSY;
820 if (!(tun->flags & IFF_MULTI_QUEUE) && tun->numqueues == 1)
821 goto out;
822
823 err = -E2BIG;
824 if (!tfile->detached &&
825 tun->numqueues + tun->numdisabled == MAX_TAP_QUEUES)
826 goto out;
827
828 err = 0;
829
830 /* Re-attach the filter to persist device */
831 if (!skip_filter && (tun->filter_attached == true)) {
832 lock_sock(tfile->socket.sk);
833 err = sk_attach_filter(&tun->fprog, tfile->socket.sk);
834 release_sock(tfile->socket.sk);
835 if (!err)
836 goto out;
837 }
838
839 if (!tfile->detached &&
840 ptr_ring_resize(&tfile->tx_ring, dev->tx_queue_len,
841 GFP_KERNEL, tun_ptr_free)) {
842 err = -ENOMEM;
843 goto out;
844 }
845
846 tfile->queue_index = tun->numqueues;
847 tfile->socket.sk->sk_shutdown &= ~RCV_SHUTDOWN;
848
849 if (tfile->detached) {
850 /* Re-attach detached tfile, updating XDP queue_index */
851 WARN_ON(!xdp_rxq_info_is_reg(&tfile->xdp_rxq));
852
853 if (tfile->xdp_rxq.queue_index != tfile->queue_index)
854 tfile->xdp_rxq.queue_index = tfile->queue_index;
855 } else {
856 /* Setup XDP RX-queue info, for new tfile getting attached */
857 err = xdp_rxq_info_reg(&tfile->xdp_rxq,
858 tun->dev, tfile->queue_index);
859 if (err < 0)
860 goto out;
861 err = xdp_rxq_info_reg_mem_model(&tfile->xdp_rxq,
862 MEM_TYPE_PAGE_SHARED, NULL);
863 if (err < 0) {
864 xdp_rxq_info_unreg(&tfile->xdp_rxq);
865 goto out;
866 }
867 err = 0;
868 }
869
870 if (tfile->detached) {
871 tun_enable_queue(tfile);
872 } else {
873 sock_hold(&tfile->sk);
874 tun_napi_init(tun, tfile, napi, napi_frags);
875 }
876
877 /* device is allowed to go away first, so no need to hold extra
878 * refcnt.
879 */
880
881 /* Publish tfile->tun and tun->tfiles only after we've fully
882 * initialized tfile; otherwise we risk using half-initialized
883 * object.
884 */
885 if (publish_tun)
886 rcu_assign_pointer(tfile->tun, tun);
887 rcu_assign_pointer(tun->tfiles[tun->numqueues], tfile);
888 tun->numqueues++;
889 tun_set_real_num_queues(tun);
890 out:
891 return err;
892 }
893
894 static struct tun_struct *tun_get(struct tun_file *tfile)
895 {
896 struct tun_struct *tun;
897
898 rcu_read_lock();
899 tun = rcu_dereference(tfile->tun);
900 if (tun)
901 dev_hold(tun->dev);
902 rcu_read_unlock();
903
904 return tun;
905 }
906
907 static void tun_put(struct tun_struct *tun)
908 {
909 dev_put(tun->dev);
910 }
911
912 /* TAP filtering */
913 static void addr_hash_set(u32 *mask, const u8 *addr)
914 {
915 int n = ether_crc(ETH_ALEN, addr) >> 26;
916 mask[n >> 5] |= (1 << (n & 31));
917 }
918
919 static unsigned int addr_hash_test(const u32 *mask, const u8 *addr)
920 {
921 int n = ether_crc(ETH_ALEN, addr) >> 26;
922 return mask[n >> 5] & (1 << (n & 31));
923 }
924
925 static int update_filter(struct tap_filter *filter, void __user *arg)
926 {
927 struct { u8 u[ETH_ALEN]; } *addr;
928 struct tun_filter uf;
929 int err, alen, n, nexact;
930
931 if (copy_from_user(&uf, arg, sizeof(uf)))
932 return -EFAULT;
933
934 if (!uf.count) {
935 /* Disabled */
936 filter->count = 0;
937 return 0;
938 }
939
940 alen = ETH_ALEN * uf.count;
941 addr = memdup_user(arg + sizeof(uf), alen);
942 if (IS_ERR(addr))
943 return PTR_ERR(addr);
944
945 /* The filter is updated without holding any locks. Which is
946 * perfectly safe. We disable it first and in the worst
947 * case we'll accept a few undesired packets. */
948 filter->count = 0;
949 wmb();
950
951 /* Use first set of addresses as an exact filter */
952 for (n = 0; n < uf.count && n < FLT_EXACT_COUNT; n++)
953 memcpy(filter->addr[n], addr[n].u, ETH_ALEN);
954
955 nexact = n;
956
957 /* Remaining multicast addresses are hashed,
958 * unicast will leave the filter disabled. */
959 memset(filter->mask, 0, sizeof(filter->mask));
960 for (; n < uf.count; n++) {
961 if (!is_multicast_ether_addr(addr[n].u)) {
962 err = 0; /* no filter */
963 goto free_addr;
964 }
965 addr_hash_set(filter->mask, addr[n].u);
966 }
967
968 /* For ALLMULTI just set the mask to all ones.
969 * This overrides the mask populated above. */
970 if ((uf.flags & TUN_FLT_ALLMULTI))
971 memset(filter->mask, ~0, sizeof(filter->mask));
972
973 /* Now enable the filter */
974 wmb();
975 filter->count = nexact;
976
977 /* Return the number of exact filters */
978 err = nexact;
979 free_addr:
980 kfree(addr);
981 return err;
982 }
983
984 /* Returns: 0 - drop, !=0 - accept */
985 static int run_filter(struct tap_filter *filter, const struct sk_buff *skb)
986 {
987 /* Cannot use eth_hdr(skb) here because skb_mac_hdr() is incorrect
988 * at this point. */
989 struct ethhdr *eh = (struct ethhdr *) skb->data;
990 int i;
991
992 /* Exact match */
993 for (i = 0; i < filter->count; i++)
994 if (ether_addr_equal(eh->h_dest, filter->addr[i]))
995 return 1;
996
997 /* Inexact match (multicast only) */
998 if (is_multicast_ether_addr(eh->h_dest))
999 return addr_hash_test(filter->mask, eh->h_dest);
1000
1001 return 0;
1002 }
1003
1004 /*
1005 * Checks whether the packet is accepted or not.
1006 * Returns: 0 - drop, !=0 - accept
1007 */
1008 static int check_filter(struct tap_filter *filter, const struct sk_buff *skb)
1009 {
1010 if (!filter->count)
1011 return 1;
1012
1013 return run_filter(filter, skb);
1014 }
1015
1016 /* Network device part of the driver */
1017
1018 static const struct ethtool_ops tun_ethtool_ops;
1019
1020 /* Net device detach from fd. */
1021 static void tun_net_uninit(struct net_device *dev)
1022 {
1023 tun_detach_all(dev);
1024 }
1025
1026 /* Net device open. */
1027 static int tun_net_open(struct net_device *dev)
1028 {
1029 netif_tx_start_all_queues(dev);
1030
1031 return 0;
1032 }
1033
1034 /* Net device close. */
1035 static int tun_net_close(struct net_device *dev)
1036 {
1037 netif_tx_stop_all_queues(dev);
1038 return 0;
1039 }
1040
1041 /* Net device start xmit */
1042 static void tun_automq_xmit(struct tun_struct *tun, struct sk_buff *skb)
1043 {
1044 #ifdef CONFIG_RPS
1045 if (tun->numqueues == 1 && static_key_false(&rps_needed)) {
1046 /* Select queue was not called for the skbuff, so we extract the
1047 * RPS hash and save it into the flow_table here.
1048 */
1049 __u32 rxhash;
1050
1051 rxhash = __skb_get_hash_symmetric(skb);
1052 if (rxhash) {
1053 struct tun_flow_entry *e;
1054 e = tun_flow_find(&tun->flows[tun_hashfn(rxhash)],
1055 rxhash);
1056 if (e)
1057 tun_flow_save_rps_rxhash(e, rxhash);
1058 }
1059 }
1060 #endif
1061 }
1062
1063 static unsigned int run_ebpf_filter(struct tun_struct *tun,
1064 struct sk_buff *skb,
1065 int len)
1066 {
1067 struct tun_prog *prog = rcu_dereference(tun->filter_prog);
1068
1069 if (prog)
1070 len = bpf_prog_run_clear_cb(prog->prog, skb);
1071
1072 return len;
1073 }
1074
1075 /* Net device start xmit */
1076 static netdev_tx_t tun_net_xmit(struct sk_buff *skb, struct net_device *dev)
1077 {
1078 struct tun_struct *tun = netdev_priv(dev);
1079 int txq = skb->queue_mapping;
1080 struct tun_file *tfile;
1081 int len = skb->len;
1082
1083 rcu_read_lock();
1084 tfile = rcu_dereference(tun->tfiles[txq]);
1085
1086 /* Drop packet if interface is not attached */
1087 if (!tfile)
1088 goto drop;
1089
1090 if (!rcu_dereference(tun->steering_prog))
1091 tun_automq_xmit(tun, skb);
1092
1093 tun_debug(KERN_INFO, tun, "tun_net_xmit %d\n", skb->len);
1094
1095 BUG_ON(!tfile);
1096
1097 /* Drop if the filter does not like it.
1098 * This is a noop if the filter is disabled.
1099 * Filter can be enabled only for the TAP devices. */
1100 if (!check_filter(&tun->txflt, skb))
1101 goto drop;
1102
1103 if (tfile->socket.sk->sk_filter &&
1104 sk_filter(tfile->socket.sk, skb))
1105 goto drop;
1106
1107 len = run_ebpf_filter(tun, skb, len);
1108 if (len == 0 || pskb_trim(skb, len))
1109 goto drop;
1110
1111 if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC)))
1112 goto drop;
1113
1114 skb_tx_timestamp(skb);
1115
1116 /* Orphan the skb - required as we might hang on to it
1117 * for indefinite time.
1118 */
1119 skb_orphan(skb);
1120
1121 nf_reset(skb);
1122
1123 if (ptr_ring_produce(&tfile->tx_ring, skb))
1124 goto drop;
1125
1126 /* Notify and wake up reader process */
1127 if (tfile->flags & TUN_FASYNC)
1128 kill_fasync(&tfile->fasync, SIGIO, POLL_IN);
1129 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
1130
1131 rcu_read_unlock();
1132 return NETDEV_TX_OK;
1133
1134 drop:
1135 this_cpu_inc(tun->pcpu_stats->tx_dropped);
1136 skb_tx_error(skb);
1137 kfree_skb(skb);
1138 rcu_read_unlock();
1139 return NET_XMIT_DROP;
1140 }
1141
1142 static void tun_net_mclist(struct net_device *dev)
1143 {
1144 /*
1145 * This callback is supposed to deal with mc filter in
1146 * _rx_ path and has nothing to do with the _tx_ path.
1147 * In rx path we always accept everything userspace gives us.
1148 */
1149 }
1150
1151 static netdev_features_t tun_net_fix_features(struct net_device *dev,
1152 netdev_features_t features)
1153 {
1154 struct tun_struct *tun = netdev_priv(dev);
1155
1156 return (features & tun->set_features) | (features & ~TUN_USER_FEATURES);
1157 }
1158
1159 static void tun_set_headroom(struct net_device *dev, int new_hr)
1160 {
1161 struct tun_struct *tun = netdev_priv(dev);
1162
1163 if (new_hr < NET_SKB_PAD)
1164 new_hr = NET_SKB_PAD;
1165
1166 tun->align = new_hr;
1167 }
1168
1169 static void
1170 tun_net_get_stats64(struct net_device *dev, struct rtnl_link_stats64 *stats)
1171 {
1172 u32 rx_dropped = 0, tx_dropped = 0, rx_frame_errors = 0;
1173 struct tun_struct *tun = netdev_priv(dev);
1174 struct tun_pcpu_stats *p;
1175 int i;
1176
1177 for_each_possible_cpu(i) {
1178 u64 rxpackets, rxbytes, txpackets, txbytes;
1179 unsigned int start;
1180
1181 p = per_cpu_ptr(tun->pcpu_stats, i);
1182 do {
1183 start = u64_stats_fetch_begin(&p->syncp);
1184 rxpackets = p->rx_packets;
1185 rxbytes = p->rx_bytes;
1186 txpackets = p->tx_packets;
1187 txbytes = p->tx_bytes;
1188 } while (u64_stats_fetch_retry(&p->syncp, start));
1189
1190 stats->rx_packets += rxpackets;
1191 stats->rx_bytes += rxbytes;
1192 stats->tx_packets += txpackets;
1193 stats->tx_bytes += txbytes;
1194
1195 /* u32 counters */
1196 rx_dropped += p->rx_dropped;
1197 rx_frame_errors += p->rx_frame_errors;
1198 tx_dropped += p->tx_dropped;
1199 }
1200 stats->rx_dropped = rx_dropped;
1201 stats->rx_frame_errors = rx_frame_errors;
1202 stats->tx_dropped = tx_dropped;
1203 }
1204
1205 static int tun_xdp_set(struct net_device *dev, struct bpf_prog *prog,
1206 struct netlink_ext_ack *extack)
1207 {
1208 struct tun_struct *tun = netdev_priv(dev);
1209 struct bpf_prog *old_prog;
1210
1211 old_prog = rtnl_dereference(tun->xdp_prog);
1212 rcu_assign_pointer(tun->xdp_prog, prog);
1213 if (old_prog)
1214 bpf_prog_put(old_prog);
1215
1216 return 0;
1217 }
1218
1219 static u32 tun_xdp_query(struct net_device *dev)
1220 {
1221 struct tun_struct *tun = netdev_priv(dev);
1222 const struct bpf_prog *xdp_prog;
1223
1224 xdp_prog = rtnl_dereference(tun->xdp_prog);
1225 if (xdp_prog)
1226 return xdp_prog->aux->id;
1227
1228 return 0;
1229 }
1230
1231 static int tun_xdp(struct net_device *dev, struct netdev_bpf *xdp)
1232 {
1233 switch (xdp->command) {
1234 case XDP_SETUP_PROG:
1235 return tun_xdp_set(dev, xdp->prog, xdp->extack);
1236 case XDP_QUERY_PROG:
1237 xdp->prog_id = tun_xdp_query(dev);
1238 return 0;
1239 default:
1240 return -EINVAL;
1241 }
1242 }
1243
1244 static const struct net_device_ops tun_netdev_ops = {
1245 .ndo_uninit = tun_net_uninit,
1246 .ndo_open = tun_net_open,
1247 .ndo_stop = tun_net_close,
1248 .ndo_start_xmit = tun_net_xmit,
1249 .ndo_fix_features = tun_net_fix_features,
1250 .ndo_select_queue = tun_select_queue,
1251 .ndo_set_rx_headroom = tun_set_headroom,
1252 .ndo_get_stats64 = tun_net_get_stats64,
1253 };
1254
1255 static void __tun_xdp_flush_tfile(struct tun_file *tfile)
1256 {
1257 /* Notify and wake up reader process */
1258 if (tfile->flags & TUN_FASYNC)
1259 kill_fasync(&tfile->fasync, SIGIO, POLL_IN);
1260 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
1261 }
1262
1263 static int tun_xdp_xmit(struct net_device *dev, int n,
1264 struct xdp_frame **frames, u32 flags)
1265 {
1266 struct tun_struct *tun = netdev_priv(dev);
1267 struct tun_file *tfile;
1268 u32 numqueues;
1269 int drops = 0;
1270 int cnt = n;
1271 int i;
1272
1273 if (unlikely(flags & ~XDP_XMIT_FLAGS_MASK))
1274 return -EINVAL;
1275
1276 rcu_read_lock();
1277
1278 resample:
1279 numqueues = READ_ONCE(tun->numqueues);
1280 if (!numqueues) {
1281 rcu_read_unlock();
1282 return -ENXIO; /* Caller will free/return all frames */
1283 }
1284
1285 tfile = rcu_dereference(tun->tfiles[smp_processor_id() %
1286 numqueues]);
1287 if (unlikely(!tfile))
1288 goto resample;
1289
1290 spin_lock(&tfile->tx_ring.producer_lock);
1291 for (i = 0; i < n; i++) {
1292 struct xdp_frame *xdp = frames[i];
1293 /* Encode the XDP flag into lowest bit for consumer to differ
1294 * XDP buffer from sk_buff.
1295 */
1296 void *frame = tun_xdp_to_ptr(xdp);
1297
1298 if (__ptr_ring_produce(&tfile->tx_ring, frame)) {
1299 this_cpu_inc(tun->pcpu_stats->tx_dropped);
1300 xdp_return_frame_rx_napi(xdp);
1301 drops++;
1302 }
1303 }
1304 spin_unlock(&tfile->tx_ring.producer_lock);
1305
1306 if (flags & XDP_XMIT_FLUSH)
1307 __tun_xdp_flush_tfile(tfile);
1308
1309 rcu_read_unlock();
1310 return cnt - drops;
1311 }
1312
1313 static int tun_xdp_tx(struct net_device *dev, struct xdp_buff *xdp)
1314 {
1315 struct xdp_frame *frame = convert_to_xdp_frame(xdp);
1316
1317 if (unlikely(!frame))
1318 return -EOVERFLOW;
1319
1320 return tun_xdp_xmit(dev, 1, &frame, XDP_XMIT_FLUSH);
1321 }
1322
1323 static const struct net_device_ops tap_netdev_ops = {
1324 .ndo_uninit = tun_net_uninit,
1325 .ndo_open = tun_net_open,
1326 .ndo_stop = tun_net_close,
1327 .ndo_start_xmit = tun_net_xmit,
1328 .ndo_fix_features = tun_net_fix_features,
1329 .ndo_set_rx_mode = tun_net_mclist,
1330 .ndo_set_mac_address = eth_mac_addr,
1331 .ndo_validate_addr = eth_validate_addr,
1332 .ndo_select_queue = tun_select_queue,
1333 .ndo_features_check = passthru_features_check,
1334 .ndo_set_rx_headroom = tun_set_headroom,
1335 .ndo_get_stats64 = tun_net_get_stats64,
1336 .ndo_bpf = tun_xdp,
1337 .ndo_xdp_xmit = tun_xdp_xmit,
1338 };
1339
1340 static void tun_flow_init(struct tun_struct *tun)
1341 {
1342 int i;
1343
1344 for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++)
1345 INIT_HLIST_HEAD(&tun->flows[i]);
1346
1347 tun->ageing_time = TUN_FLOW_EXPIRE;
1348 timer_setup(&tun->flow_gc_timer, tun_flow_cleanup, 0);
1349 mod_timer(&tun->flow_gc_timer,
1350 round_jiffies_up(jiffies + tun->ageing_time));
1351 }
1352
1353 static void tun_flow_uninit(struct tun_struct *tun)
1354 {
1355 del_timer_sync(&tun->flow_gc_timer);
1356 tun_flow_flush(tun);
1357 }
1358
1359 #define MIN_MTU 68
1360 #define MAX_MTU 65535
1361
1362 /* Initialize net device. */
1363 static void tun_net_init(struct net_device *dev)
1364 {
1365 struct tun_struct *tun = netdev_priv(dev);
1366
1367 switch (tun->flags & TUN_TYPE_MASK) {
1368 case IFF_TUN:
1369 dev->netdev_ops = &tun_netdev_ops;
1370
1371 /* Point-to-Point TUN Device */
1372 dev->hard_header_len = 0;
1373 dev->addr_len = 0;
1374 dev->mtu = 1500;
1375
1376 /* Zero header length */
1377 dev->type = ARPHRD_NONE;
1378 dev->flags = IFF_POINTOPOINT | IFF_NOARP | IFF_MULTICAST;
1379 break;
1380
1381 case IFF_TAP:
1382 dev->netdev_ops = &tap_netdev_ops;
1383 /* Ethernet TAP Device */
1384 ether_setup(dev);
1385 dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1386 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
1387
1388 eth_hw_addr_random(dev);
1389
1390 break;
1391 }
1392
1393 dev->min_mtu = MIN_MTU;
1394 dev->max_mtu = MAX_MTU - dev->hard_header_len;
1395 }
1396
1397 static bool tun_sock_writeable(struct tun_struct *tun, struct tun_file *tfile)
1398 {
1399 struct sock *sk = tfile->socket.sk;
1400
1401 return (tun->dev->flags & IFF_UP) && sock_writeable(sk);
1402 }
1403
1404 /* Character device part */
1405
1406 /* Poll */
1407 static __poll_t tun_chr_poll(struct file *file, poll_table *wait)
1408 {
1409 struct tun_file *tfile = file->private_data;
1410 struct tun_struct *tun = tun_get(tfile);
1411 struct sock *sk;
1412 __poll_t mask = 0;
1413
1414 if (!tun)
1415 return EPOLLERR;
1416
1417 sk = tfile->socket.sk;
1418
1419 tun_debug(KERN_INFO, tun, "tun_chr_poll\n");
1420
1421 poll_wait(file, sk_sleep(sk), wait);
1422
1423 if (!ptr_ring_empty(&tfile->tx_ring))
1424 mask |= EPOLLIN | EPOLLRDNORM;
1425
1426 /* Make sure SOCKWQ_ASYNC_NOSPACE is set if not writable to
1427 * guarantee EPOLLOUT to be raised by either here or
1428 * tun_sock_write_space(). Then process could get notification
1429 * after it writes to a down device and meets -EIO.
1430 */
1431 if (tun_sock_writeable(tun, tfile) ||
1432 (!test_and_set_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags) &&
1433 tun_sock_writeable(tun, tfile)))
1434 mask |= EPOLLOUT | EPOLLWRNORM;
1435
1436 if (tun->dev->reg_state != NETREG_REGISTERED)
1437 mask = EPOLLERR;
1438
1439 tun_put(tun);
1440 return mask;
1441 }
1442
1443 static struct sk_buff *tun_napi_alloc_frags(struct tun_file *tfile,
1444 size_t len,
1445 const struct iov_iter *it)
1446 {
1447 struct sk_buff *skb;
1448 size_t linear;
1449 int err;
1450 int i;
1451
1452 if (it->nr_segs > MAX_SKB_FRAGS + 1)
1453 return ERR_PTR(-ENOMEM);
1454
1455 local_bh_disable();
1456 skb = napi_get_frags(&tfile->napi);
1457 local_bh_enable();
1458 if (!skb)
1459 return ERR_PTR(-ENOMEM);
1460
1461 linear = iov_iter_single_seg_count(it);
1462 err = __skb_grow(skb, linear);
1463 if (err)
1464 goto free;
1465
1466 skb->len = len;
1467 skb->data_len = len - linear;
1468 skb->truesize += skb->data_len;
1469
1470 for (i = 1; i < it->nr_segs; i++) {
1471 struct page_frag *pfrag = &current->task_frag;
1472 size_t fragsz = it->iov[i].iov_len;
1473
1474 if (fragsz == 0 || fragsz > PAGE_SIZE) {
1475 err = -EINVAL;
1476 goto free;
1477 }
1478
1479 if (!skb_page_frag_refill(fragsz, pfrag, GFP_KERNEL)) {
1480 err = -ENOMEM;
1481 goto free;
1482 }
1483
1484 skb_fill_page_desc(skb, i - 1, pfrag->page,
1485 pfrag->offset, fragsz);
1486 page_ref_inc(pfrag->page);
1487 pfrag->offset += fragsz;
1488 }
1489
1490 return skb;
1491 free:
1492 /* frees skb and all frags allocated with napi_alloc_frag() */
1493 napi_free_frags(&tfile->napi);
1494 return ERR_PTR(err);
1495 }
1496
1497 /* prepad is the amount to reserve at front. len is length after that.
1498 * linear is a hint as to how much to copy (usually headers). */
1499 static struct sk_buff *tun_alloc_skb(struct tun_file *tfile,
1500 size_t prepad, size_t len,
1501 size_t linear, int noblock)
1502 {
1503 struct sock *sk = tfile->socket.sk;
1504 struct sk_buff *skb;
1505 int err;
1506
1507 /* Under a page? Don't bother with paged skb. */
1508 if (prepad + len < PAGE_SIZE || !linear)
1509 linear = len;
1510
1511 skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
1512 &err, 0);
1513 if (!skb)
1514 return ERR_PTR(err);
1515
1516 skb_reserve(skb, prepad);
1517 skb_put(skb, linear);
1518 skb->data_len = len - linear;
1519 skb->len += len - linear;
1520
1521 return skb;
1522 }
1523
1524 static void tun_rx_batched(struct tun_struct *tun, struct tun_file *tfile,
1525 struct sk_buff *skb, int more)
1526 {
1527 struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
1528 struct sk_buff_head process_queue;
1529 u32 rx_batched = tun->rx_batched;
1530 bool rcv = false;
1531
1532 if (!rx_batched || (!more && skb_queue_empty(queue))) {
1533 local_bh_disable();
1534 skb_record_rx_queue(skb, tfile->queue_index);
1535 netif_receive_skb(skb);
1536 local_bh_enable();
1537 return;
1538 }
1539
1540 spin_lock(&queue->lock);
1541 if (!more || skb_queue_len(queue) == rx_batched) {
1542 __skb_queue_head_init(&process_queue);
1543 skb_queue_splice_tail_init(queue, &process_queue);
1544 rcv = true;
1545 } else {
1546 __skb_queue_tail(queue, skb);
1547 }
1548 spin_unlock(&queue->lock);
1549
1550 if (rcv) {
1551 struct sk_buff *nskb;
1552
1553 local_bh_disable();
1554 while ((nskb = __skb_dequeue(&process_queue))) {
1555 skb_record_rx_queue(nskb, tfile->queue_index);
1556 netif_receive_skb(nskb);
1557 }
1558 skb_record_rx_queue(skb, tfile->queue_index);
1559 netif_receive_skb(skb);
1560 local_bh_enable();
1561 }
1562 }
1563
1564 static bool tun_can_build_skb(struct tun_struct *tun, struct tun_file *tfile,
1565 int len, int noblock, bool zerocopy)
1566 {
1567 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
1568 return false;
1569
1570 if (tfile->socket.sk->sk_sndbuf != INT_MAX)
1571 return false;
1572
1573 if (!noblock)
1574 return false;
1575
1576 if (zerocopy)
1577 return false;
1578
1579 if (SKB_DATA_ALIGN(len + TUN_RX_PAD) +
1580 SKB_DATA_ALIGN(sizeof(struct skb_shared_info)) > PAGE_SIZE)
1581 return false;
1582
1583 return true;
1584 }
1585
1586 static struct sk_buff *tun_build_skb(struct tun_struct *tun,
1587 struct tun_file *tfile,
1588 struct iov_iter *from,
1589 struct virtio_net_hdr *hdr,
1590 int len, int *skb_xdp)
1591 {
1592 struct page_frag *alloc_frag = &current->task_frag;
1593 struct sk_buff *skb;
1594 struct bpf_prog *xdp_prog;
1595 int buflen = SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
1596 unsigned int delta = 0;
1597 char *buf;
1598 size_t copied;
1599 int err, pad = TUN_RX_PAD;
1600
1601 rcu_read_lock();
1602 xdp_prog = rcu_dereference(tun->xdp_prog);
1603 if (xdp_prog)
1604 pad += TUN_HEADROOM;
1605 buflen += SKB_DATA_ALIGN(len + pad);
1606 rcu_read_unlock();
1607
1608 alloc_frag->offset = ALIGN((u64)alloc_frag->offset, SMP_CACHE_BYTES);
1609 if (unlikely(!skb_page_frag_refill(buflen, alloc_frag, GFP_KERNEL)))
1610 return ERR_PTR(-ENOMEM);
1611
1612 buf = (char *)page_address(alloc_frag->page) + alloc_frag->offset;
1613 copied = copy_page_from_iter(alloc_frag->page,
1614 alloc_frag->offset + pad,
1615 len, from);
1616 if (copied != len)
1617 return ERR_PTR(-EFAULT);
1618
1619 /* There's a small window that XDP may be set after the check
1620 * of xdp_prog above, this should be rare and for simplicity
1621 * we do XDP on skb in case the headroom is not enough.
1622 */
1623 if (hdr->gso_type || !xdp_prog)
1624 *skb_xdp = 1;
1625 else
1626 *skb_xdp = 0;
1627
1628 local_bh_disable();
1629 rcu_read_lock();
1630 xdp_prog = rcu_dereference(tun->xdp_prog);
1631 if (xdp_prog && !*skb_xdp) {
1632 struct xdp_buff xdp;
1633 void *orig_data;
1634 u32 act;
1635
1636 xdp.data_hard_start = buf;
1637 xdp.data = buf + pad;
1638 xdp_set_data_meta_invalid(&xdp);
1639 xdp.data_end = xdp.data + len;
1640 xdp.rxq = &tfile->xdp_rxq;
1641 orig_data = xdp.data;
1642 act = bpf_prog_run_xdp(xdp_prog, &xdp);
1643
1644 switch (act) {
1645 case XDP_REDIRECT:
1646 get_page(alloc_frag->page);
1647 alloc_frag->offset += buflen;
1648 err = xdp_do_redirect(tun->dev, &xdp, xdp_prog);
1649 xdp_do_flush_map();
1650 if (err)
1651 goto err_redirect;
1652 rcu_read_unlock();
1653 local_bh_enable();
1654 return NULL;
1655 case XDP_TX:
1656 get_page(alloc_frag->page);
1657 alloc_frag->offset += buflen;
1658 if (tun_xdp_tx(tun->dev, &xdp) < 0)
1659 goto err_redirect;
1660 rcu_read_unlock();
1661 local_bh_enable();
1662 return NULL;
1663 case XDP_PASS:
1664 delta = orig_data - xdp.data;
1665 len = xdp.data_end - xdp.data;
1666 break;
1667 default:
1668 bpf_warn_invalid_xdp_action(act);
1669 /* fall through */
1670 case XDP_ABORTED:
1671 trace_xdp_exception(tun->dev, xdp_prog, act);
1672 /* fall through */
1673 case XDP_DROP:
1674 goto err_xdp;
1675 }
1676 }
1677
1678 skb = build_skb(buf, buflen);
1679 if (!skb) {
1680 rcu_read_unlock();
1681 local_bh_enable();
1682 return ERR_PTR(-ENOMEM);
1683 }
1684
1685 skb_reserve(skb, pad - delta);
1686 skb_put(skb, len);
1687 skb_set_owner_w(skb, tfile->socket.sk);
1688 get_page(alloc_frag->page);
1689 alloc_frag->offset += buflen;
1690
1691 rcu_read_unlock();
1692 local_bh_enable();
1693
1694 return skb;
1695
1696 err_redirect:
1697 put_page(alloc_frag->page);
1698 err_xdp:
1699 rcu_read_unlock();
1700 local_bh_enable();
1701 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1702 return NULL;
1703 }
1704
1705 /* Get packet from user space buffer */
1706 static ssize_t tun_get_user(struct tun_struct *tun, struct tun_file *tfile,
1707 void *msg_control, struct iov_iter *from,
1708 int noblock, bool more)
1709 {
1710 struct tun_pi pi = { 0, cpu_to_be16(ETH_P_IP) };
1711 struct sk_buff *skb;
1712 size_t total_len = iov_iter_count(from);
1713 size_t len = total_len, align = tun->align, linear;
1714 struct virtio_net_hdr gso = { 0 };
1715 struct tun_pcpu_stats *stats;
1716 int good_linear;
1717 int copylen;
1718 bool zerocopy = false;
1719 int err;
1720 u32 rxhash = 0;
1721 int skb_xdp = 1;
1722 bool frags = tun_napi_frags_enabled(tfile);
1723
1724 if (!(tun->flags & IFF_NO_PI)) {
1725 if (len < sizeof(pi))
1726 return -EINVAL;
1727 len -= sizeof(pi);
1728
1729 if (!copy_from_iter_full(&pi, sizeof(pi), from))
1730 return -EFAULT;
1731 }
1732
1733 if (tun->flags & IFF_VNET_HDR) {
1734 int vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
1735
1736 if (len < vnet_hdr_sz)
1737 return -EINVAL;
1738 len -= vnet_hdr_sz;
1739
1740 if (!copy_from_iter_full(&gso, sizeof(gso), from))
1741 return -EFAULT;
1742
1743 if ((gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
1744 tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2 > tun16_to_cpu(tun, gso.hdr_len))
1745 gso.hdr_len = cpu_to_tun16(tun, tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2);
1746
1747 if (tun16_to_cpu(tun, gso.hdr_len) > len)
1748 return -EINVAL;
1749 iov_iter_advance(from, vnet_hdr_sz - sizeof(gso));
1750 }
1751
1752 if ((tun->flags & TUN_TYPE_MASK) == IFF_TAP) {
1753 align += NET_IP_ALIGN;
1754 if (unlikely(len < ETH_HLEN ||
1755 (gso.hdr_len && tun16_to_cpu(tun, gso.hdr_len) < ETH_HLEN)))
1756 return -EINVAL;
1757 }
1758
1759 good_linear = SKB_MAX_HEAD(align);
1760
1761 if (msg_control) {
1762 struct iov_iter i = *from;
1763
1764 /* There are 256 bytes to be copied in skb, so there is
1765 * enough room for skb expand head in case it is used.
1766 * The rest of the buffer is mapped from userspace.
1767 */
1768 copylen = gso.hdr_len ? tun16_to_cpu(tun, gso.hdr_len) : GOODCOPY_LEN;
1769 if (copylen > good_linear)
1770 copylen = good_linear;
1771 linear = copylen;
1772 iov_iter_advance(&i, copylen);
1773 if (iov_iter_npages(&i, INT_MAX) <= MAX_SKB_FRAGS)
1774 zerocopy = true;
1775 }
1776
1777 if (!frags && tun_can_build_skb(tun, tfile, len, noblock, zerocopy)) {
1778 /* For the packet that is not easy to be processed
1779 * (e.g gso or jumbo packet), we will do it at after
1780 * skb was created with generic XDP routine.
1781 */
1782 skb = tun_build_skb(tun, tfile, from, &gso, len, &skb_xdp);
1783 if (IS_ERR(skb)) {
1784 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1785 return PTR_ERR(skb);
1786 }
1787 if (!skb)
1788 return total_len;
1789 } else {
1790 if (!zerocopy) {
1791 copylen = len;
1792 if (tun16_to_cpu(tun, gso.hdr_len) > good_linear)
1793 linear = good_linear;
1794 else
1795 linear = tun16_to_cpu(tun, gso.hdr_len);
1796 }
1797
1798 if (frags) {
1799 mutex_lock(&tfile->napi_mutex);
1800 skb = tun_napi_alloc_frags(tfile, copylen, from);
1801 /* tun_napi_alloc_frags() enforces a layout for the skb.
1802 * If zerocopy is enabled, then this layout will be
1803 * overwritten by zerocopy_sg_from_iter().
1804 */
1805 zerocopy = false;
1806 } else {
1807 skb = tun_alloc_skb(tfile, align, copylen, linear,
1808 noblock);
1809 }
1810
1811 if (IS_ERR(skb)) {
1812 if (PTR_ERR(skb) != -EAGAIN)
1813 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1814 if (frags)
1815 mutex_unlock(&tfile->napi_mutex);
1816 return PTR_ERR(skb);
1817 }
1818
1819 if (zerocopy)
1820 err = zerocopy_sg_from_iter(skb, from);
1821 else
1822 err = skb_copy_datagram_from_iter(skb, 0, from, len);
1823
1824 if (err) {
1825 err = -EFAULT;
1826 drop:
1827 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1828 kfree_skb(skb);
1829 if (frags) {
1830 tfile->napi.skb = NULL;
1831 mutex_unlock(&tfile->napi_mutex);
1832 }
1833
1834 return err;
1835 }
1836 }
1837
1838 if (virtio_net_hdr_to_skb(skb, &gso, tun_is_little_endian(tun))) {
1839 this_cpu_inc(tun->pcpu_stats->rx_frame_errors);
1840 kfree_skb(skb);
1841 if (frags) {
1842 tfile->napi.skb = NULL;
1843 mutex_unlock(&tfile->napi_mutex);
1844 }
1845
1846 return -EINVAL;
1847 }
1848
1849 switch (tun->flags & TUN_TYPE_MASK) {
1850 case IFF_TUN:
1851 if (tun->flags & IFF_NO_PI) {
1852 u8 ip_version = skb->len ? (skb->data[0] >> 4) : 0;
1853
1854 switch (ip_version) {
1855 case 4:
1856 pi.proto = htons(ETH_P_IP);
1857 break;
1858 case 6:
1859 pi.proto = htons(ETH_P_IPV6);
1860 break;
1861 default:
1862 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1863 kfree_skb(skb);
1864 return -EINVAL;
1865 }
1866 }
1867
1868 skb_reset_mac_header(skb);
1869 skb->protocol = pi.proto;
1870 skb->dev = tun->dev;
1871 break;
1872 case IFF_TAP:
1873 if (frags && !pskb_may_pull(skb, ETH_HLEN)) {
1874 err = -ENOMEM;
1875 goto drop;
1876 }
1877 skb->protocol = eth_type_trans(skb, tun->dev);
1878 break;
1879 }
1880
1881 /* copy skb_ubuf_info for callback when skb has no error */
1882 if (zerocopy) {
1883 skb_shinfo(skb)->destructor_arg = msg_control;
1884 skb_shinfo(skb)->tx_flags |= SKBTX_DEV_ZEROCOPY;
1885 skb_shinfo(skb)->tx_flags |= SKBTX_SHARED_FRAG;
1886 } else if (msg_control) {
1887 struct ubuf_info *uarg = msg_control;
1888 uarg->callback(uarg, false);
1889 }
1890
1891 skb_reset_network_header(skb);
1892 skb_probe_transport_header(skb, 0);
1893
1894 if (skb_xdp) {
1895 struct bpf_prog *xdp_prog;
1896 int ret;
1897
1898 local_bh_disable();
1899 rcu_read_lock();
1900 xdp_prog = rcu_dereference(tun->xdp_prog);
1901 if (xdp_prog) {
1902 ret = do_xdp_generic(xdp_prog, skb);
1903 if (ret != XDP_PASS) {
1904 rcu_read_unlock();
1905 local_bh_enable();
1906 if (frags) {
1907 tfile->napi.skb = NULL;
1908 mutex_unlock(&tfile->napi_mutex);
1909 }
1910 return total_len;
1911 }
1912 }
1913 rcu_read_unlock();
1914 local_bh_enable();
1915 }
1916
1917 /* Compute the costly rx hash only if needed for flow updates.
1918 * We may get a very small possibility of OOO during switching, not
1919 * worth to optimize.
1920 */
1921 if (!rcu_access_pointer(tun->steering_prog) && tun->numqueues > 1 &&
1922 !tfile->detached)
1923 rxhash = __skb_get_hash_symmetric(skb);
1924
1925 rcu_read_lock();
1926 if (unlikely(!(tun->dev->flags & IFF_UP))) {
1927 err = -EIO;
1928 rcu_read_unlock();
1929 goto drop;
1930 }
1931
1932 if (frags) {
1933 u32 headlen;
1934
1935 /* Exercise flow dissector code path. */
1936 skb_push(skb, ETH_HLEN);
1937 headlen = eth_get_headlen(skb->data, skb_headlen(skb));
1938
1939 if (unlikely(headlen > skb_headlen(skb))) {
1940 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1941 napi_free_frags(&tfile->napi);
1942 rcu_read_unlock();
1943 mutex_unlock(&tfile->napi_mutex);
1944 WARN_ON(1);
1945 return -ENOMEM;
1946 }
1947
1948 local_bh_disable();
1949 napi_gro_frags(&tfile->napi);
1950 local_bh_enable();
1951 mutex_unlock(&tfile->napi_mutex);
1952 } else if (tfile->napi_enabled) {
1953 struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
1954 int queue_len;
1955
1956 spin_lock_bh(&queue->lock);
1957 __skb_queue_tail(queue, skb);
1958 queue_len = skb_queue_len(queue);
1959 spin_unlock(&queue->lock);
1960
1961 if (!more || queue_len > NAPI_POLL_WEIGHT)
1962 napi_schedule(&tfile->napi);
1963
1964 local_bh_enable();
1965 } else if (!IS_ENABLED(CONFIG_4KSTACKS)) {
1966 tun_rx_batched(tun, tfile, skb, more);
1967 } else {
1968 netif_rx_ni(skb);
1969 }
1970 rcu_read_unlock();
1971
1972 stats = get_cpu_ptr(tun->pcpu_stats);
1973 u64_stats_update_begin(&stats->syncp);
1974 stats->rx_packets++;
1975 stats->rx_bytes += len;
1976 u64_stats_update_end(&stats->syncp);
1977 put_cpu_ptr(stats);
1978
1979 if (rxhash)
1980 tun_flow_update(tun, rxhash, tfile);
1981
1982 return total_len;
1983 }
1984
1985 static ssize_t tun_chr_write_iter(struct kiocb *iocb, struct iov_iter *from)
1986 {
1987 struct file *file = iocb->ki_filp;
1988 struct tun_file *tfile = file->private_data;
1989 struct tun_struct *tun = tun_get(tfile);
1990 ssize_t result;
1991
1992 if (!tun)
1993 return -EBADFD;
1994
1995 result = tun_get_user(tun, tfile, NULL, from,
1996 file->f_flags & O_NONBLOCK, false);
1997
1998 tun_put(tun);
1999 return result;
2000 }
2001
2002 static ssize_t tun_put_user_xdp(struct tun_struct *tun,
2003 struct tun_file *tfile,
2004 struct xdp_frame *xdp_frame,
2005 struct iov_iter *iter)
2006 {
2007 int vnet_hdr_sz = 0;
2008 size_t size = xdp_frame->len;
2009 struct tun_pcpu_stats *stats;
2010 size_t ret;
2011
2012 if (tun->flags & IFF_VNET_HDR) {
2013 struct virtio_net_hdr gso = { 0 };
2014
2015 vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
2016 if (unlikely(iov_iter_count(iter) < vnet_hdr_sz))
2017 return -EINVAL;
2018 if (unlikely(copy_to_iter(&gso, sizeof(gso), iter) !=
2019 sizeof(gso)))
2020 return -EFAULT;
2021 iov_iter_advance(iter, vnet_hdr_sz - sizeof(gso));
2022 }
2023
2024 ret = copy_to_iter(xdp_frame->data, size, iter) + vnet_hdr_sz;
2025
2026 stats = get_cpu_ptr(tun->pcpu_stats);
2027 u64_stats_update_begin(&stats->syncp);
2028 stats->tx_packets++;
2029 stats->tx_bytes += ret;
2030 u64_stats_update_end(&stats->syncp);
2031 put_cpu_ptr(tun->pcpu_stats);
2032
2033 return ret;
2034 }
2035
2036 /* Put packet to the user space buffer */
2037 static ssize_t tun_put_user(struct tun_struct *tun,
2038 struct tun_file *tfile,
2039 struct sk_buff *skb,
2040 struct iov_iter *iter)
2041 {
2042 struct tun_pi pi = { 0, skb->protocol };
2043 struct tun_pcpu_stats *stats;
2044 ssize_t total;
2045 int vlan_offset = 0;
2046 int vlan_hlen = 0;
2047 int vnet_hdr_sz = 0;
2048
2049 if (skb_vlan_tag_present(skb))
2050 vlan_hlen = VLAN_HLEN;
2051
2052 if (tun->flags & IFF_VNET_HDR)
2053 vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
2054
2055 total = skb->len + vlan_hlen + vnet_hdr_sz;
2056
2057 if (!(tun->flags & IFF_NO_PI)) {
2058 if (iov_iter_count(iter) < sizeof(pi))
2059 return -EINVAL;
2060
2061 total += sizeof(pi);
2062 if (iov_iter_count(iter) < total) {
2063 /* Packet will be striped */
2064 pi.flags |= TUN_PKT_STRIP;
2065 }
2066
2067 if (copy_to_iter(&pi, sizeof(pi), iter) != sizeof(pi))
2068 return -EFAULT;
2069 }
2070
2071 if (vnet_hdr_sz) {
2072 struct virtio_net_hdr gso;
2073
2074 if (iov_iter_count(iter) < vnet_hdr_sz)
2075 return -EINVAL;
2076
2077 if (virtio_net_hdr_from_skb(skb, &gso,
2078 tun_is_little_endian(tun), true,
2079 vlan_hlen)) {
2080 struct skb_shared_info *sinfo = skb_shinfo(skb);
2081 pr_err("unexpected GSO type: "
2082 "0x%x, gso_size %d, hdr_len %d\n",
2083 sinfo->gso_type, tun16_to_cpu(tun, gso.gso_size),
2084 tun16_to_cpu(tun, gso.hdr_len));
2085 print_hex_dump(KERN_ERR, "tun: ",
2086 DUMP_PREFIX_NONE,
2087 16, 1, skb->head,
2088 min((int)tun16_to_cpu(tun, gso.hdr_len), 64), true);
2089 WARN_ON_ONCE(1);
2090 return -EINVAL;
2091 }
2092
2093 if (copy_to_iter(&gso, sizeof(gso), iter) != sizeof(gso))
2094 return -EFAULT;
2095
2096 iov_iter_advance(iter, vnet_hdr_sz - sizeof(gso));
2097 }
2098
2099 if (vlan_hlen) {
2100 int ret;
2101 struct veth veth;
2102
2103 veth.h_vlan_proto = skb->vlan_proto;
2104 veth.h_vlan_TCI = htons(skb_vlan_tag_get(skb));
2105
2106 vlan_offset = offsetof(struct vlan_ethhdr, h_vlan_proto);
2107
2108 ret = skb_copy_datagram_iter(skb, 0, iter, vlan_offset);
2109 if (ret || !iov_iter_count(iter))
2110 goto done;
2111
2112 ret = copy_to_iter(&veth, sizeof(veth), iter);
2113 if (ret != sizeof(veth) || !iov_iter_count(iter))
2114 goto done;
2115 }
2116
2117 skb_copy_datagram_iter(skb, vlan_offset, iter, skb->len - vlan_offset);
2118
2119 done:
2120 /* caller is in process context, */
2121 stats = get_cpu_ptr(tun->pcpu_stats);
2122 u64_stats_update_begin(&stats->syncp);
2123 stats->tx_packets++;
2124 stats->tx_bytes += skb->len + vlan_hlen;
2125 u64_stats_update_end(&stats->syncp);
2126 put_cpu_ptr(tun->pcpu_stats);
2127
2128 return total;
2129 }
2130
2131 static void *tun_ring_recv(struct tun_file *tfile, int noblock, int *err)
2132 {
2133 DECLARE_WAITQUEUE(wait, current);
2134 void *ptr = NULL;
2135 int error = 0;
2136
2137 ptr = ptr_ring_consume(&tfile->tx_ring);
2138 if (ptr)
2139 goto out;
2140 if (noblock) {
2141 error = -EAGAIN;
2142 goto out;
2143 }
2144
2145 add_wait_queue(&tfile->wq.wait, &wait);
2146
2147 while (1) {
2148 set_current_state(TASK_INTERRUPTIBLE);
2149 ptr = ptr_ring_consume(&tfile->tx_ring);
2150 if (ptr)
2151 break;
2152 if (signal_pending(current)) {
2153 error = -ERESTARTSYS;
2154 break;
2155 }
2156 if (tfile->socket.sk->sk_shutdown & RCV_SHUTDOWN) {
2157 error = -EFAULT;
2158 break;
2159 }
2160
2161 schedule();
2162 }
2163
2164 __set_current_state(TASK_RUNNING);
2165 remove_wait_queue(&tfile->wq.wait, &wait);
2166
2167 out:
2168 *err = error;
2169 return ptr;
2170 }
2171
2172 static ssize_t tun_do_read(struct tun_struct *tun, struct tun_file *tfile,
2173 struct iov_iter *to,
2174 int noblock, void *ptr)
2175 {
2176 ssize_t ret;
2177 int err;
2178
2179 tun_debug(KERN_INFO, tun, "tun_do_read\n");
2180
2181 if (!iov_iter_count(to)) {
2182 tun_ptr_free(ptr);
2183 return 0;
2184 }
2185
2186 if (!ptr) {
2187 /* Read frames from ring */
2188 ptr = tun_ring_recv(tfile, noblock, &err);
2189 if (!ptr)
2190 return err;
2191 }
2192
2193 if (tun_is_xdp_frame(ptr)) {
2194 struct xdp_frame *xdpf = tun_ptr_to_xdp(ptr);
2195
2196 ret = tun_put_user_xdp(tun, tfile, xdpf, to);
2197 xdp_return_frame(xdpf);
2198 } else {
2199 struct sk_buff *skb = ptr;
2200
2201 ret = tun_put_user(tun, tfile, skb, to);
2202 if (unlikely(ret < 0))
2203 kfree_skb(skb);
2204 else
2205 consume_skb(skb);
2206 }
2207
2208 return ret;
2209 }
2210
2211 static ssize_t tun_chr_read_iter(struct kiocb *iocb, struct iov_iter *to)
2212 {
2213 struct file *file = iocb->ki_filp;
2214 struct tun_file *tfile = file->private_data;
2215 struct tun_struct *tun = tun_get(tfile);
2216 ssize_t len = iov_iter_count(to), ret;
2217
2218 if (!tun)
2219 return -EBADFD;
2220 ret = tun_do_read(tun, tfile, to, file->f_flags & O_NONBLOCK, NULL);
2221 ret = min_t(ssize_t, ret, len);
2222 if (ret > 0)
2223 iocb->ki_pos = ret;
2224 tun_put(tun);
2225 return ret;
2226 }
2227
2228 static void tun_prog_free(struct rcu_head *rcu)
2229 {
2230 struct tun_prog *prog = container_of(rcu, struct tun_prog, rcu);
2231
2232 bpf_prog_destroy(prog->prog);
2233 kfree(prog);
2234 }
2235
2236 static int __tun_set_ebpf(struct tun_struct *tun,
2237 struct tun_prog __rcu **prog_p,
2238 struct bpf_prog *prog)
2239 {
2240 struct tun_prog *old, *new = NULL;
2241
2242 if (prog) {
2243 new = kmalloc(sizeof(*new), GFP_KERNEL);
2244 if (!new)
2245 return -ENOMEM;
2246 new->prog = prog;
2247 }
2248
2249 spin_lock_bh(&tun->lock);
2250 old = rcu_dereference_protected(*prog_p,
2251 lockdep_is_held(&tun->lock));
2252 rcu_assign_pointer(*prog_p, new);
2253 spin_unlock_bh(&tun->lock);
2254
2255 if (old)
2256 call_rcu(&old->rcu, tun_prog_free);
2257
2258 return 0;
2259 }
2260
2261 static void tun_free_netdev(struct net_device *dev)
2262 {
2263 struct tun_struct *tun = netdev_priv(dev);
2264
2265 BUG_ON(!(list_empty(&tun->disabled)));
2266 free_percpu(tun->pcpu_stats);
2267 tun_flow_uninit(tun);
2268 security_tun_dev_free_security(tun->security);
2269 __tun_set_ebpf(tun, &tun->steering_prog, NULL);
2270 __tun_set_ebpf(tun, &tun->filter_prog, NULL);
2271 }
2272
2273 static void tun_setup(struct net_device *dev)
2274 {
2275 struct tun_struct *tun = netdev_priv(dev);
2276
2277 tun->owner = INVALID_UID;
2278 tun->group = INVALID_GID;
2279 tun_default_link_ksettings(dev, &tun->link_ksettings);
2280
2281 dev->ethtool_ops = &tun_ethtool_ops;
2282 dev->needs_free_netdev = true;
2283 dev->priv_destructor = tun_free_netdev;
2284 /* We prefer our own queue length */
2285 dev->tx_queue_len = TUN_READQ_SIZE;
2286 }
2287
2288 /* Trivial set of netlink ops to allow deleting tun or tap
2289 * device with netlink.
2290 */
2291 static int tun_validate(struct nlattr *tb[], struct nlattr *data[],
2292 struct netlink_ext_ack *extack)
2293 {
2294 NL_SET_ERR_MSG(extack,
2295 "tun/tap creation via rtnetlink is not supported.");
2296 return -EOPNOTSUPP;
2297 }
2298
2299 static size_t tun_get_size(const struct net_device *dev)
2300 {
2301 BUILD_BUG_ON(sizeof(u32) != sizeof(uid_t));
2302 BUILD_BUG_ON(sizeof(u32) != sizeof(gid_t));
2303
2304 return nla_total_size(sizeof(uid_t)) + /* OWNER */
2305 nla_total_size(sizeof(gid_t)) + /* GROUP */
2306 nla_total_size(sizeof(u8)) + /* TYPE */
2307 nla_total_size(sizeof(u8)) + /* PI */
2308 nla_total_size(sizeof(u8)) + /* VNET_HDR */
2309 nla_total_size(sizeof(u8)) + /* PERSIST */
2310 nla_total_size(sizeof(u8)) + /* MULTI_QUEUE */
2311 nla_total_size(sizeof(u32)) + /* NUM_QUEUES */
2312 nla_total_size(sizeof(u32)) + /* NUM_DISABLED_QUEUES */
2313 0;
2314 }
2315
2316 static int tun_fill_info(struct sk_buff *skb, const struct net_device *dev)
2317 {
2318 struct tun_struct *tun = netdev_priv(dev);
2319
2320 if (nla_put_u8(skb, IFLA_TUN_TYPE, tun->flags & TUN_TYPE_MASK))
2321 goto nla_put_failure;
2322 if (uid_valid(tun->owner) &&
2323 nla_put_u32(skb, IFLA_TUN_OWNER,
2324 from_kuid_munged(current_user_ns(), tun->owner)))
2325 goto nla_put_failure;
2326 if (gid_valid(tun->group) &&
2327 nla_put_u32(skb, IFLA_TUN_GROUP,
2328 from_kgid_munged(current_user_ns(), tun->group)))
2329 goto nla_put_failure;
2330 if (nla_put_u8(skb, IFLA_TUN_PI, !(tun->flags & IFF_NO_PI)))
2331 goto nla_put_failure;
2332 if (nla_put_u8(skb, IFLA_TUN_VNET_HDR, !!(tun->flags & IFF_VNET_HDR)))
2333 goto nla_put_failure;
2334 if (nla_put_u8(skb, IFLA_TUN_PERSIST, !!(tun->flags & IFF_PERSIST)))
2335 goto nla_put_failure;
2336 if (nla_put_u8(skb, IFLA_TUN_MULTI_QUEUE,
2337 !!(tun->flags & IFF_MULTI_QUEUE)))
2338 goto nla_put_failure;
2339 if (tun->flags & IFF_MULTI_QUEUE) {
2340 if (nla_put_u32(skb, IFLA_TUN_NUM_QUEUES, tun->numqueues))
2341 goto nla_put_failure;
2342 if (nla_put_u32(skb, IFLA_TUN_NUM_DISABLED_QUEUES,
2343 tun->numdisabled))
2344 goto nla_put_failure;
2345 }
2346
2347 return 0;
2348
2349 nla_put_failure:
2350 return -EMSGSIZE;
2351 }
2352
2353 static struct rtnl_link_ops tun_link_ops __read_mostly = {
2354 .kind = DRV_NAME,
2355 .priv_size = sizeof(struct tun_struct),
2356 .setup = tun_setup,
2357 .validate = tun_validate,
2358 .get_size = tun_get_size,
2359 .fill_info = tun_fill_info,
2360 };
2361
2362 static void tun_sock_write_space(struct sock *sk)
2363 {
2364 struct tun_file *tfile;
2365 wait_queue_head_t *wqueue;
2366
2367 if (!sock_writeable(sk))
2368 return;
2369
2370 if (!test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags))
2371 return;
2372
2373 wqueue = sk_sleep(sk);
2374 if (wqueue && waitqueue_active(wqueue))
2375 wake_up_interruptible_sync_poll(wqueue, EPOLLOUT |
2376 EPOLLWRNORM | EPOLLWRBAND);
2377
2378 tfile = container_of(sk, struct tun_file, sk);
2379 kill_fasync(&tfile->fasync, SIGIO, POLL_OUT);
2380 }
2381
2382 static int tun_sendmsg(struct socket *sock, struct msghdr *m, size_t total_len)
2383 {
2384 int ret;
2385 struct tun_file *tfile = container_of(sock, struct tun_file, socket);
2386 struct tun_struct *tun = tun_get(tfile);
2387
2388 if (!tun)
2389 return -EBADFD;
2390
2391 ret = tun_get_user(tun, tfile, m->msg_control, &m->msg_iter,
2392 m->msg_flags & MSG_DONTWAIT,
2393 m->msg_flags & MSG_MORE);
2394 tun_put(tun);
2395 return ret;
2396 }
2397
2398 static int tun_recvmsg(struct socket *sock, struct msghdr *m, size_t total_len,
2399 int flags)
2400 {
2401 struct tun_file *tfile = container_of(sock, struct tun_file, socket);
2402 struct tun_struct *tun = tun_get(tfile);
2403 void *ptr = m->msg_control;
2404 int ret;
2405
2406 if (!tun) {
2407 ret = -EBADFD;
2408 goto out_free;
2409 }
2410
2411 if (flags & ~(MSG_DONTWAIT|MSG_TRUNC|MSG_ERRQUEUE)) {
2412 ret = -EINVAL;
2413 goto out_put_tun;
2414 }
2415 if (flags & MSG_ERRQUEUE) {
2416 ret = sock_recv_errqueue(sock->sk, m, total_len,
2417 SOL_PACKET, TUN_TX_TIMESTAMP);
2418 goto out;
2419 }
2420 ret = tun_do_read(tun, tfile, &m->msg_iter, flags & MSG_DONTWAIT, ptr);
2421 if (ret > (ssize_t)total_len) {
2422 m->msg_flags |= MSG_TRUNC;
2423 ret = flags & MSG_TRUNC ? ret : total_len;
2424 }
2425 out:
2426 tun_put(tun);
2427 return ret;
2428
2429 out_put_tun:
2430 tun_put(tun);
2431 out_free:
2432 tun_ptr_free(ptr);
2433 return ret;
2434 }
2435
2436 static int tun_ptr_peek_len(void *ptr)
2437 {
2438 if (likely(ptr)) {
2439 if (tun_is_xdp_frame(ptr)) {
2440 struct xdp_frame *xdpf = tun_ptr_to_xdp(ptr);
2441
2442 return xdpf->len;
2443 }
2444 return __skb_array_len_with_tag(ptr);
2445 } else {
2446 return 0;
2447 }
2448 }
2449
2450 static int tun_peek_len(struct socket *sock)
2451 {
2452 struct tun_file *tfile = container_of(sock, struct tun_file, socket);
2453 struct tun_struct *tun;
2454 int ret = 0;
2455
2456 tun = tun_get(tfile);
2457 if (!tun)
2458 return 0;
2459
2460 ret = PTR_RING_PEEK_CALL(&tfile->tx_ring, tun_ptr_peek_len);
2461 tun_put(tun);
2462
2463 return ret;
2464 }
2465
2466 /* Ops structure to mimic raw sockets with tun */
2467 static const struct proto_ops tun_socket_ops = {
2468 .peek_len = tun_peek_len,
2469 .sendmsg = tun_sendmsg,
2470 .recvmsg = tun_recvmsg,
2471 };
2472
2473 static struct proto tun_proto = {
2474 .name = "tun",
2475 .owner = THIS_MODULE,
2476 .obj_size = sizeof(struct tun_file),
2477 };
2478
2479 static int tun_flags(struct tun_struct *tun)
2480 {
2481 return tun->flags & (TUN_FEATURES | IFF_PERSIST | IFF_TUN | IFF_TAP);
2482 }
2483
2484 static ssize_t tun_show_flags(struct device *dev, struct device_attribute *attr,
2485 char *buf)
2486 {
2487 struct tun_struct *tun = netdev_priv(to_net_dev(dev));
2488 return sprintf(buf, "0x%x\n", tun_flags(tun));
2489 }
2490
2491 static ssize_t tun_show_owner(struct device *dev, struct device_attribute *attr,
2492 char *buf)
2493 {
2494 struct tun_struct *tun = netdev_priv(to_net_dev(dev));
2495 return uid_valid(tun->owner)?
2496 sprintf(buf, "%u\n",
2497 from_kuid_munged(current_user_ns(), tun->owner)):
2498 sprintf(buf, "-1\n");
2499 }
2500
2501 static ssize_t tun_show_group(struct device *dev, struct device_attribute *attr,
2502 char *buf)
2503 {
2504 struct tun_struct *tun = netdev_priv(to_net_dev(dev));
2505 return gid_valid(tun->group) ?
2506 sprintf(buf, "%u\n",
2507 from_kgid_munged(current_user_ns(), tun->group)):
2508 sprintf(buf, "-1\n");
2509 }
2510
2511 static DEVICE_ATTR(tun_flags, 0444, tun_show_flags, NULL);
2512 static DEVICE_ATTR(owner, 0444, tun_show_owner, NULL);
2513 static DEVICE_ATTR(group, 0444, tun_show_group, NULL);
2514
2515 static struct attribute *tun_dev_attrs[] = {
2516 &dev_attr_tun_flags.attr,
2517 &dev_attr_owner.attr,
2518 &dev_attr_group.attr,
2519 NULL
2520 };
2521
2522 static const struct attribute_group tun_attr_group = {
2523 .attrs = tun_dev_attrs
2524 };
2525
2526 static int tun_set_iff(struct net *net, struct file *file, struct ifreq *ifr)
2527 {
2528 struct tun_struct *tun;
2529 struct tun_file *tfile = file->private_data;
2530 struct net_device *dev;
2531 int err;
2532
2533 if (tfile->detached)
2534 return -EINVAL;
2535
2536 if ((ifr->ifr_flags & IFF_NAPI_FRAGS)) {
2537 if (!capable(CAP_NET_ADMIN))
2538 return -EPERM;
2539
2540 if (!(ifr->ifr_flags & IFF_NAPI) ||
2541 (ifr->ifr_flags & TUN_TYPE_MASK) != IFF_TAP)
2542 return -EINVAL;
2543 }
2544
2545 dev = __dev_get_by_name(net, ifr->ifr_name);
2546 if (dev) {
2547 if (ifr->ifr_flags & IFF_TUN_EXCL)
2548 return -EBUSY;
2549 if ((ifr->ifr_flags & IFF_TUN) && dev->netdev_ops == &tun_netdev_ops)
2550 tun = netdev_priv(dev);
2551 else if ((ifr->ifr_flags & IFF_TAP) && dev->netdev_ops == &tap_netdev_ops)
2552 tun = netdev_priv(dev);
2553 else
2554 return -EINVAL;
2555
2556 if (!!(ifr->ifr_flags & IFF_MULTI_QUEUE) !=
2557 !!(tun->flags & IFF_MULTI_QUEUE))
2558 return -EINVAL;
2559
2560 if (tun_not_capable(tun))
2561 return -EPERM;
2562 err = security_tun_dev_open(tun->security);
2563 if (err < 0)
2564 return err;
2565
2566 err = tun_attach(tun, file, ifr->ifr_flags & IFF_NOFILTER,
2567 ifr->ifr_flags & IFF_NAPI,
2568 ifr->ifr_flags & IFF_NAPI_FRAGS, true);
2569 if (err < 0)
2570 return err;
2571
2572 if (tun->flags & IFF_MULTI_QUEUE &&
2573 (tun->numqueues + tun->numdisabled > 1)) {
2574 /* One or more queue has already been attached, no need
2575 * to initialize the device again.
2576 */
2577 netdev_state_change(dev);
2578 return 0;
2579 }
2580
2581 tun->flags = (tun->flags & ~TUN_FEATURES) |
2582 (ifr->ifr_flags & TUN_FEATURES);
2583
2584 netdev_state_change(dev);
2585 } else {
2586 char *name;
2587 unsigned long flags = 0;
2588 int queues = ifr->ifr_flags & IFF_MULTI_QUEUE ?
2589 MAX_TAP_QUEUES : 1;
2590
2591 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
2592 return -EPERM;
2593 err = security_tun_dev_create();
2594 if (err < 0)
2595 return err;
2596
2597 /* Set dev type */
2598 if (ifr->ifr_flags & IFF_TUN) {
2599 /* TUN device */
2600 flags |= IFF_TUN;
2601 name = "tun%d";
2602 } else if (ifr->ifr_flags & IFF_TAP) {
2603 /* TAP device */
2604 flags |= IFF_TAP;
2605 name = "tap%d";
2606 } else
2607 return -EINVAL;
2608
2609 if (*ifr->ifr_name)
2610 name = ifr->ifr_name;
2611
2612 dev = alloc_netdev_mqs(sizeof(struct tun_struct), name,
2613 NET_NAME_UNKNOWN, tun_setup, queues,
2614 queues);
2615
2616 if (!dev)
2617 return -ENOMEM;
2618 err = dev_get_valid_name(net, dev, name);
2619 if (err < 0)
2620 goto err_free_dev;
2621
2622 dev_net_set(dev, net);
2623 dev->rtnl_link_ops = &tun_link_ops;
2624 dev->ifindex = tfile->ifindex;
2625 dev->sysfs_groups[0] = &tun_attr_group;
2626
2627 tun = netdev_priv(dev);
2628 tun->dev = dev;
2629 tun->flags = flags;
2630 tun->txflt.count = 0;
2631 tun->vnet_hdr_sz = sizeof(struct virtio_net_hdr);
2632
2633 tun->align = NET_SKB_PAD;
2634 tun->filter_attached = false;
2635 tun->sndbuf = tfile->socket.sk->sk_sndbuf;
2636 tun->rx_batched = 0;
2637 RCU_INIT_POINTER(tun->steering_prog, NULL);
2638
2639 tun->pcpu_stats = netdev_alloc_pcpu_stats(struct tun_pcpu_stats);
2640 if (!tun->pcpu_stats) {
2641 err = -ENOMEM;
2642 goto err_free_dev;
2643 }
2644
2645 spin_lock_init(&tun->lock);
2646
2647 err = security_tun_dev_alloc_security(&tun->security);
2648 if (err < 0)
2649 goto err_free_stat;
2650
2651 tun_net_init(dev);
2652 tun_flow_init(tun);
2653
2654 dev->hw_features = NETIF_F_SG | NETIF_F_FRAGLIST |
2655 TUN_USER_FEATURES | NETIF_F_HW_VLAN_CTAG_TX |
2656 NETIF_F_HW_VLAN_STAG_TX;
2657 dev->features = dev->hw_features | NETIF_F_LLTX;
2658 dev->vlan_features = dev->features &
2659 ~(NETIF_F_HW_VLAN_CTAG_TX |
2660 NETIF_F_HW_VLAN_STAG_TX);
2661
2662 tun->flags = (tun->flags & ~TUN_FEATURES) |
2663 (ifr->ifr_flags & TUN_FEATURES);
2664
2665 INIT_LIST_HEAD(&tun->disabled);
2666 err = tun_attach(tun, file, false, ifr->ifr_flags & IFF_NAPI,
2667 ifr->ifr_flags & IFF_NAPI_FRAGS, false);
2668 if (err < 0)
2669 goto err_free_flow;
2670
2671 err = register_netdevice(tun->dev);
2672 if (err < 0)
2673 goto err_detach;
2674 /* free_netdev() won't check refcnt, to aovid race
2675 * with dev_put() we need publish tun after registration.
2676 */
2677 rcu_assign_pointer(tfile->tun, tun);
2678 }
2679
2680 netif_carrier_on(tun->dev);
2681
2682 tun_debug(KERN_INFO, tun, "tun_set_iff\n");
2683
2684 /* Make sure persistent devices do not get stuck in
2685 * xoff state.
2686 */
2687 if (netif_running(tun->dev))
2688 netif_tx_wake_all_queues(tun->dev);
2689
2690 strcpy(ifr->ifr_name, tun->dev->name);
2691 return 0;
2692
2693 err_detach:
2694 tun_detach_all(dev);
2695 /* register_netdevice() already called tun_free_netdev() */
2696 goto err_free_dev;
2697
2698 err_free_flow:
2699 tun_flow_uninit(tun);
2700 security_tun_dev_free_security(tun->security);
2701 err_free_stat:
2702 free_percpu(tun->pcpu_stats);
2703 err_free_dev:
2704 free_netdev(dev);
2705 return err;
2706 }
2707
2708 static void tun_get_iff(struct net *net, struct tun_struct *tun,
2709 struct ifreq *ifr)
2710 {
2711 tun_debug(KERN_INFO, tun, "tun_get_iff\n");
2712
2713 strcpy(ifr->ifr_name, tun->dev->name);
2714
2715 ifr->ifr_flags = tun_flags(tun);
2716
2717 }
2718
2719 /* This is like a cut-down ethtool ops, except done via tun fd so no
2720 * privs required. */
2721 static int set_offload(struct tun_struct *tun, unsigned long arg)
2722 {
2723 netdev_features_t features = 0;
2724
2725 if (arg & TUN_F_CSUM) {
2726 features |= NETIF_F_HW_CSUM;
2727 arg &= ~TUN_F_CSUM;
2728
2729 if (arg & (TUN_F_TSO4|TUN_F_TSO6)) {
2730 if (arg & TUN_F_TSO_ECN) {
2731 features |= NETIF_F_TSO_ECN;
2732 arg &= ~TUN_F_TSO_ECN;
2733 }
2734 if (arg & TUN_F_TSO4)
2735 features |= NETIF_F_TSO;
2736 if (arg & TUN_F_TSO6)
2737 features |= NETIF_F_TSO6;
2738 arg &= ~(TUN_F_TSO4|TUN_F_TSO6);
2739 }
2740
2741 arg &= ~TUN_F_UFO;
2742 }
2743
2744 /* This gives the user a way to test for new features in future by
2745 * trying to set them. */
2746 if (arg)
2747 return -EINVAL;
2748
2749 tun->set_features = features;
2750 tun->dev->wanted_features &= ~TUN_USER_FEATURES;
2751 tun->dev->wanted_features |= features;
2752 netdev_update_features(tun->dev);
2753
2754 return 0;
2755 }
2756
2757 static void tun_detach_filter(struct tun_struct *tun, int n)
2758 {
2759 int i;
2760 struct tun_file *tfile;
2761
2762 for (i = 0; i < n; i++) {
2763 tfile = rtnl_dereference(tun->tfiles[i]);
2764 lock_sock(tfile->socket.sk);
2765 sk_detach_filter(tfile->socket.sk);
2766 release_sock(tfile->socket.sk);
2767 }
2768
2769 tun->filter_attached = false;
2770 }
2771
2772 static int tun_attach_filter(struct tun_struct *tun)
2773 {
2774 int i, ret = 0;
2775 struct tun_file *tfile;
2776
2777 for (i = 0; i < tun->numqueues; i++) {
2778 tfile = rtnl_dereference(tun->tfiles[i]);
2779 lock_sock(tfile->socket.sk);
2780 ret = sk_attach_filter(&tun->fprog, tfile->socket.sk);
2781 release_sock(tfile->socket.sk);
2782 if (ret) {
2783 tun_detach_filter(tun, i);
2784 return ret;
2785 }
2786 }
2787
2788 tun->filter_attached = true;
2789 return ret;
2790 }
2791
2792 static void tun_set_sndbuf(struct tun_struct *tun)
2793 {
2794 struct tun_file *tfile;
2795 int i;
2796
2797 for (i = 0; i < tun->numqueues; i++) {
2798 tfile = rtnl_dereference(tun->tfiles[i]);
2799 tfile->socket.sk->sk_sndbuf = tun->sndbuf;
2800 }
2801 }
2802
2803 static int tun_set_queue(struct file *file, struct ifreq *ifr)
2804 {
2805 struct tun_file *tfile = file->private_data;
2806 struct tun_struct *tun;
2807 int ret = 0;
2808
2809 rtnl_lock();
2810
2811 if (ifr->ifr_flags & IFF_ATTACH_QUEUE) {
2812 tun = tfile->detached;
2813 if (!tun) {
2814 ret = -EINVAL;
2815 goto unlock;
2816 }
2817 ret = security_tun_dev_attach_queue(tun->security);
2818 if (ret < 0)
2819 goto unlock;
2820 ret = tun_attach(tun, file, false, tun->flags & IFF_NAPI,
2821 tun->flags & IFF_NAPI_FRAGS, true);
2822 } else if (ifr->ifr_flags & IFF_DETACH_QUEUE) {
2823 tun = rtnl_dereference(tfile->tun);
2824 if (!tun || !(tun->flags & IFF_MULTI_QUEUE) || tfile->detached)
2825 ret = -EINVAL;
2826 else
2827 __tun_detach(tfile, false);
2828 } else
2829 ret = -EINVAL;
2830
2831 if (ret >= 0)
2832 netdev_state_change(tun->dev);
2833
2834 unlock:
2835 rtnl_unlock();
2836 return ret;
2837 }
2838
2839 static int tun_set_ebpf(struct tun_struct *tun, struct tun_prog **prog_p,
2840 void __user *data)
2841 {
2842 struct bpf_prog *prog;
2843 int fd;
2844
2845 if (copy_from_user(&fd, data, sizeof(fd)))
2846 return -EFAULT;
2847
2848 if (fd == -1) {
2849 prog = NULL;
2850 } else {
2851 prog = bpf_prog_get_type(fd, BPF_PROG_TYPE_SOCKET_FILTER);
2852 if (IS_ERR(prog))
2853 return PTR_ERR(prog);
2854 }
2855
2856 return __tun_set_ebpf(tun, prog_p, prog);
2857 }
2858
2859 static long __tun_chr_ioctl(struct file *file, unsigned int cmd,
2860 unsigned long arg, int ifreq_len)
2861 {
2862 struct tun_file *tfile = file->private_data;
2863 struct net *net = sock_net(&tfile->sk);
2864 struct tun_struct *tun;
2865 void __user* argp = (void __user*)arg;
2866 struct ifreq ifr;
2867 kuid_t owner;
2868 kgid_t group;
2869 int sndbuf;
2870 int vnet_hdr_sz;
2871 unsigned int ifindex;
2872 int le;
2873 int ret;
2874 bool do_notify = false;
2875
2876 if (cmd == TUNSETIFF || cmd == TUNSETQUEUE ||
2877 (_IOC_TYPE(cmd) == SOCK_IOC_TYPE && cmd != SIOCGSKNS)) {
2878 if (copy_from_user(&ifr, argp, ifreq_len))
2879 return -EFAULT;
2880 } else {
2881 memset(&ifr, 0, sizeof(ifr));
2882 }
2883 if (cmd == TUNGETFEATURES) {
2884 /* Currently this just means: "what IFF flags are valid?".
2885 * This is needed because we never checked for invalid flags on
2886 * TUNSETIFF.
2887 */
2888 return put_user(IFF_TUN | IFF_TAP | TUN_FEATURES,
2889 (unsigned int __user*)argp);
2890 } else if (cmd == TUNSETQUEUE) {
2891 return tun_set_queue(file, &ifr);
2892 } else if (cmd == SIOCGSKNS) {
2893 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
2894 return -EPERM;
2895 return open_related_ns(&net->ns, get_net_ns);
2896 }
2897
2898 ret = 0;
2899 rtnl_lock();
2900
2901 tun = tun_get(tfile);
2902 if (cmd == TUNSETIFF) {
2903 ret = -EEXIST;
2904 if (tun)
2905 goto unlock;
2906
2907 ifr.ifr_name[IFNAMSIZ-1] = '\0';
2908
2909 ret = tun_set_iff(net, file, &ifr);
2910
2911 if (ret)
2912 goto unlock;
2913
2914 if (copy_to_user(argp, &ifr, ifreq_len))
2915 ret = -EFAULT;
2916 goto unlock;
2917 }
2918 if (cmd == TUNSETIFINDEX) {
2919 ret = -EPERM;
2920 if (tun)
2921 goto unlock;
2922
2923 ret = -EFAULT;
2924 if (copy_from_user(&ifindex, argp, sizeof(ifindex)))
2925 goto unlock;
2926
2927 ret = 0;
2928 tfile->ifindex = ifindex;
2929 goto unlock;
2930 }
2931
2932 ret = -EBADFD;
2933 if (!tun)
2934 goto unlock;
2935
2936 tun_debug(KERN_INFO, tun, "tun_chr_ioctl cmd %u\n", cmd);
2937
2938 ret = 0;
2939 switch (cmd) {
2940 case TUNGETIFF:
2941 tun_get_iff(current->nsproxy->net_ns, tun, &ifr);
2942
2943 if (tfile->detached)
2944 ifr.ifr_flags |= IFF_DETACH_QUEUE;
2945 if (!tfile->socket.sk->sk_filter)
2946 ifr.ifr_flags |= IFF_NOFILTER;
2947
2948 if (copy_to_user(argp, &ifr, ifreq_len))
2949 ret = -EFAULT;
2950 break;
2951
2952 case TUNSETNOCSUM:
2953 /* Disable/Enable checksum */
2954
2955 /* [unimplemented] */
2956 tun_debug(KERN_INFO, tun, "ignored: set checksum %s\n",
2957 arg ? "disabled" : "enabled");
2958 break;
2959
2960 case TUNSETPERSIST:
2961 /* Disable/Enable persist mode. Keep an extra reference to the
2962 * module to prevent the module being unprobed.
2963 */
2964 if (arg && !(tun->flags & IFF_PERSIST)) {
2965 tun->flags |= IFF_PERSIST;
2966 __module_get(THIS_MODULE);
2967 do_notify = true;
2968 }
2969 if (!arg && (tun->flags & IFF_PERSIST)) {
2970 tun->flags &= ~IFF_PERSIST;
2971 module_put(THIS_MODULE);
2972 do_notify = true;
2973 }
2974
2975 tun_debug(KERN_INFO, tun, "persist %s\n",
2976 arg ? "enabled" : "disabled");
2977 break;
2978
2979 case TUNSETOWNER:
2980 /* Set owner of the device */
2981 owner = make_kuid(current_user_ns(), arg);
2982 if (!uid_valid(owner)) {
2983 ret = -EINVAL;
2984 break;
2985 }
2986 tun->owner = owner;
2987 do_notify = true;
2988 tun_debug(KERN_INFO, tun, "owner set to %u\n",
2989 from_kuid(&init_user_ns, tun->owner));
2990 break;
2991
2992 case TUNSETGROUP:
2993 /* Set group of the device */
2994 group = make_kgid(current_user_ns(), arg);
2995 if (!gid_valid(group)) {
2996 ret = -EINVAL;
2997 break;
2998 }
2999 tun->group = group;
3000 do_notify = true;
3001 tun_debug(KERN_INFO, tun, "group set to %u\n",
3002 from_kgid(&init_user_ns, tun->group));
3003 break;
3004
3005 case TUNSETLINK:
3006 /* Only allow setting the type when the interface is down */
3007 if (tun->dev->flags & IFF_UP) {
3008 tun_debug(KERN_INFO, tun,
3009 "Linktype set failed because interface is up\n");
3010 ret = -EBUSY;
3011 } else {
3012 tun->dev->type = (int) arg;
3013 tun_debug(KERN_INFO, tun, "linktype set to %d\n",
3014 tun->dev->type);
3015 ret = 0;
3016 }
3017 break;
3018
3019 #ifdef TUN_DEBUG
3020 case TUNSETDEBUG:
3021 tun->debug = arg;
3022 break;
3023 #endif
3024 case TUNSETOFFLOAD:
3025 ret = set_offload(tun, arg);
3026 break;
3027
3028 case TUNSETTXFILTER:
3029 /* Can be set only for TAPs */
3030 ret = -EINVAL;
3031 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3032 break;
3033 ret = update_filter(&tun->txflt, (void __user *)arg);
3034 break;
3035
3036 case SIOCGIFHWADDR:
3037 /* Get hw address */
3038 memcpy(ifr.ifr_hwaddr.sa_data, tun->dev->dev_addr, ETH_ALEN);
3039 ifr.ifr_hwaddr.sa_family = tun->dev->type;
3040 if (copy_to_user(argp, &ifr, ifreq_len))
3041 ret = -EFAULT;
3042 break;
3043
3044 case SIOCSIFHWADDR:
3045 /* Set hw address */
3046 tun_debug(KERN_DEBUG, tun, "set hw address: %pM\n",
3047 ifr.ifr_hwaddr.sa_data);
3048
3049 ret = dev_set_mac_address(tun->dev, &ifr.ifr_hwaddr);
3050 break;
3051
3052 case TUNGETSNDBUF:
3053 sndbuf = tfile->socket.sk->sk_sndbuf;
3054 if (copy_to_user(argp, &sndbuf, sizeof(sndbuf)))
3055 ret = -EFAULT;
3056 break;
3057
3058 case TUNSETSNDBUF:
3059 if (copy_from_user(&sndbuf, argp, sizeof(sndbuf))) {
3060 ret = -EFAULT;
3061 break;
3062 }
3063 if (sndbuf <= 0) {
3064 ret = -EINVAL;
3065 break;
3066 }
3067
3068 tun->sndbuf = sndbuf;
3069 tun_set_sndbuf(tun);
3070 break;
3071
3072 case TUNGETVNETHDRSZ:
3073 vnet_hdr_sz = tun->vnet_hdr_sz;
3074 if (copy_to_user(argp, &vnet_hdr_sz, sizeof(vnet_hdr_sz)))
3075 ret = -EFAULT;
3076 break;
3077
3078 case TUNSETVNETHDRSZ:
3079 if (copy_from_user(&vnet_hdr_sz, argp, sizeof(vnet_hdr_sz))) {
3080 ret = -EFAULT;
3081 break;
3082 }
3083 if (vnet_hdr_sz < (int)sizeof(struct virtio_net_hdr)) {
3084 ret = -EINVAL;
3085 break;
3086 }
3087
3088 tun->vnet_hdr_sz = vnet_hdr_sz;
3089 break;
3090
3091 case TUNGETVNETLE:
3092 le = !!(tun->flags & TUN_VNET_LE);
3093 if (put_user(le, (int __user *)argp))
3094 ret = -EFAULT;
3095 break;
3096
3097 case TUNSETVNETLE:
3098 if (get_user(le, (int __user *)argp)) {
3099 ret = -EFAULT;
3100 break;
3101 }
3102 if (le)
3103 tun->flags |= TUN_VNET_LE;
3104 else
3105 tun->flags &= ~TUN_VNET_LE;
3106 break;
3107
3108 case TUNGETVNETBE:
3109 ret = tun_get_vnet_be(tun, argp);
3110 break;
3111
3112 case TUNSETVNETBE:
3113 ret = tun_set_vnet_be(tun, argp);
3114 break;
3115
3116 case TUNATTACHFILTER:
3117 /* Can be set only for TAPs */
3118 ret = -EINVAL;
3119 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3120 break;
3121 ret = -EFAULT;
3122 if (copy_from_user(&tun->fprog, argp, sizeof(tun->fprog)))
3123 break;
3124
3125 ret = tun_attach_filter(tun);
3126 break;
3127
3128 case TUNDETACHFILTER:
3129 /* Can be set only for TAPs */
3130 ret = -EINVAL;
3131 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3132 break;
3133 ret = 0;
3134 tun_detach_filter(tun, tun->numqueues);
3135 break;
3136
3137 case TUNGETFILTER:
3138 ret = -EINVAL;
3139 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3140 break;
3141 ret = -EFAULT;
3142 if (copy_to_user(argp, &tun->fprog, sizeof(tun->fprog)))
3143 break;
3144 ret = 0;
3145 break;
3146
3147 case TUNSETSTEERINGEBPF:
3148 ret = tun_set_ebpf(tun, &tun->steering_prog, argp);
3149 break;
3150
3151 case TUNSETFILTEREBPF:
3152 ret = tun_set_ebpf(tun, &tun->filter_prog, argp);
3153 break;
3154
3155 default:
3156 ret = -EINVAL;
3157 break;
3158 }
3159
3160 if (do_notify)
3161 netdev_state_change(tun->dev);
3162
3163 unlock:
3164 rtnl_unlock();
3165 if (tun)
3166 tun_put(tun);
3167 return ret;
3168 }
3169
3170 static long tun_chr_ioctl(struct file *file,
3171 unsigned int cmd, unsigned long arg)
3172 {
3173 return __tun_chr_ioctl(file, cmd, arg, sizeof (struct ifreq));
3174 }
3175
3176 #ifdef CONFIG_COMPAT
3177 static long tun_chr_compat_ioctl(struct file *file,
3178 unsigned int cmd, unsigned long arg)
3179 {
3180 switch (cmd) {
3181 case TUNSETIFF:
3182 case TUNGETIFF:
3183 case TUNSETTXFILTER:
3184 case TUNGETSNDBUF:
3185 case TUNSETSNDBUF:
3186 case SIOCGIFHWADDR:
3187 case SIOCSIFHWADDR:
3188 arg = (unsigned long)compat_ptr(arg);
3189 break;
3190 default:
3191 arg = (compat_ulong_t)arg;
3192 break;
3193 }
3194
3195 /*
3196 * compat_ifreq is shorter than ifreq, so we must not access beyond
3197 * the end of that structure. All fields that are used in this
3198 * driver are compatible though, we don't need to convert the
3199 * contents.
3200 */
3201 return __tun_chr_ioctl(file, cmd, arg, sizeof(struct compat_ifreq));
3202 }
3203 #endif /* CONFIG_COMPAT */
3204
3205 static int tun_chr_fasync(int fd, struct file *file, int on)
3206 {
3207 struct tun_file *tfile = file->private_data;
3208 int ret;
3209
3210 if ((ret = fasync_helper(fd, file, on, &tfile->fasync)) < 0)
3211 goto out;
3212
3213 if (on) {
3214 __f_setown(file, task_pid(current), PIDTYPE_TGID, 0);
3215 tfile->flags |= TUN_FASYNC;
3216 } else
3217 tfile->flags &= ~TUN_FASYNC;
3218 ret = 0;
3219 out:
3220 return ret;
3221 }
3222
3223 static int tun_chr_open(struct inode *inode, struct file * file)
3224 {
3225 struct net *net = current->nsproxy->net_ns;
3226 struct tun_file *tfile;
3227
3228 DBG1(KERN_INFO, "tunX: tun_chr_open\n");
3229
3230 tfile = (struct tun_file *)sk_alloc(net, AF_UNSPEC, GFP_KERNEL,
3231 &tun_proto, 0);
3232 if (!tfile)
3233 return -ENOMEM;
3234 if (ptr_ring_init(&tfile->tx_ring, 0, GFP_KERNEL)) {
3235 sk_free(&tfile->sk);
3236 return -ENOMEM;
3237 }
3238
3239 mutex_init(&tfile->napi_mutex);
3240 RCU_INIT_POINTER(tfile->tun, NULL);
3241 tfile->flags = 0;
3242 tfile->ifindex = 0;
3243
3244 init_waitqueue_head(&tfile->wq.wait);
3245 RCU_INIT_POINTER(tfile->socket.wq, &tfile->wq);
3246
3247 tfile->socket.file = file;
3248 tfile->socket.ops = &tun_socket_ops;
3249
3250 sock_init_data(&tfile->socket, &tfile->sk);
3251
3252 tfile->sk.sk_write_space = tun_sock_write_space;
3253 tfile->sk.sk_sndbuf = INT_MAX;
3254
3255 file->private_data = tfile;
3256 INIT_LIST_HEAD(&tfile->next);
3257
3258 sock_set_flag(&tfile->sk, SOCK_ZEROCOPY);
3259
3260 return 0;
3261 }
3262
3263 static int tun_chr_close(struct inode *inode, struct file *file)
3264 {
3265 struct tun_file *tfile = file->private_data;
3266
3267 tun_detach(tfile, true);
3268
3269 return 0;
3270 }
3271
3272 #ifdef CONFIG_PROC_FS
3273 static void tun_chr_show_fdinfo(struct seq_file *m, struct file *file)
3274 {
3275 struct tun_file *tfile = file->private_data;
3276 struct tun_struct *tun;
3277 struct ifreq ifr;
3278
3279 memset(&ifr, 0, sizeof(ifr));
3280
3281 rtnl_lock();
3282 tun = tun_get(tfile);
3283 if (tun)
3284 tun_get_iff(current->nsproxy->net_ns, tun, &ifr);
3285 rtnl_unlock();
3286
3287 if (tun)
3288 tun_put(tun);
3289
3290 seq_printf(m, "iff:\t%s\n", ifr.ifr_name);
3291 }
3292 #endif
3293
3294 static const struct file_operations tun_fops = {
3295 .owner = THIS_MODULE,
3296 .llseek = no_llseek,
3297 .read_iter = tun_chr_read_iter,
3298 .write_iter = tun_chr_write_iter,
3299 .poll = tun_chr_poll,
3300 .unlocked_ioctl = tun_chr_ioctl,
3301 #ifdef CONFIG_COMPAT
3302 .compat_ioctl = tun_chr_compat_ioctl,
3303 #endif
3304 .open = tun_chr_open,
3305 .release = tun_chr_close,
3306 .fasync = tun_chr_fasync,
3307 #ifdef CONFIG_PROC_FS
3308 .show_fdinfo = tun_chr_show_fdinfo,
3309 #endif
3310 };
3311
3312 static struct miscdevice tun_miscdev = {
3313 .minor = TUN_MINOR,
3314 .name = "tun",
3315 .nodename = "net/tun",
3316 .fops = &tun_fops,
3317 };
3318
3319 /* ethtool interface */
3320
3321 static void tun_default_link_ksettings(struct net_device *dev,
3322 struct ethtool_link_ksettings *cmd)
3323 {
3324 ethtool_link_ksettings_zero_link_mode(cmd, supported);
3325 ethtool_link_ksettings_zero_link_mode(cmd, advertising);
3326 cmd->base.speed = SPEED_10;
3327 cmd->base.duplex = DUPLEX_FULL;
3328 cmd->base.port = PORT_TP;
3329 cmd->base.phy_address = 0;
3330 cmd->base.autoneg = AUTONEG_DISABLE;
3331 }
3332
3333 static int tun_get_link_ksettings(struct net_device *dev,
3334 struct ethtool_link_ksettings *cmd)
3335 {
3336 struct tun_struct *tun = netdev_priv(dev);
3337
3338 memcpy(cmd, &tun->link_ksettings, sizeof(*cmd));
3339 return 0;
3340 }
3341
3342 static int tun_set_link_ksettings(struct net_device *dev,
3343 const struct ethtool_link_ksettings *cmd)
3344 {
3345 struct tun_struct *tun = netdev_priv(dev);
3346
3347 memcpy(&tun->link_ksettings, cmd, sizeof(*cmd));
3348 return 0;
3349 }
3350
3351 static void tun_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
3352 {
3353 struct tun_struct *tun = netdev_priv(dev);
3354
3355 strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
3356 strlcpy(info->version, DRV_VERSION, sizeof(info->version));
3357
3358 switch (tun->flags & TUN_TYPE_MASK) {
3359 case IFF_TUN:
3360 strlcpy(info->bus_info, "tun", sizeof(info->bus_info));
3361 break;
3362 case IFF_TAP:
3363 strlcpy(info->bus_info, "tap", sizeof(info->bus_info));
3364 break;
3365 }
3366 }
3367
3368 static u32 tun_get_msglevel(struct net_device *dev)
3369 {
3370 #ifdef TUN_DEBUG
3371 struct tun_struct *tun = netdev_priv(dev);
3372 return tun->debug;
3373 #else
3374 return -EOPNOTSUPP;
3375 #endif
3376 }
3377
3378 static void tun_set_msglevel(struct net_device *dev, u32 value)
3379 {
3380 #ifdef TUN_DEBUG
3381 struct tun_struct *tun = netdev_priv(dev);
3382 tun->debug = value;
3383 #endif
3384 }
3385
3386 static int tun_get_coalesce(struct net_device *dev,
3387 struct ethtool_coalesce *ec)
3388 {
3389 struct tun_struct *tun = netdev_priv(dev);
3390
3391 ec->rx_max_coalesced_frames = tun->rx_batched;
3392
3393 return 0;
3394 }
3395
3396 static int tun_set_coalesce(struct net_device *dev,
3397 struct ethtool_coalesce *ec)
3398 {
3399 struct tun_struct *tun = netdev_priv(dev);
3400
3401 if (ec->rx_max_coalesced_frames > NAPI_POLL_WEIGHT)
3402 tun->rx_batched = NAPI_POLL_WEIGHT;
3403 else
3404 tun->rx_batched = ec->rx_max_coalesced_frames;
3405
3406 return 0;
3407 }
3408
3409 static const struct ethtool_ops tun_ethtool_ops = {
3410 .get_drvinfo = tun_get_drvinfo,
3411 .get_msglevel = tun_get_msglevel,
3412 .set_msglevel = tun_set_msglevel,
3413 .get_link = ethtool_op_get_link,
3414 .get_ts_info = ethtool_op_get_ts_info,
3415 .get_coalesce = tun_get_coalesce,
3416 .set_coalesce = tun_set_coalesce,
3417 .get_link_ksettings = tun_get_link_ksettings,
3418 .set_link_ksettings = tun_set_link_ksettings,
3419 };
3420
3421 static int tun_queue_resize(struct tun_struct *tun)
3422 {
3423 struct net_device *dev = tun->dev;
3424 struct tun_file *tfile;
3425 struct ptr_ring **rings;
3426 int n = tun->numqueues + tun->numdisabled;
3427 int ret, i;
3428
3429 rings = kmalloc_array(n, sizeof(*rings), GFP_KERNEL);
3430 if (!rings)
3431 return -ENOMEM;
3432
3433 for (i = 0; i < tun->numqueues; i++) {
3434 tfile = rtnl_dereference(tun->tfiles[i]);
3435 rings[i] = &tfile->tx_ring;
3436 }
3437 list_for_each_entry(tfile, &tun->disabled, next)
3438 rings[i++] = &tfile->tx_ring;
3439
3440 ret = ptr_ring_resize_multiple(rings, n,
3441 dev->tx_queue_len, GFP_KERNEL,
3442 tun_ptr_free);
3443
3444 kfree(rings);
3445 return ret;
3446 }
3447
3448 static int tun_device_event(struct notifier_block *unused,
3449 unsigned long event, void *ptr)
3450 {
3451 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
3452 struct tun_struct *tun = netdev_priv(dev);
3453 int i;
3454
3455 if (dev->rtnl_link_ops != &tun_link_ops)
3456 return NOTIFY_DONE;
3457
3458 switch (event) {
3459 case NETDEV_CHANGE_TX_QUEUE_LEN:
3460 if (tun_queue_resize(tun))
3461 return NOTIFY_BAD;
3462 break;
3463 case NETDEV_UP:
3464 for (i = 0; i < tun->numqueues; i++) {
3465 struct tun_file *tfile;
3466
3467 tfile = rtnl_dereference(tun->tfiles[i]);
3468 tfile->socket.sk->sk_write_space(tfile->socket.sk);
3469 }
3470 break;
3471 default:
3472 break;
3473 }
3474
3475 return NOTIFY_DONE;
3476 }
3477
3478 static struct notifier_block tun_notifier_block __read_mostly = {
3479 .notifier_call = tun_device_event,
3480 };
3481
3482 static int __init tun_init(void)
3483 {
3484 int ret = 0;
3485
3486 pr_info("%s, %s\n", DRV_DESCRIPTION, DRV_VERSION);
3487
3488 ret = rtnl_link_register(&tun_link_ops);
3489 if (ret) {
3490 pr_err("Can't register link_ops\n");
3491 goto err_linkops;
3492 }
3493
3494 ret = misc_register(&tun_miscdev);
3495 if (ret) {
3496 pr_err("Can't register misc device %d\n", TUN_MINOR);
3497 goto err_misc;
3498 }
3499
3500 ret = register_netdevice_notifier(&tun_notifier_block);
3501 if (ret) {
3502 pr_err("Can't register netdevice notifier\n");
3503 goto err_notifier;
3504 }
3505
3506 return 0;
3507
3508 err_notifier:
3509 misc_deregister(&tun_miscdev);
3510 err_misc:
3511 rtnl_link_unregister(&tun_link_ops);
3512 err_linkops:
3513 return ret;
3514 }
3515
3516 static void tun_cleanup(void)
3517 {
3518 misc_deregister(&tun_miscdev);
3519 rtnl_link_unregister(&tun_link_ops);
3520 unregister_netdevice_notifier(&tun_notifier_block);
3521 }
3522
3523 /* Get an underlying socket object from tun file. Returns error unless file is
3524 * attached to a device. The returned object works like a packet socket, it
3525 * can be used for sock_sendmsg/sock_recvmsg. The caller is responsible for
3526 * holding a reference to the file for as long as the socket is in use. */
3527 struct socket *tun_get_socket(struct file *file)
3528 {
3529 struct tun_file *tfile;
3530 if (file->f_op != &tun_fops)
3531 return ERR_PTR(-EINVAL);
3532 tfile = file->private_data;
3533 if (!tfile)
3534 return ERR_PTR(-EBADFD);
3535 return &tfile->socket;
3536 }
3537 EXPORT_SYMBOL_GPL(tun_get_socket);
3538
3539 struct ptr_ring *tun_get_tx_ring(struct file *file)
3540 {
3541 struct tun_file *tfile;
3542
3543 if (file->f_op != &tun_fops)
3544 return ERR_PTR(-EINVAL);
3545 tfile = file->private_data;
3546 if (!tfile)
3547 return ERR_PTR(-EBADFD);
3548 return &tfile->tx_ring;
3549 }
3550 EXPORT_SYMBOL_GPL(tun_get_tx_ring);
3551
3552 module_init(tun_init);
3553 module_exit(tun_cleanup);
3554 MODULE_DESCRIPTION(DRV_DESCRIPTION);
3555 MODULE_AUTHOR(DRV_COPYRIGHT);
3556 MODULE_LICENSE("GPL");
3557 MODULE_ALIAS_MISCDEV(TUN_MINOR);
3558 MODULE_ALIAS("devname:net/tun");
Linux with added FPC-III support