?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Merge branch 'x86-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel...
[cris-mirror.git] / drivers / net / tun.c
blobb52258c327d2e1d7c7476de345e49f082909c246
1 /*
2 * TUN - Universal TUN/TAP device driver.
3 * Copyright (C) 1999-2002 Maxim Krasnyansky <maxk@qualcomm.com>
4 *
5 * This program is free software; you can redistribute it and/or modify
6 * it under the terms of the GNU General Public License as published by
7 * the Free Software Foundation; either version 2 of the License, or
8 * (at your option) any later version.
9 *
10 * This program is distributed in the hope that it will be useful,
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
13 * GNU General Public License for more details.
14 *
15 * $Id: tun.c,v 1.15 2002/03/01 02:44:24 maxk Exp $
16 */
17
18 /*
19 * Changes:
20 *
21 * Mike Kershaw <dragorn@kismetwireless.net> 2005/08/14
22 * Add TUNSETLINK ioctl to set the link encapsulation
23 *
24 * Mark Smith <markzzzsmith@yahoo.com.au>
25 * Use eth_random_addr() for tap MAC address.
26 *
27 * Harald Roelle <harald.roelle@ifi.lmu.de> 2004/04/20
28 * Fixes in packet dropping, queue length setting and queue wakeup.
29 * Increased default tx queue length.
30 * Added ethtool API.
31 * Minor cleanups
32 *
33 * Daniel Podlejski <underley@underley.eu.org>
34 * Modifications for 2.3.99-pre5 kernel.
35 */
36
37 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
38
39 #define DRV_NAME "tun"
40 #define DRV_VERSION "1.6"
41 #define DRV_DESCRIPTION "Universal TUN/TAP device driver"
42 #define DRV_COPYRIGHT "(C) 1999-2004 Max Krasnyansky <maxk@qualcomm.com>"
43
44 #include <linux/module.h>
45 #include <linux/errno.h>
46 #include <linux/kernel.h>
47 #include <linux/sched/signal.h>
48 #include <linux/major.h>
49 #include <linux/slab.h>
50 #include <linux/poll.h>
51 #include <linux/fcntl.h>
52 #include <linux/init.h>
53 #include <linux/skbuff.h>
54 #include <linux/netdevice.h>
55 #include <linux/etherdevice.h>
56 #include <linux/miscdevice.h>
57 #include <linux/ethtool.h>
58 #include <linux/rtnetlink.h>
59 #include <linux/compat.h>
60 #include <linux/if.h>
61 #include <linux/if_arp.h>
62 #include <linux/if_ether.h>
63 #include <linux/if_tun.h>
64 #include <linux/if_vlan.h>
65 #include <linux/crc32.h>
66 #include <linux/nsproxy.h>
67 #include <linux/virtio_net.h>
68 #include <linux/rcupdate.h>
69 #include <net/net_namespace.h>
70 #include <net/netns/generic.h>
71 #include <net/rtnetlink.h>
72 #include <net/sock.h>
73 #include <linux/seq_file.h>
74 #include <linux/uio.h>
75 #include <linux/skb_array.h>
76 #include <linux/bpf.h>
77 #include <linux/bpf_trace.h>
78 #include <linux/mutex.h>
79
80 #include <linux/uaccess.h>
81
82 /* Uncomment to enable debugging */
83 /* #define TUN_DEBUG 1 */
84
85 #ifdef TUN_DEBUG
86 static int debug;
87
88 #define tun_debug(level, tun, fmt, args...) \
89 do { \
90 if (tun->debug) \
91 netdev_printk(level, tun->dev, fmt, ##args); \
92 } while (0)
93 #define DBG1(level, fmt, args...) \
94 do { \
95 if (debug == 2) \
96 printk(level fmt, ##args); \
97 } while (0)
98 #else
99 #define tun_debug(level, tun, fmt, args...) \
100 do { \
101 if (0) \
102 netdev_printk(level, tun->dev, fmt, ##args); \
103 } while (0)
104 #define DBG1(level, fmt, args...) \
105 do { \
106 if (0) \
107 printk(level fmt, ##args); \
108 } while (0)
109 #endif
110
111 #define TUN_HEADROOM 256
112 #define TUN_RX_PAD (NET_IP_ALIGN + NET_SKB_PAD)
113
114 /* TUN device flags */
115
116 /* IFF_ATTACH_QUEUE is never stored in device flags,
117 * overload it to mean fasync when stored there.
118 */
119 #define TUN_FASYNC IFF_ATTACH_QUEUE
120 /* High bits in flags field are unused. */
121 #define TUN_VNET_LE 0x80000000
122 #define TUN_VNET_BE 0x40000000
123
124 #define TUN_FEATURES (IFF_NO_PI | IFF_ONE_QUEUE | IFF_VNET_HDR | \
125 IFF_MULTI_QUEUE | IFF_NAPI | IFF_NAPI_FRAGS)
126
127 #define GOODCOPY_LEN 128
128
129 #define FLT_EXACT_COUNT 8
130 struct tap_filter {
131 unsigned int count; /* Number of addrs. Zero means disabled */
132 u32 mask[2]; /* Mask of the hashed addrs */
133 unsigned char addr[FLT_EXACT_COUNT][ETH_ALEN];
134 };
135
136 /* MAX_TAP_QUEUES 256 is chosen to allow rx/tx queues to be equal
137 * to max number of VCPUs in guest. */
138 #define MAX_TAP_QUEUES 256
139 #define MAX_TAP_FLOWS 4096
140
141 #define TUN_FLOW_EXPIRE (3 * HZ)
142
143 struct tun_pcpu_stats {
144 u64 rx_packets;
145 u64 rx_bytes;
146 u64 tx_packets;
147 u64 tx_bytes;
148 struct u64_stats_sync syncp;
149 u32 rx_dropped;
150 u32 tx_dropped;
151 u32 rx_frame_errors;
152 };
153
154 /* A tun_file connects an open character device to a tuntap netdevice. It
155 * also contains all socket related structures (except sock_fprog and tap_filter)
156 * to serve as one transmit queue for tuntap device. The sock_fprog and
157 * tap_filter were kept in tun_struct since they were used for filtering for the
158 * netdevice not for a specific queue (at least I didn't see the requirement for
159 * this).
160 *
161 * RCU usage:
162 * The tun_file and tun_struct are loosely coupled, the pointer from one to the
163 * other can only be read while rcu_read_lock or rtnl_lock is held.
164 */
165 struct tun_file {
166 struct sock sk;
167 struct socket socket;
168 struct socket_wq wq;
169 struct tun_struct __rcu *tun;
170 struct fasync_struct *fasync;
171 /* only used for fasnyc */
172 unsigned int flags;
173 union {
174 u16 queue_index;
175 unsigned int ifindex;
176 };
177 struct napi_struct napi;
178 bool napi_enabled;
179 struct mutex napi_mutex; /* Protects access to the above napi */
180 struct list_head next;
181 struct tun_struct *detached;
182 struct ptr_ring tx_ring;
183 struct xdp_rxq_info xdp_rxq;
184 int xdp_pending_pkts;
185 };
186
187 struct tun_flow_entry {
188 struct hlist_node hash_link;
189 struct rcu_head rcu;
190 struct tun_struct *tun;
191
192 u32 rxhash;
193 u32 rps_rxhash;
194 int queue_index;
195 unsigned long updated;
196 };
197
198 #define TUN_NUM_FLOW_ENTRIES 1024
199
200 struct tun_prog {
201 struct rcu_head rcu;
202 struct bpf_prog *prog;
203 };
204
205 /* Since the socket were moved to tun_file, to preserve the behavior of persist
206 * device, socket filter, sndbuf and vnet header size were restore when the
207 * file were attached to a persist device.
208 */
209 struct tun_struct {
210 struct tun_file __rcu *tfiles[MAX_TAP_QUEUES];
211 unsigned int numqueues;
212 unsigned int flags;
213 kuid_t owner;
214 kgid_t group;
215
216 struct net_device *dev;
217 netdev_features_t set_features;
218 #define TUN_USER_FEATURES (NETIF_F_HW_CSUM|NETIF_F_TSO_ECN|NETIF_F_TSO| \
219 NETIF_F_TSO6)
220
221 int align;
222 int vnet_hdr_sz;
223 int sndbuf;
224 struct tap_filter txflt;
225 struct sock_fprog fprog;
226 /* protected by rtnl lock */
227 bool filter_attached;
228 #ifdef TUN_DEBUG
229 int debug;
230 #endif
231 spinlock_t lock;
232 struct hlist_head flows[TUN_NUM_FLOW_ENTRIES];
233 struct timer_list flow_gc_timer;
234 unsigned long ageing_time;
235 unsigned int numdisabled;
236 struct list_head disabled;
237 void *security;
238 u32 flow_count;
239 u32 rx_batched;
240 struct tun_pcpu_stats __percpu *pcpu_stats;
241 struct bpf_prog __rcu *xdp_prog;
242 struct tun_prog __rcu *steering_prog;
243 struct tun_prog __rcu *filter_prog;
244 };
245
246 struct veth {
247 __be16 h_vlan_proto;
248 __be16 h_vlan_TCI;
249 };
250
251 bool tun_is_xdp_buff(void *ptr)
252 {
253 return (unsigned long)ptr & TUN_XDP_FLAG;
254 }
255 EXPORT_SYMBOL(tun_is_xdp_buff);
256
257 void *tun_xdp_to_ptr(void *ptr)
258 {
259 return (void *)((unsigned long)ptr | TUN_XDP_FLAG);
260 }
261 EXPORT_SYMBOL(tun_xdp_to_ptr);
262
263 void *tun_ptr_to_xdp(void *ptr)
264 {
265 return (void *)((unsigned long)ptr & ~TUN_XDP_FLAG);
266 }
267 EXPORT_SYMBOL(tun_ptr_to_xdp);
268
269 static int tun_napi_receive(struct napi_struct *napi, int budget)
270 {
271 struct tun_file *tfile = container_of(napi, struct tun_file, napi);
272 struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
273 struct sk_buff_head process_queue;
274 struct sk_buff *skb;
275 int received = 0;
276
277 __skb_queue_head_init(&process_queue);
278
279 spin_lock(&queue->lock);
280 skb_queue_splice_tail_init(queue, &process_queue);
281 spin_unlock(&queue->lock);
282
283 while (received < budget && (skb = __skb_dequeue(&process_queue))) {
284 napi_gro_receive(napi, skb);
285 ++received;
286 }
287
288 if (!skb_queue_empty(&process_queue)) {
289 spin_lock(&queue->lock);
290 skb_queue_splice(&process_queue, queue);
291 spin_unlock(&queue->lock);
292 }
293
294 return received;
295 }
296
297 static int tun_napi_poll(struct napi_struct *napi, int budget)
298 {
299 unsigned int received;
300
301 received = tun_napi_receive(napi, budget);
302
303 if (received < budget)
304 napi_complete_done(napi, received);
305
306 return received;
307 }
308
309 static void tun_napi_init(struct tun_struct *tun, struct tun_file *tfile,
310 bool napi_en)
311 {
312 tfile->napi_enabled = napi_en;
313 if (napi_en) {
314 netif_napi_add(tun->dev, &tfile->napi, tun_napi_poll,
315 NAPI_POLL_WEIGHT);
316 napi_enable(&tfile->napi);
317 mutex_init(&tfile->napi_mutex);
318 }
319 }
320
321 static void tun_napi_disable(struct tun_struct *tun, struct tun_file *tfile)
322 {
323 if (tfile->napi_enabled)
324 napi_disable(&tfile->napi);
325 }
326
327 static void tun_napi_del(struct tun_struct *tun, struct tun_file *tfile)
328 {
329 if (tfile->napi_enabled)
330 netif_napi_del(&tfile->napi);
331 }
332
333 static bool tun_napi_frags_enabled(const struct tun_struct *tun)
334 {
335 return READ_ONCE(tun->flags) & IFF_NAPI_FRAGS;
336 }
337
338 #ifdef CONFIG_TUN_VNET_CROSS_LE
339 static inline bool tun_legacy_is_little_endian(struct tun_struct *tun)
340 {
341 return tun->flags & TUN_VNET_BE ? false :
342 virtio_legacy_is_little_endian();
343 }
344
345 static long tun_get_vnet_be(struct tun_struct *tun, int __user *argp)
346 {
347 int be = !!(tun->flags & TUN_VNET_BE);
348
349 if (put_user(be, argp))
350 return -EFAULT;
351
352 return 0;
353 }
354
355 static long tun_set_vnet_be(struct tun_struct *tun, int __user *argp)
356 {
357 int be;
358
359 if (get_user(be, argp))
360 return -EFAULT;
361
362 if (be)
363 tun->flags |= TUN_VNET_BE;
364 else
365 tun->flags &= ~TUN_VNET_BE;
366
367 return 0;
368 }
369 #else
370 static inline bool tun_legacy_is_little_endian(struct tun_struct *tun)
371 {
372 return virtio_legacy_is_little_endian();
373 }
374
375 static long tun_get_vnet_be(struct tun_struct *tun, int __user *argp)
376 {
377 return -EINVAL;
378 }
379
380 static long tun_set_vnet_be(struct tun_struct *tun, int __user *argp)
381 {
382 return -EINVAL;
383 }
384 #endif /* CONFIG_TUN_VNET_CROSS_LE */
385
386 static inline bool tun_is_little_endian(struct tun_struct *tun)
387 {
388 return tun->flags & TUN_VNET_LE ||
389 tun_legacy_is_little_endian(tun);
390 }
391
392 static inline u16 tun16_to_cpu(struct tun_struct *tun, __virtio16 val)
393 {
394 return __virtio16_to_cpu(tun_is_little_endian(tun), val);
395 }
396
397 static inline __virtio16 cpu_to_tun16(struct tun_struct *tun, u16 val)
398 {
399 return __cpu_to_virtio16(tun_is_little_endian(tun), val);
400 }
401
402 static inline u32 tun_hashfn(u32 rxhash)
403 {
404 return rxhash & 0x3ff;
405 }
406
407 static struct tun_flow_entry *tun_flow_find(struct hlist_head *head, u32 rxhash)
408 {
409 struct tun_flow_entry *e;
410
411 hlist_for_each_entry_rcu(e, head, hash_link) {
412 if (e->rxhash == rxhash)
413 return e;
414 }
415 return NULL;
416 }
417
418 static struct tun_flow_entry *tun_flow_create(struct tun_struct *tun,
419 struct hlist_head *head,
420 u32 rxhash, u16 queue_index)
421 {
422 struct tun_flow_entry *e = kmalloc(sizeof(*e), GFP_ATOMIC);
423
424 if (e) {
425 tun_debug(KERN_INFO, tun, "create flow: hash %u index %u\n",
426 rxhash, queue_index);
427 e->updated = jiffies;
428 e->rxhash = rxhash;
429 e->rps_rxhash = 0;
430 e->queue_index = queue_index;
431 e->tun = tun;
432 hlist_add_head_rcu(&e->hash_link, head);
433 ++tun->flow_count;
434 }
435 return e;
436 }
437
438 static void tun_flow_delete(struct tun_struct *tun, struct tun_flow_entry *e)
439 {
440 tun_debug(KERN_INFO, tun, "delete flow: hash %u index %u\n",
441 e->rxhash, e->queue_index);
442 hlist_del_rcu(&e->hash_link);
443 kfree_rcu(e, rcu);
444 --tun->flow_count;
445 }
446
447 static void tun_flow_flush(struct tun_struct *tun)
448 {
449 int i;
450
451 spin_lock_bh(&tun->lock);
452 for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
453 struct tun_flow_entry *e;
454 struct hlist_node *n;
455
456 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link)
457 tun_flow_delete(tun, e);
458 }
459 spin_unlock_bh(&tun->lock);
460 }
461
462 static void tun_flow_delete_by_queue(struct tun_struct *tun, u16 queue_index)
463 {
464 int i;
465
466 spin_lock_bh(&tun->lock);
467 for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
468 struct tun_flow_entry *e;
469 struct hlist_node *n;
470
471 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
472 if (e->queue_index == queue_index)
473 tun_flow_delete(tun, e);
474 }
475 }
476 spin_unlock_bh(&tun->lock);
477 }
478
479 static void tun_flow_cleanup(struct timer_list *t)
480 {
481 struct tun_struct *tun = from_timer(tun, t, flow_gc_timer);
482 unsigned long delay = tun->ageing_time;
483 unsigned long next_timer = jiffies + delay;
484 unsigned long count = 0;
485 int i;
486
487 tun_debug(KERN_INFO, tun, "tun_flow_cleanup\n");
488
489 spin_lock(&tun->lock);
490 for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
491 struct tun_flow_entry *e;
492 struct hlist_node *n;
493
494 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
495 unsigned long this_timer;
496
497 this_timer = e->updated + delay;
498 if (time_before_eq(this_timer, jiffies)) {
499 tun_flow_delete(tun, e);
500 continue;
501 }
502 count++;
503 if (time_before(this_timer, next_timer))
504 next_timer = this_timer;
505 }
506 }
507
508 if (count)
509 mod_timer(&tun->flow_gc_timer, round_jiffies_up(next_timer));
510 spin_unlock(&tun->lock);
511 }
512
513 static void tun_flow_update(struct tun_struct *tun, u32 rxhash,
514 struct tun_file *tfile)
515 {
516 struct hlist_head *head;
517 struct tun_flow_entry *e;
518 unsigned long delay = tun->ageing_time;
519 u16 queue_index = tfile->queue_index;
520
521 if (!rxhash)
522 return;
523 else
524 head = &tun->flows[tun_hashfn(rxhash)];
525
526 rcu_read_lock();
527
528 /* We may get a very small possibility of OOO during switching, not
529 * worth to optimize.*/
530 if (tun->numqueues == 1 || tfile->detached)
531 goto unlock;
532
533 e = tun_flow_find(head, rxhash);
534 if (likely(e)) {
535 /* TODO: keep queueing to old queue until it's empty? */
536 e->queue_index = queue_index;
537 e->updated = jiffies;
538 sock_rps_record_flow_hash(e->rps_rxhash);
539 } else {
540 spin_lock_bh(&tun->lock);
541 if (!tun_flow_find(head, rxhash) &&
542 tun->flow_count < MAX_TAP_FLOWS)
543 tun_flow_create(tun, head, rxhash, queue_index);
544
545 if (!timer_pending(&tun->flow_gc_timer))
546 mod_timer(&tun->flow_gc_timer,
547 round_jiffies_up(jiffies + delay));
548 spin_unlock_bh(&tun->lock);
549 }
550
551 unlock:
552 rcu_read_unlock();
553 }
554
555 /**
556 * Save the hash received in the stack receive path and update the
557 * flow_hash table accordingly.
558 */
559 static inline void tun_flow_save_rps_rxhash(struct tun_flow_entry *e, u32 hash)
560 {
561 if (unlikely(e->rps_rxhash != hash))
562 e->rps_rxhash = hash;
563 }
564
565 /* We try to identify a flow through its rxhash first. The reason that
566 * we do not check rxq no. is because some cards(e.g 82599), chooses
567 * the rxq based on the txq where the last packet of the flow comes. As
568 * the userspace application move between processors, we may get a
569 * different rxq no. here. If we could not get rxhash, then we would
570 * hope the rxq no. may help here.
571 */
572 static u16 tun_automq_select_queue(struct tun_struct *tun, struct sk_buff *skb)
573 {
574 struct tun_flow_entry *e;
575 u32 txq = 0;
576 u32 numqueues = 0;
577
578 numqueues = READ_ONCE(tun->numqueues);
579
580 txq = __skb_get_hash_symmetric(skb);
581 if (txq) {
582 e = tun_flow_find(&tun->flows[tun_hashfn(txq)], txq);
583 if (e) {
584 tun_flow_save_rps_rxhash(e, txq);
585 txq = e->queue_index;
586 } else
587 /* use multiply and shift instead of expensive divide */
588 txq = ((u64)txq * numqueues) >> 32;
589 } else if (likely(skb_rx_queue_recorded(skb))) {
590 txq = skb_get_rx_queue(skb);
591 while (unlikely(txq >= numqueues))
592 txq -= numqueues;
593 }
594
595 return txq;
596 }
597
598 static u16 tun_ebpf_select_queue(struct tun_struct *tun, struct sk_buff *skb)
599 {
600 struct tun_prog *prog;
601 u16 ret = 0;
602
603 prog = rcu_dereference(tun->steering_prog);
604 if (prog)
605 ret = bpf_prog_run_clear_cb(prog->prog, skb);
606
607 return ret % tun->numqueues;
608 }
609
610 static u16 tun_select_queue(struct net_device *dev, struct sk_buff *skb,
611 void *accel_priv, select_queue_fallback_t fallback)
612 {
613 struct tun_struct *tun = netdev_priv(dev);
614 u16 ret;
615
616 rcu_read_lock();
617 if (rcu_dereference(tun->steering_prog))
618 ret = tun_ebpf_select_queue(tun, skb);
619 else
620 ret = tun_automq_select_queue(tun, skb);
621 rcu_read_unlock();
622
623 return ret;
624 }
625
626 static inline bool tun_not_capable(struct tun_struct *tun)
627 {
628 const struct cred *cred = current_cred();
629 struct net *net = dev_net(tun->dev);
630
631 return ((uid_valid(tun->owner) && !uid_eq(cred->euid, tun->owner)) ||
632 (gid_valid(tun->group) && !in_egroup_p(tun->group))) &&
633 !ns_capable(net->user_ns, CAP_NET_ADMIN);
634 }
635
636 static void tun_set_real_num_queues(struct tun_struct *tun)
637 {
638 netif_set_real_num_tx_queues(tun->dev, tun->numqueues);
639 netif_set_real_num_rx_queues(tun->dev, tun->numqueues);
640 }
641
642 static void tun_disable_queue(struct tun_struct *tun, struct tun_file *tfile)
643 {
644 tfile->detached = tun;
645 list_add_tail(&tfile->next, &tun->disabled);
646 ++tun->numdisabled;
647 }
648
649 static struct tun_struct *tun_enable_queue(struct tun_file *tfile)
650 {
651 struct tun_struct *tun = tfile->detached;
652
653 tfile->detached = NULL;
654 list_del_init(&tfile->next);
655 --tun->numdisabled;
656 return tun;
657 }
658
659 static void tun_ptr_free(void *ptr)
660 {
661 if (!ptr)
662 return;
663 if (tun_is_xdp_buff(ptr)) {
664 struct xdp_buff *xdp = tun_ptr_to_xdp(ptr);
665
666 put_page(virt_to_head_page(xdp->data));
667 } else {
668 __skb_array_destroy_skb(ptr);
669 }
670 }
671
672 static void tun_queue_purge(struct tun_file *tfile)
673 {
674 void *ptr;
675
676 while ((ptr = ptr_ring_consume(&tfile->tx_ring)) != NULL)
677 tun_ptr_free(ptr);
678
679 skb_queue_purge(&tfile->sk.sk_write_queue);
680 skb_queue_purge(&tfile->sk.sk_error_queue);
681 }
682
683 static void tun_cleanup_tx_ring(struct tun_file *tfile)
684 {
685 if (tfile->tx_ring.queue) {
686 ptr_ring_cleanup(&tfile->tx_ring, tun_ptr_free);
687 xdp_rxq_info_unreg(&tfile->xdp_rxq);
688 memset(&tfile->tx_ring, 0, sizeof(tfile->tx_ring));
689 }
690 }
691
692 static void __tun_detach(struct tun_file *tfile, bool clean)
693 {
694 struct tun_file *ntfile;
695 struct tun_struct *tun;
696
697 tun = rtnl_dereference(tfile->tun);
698
699 if (tun && clean) {
700 tun_napi_disable(tun, tfile);
701 tun_napi_del(tun, tfile);
702 }
703
704 if (tun && !tfile->detached) {
705 u16 index = tfile->queue_index;
706 BUG_ON(index >= tun->numqueues);
707
708 rcu_assign_pointer(tun->tfiles[index],
709 tun->tfiles[tun->numqueues - 1]);
710 ntfile = rtnl_dereference(tun->tfiles[index]);
711 ntfile->queue_index = index;
712
713 --tun->numqueues;
714 if (clean) {
715 RCU_INIT_POINTER(tfile->tun, NULL);
716 sock_put(&tfile->sk);
717 } else
718 tun_disable_queue(tun, tfile);
719
720 synchronize_net();
721 tun_flow_delete_by_queue(tun, tun->numqueues + 1);
722 /* Drop read queue */
723 tun_queue_purge(tfile);
724 tun_set_real_num_queues(tun);
725 } else if (tfile->detached && clean) {
726 tun = tun_enable_queue(tfile);
727 sock_put(&tfile->sk);
728 }
729
730 if (clean) {
731 if (tun && tun->numqueues == 0 && tun->numdisabled == 0) {
732 netif_carrier_off(tun->dev);
733
734 if (!(tun->flags & IFF_PERSIST) &&
735 tun->dev->reg_state == NETREG_REGISTERED)
736 unregister_netdevice(tun->dev);
737 }
738 tun_cleanup_tx_ring(tfile);
739 sock_put(&tfile->sk);
740 }
741 }
742
743 static void tun_detach(struct tun_file *tfile, bool clean)
744 {
745 rtnl_lock();
746 __tun_detach(tfile, clean);
747 rtnl_unlock();
748 }
749
750 static void tun_detach_all(struct net_device *dev)
751 {
752 struct tun_struct *tun = netdev_priv(dev);
753 struct tun_file *tfile, *tmp;
754 int i, n = tun->numqueues;
755
756 for (i = 0; i < n; i++) {
757 tfile = rtnl_dereference(tun->tfiles[i]);
758 BUG_ON(!tfile);
759 tun_napi_disable(tun, tfile);
760 tfile->socket.sk->sk_shutdown = RCV_SHUTDOWN;
761 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
762 RCU_INIT_POINTER(tfile->tun, NULL);
763 --tun->numqueues;
764 }
765 list_for_each_entry(tfile, &tun->disabled, next) {
766 tfile->socket.sk->sk_shutdown = RCV_SHUTDOWN;
767 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
768 RCU_INIT_POINTER(tfile->tun, NULL);
769 }
770 BUG_ON(tun->numqueues != 0);
771
772 synchronize_net();
773 for (i = 0; i < n; i++) {
774 tfile = rtnl_dereference(tun->tfiles[i]);
775 tun_napi_del(tun, tfile);
776 /* Drop read queue */
777 tun_queue_purge(tfile);
778 sock_put(&tfile->sk);
779 tun_cleanup_tx_ring(tfile);
780 }
781 list_for_each_entry_safe(tfile, tmp, &tun->disabled, next) {
782 tun_enable_queue(tfile);
783 tun_queue_purge(tfile);
784 sock_put(&tfile->sk);
785 tun_cleanup_tx_ring(tfile);
786 }
787 BUG_ON(tun->numdisabled != 0);
788
789 if (tun->flags & IFF_PERSIST)
790 module_put(THIS_MODULE);
791 }
792
793 static int tun_attach(struct tun_struct *tun, struct file *file,
794 bool skip_filter, bool napi)
795 {
796 struct tun_file *tfile = file->private_data;
797 struct net_device *dev = tun->dev;
798 int err;
799
800 err = security_tun_dev_attach(tfile->socket.sk, tun->security);
801 if (err < 0)
802 goto out;
803
804 err = -EINVAL;
805 if (rtnl_dereference(tfile->tun) && !tfile->detached)
806 goto out;
807
808 err = -EBUSY;
809 if (!(tun->flags & IFF_MULTI_QUEUE) && tun->numqueues == 1)
810 goto out;
811
812 err = -E2BIG;
813 if (!tfile->detached &&
814 tun->numqueues + tun->numdisabled == MAX_TAP_QUEUES)
815 goto out;
816
817 err = 0;
818
819 /* Re-attach the filter to persist device */
820 if (!skip_filter && (tun->filter_attached == true)) {
821 lock_sock(tfile->socket.sk);
822 err = sk_attach_filter(&tun->fprog, tfile->socket.sk);
823 release_sock(tfile->socket.sk);
824 if (!err)
825 goto out;
826 }
827
828 if (!tfile->detached &&
829 ptr_ring_init(&tfile->tx_ring, dev->tx_queue_len, GFP_KERNEL)) {
830 err = -ENOMEM;
831 goto out;
832 }
833
834 tfile->queue_index = tun->numqueues;
835 tfile->socket.sk->sk_shutdown &= ~RCV_SHUTDOWN;
836
837 if (tfile->detached) {
838 /* Re-attach detached tfile, updating XDP queue_index */
839 WARN_ON(!xdp_rxq_info_is_reg(&tfile->xdp_rxq));
840
841 if (tfile->xdp_rxq.queue_index != tfile->queue_index)
842 tfile->xdp_rxq.queue_index = tfile->queue_index;
843 } else {
844 /* Setup XDP RX-queue info, for new tfile getting attached */
845 err = xdp_rxq_info_reg(&tfile->xdp_rxq,
846 tun->dev, tfile->queue_index);
847 if (err < 0)
848 goto out;
849 err = 0;
850 }
851
852 rcu_assign_pointer(tfile->tun, tun);
853 rcu_assign_pointer(tun->tfiles[tun->numqueues], tfile);
854 tun->numqueues++;
855
856 if (tfile->detached) {
857 tun_enable_queue(tfile);
858 } else {
859 sock_hold(&tfile->sk);
860 tun_napi_init(tun, tfile, napi);
861 }
862
863 tun_set_real_num_queues(tun);
864
865 /* device is allowed to go away first, so no need to hold extra
866 * refcnt.
867 */
868
869 out:
870 return err;
871 }
872
873 static struct tun_struct *tun_get(struct tun_file *tfile)
874 {
875 struct tun_struct *tun;
876
877 rcu_read_lock();
878 tun = rcu_dereference(tfile->tun);
879 if (tun)
880 dev_hold(tun->dev);
881 rcu_read_unlock();
882
883 return tun;
884 }
885
886 static void tun_put(struct tun_struct *tun)
887 {
888 dev_put(tun->dev);
889 }
890
891 /* TAP filtering */
892 static void addr_hash_set(u32 *mask, const u8 *addr)
893 {
894 int n = ether_crc(ETH_ALEN, addr) >> 26;
895 mask[n >> 5] |= (1 << (n & 31));
896 }
897
898 static unsigned int addr_hash_test(const u32 *mask, const u8 *addr)
899 {
900 int n = ether_crc(ETH_ALEN, addr) >> 26;
901 return mask[n >> 5] & (1 << (n & 31));
902 }
903
904 static int update_filter(struct tap_filter *filter, void __user *arg)
905 {
906 struct { u8 u[ETH_ALEN]; } *addr;
907 struct tun_filter uf;
908 int err, alen, n, nexact;
909
910 if (copy_from_user(&uf, arg, sizeof(uf)))
911 return -EFAULT;
912
913 if (!uf.count) {
914 /* Disabled */
915 filter->count = 0;
916 return 0;
917 }
918
919 alen = ETH_ALEN * uf.count;
920 addr = memdup_user(arg + sizeof(uf), alen);
921 if (IS_ERR(addr))
922 return PTR_ERR(addr);
923
924 /* The filter is updated without holding any locks. Which is
925 * perfectly safe. We disable it first and in the worst
926 * case we'll accept a few undesired packets. */
927 filter->count = 0;
928 wmb();
929
930 /* Use first set of addresses as an exact filter */
931 for (n = 0; n < uf.count && n < FLT_EXACT_COUNT; n++)
932 memcpy(filter->addr[n], addr[n].u, ETH_ALEN);
933
934 nexact = n;
935
936 /* Remaining multicast addresses are hashed,
937 * unicast will leave the filter disabled. */
938 memset(filter->mask, 0, sizeof(filter->mask));
939 for (; n < uf.count; n++) {
940 if (!is_multicast_ether_addr(addr[n].u)) {
941 err = 0; /* no filter */
942 goto free_addr;
943 }
944 addr_hash_set(filter->mask, addr[n].u);
945 }
946
947 /* For ALLMULTI just set the mask to all ones.
948 * This overrides the mask populated above. */
949 if ((uf.flags & TUN_FLT_ALLMULTI))
950 memset(filter->mask, ~0, sizeof(filter->mask));
951
952 /* Now enable the filter */
953 wmb();
954 filter->count = nexact;
955
956 /* Return the number of exact filters */
957 err = nexact;
958 free_addr:
959 kfree(addr);
960 return err;
961 }
962
963 /* Returns: 0 - drop, !=0 - accept */
964 static int run_filter(struct tap_filter *filter, const struct sk_buff *skb)
965 {
966 /* Cannot use eth_hdr(skb) here because skb_mac_hdr() is incorrect
967 * at this point. */
968 struct ethhdr *eh = (struct ethhdr *) skb->data;
969 int i;
970
971 /* Exact match */
972 for (i = 0; i < filter->count; i++)
973 if (ether_addr_equal(eh->h_dest, filter->addr[i]))
974 return 1;
975
976 /* Inexact match (multicast only) */
977 if (is_multicast_ether_addr(eh->h_dest))
978 return addr_hash_test(filter->mask, eh->h_dest);
979
980 return 0;
981 }
982
983 /*
984 * Checks whether the packet is accepted or not.
985 * Returns: 0 - drop, !=0 - accept
986 */
987 static int check_filter(struct tap_filter *filter, const struct sk_buff *skb)
988 {
989 if (!filter->count)
990 return 1;
991
992 return run_filter(filter, skb);
993 }
994
995 /* Network device part of the driver */
996
997 static const struct ethtool_ops tun_ethtool_ops;
998
999 /* Net device detach from fd. */
1000 static void tun_net_uninit(struct net_device *dev)
1001 {
1002 tun_detach_all(dev);
1003 }
1004
1005 /* Net device open. */
1006 static int tun_net_open(struct net_device *dev)
1007 {
1008 struct tun_struct *tun = netdev_priv(dev);
1009 int i;
1010
1011 netif_tx_start_all_queues(dev);
1012
1013 for (i = 0; i < tun->numqueues; i++) {
1014 struct tun_file *tfile;
1015
1016 tfile = rtnl_dereference(tun->tfiles[i]);
1017 tfile->socket.sk->sk_write_space(tfile->socket.sk);
1018 }
1019
1020 return 0;
1021 }
1022
1023 /* Net device close. */
1024 static int tun_net_close(struct net_device *dev)
1025 {
1026 netif_tx_stop_all_queues(dev);
1027 return 0;
1028 }
1029
1030 /* Net device start xmit */
1031 static void tun_automq_xmit(struct tun_struct *tun, struct sk_buff *skb)
1032 {
1033 #ifdef CONFIG_RPS
1034 if (tun->numqueues == 1 && static_key_false(&rps_needed)) {
1035 /* Select queue was not called for the skbuff, so we extract the
1036 * RPS hash and save it into the flow_table here.
1037 */
1038 __u32 rxhash;
1039
1040 rxhash = __skb_get_hash_symmetric(skb);
1041 if (rxhash) {
1042 struct tun_flow_entry *e;
1043 e = tun_flow_find(&tun->flows[tun_hashfn(rxhash)],
1044 rxhash);
1045 if (e)
1046 tun_flow_save_rps_rxhash(e, rxhash);
1047 }
1048 }
1049 #endif
1050 }
1051
1052 static unsigned int run_ebpf_filter(struct tun_struct *tun,
1053 struct sk_buff *skb,
1054 int len)
1055 {
1056 struct tun_prog *prog = rcu_dereference(tun->filter_prog);
1057
1058 if (prog)
1059 len = bpf_prog_run_clear_cb(prog->prog, skb);
1060
1061 return len;
1062 }
1063
1064 /* Net device start xmit */
1065 static netdev_tx_t tun_net_xmit(struct sk_buff *skb, struct net_device *dev)
1066 {
1067 struct tun_struct *tun = netdev_priv(dev);
1068 int txq = skb->queue_mapping;
1069 struct tun_file *tfile;
1070 int len = skb->len;
1071
1072 rcu_read_lock();
1073 tfile = rcu_dereference(tun->tfiles[txq]);
1074
1075 /* Drop packet if interface is not attached */
1076 if (txq >= tun->numqueues)
1077 goto drop;
1078
1079 if (!rcu_dereference(tun->steering_prog))
1080 tun_automq_xmit(tun, skb);
1081
1082 tun_debug(KERN_INFO, tun, "tun_net_xmit %d\n", skb->len);
1083
1084 BUG_ON(!tfile);
1085
1086 /* Drop if the filter does not like it.
1087 * This is a noop if the filter is disabled.
1088 * Filter can be enabled only for the TAP devices. */
1089 if (!check_filter(&tun->txflt, skb))
1090 goto drop;
1091
1092 if (tfile->socket.sk->sk_filter &&
1093 sk_filter(tfile->socket.sk, skb))
1094 goto drop;
1095
1096 len = run_ebpf_filter(tun, skb, len);
1097
1098 /* Trim extra bytes since we may insert vlan proto & TCI
1099 * in tun_put_user().
1100 */
1101 len -= skb_vlan_tag_present(skb) ? sizeof(struct veth) : 0;
1102 if (len <= 0 || pskb_trim(skb, len))
1103 goto drop;
1104
1105 if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC)))
1106 goto drop;
1107
1108 skb_tx_timestamp(skb);
1109
1110 /* Orphan the skb - required as we might hang on to it
1111 * for indefinite time.
1112 */
1113 skb_orphan(skb);
1114
1115 nf_reset(skb);
1116
1117 if (ptr_ring_produce(&tfile->tx_ring, skb))
1118 goto drop;
1119
1120 /* Notify and wake up reader process */
1121 if (tfile->flags & TUN_FASYNC)
1122 kill_fasync(&tfile->fasync, SIGIO, POLL_IN);
1123 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
1124
1125 rcu_read_unlock();
1126 return NETDEV_TX_OK;
1127
1128 drop:
1129 this_cpu_inc(tun->pcpu_stats->tx_dropped);
1130 skb_tx_error(skb);
1131 kfree_skb(skb);
1132 rcu_read_unlock();
1133 return NET_XMIT_DROP;
1134 }
1135
1136 static void tun_net_mclist(struct net_device *dev)
1137 {
1138 /*
1139 * This callback is supposed to deal with mc filter in
1140 * _rx_ path and has nothing to do with the _tx_ path.
1141 * In rx path we always accept everything userspace gives us.
1142 */
1143 }
1144
1145 static netdev_features_t tun_net_fix_features(struct net_device *dev,
1146 netdev_features_t features)
1147 {
1148 struct tun_struct *tun = netdev_priv(dev);
1149
1150 return (features & tun->set_features) | (features & ~TUN_USER_FEATURES);
1151 }
1152 #ifdef CONFIG_NET_POLL_CONTROLLER
1153 static void tun_poll_controller(struct net_device *dev)
1154 {
1155 /*
1156 * Tun only receives frames when:
1157 * 1) the char device endpoint gets data from user space
1158 * 2) the tun socket gets a sendmsg call from user space
1159 * If NAPI is not enabled, since both of those are synchronous
1160 * operations, we are guaranteed never to have pending data when we poll
1161 * for it so there is nothing to do here but return.
1162 * We need this though so netpoll recognizes us as an interface that
1163 * supports polling, which enables bridge devices in virt setups to
1164 * still use netconsole
1165 * If NAPI is enabled, however, we need to schedule polling for all
1166 * queues unless we are using napi_gro_frags(), which we call in
1167 * process context and not in NAPI context.
1168 */
1169 struct tun_struct *tun = netdev_priv(dev);
1170
1171 if (tun->flags & IFF_NAPI) {
1172 struct tun_file *tfile;
1173 int i;
1174
1175 if (tun_napi_frags_enabled(tun))
1176 return;
1177
1178 rcu_read_lock();
1179 for (i = 0; i < tun->numqueues; i++) {
1180 tfile = rcu_dereference(tun->tfiles[i]);
1181 if (tfile->napi_enabled)
1182 napi_schedule(&tfile->napi);
1183 }
1184 rcu_read_unlock();
1185 }
1186 return;
1187 }
1188 #endif
1189
1190 static void tun_set_headroom(struct net_device *dev, int new_hr)
1191 {
1192 struct tun_struct *tun = netdev_priv(dev);
1193
1194 if (new_hr < NET_SKB_PAD)
1195 new_hr = NET_SKB_PAD;
1196
1197 tun->align = new_hr;
1198 }
1199
1200 static void
1201 tun_net_get_stats64(struct net_device *dev, struct rtnl_link_stats64 *stats)
1202 {
1203 u32 rx_dropped = 0, tx_dropped = 0, rx_frame_errors = 0;
1204 struct tun_struct *tun = netdev_priv(dev);
1205 struct tun_pcpu_stats *p;
1206 int i;
1207
1208 for_each_possible_cpu(i) {
1209 u64 rxpackets, rxbytes, txpackets, txbytes;
1210 unsigned int start;
1211
1212 p = per_cpu_ptr(tun->pcpu_stats, i);
1213 do {
1214 start = u64_stats_fetch_begin(&p->syncp);
1215 rxpackets = p->rx_packets;
1216 rxbytes = p->rx_bytes;
1217 txpackets = p->tx_packets;
1218 txbytes = p->tx_bytes;
1219 } while (u64_stats_fetch_retry(&p->syncp, start));
1220
1221 stats->rx_packets += rxpackets;
1222 stats->rx_bytes += rxbytes;
1223 stats->tx_packets += txpackets;
1224 stats->tx_bytes += txbytes;
1225
1226 /* u32 counters */
1227 rx_dropped += p->rx_dropped;
1228 rx_frame_errors += p->rx_frame_errors;
1229 tx_dropped += p->tx_dropped;
1230 }
1231 stats->rx_dropped = rx_dropped;
1232 stats->rx_frame_errors = rx_frame_errors;
1233 stats->tx_dropped = tx_dropped;
1234 }
1235
1236 static int tun_xdp_set(struct net_device *dev, struct bpf_prog *prog,
1237 struct netlink_ext_ack *extack)
1238 {
1239 struct tun_struct *tun = netdev_priv(dev);
1240 struct bpf_prog *old_prog;
1241
1242 old_prog = rtnl_dereference(tun->xdp_prog);
1243 rcu_assign_pointer(tun->xdp_prog, prog);
1244 if (old_prog)
1245 bpf_prog_put(old_prog);
1246
1247 return 0;
1248 }
1249
1250 static u32 tun_xdp_query(struct net_device *dev)
1251 {
1252 struct tun_struct *tun = netdev_priv(dev);
1253 const struct bpf_prog *xdp_prog;
1254
1255 xdp_prog = rtnl_dereference(tun->xdp_prog);
1256 if (xdp_prog)
1257 return xdp_prog->aux->id;
1258
1259 return 0;
1260 }
1261
1262 static int tun_xdp(struct net_device *dev, struct netdev_bpf *xdp)
1263 {
1264 switch (xdp->command) {
1265 case XDP_SETUP_PROG:
1266 return tun_xdp_set(dev, xdp->prog, xdp->extack);
1267 case XDP_QUERY_PROG:
1268 xdp->prog_id = tun_xdp_query(dev);
1269 xdp->prog_attached = !!xdp->prog_id;
1270 return 0;
1271 default:
1272 return -EINVAL;
1273 }
1274 }
1275
1276 static const struct net_device_ops tun_netdev_ops = {
1277 .ndo_uninit = tun_net_uninit,
1278 .ndo_open = tun_net_open,
1279 .ndo_stop = tun_net_close,
1280 .ndo_start_xmit = tun_net_xmit,
1281 .ndo_fix_features = tun_net_fix_features,
1282 .ndo_select_queue = tun_select_queue,
1283 #ifdef CONFIG_NET_POLL_CONTROLLER
1284 .ndo_poll_controller = tun_poll_controller,
1285 #endif
1286 .ndo_set_rx_headroom = tun_set_headroom,
1287 .ndo_get_stats64 = tun_net_get_stats64,
1288 };
1289
1290 static int tun_xdp_xmit(struct net_device *dev, struct xdp_buff *xdp)
1291 {
1292 struct tun_struct *tun = netdev_priv(dev);
1293 struct xdp_buff *buff = xdp->data_hard_start;
1294 int headroom = xdp->data - xdp->data_hard_start;
1295 struct tun_file *tfile;
1296 u32 numqueues;
1297 int ret = 0;
1298
1299 /* Assure headroom is available and buff is properly aligned */
1300 if (unlikely(headroom < sizeof(*xdp) || tun_is_xdp_buff(xdp)))
1301 return -ENOSPC;
1302
1303 *buff = *xdp;
1304
1305 rcu_read_lock();
1306
1307 numqueues = READ_ONCE(tun->numqueues);
1308 if (!numqueues) {
1309 ret = -ENOSPC;
1310 goto out;
1311 }
1312
1313 tfile = rcu_dereference(tun->tfiles[smp_processor_id() %
1314 numqueues]);
1315 /* Encode the XDP flag into lowest bit for consumer to differ
1316 * XDP buffer from sk_buff.
1317 */
1318 if (ptr_ring_produce(&tfile->tx_ring, tun_xdp_to_ptr(buff))) {
1319 this_cpu_inc(tun->pcpu_stats->tx_dropped);
1320 ret = -ENOSPC;
1321 }
1322
1323 out:
1324 rcu_read_unlock();
1325 return ret;
1326 }
1327
1328 static void tun_xdp_flush(struct net_device *dev)
1329 {
1330 struct tun_struct *tun = netdev_priv(dev);
1331 struct tun_file *tfile;
1332 u32 numqueues;
1333
1334 rcu_read_lock();
1335
1336 numqueues = READ_ONCE(tun->numqueues);
1337 if (!numqueues)
1338 goto out;
1339
1340 tfile = rcu_dereference(tun->tfiles[smp_processor_id() %
1341 numqueues]);
1342 /* Notify and wake up reader process */
1343 if (tfile->flags & TUN_FASYNC)
1344 kill_fasync(&tfile->fasync, SIGIO, POLL_IN);
1345 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
1346
1347 out:
1348 rcu_read_unlock();
1349 }
1350
1351 static const struct net_device_ops tap_netdev_ops = {
1352 .ndo_uninit = tun_net_uninit,
1353 .ndo_open = tun_net_open,
1354 .ndo_stop = tun_net_close,
1355 .ndo_start_xmit = tun_net_xmit,
1356 .ndo_fix_features = tun_net_fix_features,
1357 .ndo_set_rx_mode = tun_net_mclist,
1358 .ndo_set_mac_address = eth_mac_addr,
1359 .ndo_validate_addr = eth_validate_addr,
1360 .ndo_select_queue = tun_select_queue,
1361 #ifdef CONFIG_NET_POLL_CONTROLLER
1362 .ndo_poll_controller = tun_poll_controller,
1363 #endif
1364 .ndo_features_check = passthru_features_check,
1365 .ndo_set_rx_headroom = tun_set_headroom,
1366 .ndo_get_stats64 = tun_net_get_stats64,
1367 .ndo_bpf = tun_xdp,
1368 .ndo_xdp_xmit = tun_xdp_xmit,
1369 .ndo_xdp_flush = tun_xdp_flush,
1370 };
1371
1372 static void tun_flow_init(struct tun_struct *tun)
1373 {
1374 int i;
1375
1376 for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++)
1377 INIT_HLIST_HEAD(&tun->flows[i]);
1378
1379 tun->ageing_time = TUN_FLOW_EXPIRE;
1380 timer_setup(&tun->flow_gc_timer, tun_flow_cleanup, 0);
1381 mod_timer(&tun->flow_gc_timer,
1382 round_jiffies_up(jiffies + tun->ageing_time));
1383 }
1384
1385 static void tun_flow_uninit(struct tun_struct *tun)
1386 {
1387 del_timer_sync(&tun->flow_gc_timer);
1388 tun_flow_flush(tun);
1389 }
1390
1391 #define MIN_MTU 68
1392 #define MAX_MTU 65535
1393
1394 /* Initialize net device. */
1395 static void tun_net_init(struct net_device *dev)
1396 {
1397 struct tun_struct *tun = netdev_priv(dev);
1398
1399 switch (tun->flags & TUN_TYPE_MASK) {
1400 case IFF_TUN:
1401 dev->netdev_ops = &tun_netdev_ops;
1402
1403 /* Point-to-Point TUN Device */
1404 dev->hard_header_len = 0;
1405 dev->addr_len = 0;
1406 dev->mtu = 1500;
1407
1408 /* Zero header length */
1409 dev->type = ARPHRD_NONE;
1410 dev->flags = IFF_POINTOPOINT | IFF_NOARP | IFF_MULTICAST;
1411 break;
1412
1413 case IFF_TAP:
1414 dev->netdev_ops = &tap_netdev_ops;
1415 /* Ethernet TAP Device */
1416 ether_setup(dev);
1417 dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1418 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
1419
1420 eth_hw_addr_random(dev);
1421
1422 break;
1423 }
1424
1425 dev->min_mtu = MIN_MTU;
1426 dev->max_mtu = MAX_MTU - dev->hard_header_len;
1427 }
1428
1429 /* Character device part */
1430
1431 /* Poll */
1432 static __poll_t tun_chr_poll(struct file *file, poll_table *wait)
1433 {
1434 struct tun_file *tfile = file->private_data;
1435 struct tun_struct *tun = tun_get(tfile);
1436 struct sock *sk;
1437 __poll_t mask = 0;
1438
1439 if (!tun)
1440 return EPOLLERR;
1441
1442 sk = tfile->socket.sk;
1443
1444 tun_debug(KERN_INFO, tun, "tun_chr_poll\n");
1445
1446 poll_wait(file, sk_sleep(sk), wait);
1447
1448 if (!ptr_ring_empty(&tfile->tx_ring))
1449 mask |= EPOLLIN | EPOLLRDNORM;
1450
1451 if (tun->dev->flags & IFF_UP &&
1452 (sock_writeable(sk) ||
1453 (!test_and_set_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags) &&
1454 sock_writeable(sk))))
1455 mask |= EPOLLOUT | EPOLLWRNORM;
1456
1457 if (tun->dev->reg_state != NETREG_REGISTERED)
1458 mask = EPOLLERR;
1459
1460 tun_put(tun);
1461 return mask;
1462 }
1463
1464 static struct sk_buff *tun_napi_alloc_frags(struct tun_file *tfile,
1465 size_t len,
1466 const struct iov_iter *it)
1467 {
1468 struct sk_buff *skb;
1469 size_t linear;
1470 int err;
1471 int i;
1472
1473 if (it->nr_segs > MAX_SKB_FRAGS + 1)
1474 return ERR_PTR(-ENOMEM);
1475
1476 local_bh_disable();
1477 skb = napi_get_frags(&tfile->napi);
1478 local_bh_enable();
1479 if (!skb)
1480 return ERR_PTR(-ENOMEM);
1481
1482 linear = iov_iter_single_seg_count(it);
1483 err = __skb_grow(skb, linear);
1484 if (err)
1485 goto free;
1486
1487 skb->len = len;
1488 skb->data_len = len - linear;
1489 skb->truesize += skb->data_len;
1490
1491 for (i = 1; i < it->nr_segs; i++) {
1492 struct page_frag *pfrag = &current->task_frag;
1493 size_t fragsz = it->iov[i].iov_len;
1494
1495 if (fragsz == 0 || fragsz > PAGE_SIZE) {
1496 err = -EINVAL;
1497 goto free;
1498 }
1499
1500 if (!skb_page_frag_refill(fragsz, pfrag, GFP_KERNEL)) {
1501 err = -ENOMEM;
1502 goto free;
1503 }
1504
1505 skb_fill_page_desc(skb, i - 1, pfrag->page,
1506 pfrag->offset, fragsz);
1507 page_ref_inc(pfrag->page);
1508 pfrag->offset += fragsz;
1509 }
1510
1511 return skb;
1512 free:
1513 /* frees skb and all frags allocated with napi_alloc_frag() */
1514 napi_free_frags(&tfile->napi);
1515 return ERR_PTR(err);
1516 }
1517
1518 /* prepad is the amount to reserve at front. len is length after that.
1519 * linear is a hint as to how much to copy (usually headers). */
1520 static struct sk_buff *tun_alloc_skb(struct tun_file *tfile,
1521 size_t prepad, size_t len,
1522 size_t linear, int noblock)
1523 {
1524 struct sock *sk = tfile->socket.sk;
1525 struct sk_buff *skb;
1526 int err;
1527
1528 /* Under a page? Don't bother with paged skb. */
1529 if (prepad + len < PAGE_SIZE || !linear)
1530 linear = len;
1531
1532 skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
1533 &err, 0);
1534 if (!skb)
1535 return ERR_PTR(err);
1536
1537 skb_reserve(skb, prepad);
1538 skb_put(skb, linear);
1539 skb->data_len = len - linear;
1540 skb->len += len - linear;
1541
1542 return skb;
1543 }
1544
1545 static void tun_rx_batched(struct tun_struct *tun, struct tun_file *tfile,
1546 struct sk_buff *skb, int more)
1547 {
1548 struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
1549 struct sk_buff_head process_queue;
1550 u32 rx_batched = tun->rx_batched;
1551 bool rcv = false;
1552
1553 if (!rx_batched || (!more && skb_queue_empty(queue))) {
1554 local_bh_disable();
1555 netif_receive_skb(skb);
1556 local_bh_enable();
1557 return;
1558 }
1559
1560 spin_lock(&queue->lock);
1561 if (!more || skb_queue_len(queue) == rx_batched) {
1562 __skb_queue_head_init(&process_queue);
1563 skb_queue_splice_tail_init(queue, &process_queue);
1564 rcv = true;
1565 } else {
1566 __skb_queue_tail(queue, skb);
1567 }
1568 spin_unlock(&queue->lock);
1569
1570 if (rcv) {
1571 struct sk_buff *nskb;
1572
1573 local_bh_disable();
1574 while ((nskb = __skb_dequeue(&process_queue)))
1575 netif_receive_skb(nskb);
1576 netif_receive_skb(skb);
1577 local_bh_enable();
1578 }
1579 }
1580
1581 static bool tun_can_build_skb(struct tun_struct *tun, struct tun_file *tfile,
1582 int len, int noblock, bool zerocopy)
1583 {
1584 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
1585 return false;
1586
1587 if (tfile->socket.sk->sk_sndbuf != INT_MAX)
1588 return false;
1589
1590 if (!noblock)
1591 return false;
1592
1593 if (zerocopy)
1594 return false;
1595
1596 if (SKB_DATA_ALIGN(len + TUN_RX_PAD) +
1597 SKB_DATA_ALIGN(sizeof(struct skb_shared_info)) > PAGE_SIZE)
1598 return false;
1599
1600 return true;
1601 }
1602
1603 static struct sk_buff *tun_build_skb(struct tun_struct *tun,
1604 struct tun_file *tfile,
1605 struct iov_iter *from,
1606 struct virtio_net_hdr *hdr,
1607 int len, int *skb_xdp)
1608 {
1609 struct page_frag *alloc_frag = &current->task_frag;
1610 struct sk_buff *skb;
1611 struct bpf_prog *xdp_prog;
1612 int buflen = SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
1613 unsigned int delta = 0;
1614 char *buf;
1615 size_t copied;
1616 bool xdp_xmit = false;
1617 int err, pad = TUN_RX_PAD;
1618
1619 rcu_read_lock();
1620 xdp_prog = rcu_dereference(tun->xdp_prog);
1621 if (xdp_prog)
1622 pad += TUN_HEADROOM;
1623 buflen += SKB_DATA_ALIGN(len + pad);
1624 rcu_read_unlock();
1625
1626 alloc_frag->offset = ALIGN((u64)alloc_frag->offset, SMP_CACHE_BYTES);
1627 if (unlikely(!skb_page_frag_refill(buflen, alloc_frag, GFP_KERNEL)))
1628 return ERR_PTR(-ENOMEM);
1629
1630 buf = (char *)page_address(alloc_frag->page) + alloc_frag->offset;
1631 copied = copy_page_from_iter(alloc_frag->page,
1632 alloc_frag->offset + pad,
1633 len, from);
1634 if (copied != len)
1635 return ERR_PTR(-EFAULT);
1636
1637 /* There's a small window that XDP may be set after the check
1638 * of xdp_prog above, this should be rare and for simplicity
1639 * we do XDP on skb in case the headroom is not enough.
1640 */
1641 if (hdr->gso_type || !xdp_prog)
1642 *skb_xdp = 1;
1643 else
1644 *skb_xdp = 0;
1645
1646 rcu_read_lock();
1647 xdp_prog = rcu_dereference(tun->xdp_prog);
1648 if (xdp_prog && !*skb_xdp) {
1649 struct xdp_buff xdp;
1650 void *orig_data;
1651 u32 act;
1652
1653 xdp.data_hard_start = buf;
1654 xdp.data = buf + pad;
1655 xdp_set_data_meta_invalid(&xdp);
1656 xdp.data_end = xdp.data + len;
1657 xdp.rxq = &tfile->xdp_rxq;
1658 orig_data = xdp.data;
1659 act = bpf_prog_run_xdp(xdp_prog, &xdp);
1660
1661 switch (act) {
1662 case XDP_REDIRECT:
1663 get_page(alloc_frag->page);
1664 alloc_frag->offset += buflen;
1665 ++tfile->xdp_pending_pkts;
1666 err = xdp_do_redirect(tun->dev, &xdp, xdp_prog);
1667 if (err)
1668 goto err_redirect;
1669 rcu_read_unlock();
1670 return NULL;
1671 case XDP_TX:
1672 xdp_xmit = true;
1673 /* fall through */
1674 case XDP_PASS:
1675 delta = orig_data - xdp.data;
1676 break;
1677 default:
1678 bpf_warn_invalid_xdp_action(act);
1679 /* fall through */
1680 case XDP_ABORTED:
1681 trace_xdp_exception(tun->dev, xdp_prog, act);
1682 /* fall through */
1683 case XDP_DROP:
1684 goto err_xdp;
1685 }
1686 }
1687
1688 skb = build_skb(buf, buflen);
1689 if (!skb) {
1690 rcu_read_unlock();
1691 return ERR_PTR(-ENOMEM);
1692 }
1693
1694 skb_reserve(skb, pad - delta);
1695 skb_put(skb, len + delta);
1696 get_page(alloc_frag->page);
1697 alloc_frag->offset += buflen;
1698
1699 if (xdp_xmit) {
1700 skb->dev = tun->dev;
1701 generic_xdp_tx(skb, xdp_prog);
1702 rcu_read_unlock();
1703 return NULL;
1704 }
1705
1706 rcu_read_unlock();
1707
1708 return skb;
1709
1710 err_redirect:
1711 put_page(alloc_frag->page);
1712 err_xdp:
1713 rcu_read_unlock();
1714 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1715 return NULL;
1716 }
1717
1718 /* Get packet from user space buffer */
1719 static ssize_t tun_get_user(struct tun_struct *tun, struct tun_file *tfile,
1720 void *msg_control, struct iov_iter *from,
1721 int noblock, bool more)
1722 {
1723 struct tun_pi pi = { 0, cpu_to_be16(ETH_P_IP) };
1724 struct sk_buff *skb;
1725 size_t total_len = iov_iter_count(from);
1726 size_t len = total_len, align = tun->align, linear;
1727 struct virtio_net_hdr gso = { 0 };
1728 struct tun_pcpu_stats *stats;
1729 int good_linear;
1730 int copylen;
1731 bool zerocopy = false;
1732 int err;
1733 u32 rxhash = 0;
1734 int skb_xdp = 1;
1735 bool frags = tun_napi_frags_enabled(tun);
1736
1737 if (!(tun->dev->flags & IFF_UP))
1738 return -EIO;
1739
1740 if (!(tun->flags & IFF_NO_PI)) {
1741 if (len < sizeof(pi))
1742 return -EINVAL;
1743 len -= sizeof(pi);
1744
1745 if (!copy_from_iter_full(&pi, sizeof(pi), from))
1746 return -EFAULT;
1747 }
1748
1749 if (tun->flags & IFF_VNET_HDR) {
1750 int vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
1751
1752 if (len < vnet_hdr_sz)
1753 return -EINVAL;
1754 len -= vnet_hdr_sz;
1755
1756 if (!copy_from_iter_full(&gso, sizeof(gso), from))
1757 return -EFAULT;
1758
1759 if ((gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
1760 tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2 > tun16_to_cpu(tun, gso.hdr_len))
1761 gso.hdr_len = cpu_to_tun16(tun, tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2);
1762
1763 if (tun16_to_cpu(tun, gso.hdr_len) > len)
1764 return -EINVAL;
1765 iov_iter_advance(from, vnet_hdr_sz - sizeof(gso));
1766 }
1767
1768 if ((tun->flags & TUN_TYPE_MASK) == IFF_TAP) {
1769 align += NET_IP_ALIGN;
1770 if (unlikely(len < ETH_HLEN ||
1771 (gso.hdr_len && tun16_to_cpu(tun, gso.hdr_len) < ETH_HLEN)))
1772 return -EINVAL;
1773 }
1774
1775 good_linear = SKB_MAX_HEAD(align);
1776
1777 if (msg_control) {
1778 struct iov_iter i = *from;
1779
1780 /* There are 256 bytes to be copied in skb, so there is
1781 * enough room for skb expand head in case it is used.
1782 * The rest of the buffer is mapped from userspace.
1783 */
1784 copylen = gso.hdr_len ? tun16_to_cpu(tun, gso.hdr_len) : GOODCOPY_LEN;
1785 if (copylen > good_linear)
1786 copylen = good_linear;
1787 linear = copylen;
1788 iov_iter_advance(&i, copylen);
1789 if (iov_iter_npages(&i, INT_MAX) <= MAX_SKB_FRAGS)
1790 zerocopy = true;
1791 }
1792
1793 if (!frags && tun_can_build_skb(tun, tfile, len, noblock, zerocopy)) {
1794 /* For the packet that is not easy to be processed
1795 * (e.g gso or jumbo packet), we will do it at after
1796 * skb was created with generic XDP routine.
1797 */
1798 skb = tun_build_skb(tun, tfile, from, &gso, len, &skb_xdp);
1799 if (IS_ERR(skb)) {
1800 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1801 return PTR_ERR(skb);
1802 }
1803 if (!skb)
1804 return total_len;
1805 } else {
1806 if (!zerocopy) {
1807 copylen = len;
1808 if (tun16_to_cpu(tun, gso.hdr_len) > good_linear)
1809 linear = good_linear;
1810 else
1811 linear = tun16_to_cpu(tun, gso.hdr_len);
1812 }
1813
1814 if (frags) {
1815 mutex_lock(&tfile->napi_mutex);
1816 skb = tun_napi_alloc_frags(tfile, copylen, from);
1817 /* tun_napi_alloc_frags() enforces a layout for the skb.
1818 * If zerocopy is enabled, then this layout will be
1819 * overwritten by zerocopy_sg_from_iter().
1820 */
1821 zerocopy = false;
1822 } else {
1823 skb = tun_alloc_skb(tfile, align, copylen, linear,
1824 noblock);
1825 }
1826
1827 if (IS_ERR(skb)) {
1828 if (PTR_ERR(skb) != -EAGAIN)
1829 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1830 if (frags)
1831 mutex_unlock(&tfile->napi_mutex);
1832 return PTR_ERR(skb);
1833 }
1834
1835 if (zerocopy)
1836 err = zerocopy_sg_from_iter(skb, from);
1837 else
1838 err = skb_copy_datagram_from_iter(skb, 0, from, len);
1839
1840 if (err) {
1841 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1842 kfree_skb(skb);
1843 if (frags) {
1844 tfile->napi.skb = NULL;
1845 mutex_unlock(&tfile->napi_mutex);
1846 }
1847
1848 return -EFAULT;
1849 }
1850 }
1851
1852 if (virtio_net_hdr_to_skb(skb, &gso, tun_is_little_endian(tun))) {
1853 this_cpu_inc(tun->pcpu_stats->rx_frame_errors);
1854 kfree_skb(skb);
1855 if (frags) {
1856 tfile->napi.skb = NULL;
1857 mutex_unlock(&tfile->napi_mutex);
1858 }
1859
1860 return -EINVAL;
1861 }
1862
1863 switch (tun->flags & TUN_TYPE_MASK) {
1864 case IFF_TUN:
1865 if (tun->flags & IFF_NO_PI) {
1866 u8 ip_version = skb->len ? (skb->data[0] >> 4) : 0;
1867
1868 switch (ip_version) {
1869 case 4:
1870 pi.proto = htons(ETH_P_IP);
1871 break;
1872 case 6:
1873 pi.proto = htons(ETH_P_IPV6);
1874 break;
1875 default:
1876 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1877 kfree_skb(skb);
1878 return -EINVAL;
1879 }
1880 }
1881
1882 skb_reset_mac_header(skb);
1883 skb->protocol = pi.proto;
1884 skb->dev = tun->dev;
1885 break;
1886 case IFF_TAP:
1887 if (!frags)
1888 skb->protocol = eth_type_trans(skb, tun->dev);
1889 break;
1890 }
1891
1892 /* copy skb_ubuf_info for callback when skb has no error */
1893 if (zerocopy) {
1894 skb_shinfo(skb)->destructor_arg = msg_control;
1895 skb_shinfo(skb)->tx_flags |= SKBTX_DEV_ZEROCOPY;
1896 skb_shinfo(skb)->tx_flags |= SKBTX_SHARED_FRAG;
1897 } else if (msg_control) {
1898 struct ubuf_info *uarg = msg_control;
1899 uarg->callback(uarg, false);
1900 }
1901
1902 skb_reset_network_header(skb);
1903 skb_probe_transport_header(skb, 0);
1904
1905 if (skb_xdp) {
1906 struct bpf_prog *xdp_prog;
1907 int ret;
1908
1909 rcu_read_lock();
1910 xdp_prog = rcu_dereference(tun->xdp_prog);
1911 if (xdp_prog) {
1912 ret = do_xdp_generic(xdp_prog, skb);
1913 if (ret != XDP_PASS) {
1914 rcu_read_unlock();
1915 return total_len;
1916 }
1917 }
1918 rcu_read_unlock();
1919 }
1920
1921 rcu_read_lock();
1922 if (!rcu_dereference(tun->steering_prog))
1923 rxhash = __skb_get_hash_symmetric(skb);
1924 rcu_read_unlock();
1925
1926 if (frags) {
1927 /* Exercise flow dissector code path. */
1928 u32 headlen = eth_get_headlen(skb->data, skb_headlen(skb));
1929
1930 if (unlikely(headlen > skb_headlen(skb))) {
1931 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1932 napi_free_frags(&tfile->napi);
1933 mutex_unlock(&tfile->napi_mutex);
1934 WARN_ON(1);
1935 return -ENOMEM;
1936 }
1937
1938 local_bh_disable();
1939 napi_gro_frags(&tfile->napi);
1940 local_bh_enable();
1941 mutex_unlock(&tfile->napi_mutex);
1942 } else if (tfile->napi_enabled) {
1943 struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
1944 int queue_len;
1945
1946 spin_lock_bh(&queue->lock);
1947 __skb_queue_tail(queue, skb);
1948 queue_len = skb_queue_len(queue);
1949 spin_unlock(&queue->lock);
1950
1951 if (!more || queue_len > NAPI_POLL_WEIGHT)
1952 napi_schedule(&tfile->napi);
1953
1954 local_bh_enable();
1955 } else if (!IS_ENABLED(CONFIG_4KSTACKS)) {
1956 tun_rx_batched(tun, tfile, skb, more);
1957 } else {
1958 netif_rx_ni(skb);
1959 }
1960
1961 stats = get_cpu_ptr(tun->pcpu_stats);
1962 u64_stats_update_begin(&stats->syncp);
1963 stats->rx_packets++;
1964 stats->rx_bytes += len;
1965 u64_stats_update_end(&stats->syncp);
1966 put_cpu_ptr(stats);
1967
1968 if (rxhash)
1969 tun_flow_update(tun, rxhash, tfile);
1970
1971 return total_len;
1972 }
1973
1974 static ssize_t tun_chr_write_iter(struct kiocb *iocb, struct iov_iter *from)
1975 {
1976 struct file *file = iocb->ki_filp;
1977 struct tun_file *tfile = file->private_data;
1978 struct tun_struct *tun = tun_get(tfile);
1979 ssize_t result;
1980
1981 if (!tun)
1982 return -EBADFD;
1983
1984 result = tun_get_user(tun, tfile, NULL, from,
1985 file->f_flags & O_NONBLOCK, false);
1986
1987 if (tfile->xdp_pending_pkts) {
1988 tfile->xdp_pending_pkts = 0;
1989 xdp_do_flush_map();
1990 }
1991
1992 tun_put(tun);
1993 return result;
1994 }
1995
1996 static ssize_t tun_put_user_xdp(struct tun_struct *tun,
1997 struct tun_file *tfile,
1998 struct xdp_buff *xdp,
1999 struct iov_iter *iter)
2000 {
2001 int vnet_hdr_sz = 0;
2002 size_t size = xdp->data_end - xdp->data;
2003 struct tun_pcpu_stats *stats;
2004 size_t ret;
2005
2006 if (tun->flags & IFF_VNET_HDR) {
2007 struct virtio_net_hdr gso = { 0 };
2008
2009 vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
2010 if (unlikely(iov_iter_count(iter) < vnet_hdr_sz))
2011 return -EINVAL;
2012 if (unlikely(copy_to_iter(&gso, sizeof(gso), iter) !=
2013 sizeof(gso)))
2014 return -EFAULT;
2015 iov_iter_advance(iter, vnet_hdr_sz - sizeof(gso));
2016 }
2017
2018 ret = copy_to_iter(xdp->data, size, iter) + vnet_hdr_sz;
2019
2020 stats = get_cpu_ptr(tun->pcpu_stats);
2021 u64_stats_update_begin(&stats->syncp);
2022 stats->tx_packets++;
2023 stats->tx_bytes += ret;
2024 u64_stats_update_end(&stats->syncp);
2025 put_cpu_ptr(tun->pcpu_stats);
2026
2027 return ret;
2028 }
2029
2030 /* Put packet to the user space buffer */
2031 static ssize_t tun_put_user(struct tun_struct *tun,
2032 struct tun_file *tfile,
2033 struct sk_buff *skb,
2034 struct iov_iter *iter)
2035 {
2036 struct tun_pi pi = { 0, skb->protocol };
2037 struct tun_pcpu_stats *stats;
2038 ssize_t total;
2039 int vlan_offset = 0;
2040 int vlan_hlen = 0;
2041 int vnet_hdr_sz = 0;
2042
2043 if (skb_vlan_tag_present(skb))
2044 vlan_hlen = VLAN_HLEN;
2045
2046 if (tun->flags & IFF_VNET_HDR)
2047 vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
2048
2049 total = skb->len + vlan_hlen + vnet_hdr_sz;
2050
2051 if (!(tun->flags & IFF_NO_PI)) {
2052 if (iov_iter_count(iter) < sizeof(pi))
2053 return -EINVAL;
2054
2055 total += sizeof(pi);
2056 if (iov_iter_count(iter) < total) {
2057 /* Packet will be striped */
2058 pi.flags |= TUN_PKT_STRIP;
2059 }
2060
2061 if (copy_to_iter(&pi, sizeof(pi), iter) != sizeof(pi))
2062 return -EFAULT;
2063 }
2064
2065 if (vnet_hdr_sz) {
2066 struct virtio_net_hdr gso;
2067
2068 if (iov_iter_count(iter) < vnet_hdr_sz)
2069 return -EINVAL;
2070
2071 if (virtio_net_hdr_from_skb(skb, &gso,
2072 tun_is_little_endian(tun), true)) {
2073 struct skb_shared_info *sinfo = skb_shinfo(skb);
2074 pr_err("unexpected GSO type: "
2075 "0x%x, gso_size %d, hdr_len %d\n",
2076 sinfo->gso_type, tun16_to_cpu(tun, gso.gso_size),
2077 tun16_to_cpu(tun, gso.hdr_len));
2078 print_hex_dump(KERN_ERR, "tun: ",
2079 DUMP_PREFIX_NONE,
2080 16, 1, skb->head,
2081 min((int)tun16_to_cpu(tun, gso.hdr_len), 64), true);
2082 WARN_ON_ONCE(1);
2083 return -EINVAL;
2084 }
2085
2086 if (copy_to_iter(&gso, sizeof(gso), iter) != sizeof(gso))
2087 return -EFAULT;
2088
2089 iov_iter_advance(iter, vnet_hdr_sz - sizeof(gso));
2090 }
2091
2092 if (vlan_hlen) {
2093 int ret;
2094 struct veth veth;
2095
2096 veth.h_vlan_proto = skb->vlan_proto;
2097 veth.h_vlan_TCI = htons(skb_vlan_tag_get(skb));
2098
2099 vlan_offset = offsetof(struct vlan_ethhdr, h_vlan_proto);
2100
2101 ret = skb_copy_datagram_iter(skb, 0, iter, vlan_offset);
2102 if (ret || !iov_iter_count(iter))
2103 goto done;
2104
2105 ret = copy_to_iter(&veth, sizeof(veth), iter);
2106 if (ret != sizeof(veth) || !iov_iter_count(iter))
2107 goto done;
2108 }
2109
2110 skb_copy_datagram_iter(skb, vlan_offset, iter, skb->len - vlan_offset);
2111
2112 done:
2113 /* caller is in process context, */
2114 stats = get_cpu_ptr(tun->pcpu_stats);
2115 u64_stats_update_begin(&stats->syncp);
2116 stats->tx_packets++;
2117 stats->tx_bytes += skb->len + vlan_hlen;
2118 u64_stats_update_end(&stats->syncp);
2119 put_cpu_ptr(tun->pcpu_stats);
2120
2121 return total;
2122 }
2123
2124 static void *tun_ring_recv(struct tun_file *tfile, int noblock, int *err)
2125 {
2126 DECLARE_WAITQUEUE(wait, current);
2127 void *ptr = NULL;
2128 int error = 0;
2129
2130 ptr = ptr_ring_consume(&tfile->tx_ring);
2131 if (ptr)
2132 goto out;
2133 if (noblock) {
2134 error = -EAGAIN;
2135 goto out;
2136 }
2137
2138 add_wait_queue(&tfile->wq.wait, &wait);
2139 current->state = TASK_INTERRUPTIBLE;
2140
2141 while (1) {
2142 ptr = ptr_ring_consume(&tfile->tx_ring);
2143 if (ptr)
2144 break;
2145 if (signal_pending(current)) {
2146 error = -ERESTARTSYS;
2147 break;
2148 }
2149 if (tfile->socket.sk->sk_shutdown & RCV_SHUTDOWN) {
2150 error = -EFAULT;
2151 break;
2152 }
2153
2154 schedule();
2155 }
2156
2157 current->state = TASK_RUNNING;
2158 remove_wait_queue(&tfile->wq.wait, &wait);
2159
2160 out:
2161 *err = error;
2162 return ptr;
2163 }
2164
2165 static ssize_t tun_do_read(struct tun_struct *tun, struct tun_file *tfile,
2166 struct iov_iter *to,
2167 int noblock, void *ptr)
2168 {
2169 ssize_t ret;
2170 int err;
2171
2172 tun_debug(KERN_INFO, tun, "tun_do_read\n");
2173
2174 if (!iov_iter_count(to)) {
2175 tun_ptr_free(ptr);
2176 return 0;
2177 }
2178
2179 if (!ptr) {
2180 /* Read frames from ring */
2181 ptr = tun_ring_recv(tfile, noblock, &err);
2182 if (!ptr)
2183 return err;
2184 }
2185
2186 if (tun_is_xdp_buff(ptr)) {
2187 struct xdp_buff *xdp = tun_ptr_to_xdp(ptr);
2188
2189 ret = tun_put_user_xdp(tun, tfile, xdp, to);
2190 put_page(virt_to_head_page(xdp->data));
2191 } else {
2192 struct sk_buff *skb = ptr;
2193
2194 ret = tun_put_user(tun, tfile, skb, to);
2195 if (unlikely(ret < 0))
2196 kfree_skb(skb);
2197 else
2198 consume_skb(skb);
2199 }
2200
2201 return ret;
2202 }
2203
2204 static ssize_t tun_chr_read_iter(struct kiocb *iocb, struct iov_iter *to)
2205 {
2206 struct file *file = iocb->ki_filp;
2207 struct tun_file *tfile = file->private_data;
2208 struct tun_struct *tun = tun_get(tfile);
2209 ssize_t len = iov_iter_count(to), ret;
2210
2211 if (!tun)
2212 return -EBADFD;
2213 ret = tun_do_read(tun, tfile, to, file->f_flags & O_NONBLOCK, NULL);
2214 ret = min_t(ssize_t, ret, len);
2215 if (ret > 0)
2216 iocb->ki_pos = ret;
2217 tun_put(tun);
2218 return ret;
2219 }
2220
2221 static void tun_prog_free(struct rcu_head *rcu)
2222 {
2223 struct tun_prog *prog = container_of(rcu, struct tun_prog, rcu);
2224
2225 bpf_prog_destroy(prog->prog);
2226 kfree(prog);
2227 }
2228
2229 static int __tun_set_ebpf(struct tun_struct *tun,
2230 struct tun_prog __rcu **prog_p,
2231 struct bpf_prog *prog)
2232 {
2233 struct tun_prog *old, *new = NULL;
2234
2235 if (prog) {
2236 new = kmalloc(sizeof(*new), GFP_KERNEL);
2237 if (!new)
2238 return -ENOMEM;
2239 new->prog = prog;
2240 }
2241
2242 spin_lock_bh(&tun->lock);
2243 old = rcu_dereference_protected(*prog_p,
2244 lockdep_is_held(&tun->lock));
2245 rcu_assign_pointer(*prog_p, new);
2246 spin_unlock_bh(&tun->lock);
2247
2248 if (old)
2249 call_rcu(&old->rcu, tun_prog_free);
2250
2251 return 0;
2252 }
2253
2254 static void tun_free_netdev(struct net_device *dev)
2255 {
2256 struct tun_struct *tun = netdev_priv(dev);
2257
2258 BUG_ON(!(list_empty(&tun->disabled)));
2259 free_percpu(tun->pcpu_stats);
2260 tun_flow_uninit(tun);
2261 security_tun_dev_free_security(tun->security);
2262 __tun_set_ebpf(tun, &tun->steering_prog, NULL);
2263 __tun_set_ebpf(tun, &tun->filter_prog, NULL);
2264 }
2265
2266 static void tun_setup(struct net_device *dev)
2267 {
2268 struct tun_struct *tun = netdev_priv(dev);
2269
2270 tun->owner = INVALID_UID;
2271 tun->group = INVALID_GID;
2272
2273 dev->ethtool_ops = &tun_ethtool_ops;
2274 dev->needs_free_netdev = true;
2275 dev->priv_destructor = tun_free_netdev;
2276 /* We prefer our own queue length */
2277 dev->tx_queue_len = TUN_READQ_SIZE;
2278 }
2279
2280 /* Trivial set of netlink ops to allow deleting tun or tap
2281 * device with netlink.
2282 */
2283 static int tun_validate(struct nlattr *tb[], struct nlattr *data[],
2284 struct netlink_ext_ack *extack)
2285 {
2286 return -EINVAL;
2287 }
2288
2289 static struct rtnl_link_ops tun_link_ops __read_mostly = {
2290 .kind = DRV_NAME,
2291 .priv_size = sizeof(struct tun_struct),
2292 .setup = tun_setup,
2293 .validate = tun_validate,
2294 };
2295
2296 static void tun_sock_write_space(struct sock *sk)
2297 {
2298 struct tun_file *tfile;
2299 wait_queue_head_t *wqueue;
2300
2301 if (!sock_writeable(sk))
2302 return;
2303
2304 if (!test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags))
2305 return;
2306
2307 wqueue = sk_sleep(sk);
2308 if (wqueue && waitqueue_active(wqueue))
2309 wake_up_interruptible_sync_poll(wqueue, EPOLLOUT |
2310 EPOLLWRNORM | EPOLLWRBAND);
2311
2312 tfile = container_of(sk, struct tun_file, sk);
2313 kill_fasync(&tfile->fasync, SIGIO, POLL_OUT);
2314 }
2315
2316 static int tun_sendmsg(struct socket *sock, struct msghdr *m, size_t total_len)
2317 {
2318 int ret;
2319 struct tun_file *tfile = container_of(sock, struct tun_file, socket);
2320 struct tun_struct *tun = tun_get(tfile);
2321
2322 if (!tun)
2323 return -EBADFD;
2324
2325 ret = tun_get_user(tun, tfile, m->msg_control, &m->msg_iter,
2326 m->msg_flags & MSG_DONTWAIT,
2327 m->msg_flags & MSG_MORE);
2328
2329 if (tfile->xdp_pending_pkts >= NAPI_POLL_WEIGHT ||
2330 !(m->msg_flags & MSG_MORE)) {
2331 tfile->xdp_pending_pkts = 0;
2332 xdp_do_flush_map();
2333 }
2334
2335 tun_put(tun);
2336 return ret;
2337 }
2338
2339 static int tun_recvmsg(struct socket *sock, struct msghdr *m, size_t total_len,
2340 int flags)
2341 {
2342 struct tun_file *tfile = container_of(sock, struct tun_file, socket);
2343 struct tun_struct *tun = tun_get(tfile);
2344 void *ptr = m->msg_control;
2345 int ret;
2346
2347 if (!tun) {
2348 ret = -EBADFD;
2349 goto out_free;
2350 }
2351
2352 if (flags & ~(MSG_DONTWAIT|MSG_TRUNC|MSG_ERRQUEUE)) {
2353 ret = -EINVAL;
2354 goto out_put_tun;
2355 }
2356 if (flags & MSG_ERRQUEUE) {
2357 ret = sock_recv_errqueue(sock->sk, m, total_len,
2358 SOL_PACKET, TUN_TX_TIMESTAMP);
2359 goto out;
2360 }
2361 ret = tun_do_read(tun, tfile, &m->msg_iter, flags & MSG_DONTWAIT, ptr);
2362 if (ret > (ssize_t)total_len) {
2363 m->msg_flags |= MSG_TRUNC;
2364 ret = flags & MSG_TRUNC ? ret : total_len;
2365 }
2366 out:
2367 tun_put(tun);
2368 return ret;
2369
2370 out_put_tun:
2371 tun_put(tun);
2372 out_free:
2373 tun_ptr_free(ptr);
2374 return ret;
2375 }
2376
2377 static int tun_ptr_peek_len(void *ptr)
2378 {
2379 if (likely(ptr)) {
2380 if (tun_is_xdp_buff(ptr)) {
2381 struct xdp_buff *xdp = tun_ptr_to_xdp(ptr);
2382
2383 return xdp->data_end - xdp->data;
2384 }
2385 return __skb_array_len_with_tag(ptr);
2386 } else {
2387 return 0;
2388 }
2389 }
2390
2391 static int tun_peek_len(struct socket *sock)
2392 {
2393 struct tun_file *tfile = container_of(sock, struct tun_file, socket);
2394 struct tun_struct *tun;
2395 int ret = 0;
2396
2397 tun = tun_get(tfile);
2398 if (!tun)
2399 return 0;
2400
2401 ret = PTR_RING_PEEK_CALL(&tfile->tx_ring, tun_ptr_peek_len);
2402 tun_put(tun);
2403
2404 return ret;
2405 }
2406
2407 /* Ops structure to mimic raw sockets with tun */
2408 static const struct proto_ops tun_socket_ops = {
2409 .peek_len = tun_peek_len,
2410 .sendmsg = tun_sendmsg,
2411 .recvmsg = tun_recvmsg,
2412 };
2413
2414 static struct proto tun_proto = {
2415 .name = "tun",
2416 .owner = THIS_MODULE,
2417 .obj_size = sizeof(struct tun_file),
2418 };
2419
2420 static int tun_flags(struct tun_struct *tun)
2421 {
2422 return tun->flags & (TUN_FEATURES | IFF_PERSIST | IFF_TUN | IFF_TAP);
2423 }
2424
2425 static ssize_t tun_show_flags(struct device *dev, struct device_attribute *attr,
2426 char *buf)
2427 {
2428 struct tun_struct *tun = netdev_priv(to_net_dev(dev));
2429 return sprintf(buf, "0x%x\n", tun_flags(tun));
2430 }
2431
2432 static ssize_t tun_show_owner(struct device *dev, struct device_attribute *attr,
2433 char *buf)
2434 {
2435 struct tun_struct *tun = netdev_priv(to_net_dev(dev));
2436 return uid_valid(tun->owner)?
2437 sprintf(buf, "%u\n",
2438 from_kuid_munged(current_user_ns(), tun->owner)):
2439 sprintf(buf, "-1\n");
2440 }
2441
2442 static ssize_t tun_show_group(struct device *dev, struct device_attribute *attr,
2443 char *buf)
2444 {
2445 struct tun_struct *tun = netdev_priv(to_net_dev(dev));
2446 return gid_valid(tun->group) ?
2447 sprintf(buf, "%u\n",
2448 from_kgid_munged(current_user_ns(), tun->group)):
2449 sprintf(buf, "-1\n");
2450 }
2451
2452 static DEVICE_ATTR(tun_flags, 0444, tun_show_flags, NULL);
2453 static DEVICE_ATTR(owner, 0444, tun_show_owner, NULL);
2454 static DEVICE_ATTR(group, 0444, tun_show_group, NULL);
2455
2456 static struct attribute *tun_dev_attrs[] = {
2457 &dev_attr_tun_flags.attr,
2458 &dev_attr_owner.attr,
2459 &dev_attr_group.attr,
2460 NULL
2461 };
2462
2463 static const struct attribute_group tun_attr_group = {
2464 .attrs = tun_dev_attrs
2465 };
2466
2467 static int tun_set_iff(struct net *net, struct file *file, struct ifreq *ifr)
2468 {
2469 struct tun_struct *tun;
2470 struct tun_file *tfile = file->private_data;
2471 struct net_device *dev;
2472 int err;
2473
2474 if (tfile->detached)
2475 return -EINVAL;
2476
2477 if ((ifr->ifr_flags & IFF_NAPI_FRAGS)) {
2478 if (!capable(CAP_NET_ADMIN))
2479 return -EPERM;
2480
2481 if (!(ifr->ifr_flags & IFF_NAPI) ||
2482 (ifr->ifr_flags & TUN_TYPE_MASK) != IFF_TAP)
2483 return -EINVAL;
2484 }
2485
2486 dev = __dev_get_by_name(net, ifr->ifr_name);
2487 if (dev) {
2488 if (ifr->ifr_flags & IFF_TUN_EXCL)
2489 return -EBUSY;
2490 if ((ifr->ifr_flags & IFF_TUN) && dev->netdev_ops == &tun_netdev_ops)
2491 tun = netdev_priv(dev);
2492 else if ((ifr->ifr_flags & IFF_TAP) && dev->netdev_ops == &tap_netdev_ops)
2493 tun = netdev_priv(dev);
2494 else
2495 return -EINVAL;
2496
2497 if (!!(ifr->ifr_flags & IFF_MULTI_QUEUE) !=
2498 !!(tun->flags & IFF_MULTI_QUEUE))
2499 return -EINVAL;
2500
2501 if (tun_not_capable(tun))
2502 return -EPERM;
2503 err = security_tun_dev_open(tun->security);
2504 if (err < 0)
2505 return err;
2506
2507 err = tun_attach(tun, file, ifr->ifr_flags & IFF_NOFILTER,
2508 ifr->ifr_flags & IFF_NAPI);
2509 if (err < 0)
2510 return err;
2511
2512 if (tun->flags & IFF_MULTI_QUEUE &&
2513 (tun->numqueues + tun->numdisabled > 1)) {
2514 /* One or more queue has already been attached, no need
2515 * to initialize the device again.
2516 */
2517 return 0;
2518 }
2519 }
2520 else {
2521 char *name;
2522 unsigned long flags = 0;
2523 int queues = ifr->ifr_flags & IFF_MULTI_QUEUE ?
2524 MAX_TAP_QUEUES : 1;
2525
2526 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
2527 return -EPERM;
2528 err = security_tun_dev_create();
2529 if (err < 0)
2530 return err;
2531
2532 /* Set dev type */
2533 if (ifr->ifr_flags & IFF_TUN) {
2534 /* TUN device */
2535 flags |= IFF_TUN;
2536 name = "tun%d";
2537 } else if (ifr->ifr_flags & IFF_TAP) {
2538 /* TAP device */
2539 flags |= IFF_TAP;
2540 name = "tap%d";
2541 } else
2542 return -EINVAL;
2543
2544 if (*ifr->ifr_name)
2545 name = ifr->ifr_name;
2546
2547 dev = alloc_netdev_mqs(sizeof(struct tun_struct), name,
2548 NET_NAME_UNKNOWN, tun_setup, queues,
2549 queues);
2550
2551 if (!dev)
2552 return -ENOMEM;
2553 err = dev_get_valid_name(net, dev, name);
2554 if (err < 0)
2555 goto err_free_dev;
2556
2557 dev_net_set(dev, net);
2558 dev->rtnl_link_ops = &tun_link_ops;
2559 dev->ifindex = tfile->ifindex;
2560 dev->sysfs_groups[0] = &tun_attr_group;
2561
2562 tun = netdev_priv(dev);
2563 tun->dev = dev;
2564 tun->flags = flags;
2565 tun->txflt.count = 0;
2566 tun->vnet_hdr_sz = sizeof(struct virtio_net_hdr);
2567
2568 tun->align = NET_SKB_PAD;
2569 tun->filter_attached = false;
2570 tun->sndbuf = tfile->socket.sk->sk_sndbuf;
2571 tun->rx_batched = 0;
2572 RCU_INIT_POINTER(tun->steering_prog, NULL);
2573
2574 tun->pcpu_stats = netdev_alloc_pcpu_stats(struct tun_pcpu_stats);
2575 if (!tun->pcpu_stats) {
2576 err = -ENOMEM;
2577 goto err_free_dev;
2578 }
2579
2580 spin_lock_init(&tun->lock);
2581
2582 err = security_tun_dev_alloc_security(&tun->security);
2583 if (err < 0)
2584 goto err_free_stat;
2585
2586 tun_net_init(dev);
2587 tun_flow_init(tun);
2588
2589 dev->hw_features = NETIF_F_SG | NETIF_F_FRAGLIST |
2590 TUN_USER_FEATURES | NETIF_F_HW_VLAN_CTAG_TX |
2591 NETIF_F_HW_VLAN_STAG_TX;
2592 dev->features = dev->hw_features | NETIF_F_LLTX;
2593 dev->vlan_features = dev->features &
2594 ~(NETIF_F_HW_VLAN_CTAG_TX |
2595 NETIF_F_HW_VLAN_STAG_TX);
2596
2597 INIT_LIST_HEAD(&tun->disabled);
2598 err = tun_attach(tun, file, false, ifr->ifr_flags & IFF_NAPI);
2599 if (err < 0)
2600 goto err_free_flow;
2601
2602 err = register_netdevice(tun->dev);
2603 if (err < 0)
2604 goto err_detach;
2605 }
2606
2607 netif_carrier_on(tun->dev);
2608
2609 tun_debug(KERN_INFO, tun, "tun_set_iff\n");
2610
2611 tun->flags = (tun->flags & ~TUN_FEATURES) |
2612 (ifr->ifr_flags & TUN_FEATURES);
2613
2614 /* Make sure persistent devices do not get stuck in
2615 * xoff state.
2616 */
2617 if (netif_running(tun->dev))
2618 netif_tx_wake_all_queues(tun->dev);
2619
2620 strcpy(ifr->ifr_name, tun->dev->name);
2621 return 0;
2622
2623 err_detach:
2624 tun_detach_all(dev);
2625 /* register_netdevice() already called tun_free_netdev() */
2626 goto err_free_dev;
2627
2628 err_free_flow:
2629 tun_flow_uninit(tun);
2630 security_tun_dev_free_security(tun->security);
2631 err_free_stat:
2632 free_percpu(tun->pcpu_stats);
2633 err_free_dev:
2634 free_netdev(dev);
2635 return err;
2636 }
2637
2638 static void tun_get_iff(struct net *net, struct tun_struct *tun,
2639 struct ifreq *ifr)
2640 {
2641 tun_debug(KERN_INFO, tun, "tun_get_iff\n");
2642
2643 strcpy(ifr->ifr_name, tun->dev->name);
2644
2645 ifr->ifr_flags = tun_flags(tun);
2646
2647 }
2648
2649 /* This is like a cut-down ethtool ops, except done via tun fd so no
2650 * privs required. */
2651 static int set_offload(struct tun_struct *tun, unsigned long arg)
2652 {
2653 netdev_features_t features = 0;
2654
2655 if (arg & TUN_F_CSUM) {
2656 features |= NETIF_F_HW_CSUM;
2657 arg &= ~TUN_F_CSUM;
2658
2659 if (arg & (TUN_F_TSO4|TUN_F_TSO6)) {
2660 if (arg & TUN_F_TSO_ECN) {
2661 features |= NETIF_F_TSO_ECN;
2662 arg &= ~TUN_F_TSO_ECN;
2663 }
2664 if (arg & TUN_F_TSO4)
2665 features |= NETIF_F_TSO;
2666 if (arg & TUN_F_TSO6)
2667 features |= NETIF_F_TSO6;
2668 arg &= ~(TUN_F_TSO4|TUN_F_TSO6);
2669 }
2670
2671 arg &= ~TUN_F_UFO;
2672 }
2673
2674 /* This gives the user a way to test for new features in future by
2675 * trying to set them. */
2676 if (arg)
2677 return -EINVAL;
2678
2679 tun->set_features = features;
2680 tun->dev->wanted_features &= ~TUN_USER_FEATURES;
2681 tun->dev->wanted_features |= features;
2682 netdev_update_features(tun->dev);
2683
2684 return 0;
2685 }
2686
2687 static void tun_detach_filter(struct tun_struct *tun, int n)
2688 {
2689 int i;
2690 struct tun_file *tfile;
2691
2692 for (i = 0; i < n; i++) {
2693 tfile = rtnl_dereference(tun->tfiles[i]);
2694 lock_sock(tfile->socket.sk);
2695 sk_detach_filter(tfile->socket.sk);
2696 release_sock(tfile->socket.sk);
2697 }
2698
2699 tun->filter_attached = false;
2700 }
2701
2702 static int tun_attach_filter(struct tun_struct *tun)
2703 {
2704 int i, ret = 0;
2705 struct tun_file *tfile;
2706
2707 for (i = 0; i < tun->numqueues; i++) {
2708 tfile = rtnl_dereference(tun->tfiles[i]);
2709 lock_sock(tfile->socket.sk);
2710 ret = sk_attach_filter(&tun->fprog, tfile->socket.sk);
2711 release_sock(tfile->socket.sk);
2712 if (ret) {
2713 tun_detach_filter(tun, i);
2714 return ret;
2715 }
2716 }
2717
2718 tun->filter_attached = true;
2719 return ret;
2720 }
2721
2722 static void tun_set_sndbuf(struct tun_struct *tun)
2723 {
2724 struct tun_file *tfile;
2725 int i;
2726
2727 for (i = 0; i < tun->numqueues; i++) {
2728 tfile = rtnl_dereference(tun->tfiles[i]);
2729 tfile->socket.sk->sk_sndbuf = tun->sndbuf;
2730 }
2731 }
2732
2733 static int tun_set_queue(struct file *file, struct ifreq *ifr)
2734 {
2735 struct tun_file *tfile = file->private_data;
2736 struct tun_struct *tun;
2737 int ret = 0;
2738
2739 rtnl_lock();
2740
2741 if (ifr->ifr_flags & IFF_ATTACH_QUEUE) {
2742 tun = tfile->detached;
2743 if (!tun) {
2744 ret = -EINVAL;
2745 goto unlock;
2746 }
2747 ret = security_tun_dev_attach_queue(tun->security);
2748 if (ret < 0)
2749 goto unlock;
2750 ret = tun_attach(tun, file, false, tun->flags & IFF_NAPI);
2751 } else if (ifr->ifr_flags & IFF_DETACH_QUEUE) {
2752 tun = rtnl_dereference(tfile->tun);
2753 if (!tun || !(tun->flags & IFF_MULTI_QUEUE) || tfile->detached)
2754 ret = -EINVAL;
2755 else
2756 __tun_detach(tfile, false);
2757 } else
2758 ret = -EINVAL;
2759
2760 unlock:
2761 rtnl_unlock();
2762 return ret;
2763 }
2764
2765 static int tun_set_ebpf(struct tun_struct *tun, struct tun_prog **prog_p,
2766 void __user *data)
2767 {
2768 struct bpf_prog *prog;
2769 int fd;
2770
2771 if (copy_from_user(&fd, data, sizeof(fd)))
2772 return -EFAULT;
2773
2774 if (fd == -1) {
2775 prog = NULL;
2776 } else {
2777 prog = bpf_prog_get_type(fd, BPF_PROG_TYPE_SOCKET_FILTER);
2778 if (IS_ERR(prog))
2779 return PTR_ERR(prog);
2780 }
2781
2782 return __tun_set_ebpf(tun, prog_p, prog);
2783 }
2784
2785 static long __tun_chr_ioctl(struct file *file, unsigned int cmd,
2786 unsigned long arg, int ifreq_len)
2787 {
2788 struct tun_file *tfile = file->private_data;
2789 struct tun_struct *tun;
2790 void __user* argp = (void __user*)arg;
2791 struct ifreq ifr;
2792 kuid_t owner;
2793 kgid_t group;
2794 int sndbuf;
2795 int vnet_hdr_sz;
2796 unsigned int ifindex;
2797 int le;
2798 int ret;
2799
2800 if (cmd == TUNSETIFF || cmd == TUNSETQUEUE || _IOC_TYPE(cmd) == SOCK_IOC_TYPE) {
2801 if (copy_from_user(&ifr, argp, ifreq_len))
2802 return -EFAULT;
2803 } else {
2804 memset(&ifr, 0, sizeof(ifr));
2805 }
2806 if (cmd == TUNGETFEATURES) {
2807 /* Currently this just means: "what IFF flags are valid?".
2808 * This is needed because we never checked for invalid flags on
2809 * TUNSETIFF.
2810 */
2811 return put_user(IFF_TUN | IFF_TAP | TUN_FEATURES,
2812 (unsigned int __user*)argp);
2813 } else if (cmd == TUNSETQUEUE)
2814 return tun_set_queue(file, &ifr);
2815
2816 ret = 0;
2817 rtnl_lock();
2818
2819 tun = tun_get(tfile);
2820 if (cmd == TUNSETIFF) {
2821 ret = -EEXIST;
2822 if (tun)
2823 goto unlock;
2824
2825 ifr.ifr_name[IFNAMSIZ-1] = '\0';
2826
2827 ret = tun_set_iff(sock_net(&tfile->sk), file, &ifr);
2828
2829 if (ret)
2830 goto unlock;
2831
2832 if (copy_to_user(argp, &ifr, ifreq_len))
2833 ret = -EFAULT;
2834 goto unlock;
2835 }
2836 if (cmd == TUNSETIFINDEX) {
2837 ret = -EPERM;
2838 if (tun)
2839 goto unlock;
2840
2841 ret = -EFAULT;
2842 if (copy_from_user(&ifindex, argp, sizeof(ifindex)))
2843 goto unlock;
2844
2845 ret = 0;
2846 tfile->ifindex = ifindex;
2847 goto unlock;
2848 }
2849
2850 ret = -EBADFD;
2851 if (!tun)
2852 goto unlock;
2853
2854 tun_debug(KERN_INFO, tun, "tun_chr_ioctl cmd %u\n", cmd);
2855
2856 ret = 0;
2857 switch (cmd) {
2858 case TUNGETIFF:
2859 tun_get_iff(current->nsproxy->net_ns, tun, &ifr);
2860
2861 if (tfile->detached)
2862 ifr.ifr_flags |= IFF_DETACH_QUEUE;
2863 if (!tfile->socket.sk->sk_filter)
2864 ifr.ifr_flags |= IFF_NOFILTER;
2865
2866 if (copy_to_user(argp, &ifr, ifreq_len))
2867 ret = -EFAULT;
2868 break;
2869
2870 case TUNSETNOCSUM:
2871 /* Disable/Enable checksum */
2872
2873 /* [unimplemented] */
2874 tun_debug(KERN_INFO, tun, "ignored: set checksum %s\n",
2875 arg ? "disabled" : "enabled");
2876 break;
2877
2878 case TUNSETPERSIST:
2879 /* Disable/Enable persist mode. Keep an extra reference to the
2880 * module to prevent the module being unprobed.
2881 */
2882 if (arg && !(tun->flags & IFF_PERSIST)) {
2883 tun->flags |= IFF_PERSIST;
2884 __module_get(THIS_MODULE);
2885 }
2886 if (!arg && (tun->flags & IFF_PERSIST)) {
2887 tun->flags &= ~IFF_PERSIST;
2888 module_put(THIS_MODULE);
2889 }
2890
2891 tun_debug(KERN_INFO, tun, "persist %s\n",
2892 arg ? "enabled" : "disabled");
2893 break;
2894
2895 case TUNSETOWNER:
2896 /* Set owner of the device */
2897 owner = make_kuid(current_user_ns(), arg);
2898 if (!uid_valid(owner)) {
2899 ret = -EINVAL;
2900 break;
2901 }
2902 tun->owner = owner;
2903 tun_debug(KERN_INFO, tun, "owner set to %u\n",
2904 from_kuid(&init_user_ns, tun->owner));
2905 break;
2906
2907 case TUNSETGROUP:
2908 /* Set group of the device */
2909 group = make_kgid(current_user_ns(), arg);
2910 if (!gid_valid(group)) {
2911 ret = -EINVAL;
2912 break;
2913 }
2914 tun->group = group;
2915 tun_debug(KERN_INFO, tun, "group set to %u\n",
2916 from_kgid(&init_user_ns, tun->group));
2917 break;
2918
2919 case TUNSETLINK:
2920 /* Only allow setting the type when the interface is down */
2921 if (tun->dev->flags & IFF_UP) {
2922 tun_debug(KERN_INFO, tun,
2923 "Linktype set failed because interface is up\n");
2924 ret = -EBUSY;
2925 } else {
2926 tun->dev->type = (int) arg;
2927 tun_debug(KERN_INFO, tun, "linktype set to %d\n",
2928 tun->dev->type);
2929 ret = 0;
2930 }
2931 break;
2932
2933 #ifdef TUN_DEBUG
2934 case TUNSETDEBUG:
2935 tun->debug = arg;
2936 break;
2937 #endif
2938 case TUNSETOFFLOAD:
2939 ret = set_offload(tun, arg);
2940 break;
2941
2942 case TUNSETTXFILTER:
2943 /* Can be set only for TAPs */
2944 ret = -EINVAL;
2945 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
2946 break;
2947 ret = update_filter(&tun->txflt, (void __user *)arg);
2948 break;
2949
2950 case SIOCGIFHWADDR:
2951 /* Get hw address */
2952 memcpy(ifr.ifr_hwaddr.sa_data, tun->dev->dev_addr, ETH_ALEN);
2953 ifr.ifr_hwaddr.sa_family = tun->dev->type;
2954 if (copy_to_user(argp, &ifr, ifreq_len))
2955 ret = -EFAULT;
2956 break;
2957
2958 case SIOCSIFHWADDR:
2959 /* Set hw address */
2960 tun_debug(KERN_DEBUG, tun, "set hw address: %pM\n",
2961 ifr.ifr_hwaddr.sa_data);
2962
2963 ret = dev_set_mac_address(tun->dev, &ifr.ifr_hwaddr);
2964 break;
2965
2966 case TUNGETSNDBUF:
2967 sndbuf = tfile->socket.sk->sk_sndbuf;
2968 if (copy_to_user(argp, &sndbuf, sizeof(sndbuf)))
2969 ret = -EFAULT;
2970 break;
2971
2972 case TUNSETSNDBUF:
2973 if (copy_from_user(&sndbuf, argp, sizeof(sndbuf))) {
2974 ret = -EFAULT;
2975 break;
2976 }
2977 if (sndbuf <= 0) {
2978 ret = -EINVAL;
2979 break;
2980 }
2981
2982 tun->sndbuf = sndbuf;
2983 tun_set_sndbuf(tun);
2984 break;
2985
2986 case TUNGETVNETHDRSZ:
2987 vnet_hdr_sz = tun->vnet_hdr_sz;
2988 if (copy_to_user(argp, &vnet_hdr_sz, sizeof(vnet_hdr_sz)))
2989 ret = -EFAULT;
2990 break;
2991
2992 case TUNSETVNETHDRSZ:
2993 if (copy_from_user(&vnet_hdr_sz, argp, sizeof(vnet_hdr_sz))) {
2994 ret = -EFAULT;
2995 break;
2996 }
2997 if (vnet_hdr_sz < (int)sizeof(struct virtio_net_hdr)) {
2998 ret = -EINVAL;
2999 break;
3000 }
3001
3002 tun->vnet_hdr_sz = vnet_hdr_sz;
3003 break;
3004
3005 case TUNGETVNETLE:
3006 le = !!(tun->flags & TUN_VNET_LE);
3007 if (put_user(le, (int __user *)argp))
3008 ret = -EFAULT;
3009 break;
3010
3011 case TUNSETVNETLE:
3012 if (get_user(le, (int __user *)argp)) {
3013 ret = -EFAULT;
3014 break;
3015 }
3016 if (le)
3017 tun->flags |= TUN_VNET_LE;
3018 else
3019 tun->flags &= ~TUN_VNET_LE;
3020 break;
3021
3022 case TUNGETVNETBE:
3023 ret = tun_get_vnet_be(tun, argp);
3024 break;
3025
3026 case TUNSETVNETBE:
3027 ret = tun_set_vnet_be(tun, argp);
3028 break;
3029
3030 case TUNATTACHFILTER:
3031 /* Can be set only for TAPs */
3032 ret = -EINVAL;
3033 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3034 break;
3035 ret = -EFAULT;
3036 if (copy_from_user(&tun->fprog, argp, sizeof(tun->fprog)))
3037 break;
3038
3039 ret = tun_attach_filter(tun);
3040 break;
3041
3042 case TUNDETACHFILTER:
3043 /* Can be set only for TAPs */
3044 ret = -EINVAL;
3045 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3046 break;
3047 ret = 0;
3048 tun_detach_filter(tun, tun->numqueues);
3049 break;
3050
3051 case TUNGETFILTER:
3052 ret = -EINVAL;
3053 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3054 break;
3055 ret = -EFAULT;
3056 if (copy_to_user(argp, &tun->fprog, sizeof(tun->fprog)))
3057 break;
3058 ret = 0;
3059 break;
3060
3061 case TUNSETSTEERINGEBPF:
3062 ret = tun_set_ebpf(tun, &tun->steering_prog, argp);
3063 break;
3064
3065 case TUNSETFILTEREBPF:
3066 ret = tun_set_ebpf(tun, &tun->filter_prog, argp);
3067 break;
3068
3069 default:
3070 ret = -EINVAL;
3071 break;
3072 }
3073
3074 unlock:
3075 rtnl_unlock();
3076 if (tun)
3077 tun_put(tun);
3078 return ret;
3079 }
3080
3081 static long tun_chr_ioctl(struct file *file,
3082 unsigned int cmd, unsigned long arg)
3083 {
3084 return __tun_chr_ioctl(file, cmd, arg, sizeof (struct ifreq));
3085 }
3086
3087 #ifdef CONFIG_COMPAT
3088 static long tun_chr_compat_ioctl(struct file *file,
3089 unsigned int cmd, unsigned long arg)
3090 {
3091 switch (cmd) {
3092 case TUNSETIFF:
3093 case TUNGETIFF:
3094 case TUNSETTXFILTER:
3095 case TUNGETSNDBUF:
3096 case TUNSETSNDBUF:
3097 case SIOCGIFHWADDR:
3098 case SIOCSIFHWADDR:
3099 arg = (unsigned long)compat_ptr(arg);
3100 break;
3101 default:
3102 arg = (compat_ulong_t)arg;
3103 break;
3104 }
3105
3106 /*
3107 * compat_ifreq is shorter than ifreq, so we must not access beyond
3108 * the end of that structure. All fields that are used in this
3109 * driver are compatible though, we don't need to convert the
3110 * contents.
3111 */
3112 return __tun_chr_ioctl(file, cmd, arg, sizeof(struct compat_ifreq));
3113 }
3114 #endif /* CONFIG_COMPAT */
3115
3116 static int tun_chr_fasync(int fd, struct file *file, int on)
3117 {
3118 struct tun_file *tfile = file->private_data;
3119 int ret;
3120
3121 if ((ret = fasync_helper(fd, file, on, &tfile->fasync)) < 0)
3122 goto out;
3123
3124 if (on) {
3125 __f_setown(file, task_pid(current), PIDTYPE_PID, 0);
3126 tfile->flags |= TUN_FASYNC;
3127 } else
3128 tfile->flags &= ~TUN_FASYNC;
3129 ret = 0;
3130 out:
3131 return ret;
3132 }
3133
3134 static int tun_chr_open(struct inode *inode, struct file * file)
3135 {
3136 struct net *net = current->nsproxy->net_ns;
3137 struct tun_file *tfile;
3138
3139 DBG1(KERN_INFO, "tunX: tun_chr_open\n");
3140
3141 tfile = (struct tun_file *)sk_alloc(net, AF_UNSPEC, GFP_KERNEL,
3142 &tun_proto, 0);
3143 if (!tfile)
3144 return -ENOMEM;
3145 RCU_INIT_POINTER(tfile->tun, NULL);
3146 tfile->flags = 0;
3147 tfile->ifindex = 0;
3148
3149 init_waitqueue_head(&tfile->wq.wait);
3150 RCU_INIT_POINTER(tfile->socket.wq, &tfile->wq);
3151
3152 tfile->socket.file = file;
3153 tfile->socket.ops = &tun_socket_ops;
3154
3155 sock_init_data(&tfile->socket, &tfile->sk);
3156
3157 tfile->sk.sk_write_space = tun_sock_write_space;
3158 tfile->sk.sk_sndbuf = INT_MAX;
3159
3160 file->private_data = tfile;
3161 INIT_LIST_HEAD(&tfile->next);
3162
3163 sock_set_flag(&tfile->sk, SOCK_ZEROCOPY);
3164
3165 memset(&tfile->tx_ring, 0, sizeof(tfile->tx_ring));
3166 tfile->xdp_pending_pkts = 0;
3167
3168 return 0;
3169 }
3170
3171 static int tun_chr_close(struct inode *inode, struct file *file)
3172 {
3173 struct tun_file *tfile = file->private_data;
3174
3175 tun_detach(tfile, true);
3176
3177 return 0;
3178 }
3179
3180 #ifdef CONFIG_PROC_FS
3181 static void tun_chr_show_fdinfo(struct seq_file *m, struct file *file)
3182 {
3183 struct tun_file *tfile = file->private_data;
3184 struct tun_struct *tun;
3185 struct ifreq ifr;
3186
3187 memset(&ifr, 0, sizeof(ifr));
3188
3189 rtnl_lock();
3190 tun = tun_get(tfile);
3191 if (tun)
3192 tun_get_iff(current->nsproxy->net_ns, tun, &ifr);
3193 rtnl_unlock();
3194
3195 if (tun)
3196 tun_put(tun);
3197
3198 seq_printf(m, "iff:\t%s\n", ifr.ifr_name);
3199 }
3200 #endif
3201
3202 static const struct file_operations tun_fops = {
3203 .owner = THIS_MODULE,
3204 .llseek = no_llseek,
3205 .read_iter = tun_chr_read_iter,
3206 .write_iter = tun_chr_write_iter,
3207 .poll = tun_chr_poll,
3208 .unlocked_ioctl = tun_chr_ioctl,
3209 #ifdef CONFIG_COMPAT
3210 .compat_ioctl = tun_chr_compat_ioctl,
3211 #endif
3212 .open = tun_chr_open,
3213 .release = tun_chr_close,
3214 .fasync = tun_chr_fasync,
3215 #ifdef CONFIG_PROC_FS
3216 .show_fdinfo = tun_chr_show_fdinfo,
3217 #endif
3218 };
3219
3220 static struct miscdevice tun_miscdev = {
3221 .minor = TUN_MINOR,
3222 .name = "tun",
3223 .nodename = "net/tun",
3224 .fops = &tun_fops,
3225 };
3226
3227 /* ethtool interface */
3228
3229 static int tun_get_link_ksettings(struct net_device *dev,
3230 struct ethtool_link_ksettings *cmd)
3231 {
3232 ethtool_link_ksettings_zero_link_mode(cmd, supported);
3233 ethtool_link_ksettings_zero_link_mode(cmd, advertising);
3234 cmd->base.speed = SPEED_10;
3235 cmd->base.duplex = DUPLEX_FULL;
3236 cmd->base.port = PORT_TP;
3237 cmd->base.phy_address = 0;
3238 cmd->base.autoneg = AUTONEG_DISABLE;
3239 return 0;
3240 }
3241
3242 static void tun_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
3243 {
3244 struct tun_struct *tun = netdev_priv(dev);
3245
3246 strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
3247 strlcpy(info->version, DRV_VERSION, sizeof(info->version));
3248
3249 switch (tun->flags & TUN_TYPE_MASK) {
3250 case IFF_TUN:
3251 strlcpy(info->bus_info, "tun", sizeof(info->bus_info));
3252 break;
3253 case IFF_TAP:
3254 strlcpy(info->bus_info, "tap", sizeof(info->bus_info));
3255 break;
3256 }
3257 }
3258
3259 static u32 tun_get_msglevel(struct net_device *dev)
3260 {
3261 #ifdef TUN_DEBUG
3262 struct tun_struct *tun = netdev_priv(dev);
3263 return tun->debug;
3264 #else
3265 return -EOPNOTSUPP;
3266 #endif
3267 }
3268
3269 static void tun_set_msglevel(struct net_device *dev, u32 value)
3270 {
3271 #ifdef TUN_DEBUG
3272 struct tun_struct *tun = netdev_priv(dev);
3273 tun->debug = value;
3274 #endif
3275 }
3276
3277 static int tun_get_coalesce(struct net_device *dev,
3278 struct ethtool_coalesce *ec)
3279 {
3280 struct tun_struct *tun = netdev_priv(dev);
3281
3282 ec->rx_max_coalesced_frames = tun->rx_batched;
3283
3284 return 0;
3285 }
3286
3287 static int tun_set_coalesce(struct net_device *dev,
3288 struct ethtool_coalesce *ec)
3289 {
3290 struct tun_struct *tun = netdev_priv(dev);
3291
3292 if (ec->rx_max_coalesced_frames > NAPI_POLL_WEIGHT)
3293 tun->rx_batched = NAPI_POLL_WEIGHT;
3294 else
3295 tun->rx_batched = ec->rx_max_coalesced_frames;
3296
3297 return 0;
3298 }
3299
3300 static const struct ethtool_ops tun_ethtool_ops = {
3301 .get_drvinfo = tun_get_drvinfo,
3302 .get_msglevel = tun_get_msglevel,
3303 .set_msglevel = tun_set_msglevel,
3304 .get_link = ethtool_op_get_link,
3305 .get_ts_info = ethtool_op_get_ts_info,
3306 .get_coalesce = tun_get_coalesce,
3307 .set_coalesce = tun_set_coalesce,
3308 .get_link_ksettings = tun_get_link_ksettings,
3309 };
3310
3311 static int tun_queue_resize(struct tun_struct *tun)
3312 {
3313 struct net_device *dev = tun->dev;
3314 struct tun_file *tfile;
3315 struct ptr_ring **rings;
3316 int n = tun->numqueues + tun->numdisabled;
3317 int ret, i;
3318
3319 rings = kmalloc_array(n, sizeof(*rings), GFP_KERNEL);
3320 if (!rings)
3321 return -ENOMEM;
3322
3323 for (i = 0; i < tun->numqueues; i++) {
3324 tfile = rtnl_dereference(tun->tfiles[i]);
3325 rings[i] = &tfile->tx_ring;
3326 }
3327 list_for_each_entry(tfile, &tun->disabled, next)
3328 rings[i++] = &tfile->tx_ring;
3329
3330 ret = ptr_ring_resize_multiple(rings, n,
3331 dev->tx_queue_len, GFP_KERNEL,
3332 tun_ptr_free);
3333
3334 kfree(rings);
3335 return ret;
3336 }
3337
3338 static int tun_device_event(struct notifier_block *unused,
3339 unsigned long event, void *ptr)
3340 {
3341 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
3342 struct tun_struct *tun = netdev_priv(dev);
3343
3344 if (dev->rtnl_link_ops != &tun_link_ops)
3345 return NOTIFY_DONE;
3346
3347 switch (event) {
3348 case NETDEV_CHANGE_TX_QUEUE_LEN:
3349 if (tun_queue_resize(tun))
3350 return NOTIFY_BAD;
3351 break;
3352 default:
3353 break;
3354 }
3355
3356 return NOTIFY_DONE;
3357 }
3358
3359 static struct notifier_block tun_notifier_block __read_mostly = {
3360 .notifier_call = tun_device_event,
3361 };
3362
3363 static int __init tun_init(void)
3364 {
3365 int ret = 0;
3366
3367 pr_info("%s, %s\n", DRV_DESCRIPTION, DRV_VERSION);
3368
3369 ret = rtnl_link_register(&tun_link_ops);
3370 if (ret) {
3371 pr_err("Can't register link_ops\n");
3372 goto err_linkops;
3373 }
3374
3375 ret = misc_register(&tun_miscdev);
3376 if (ret) {
3377 pr_err("Can't register misc device %d\n", TUN_MINOR);
3378 goto err_misc;
3379 }
3380
3381 ret = register_netdevice_notifier(&tun_notifier_block);
3382 if (ret) {
3383 pr_err("Can't register netdevice notifier\n");
3384 goto err_notifier;
3385 }
3386
3387 return 0;
3388
3389 err_notifier:
3390 misc_deregister(&tun_miscdev);
3391 err_misc:
3392 rtnl_link_unregister(&tun_link_ops);
3393 err_linkops:
3394 return ret;
3395 }
3396
3397 static void tun_cleanup(void)
3398 {
3399 misc_deregister(&tun_miscdev);
3400 rtnl_link_unregister(&tun_link_ops);
3401 unregister_netdevice_notifier(&tun_notifier_block);
3402 }
3403
3404 /* Get an underlying socket object from tun file. Returns error unless file is
3405 * attached to a device. The returned object works like a packet socket, it
3406 * can be used for sock_sendmsg/sock_recvmsg. The caller is responsible for
3407 * holding a reference to the file for as long as the socket is in use. */
3408 struct socket *tun_get_socket(struct file *file)
3409 {
3410 struct tun_file *tfile;
3411 if (file->f_op != &tun_fops)
3412 return ERR_PTR(-EINVAL);
3413 tfile = file->private_data;
3414 if (!tfile)
3415 return ERR_PTR(-EBADFD);
3416 return &tfile->socket;
3417 }
3418 EXPORT_SYMBOL_GPL(tun_get_socket);
3419
3420 struct ptr_ring *tun_get_tx_ring(struct file *file)
3421 {
3422 struct tun_file *tfile;
3423
3424 if (file->f_op != &tun_fops)
3425 return ERR_PTR(-EINVAL);
3426 tfile = file->private_data;
3427 if (!tfile)
3428 return ERR_PTR(-EBADFD);
3429 return &tfile->tx_ring;
3430 }
3431 EXPORT_SYMBOL_GPL(tun_get_tx_ring);
3432
3433 module_init(tun_init);
3434 module_exit(tun_cleanup);
3435 MODULE_DESCRIPTION(DRV_DESCRIPTION);
3436 MODULE_AUTHOR(DRV_COPYRIGHT);
3437 MODULE_LICENSE("GPL");
3438 MODULE_ALIAS_MISCDEV(TUN_MINOR);
3439 MODULE_ALIAS("devname:net/tun");
CRIS linux kernel tree