Loading...
1// SPDX-License-Identifier: GPL-2.0-or-later
2/*
3 * TUN - Universal TUN/TAP device driver.
4 * Copyright (C) 1999-2002 Maxim Krasnyansky <maxk@qualcomm.com>
5 *
6 * $Id: tun.c,v 1.15 2002/03/01 02:44:24 maxk Exp $
7 */
8
9/*
10 * Changes:
11 *
12 * Mike Kershaw <dragorn@kismetwireless.net> 2005/08/14
13 * Add TUNSETLINK ioctl to set the link encapsulation
14 *
15 * Mark Smith <markzzzsmith@yahoo.com.au>
16 * Use eth_random_addr() for tap MAC address.
17 *
18 * Harald Roelle <harald.roelle@ifi.lmu.de> 2004/04/20
19 * Fixes in packet dropping, queue length setting and queue wakeup.
20 * Increased default tx queue length.
21 * Added ethtool API.
22 * Minor cleanups
23 *
24 * Daniel Podlejski <underley@underley.eu.org>
25 * Modifications for 2.3.99-pre5 kernel.
26 */
27
28#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
29
30#define DRV_NAME "tun"
31#define DRV_VERSION "1.6"
32#define DRV_DESCRIPTION "Universal TUN/TAP device driver"
33#define DRV_COPYRIGHT "(C) 1999-2004 Max Krasnyansky <maxk@qualcomm.com>"
34
35#include <linux/module.h>
36#include <linux/errno.h>
37#include <linux/kernel.h>
38#include <linux/sched/signal.h>
39#include <linux/major.h>
40#include <linux/slab.h>
41#include <linux/poll.h>
42#include <linux/fcntl.h>
43#include <linux/init.h>
44#include <linux/skbuff.h>
45#include <linux/netdevice.h>
46#include <linux/etherdevice.h>
47#include <linux/miscdevice.h>
48#include <linux/ethtool.h>
49#include <linux/rtnetlink.h>
50#include <linux/compat.h>
51#include <linux/if.h>
52#include <linux/if_arp.h>
53#include <linux/if_ether.h>
54#include <linux/if_tun.h>
55#include <linux/if_vlan.h>
56#include <linux/crc32.h>
57#include <linux/math.h>
58#include <linux/nsproxy.h>
59#include <linux/virtio_net.h>
60#include <linux/rcupdate.h>
61#include <net/net_namespace.h>
62#include <net/netns/generic.h>
63#include <net/rtnetlink.h>
64#include <net/sock.h>
65#include <net/xdp.h>
66#include <net/ip_tunnels.h>
67#include <linux/seq_file.h>
68#include <linux/uio.h>
69#include <linux/skb_array.h>
70#include <linux/bpf.h>
71#include <linux/bpf_trace.h>
72#include <linux/mutex.h>
73#include <linux/ieee802154.h>
74#include <uapi/linux/if_ltalk.h>
75#include <uapi/linux/if_fddi.h>
76#include <uapi/linux/if_hippi.h>
77#include <uapi/linux/if_fc.h>
78#include <net/ax25.h>
79#include <net/rose.h>
80#include <net/6lowpan.h>
81#include <net/rps.h>
82
83#include <linux/uaccess.h>
84#include <linux/proc_fs.h>
85
86static void tun_default_link_ksettings(struct net_device *dev,
87 struct ethtool_link_ksettings *cmd);
88
89#define TUN_RX_PAD (NET_IP_ALIGN + NET_SKB_PAD)
90
91/* TUN device flags */
92
93/* IFF_ATTACH_QUEUE is never stored in device flags,
94 * overload it to mean fasync when stored there.
95 */
96#define TUN_FASYNC IFF_ATTACH_QUEUE
97/* High bits in flags field are unused. */
98#define TUN_VNET_LE 0x80000000
99#define TUN_VNET_BE 0x40000000
100
101#define TUN_FEATURES (IFF_NO_PI | IFF_ONE_QUEUE | IFF_VNET_HDR | \
102 IFF_MULTI_QUEUE | IFF_NAPI | IFF_NAPI_FRAGS)
103
104#define GOODCOPY_LEN 128
105
106#define FLT_EXACT_COUNT 8
107struct tap_filter {
108 unsigned int count; /* Number of addrs. Zero means disabled */
109 u32 mask[2]; /* Mask of the hashed addrs */
110 unsigned char addr[FLT_EXACT_COUNT][ETH_ALEN];
111};
112
113/* MAX_TAP_QUEUES 256 is chosen to allow rx/tx queues to be equal
114 * to max number of VCPUs in guest. */
115#define MAX_TAP_QUEUES 256
116#define MAX_TAP_FLOWS 4096
117
118#define TUN_FLOW_EXPIRE (3 * HZ)
119
120/* A tun_file connects an open character device to a tuntap netdevice. It
121 * also contains all socket related structures (except sock_fprog and tap_filter)
122 * to serve as one transmit queue for tuntap device. The sock_fprog and
123 * tap_filter were kept in tun_struct since they were used for filtering for the
124 * netdevice not for a specific queue (at least I didn't see the requirement for
125 * this).
126 *
127 * RCU usage:
128 * The tun_file and tun_struct are loosely coupled, the pointer from one to the
129 * other can only be read while rcu_read_lock or rtnl_lock is held.
130 */
131struct tun_file {
132 struct sock sk;
133 struct socket socket;
134 struct tun_struct __rcu *tun;
135 struct fasync_struct *fasync;
136 /* only used for fasnyc */
137 unsigned int flags;
138 union {
139 u16 queue_index;
140 unsigned int ifindex;
141 };
142 struct napi_struct napi;
143 bool napi_enabled;
144 bool napi_frags_enabled;
145 struct mutex napi_mutex; /* Protects access to the above napi */
146 struct list_head next;
147 struct tun_struct *detached;
148 struct ptr_ring tx_ring;
149 struct xdp_rxq_info xdp_rxq;
150};
151
152struct tun_page {
153 struct page *page;
154 int count;
155};
156
157struct tun_flow_entry {
158 struct hlist_node hash_link;
159 struct rcu_head rcu;
160 struct tun_struct *tun;
161
162 u32 rxhash;
163 u32 rps_rxhash;
164 int queue_index;
165 unsigned long updated ____cacheline_aligned_in_smp;
166};
167
168#define TUN_NUM_FLOW_ENTRIES 1024
169#define TUN_MASK_FLOW_ENTRIES (TUN_NUM_FLOW_ENTRIES - 1)
170
171struct tun_prog {
172 struct rcu_head rcu;
173 struct bpf_prog *prog;
174};
175
176/* Since the socket were moved to tun_file, to preserve the behavior of persist
177 * device, socket filter, sndbuf and vnet header size were restore when the
178 * file were attached to a persist device.
179 */
180struct tun_struct {
181 struct tun_file __rcu *tfiles[MAX_TAP_QUEUES];
182 unsigned int numqueues;
183 unsigned int flags;
184 kuid_t owner;
185 kgid_t group;
186
187 struct net_device *dev;
188 netdev_features_t set_features;
189#define TUN_USER_FEATURES (NETIF_F_HW_CSUM|NETIF_F_TSO_ECN|NETIF_F_TSO| \
190 NETIF_F_TSO6 | NETIF_F_GSO_UDP_L4)
191
192 int align;
193 int vnet_hdr_sz;
194 int sndbuf;
195 struct tap_filter txflt;
196 struct sock_fprog fprog;
197 /* protected by rtnl lock */
198 bool filter_attached;
199 u32 msg_enable;
200 spinlock_t lock;
201 struct hlist_head flows[TUN_NUM_FLOW_ENTRIES];
202 struct timer_list flow_gc_timer;
203 unsigned long ageing_time;
204 unsigned int numdisabled;
205 struct list_head disabled;
206 void *security;
207 u32 flow_count;
208 u32 rx_batched;
209 atomic_long_t rx_frame_errors;
210 struct bpf_prog __rcu *xdp_prog;
211 struct tun_prog __rcu *steering_prog;
212 struct tun_prog __rcu *filter_prog;
213 struct ethtool_link_ksettings link_ksettings;
214 /* init args */
215 struct file *file;
216 struct ifreq *ifr;
217};
218
219struct veth {
220 __be16 h_vlan_proto;
221 __be16 h_vlan_TCI;
222};
223
224static void tun_flow_init(struct tun_struct *tun);
225static void tun_flow_uninit(struct tun_struct *tun);
226
227static int tun_napi_receive(struct napi_struct *napi, int budget)
228{
229 struct tun_file *tfile = container_of(napi, struct tun_file, napi);
230 struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
231 struct sk_buff_head process_queue;
232 struct sk_buff *skb;
233 int received = 0;
234
235 __skb_queue_head_init(&process_queue);
236
237 spin_lock(&queue->lock);
238 skb_queue_splice_tail_init(queue, &process_queue);
239 spin_unlock(&queue->lock);
240
241 while (received < budget && (skb = __skb_dequeue(&process_queue))) {
242 napi_gro_receive(napi, skb);
243 ++received;
244 }
245
246 if (!skb_queue_empty(&process_queue)) {
247 spin_lock(&queue->lock);
248 skb_queue_splice(&process_queue, queue);
249 spin_unlock(&queue->lock);
250 }
251
252 return received;
253}
254
255static int tun_napi_poll(struct napi_struct *napi, int budget)
256{
257 unsigned int received;
258
259 received = tun_napi_receive(napi, budget);
260
261 if (received < budget)
262 napi_complete_done(napi, received);
263
264 return received;
265}
266
267static void tun_napi_init(struct tun_struct *tun, struct tun_file *tfile,
268 bool napi_en, bool napi_frags)
269{
270 tfile->napi_enabled = napi_en;
271 tfile->napi_frags_enabled = napi_en && napi_frags;
272 if (napi_en) {
273 netif_napi_add_tx(tun->dev, &tfile->napi, tun_napi_poll);
274 napi_enable(&tfile->napi);
275 }
276}
277
278static void tun_napi_enable(struct tun_file *tfile)
279{
280 if (tfile->napi_enabled)
281 napi_enable(&tfile->napi);
282}
283
284static void tun_napi_disable(struct tun_file *tfile)
285{
286 if (tfile->napi_enabled)
287 napi_disable(&tfile->napi);
288}
289
290static void tun_napi_del(struct tun_file *tfile)
291{
292 if (tfile->napi_enabled)
293 netif_napi_del(&tfile->napi);
294}
295
296static bool tun_napi_frags_enabled(const struct tun_file *tfile)
297{
298 return tfile->napi_frags_enabled;
299}
300
301#ifdef CONFIG_TUN_VNET_CROSS_LE
302static inline bool tun_legacy_is_little_endian(struct tun_struct *tun)
303{
304 return tun->flags & TUN_VNET_BE ? false :
305 virtio_legacy_is_little_endian();
306}
307
308static long tun_get_vnet_be(struct tun_struct *tun, int __user *argp)
309{
310 int be = !!(tun->flags & TUN_VNET_BE);
311
312 if (put_user(be, argp))
313 return -EFAULT;
314
315 return 0;
316}
317
318static long tun_set_vnet_be(struct tun_struct *tun, int __user *argp)
319{
320 int be;
321
322 if (get_user(be, argp))
323 return -EFAULT;
324
325 if (be)
326 tun->flags |= TUN_VNET_BE;
327 else
328 tun->flags &= ~TUN_VNET_BE;
329
330 return 0;
331}
332#else
333static inline bool tun_legacy_is_little_endian(struct tun_struct *tun)
334{
335 return virtio_legacy_is_little_endian();
336}
337
338static long tun_get_vnet_be(struct tun_struct *tun, int __user *argp)
339{
340 return -EINVAL;
341}
342
343static long tun_set_vnet_be(struct tun_struct *tun, int __user *argp)
344{
345 return -EINVAL;
346}
347#endif /* CONFIG_TUN_VNET_CROSS_LE */
348
349static inline bool tun_is_little_endian(struct tun_struct *tun)
350{
351 return tun->flags & TUN_VNET_LE ||
352 tun_legacy_is_little_endian(tun);
353}
354
355static inline u16 tun16_to_cpu(struct tun_struct *tun, __virtio16 val)
356{
357 return __virtio16_to_cpu(tun_is_little_endian(tun), val);
358}
359
360static inline __virtio16 cpu_to_tun16(struct tun_struct *tun, u16 val)
361{
362 return __cpu_to_virtio16(tun_is_little_endian(tun), val);
363}
364
365static inline u32 tun_hashfn(u32 rxhash)
366{
367 return rxhash & TUN_MASK_FLOW_ENTRIES;
368}
369
370static struct tun_flow_entry *tun_flow_find(struct hlist_head *head, u32 rxhash)
371{
372 struct tun_flow_entry *e;
373
374 hlist_for_each_entry_rcu(e, head, hash_link) {
375 if (e->rxhash == rxhash)
376 return e;
377 }
378 return NULL;
379}
380
381static struct tun_flow_entry *tun_flow_create(struct tun_struct *tun,
382 struct hlist_head *head,
383 u32 rxhash, u16 queue_index)
384{
385 struct tun_flow_entry *e = kmalloc(sizeof(*e), GFP_ATOMIC);
386
387 if (e) {
388 netif_info(tun, tx_queued, tun->dev,
389 "create flow: hash %u index %u\n",
390 rxhash, queue_index);
391 e->updated = jiffies;
392 e->rxhash = rxhash;
393 e->rps_rxhash = 0;
394 e->queue_index = queue_index;
395 e->tun = tun;
396 hlist_add_head_rcu(&e->hash_link, head);
397 ++tun->flow_count;
398 }
399 return e;
400}
401
402static void tun_flow_delete(struct tun_struct *tun, struct tun_flow_entry *e)
403{
404 netif_info(tun, tx_queued, tun->dev, "delete flow: hash %u index %u\n",
405 e->rxhash, e->queue_index);
406 hlist_del_rcu(&e->hash_link);
407 kfree_rcu(e, rcu);
408 --tun->flow_count;
409}
410
411static void tun_flow_flush(struct tun_struct *tun)
412{
413 int i;
414
415 spin_lock_bh(&tun->lock);
416 for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
417 struct tun_flow_entry *e;
418 struct hlist_node *n;
419
420 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link)
421 tun_flow_delete(tun, e);
422 }
423 spin_unlock_bh(&tun->lock);
424}
425
426static void tun_flow_delete_by_queue(struct tun_struct *tun, u16 queue_index)
427{
428 int i;
429
430 spin_lock_bh(&tun->lock);
431 for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
432 struct tun_flow_entry *e;
433 struct hlist_node *n;
434
435 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
436 if (e->queue_index == queue_index)
437 tun_flow_delete(tun, e);
438 }
439 }
440 spin_unlock_bh(&tun->lock);
441}
442
443static void tun_flow_cleanup(struct timer_list *t)
444{
445 struct tun_struct *tun = from_timer(tun, t, flow_gc_timer);
446 unsigned long delay = tun->ageing_time;
447 unsigned long next_timer = jiffies + delay;
448 unsigned long count = 0;
449 int i;
450
451 spin_lock(&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 unsigned long this_timer;
458
459 this_timer = e->updated + delay;
460 if (time_before_eq(this_timer, jiffies)) {
461 tun_flow_delete(tun, e);
462 continue;
463 }
464 count++;
465 if (time_before(this_timer, next_timer))
466 next_timer = this_timer;
467 }
468 }
469
470 if (count)
471 mod_timer(&tun->flow_gc_timer, round_jiffies_up(next_timer));
472 spin_unlock(&tun->lock);
473}
474
475static void tun_flow_update(struct tun_struct *tun, u32 rxhash,
476 struct tun_file *tfile)
477{
478 struct hlist_head *head;
479 struct tun_flow_entry *e;
480 unsigned long delay = tun->ageing_time;
481 u16 queue_index = tfile->queue_index;
482
483 head = &tun->flows[tun_hashfn(rxhash)];
484
485 rcu_read_lock();
486
487 e = tun_flow_find(head, rxhash);
488 if (likely(e)) {
489 /* TODO: keep queueing to old queue until it's empty? */
490 if (READ_ONCE(e->queue_index) != queue_index)
491 WRITE_ONCE(e->queue_index, queue_index);
492 if (e->updated != jiffies)
493 e->updated = jiffies;
494 sock_rps_record_flow_hash(e->rps_rxhash);
495 } else {
496 spin_lock_bh(&tun->lock);
497 if (!tun_flow_find(head, rxhash) &&
498 tun->flow_count < MAX_TAP_FLOWS)
499 tun_flow_create(tun, head, rxhash, queue_index);
500
501 if (!timer_pending(&tun->flow_gc_timer))
502 mod_timer(&tun->flow_gc_timer,
503 round_jiffies_up(jiffies + delay));
504 spin_unlock_bh(&tun->lock);
505 }
506
507 rcu_read_unlock();
508}
509
510/* Save the hash received in the stack receive path and update the
511 * flow_hash table accordingly.
512 */
513static inline void tun_flow_save_rps_rxhash(struct tun_flow_entry *e, u32 hash)
514{
515 if (unlikely(e->rps_rxhash != hash))
516 e->rps_rxhash = hash;
517}
518
519/* We try to identify a flow through its rxhash. The reason that
520 * we do not check rxq no. is because some cards(e.g 82599), chooses
521 * the rxq based on the txq where the last packet of the flow comes. As
522 * the userspace application move between processors, we may get a
523 * different rxq no. here.
524 */
525static u16 tun_automq_select_queue(struct tun_struct *tun, struct sk_buff *skb)
526{
527 struct tun_flow_entry *e;
528 u32 txq, numqueues;
529
530 numqueues = READ_ONCE(tun->numqueues);
531
532 txq = __skb_get_hash_symmetric(skb);
533 e = tun_flow_find(&tun->flows[tun_hashfn(txq)], txq);
534 if (e) {
535 tun_flow_save_rps_rxhash(e, txq);
536 txq = e->queue_index;
537 } else {
538 txq = reciprocal_scale(txq, numqueues);
539 }
540
541 return txq;
542}
543
544static u16 tun_ebpf_select_queue(struct tun_struct *tun, struct sk_buff *skb)
545{
546 struct tun_prog *prog;
547 u32 numqueues;
548 u16 ret = 0;
549
550 numqueues = READ_ONCE(tun->numqueues);
551 if (!numqueues)
552 return 0;
553
554 prog = rcu_dereference(tun->steering_prog);
555 if (prog)
556 ret = bpf_prog_run_clear_cb(prog->prog, skb);
557
558 return ret % numqueues;
559}
560
561static u16 tun_select_queue(struct net_device *dev, struct sk_buff *skb,
562 struct net_device *sb_dev)
563{
564 struct tun_struct *tun = netdev_priv(dev);
565 u16 ret;
566
567 rcu_read_lock();
568 if (rcu_dereference(tun->steering_prog))
569 ret = tun_ebpf_select_queue(tun, skb);
570 else
571 ret = tun_automq_select_queue(tun, skb);
572 rcu_read_unlock();
573
574 return ret;
575}
576
577static inline bool tun_not_capable(struct tun_struct *tun)
578{
579 const struct cred *cred = current_cred();
580 struct net *net = dev_net(tun->dev);
581
582 return ((uid_valid(tun->owner) && !uid_eq(cred->euid, tun->owner)) ||
583 (gid_valid(tun->group) && !in_egroup_p(tun->group))) &&
584 !ns_capable(net->user_ns, CAP_NET_ADMIN);
585}
586
587static void tun_set_real_num_queues(struct tun_struct *tun)
588{
589 netif_set_real_num_tx_queues(tun->dev, tun->numqueues);
590 netif_set_real_num_rx_queues(tun->dev, tun->numqueues);
591}
592
593static void tun_disable_queue(struct tun_struct *tun, struct tun_file *tfile)
594{
595 tfile->detached = tun;
596 list_add_tail(&tfile->next, &tun->disabled);
597 ++tun->numdisabled;
598}
599
600static struct tun_struct *tun_enable_queue(struct tun_file *tfile)
601{
602 struct tun_struct *tun = tfile->detached;
603
604 tfile->detached = NULL;
605 list_del_init(&tfile->next);
606 --tun->numdisabled;
607 return tun;
608}
609
610void tun_ptr_free(void *ptr)
611{
612 if (!ptr)
613 return;
614 if (tun_is_xdp_frame(ptr)) {
615 struct xdp_frame *xdpf = tun_ptr_to_xdp(ptr);
616
617 xdp_return_frame(xdpf);
618 } else {
619 __skb_array_destroy_skb(ptr);
620 }
621}
622EXPORT_SYMBOL_GPL(tun_ptr_free);
623
624static void tun_queue_purge(struct tun_file *tfile)
625{
626 void *ptr;
627
628 while ((ptr = ptr_ring_consume(&tfile->tx_ring)) != NULL)
629 tun_ptr_free(ptr);
630
631 skb_queue_purge(&tfile->sk.sk_write_queue);
632 skb_queue_purge(&tfile->sk.sk_error_queue);
633}
634
635static void __tun_detach(struct tun_file *tfile, bool clean)
636{
637 struct tun_file *ntfile;
638 struct tun_struct *tun;
639
640 tun = rtnl_dereference(tfile->tun);
641
642 if (tun && clean) {
643 if (!tfile->detached)
644 tun_napi_disable(tfile);
645 tun_napi_del(tfile);
646 }
647
648 if (tun && !tfile->detached) {
649 u16 index = tfile->queue_index;
650 BUG_ON(index >= tun->numqueues);
651
652 rcu_assign_pointer(tun->tfiles[index],
653 tun->tfiles[tun->numqueues - 1]);
654 ntfile = rtnl_dereference(tun->tfiles[index]);
655 ntfile->queue_index = index;
656 ntfile->xdp_rxq.queue_index = index;
657 rcu_assign_pointer(tun->tfiles[tun->numqueues - 1],
658 NULL);
659
660 --tun->numqueues;
661 if (clean) {
662 RCU_INIT_POINTER(tfile->tun, NULL);
663 sock_put(&tfile->sk);
664 } else {
665 tun_disable_queue(tun, tfile);
666 tun_napi_disable(tfile);
667 }
668
669 synchronize_net();
670 tun_flow_delete_by_queue(tun, tun->numqueues + 1);
671 /* Drop read queue */
672 tun_queue_purge(tfile);
673 tun_set_real_num_queues(tun);
674 } else if (tfile->detached && clean) {
675 tun = tun_enable_queue(tfile);
676 sock_put(&tfile->sk);
677 }
678
679 if (clean) {
680 if (tun && tun->numqueues == 0 && tun->numdisabled == 0) {
681 netif_carrier_off(tun->dev);
682
683 if (!(tun->flags & IFF_PERSIST) &&
684 tun->dev->reg_state == NETREG_REGISTERED)
685 unregister_netdevice(tun->dev);
686 }
687 if (tun)
688 xdp_rxq_info_unreg(&tfile->xdp_rxq);
689 ptr_ring_cleanup(&tfile->tx_ring, tun_ptr_free);
690 }
691}
692
693static void tun_detach(struct tun_file *tfile, bool clean)
694{
695 struct tun_struct *tun;
696 struct net_device *dev;
697
698 rtnl_lock();
699 tun = rtnl_dereference(tfile->tun);
700 dev = tun ? tun->dev : NULL;
701 __tun_detach(tfile, clean);
702 if (dev)
703 netdev_state_change(dev);
704 rtnl_unlock();
705
706 if (clean)
707 sock_put(&tfile->sk);
708}
709
710static void tun_detach_all(struct net_device *dev)
711{
712 struct tun_struct *tun = netdev_priv(dev);
713 struct tun_file *tfile, *tmp;
714 int i, n = tun->numqueues;
715
716 for (i = 0; i < n; i++) {
717 tfile = rtnl_dereference(tun->tfiles[i]);
718 BUG_ON(!tfile);
719 tun_napi_disable(tfile);
720 tfile->socket.sk->sk_shutdown = RCV_SHUTDOWN;
721 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
722 RCU_INIT_POINTER(tfile->tun, NULL);
723 --tun->numqueues;
724 }
725 list_for_each_entry(tfile, &tun->disabled, next) {
726 tfile->socket.sk->sk_shutdown = RCV_SHUTDOWN;
727 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
728 RCU_INIT_POINTER(tfile->tun, NULL);
729 }
730 BUG_ON(tun->numqueues != 0);
731
732 synchronize_net();
733 for (i = 0; i < n; i++) {
734 tfile = rtnl_dereference(tun->tfiles[i]);
735 tun_napi_del(tfile);
736 /* Drop read queue */
737 tun_queue_purge(tfile);
738 xdp_rxq_info_unreg(&tfile->xdp_rxq);
739 sock_put(&tfile->sk);
740 }
741 list_for_each_entry_safe(tfile, tmp, &tun->disabled, next) {
742 tun_napi_del(tfile);
743 tun_enable_queue(tfile);
744 tun_queue_purge(tfile);
745 xdp_rxq_info_unreg(&tfile->xdp_rxq);
746 sock_put(&tfile->sk);
747 }
748 BUG_ON(tun->numdisabled != 0);
749
750 if (tun->flags & IFF_PERSIST)
751 module_put(THIS_MODULE);
752}
753
754static int tun_attach(struct tun_struct *tun, struct file *file,
755 bool skip_filter, bool napi, bool napi_frags,
756 bool publish_tun)
757{
758 struct tun_file *tfile = file->private_data;
759 struct net_device *dev = tun->dev;
760 int err;
761
762 err = security_tun_dev_attach(tfile->socket.sk, tun->security);
763 if (err < 0)
764 goto out;
765
766 err = -EINVAL;
767 if (rtnl_dereference(tfile->tun) && !tfile->detached)
768 goto out;
769
770 err = -EBUSY;
771 if (!(tun->flags & IFF_MULTI_QUEUE) && tun->numqueues == 1)
772 goto out;
773
774 err = -E2BIG;
775 if (!tfile->detached &&
776 tun->numqueues + tun->numdisabled == MAX_TAP_QUEUES)
777 goto out;
778
779 err = 0;
780
781 /* Re-attach the filter to persist device */
782 if (!skip_filter && (tun->filter_attached == true)) {
783 lock_sock(tfile->socket.sk);
784 err = sk_attach_filter(&tun->fprog, tfile->socket.sk);
785 release_sock(tfile->socket.sk);
786 if (!err)
787 goto out;
788 }
789
790 if (!tfile->detached &&
791 ptr_ring_resize(&tfile->tx_ring, dev->tx_queue_len,
792 GFP_KERNEL, tun_ptr_free)) {
793 err = -ENOMEM;
794 goto out;
795 }
796
797 tfile->queue_index = tun->numqueues;
798 tfile->socket.sk->sk_shutdown &= ~RCV_SHUTDOWN;
799
800 if (tfile->detached) {
801 /* Re-attach detached tfile, updating XDP queue_index */
802 WARN_ON(!xdp_rxq_info_is_reg(&tfile->xdp_rxq));
803
804 if (tfile->xdp_rxq.queue_index != tfile->queue_index)
805 tfile->xdp_rxq.queue_index = tfile->queue_index;
806 } else {
807 /* Setup XDP RX-queue info, for new tfile getting attached */
808 err = xdp_rxq_info_reg(&tfile->xdp_rxq,
809 tun->dev, tfile->queue_index, 0);
810 if (err < 0)
811 goto out;
812 err = xdp_rxq_info_reg_mem_model(&tfile->xdp_rxq,
813 MEM_TYPE_PAGE_SHARED, NULL);
814 if (err < 0) {
815 xdp_rxq_info_unreg(&tfile->xdp_rxq);
816 goto out;
817 }
818 err = 0;
819 }
820
821 if (tfile->detached) {
822 tun_enable_queue(tfile);
823 tun_napi_enable(tfile);
824 } else {
825 sock_hold(&tfile->sk);
826 tun_napi_init(tun, tfile, napi, napi_frags);
827 }
828
829 if (rtnl_dereference(tun->xdp_prog))
830 sock_set_flag(&tfile->sk, SOCK_XDP);
831
832 /* device is allowed to go away first, so no need to hold extra
833 * refcnt.
834 */
835
836 /* Publish tfile->tun and tun->tfiles only after we've fully
837 * initialized tfile; otherwise we risk using half-initialized
838 * object.
839 */
840 if (publish_tun)
841 rcu_assign_pointer(tfile->tun, tun);
842 rcu_assign_pointer(tun->tfiles[tun->numqueues], tfile);
843 tun->numqueues++;
844 tun_set_real_num_queues(tun);
845out:
846 return err;
847}
848
849static struct tun_struct *tun_get(struct tun_file *tfile)
850{
851 struct tun_struct *tun;
852
853 rcu_read_lock();
854 tun = rcu_dereference(tfile->tun);
855 if (tun)
856 dev_hold(tun->dev);
857 rcu_read_unlock();
858
859 return tun;
860}
861
862static void tun_put(struct tun_struct *tun)
863{
864 dev_put(tun->dev);
865}
866
867/* TAP filtering */
868static void addr_hash_set(u32 *mask, const u8 *addr)
869{
870 int n = ether_crc(ETH_ALEN, addr) >> 26;
871 mask[n >> 5] |= (1 << (n & 31));
872}
873
874static unsigned int addr_hash_test(const u32 *mask, const u8 *addr)
875{
876 int n = ether_crc(ETH_ALEN, addr) >> 26;
877 return mask[n >> 5] & (1 << (n & 31));
878}
879
880static int update_filter(struct tap_filter *filter, void __user *arg)
881{
882 struct { u8 u[ETH_ALEN]; } *addr;
883 struct tun_filter uf;
884 int err, alen, n, nexact;
885
886 if (copy_from_user(&uf, arg, sizeof(uf)))
887 return -EFAULT;
888
889 if (!uf.count) {
890 /* Disabled */
891 filter->count = 0;
892 return 0;
893 }
894
895 alen = ETH_ALEN * uf.count;
896 addr = memdup_user(arg + sizeof(uf), alen);
897 if (IS_ERR(addr))
898 return PTR_ERR(addr);
899
900 /* The filter is updated without holding any locks. Which is
901 * perfectly safe. We disable it first and in the worst
902 * case we'll accept a few undesired packets. */
903 filter->count = 0;
904 wmb();
905
906 /* Use first set of addresses as an exact filter */
907 for (n = 0; n < uf.count && n < FLT_EXACT_COUNT; n++)
908 memcpy(filter->addr[n], addr[n].u, ETH_ALEN);
909
910 nexact = n;
911
912 /* Remaining multicast addresses are hashed,
913 * unicast will leave the filter disabled. */
914 memset(filter->mask, 0, sizeof(filter->mask));
915 for (; n < uf.count; n++) {
916 if (!is_multicast_ether_addr(addr[n].u)) {
917 err = 0; /* no filter */
918 goto free_addr;
919 }
920 addr_hash_set(filter->mask, addr[n].u);
921 }
922
923 /* For ALLMULTI just set the mask to all ones.
924 * This overrides the mask populated above. */
925 if ((uf.flags & TUN_FLT_ALLMULTI))
926 memset(filter->mask, ~0, sizeof(filter->mask));
927
928 /* Now enable the filter */
929 wmb();
930 filter->count = nexact;
931
932 /* Return the number of exact filters */
933 err = nexact;
934free_addr:
935 kfree(addr);
936 return err;
937}
938
939/* Returns: 0 - drop, !=0 - accept */
940static int run_filter(struct tap_filter *filter, const struct sk_buff *skb)
941{
942 /* Cannot use eth_hdr(skb) here because skb_mac_hdr() is incorrect
943 * at this point. */
944 struct ethhdr *eh = (struct ethhdr *) skb->data;
945 int i;
946
947 /* Exact match */
948 for (i = 0; i < filter->count; i++)
949 if (ether_addr_equal(eh->h_dest, filter->addr[i]))
950 return 1;
951
952 /* Inexact match (multicast only) */
953 if (is_multicast_ether_addr(eh->h_dest))
954 return addr_hash_test(filter->mask, eh->h_dest);
955
956 return 0;
957}
958
959/*
960 * Checks whether the packet is accepted or not.
961 * Returns: 0 - drop, !=0 - accept
962 */
963static int check_filter(struct tap_filter *filter, const struct sk_buff *skb)
964{
965 if (!filter->count)
966 return 1;
967
968 return run_filter(filter, skb);
969}
970
971/* Network device part of the driver */
972
973static const struct ethtool_ops tun_ethtool_ops;
974
975static int tun_net_init(struct net_device *dev)
976{
977 struct tun_struct *tun = netdev_priv(dev);
978 struct ifreq *ifr = tun->ifr;
979 int err;
980
981 spin_lock_init(&tun->lock);
982
983 err = security_tun_dev_alloc_security(&tun->security);
984 if (err < 0)
985 return err;
986
987 tun_flow_init(tun);
988
989 dev->pcpu_stat_type = NETDEV_PCPU_STAT_TSTATS;
990 dev->hw_features = NETIF_F_SG | NETIF_F_FRAGLIST |
991 TUN_USER_FEATURES | NETIF_F_HW_VLAN_CTAG_TX |
992 NETIF_F_HW_VLAN_STAG_TX;
993 dev->features = dev->hw_features;
994 dev->vlan_features = dev->features &
995 ~(NETIF_F_HW_VLAN_CTAG_TX |
996 NETIF_F_HW_VLAN_STAG_TX);
997 dev->lltx = true;
998
999 tun->flags = (tun->flags & ~TUN_FEATURES) |
1000 (ifr->ifr_flags & TUN_FEATURES);
1001
1002 INIT_LIST_HEAD(&tun->disabled);
1003 err = tun_attach(tun, tun->file, false, ifr->ifr_flags & IFF_NAPI,
1004 ifr->ifr_flags & IFF_NAPI_FRAGS, false);
1005 if (err < 0) {
1006 tun_flow_uninit(tun);
1007 security_tun_dev_free_security(tun->security);
1008 return err;
1009 }
1010 return 0;
1011}
1012
1013/* Net device detach from fd. */
1014static void tun_net_uninit(struct net_device *dev)
1015{
1016 tun_detach_all(dev);
1017}
1018
1019/* Net device open. */
1020static int tun_net_open(struct net_device *dev)
1021{
1022 netif_tx_start_all_queues(dev);
1023
1024 return 0;
1025}
1026
1027/* Net device close. */
1028static int tun_net_close(struct net_device *dev)
1029{
1030 netif_tx_stop_all_queues(dev);
1031 return 0;
1032}
1033
1034/* Net device start xmit */
1035static void tun_automq_xmit(struct tun_struct *tun, struct sk_buff *skb)
1036{
1037#ifdef CONFIG_RPS
1038 if (tun->numqueues == 1 && static_branch_unlikely(&rps_needed)) {
1039 /* Select queue was not called for the skbuff, so we extract the
1040 * RPS hash and save it into the flow_table here.
1041 */
1042 struct tun_flow_entry *e;
1043 __u32 rxhash;
1044
1045 rxhash = __skb_get_hash_symmetric(skb);
1046 e = tun_flow_find(&tun->flows[tun_hashfn(rxhash)], rxhash);
1047 if (e)
1048 tun_flow_save_rps_rxhash(e, rxhash);
1049 }
1050#endif
1051}
1052
1053static unsigned int run_ebpf_filter(struct tun_struct *tun,
1054 struct sk_buff *skb,
1055 int len)
1056{
1057 struct tun_prog *prog = rcu_dereference(tun->filter_prog);
1058
1059 if (prog)
1060 len = bpf_prog_run_clear_cb(prog->prog, skb);
1061
1062 return len;
1063}
1064
1065/* Net device start xmit */
1066static netdev_tx_t tun_net_xmit(struct sk_buff *skb, struct net_device *dev)
1067{
1068 struct tun_struct *tun = netdev_priv(dev);
1069 enum skb_drop_reason drop_reason;
1070 int txq = skb->queue_mapping;
1071 struct netdev_queue *queue;
1072 struct tun_file *tfile;
1073 int len = skb->len;
1074
1075 rcu_read_lock();
1076 tfile = rcu_dereference(tun->tfiles[txq]);
1077
1078 /* Drop packet if interface is not attached */
1079 if (!tfile) {
1080 drop_reason = SKB_DROP_REASON_DEV_READY;
1081 goto drop;
1082 }
1083
1084 if (!rcu_dereference(tun->steering_prog))
1085 tun_automq_xmit(tun, skb);
1086
1087 netif_info(tun, tx_queued, tun->dev, "%s %d\n", __func__, skb->len);
1088
1089 /* Drop if the filter does not like it.
1090 * This is a noop if the filter is disabled.
1091 * Filter can be enabled only for the TAP devices. */
1092 if (!check_filter(&tun->txflt, skb)) {
1093 drop_reason = SKB_DROP_REASON_TAP_TXFILTER;
1094 goto drop;
1095 }
1096
1097 if (tfile->socket.sk->sk_filter &&
1098 sk_filter(tfile->socket.sk, skb)) {
1099 drop_reason = SKB_DROP_REASON_SOCKET_FILTER;
1100 goto drop;
1101 }
1102
1103 len = run_ebpf_filter(tun, skb, len);
1104 if (len == 0) {
1105 drop_reason = SKB_DROP_REASON_TAP_FILTER;
1106 goto drop;
1107 }
1108
1109 if (pskb_trim(skb, len)) {
1110 drop_reason = SKB_DROP_REASON_NOMEM;
1111 goto drop;
1112 }
1113
1114 if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC))) {
1115 drop_reason = SKB_DROP_REASON_SKB_UCOPY_FAULT;
1116 goto drop;
1117 }
1118
1119 skb_tx_timestamp(skb);
1120
1121 /* Orphan the skb - required as we might hang on to it
1122 * for indefinite time.
1123 */
1124 skb_orphan(skb);
1125
1126 nf_reset_ct(skb);
1127
1128 if (ptr_ring_produce(&tfile->tx_ring, skb)) {
1129 drop_reason = SKB_DROP_REASON_FULL_RING;
1130 goto drop;
1131 }
1132
1133 /* dev->lltx requires to do our own update of trans_start */
1134 queue = netdev_get_tx_queue(dev, txq);
1135 txq_trans_cond_update(queue);
1136
1137 /* Notify and wake up reader process */
1138 if (tfile->flags & TUN_FASYNC)
1139 kill_fasync(&tfile->fasync, SIGIO, POLL_IN);
1140 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
1141
1142 rcu_read_unlock();
1143 return NETDEV_TX_OK;
1144
1145drop:
1146 dev_core_stats_tx_dropped_inc(dev);
1147 skb_tx_error(skb);
1148 kfree_skb_reason(skb, drop_reason);
1149 rcu_read_unlock();
1150 return NET_XMIT_DROP;
1151}
1152
1153static void tun_net_mclist(struct net_device *dev)
1154{
1155 /*
1156 * This callback is supposed to deal with mc filter in
1157 * _rx_ path and has nothing to do with the _tx_ path.
1158 * In rx path we always accept everything userspace gives us.
1159 */
1160}
1161
1162static netdev_features_t tun_net_fix_features(struct net_device *dev,
1163 netdev_features_t features)
1164{
1165 struct tun_struct *tun = netdev_priv(dev);
1166
1167 return (features & tun->set_features) | (features & ~TUN_USER_FEATURES);
1168}
1169
1170static void tun_set_headroom(struct net_device *dev, int new_hr)
1171{
1172 struct tun_struct *tun = netdev_priv(dev);
1173
1174 if (new_hr < NET_SKB_PAD)
1175 new_hr = NET_SKB_PAD;
1176
1177 tun->align = new_hr;
1178}
1179
1180static void
1181tun_net_get_stats64(struct net_device *dev, struct rtnl_link_stats64 *stats)
1182{
1183 struct tun_struct *tun = netdev_priv(dev);
1184
1185 dev_get_tstats64(dev, stats);
1186
1187 stats->rx_frame_errors +=
1188 (unsigned long)atomic_long_read(&tun->rx_frame_errors);
1189}
1190
1191static int tun_xdp_set(struct net_device *dev, struct bpf_prog *prog,
1192 struct netlink_ext_ack *extack)
1193{
1194 struct tun_struct *tun = netdev_priv(dev);
1195 struct tun_file *tfile;
1196 struct bpf_prog *old_prog;
1197 int i;
1198
1199 old_prog = rtnl_dereference(tun->xdp_prog);
1200 rcu_assign_pointer(tun->xdp_prog, prog);
1201 if (old_prog)
1202 bpf_prog_put(old_prog);
1203
1204 for (i = 0; i < tun->numqueues; i++) {
1205 tfile = rtnl_dereference(tun->tfiles[i]);
1206 if (prog)
1207 sock_set_flag(&tfile->sk, SOCK_XDP);
1208 else
1209 sock_reset_flag(&tfile->sk, SOCK_XDP);
1210 }
1211 list_for_each_entry(tfile, &tun->disabled, next) {
1212 if (prog)
1213 sock_set_flag(&tfile->sk, SOCK_XDP);
1214 else
1215 sock_reset_flag(&tfile->sk, SOCK_XDP);
1216 }
1217
1218 return 0;
1219}
1220
1221static int tun_xdp(struct net_device *dev, struct netdev_bpf *xdp)
1222{
1223 switch (xdp->command) {
1224 case XDP_SETUP_PROG:
1225 return tun_xdp_set(dev, xdp->prog, xdp->extack);
1226 default:
1227 return -EINVAL;
1228 }
1229}
1230
1231static int tun_net_change_carrier(struct net_device *dev, bool new_carrier)
1232{
1233 if (new_carrier) {
1234 struct tun_struct *tun = netdev_priv(dev);
1235
1236 if (!tun->numqueues)
1237 return -EPERM;
1238
1239 netif_carrier_on(dev);
1240 } else {
1241 netif_carrier_off(dev);
1242 }
1243 return 0;
1244}
1245
1246static const struct net_device_ops tun_netdev_ops = {
1247 .ndo_init = tun_net_init,
1248 .ndo_uninit = tun_net_uninit,
1249 .ndo_open = tun_net_open,
1250 .ndo_stop = tun_net_close,
1251 .ndo_start_xmit = tun_net_xmit,
1252 .ndo_fix_features = tun_net_fix_features,
1253 .ndo_select_queue = tun_select_queue,
1254 .ndo_set_rx_headroom = tun_set_headroom,
1255 .ndo_get_stats64 = tun_net_get_stats64,
1256 .ndo_change_carrier = tun_net_change_carrier,
1257};
1258
1259static void __tun_xdp_flush_tfile(struct tun_file *tfile)
1260{
1261 /* Notify and wake up reader process */
1262 if (tfile->flags & TUN_FASYNC)
1263 kill_fasync(&tfile->fasync, SIGIO, POLL_IN);
1264 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
1265}
1266
1267static int tun_xdp_xmit(struct net_device *dev, int n,
1268 struct xdp_frame **frames, u32 flags)
1269{
1270 struct tun_struct *tun = netdev_priv(dev);
1271 struct tun_file *tfile;
1272 u32 numqueues;
1273 int nxmit = 0;
1274 int i;
1275
1276 if (unlikely(flags & ~XDP_XMIT_FLAGS_MASK))
1277 return -EINVAL;
1278
1279 rcu_read_lock();
1280
1281resample:
1282 numqueues = READ_ONCE(tun->numqueues);
1283 if (!numqueues) {
1284 rcu_read_unlock();
1285 return -ENXIO; /* Caller will free/return all frames */
1286 }
1287
1288 tfile = rcu_dereference(tun->tfiles[smp_processor_id() %
1289 numqueues]);
1290 if (unlikely(!tfile))
1291 goto resample;
1292
1293 spin_lock(&tfile->tx_ring.producer_lock);
1294 for (i = 0; i < n; i++) {
1295 struct xdp_frame *xdp = frames[i];
1296 /* Encode the XDP flag into lowest bit for consumer to differ
1297 * XDP buffer from sk_buff.
1298 */
1299 void *frame = tun_xdp_to_ptr(xdp);
1300
1301 if (__ptr_ring_produce(&tfile->tx_ring, frame)) {
1302 dev_core_stats_tx_dropped_inc(dev);
1303 break;
1304 }
1305 nxmit++;
1306 }
1307 spin_unlock(&tfile->tx_ring.producer_lock);
1308
1309 if (flags & XDP_XMIT_FLUSH)
1310 __tun_xdp_flush_tfile(tfile);
1311
1312 rcu_read_unlock();
1313 return nxmit;
1314}
1315
1316static int tun_xdp_tx(struct net_device *dev, struct xdp_buff *xdp)
1317{
1318 struct xdp_frame *frame = xdp_convert_buff_to_frame(xdp);
1319 int nxmit;
1320
1321 if (unlikely(!frame))
1322 return -EOVERFLOW;
1323
1324 nxmit = tun_xdp_xmit(dev, 1, &frame, XDP_XMIT_FLUSH);
1325 if (!nxmit)
1326 xdp_return_frame_rx_napi(frame);
1327 return nxmit;
1328}
1329
1330static const struct net_device_ops tap_netdev_ops = {
1331 .ndo_init = tun_net_init,
1332 .ndo_uninit = tun_net_uninit,
1333 .ndo_open = tun_net_open,
1334 .ndo_stop = tun_net_close,
1335 .ndo_start_xmit = tun_net_xmit,
1336 .ndo_fix_features = tun_net_fix_features,
1337 .ndo_set_rx_mode = tun_net_mclist,
1338 .ndo_set_mac_address = eth_mac_addr,
1339 .ndo_validate_addr = eth_validate_addr,
1340 .ndo_select_queue = tun_select_queue,
1341 .ndo_features_check = passthru_features_check,
1342 .ndo_set_rx_headroom = tun_set_headroom,
1343 .ndo_bpf = tun_xdp,
1344 .ndo_xdp_xmit = tun_xdp_xmit,
1345 .ndo_change_carrier = tun_net_change_carrier,
1346};
1347
1348static void tun_flow_init(struct tun_struct *tun)
1349{
1350 int i;
1351
1352 for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++)
1353 INIT_HLIST_HEAD(&tun->flows[i]);
1354
1355 tun->ageing_time = TUN_FLOW_EXPIRE;
1356 timer_setup(&tun->flow_gc_timer, tun_flow_cleanup, 0);
1357 mod_timer(&tun->flow_gc_timer,
1358 round_jiffies_up(jiffies + tun->ageing_time));
1359}
1360
1361static void tun_flow_uninit(struct tun_struct *tun)
1362{
1363 del_timer_sync(&tun->flow_gc_timer);
1364 tun_flow_flush(tun);
1365}
1366
1367#define MIN_MTU 68
1368#define MAX_MTU 65535
1369
1370/* Initialize net device. */
1371static void tun_net_initialize(struct net_device *dev)
1372{
1373 struct tun_struct *tun = netdev_priv(dev);
1374
1375 switch (tun->flags & TUN_TYPE_MASK) {
1376 case IFF_TUN:
1377 dev->netdev_ops = &tun_netdev_ops;
1378 dev->header_ops = &ip_tunnel_header_ops;
1379
1380 /* Point-to-Point TUN Device */
1381 dev->hard_header_len = 0;
1382 dev->addr_len = 0;
1383 dev->mtu = 1500;
1384
1385 /* Zero header length */
1386 dev->type = ARPHRD_NONE;
1387 dev->flags = IFF_POINTOPOINT | IFF_NOARP | IFF_MULTICAST;
1388 break;
1389
1390 case IFF_TAP:
1391 dev->netdev_ops = &tap_netdev_ops;
1392 /* Ethernet TAP Device */
1393 ether_setup(dev);
1394 dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1395 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
1396
1397 eth_hw_addr_random(dev);
1398
1399 /* Currently tun does not support XDP, only tap does. */
1400 dev->xdp_features = NETDEV_XDP_ACT_BASIC |
1401 NETDEV_XDP_ACT_REDIRECT |
1402 NETDEV_XDP_ACT_NDO_XMIT;
1403
1404 break;
1405 }
1406
1407 dev->min_mtu = MIN_MTU;
1408 dev->max_mtu = MAX_MTU - dev->hard_header_len;
1409}
1410
1411static bool tun_sock_writeable(struct tun_struct *tun, struct tun_file *tfile)
1412{
1413 struct sock *sk = tfile->socket.sk;
1414
1415 return (tun->dev->flags & IFF_UP) && sock_writeable(sk);
1416}
1417
1418/* Character device part */
1419
1420/* Poll */
1421static __poll_t tun_chr_poll(struct file *file, poll_table *wait)
1422{
1423 struct tun_file *tfile = file->private_data;
1424 struct tun_struct *tun = tun_get(tfile);
1425 struct sock *sk;
1426 __poll_t mask = 0;
1427
1428 if (!tun)
1429 return EPOLLERR;
1430
1431 sk = tfile->socket.sk;
1432
1433 poll_wait(file, sk_sleep(sk), wait);
1434
1435 if (!ptr_ring_empty(&tfile->tx_ring))
1436 mask |= EPOLLIN | EPOLLRDNORM;
1437
1438 /* Make sure SOCKWQ_ASYNC_NOSPACE is set if not writable to
1439 * guarantee EPOLLOUT to be raised by either here or
1440 * tun_sock_write_space(). Then process could get notification
1441 * after it writes to a down device and meets -EIO.
1442 */
1443 if (tun_sock_writeable(tun, tfile) ||
1444 (!test_and_set_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags) &&
1445 tun_sock_writeable(tun, tfile)))
1446 mask |= EPOLLOUT | EPOLLWRNORM;
1447
1448 if (tun->dev->reg_state != NETREG_REGISTERED)
1449 mask = EPOLLERR;
1450
1451 tun_put(tun);
1452 return mask;
1453}
1454
1455static struct sk_buff *tun_napi_alloc_frags(struct tun_file *tfile,
1456 size_t len,
1457 const struct iov_iter *it)
1458{
1459 struct sk_buff *skb;
1460 size_t linear;
1461 int err;
1462 int i;
1463
1464 if (it->nr_segs > MAX_SKB_FRAGS + 1 ||
1465 len > (ETH_MAX_MTU - NET_SKB_PAD - NET_IP_ALIGN))
1466 return ERR_PTR(-EMSGSIZE);
1467
1468 local_bh_disable();
1469 skb = napi_get_frags(&tfile->napi);
1470 local_bh_enable();
1471 if (!skb)
1472 return ERR_PTR(-ENOMEM);
1473
1474 linear = iov_iter_single_seg_count(it);
1475 err = __skb_grow(skb, linear);
1476 if (err)
1477 goto free;
1478
1479 skb->len = len;
1480 skb->data_len = len - linear;
1481 skb->truesize += skb->data_len;
1482
1483 for (i = 1; i < it->nr_segs; i++) {
1484 const struct iovec *iov = iter_iov(it) + i;
1485 size_t fragsz = iov->iov_len;
1486 struct page *page;
1487 void *frag;
1488
1489 if (fragsz == 0 || fragsz > PAGE_SIZE) {
1490 err = -EINVAL;
1491 goto free;
1492 }
1493 frag = netdev_alloc_frag(fragsz);
1494 if (!frag) {
1495 err = -ENOMEM;
1496 goto free;
1497 }
1498 page = virt_to_head_page(frag);
1499 skb_fill_page_desc(skb, i - 1, page,
1500 frag - page_address(page), fragsz);
1501 }
1502
1503 return skb;
1504free:
1505 /* frees skb and all frags allocated with napi_alloc_frag() */
1506 napi_free_frags(&tfile->napi);
1507 return ERR_PTR(err);
1508}
1509
1510/* prepad is the amount to reserve at front. len is length after that.
1511 * linear is a hint as to how much to copy (usually headers). */
1512static struct sk_buff *tun_alloc_skb(struct tun_file *tfile,
1513 size_t prepad, size_t len,
1514 size_t linear, int noblock)
1515{
1516 struct sock *sk = tfile->socket.sk;
1517 struct sk_buff *skb;
1518 int err;
1519
1520 /* Under a page? Don't bother with paged skb. */
1521 if (prepad + len < PAGE_SIZE)
1522 linear = len;
1523
1524 if (len - linear > MAX_SKB_FRAGS * (PAGE_SIZE << PAGE_ALLOC_COSTLY_ORDER))
1525 linear = len - MAX_SKB_FRAGS * (PAGE_SIZE << PAGE_ALLOC_COSTLY_ORDER);
1526 skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
1527 &err, PAGE_ALLOC_COSTLY_ORDER);
1528 if (!skb)
1529 return ERR_PTR(err);
1530
1531 skb_reserve(skb, prepad);
1532 skb_put(skb, linear);
1533 skb->data_len = len - linear;
1534 skb->len += len - linear;
1535
1536 return skb;
1537}
1538
1539static void tun_rx_batched(struct tun_struct *tun, struct tun_file *tfile,
1540 struct sk_buff *skb, int more)
1541{
1542 struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
1543 struct sk_buff_head process_queue;
1544 u32 rx_batched = tun->rx_batched;
1545 bool rcv = false;
1546
1547 if (!rx_batched || (!more && skb_queue_empty(queue))) {
1548 local_bh_disable();
1549 skb_record_rx_queue(skb, tfile->queue_index);
1550 netif_receive_skb(skb);
1551 local_bh_enable();
1552 return;
1553 }
1554
1555 spin_lock(&queue->lock);
1556 if (!more || skb_queue_len(queue) == rx_batched) {
1557 __skb_queue_head_init(&process_queue);
1558 skb_queue_splice_tail_init(queue, &process_queue);
1559 rcv = true;
1560 } else {
1561 __skb_queue_tail(queue, skb);
1562 }
1563 spin_unlock(&queue->lock);
1564
1565 if (rcv) {
1566 struct sk_buff *nskb;
1567
1568 local_bh_disable();
1569 while ((nskb = __skb_dequeue(&process_queue))) {
1570 skb_record_rx_queue(nskb, tfile->queue_index);
1571 netif_receive_skb(nskb);
1572 }
1573 skb_record_rx_queue(skb, tfile->queue_index);
1574 netif_receive_skb(skb);
1575 local_bh_enable();
1576 }
1577}
1578
1579static bool tun_can_build_skb(struct tun_struct *tun, struct tun_file *tfile,
1580 int len, int noblock, bool zerocopy)
1581{
1582 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
1583 return false;
1584
1585 if (tfile->socket.sk->sk_sndbuf != INT_MAX)
1586 return false;
1587
1588 if (!noblock)
1589 return false;
1590
1591 if (zerocopy)
1592 return false;
1593
1594 if (SKB_DATA_ALIGN(len + TUN_RX_PAD + XDP_PACKET_HEADROOM) +
1595 SKB_DATA_ALIGN(sizeof(struct skb_shared_info)) > PAGE_SIZE)
1596 return false;
1597
1598 return true;
1599}
1600
1601static struct sk_buff *__tun_build_skb(struct tun_file *tfile,
1602 struct page_frag *alloc_frag, char *buf,
1603 int buflen, int len, int pad)
1604{
1605 struct sk_buff *skb = build_skb(buf, buflen);
1606
1607 if (!skb)
1608 return ERR_PTR(-ENOMEM);
1609
1610 skb_reserve(skb, pad);
1611 skb_put(skb, len);
1612 skb_set_owner_w(skb, tfile->socket.sk);
1613
1614 get_page(alloc_frag->page);
1615 alloc_frag->offset += buflen;
1616
1617 return skb;
1618}
1619
1620static int tun_xdp_act(struct tun_struct *tun, struct bpf_prog *xdp_prog,
1621 struct xdp_buff *xdp, u32 act)
1622{
1623 int err;
1624
1625 switch (act) {
1626 case XDP_REDIRECT:
1627 err = xdp_do_redirect(tun->dev, xdp, xdp_prog);
1628 if (err) {
1629 dev_core_stats_rx_dropped_inc(tun->dev);
1630 return err;
1631 }
1632 dev_sw_netstats_rx_add(tun->dev, xdp->data_end - xdp->data);
1633 break;
1634 case XDP_TX:
1635 err = tun_xdp_tx(tun->dev, xdp);
1636 if (err < 0) {
1637 dev_core_stats_rx_dropped_inc(tun->dev);
1638 return err;
1639 }
1640 dev_sw_netstats_rx_add(tun->dev, xdp->data_end - xdp->data);
1641 break;
1642 case XDP_PASS:
1643 break;
1644 default:
1645 bpf_warn_invalid_xdp_action(tun->dev, xdp_prog, act);
1646 fallthrough;
1647 case XDP_ABORTED:
1648 trace_xdp_exception(tun->dev, xdp_prog, act);
1649 fallthrough;
1650 case XDP_DROP:
1651 dev_core_stats_rx_dropped_inc(tun->dev);
1652 break;
1653 }
1654
1655 return act;
1656}
1657
1658static struct sk_buff *tun_build_skb(struct tun_struct *tun,
1659 struct tun_file *tfile,
1660 struct iov_iter *from,
1661 struct virtio_net_hdr *hdr,
1662 int len, int *skb_xdp)
1663{
1664 struct page_frag *alloc_frag = ¤t->task_frag;
1665 struct bpf_net_context __bpf_net_ctx, *bpf_net_ctx;
1666 struct bpf_prog *xdp_prog;
1667 int buflen = SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
1668 char *buf;
1669 size_t copied;
1670 int pad = TUN_RX_PAD;
1671 int err = 0;
1672
1673 rcu_read_lock();
1674 xdp_prog = rcu_dereference(tun->xdp_prog);
1675 if (xdp_prog)
1676 pad += XDP_PACKET_HEADROOM;
1677 buflen += SKB_DATA_ALIGN(len + pad);
1678 rcu_read_unlock();
1679
1680 alloc_frag->offset = ALIGN((u64)alloc_frag->offset, SMP_CACHE_BYTES);
1681 if (unlikely(!skb_page_frag_refill(buflen, alloc_frag, GFP_KERNEL)))
1682 return ERR_PTR(-ENOMEM);
1683
1684 buf = (char *)page_address(alloc_frag->page) + alloc_frag->offset;
1685 copied = copy_page_from_iter(alloc_frag->page,
1686 alloc_frag->offset + pad,
1687 len, from);
1688 if (copied != len)
1689 return ERR_PTR(-EFAULT);
1690
1691 /* There's a small window that XDP may be set after the check
1692 * of xdp_prog above, this should be rare and for simplicity
1693 * we do XDP on skb in case the headroom is not enough.
1694 */
1695 if (hdr->gso_type || !xdp_prog) {
1696 *skb_xdp = 1;
1697 return __tun_build_skb(tfile, alloc_frag, buf, buflen, len,
1698 pad);
1699 }
1700
1701 *skb_xdp = 0;
1702
1703 local_bh_disable();
1704 rcu_read_lock();
1705 bpf_net_ctx = bpf_net_ctx_set(&__bpf_net_ctx);
1706 xdp_prog = rcu_dereference(tun->xdp_prog);
1707 if (xdp_prog) {
1708 struct xdp_buff xdp;
1709 u32 act;
1710
1711 xdp_init_buff(&xdp, buflen, &tfile->xdp_rxq);
1712 xdp_prepare_buff(&xdp, buf, pad, len, false);
1713
1714 act = bpf_prog_run_xdp(xdp_prog, &xdp);
1715 if (act == XDP_REDIRECT || act == XDP_TX) {
1716 get_page(alloc_frag->page);
1717 alloc_frag->offset += buflen;
1718 }
1719 err = tun_xdp_act(tun, xdp_prog, &xdp, act);
1720 if (err < 0) {
1721 if (act == XDP_REDIRECT || act == XDP_TX)
1722 put_page(alloc_frag->page);
1723 goto out;
1724 }
1725
1726 if (err == XDP_REDIRECT)
1727 xdp_do_flush();
1728 if (err != XDP_PASS)
1729 goto out;
1730
1731 pad = xdp.data - xdp.data_hard_start;
1732 len = xdp.data_end - xdp.data;
1733 }
1734 bpf_net_ctx_clear(bpf_net_ctx);
1735 rcu_read_unlock();
1736 local_bh_enable();
1737
1738 return __tun_build_skb(tfile, alloc_frag, buf, buflen, len, pad);
1739
1740out:
1741 bpf_net_ctx_clear(bpf_net_ctx);
1742 rcu_read_unlock();
1743 local_bh_enable();
1744 return NULL;
1745}
1746
1747/* Get packet from user space buffer */
1748static ssize_t tun_get_user(struct tun_struct *tun, struct tun_file *tfile,
1749 void *msg_control, struct iov_iter *from,
1750 int noblock, bool more)
1751{
1752 struct tun_pi pi = { 0, cpu_to_be16(ETH_P_IP) };
1753 struct sk_buff *skb;
1754 size_t total_len = iov_iter_count(from);
1755 size_t len = total_len, align = tun->align, linear;
1756 struct virtio_net_hdr gso = { 0 };
1757 int good_linear;
1758 int copylen;
1759 bool zerocopy = false;
1760 int err;
1761 u32 rxhash = 0;
1762 int skb_xdp = 1;
1763 bool frags = tun_napi_frags_enabled(tfile);
1764 enum skb_drop_reason drop_reason = SKB_DROP_REASON_NOT_SPECIFIED;
1765
1766 if (!(tun->flags & IFF_NO_PI)) {
1767 if (len < sizeof(pi))
1768 return -EINVAL;
1769 len -= sizeof(pi);
1770
1771 if (!copy_from_iter_full(&pi, sizeof(pi), from))
1772 return -EFAULT;
1773 }
1774
1775 if (tun->flags & IFF_VNET_HDR) {
1776 int vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
1777
1778 if (len < vnet_hdr_sz)
1779 return -EINVAL;
1780 len -= vnet_hdr_sz;
1781
1782 if (!copy_from_iter_full(&gso, sizeof(gso), from))
1783 return -EFAULT;
1784
1785 if ((gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
1786 tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2 > tun16_to_cpu(tun, gso.hdr_len))
1787 gso.hdr_len = cpu_to_tun16(tun, tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2);
1788
1789 if (tun16_to_cpu(tun, gso.hdr_len) > len)
1790 return -EINVAL;
1791 iov_iter_advance(from, vnet_hdr_sz - sizeof(gso));
1792 }
1793
1794 if ((tun->flags & TUN_TYPE_MASK) == IFF_TAP) {
1795 align += NET_IP_ALIGN;
1796 if (unlikely(len < ETH_HLEN ||
1797 (gso.hdr_len && tun16_to_cpu(tun, gso.hdr_len) < ETH_HLEN)))
1798 return -EINVAL;
1799 }
1800
1801 good_linear = SKB_MAX_HEAD(align);
1802
1803 if (msg_control) {
1804 struct iov_iter i = *from;
1805
1806 /* There are 256 bytes to be copied in skb, so there is
1807 * enough room for skb expand head in case it is used.
1808 * The rest of the buffer is mapped from userspace.
1809 */
1810 copylen = gso.hdr_len ? tun16_to_cpu(tun, gso.hdr_len) : GOODCOPY_LEN;
1811 if (copylen > good_linear)
1812 copylen = good_linear;
1813 linear = copylen;
1814 iov_iter_advance(&i, copylen);
1815 if (iov_iter_npages(&i, INT_MAX) <= MAX_SKB_FRAGS)
1816 zerocopy = true;
1817 }
1818
1819 if (!frags && tun_can_build_skb(tun, tfile, len, noblock, zerocopy)) {
1820 /* For the packet that is not easy to be processed
1821 * (e.g gso or jumbo packet), we will do it at after
1822 * skb was created with generic XDP routine.
1823 */
1824 skb = tun_build_skb(tun, tfile, from, &gso, len, &skb_xdp);
1825 err = PTR_ERR_OR_ZERO(skb);
1826 if (err)
1827 goto drop;
1828 if (!skb)
1829 return total_len;
1830 } else {
1831 if (!zerocopy) {
1832 copylen = len;
1833 if (tun16_to_cpu(tun, gso.hdr_len) > good_linear)
1834 linear = good_linear;
1835 else
1836 linear = tun16_to_cpu(tun, gso.hdr_len);
1837 }
1838
1839 if (frags) {
1840 mutex_lock(&tfile->napi_mutex);
1841 skb = tun_napi_alloc_frags(tfile, copylen, from);
1842 /* tun_napi_alloc_frags() enforces a layout for the skb.
1843 * If zerocopy is enabled, then this layout will be
1844 * overwritten by zerocopy_sg_from_iter().
1845 */
1846 zerocopy = false;
1847 } else {
1848 if (!linear)
1849 linear = min_t(size_t, good_linear, copylen);
1850
1851 skb = tun_alloc_skb(tfile, align, copylen, linear,
1852 noblock);
1853 }
1854
1855 err = PTR_ERR_OR_ZERO(skb);
1856 if (err)
1857 goto drop;
1858
1859 if (zerocopy)
1860 err = zerocopy_sg_from_iter(skb, from);
1861 else
1862 err = skb_copy_datagram_from_iter(skb, 0, from, len);
1863
1864 if (err) {
1865 err = -EFAULT;
1866 drop_reason = SKB_DROP_REASON_SKB_UCOPY_FAULT;
1867 goto drop;
1868 }
1869 }
1870
1871 if (virtio_net_hdr_to_skb(skb, &gso, tun_is_little_endian(tun))) {
1872 atomic_long_inc(&tun->rx_frame_errors);
1873 err = -EINVAL;
1874 goto free_skb;
1875 }
1876
1877 switch (tun->flags & TUN_TYPE_MASK) {
1878 case IFF_TUN:
1879 if (tun->flags & IFF_NO_PI) {
1880 u8 ip_version = skb->len ? (skb->data[0] >> 4) : 0;
1881
1882 switch (ip_version) {
1883 case 4:
1884 pi.proto = htons(ETH_P_IP);
1885 break;
1886 case 6:
1887 pi.proto = htons(ETH_P_IPV6);
1888 break;
1889 default:
1890 err = -EINVAL;
1891 goto drop;
1892 }
1893 }
1894
1895 skb_reset_mac_header(skb);
1896 skb->protocol = pi.proto;
1897 skb->dev = tun->dev;
1898 break;
1899 case IFF_TAP:
1900 if (frags && !pskb_may_pull(skb, ETH_HLEN)) {
1901 err = -ENOMEM;
1902 drop_reason = SKB_DROP_REASON_HDR_TRUNC;
1903 goto drop;
1904 }
1905 skb->protocol = eth_type_trans(skb, tun->dev);
1906 break;
1907 }
1908
1909 /* copy skb_ubuf_info for callback when skb has no error */
1910 if (zerocopy) {
1911 skb_zcopy_init(skb, msg_control);
1912 } else if (msg_control) {
1913 struct ubuf_info *uarg = msg_control;
1914 uarg->ops->complete(NULL, uarg, false);
1915 }
1916
1917 skb_reset_network_header(skb);
1918 skb_probe_transport_header(skb);
1919 skb_record_rx_queue(skb, tfile->queue_index);
1920
1921 if (skb_xdp) {
1922 struct bpf_prog *xdp_prog;
1923 int ret;
1924
1925 local_bh_disable();
1926 rcu_read_lock();
1927 xdp_prog = rcu_dereference(tun->xdp_prog);
1928 if (xdp_prog) {
1929 ret = do_xdp_generic(xdp_prog, &skb);
1930 if (ret != XDP_PASS) {
1931 rcu_read_unlock();
1932 local_bh_enable();
1933 goto unlock_frags;
1934 }
1935 }
1936 rcu_read_unlock();
1937 local_bh_enable();
1938 }
1939
1940 /* Compute the costly rx hash only if needed for flow updates.
1941 * We may get a very small possibility of OOO during switching, not
1942 * worth to optimize.
1943 */
1944 if (!rcu_access_pointer(tun->steering_prog) && tun->numqueues > 1 &&
1945 !tfile->detached)
1946 rxhash = __skb_get_hash_symmetric(skb);
1947
1948 rcu_read_lock();
1949 if (unlikely(!(tun->dev->flags & IFF_UP))) {
1950 err = -EIO;
1951 rcu_read_unlock();
1952 drop_reason = SKB_DROP_REASON_DEV_READY;
1953 goto drop;
1954 }
1955
1956 if (frags) {
1957 u32 headlen;
1958
1959 /* Exercise flow dissector code path. */
1960 skb_push(skb, ETH_HLEN);
1961 headlen = eth_get_headlen(tun->dev, skb->data,
1962 skb_headlen(skb));
1963
1964 if (unlikely(headlen > skb_headlen(skb))) {
1965 WARN_ON_ONCE(1);
1966 err = -ENOMEM;
1967 dev_core_stats_rx_dropped_inc(tun->dev);
1968napi_busy:
1969 napi_free_frags(&tfile->napi);
1970 rcu_read_unlock();
1971 mutex_unlock(&tfile->napi_mutex);
1972 return err;
1973 }
1974
1975 if (likely(napi_schedule_prep(&tfile->napi))) {
1976 local_bh_disable();
1977 napi_gro_frags(&tfile->napi);
1978 napi_complete(&tfile->napi);
1979 local_bh_enable();
1980 } else {
1981 err = -EBUSY;
1982 goto napi_busy;
1983 }
1984 mutex_unlock(&tfile->napi_mutex);
1985 } else if (tfile->napi_enabled) {
1986 struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
1987 int queue_len;
1988
1989 spin_lock_bh(&queue->lock);
1990
1991 if (unlikely(tfile->detached)) {
1992 spin_unlock_bh(&queue->lock);
1993 rcu_read_unlock();
1994 err = -EBUSY;
1995 goto free_skb;
1996 }
1997
1998 __skb_queue_tail(queue, skb);
1999 queue_len = skb_queue_len(queue);
2000 spin_unlock(&queue->lock);
2001
2002 if (!more || queue_len > NAPI_POLL_WEIGHT)
2003 napi_schedule(&tfile->napi);
2004
2005 local_bh_enable();
2006 } else if (!IS_ENABLED(CONFIG_4KSTACKS)) {
2007 tun_rx_batched(tun, tfile, skb, more);
2008 } else {
2009 netif_rx(skb);
2010 }
2011 rcu_read_unlock();
2012
2013 preempt_disable();
2014 dev_sw_netstats_rx_add(tun->dev, len);
2015 preempt_enable();
2016
2017 if (rxhash)
2018 tun_flow_update(tun, rxhash, tfile);
2019
2020 return total_len;
2021
2022drop:
2023 if (err != -EAGAIN)
2024 dev_core_stats_rx_dropped_inc(tun->dev);
2025
2026free_skb:
2027 if (!IS_ERR_OR_NULL(skb))
2028 kfree_skb_reason(skb, drop_reason);
2029
2030unlock_frags:
2031 if (frags) {
2032 tfile->napi.skb = NULL;
2033 mutex_unlock(&tfile->napi_mutex);
2034 }
2035
2036 return err ?: total_len;
2037}
2038
2039static ssize_t tun_chr_write_iter(struct kiocb *iocb, struct iov_iter *from)
2040{
2041 struct file *file = iocb->ki_filp;
2042 struct tun_file *tfile = file->private_data;
2043 struct tun_struct *tun = tun_get(tfile);
2044 ssize_t result;
2045 int noblock = 0;
2046
2047 if (!tun)
2048 return -EBADFD;
2049
2050 if ((file->f_flags & O_NONBLOCK) || (iocb->ki_flags & IOCB_NOWAIT))
2051 noblock = 1;
2052
2053 result = tun_get_user(tun, tfile, NULL, from, noblock, false);
2054
2055 tun_put(tun);
2056 return result;
2057}
2058
2059static ssize_t tun_put_user_xdp(struct tun_struct *tun,
2060 struct tun_file *tfile,
2061 struct xdp_frame *xdp_frame,
2062 struct iov_iter *iter)
2063{
2064 int vnet_hdr_sz = 0;
2065 size_t size = xdp_frame->len;
2066 size_t ret;
2067
2068 if (tun->flags & IFF_VNET_HDR) {
2069 struct virtio_net_hdr gso = { 0 };
2070
2071 vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
2072 if (unlikely(iov_iter_count(iter) < vnet_hdr_sz))
2073 return -EINVAL;
2074 if (unlikely(copy_to_iter(&gso, sizeof(gso), iter) !=
2075 sizeof(gso)))
2076 return -EFAULT;
2077 iov_iter_advance(iter, vnet_hdr_sz - sizeof(gso));
2078 }
2079
2080 ret = copy_to_iter(xdp_frame->data, size, iter) + vnet_hdr_sz;
2081
2082 preempt_disable();
2083 dev_sw_netstats_tx_add(tun->dev, 1, ret);
2084 preempt_enable();
2085
2086 return ret;
2087}
2088
2089/* Put packet to the user space buffer */
2090static ssize_t tun_put_user(struct tun_struct *tun,
2091 struct tun_file *tfile,
2092 struct sk_buff *skb,
2093 struct iov_iter *iter)
2094{
2095 struct tun_pi pi = { 0, skb->protocol };
2096 ssize_t total;
2097 int vlan_offset = 0;
2098 int vlan_hlen = 0;
2099 int vnet_hdr_sz = 0;
2100
2101 if (skb_vlan_tag_present(skb))
2102 vlan_hlen = VLAN_HLEN;
2103
2104 if (tun->flags & IFF_VNET_HDR)
2105 vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
2106
2107 total = skb->len + vlan_hlen + vnet_hdr_sz;
2108
2109 if (!(tun->flags & IFF_NO_PI)) {
2110 if (iov_iter_count(iter) < sizeof(pi))
2111 return -EINVAL;
2112
2113 total += sizeof(pi);
2114 if (iov_iter_count(iter) < total) {
2115 /* Packet will be striped */
2116 pi.flags |= TUN_PKT_STRIP;
2117 }
2118
2119 if (copy_to_iter(&pi, sizeof(pi), iter) != sizeof(pi))
2120 return -EFAULT;
2121 }
2122
2123 if (vnet_hdr_sz) {
2124 struct virtio_net_hdr gso;
2125
2126 if (iov_iter_count(iter) < vnet_hdr_sz)
2127 return -EINVAL;
2128
2129 if (virtio_net_hdr_from_skb(skb, &gso,
2130 tun_is_little_endian(tun), true,
2131 vlan_hlen)) {
2132 struct skb_shared_info *sinfo = skb_shinfo(skb);
2133
2134 if (net_ratelimit()) {
2135 netdev_err(tun->dev, "unexpected GSO type: 0x%x, gso_size %d, hdr_len %d\n",
2136 sinfo->gso_type, tun16_to_cpu(tun, gso.gso_size),
2137 tun16_to_cpu(tun, gso.hdr_len));
2138 print_hex_dump(KERN_ERR, "tun: ",
2139 DUMP_PREFIX_NONE,
2140 16, 1, skb->head,
2141 min((int)tun16_to_cpu(tun, gso.hdr_len), 64), true);
2142 }
2143 WARN_ON_ONCE(1);
2144 return -EINVAL;
2145 }
2146
2147 if (copy_to_iter(&gso, sizeof(gso), iter) != sizeof(gso))
2148 return -EFAULT;
2149
2150 iov_iter_advance(iter, vnet_hdr_sz - sizeof(gso));
2151 }
2152
2153 if (vlan_hlen) {
2154 int ret;
2155 struct veth veth;
2156
2157 veth.h_vlan_proto = skb->vlan_proto;
2158 veth.h_vlan_TCI = htons(skb_vlan_tag_get(skb));
2159
2160 vlan_offset = offsetof(struct vlan_ethhdr, h_vlan_proto);
2161
2162 ret = skb_copy_datagram_iter(skb, 0, iter, vlan_offset);
2163 if (ret || !iov_iter_count(iter))
2164 goto done;
2165
2166 ret = copy_to_iter(&veth, sizeof(veth), iter);
2167 if (ret != sizeof(veth) || !iov_iter_count(iter))
2168 goto done;
2169 }
2170
2171 skb_copy_datagram_iter(skb, vlan_offset, iter, skb->len - vlan_offset);
2172
2173done:
2174 /* caller is in process context, */
2175 preempt_disable();
2176 dev_sw_netstats_tx_add(tun->dev, 1, skb->len + vlan_hlen);
2177 preempt_enable();
2178
2179 return total;
2180}
2181
2182static void *tun_ring_recv(struct tun_file *tfile, int noblock, int *err)
2183{
2184 DECLARE_WAITQUEUE(wait, current);
2185 void *ptr = NULL;
2186 int error = 0;
2187
2188 ptr = ptr_ring_consume(&tfile->tx_ring);
2189 if (ptr)
2190 goto out;
2191 if (noblock) {
2192 error = -EAGAIN;
2193 goto out;
2194 }
2195
2196 add_wait_queue(&tfile->socket.wq.wait, &wait);
2197
2198 while (1) {
2199 set_current_state(TASK_INTERRUPTIBLE);
2200 ptr = ptr_ring_consume(&tfile->tx_ring);
2201 if (ptr)
2202 break;
2203 if (signal_pending(current)) {
2204 error = -ERESTARTSYS;
2205 break;
2206 }
2207 if (tfile->socket.sk->sk_shutdown & RCV_SHUTDOWN) {
2208 error = -EFAULT;
2209 break;
2210 }
2211
2212 schedule();
2213 }
2214
2215 __set_current_state(TASK_RUNNING);
2216 remove_wait_queue(&tfile->socket.wq.wait, &wait);
2217
2218out:
2219 *err = error;
2220 return ptr;
2221}
2222
2223static ssize_t tun_do_read(struct tun_struct *tun, struct tun_file *tfile,
2224 struct iov_iter *to,
2225 int noblock, void *ptr)
2226{
2227 ssize_t ret;
2228 int err;
2229
2230 if (!iov_iter_count(to)) {
2231 tun_ptr_free(ptr);
2232 return 0;
2233 }
2234
2235 if (!ptr) {
2236 /* Read frames from ring */
2237 ptr = tun_ring_recv(tfile, noblock, &err);
2238 if (!ptr)
2239 return err;
2240 }
2241
2242 if (tun_is_xdp_frame(ptr)) {
2243 struct xdp_frame *xdpf = tun_ptr_to_xdp(ptr);
2244
2245 ret = tun_put_user_xdp(tun, tfile, xdpf, to);
2246 xdp_return_frame(xdpf);
2247 } else {
2248 struct sk_buff *skb = ptr;
2249
2250 ret = tun_put_user(tun, tfile, skb, to);
2251 if (unlikely(ret < 0))
2252 kfree_skb(skb);
2253 else
2254 consume_skb(skb);
2255 }
2256
2257 return ret;
2258}
2259
2260static ssize_t tun_chr_read_iter(struct kiocb *iocb, struct iov_iter *to)
2261{
2262 struct file *file = iocb->ki_filp;
2263 struct tun_file *tfile = file->private_data;
2264 struct tun_struct *tun = tun_get(tfile);
2265 ssize_t len = iov_iter_count(to), ret;
2266 int noblock = 0;
2267
2268 if (!tun)
2269 return -EBADFD;
2270
2271 if ((file->f_flags & O_NONBLOCK) || (iocb->ki_flags & IOCB_NOWAIT))
2272 noblock = 1;
2273
2274 ret = tun_do_read(tun, tfile, to, noblock, NULL);
2275 ret = min_t(ssize_t, ret, len);
2276 if (ret > 0)
2277 iocb->ki_pos = ret;
2278 tun_put(tun);
2279 return ret;
2280}
2281
2282static void tun_prog_free(struct rcu_head *rcu)
2283{
2284 struct tun_prog *prog = container_of(rcu, struct tun_prog, rcu);
2285
2286 bpf_prog_destroy(prog->prog);
2287 kfree(prog);
2288}
2289
2290static int __tun_set_ebpf(struct tun_struct *tun,
2291 struct tun_prog __rcu **prog_p,
2292 struct bpf_prog *prog)
2293{
2294 struct tun_prog *old, *new = NULL;
2295
2296 if (prog) {
2297 new = kmalloc(sizeof(*new), GFP_KERNEL);
2298 if (!new)
2299 return -ENOMEM;
2300 new->prog = prog;
2301 }
2302
2303 spin_lock_bh(&tun->lock);
2304 old = rcu_dereference_protected(*prog_p,
2305 lockdep_is_held(&tun->lock));
2306 rcu_assign_pointer(*prog_p, new);
2307 spin_unlock_bh(&tun->lock);
2308
2309 if (old)
2310 call_rcu(&old->rcu, tun_prog_free);
2311
2312 return 0;
2313}
2314
2315static void tun_free_netdev(struct net_device *dev)
2316{
2317 struct tun_struct *tun = netdev_priv(dev);
2318
2319 BUG_ON(!(list_empty(&tun->disabled)));
2320
2321 tun_flow_uninit(tun);
2322 security_tun_dev_free_security(tun->security);
2323 __tun_set_ebpf(tun, &tun->steering_prog, NULL);
2324 __tun_set_ebpf(tun, &tun->filter_prog, NULL);
2325}
2326
2327static void tun_setup(struct net_device *dev)
2328{
2329 struct tun_struct *tun = netdev_priv(dev);
2330
2331 tun->owner = INVALID_UID;
2332 tun->group = INVALID_GID;
2333 tun_default_link_ksettings(dev, &tun->link_ksettings);
2334
2335 dev->ethtool_ops = &tun_ethtool_ops;
2336 dev->needs_free_netdev = true;
2337 dev->priv_destructor = tun_free_netdev;
2338 /* We prefer our own queue length */
2339 dev->tx_queue_len = TUN_READQ_SIZE;
2340}
2341
2342/* Trivial set of netlink ops to allow deleting tun or tap
2343 * device with netlink.
2344 */
2345static int tun_validate(struct nlattr *tb[], struct nlattr *data[],
2346 struct netlink_ext_ack *extack)
2347{
2348 NL_SET_ERR_MSG(extack,
2349 "tun/tap creation via rtnetlink is not supported.");
2350 return -EOPNOTSUPP;
2351}
2352
2353static size_t tun_get_size(const struct net_device *dev)
2354{
2355 BUILD_BUG_ON(sizeof(u32) != sizeof(uid_t));
2356 BUILD_BUG_ON(sizeof(u32) != sizeof(gid_t));
2357
2358 return nla_total_size(sizeof(uid_t)) + /* OWNER */
2359 nla_total_size(sizeof(gid_t)) + /* GROUP */
2360 nla_total_size(sizeof(u8)) + /* TYPE */
2361 nla_total_size(sizeof(u8)) + /* PI */
2362 nla_total_size(sizeof(u8)) + /* VNET_HDR */
2363 nla_total_size(sizeof(u8)) + /* PERSIST */
2364 nla_total_size(sizeof(u8)) + /* MULTI_QUEUE */
2365 nla_total_size(sizeof(u32)) + /* NUM_QUEUES */
2366 nla_total_size(sizeof(u32)) + /* NUM_DISABLED_QUEUES */
2367 0;
2368}
2369
2370static int tun_fill_info(struct sk_buff *skb, const struct net_device *dev)
2371{
2372 struct tun_struct *tun = netdev_priv(dev);
2373
2374 if (nla_put_u8(skb, IFLA_TUN_TYPE, tun->flags & TUN_TYPE_MASK))
2375 goto nla_put_failure;
2376 if (uid_valid(tun->owner) &&
2377 nla_put_u32(skb, IFLA_TUN_OWNER,
2378 from_kuid_munged(current_user_ns(), tun->owner)))
2379 goto nla_put_failure;
2380 if (gid_valid(tun->group) &&
2381 nla_put_u32(skb, IFLA_TUN_GROUP,
2382 from_kgid_munged(current_user_ns(), tun->group)))
2383 goto nla_put_failure;
2384 if (nla_put_u8(skb, IFLA_TUN_PI, !(tun->flags & IFF_NO_PI)))
2385 goto nla_put_failure;
2386 if (nla_put_u8(skb, IFLA_TUN_VNET_HDR, !!(tun->flags & IFF_VNET_HDR)))
2387 goto nla_put_failure;
2388 if (nla_put_u8(skb, IFLA_TUN_PERSIST, !!(tun->flags & IFF_PERSIST)))
2389 goto nla_put_failure;
2390 if (nla_put_u8(skb, IFLA_TUN_MULTI_QUEUE,
2391 !!(tun->flags & IFF_MULTI_QUEUE)))
2392 goto nla_put_failure;
2393 if (tun->flags & IFF_MULTI_QUEUE) {
2394 if (nla_put_u32(skb, IFLA_TUN_NUM_QUEUES, tun->numqueues))
2395 goto nla_put_failure;
2396 if (nla_put_u32(skb, IFLA_TUN_NUM_DISABLED_QUEUES,
2397 tun->numdisabled))
2398 goto nla_put_failure;
2399 }
2400
2401 return 0;
2402
2403nla_put_failure:
2404 return -EMSGSIZE;
2405}
2406
2407static struct rtnl_link_ops tun_link_ops __read_mostly = {
2408 .kind = DRV_NAME,
2409 .priv_size = sizeof(struct tun_struct),
2410 .setup = tun_setup,
2411 .validate = tun_validate,
2412 .get_size = tun_get_size,
2413 .fill_info = tun_fill_info,
2414};
2415
2416static void tun_sock_write_space(struct sock *sk)
2417{
2418 struct tun_file *tfile;
2419 wait_queue_head_t *wqueue;
2420
2421 if (!sock_writeable(sk))
2422 return;
2423
2424 if (!test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags))
2425 return;
2426
2427 wqueue = sk_sleep(sk);
2428 if (wqueue && waitqueue_active(wqueue))
2429 wake_up_interruptible_sync_poll(wqueue, EPOLLOUT |
2430 EPOLLWRNORM | EPOLLWRBAND);
2431
2432 tfile = container_of(sk, struct tun_file, sk);
2433 kill_fasync(&tfile->fasync, SIGIO, POLL_OUT);
2434}
2435
2436static void tun_put_page(struct tun_page *tpage)
2437{
2438 if (tpage->page)
2439 __page_frag_cache_drain(tpage->page, tpage->count);
2440}
2441
2442static int tun_xdp_one(struct tun_struct *tun,
2443 struct tun_file *tfile,
2444 struct xdp_buff *xdp, int *flush,
2445 struct tun_page *tpage)
2446{
2447 unsigned int datasize = xdp->data_end - xdp->data;
2448 struct tun_xdp_hdr *hdr = xdp->data_hard_start;
2449 struct virtio_net_hdr *gso = &hdr->gso;
2450 struct bpf_prog *xdp_prog;
2451 struct sk_buff *skb = NULL;
2452 struct sk_buff_head *queue;
2453 u32 rxhash = 0, act;
2454 int buflen = hdr->buflen;
2455 int ret = 0;
2456 bool skb_xdp = false;
2457 struct page *page;
2458
2459 if (unlikely(datasize < ETH_HLEN))
2460 return -EINVAL;
2461
2462 xdp_prog = rcu_dereference(tun->xdp_prog);
2463 if (xdp_prog) {
2464 if (gso->gso_type) {
2465 skb_xdp = true;
2466 goto build;
2467 }
2468
2469 xdp_init_buff(xdp, buflen, &tfile->xdp_rxq);
2470 xdp_set_data_meta_invalid(xdp);
2471
2472 act = bpf_prog_run_xdp(xdp_prog, xdp);
2473 ret = tun_xdp_act(tun, xdp_prog, xdp, act);
2474 if (ret < 0) {
2475 put_page(virt_to_head_page(xdp->data));
2476 return ret;
2477 }
2478
2479 switch (ret) {
2480 case XDP_REDIRECT:
2481 *flush = true;
2482 fallthrough;
2483 case XDP_TX:
2484 return 0;
2485 case XDP_PASS:
2486 break;
2487 default:
2488 page = virt_to_head_page(xdp->data);
2489 if (tpage->page == page) {
2490 ++tpage->count;
2491 } else {
2492 tun_put_page(tpage);
2493 tpage->page = page;
2494 tpage->count = 1;
2495 }
2496 return 0;
2497 }
2498 }
2499
2500build:
2501 skb = build_skb(xdp->data_hard_start, buflen);
2502 if (!skb) {
2503 ret = -ENOMEM;
2504 goto out;
2505 }
2506
2507 skb_reserve(skb, xdp->data - xdp->data_hard_start);
2508 skb_put(skb, xdp->data_end - xdp->data);
2509
2510 if (virtio_net_hdr_to_skb(skb, gso, tun_is_little_endian(tun))) {
2511 atomic_long_inc(&tun->rx_frame_errors);
2512 kfree_skb(skb);
2513 ret = -EINVAL;
2514 goto out;
2515 }
2516
2517 skb->protocol = eth_type_trans(skb, tun->dev);
2518 skb_reset_network_header(skb);
2519 skb_probe_transport_header(skb);
2520 skb_record_rx_queue(skb, tfile->queue_index);
2521
2522 if (skb_xdp) {
2523 ret = do_xdp_generic(xdp_prog, &skb);
2524 if (ret != XDP_PASS) {
2525 ret = 0;
2526 goto out;
2527 }
2528 }
2529
2530 if (!rcu_dereference(tun->steering_prog) && tun->numqueues > 1 &&
2531 !tfile->detached)
2532 rxhash = __skb_get_hash_symmetric(skb);
2533
2534 if (tfile->napi_enabled) {
2535 queue = &tfile->sk.sk_write_queue;
2536 spin_lock(&queue->lock);
2537
2538 if (unlikely(tfile->detached)) {
2539 spin_unlock(&queue->lock);
2540 kfree_skb(skb);
2541 return -EBUSY;
2542 }
2543
2544 __skb_queue_tail(queue, skb);
2545 spin_unlock(&queue->lock);
2546 ret = 1;
2547 } else {
2548 netif_receive_skb(skb);
2549 ret = 0;
2550 }
2551
2552 /* No need to disable preemption here since this function is
2553 * always called with bh disabled
2554 */
2555 dev_sw_netstats_rx_add(tun->dev, datasize);
2556
2557 if (rxhash)
2558 tun_flow_update(tun, rxhash, tfile);
2559
2560out:
2561 return ret;
2562}
2563
2564static int tun_sendmsg(struct socket *sock, struct msghdr *m, size_t total_len)
2565{
2566 int ret, i;
2567 struct tun_file *tfile = container_of(sock, struct tun_file, socket);
2568 struct tun_struct *tun = tun_get(tfile);
2569 struct tun_msg_ctl *ctl = m->msg_control;
2570 struct xdp_buff *xdp;
2571
2572 if (!tun)
2573 return -EBADFD;
2574
2575 if (m->msg_controllen == sizeof(struct tun_msg_ctl) &&
2576 ctl && ctl->type == TUN_MSG_PTR) {
2577 struct bpf_net_context __bpf_net_ctx, *bpf_net_ctx;
2578 struct tun_page tpage;
2579 int n = ctl->num;
2580 int flush = 0, queued = 0;
2581
2582 memset(&tpage, 0, sizeof(tpage));
2583
2584 local_bh_disable();
2585 rcu_read_lock();
2586 bpf_net_ctx = bpf_net_ctx_set(&__bpf_net_ctx);
2587
2588 for (i = 0; i < n; i++) {
2589 xdp = &((struct xdp_buff *)ctl->ptr)[i];
2590 ret = tun_xdp_one(tun, tfile, xdp, &flush, &tpage);
2591 if (ret > 0)
2592 queued += ret;
2593 }
2594
2595 if (flush)
2596 xdp_do_flush();
2597
2598 if (tfile->napi_enabled && queued > 0)
2599 napi_schedule(&tfile->napi);
2600
2601 bpf_net_ctx_clear(bpf_net_ctx);
2602 rcu_read_unlock();
2603 local_bh_enable();
2604
2605 tun_put_page(&tpage);
2606
2607 ret = total_len;
2608 goto out;
2609 }
2610
2611 ret = tun_get_user(tun, tfile, ctl ? ctl->ptr : NULL, &m->msg_iter,
2612 m->msg_flags & MSG_DONTWAIT,
2613 m->msg_flags & MSG_MORE);
2614out:
2615 tun_put(tun);
2616 return ret;
2617}
2618
2619static int tun_recvmsg(struct socket *sock, struct msghdr *m, size_t total_len,
2620 int flags)
2621{
2622 struct tun_file *tfile = container_of(sock, struct tun_file, socket);
2623 struct tun_struct *tun = tun_get(tfile);
2624 void *ptr = m->msg_control;
2625 int ret;
2626
2627 if (!tun) {
2628 ret = -EBADFD;
2629 goto out_free;
2630 }
2631
2632 if (flags & ~(MSG_DONTWAIT|MSG_TRUNC|MSG_ERRQUEUE)) {
2633 ret = -EINVAL;
2634 goto out_put_tun;
2635 }
2636 if (flags & MSG_ERRQUEUE) {
2637 ret = sock_recv_errqueue(sock->sk, m, total_len,
2638 SOL_PACKET, TUN_TX_TIMESTAMP);
2639 goto out;
2640 }
2641 ret = tun_do_read(tun, tfile, &m->msg_iter, flags & MSG_DONTWAIT, ptr);
2642 if (ret > (ssize_t)total_len) {
2643 m->msg_flags |= MSG_TRUNC;
2644 ret = flags & MSG_TRUNC ? ret : total_len;
2645 }
2646out:
2647 tun_put(tun);
2648 return ret;
2649
2650out_put_tun:
2651 tun_put(tun);
2652out_free:
2653 tun_ptr_free(ptr);
2654 return ret;
2655}
2656
2657static int tun_ptr_peek_len(void *ptr)
2658{
2659 if (likely(ptr)) {
2660 if (tun_is_xdp_frame(ptr)) {
2661 struct xdp_frame *xdpf = tun_ptr_to_xdp(ptr);
2662
2663 return xdpf->len;
2664 }
2665 return __skb_array_len_with_tag(ptr);
2666 } else {
2667 return 0;
2668 }
2669}
2670
2671static int tun_peek_len(struct socket *sock)
2672{
2673 struct tun_file *tfile = container_of(sock, struct tun_file, socket);
2674 struct tun_struct *tun;
2675 int ret = 0;
2676
2677 tun = tun_get(tfile);
2678 if (!tun)
2679 return 0;
2680
2681 ret = PTR_RING_PEEK_CALL(&tfile->tx_ring, tun_ptr_peek_len);
2682 tun_put(tun);
2683
2684 return ret;
2685}
2686
2687/* Ops structure to mimic raw sockets with tun */
2688static const struct proto_ops tun_socket_ops = {
2689 .peek_len = tun_peek_len,
2690 .sendmsg = tun_sendmsg,
2691 .recvmsg = tun_recvmsg,
2692};
2693
2694static struct proto tun_proto = {
2695 .name = "tun",
2696 .owner = THIS_MODULE,
2697 .obj_size = sizeof(struct tun_file),
2698};
2699
2700static int tun_flags(struct tun_struct *tun)
2701{
2702 return tun->flags & (TUN_FEATURES | IFF_PERSIST | IFF_TUN | IFF_TAP);
2703}
2704
2705static ssize_t tun_flags_show(struct device *dev, struct device_attribute *attr,
2706 char *buf)
2707{
2708 struct tun_struct *tun = netdev_priv(to_net_dev(dev));
2709 return sysfs_emit(buf, "0x%x\n", tun_flags(tun));
2710}
2711
2712static ssize_t owner_show(struct device *dev, struct device_attribute *attr,
2713 char *buf)
2714{
2715 struct tun_struct *tun = netdev_priv(to_net_dev(dev));
2716 return uid_valid(tun->owner)?
2717 sysfs_emit(buf, "%u\n",
2718 from_kuid_munged(current_user_ns(), tun->owner)) :
2719 sysfs_emit(buf, "-1\n");
2720}
2721
2722static ssize_t group_show(struct device *dev, struct device_attribute *attr,
2723 char *buf)
2724{
2725 struct tun_struct *tun = netdev_priv(to_net_dev(dev));
2726 return gid_valid(tun->group) ?
2727 sysfs_emit(buf, "%u\n",
2728 from_kgid_munged(current_user_ns(), tun->group)) :
2729 sysfs_emit(buf, "-1\n");
2730}
2731
2732static DEVICE_ATTR_RO(tun_flags);
2733static DEVICE_ATTR_RO(owner);
2734static DEVICE_ATTR_RO(group);
2735
2736static struct attribute *tun_dev_attrs[] = {
2737 &dev_attr_tun_flags.attr,
2738 &dev_attr_owner.attr,
2739 &dev_attr_group.attr,
2740 NULL
2741};
2742
2743static const struct attribute_group tun_attr_group = {
2744 .attrs = tun_dev_attrs
2745};
2746
2747static int tun_set_iff(struct net *net, struct file *file, struct ifreq *ifr)
2748{
2749 struct tun_struct *tun;
2750 struct tun_file *tfile = file->private_data;
2751 struct net_device *dev;
2752 int err;
2753
2754 if (tfile->detached)
2755 return -EINVAL;
2756
2757 if ((ifr->ifr_flags & IFF_NAPI_FRAGS)) {
2758 if (!capable(CAP_NET_ADMIN))
2759 return -EPERM;
2760
2761 if (!(ifr->ifr_flags & IFF_NAPI) ||
2762 (ifr->ifr_flags & TUN_TYPE_MASK) != IFF_TAP)
2763 return -EINVAL;
2764 }
2765
2766 dev = __dev_get_by_name(net, ifr->ifr_name);
2767 if (dev) {
2768 if (ifr->ifr_flags & IFF_TUN_EXCL)
2769 return -EBUSY;
2770 if ((ifr->ifr_flags & IFF_TUN) && dev->netdev_ops == &tun_netdev_ops)
2771 tun = netdev_priv(dev);
2772 else if ((ifr->ifr_flags & IFF_TAP) && dev->netdev_ops == &tap_netdev_ops)
2773 tun = netdev_priv(dev);
2774 else
2775 return -EINVAL;
2776
2777 if (!!(ifr->ifr_flags & IFF_MULTI_QUEUE) !=
2778 !!(tun->flags & IFF_MULTI_QUEUE))
2779 return -EINVAL;
2780
2781 if (tun_not_capable(tun))
2782 return -EPERM;
2783 err = security_tun_dev_open(tun->security);
2784 if (err < 0)
2785 return err;
2786
2787 err = tun_attach(tun, file, ifr->ifr_flags & IFF_NOFILTER,
2788 ifr->ifr_flags & IFF_NAPI,
2789 ifr->ifr_flags & IFF_NAPI_FRAGS, true);
2790 if (err < 0)
2791 return err;
2792
2793 if (tun->flags & IFF_MULTI_QUEUE &&
2794 (tun->numqueues + tun->numdisabled > 1)) {
2795 /* One or more queue has already been attached, no need
2796 * to initialize the device again.
2797 */
2798 netdev_state_change(dev);
2799 return 0;
2800 }
2801
2802 tun->flags = (tun->flags & ~TUN_FEATURES) |
2803 (ifr->ifr_flags & TUN_FEATURES);
2804
2805 netdev_state_change(dev);
2806 } else {
2807 char *name;
2808 unsigned long flags = 0;
2809 int queues = ifr->ifr_flags & IFF_MULTI_QUEUE ?
2810 MAX_TAP_QUEUES : 1;
2811
2812 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
2813 return -EPERM;
2814 err = security_tun_dev_create();
2815 if (err < 0)
2816 return err;
2817
2818 /* Set dev type */
2819 if (ifr->ifr_flags & IFF_TUN) {
2820 /* TUN device */
2821 flags |= IFF_TUN;
2822 name = "tun%d";
2823 } else if (ifr->ifr_flags & IFF_TAP) {
2824 /* TAP device */
2825 flags |= IFF_TAP;
2826 name = "tap%d";
2827 } else
2828 return -EINVAL;
2829
2830 if (*ifr->ifr_name)
2831 name = ifr->ifr_name;
2832
2833 dev = alloc_netdev_mqs(sizeof(struct tun_struct), name,
2834 NET_NAME_UNKNOWN, tun_setup, queues,
2835 queues);
2836
2837 if (!dev)
2838 return -ENOMEM;
2839
2840 dev_net_set(dev, net);
2841 dev->rtnl_link_ops = &tun_link_ops;
2842 dev->ifindex = tfile->ifindex;
2843 dev->sysfs_groups[0] = &tun_attr_group;
2844
2845 tun = netdev_priv(dev);
2846 tun->dev = dev;
2847 tun->flags = flags;
2848 tun->txflt.count = 0;
2849 tun->vnet_hdr_sz = sizeof(struct virtio_net_hdr);
2850
2851 tun->align = NET_SKB_PAD;
2852 tun->filter_attached = false;
2853 tun->sndbuf = tfile->socket.sk->sk_sndbuf;
2854 tun->rx_batched = 0;
2855 RCU_INIT_POINTER(tun->steering_prog, NULL);
2856
2857 tun->ifr = ifr;
2858 tun->file = file;
2859
2860 tun_net_initialize(dev);
2861
2862 err = register_netdevice(tun->dev);
2863 if (err < 0) {
2864 free_netdev(dev);
2865 return err;
2866 }
2867 /* free_netdev() won't check refcnt, to avoid race
2868 * with dev_put() we need publish tun after registration.
2869 */
2870 rcu_assign_pointer(tfile->tun, tun);
2871 }
2872
2873 if (ifr->ifr_flags & IFF_NO_CARRIER)
2874 netif_carrier_off(tun->dev);
2875 else
2876 netif_carrier_on(tun->dev);
2877
2878 /* Make sure persistent devices do not get stuck in
2879 * xoff state.
2880 */
2881 if (netif_running(tun->dev))
2882 netif_tx_wake_all_queues(tun->dev);
2883
2884 strcpy(ifr->ifr_name, tun->dev->name);
2885 return 0;
2886}
2887
2888static void tun_get_iff(struct tun_struct *tun, struct ifreq *ifr)
2889{
2890 strcpy(ifr->ifr_name, tun->dev->name);
2891
2892 ifr->ifr_flags = tun_flags(tun);
2893
2894}
2895
2896/* This is like a cut-down ethtool ops, except done via tun fd so no
2897 * privs required. */
2898static int set_offload(struct tun_struct *tun, unsigned long arg)
2899{
2900 netdev_features_t features = 0;
2901
2902 if (arg & TUN_F_CSUM) {
2903 features |= NETIF_F_HW_CSUM;
2904 arg &= ~TUN_F_CSUM;
2905
2906 if (arg & (TUN_F_TSO4|TUN_F_TSO6)) {
2907 if (arg & TUN_F_TSO_ECN) {
2908 features |= NETIF_F_TSO_ECN;
2909 arg &= ~TUN_F_TSO_ECN;
2910 }
2911 if (arg & TUN_F_TSO4)
2912 features |= NETIF_F_TSO;
2913 if (arg & TUN_F_TSO6)
2914 features |= NETIF_F_TSO6;
2915 arg &= ~(TUN_F_TSO4|TUN_F_TSO6);
2916 }
2917
2918 arg &= ~TUN_F_UFO;
2919
2920 /* TODO: for now USO4 and USO6 should work simultaneously */
2921 if (arg & TUN_F_USO4 && arg & TUN_F_USO6) {
2922 features |= NETIF_F_GSO_UDP_L4;
2923 arg &= ~(TUN_F_USO4 | TUN_F_USO6);
2924 }
2925 }
2926
2927 /* This gives the user a way to test for new features in future by
2928 * trying to set them. */
2929 if (arg)
2930 return -EINVAL;
2931
2932 tun->set_features = features;
2933 tun->dev->wanted_features &= ~TUN_USER_FEATURES;
2934 tun->dev->wanted_features |= features;
2935 netdev_update_features(tun->dev);
2936
2937 return 0;
2938}
2939
2940static void tun_detach_filter(struct tun_struct *tun, int n)
2941{
2942 int i;
2943 struct tun_file *tfile;
2944
2945 for (i = 0; i < n; i++) {
2946 tfile = rtnl_dereference(tun->tfiles[i]);
2947 lock_sock(tfile->socket.sk);
2948 sk_detach_filter(tfile->socket.sk);
2949 release_sock(tfile->socket.sk);
2950 }
2951
2952 tun->filter_attached = false;
2953}
2954
2955static int tun_attach_filter(struct tun_struct *tun)
2956{
2957 int i, ret = 0;
2958 struct tun_file *tfile;
2959
2960 for (i = 0; i < tun->numqueues; i++) {
2961 tfile = rtnl_dereference(tun->tfiles[i]);
2962 lock_sock(tfile->socket.sk);
2963 ret = sk_attach_filter(&tun->fprog, tfile->socket.sk);
2964 release_sock(tfile->socket.sk);
2965 if (ret) {
2966 tun_detach_filter(tun, i);
2967 return ret;
2968 }
2969 }
2970
2971 tun->filter_attached = true;
2972 return ret;
2973}
2974
2975static void tun_set_sndbuf(struct tun_struct *tun)
2976{
2977 struct tun_file *tfile;
2978 int i;
2979
2980 for (i = 0; i < tun->numqueues; i++) {
2981 tfile = rtnl_dereference(tun->tfiles[i]);
2982 tfile->socket.sk->sk_sndbuf = tun->sndbuf;
2983 }
2984}
2985
2986static int tun_set_queue(struct file *file, struct ifreq *ifr)
2987{
2988 struct tun_file *tfile = file->private_data;
2989 struct tun_struct *tun;
2990 int ret = 0;
2991
2992 rtnl_lock();
2993
2994 if (ifr->ifr_flags & IFF_ATTACH_QUEUE) {
2995 tun = tfile->detached;
2996 if (!tun) {
2997 ret = -EINVAL;
2998 goto unlock;
2999 }
3000 ret = security_tun_dev_attach_queue(tun->security);
3001 if (ret < 0)
3002 goto unlock;
3003 ret = tun_attach(tun, file, false, tun->flags & IFF_NAPI,
3004 tun->flags & IFF_NAPI_FRAGS, true);
3005 } else if (ifr->ifr_flags & IFF_DETACH_QUEUE) {
3006 tun = rtnl_dereference(tfile->tun);
3007 if (!tun || !(tun->flags & IFF_MULTI_QUEUE) || tfile->detached)
3008 ret = -EINVAL;
3009 else
3010 __tun_detach(tfile, false);
3011 } else
3012 ret = -EINVAL;
3013
3014 if (ret >= 0)
3015 netdev_state_change(tun->dev);
3016
3017unlock:
3018 rtnl_unlock();
3019 return ret;
3020}
3021
3022static int tun_set_ebpf(struct tun_struct *tun, struct tun_prog __rcu **prog_p,
3023 void __user *data)
3024{
3025 struct bpf_prog *prog;
3026 int fd;
3027
3028 if (copy_from_user(&fd, data, sizeof(fd)))
3029 return -EFAULT;
3030
3031 if (fd == -1) {
3032 prog = NULL;
3033 } else {
3034 prog = bpf_prog_get_type(fd, BPF_PROG_TYPE_SOCKET_FILTER);
3035 if (IS_ERR(prog))
3036 return PTR_ERR(prog);
3037 }
3038
3039 return __tun_set_ebpf(tun, prog_p, prog);
3040}
3041
3042/* Return correct value for tun->dev->addr_len based on tun->dev->type. */
3043static unsigned char tun_get_addr_len(unsigned short type)
3044{
3045 switch (type) {
3046 case ARPHRD_IP6GRE:
3047 case ARPHRD_TUNNEL6:
3048 return sizeof(struct in6_addr);
3049 case ARPHRD_IPGRE:
3050 case ARPHRD_TUNNEL:
3051 case ARPHRD_SIT:
3052 return 4;
3053 case ARPHRD_ETHER:
3054 return ETH_ALEN;
3055 case ARPHRD_IEEE802154:
3056 case ARPHRD_IEEE802154_MONITOR:
3057 return IEEE802154_EXTENDED_ADDR_LEN;
3058 case ARPHRD_PHONET_PIPE:
3059 case ARPHRD_PPP:
3060 case ARPHRD_NONE:
3061 return 0;
3062 case ARPHRD_6LOWPAN:
3063 return EUI64_ADDR_LEN;
3064 case ARPHRD_FDDI:
3065 return FDDI_K_ALEN;
3066 case ARPHRD_HIPPI:
3067 return HIPPI_ALEN;
3068 case ARPHRD_IEEE802:
3069 return FC_ALEN;
3070 case ARPHRD_ROSE:
3071 return ROSE_ADDR_LEN;
3072 case ARPHRD_NETROM:
3073 return AX25_ADDR_LEN;
3074 case ARPHRD_LOCALTLK:
3075 return LTALK_ALEN;
3076 default:
3077 return 0;
3078 }
3079}
3080
3081static long __tun_chr_ioctl(struct file *file, unsigned int cmd,
3082 unsigned long arg, int ifreq_len)
3083{
3084 struct tun_file *tfile = file->private_data;
3085 struct net *net = sock_net(&tfile->sk);
3086 struct tun_struct *tun;
3087 void __user* argp = (void __user*)arg;
3088 unsigned int carrier;
3089 struct ifreq ifr;
3090 kuid_t owner;
3091 kgid_t group;
3092 int ifindex;
3093 int sndbuf;
3094 int vnet_hdr_sz;
3095 int le;
3096 int ret;
3097 bool do_notify = false;
3098
3099 if (cmd == TUNSETIFF || cmd == TUNSETQUEUE ||
3100 (_IOC_TYPE(cmd) == SOCK_IOC_TYPE && cmd != SIOCGSKNS)) {
3101 if (copy_from_user(&ifr, argp, ifreq_len))
3102 return -EFAULT;
3103 } else {
3104 memset(&ifr, 0, sizeof(ifr));
3105 }
3106 if (cmd == TUNGETFEATURES) {
3107 /* Currently this just means: "what IFF flags are valid?".
3108 * This is needed because we never checked for invalid flags on
3109 * TUNSETIFF.
3110 */
3111 return put_user(IFF_TUN | IFF_TAP | IFF_NO_CARRIER |
3112 TUN_FEATURES, (unsigned int __user*)argp);
3113 } else if (cmd == TUNSETQUEUE) {
3114 return tun_set_queue(file, &ifr);
3115 } else if (cmd == SIOCGSKNS) {
3116 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
3117 return -EPERM;
3118 return open_related_ns(&net->ns, get_net_ns);
3119 }
3120
3121 rtnl_lock();
3122
3123 tun = tun_get(tfile);
3124 if (cmd == TUNSETIFF) {
3125 ret = -EEXIST;
3126 if (tun)
3127 goto unlock;
3128
3129 ifr.ifr_name[IFNAMSIZ-1] = '\0';
3130
3131 ret = tun_set_iff(net, file, &ifr);
3132
3133 if (ret)
3134 goto unlock;
3135
3136 if (copy_to_user(argp, &ifr, ifreq_len))
3137 ret = -EFAULT;
3138 goto unlock;
3139 }
3140 if (cmd == TUNSETIFINDEX) {
3141 ret = -EPERM;
3142 if (tun)
3143 goto unlock;
3144
3145 ret = -EFAULT;
3146 if (copy_from_user(&ifindex, argp, sizeof(ifindex)))
3147 goto unlock;
3148 ret = -EINVAL;
3149 if (ifindex < 0)
3150 goto unlock;
3151 ret = 0;
3152 tfile->ifindex = ifindex;
3153 goto unlock;
3154 }
3155
3156 ret = -EBADFD;
3157 if (!tun)
3158 goto unlock;
3159
3160 netif_info(tun, drv, tun->dev, "tun_chr_ioctl cmd %u\n", cmd);
3161
3162 net = dev_net(tun->dev);
3163 ret = 0;
3164 switch (cmd) {
3165 case TUNGETIFF:
3166 tun_get_iff(tun, &ifr);
3167
3168 if (tfile->detached)
3169 ifr.ifr_flags |= IFF_DETACH_QUEUE;
3170 if (!tfile->socket.sk->sk_filter)
3171 ifr.ifr_flags |= IFF_NOFILTER;
3172
3173 if (copy_to_user(argp, &ifr, ifreq_len))
3174 ret = -EFAULT;
3175 break;
3176
3177 case TUNSETNOCSUM:
3178 /* Disable/Enable checksum */
3179
3180 /* [unimplemented] */
3181 netif_info(tun, drv, tun->dev, "ignored: set checksum %s\n",
3182 arg ? "disabled" : "enabled");
3183 break;
3184
3185 case TUNSETPERSIST:
3186 /* Disable/Enable persist mode. Keep an extra reference to the
3187 * module to prevent the module being unprobed.
3188 */
3189 if (arg && !(tun->flags & IFF_PERSIST)) {
3190 tun->flags |= IFF_PERSIST;
3191 __module_get(THIS_MODULE);
3192 do_notify = true;
3193 }
3194 if (!arg && (tun->flags & IFF_PERSIST)) {
3195 tun->flags &= ~IFF_PERSIST;
3196 module_put(THIS_MODULE);
3197 do_notify = true;
3198 }
3199
3200 netif_info(tun, drv, tun->dev, "persist %s\n",
3201 arg ? "enabled" : "disabled");
3202 break;
3203
3204 case TUNSETOWNER:
3205 /* Set owner of the device */
3206 owner = make_kuid(current_user_ns(), arg);
3207 if (!uid_valid(owner)) {
3208 ret = -EINVAL;
3209 break;
3210 }
3211 tun->owner = owner;
3212 do_notify = true;
3213 netif_info(tun, drv, tun->dev, "owner set to %u\n",
3214 from_kuid(&init_user_ns, tun->owner));
3215 break;
3216
3217 case TUNSETGROUP:
3218 /* Set group of the device */
3219 group = make_kgid(current_user_ns(), arg);
3220 if (!gid_valid(group)) {
3221 ret = -EINVAL;
3222 break;
3223 }
3224 tun->group = group;
3225 do_notify = true;
3226 netif_info(tun, drv, tun->dev, "group set to %u\n",
3227 from_kgid(&init_user_ns, tun->group));
3228 break;
3229
3230 case TUNSETLINK:
3231 /* Only allow setting the type when the interface is down */
3232 if (tun->dev->flags & IFF_UP) {
3233 netif_info(tun, drv, tun->dev,
3234 "Linktype set failed because interface is up\n");
3235 ret = -EBUSY;
3236 } else {
3237 ret = call_netdevice_notifiers(NETDEV_PRE_TYPE_CHANGE,
3238 tun->dev);
3239 ret = notifier_to_errno(ret);
3240 if (ret) {
3241 netif_info(tun, drv, tun->dev,
3242 "Refused to change device type\n");
3243 break;
3244 }
3245 tun->dev->type = (int) arg;
3246 tun->dev->addr_len = tun_get_addr_len(tun->dev->type);
3247 netif_info(tun, drv, tun->dev, "linktype set to %d\n",
3248 tun->dev->type);
3249 call_netdevice_notifiers(NETDEV_POST_TYPE_CHANGE,
3250 tun->dev);
3251 }
3252 break;
3253
3254 case TUNSETDEBUG:
3255 tun->msg_enable = (u32)arg;
3256 break;
3257
3258 case TUNSETOFFLOAD:
3259 ret = set_offload(tun, arg);
3260 break;
3261
3262 case TUNSETTXFILTER:
3263 /* Can be set only for TAPs */
3264 ret = -EINVAL;
3265 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3266 break;
3267 ret = update_filter(&tun->txflt, (void __user *)arg);
3268 break;
3269
3270 case SIOCGIFHWADDR:
3271 /* Get hw address */
3272 dev_get_mac_address(&ifr.ifr_hwaddr, net, tun->dev->name);
3273 if (copy_to_user(argp, &ifr, ifreq_len))
3274 ret = -EFAULT;
3275 break;
3276
3277 case SIOCSIFHWADDR:
3278 /* Set hw address */
3279 ret = dev_set_mac_address_user(tun->dev, &ifr.ifr_hwaddr, NULL);
3280 break;
3281
3282 case TUNGETSNDBUF:
3283 sndbuf = tfile->socket.sk->sk_sndbuf;
3284 if (copy_to_user(argp, &sndbuf, sizeof(sndbuf)))
3285 ret = -EFAULT;
3286 break;
3287
3288 case TUNSETSNDBUF:
3289 if (copy_from_user(&sndbuf, argp, sizeof(sndbuf))) {
3290 ret = -EFAULT;
3291 break;
3292 }
3293 if (sndbuf <= 0) {
3294 ret = -EINVAL;
3295 break;
3296 }
3297
3298 tun->sndbuf = sndbuf;
3299 tun_set_sndbuf(tun);
3300 break;
3301
3302 case TUNGETVNETHDRSZ:
3303 vnet_hdr_sz = tun->vnet_hdr_sz;
3304 if (copy_to_user(argp, &vnet_hdr_sz, sizeof(vnet_hdr_sz)))
3305 ret = -EFAULT;
3306 break;
3307
3308 case TUNSETVNETHDRSZ:
3309 if (copy_from_user(&vnet_hdr_sz, argp, sizeof(vnet_hdr_sz))) {
3310 ret = -EFAULT;
3311 break;
3312 }
3313 if (vnet_hdr_sz < (int)sizeof(struct virtio_net_hdr)) {
3314 ret = -EINVAL;
3315 break;
3316 }
3317
3318 tun->vnet_hdr_sz = vnet_hdr_sz;
3319 break;
3320
3321 case TUNGETVNETLE:
3322 le = !!(tun->flags & TUN_VNET_LE);
3323 if (put_user(le, (int __user *)argp))
3324 ret = -EFAULT;
3325 break;
3326
3327 case TUNSETVNETLE:
3328 if (get_user(le, (int __user *)argp)) {
3329 ret = -EFAULT;
3330 break;
3331 }
3332 if (le)
3333 tun->flags |= TUN_VNET_LE;
3334 else
3335 tun->flags &= ~TUN_VNET_LE;
3336 break;
3337
3338 case TUNGETVNETBE:
3339 ret = tun_get_vnet_be(tun, argp);
3340 break;
3341
3342 case TUNSETVNETBE:
3343 ret = tun_set_vnet_be(tun, argp);
3344 break;
3345
3346 case TUNATTACHFILTER:
3347 /* Can be set only for TAPs */
3348 ret = -EINVAL;
3349 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3350 break;
3351 ret = -EFAULT;
3352 if (copy_from_user(&tun->fprog, argp, sizeof(tun->fprog)))
3353 break;
3354
3355 ret = tun_attach_filter(tun);
3356 break;
3357
3358 case TUNDETACHFILTER:
3359 /* Can be set only for TAPs */
3360 ret = -EINVAL;
3361 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3362 break;
3363 ret = 0;
3364 tun_detach_filter(tun, tun->numqueues);
3365 break;
3366
3367 case TUNGETFILTER:
3368 ret = -EINVAL;
3369 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3370 break;
3371 ret = -EFAULT;
3372 if (copy_to_user(argp, &tun->fprog, sizeof(tun->fprog)))
3373 break;
3374 ret = 0;
3375 break;
3376
3377 case TUNSETSTEERINGEBPF:
3378 ret = tun_set_ebpf(tun, &tun->steering_prog, argp);
3379 break;
3380
3381 case TUNSETFILTEREBPF:
3382 ret = tun_set_ebpf(tun, &tun->filter_prog, argp);
3383 break;
3384
3385 case TUNSETCARRIER:
3386 ret = -EFAULT;
3387 if (copy_from_user(&carrier, argp, sizeof(carrier)))
3388 goto unlock;
3389
3390 ret = tun_net_change_carrier(tun->dev, (bool)carrier);
3391 break;
3392
3393 case TUNGETDEVNETNS:
3394 ret = -EPERM;
3395 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
3396 goto unlock;
3397 ret = open_related_ns(&net->ns, get_net_ns);
3398 break;
3399
3400 default:
3401 ret = -EINVAL;
3402 break;
3403 }
3404
3405 if (do_notify)
3406 netdev_state_change(tun->dev);
3407
3408unlock:
3409 rtnl_unlock();
3410 if (tun)
3411 tun_put(tun);
3412 return ret;
3413}
3414
3415static long tun_chr_ioctl(struct file *file,
3416 unsigned int cmd, unsigned long arg)
3417{
3418 return __tun_chr_ioctl(file, cmd, arg, sizeof (struct ifreq));
3419}
3420
3421#ifdef CONFIG_COMPAT
3422static long tun_chr_compat_ioctl(struct file *file,
3423 unsigned int cmd, unsigned long arg)
3424{
3425 switch (cmd) {
3426 case TUNSETIFF:
3427 case TUNGETIFF:
3428 case TUNSETTXFILTER:
3429 case TUNGETSNDBUF:
3430 case TUNSETSNDBUF:
3431 case SIOCGIFHWADDR:
3432 case SIOCSIFHWADDR:
3433 arg = (unsigned long)compat_ptr(arg);
3434 break;
3435 default:
3436 arg = (compat_ulong_t)arg;
3437 break;
3438 }
3439
3440 /*
3441 * compat_ifreq is shorter than ifreq, so we must not access beyond
3442 * the end of that structure. All fields that are used in this
3443 * driver are compatible though, we don't need to convert the
3444 * contents.
3445 */
3446 return __tun_chr_ioctl(file, cmd, arg, sizeof(struct compat_ifreq));
3447}
3448#endif /* CONFIG_COMPAT */
3449
3450static int tun_chr_fasync(int fd, struct file *file, int on)
3451{
3452 struct tun_file *tfile = file->private_data;
3453 int ret;
3454
3455 if (on) {
3456 ret = file_f_owner_allocate(file);
3457 if (ret)
3458 goto out;
3459 }
3460
3461 if ((ret = fasync_helper(fd, file, on, &tfile->fasync)) < 0)
3462 goto out;
3463
3464 if (on) {
3465 __f_setown(file, task_pid(current), PIDTYPE_TGID, 0);
3466 tfile->flags |= TUN_FASYNC;
3467 } else
3468 tfile->flags &= ~TUN_FASYNC;
3469 ret = 0;
3470out:
3471 return ret;
3472}
3473
3474static int tun_chr_open(struct inode *inode, struct file * file)
3475{
3476 struct net *net = current->nsproxy->net_ns;
3477 struct tun_file *tfile;
3478
3479 tfile = (struct tun_file *)sk_alloc(net, AF_UNSPEC, GFP_KERNEL,
3480 &tun_proto, 0);
3481 if (!tfile)
3482 return -ENOMEM;
3483 if (ptr_ring_init(&tfile->tx_ring, 0, GFP_KERNEL)) {
3484 sk_free(&tfile->sk);
3485 return -ENOMEM;
3486 }
3487
3488 mutex_init(&tfile->napi_mutex);
3489 RCU_INIT_POINTER(tfile->tun, NULL);
3490 tfile->flags = 0;
3491 tfile->ifindex = 0;
3492
3493 init_waitqueue_head(&tfile->socket.wq.wait);
3494
3495 tfile->socket.file = file;
3496 tfile->socket.ops = &tun_socket_ops;
3497
3498 sock_init_data_uid(&tfile->socket, &tfile->sk, current_fsuid());
3499
3500 tfile->sk.sk_write_space = tun_sock_write_space;
3501 tfile->sk.sk_sndbuf = INT_MAX;
3502
3503 file->private_data = tfile;
3504 INIT_LIST_HEAD(&tfile->next);
3505
3506 sock_set_flag(&tfile->sk, SOCK_ZEROCOPY);
3507
3508 /* tun groks IOCB_NOWAIT just fine, mark it as such */
3509 file->f_mode |= FMODE_NOWAIT;
3510 return 0;
3511}
3512
3513static int tun_chr_close(struct inode *inode, struct file *file)
3514{
3515 struct tun_file *tfile = file->private_data;
3516
3517 tun_detach(tfile, true);
3518
3519 return 0;
3520}
3521
3522#ifdef CONFIG_PROC_FS
3523static void tun_chr_show_fdinfo(struct seq_file *m, struct file *file)
3524{
3525 struct tun_file *tfile = file->private_data;
3526 struct tun_struct *tun;
3527 struct ifreq ifr;
3528
3529 memset(&ifr, 0, sizeof(ifr));
3530
3531 rtnl_lock();
3532 tun = tun_get(tfile);
3533 if (tun)
3534 tun_get_iff(tun, &ifr);
3535 rtnl_unlock();
3536
3537 if (tun)
3538 tun_put(tun);
3539
3540 seq_printf(m, "iff:\t%s\n", ifr.ifr_name);
3541}
3542#endif
3543
3544static const struct file_operations tun_fops = {
3545 .owner = THIS_MODULE,
3546 .read_iter = tun_chr_read_iter,
3547 .write_iter = tun_chr_write_iter,
3548 .poll = tun_chr_poll,
3549 .unlocked_ioctl = tun_chr_ioctl,
3550#ifdef CONFIG_COMPAT
3551 .compat_ioctl = tun_chr_compat_ioctl,
3552#endif
3553 .open = tun_chr_open,
3554 .release = tun_chr_close,
3555 .fasync = tun_chr_fasync,
3556#ifdef CONFIG_PROC_FS
3557 .show_fdinfo = tun_chr_show_fdinfo,
3558#endif
3559};
3560
3561static struct miscdevice tun_miscdev = {
3562 .minor = TUN_MINOR,
3563 .name = "tun",
3564 .nodename = "net/tun",
3565 .fops = &tun_fops,
3566};
3567
3568/* ethtool interface */
3569
3570static void tun_default_link_ksettings(struct net_device *dev,
3571 struct ethtool_link_ksettings *cmd)
3572{
3573 ethtool_link_ksettings_zero_link_mode(cmd, supported);
3574 ethtool_link_ksettings_zero_link_mode(cmd, advertising);
3575 cmd->base.speed = SPEED_10000;
3576 cmd->base.duplex = DUPLEX_FULL;
3577 cmd->base.port = PORT_TP;
3578 cmd->base.phy_address = 0;
3579 cmd->base.autoneg = AUTONEG_DISABLE;
3580}
3581
3582static int tun_get_link_ksettings(struct net_device *dev,
3583 struct ethtool_link_ksettings *cmd)
3584{
3585 struct tun_struct *tun = netdev_priv(dev);
3586
3587 memcpy(cmd, &tun->link_ksettings, sizeof(*cmd));
3588 return 0;
3589}
3590
3591static int tun_set_link_ksettings(struct net_device *dev,
3592 const struct ethtool_link_ksettings *cmd)
3593{
3594 struct tun_struct *tun = netdev_priv(dev);
3595
3596 memcpy(&tun->link_ksettings, cmd, sizeof(*cmd));
3597 return 0;
3598}
3599
3600static void tun_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
3601{
3602 struct tun_struct *tun = netdev_priv(dev);
3603
3604 strscpy(info->driver, DRV_NAME, sizeof(info->driver));
3605 strscpy(info->version, DRV_VERSION, sizeof(info->version));
3606
3607 switch (tun->flags & TUN_TYPE_MASK) {
3608 case IFF_TUN:
3609 strscpy(info->bus_info, "tun", sizeof(info->bus_info));
3610 break;
3611 case IFF_TAP:
3612 strscpy(info->bus_info, "tap", sizeof(info->bus_info));
3613 break;
3614 }
3615}
3616
3617static u32 tun_get_msglevel(struct net_device *dev)
3618{
3619 struct tun_struct *tun = netdev_priv(dev);
3620
3621 return tun->msg_enable;
3622}
3623
3624static void tun_set_msglevel(struct net_device *dev, u32 value)
3625{
3626 struct tun_struct *tun = netdev_priv(dev);
3627
3628 tun->msg_enable = value;
3629}
3630
3631static int tun_get_coalesce(struct net_device *dev,
3632 struct ethtool_coalesce *ec,
3633 struct kernel_ethtool_coalesce *kernel_coal,
3634 struct netlink_ext_ack *extack)
3635{
3636 struct tun_struct *tun = netdev_priv(dev);
3637
3638 ec->rx_max_coalesced_frames = tun->rx_batched;
3639
3640 return 0;
3641}
3642
3643static int tun_set_coalesce(struct net_device *dev,
3644 struct ethtool_coalesce *ec,
3645 struct kernel_ethtool_coalesce *kernel_coal,
3646 struct netlink_ext_ack *extack)
3647{
3648 struct tun_struct *tun = netdev_priv(dev);
3649
3650 if (ec->rx_max_coalesced_frames > NAPI_POLL_WEIGHT)
3651 tun->rx_batched = NAPI_POLL_WEIGHT;
3652 else
3653 tun->rx_batched = ec->rx_max_coalesced_frames;
3654
3655 return 0;
3656}
3657
3658static void tun_get_channels(struct net_device *dev,
3659 struct ethtool_channels *channels)
3660{
3661 struct tun_struct *tun = netdev_priv(dev);
3662
3663 channels->combined_count = tun->numqueues;
3664 channels->max_combined = tun->flags & IFF_MULTI_QUEUE ? MAX_TAP_QUEUES : 1;
3665}
3666
3667static const struct ethtool_ops tun_ethtool_ops = {
3668 .supported_coalesce_params = ETHTOOL_COALESCE_RX_MAX_FRAMES,
3669 .get_drvinfo = tun_get_drvinfo,
3670 .get_msglevel = tun_get_msglevel,
3671 .set_msglevel = tun_set_msglevel,
3672 .get_link = ethtool_op_get_link,
3673 .get_channels = tun_get_channels,
3674 .get_ts_info = ethtool_op_get_ts_info,
3675 .get_coalesce = tun_get_coalesce,
3676 .set_coalesce = tun_set_coalesce,
3677 .get_link_ksettings = tun_get_link_ksettings,
3678 .set_link_ksettings = tun_set_link_ksettings,
3679};
3680
3681static int tun_queue_resize(struct tun_struct *tun)
3682{
3683 struct net_device *dev = tun->dev;
3684 struct tun_file *tfile;
3685 struct ptr_ring **rings;
3686 int n = tun->numqueues + tun->numdisabled;
3687 int ret, i;
3688
3689 rings = kmalloc_array(n, sizeof(*rings), GFP_KERNEL);
3690 if (!rings)
3691 return -ENOMEM;
3692
3693 for (i = 0; i < tun->numqueues; i++) {
3694 tfile = rtnl_dereference(tun->tfiles[i]);
3695 rings[i] = &tfile->tx_ring;
3696 }
3697 list_for_each_entry(tfile, &tun->disabled, next)
3698 rings[i++] = &tfile->tx_ring;
3699
3700 ret = ptr_ring_resize_multiple_bh(rings, n,
3701 dev->tx_queue_len, GFP_KERNEL,
3702 tun_ptr_free);
3703
3704 kfree(rings);
3705 return ret;
3706}
3707
3708static int tun_device_event(struct notifier_block *unused,
3709 unsigned long event, void *ptr)
3710{
3711 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
3712 struct tun_struct *tun = netdev_priv(dev);
3713 int i;
3714
3715 if (dev->rtnl_link_ops != &tun_link_ops)
3716 return NOTIFY_DONE;
3717
3718 switch (event) {
3719 case NETDEV_CHANGE_TX_QUEUE_LEN:
3720 if (tun_queue_resize(tun))
3721 return NOTIFY_BAD;
3722 break;
3723 case NETDEV_UP:
3724 for (i = 0; i < tun->numqueues; i++) {
3725 struct tun_file *tfile;
3726
3727 tfile = rtnl_dereference(tun->tfiles[i]);
3728 tfile->socket.sk->sk_write_space(tfile->socket.sk);
3729 }
3730 break;
3731 default:
3732 break;
3733 }
3734
3735 return NOTIFY_DONE;
3736}
3737
3738static struct notifier_block tun_notifier_block __read_mostly = {
3739 .notifier_call = tun_device_event,
3740};
3741
3742static int __init tun_init(void)
3743{
3744 int ret = 0;
3745
3746 pr_info("%s, %s\n", DRV_DESCRIPTION, DRV_VERSION);
3747
3748 ret = rtnl_link_register(&tun_link_ops);
3749 if (ret) {
3750 pr_err("Can't register link_ops\n");
3751 goto err_linkops;
3752 }
3753
3754 ret = misc_register(&tun_miscdev);
3755 if (ret) {
3756 pr_err("Can't register misc device %d\n", TUN_MINOR);
3757 goto err_misc;
3758 }
3759
3760 ret = register_netdevice_notifier(&tun_notifier_block);
3761 if (ret) {
3762 pr_err("Can't register netdevice notifier\n");
3763 goto err_notifier;
3764 }
3765
3766 return 0;
3767
3768err_notifier:
3769 misc_deregister(&tun_miscdev);
3770err_misc:
3771 rtnl_link_unregister(&tun_link_ops);
3772err_linkops:
3773 return ret;
3774}
3775
3776static void __exit tun_cleanup(void)
3777{
3778 misc_deregister(&tun_miscdev);
3779 rtnl_link_unregister(&tun_link_ops);
3780 unregister_netdevice_notifier(&tun_notifier_block);
3781}
3782
3783/* Get an underlying socket object from tun file. Returns error unless file is
3784 * attached to a device. The returned object works like a packet socket, it
3785 * can be used for sock_sendmsg/sock_recvmsg. The caller is responsible for
3786 * holding a reference to the file for as long as the socket is in use. */
3787struct socket *tun_get_socket(struct file *file)
3788{
3789 struct tun_file *tfile;
3790 if (file->f_op != &tun_fops)
3791 return ERR_PTR(-EINVAL);
3792 tfile = file->private_data;
3793 if (!tfile)
3794 return ERR_PTR(-EBADFD);
3795 return &tfile->socket;
3796}
3797EXPORT_SYMBOL_GPL(tun_get_socket);
3798
3799struct ptr_ring *tun_get_tx_ring(struct file *file)
3800{
3801 struct tun_file *tfile;
3802
3803 if (file->f_op != &tun_fops)
3804 return ERR_PTR(-EINVAL);
3805 tfile = file->private_data;
3806 if (!tfile)
3807 return ERR_PTR(-EBADFD);
3808 return &tfile->tx_ring;
3809}
3810EXPORT_SYMBOL_GPL(tun_get_tx_ring);
3811
3812module_init(tun_init);
3813module_exit(tun_cleanup);
3814MODULE_DESCRIPTION(DRV_DESCRIPTION);
3815MODULE_AUTHOR(DRV_COPYRIGHT);
3816MODULE_LICENSE("GPL");
3817MODULE_ALIAS_MISCDEV(TUN_MINOR);
3818MODULE_ALIAS("devname:net/tun");
1// SPDX-License-Identifier: GPL-2.0-or-later
2/*
3 * TUN - Universal TUN/TAP device driver.
4 * Copyright (C) 1999-2002 Maxim Krasnyansky <maxk@qualcomm.com>
5 *
6 * $Id: tun.c,v 1.15 2002/03/01 02:44:24 maxk Exp $
7 */
8
9/*
10 * Changes:
11 *
12 * Mike Kershaw <dragorn@kismetwireless.net> 2005/08/14
13 * Add TUNSETLINK ioctl to set the link encapsulation
14 *
15 * Mark Smith <markzzzsmith@yahoo.com.au>
16 * Use eth_random_addr() for tap MAC address.
17 *
18 * Harald Roelle <harald.roelle@ifi.lmu.de> 2004/04/20
19 * Fixes in packet dropping, queue length setting and queue wakeup.
20 * Increased default tx queue length.
21 * Added ethtool API.
22 * Minor cleanups
23 *
24 * Daniel Podlejski <underley@underley.eu.org>
25 * Modifications for 2.3.99-pre5 kernel.
26 */
27
28#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
29
30#define DRV_NAME "tun"
31#define DRV_VERSION "1.6"
32#define DRV_DESCRIPTION "Universal TUN/TAP device driver"
33#define DRV_COPYRIGHT "(C) 1999-2004 Max Krasnyansky <maxk@qualcomm.com>"
34
35#include <linux/module.h>
36#include <linux/errno.h>
37#include <linux/kernel.h>
38#include <linux/sched/signal.h>
39#include <linux/major.h>
40#include <linux/slab.h>
41#include <linux/poll.h>
42#include <linux/fcntl.h>
43#include <linux/init.h>
44#include <linux/skbuff.h>
45#include <linux/netdevice.h>
46#include <linux/etherdevice.h>
47#include <linux/miscdevice.h>
48#include <linux/ethtool.h>
49#include <linux/rtnetlink.h>
50#include <linux/compat.h>
51#include <linux/if.h>
52#include <linux/if_arp.h>
53#include <linux/if_ether.h>
54#include <linux/if_tun.h>
55#include <linux/if_vlan.h>
56#include <linux/crc32.h>
57#include <linux/nsproxy.h>
58#include <linux/virtio_net.h>
59#include <linux/rcupdate.h>
60#include <net/net_namespace.h>
61#include <net/netns/generic.h>
62#include <net/rtnetlink.h>
63#include <net/sock.h>
64#include <net/xdp.h>
65#include <net/ip_tunnels.h>
66#include <linux/seq_file.h>
67#include <linux/uio.h>
68#include <linux/skb_array.h>
69#include <linux/bpf.h>
70#include <linux/bpf_trace.h>
71#include <linux/mutex.h>
72#include <linux/ieee802154.h>
73#include <linux/if_ltalk.h>
74#include <uapi/linux/if_fddi.h>
75#include <uapi/linux/if_hippi.h>
76#include <uapi/linux/if_fc.h>
77#include <net/ax25.h>
78#include <net/rose.h>
79#include <net/6lowpan.h>
80
81#include <linux/uaccess.h>
82#include <linux/proc_fs.h>
83
84static void tun_default_link_ksettings(struct net_device *dev,
85 struct ethtool_link_ksettings *cmd);
86
87#define TUN_RX_PAD (NET_IP_ALIGN + NET_SKB_PAD)
88
89/* TUN device flags */
90
91/* IFF_ATTACH_QUEUE is never stored in device flags,
92 * overload it to mean fasync when stored there.
93 */
94#define TUN_FASYNC IFF_ATTACH_QUEUE
95/* High bits in flags field are unused. */
96#define TUN_VNET_LE 0x80000000
97#define TUN_VNET_BE 0x40000000
98
99#define TUN_FEATURES (IFF_NO_PI | IFF_ONE_QUEUE | IFF_VNET_HDR | \
100 IFF_MULTI_QUEUE | IFF_NAPI | IFF_NAPI_FRAGS)
101
102#define GOODCOPY_LEN 128
103
104#define FLT_EXACT_COUNT 8
105struct tap_filter {
106 unsigned int count; /* Number of addrs. Zero means disabled */
107 u32 mask[2]; /* Mask of the hashed addrs */
108 unsigned char addr[FLT_EXACT_COUNT][ETH_ALEN];
109};
110
111/* MAX_TAP_QUEUES 256 is chosen to allow rx/tx queues to be equal
112 * to max number of VCPUs in guest. */
113#define MAX_TAP_QUEUES 256
114#define MAX_TAP_FLOWS 4096
115
116#define TUN_FLOW_EXPIRE (3 * HZ)
117
118/* A tun_file connects an open character device to a tuntap netdevice. It
119 * also contains all socket related structures (except sock_fprog and tap_filter)
120 * to serve as one transmit queue for tuntap device. The sock_fprog and
121 * tap_filter were kept in tun_struct since they were used for filtering for the
122 * netdevice not for a specific queue (at least I didn't see the requirement for
123 * this).
124 *
125 * RCU usage:
126 * The tun_file and tun_struct are loosely coupled, the pointer from one to the
127 * other can only be read while rcu_read_lock or rtnl_lock is held.
128 */
129struct tun_file {
130 struct sock sk;
131 struct socket socket;
132 struct tun_struct __rcu *tun;
133 struct fasync_struct *fasync;
134 /* only used for fasnyc */
135 unsigned int flags;
136 union {
137 u16 queue_index;
138 unsigned int ifindex;
139 };
140 struct napi_struct napi;
141 bool napi_enabled;
142 bool napi_frags_enabled;
143 struct mutex napi_mutex; /* Protects access to the above napi */
144 struct list_head next;
145 struct tun_struct *detached;
146 struct ptr_ring tx_ring;
147 struct xdp_rxq_info xdp_rxq;
148};
149
150struct tun_page {
151 struct page *page;
152 int count;
153};
154
155struct tun_flow_entry {
156 struct hlist_node hash_link;
157 struct rcu_head rcu;
158 struct tun_struct *tun;
159
160 u32 rxhash;
161 u32 rps_rxhash;
162 int queue_index;
163 unsigned long updated ____cacheline_aligned_in_smp;
164};
165
166#define TUN_NUM_FLOW_ENTRIES 1024
167#define TUN_MASK_FLOW_ENTRIES (TUN_NUM_FLOW_ENTRIES - 1)
168
169struct tun_prog {
170 struct rcu_head rcu;
171 struct bpf_prog *prog;
172};
173
174/* Since the socket were moved to tun_file, to preserve the behavior of persist
175 * device, socket filter, sndbuf and vnet header size were restore when the
176 * file were attached to a persist device.
177 */
178struct tun_struct {
179 struct tun_file __rcu *tfiles[MAX_TAP_QUEUES];
180 unsigned int numqueues;
181 unsigned int flags;
182 kuid_t owner;
183 kgid_t group;
184
185 struct net_device *dev;
186 netdev_features_t set_features;
187#define TUN_USER_FEATURES (NETIF_F_HW_CSUM|NETIF_F_TSO_ECN|NETIF_F_TSO| \
188 NETIF_F_TSO6)
189
190 int align;
191 int vnet_hdr_sz;
192 int sndbuf;
193 struct tap_filter txflt;
194 struct sock_fprog fprog;
195 /* protected by rtnl lock */
196 bool filter_attached;
197 u32 msg_enable;
198 spinlock_t lock;
199 struct hlist_head flows[TUN_NUM_FLOW_ENTRIES];
200 struct timer_list flow_gc_timer;
201 unsigned long ageing_time;
202 unsigned int numdisabled;
203 struct list_head disabled;
204 void *security;
205 u32 flow_count;
206 u32 rx_batched;
207 atomic_long_t rx_frame_errors;
208 struct bpf_prog __rcu *xdp_prog;
209 struct tun_prog __rcu *steering_prog;
210 struct tun_prog __rcu *filter_prog;
211 struct ethtool_link_ksettings link_ksettings;
212};
213
214struct veth {
215 __be16 h_vlan_proto;
216 __be16 h_vlan_TCI;
217};
218
219static int tun_napi_receive(struct napi_struct *napi, int budget)
220{
221 struct tun_file *tfile = container_of(napi, struct tun_file, napi);
222 struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
223 struct sk_buff_head process_queue;
224 struct sk_buff *skb;
225 int received = 0;
226
227 __skb_queue_head_init(&process_queue);
228
229 spin_lock(&queue->lock);
230 skb_queue_splice_tail_init(queue, &process_queue);
231 spin_unlock(&queue->lock);
232
233 while (received < budget && (skb = __skb_dequeue(&process_queue))) {
234 napi_gro_receive(napi, skb);
235 ++received;
236 }
237
238 if (!skb_queue_empty(&process_queue)) {
239 spin_lock(&queue->lock);
240 skb_queue_splice(&process_queue, queue);
241 spin_unlock(&queue->lock);
242 }
243
244 return received;
245}
246
247static int tun_napi_poll(struct napi_struct *napi, int budget)
248{
249 unsigned int received;
250
251 received = tun_napi_receive(napi, budget);
252
253 if (received < budget)
254 napi_complete_done(napi, received);
255
256 return received;
257}
258
259static void tun_napi_init(struct tun_struct *tun, struct tun_file *tfile,
260 bool napi_en, bool napi_frags)
261{
262 tfile->napi_enabled = napi_en;
263 tfile->napi_frags_enabled = napi_en && napi_frags;
264 if (napi_en) {
265 netif_tx_napi_add(tun->dev, &tfile->napi, tun_napi_poll,
266 NAPI_POLL_WEIGHT);
267 napi_enable(&tfile->napi);
268 }
269}
270
271static void tun_napi_disable(struct tun_file *tfile)
272{
273 if (tfile->napi_enabled)
274 napi_disable(&tfile->napi);
275}
276
277static void tun_napi_del(struct tun_file *tfile)
278{
279 if (tfile->napi_enabled)
280 netif_napi_del(&tfile->napi);
281}
282
283static bool tun_napi_frags_enabled(const struct tun_file *tfile)
284{
285 return tfile->napi_frags_enabled;
286}
287
288#ifdef CONFIG_TUN_VNET_CROSS_LE
289static inline bool tun_legacy_is_little_endian(struct tun_struct *tun)
290{
291 return tun->flags & TUN_VNET_BE ? false :
292 virtio_legacy_is_little_endian();
293}
294
295static long tun_get_vnet_be(struct tun_struct *tun, int __user *argp)
296{
297 int be = !!(tun->flags & TUN_VNET_BE);
298
299 if (put_user(be, argp))
300 return -EFAULT;
301
302 return 0;
303}
304
305static long tun_set_vnet_be(struct tun_struct *tun, int __user *argp)
306{
307 int be;
308
309 if (get_user(be, argp))
310 return -EFAULT;
311
312 if (be)
313 tun->flags |= TUN_VNET_BE;
314 else
315 tun->flags &= ~TUN_VNET_BE;
316
317 return 0;
318}
319#else
320static inline bool tun_legacy_is_little_endian(struct tun_struct *tun)
321{
322 return virtio_legacy_is_little_endian();
323}
324
325static long tun_get_vnet_be(struct tun_struct *tun, int __user *argp)
326{
327 return -EINVAL;
328}
329
330static long tun_set_vnet_be(struct tun_struct *tun, int __user *argp)
331{
332 return -EINVAL;
333}
334#endif /* CONFIG_TUN_VNET_CROSS_LE */
335
336static inline bool tun_is_little_endian(struct tun_struct *tun)
337{
338 return tun->flags & TUN_VNET_LE ||
339 tun_legacy_is_little_endian(tun);
340}
341
342static inline u16 tun16_to_cpu(struct tun_struct *tun, __virtio16 val)
343{
344 return __virtio16_to_cpu(tun_is_little_endian(tun), val);
345}
346
347static inline __virtio16 cpu_to_tun16(struct tun_struct *tun, u16 val)
348{
349 return __cpu_to_virtio16(tun_is_little_endian(tun), val);
350}
351
352static inline u32 tun_hashfn(u32 rxhash)
353{
354 return rxhash & TUN_MASK_FLOW_ENTRIES;
355}
356
357static struct tun_flow_entry *tun_flow_find(struct hlist_head *head, u32 rxhash)
358{
359 struct tun_flow_entry *e;
360
361 hlist_for_each_entry_rcu(e, head, hash_link) {
362 if (e->rxhash == rxhash)
363 return e;
364 }
365 return NULL;
366}
367
368static struct tun_flow_entry *tun_flow_create(struct tun_struct *tun,
369 struct hlist_head *head,
370 u32 rxhash, u16 queue_index)
371{
372 struct tun_flow_entry *e = kmalloc(sizeof(*e), GFP_ATOMIC);
373
374 if (e) {
375 netif_info(tun, tx_queued, tun->dev,
376 "create flow: hash %u index %u\n",
377 rxhash, queue_index);
378 e->updated = jiffies;
379 e->rxhash = rxhash;
380 e->rps_rxhash = 0;
381 e->queue_index = queue_index;
382 e->tun = tun;
383 hlist_add_head_rcu(&e->hash_link, head);
384 ++tun->flow_count;
385 }
386 return e;
387}
388
389static void tun_flow_delete(struct tun_struct *tun, struct tun_flow_entry *e)
390{
391 netif_info(tun, tx_queued, tun->dev, "delete flow: hash %u index %u\n",
392 e->rxhash, e->queue_index);
393 hlist_del_rcu(&e->hash_link);
394 kfree_rcu(e, rcu);
395 --tun->flow_count;
396}
397
398static void tun_flow_flush(struct tun_struct *tun)
399{
400 int i;
401
402 spin_lock_bh(&tun->lock);
403 for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
404 struct tun_flow_entry *e;
405 struct hlist_node *n;
406
407 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link)
408 tun_flow_delete(tun, e);
409 }
410 spin_unlock_bh(&tun->lock);
411}
412
413static void tun_flow_delete_by_queue(struct tun_struct *tun, u16 queue_index)
414{
415 int i;
416
417 spin_lock_bh(&tun->lock);
418 for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
419 struct tun_flow_entry *e;
420 struct hlist_node *n;
421
422 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
423 if (e->queue_index == queue_index)
424 tun_flow_delete(tun, e);
425 }
426 }
427 spin_unlock_bh(&tun->lock);
428}
429
430static void tun_flow_cleanup(struct timer_list *t)
431{
432 struct tun_struct *tun = from_timer(tun, t, flow_gc_timer);
433 unsigned long delay = tun->ageing_time;
434 unsigned long next_timer = jiffies + delay;
435 unsigned long count = 0;
436 int i;
437
438 spin_lock(&tun->lock);
439 for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
440 struct tun_flow_entry *e;
441 struct hlist_node *n;
442
443 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
444 unsigned long this_timer;
445
446 this_timer = e->updated + delay;
447 if (time_before_eq(this_timer, jiffies)) {
448 tun_flow_delete(tun, e);
449 continue;
450 }
451 count++;
452 if (time_before(this_timer, next_timer))
453 next_timer = this_timer;
454 }
455 }
456
457 if (count)
458 mod_timer(&tun->flow_gc_timer, round_jiffies_up(next_timer));
459 spin_unlock(&tun->lock);
460}
461
462static void tun_flow_update(struct tun_struct *tun, u32 rxhash,
463 struct tun_file *tfile)
464{
465 struct hlist_head *head;
466 struct tun_flow_entry *e;
467 unsigned long delay = tun->ageing_time;
468 u16 queue_index = tfile->queue_index;
469
470 head = &tun->flows[tun_hashfn(rxhash)];
471
472 rcu_read_lock();
473
474 e = tun_flow_find(head, rxhash);
475 if (likely(e)) {
476 /* TODO: keep queueing to old queue until it's empty? */
477 if (READ_ONCE(e->queue_index) != queue_index)
478 WRITE_ONCE(e->queue_index, queue_index);
479 if (e->updated != jiffies)
480 e->updated = jiffies;
481 sock_rps_record_flow_hash(e->rps_rxhash);
482 } else {
483 spin_lock_bh(&tun->lock);
484 if (!tun_flow_find(head, rxhash) &&
485 tun->flow_count < MAX_TAP_FLOWS)
486 tun_flow_create(tun, head, rxhash, queue_index);
487
488 if (!timer_pending(&tun->flow_gc_timer))
489 mod_timer(&tun->flow_gc_timer,
490 round_jiffies_up(jiffies + delay));
491 spin_unlock_bh(&tun->lock);
492 }
493
494 rcu_read_unlock();
495}
496
497/* Save the hash received in the stack receive path and update the
498 * flow_hash table accordingly.
499 */
500static inline void tun_flow_save_rps_rxhash(struct tun_flow_entry *e, u32 hash)
501{
502 if (unlikely(e->rps_rxhash != hash))
503 e->rps_rxhash = hash;
504}
505
506/* We try to identify a flow through its rxhash. The reason that
507 * we do not check rxq no. is because some cards(e.g 82599), chooses
508 * the rxq based on the txq where the last packet of the flow comes. As
509 * the userspace application move between processors, we may get a
510 * different rxq no. here.
511 */
512static u16 tun_automq_select_queue(struct tun_struct *tun, struct sk_buff *skb)
513{
514 struct tun_flow_entry *e;
515 u32 txq = 0;
516 u32 numqueues = 0;
517
518 numqueues = READ_ONCE(tun->numqueues);
519
520 txq = __skb_get_hash_symmetric(skb);
521 e = tun_flow_find(&tun->flows[tun_hashfn(txq)], txq);
522 if (e) {
523 tun_flow_save_rps_rxhash(e, txq);
524 txq = e->queue_index;
525 } else {
526 /* use multiply and shift instead of expensive divide */
527 txq = ((u64)txq * numqueues) >> 32;
528 }
529
530 return txq;
531}
532
533static u16 tun_ebpf_select_queue(struct tun_struct *tun, struct sk_buff *skb)
534{
535 struct tun_prog *prog;
536 u32 numqueues;
537 u16 ret = 0;
538
539 numqueues = READ_ONCE(tun->numqueues);
540 if (!numqueues)
541 return 0;
542
543 prog = rcu_dereference(tun->steering_prog);
544 if (prog)
545 ret = bpf_prog_run_clear_cb(prog->prog, skb);
546
547 return ret % numqueues;
548}
549
550static u16 tun_select_queue(struct net_device *dev, struct sk_buff *skb,
551 struct net_device *sb_dev)
552{
553 struct tun_struct *tun = netdev_priv(dev);
554 u16 ret;
555
556 rcu_read_lock();
557 if (rcu_dereference(tun->steering_prog))
558 ret = tun_ebpf_select_queue(tun, skb);
559 else
560 ret = tun_automq_select_queue(tun, skb);
561 rcu_read_unlock();
562
563 return ret;
564}
565
566static inline bool tun_not_capable(struct tun_struct *tun)
567{
568 const struct cred *cred = current_cred();
569 struct net *net = dev_net(tun->dev);
570
571 return ((uid_valid(tun->owner) && !uid_eq(cred->euid, tun->owner)) ||
572 (gid_valid(tun->group) && !in_egroup_p(tun->group))) &&
573 !ns_capable(net->user_ns, CAP_NET_ADMIN);
574}
575
576static void tun_set_real_num_queues(struct tun_struct *tun)
577{
578 netif_set_real_num_tx_queues(tun->dev, tun->numqueues);
579 netif_set_real_num_rx_queues(tun->dev, tun->numqueues);
580}
581
582static void tun_disable_queue(struct tun_struct *tun, struct tun_file *tfile)
583{
584 tfile->detached = tun;
585 list_add_tail(&tfile->next, &tun->disabled);
586 ++tun->numdisabled;
587}
588
589static struct tun_struct *tun_enable_queue(struct tun_file *tfile)
590{
591 struct tun_struct *tun = tfile->detached;
592
593 tfile->detached = NULL;
594 list_del_init(&tfile->next);
595 --tun->numdisabled;
596 return tun;
597}
598
599void tun_ptr_free(void *ptr)
600{
601 if (!ptr)
602 return;
603 if (tun_is_xdp_frame(ptr)) {
604 struct xdp_frame *xdpf = tun_ptr_to_xdp(ptr);
605
606 xdp_return_frame(xdpf);
607 } else {
608 __skb_array_destroy_skb(ptr);
609 }
610}
611EXPORT_SYMBOL_GPL(tun_ptr_free);
612
613static void tun_queue_purge(struct tun_file *tfile)
614{
615 void *ptr;
616
617 while ((ptr = ptr_ring_consume(&tfile->tx_ring)) != NULL)
618 tun_ptr_free(ptr);
619
620 skb_queue_purge(&tfile->sk.sk_write_queue);
621 skb_queue_purge(&tfile->sk.sk_error_queue);
622}
623
624static void __tun_detach(struct tun_file *tfile, bool clean)
625{
626 struct tun_file *ntfile;
627 struct tun_struct *tun;
628
629 tun = rtnl_dereference(tfile->tun);
630
631 if (tun && clean) {
632 tun_napi_disable(tfile);
633 tun_napi_del(tfile);
634 }
635
636 if (tun && !tfile->detached) {
637 u16 index = tfile->queue_index;
638 BUG_ON(index >= tun->numqueues);
639
640 rcu_assign_pointer(tun->tfiles[index],
641 tun->tfiles[tun->numqueues - 1]);
642 ntfile = rtnl_dereference(tun->tfiles[index]);
643 ntfile->queue_index = index;
644 rcu_assign_pointer(tun->tfiles[tun->numqueues - 1],
645 NULL);
646
647 --tun->numqueues;
648 if (clean) {
649 RCU_INIT_POINTER(tfile->tun, NULL);
650 sock_put(&tfile->sk);
651 } else
652 tun_disable_queue(tun, tfile);
653
654 synchronize_net();
655 tun_flow_delete_by_queue(tun, tun->numqueues + 1);
656 /* Drop read queue */
657 tun_queue_purge(tfile);
658 tun_set_real_num_queues(tun);
659 } else if (tfile->detached && clean) {
660 tun = tun_enable_queue(tfile);
661 sock_put(&tfile->sk);
662 }
663
664 if (clean) {
665 if (tun && tun->numqueues == 0 && tun->numdisabled == 0) {
666 netif_carrier_off(tun->dev);
667
668 if (!(tun->flags & IFF_PERSIST) &&
669 tun->dev->reg_state == NETREG_REGISTERED)
670 unregister_netdevice(tun->dev);
671 }
672 if (tun)
673 xdp_rxq_info_unreg(&tfile->xdp_rxq);
674 ptr_ring_cleanup(&tfile->tx_ring, tun_ptr_free);
675 sock_put(&tfile->sk);
676 }
677}
678
679static void tun_detach(struct tun_file *tfile, bool clean)
680{
681 struct tun_struct *tun;
682 struct net_device *dev;
683
684 rtnl_lock();
685 tun = rtnl_dereference(tfile->tun);
686 dev = tun ? tun->dev : NULL;
687 __tun_detach(tfile, clean);
688 if (dev)
689 netdev_state_change(dev);
690 rtnl_unlock();
691}
692
693static void tun_detach_all(struct net_device *dev)
694{
695 struct tun_struct *tun = netdev_priv(dev);
696 struct tun_file *tfile, *tmp;
697 int i, n = tun->numqueues;
698
699 for (i = 0; i < n; i++) {
700 tfile = rtnl_dereference(tun->tfiles[i]);
701 BUG_ON(!tfile);
702 tun_napi_disable(tfile);
703 tfile->socket.sk->sk_shutdown = RCV_SHUTDOWN;
704 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
705 RCU_INIT_POINTER(tfile->tun, NULL);
706 --tun->numqueues;
707 }
708 list_for_each_entry(tfile, &tun->disabled, next) {
709 tfile->socket.sk->sk_shutdown = RCV_SHUTDOWN;
710 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
711 RCU_INIT_POINTER(tfile->tun, NULL);
712 }
713 BUG_ON(tun->numqueues != 0);
714
715 synchronize_net();
716 for (i = 0; i < n; i++) {
717 tfile = rtnl_dereference(tun->tfiles[i]);
718 tun_napi_del(tfile);
719 /* Drop read queue */
720 tun_queue_purge(tfile);
721 xdp_rxq_info_unreg(&tfile->xdp_rxq);
722 sock_put(&tfile->sk);
723 }
724 list_for_each_entry_safe(tfile, tmp, &tun->disabled, next) {
725 tun_enable_queue(tfile);
726 tun_queue_purge(tfile);
727 xdp_rxq_info_unreg(&tfile->xdp_rxq);
728 sock_put(&tfile->sk);
729 }
730 BUG_ON(tun->numdisabled != 0);
731
732 if (tun->flags & IFF_PERSIST)
733 module_put(THIS_MODULE);
734}
735
736static int tun_attach(struct tun_struct *tun, struct file *file,
737 bool skip_filter, bool napi, bool napi_frags,
738 bool publish_tun)
739{
740 struct tun_file *tfile = file->private_data;
741 struct net_device *dev = tun->dev;
742 int err;
743
744 err = security_tun_dev_attach(tfile->socket.sk, tun->security);
745 if (err < 0)
746 goto out;
747
748 err = -EINVAL;
749 if (rtnl_dereference(tfile->tun) && !tfile->detached)
750 goto out;
751
752 err = -EBUSY;
753 if (!(tun->flags & IFF_MULTI_QUEUE) && tun->numqueues == 1)
754 goto out;
755
756 err = -E2BIG;
757 if (!tfile->detached &&
758 tun->numqueues + tun->numdisabled == MAX_TAP_QUEUES)
759 goto out;
760
761 err = 0;
762
763 /* Re-attach the filter to persist device */
764 if (!skip_filter && (tun->filter_attached == true)) {
765 lock_sock(tfile->socket.sk);
766 err = sk_attach_filter(&tun->fprog, tfile->socket.sk);
767 release_sock(tfile->socket.sk);
768 if (!err)
769 goto out;
770 }
771
772 if (!tfile->detached &&
773 ptr_ring_resize(&tfile->tx_ring, dev->tx_queue_len,
774 GFP_KERNEL, tun_ptr_free)) {
775 err = -ENOMEM;
776 goto out;
777 }
778
779 tfile->queue_index = tun->numqueues;
780 tfile->socket.sk->sk_shutdown &= ~RCV_SHUTDOWN;
781
782 if (tfile->detached) {
783 /* Re-attach detached tfile, updating XDP queue_index */
784 WARN_ON(!xdp_rxq_info_is_reg(&tfile->xdp_rxq));
785
786 if (tfile->xdp_rxq.queue_index != tfile->queue_index)
787 tfile->xdp_rxq.queue_index = tfile->queue_index;
788 } else {
789 /* Setup XDP RX-queue info, for new tfile getting attached */
790 err = xdp_rxq_info_reg(&tfile->xdp_rxq,
791 tun->dev, tfile->queue_index, 0);
792 if (err < 0)
793 goto out;
794 err = xdp_rxq_info_reg_mem_model(&tfile->xdp_rxq,
795 MEM_TYPE_PAGE_SHARED, NULL);
796 if (err < 0) {
797 xdp_rxq_info_unreg(&tfile->xdp_rxq);
798 goto out;
799 }
800 err = 0;
801 }
802
803 if (tfile->detached) {
804 tun_enable_queue(tfile);
805 } else {
806 sock_hold(&tfile->sk);
807 tun_napi_init(tun, tfile, napi, napi_frags);
808 }
809
810 if (rtnl_dereference(tun->xdp_prog))
811 sock_set_flag(&tfile->sk, SOCK_XDP);
812
813 /* device is allowed to go away first, so no need to hold extra
814 * refcnt.
815 */
816
817 /* Publish tfile->tun and tun->tfiles only after we've fully
818 * initialized tfile; otherwise we risk using half-initialized
819 * object.
820 */
821 if (publish_tun)
822 rcu_assign_pointer(tfile->tun, tun);
823 rcu_assign_pointer(tun->tfiles[tun->numqueues], tfile);
824 tun->numqueues++;
825 tun_set_real_num_queues(tun);
826out:
827 return err;
828}
829
830static struct tun_struct *tun_get(struct tun_file *tfile)
831{
832 struct tun_struct *tun;
833
834 rcu_read_lock();
835 tun = rcu_dereference(tfile->tun);
836 if (tun)
837 dev_hold(tun->dev);
838 rcu_read_unlock();
839
840 return tun;
841}
842
843static void tun_put(struct tun_struct *tun)
844{
845 dev_put(tun->dev);
846}
847
848/* TAP filtering */
849static void addr_hash_set(u32 *mask, const u8 *addr)
850{
851 int n = ether_crc(ETH_ALEN, addr) >> 26;
852 mask[n >> 5] |= (1 << (n & 31));
853}
854
855static unsigned int addr_hash_test(const u32 *mask, const u8 *addr)
856{
857 int n = ether_crc(ETH_ALEN, addr) >> 26;
858 return mask[n >> 5] & (1 << (n & 31));
859}
860
861static int update_filter(struct tap_filter *filter, void __user *arg)
862{
863 struct { u8 u[ETH_ALEN]; } *addr;
864 struct tun_filter uf;
865 int err, alen, n, nexact;
866
867 if (copy_from_user(&uf, arg, sizeof(uf)))
868 return -EFAULT;
869
870 if (!uf.count) {
871 /* Disabled */
872 filter->count = 0;
873 return 0;
874 }
875
876 alen = ETH_ALEN * uf.count;
877 addr = memdup_user(arg + sizeof(uf), alen);
878 if (IS_ERR(addr))
879 return PTR_ERR(addr);
880
881 /* The filter is updated without holding any locks. Which is
882 * perfectly safe. We disable it first and in the worst
883 * case we'll accept a few undesired packets. */
884 filter->count = 0;
885 wmb();
886
887 /* Use first set of addresses as an exact filter */
888 for (n = 0; n < uf.count && n < FLT_EXACT_COUNT; n++)
889 memcpy(filter->addr[n], addr[n].u, ETH_ALEN);
890
891 nexact = n;
892
893 /* Remaining multicast addresses are hashed,
894 * unicast will leave the filter disabled. */
895 memset(filter->mask, 0, sizeof(filter->mask));
896 for (; n < uf.count; n++) {
897 if (!is_multicast_ether_addr(addr[n].u)) {
898 err = 0; /* no filter */
899 goto free_addr;
900 }
901 addr_hash_set(filter->mask, addr[n].u);
902 }
903
904 /* For ALLMULTI just set the mask to all ones.
905 * This overrides the mask populated above. */
906 if ((uf.flags & TUN_FLT_ALLMULTI))
907 memset(filter->mask, ~0, sizeof(filter->mask));
908
909 /* Now enable the filter */
910 wmb();
911 filter->count = nexact;
912
913 /* Return the number of exact filters */
914 err = nexact;
915free_addr:
916 kfree(addr);
917 return err;
918}
919
920/* Returns: 0 - drop, !=0 - accept */
921static int run_filter(struct tap_filter *filter, const struct sk_buff *skb)
922{
923 /* Cannot use eth_hdr(skb) here because skb_mac_hdr() is incorrect
924 * at this point. */
925 struct ethhdr *eh = (struct ethhdr *) skb->data;
926 int i;
927
928 /* Exact match */
929 for (i = 0; i < filter->count; i++)
930 if (ether_addr_equal(eh->h_dest, filter->addr[i]))
931 return 1;
932
933 /* Inexact match (multicast only) */
934 if (is_multicast_ether_addr(eh->h_dest))
935 return addr_hash_test(filter->mask, eh->h_dest);
936
937 return 0;
938}
939
940/*
941 * Checks whether the packet is accepted or not.
942 * Returns: 0 - drop, !=0 - accept
943 */
944static int check_filter(struct tap_filter *filter, const struct sk_buff *skb)
945{
946 if (!filter->count)
947 return 1;
948
949 return run_filter(filter, skb);
950}
951
952/* Network device part of the driver */
953
954static const struct ethtool_ops tun_ethtool_ops;
955
956/* Net device detach from fd. */
957static void tun_net_uninit(struct net_device *dev)
958{
959 tun_detach_all(dev);
960}
961
962/* Net device open. */
963static int tun_net_open(struct net_device *dev)
964{
965 netif_tx_start_all_queues(dev);
966
967 return 0;
968}
969
970/* Net device close. */
971static int tun_net_close(struct net_device *dev)
972{
973 netif_tx_stop_all_queues(dev);
974 return 0;
975}
976
977/* Net device start xmit */
978static void tun_automq_xmit(struct tun_struct *tun, struct sk_buff *skb)
979{
980#ifdef CONFIG_RPS
981 if (tun->numqueues == 1 && static_branch_unlikely(&rps_needed)) {
982 /* Select queue was not called for the skbuff, so we extract the
983 * RPS hash and save it into the flow_table here.
984 */
985 struct tun_flow_entry *e;
986 __u32 rxhash;
987
988 rxhash = __skb_get_hash_symmetric(skb);
989 e = tun_flow_find(&tun->flows[tun_hashfn(rxhash)], rxhash);
990 if (e)
991 tun_flow_save_rps_rxhash(e, rxhash);
992 }
993#endif
994}
995
996static unsigned int run_ebpf_filter(struct tun_struct *tun,
997 struct sk_buff *skb,
998 int len)
999{
1000 struct tun_prog *prog = rcu_dereference(tun->filter_prog);
1001
1002 if (prog)
1003 len = bpf_prog_run_clear_cb(prog->prog, skb);
1004
1005 return len;
1006}
1007
1008/* Net device start xmit */
1009static netdev_tx_t tun_net_xmit(struct sk_buff *skb, struct net_device *dev)
1010{
1011 struct tun_struct *tun = netdev_priv(dev);
1012 int txq = skb->queue_mapping;
1013 struct tun_file *tfile;
1014 int len = skb->len;
1015
1016 rcu_read_lock();
1017 tfile = rcu_dereference(tun->tfiles[txq]);
1018
1019 /* Drop packet if interface is not attached */
1020 if (!tfile)
1021 goto drop;
1022
1023 if (!rcu_dereference(tun->steering_prog))
1024 tun_automq_xmit(tun, skb);
1025
1026 netif_info(tun, tx_queued, tun->dev, "%s %d\n", __func__, skb->len);
1027
1028 /* Drop if the filter does not like it.
1029 * This is a noop if the filter is disabled.
1030 * Filter can be enabled only for the TAP devices. */
1031 if (!check_filter(&tun->txflt, skb))
1032 goto drop;
1033
1034 if (tfile->socket.sk->sk_filter &&
1035 sk_filter(tfile->socket.sk, skb))
1036 goto drop;
1037
1038 len = run_ebpf_filter(tun, skb, len);
1039 if (len == 0 || pskb_trim(skb, len))
1040 goto drop;
1041
1042 if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC)))
1043 goto drop;
1044
1045 skb_tx_timestamp(skb);
1046
1047 /* Orphan the skb - required as we might hang on to it
1048 * for indefinite time.
1049 */
1050 skb_orphan(skb);
1051
1052 nf_reset_ct(skb);
1053
1054 if (ptr_ring_produce(&tfile->tx_ring, skb))
1055 goto drop;
1056
1057 /* Notify and wake up reader process */
1058 if (tfile->flags & TUN_FASYNC)
1059 kill_fasync(&tfile->fasync, SIGIO, POLL_IN);
1060 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
1061
1062 rcu_read_unlock();
1063 return NETDEV_TX_OK;
1064
1065drop:
1066 atomic_long_inc(&dev->tx_dropped);
1067 skb_tx_error(skb);
1068 kfree_skb(skb);
1069 rcu_read_unlock();
1070 return NET_XMIT_DROP;
1071}
1072
1073static void tun_net_mclist(struct net_device *dev)
1074{
1075 /*
1076 * This callback is supposed to deal with mc filter in
1077 * _rx_ path and has nothing to do with the _tx_ path.
1078 * In rx path we always accept everything userspace gives us.
1079 */
1080}
1081
1082static netdev_features_t tun_net_fix_features(struct net_device *dev,
1083 netdev_features_t features)
1084{
1085 struct tun_struct *tun = netdev_priv(dev);
1086
1087 return (features & tun->set_features) | (features & ~TUN_USER_FEATURES);
1088}
1089
1090static void tun_set_headroom(struct net_device *dev, int new_hr)
1091{
1092 struct tun_struct *tun = netdev_priv(dev);
1093
1094 if (new_hr < NET_SKB_PAD)
1095 new_hr = NET_SKB_PAD;
1096
1097 tun->align = new_hr;
1098}
1099
1100static void
1101tun_net_get_stats64(struct net_device *dev, struct rtnl_link_stats64 *stats)
1102{
1103 struct tun_struct *tun = netdev_priv(dev);
1104
1105 dev_get_tstats64(dev, stats);
1106
1107 stats->rx_frame_errors +=
1108 (unsigned long)atomic_long_read(&tun->rx_frame_errors);
1109}
1110
1111static int tun_xdp_set(struct net_device *dev, struct bpf_prog *prog,
1112 struct netlink_ext_ack *extack)
1113{
1114 struct tun_struct *tun = netdev_priv(dev);
1115 struct tun_file *tfile;
1116 struct bpf_prog *old_prog;
1117 int i;
1118
1119 old_prog = rtnl_dereference(tun->xdp_prog);
1120 rcu_assign_pointer(tun->xdp_prog, prog);
1121 if (old_prog)
1122 bpf_prog_put(old_prog);
1123
1124 for (i = 0; i < tun->numqueues; i++) {
1125 tfile = rtnl_dereference(tun->tfiles[i]);
1126 if (prog)
1127 sock_set_flag(&tfile->sk, SOCK_XDP);
1128 else
1129 sock_reset_flag(&tfile->sk, SOCK_XDP);
1130 }
1131 list_for_each_entry(tfile, &tun->disabled, next) {
1132 if (prog)
1133 sock_set_flag(&tfile->sk, SOCK_XDP);
1134 else
1135 sock_reset_flag(&tfile->sk, SOCK_XDP);
1136 }
1137
1138 return 0;
1139}
1140
1141static int tun_xdp(struct net_device *dev, struct netdev_bpf *xdp)
1142{
1143 switch (xdp->command) {
1144 case XDP_SETUP_PROG:
1145 return tun_xdp_set(dev, xdp->prog, xdp->extack);
1146 default:
1147 return -EINVAL;
1148 }
1149}
1150
1151static int tun_net_change_carrier(struct net_device *dev, bool new_carrier)
1152{
1153 if (new_carrier) {
1154 struct tun_struct *tun = netdev_priv(dev);
1155
1156 if (!tun->numqueues)
1157 return -EPERM;
1158
1159 netif_carrier_on(dev);
1160 } else {
1161 netif_carrier_off(dev);
1162 }
1163 return 0;
1164}
1165
1166static const struct net_device_ops tun_netdev_ops = {
1167 .ndo_uninit = tun_net_uninit,
1168 .ndo_open = tun_net_open,
1169 .ndo_stop = tun_net_close,
1170 .ndo_start_xmit = tun_net_xmit,
1171 .ndo_fix_features = tun_net_fix_features,
1172 .ndo_select_queue = tun_select_queue,
1173 .ndo_set_rx_headroom = tun_set_headroom,
1174 .ndo_get_stats64 = tun_net_get_stats64,
1175 .ndo_change_carrier = tun_net_change_carrier,
1176};
1177
1178static void __tun_xdp_flush_tfile(struct tun_file *tfile)
1179{
1180 /* Notify and wake up reader process */
1181 if (tfile->flags & TUN_FASYNC)
1182 kill_fasync(&tfile->fasync, SIGIO, POLL_IN);
1183 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
1184}
1185
1186static int tun_xdp_xmit(struct net_device *dev, int n,
1187 struct xdp_frame **frames, u32 flags)
1188{
1189 struct tun_struct *tun = netdev_priv(dev);
1190 struct tun_file *tfile;
1191 u32 numqueues;
1192 int nxmit = 0;
1193 int i;
1194
1195 if (unlikely(flags & ~XDP_XMIT_FLAGS_MASK))
1196 return -EINVAL;
1197
1198 rcu_read_lock();
1199
1200resample:
1201 numqueues = READ_ONCE(tun->numqueues);
1202 if (!numqueues) {
1203 rcu_read_unlock();
1204 return -ENXIO; /* Caller will free/return all frames */
1205 }
1206
1207 tfile = rcu_dereference(tun->tfiles[smp_processor_id() %
1208 numqueues]);
1209 if (unlikely(!tfile))
1210 goto resample;
1211
1212 spin_lock(&tfile->tx_ring.producer_lock);
1213 for (i = 0; i < n; i++) {
1214 struct xdp_frame *xdp = frames[i];
1215 /* Encode the XDP flag into lowest bit for consumer to differ
1216 * XDP buffer from sk_buff.
1217 */
1218 void *frame = tun_xdp_to_ptr(xdp);
1219
1220 if (__ptr_ring_produce(&tfile->tx_ring, frame)) {
1221 atomic_long_inc(&dev->tx_dropped);
1222 break;
1223 }
1224 nxmit++;
1225 }
1226 spin_unlock(&tfile->tx_ring.producer_lock);
1227
1228 if (flags & XDP_XMIT_FLUSH)
1229 __tun_xdp_flush_tfile(tfile);
1230
1231 rcu_read_unlock();
1232 return nxmit;
1233}
1234
1235static int tun_xdp_tx(struct net_device *dev, struct xdp_buff *xdp)
1236{
1237 struct xdp_frame *frame = xdp_convert_buff_to_frame(xdp);
1238 int nxmit;
1239
1240 if (unlikely(!frame))
1241 return -EOVERFLOW;
1242
1243 nxmit = tun_xdp_xmit(dev, 1, &frame, XDP_XMIT_FLUSH);
1244 if (!nxmit)
1245 xdp_return_frame_rx_napi(frame);
1246 return nxmit;
1247}
1248
1249static const struct net_device_ops tap_netdev_ops = {
1250 .ndo_uninit = tun_net_uninit,
1251 .ndo_open = tun_net_open,
1252 .ndo_stop = tun_net_close,
1253 .ndo_start_xmit = tun_net_xmit,
1254 .ndo_fix_features = tun_net_fix_features,
1255 .ndo_set_rx_mode = tun_net_mclist,
1256 .ndo_set_mac_address = eth_mac_addr,
1257 .ndo_validate_addr = eth_validate_addr,
1258 .ndo_select_queue = tun_select_queue,
1259 .ndo_features_check = passthru_features_check,
1260 .ndo_set_rx_headroom = tun_set_headroom,
1261 .ndo_get_stats64 = dev_get_tstats64,
1262 .ndo_bpf = tun_xdp,
1263 .ndo_xdp_xmit = tun_xdp_xmit,
1264 .ndo_change_carrier = tun_net_change_carrier,
1265};
1266
1267static void tun_flow_init(struct tun_struct *tun)
1268{
1269 int i;
1270
1271 for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++)
1272 INIT_HLIST_HEAD(&tun->flows[i]);
1273
1274 tun->ageing_time = TUN_FLOW_EXPIRE;
1275 timer_setup(&tun->flow_gc_timer, tun_flow_cleanup, 0);
1276 mod_timer(&tun->flow_gc_timer,
1277 round_jiffies_up(jiffies + tun->ageing_time));
1278}
1279
1280static void tun_flow_uninit(struct tun_struct *tun)
1281{
1282 del_timer_sync(&tun->flow_gc_timer);
1283 tun_flow_flush(tun);
1284}
1285
1286#define MIN_MTU 68
1287#define MAX_MTU 65535
1288
1289/* Initialize net device. */
1290static void tun_net_init(struct net_device *dev)
1291{
1292 struct tun_struct *tun = netdev_priv(dev);
1293
1294 switch (tun->flags & TUN_TYPE_MASK) {
1295 case IFF_TUN:
1296 dev->netdev_ops = &tun_netdev_ops;
1297 dev->header_ops = &ip_tunnel_header_ops;
1298
1299 /* Point-to-Point TUN Device */
1300 dev->hard_header_len = 0;
1301 dev->addr_len = 0;
1302 dev->mtu = 1500;
1303
1304 /* Zero header length */
1305 dev->type = ARPHRD_NONE;
1306 dev->flags = IFF_POINTOPOINT | IFF_NOARP | IFF_MULTICAST;
1307 break;
1308
1309 case IFF_TAP:
1310 dev->netdev_ops = &tap_netdev_ops;
1311 /* Ethernet TAP Device */
1312 ether_setup(dev);
1313 dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1314 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
1315
1316 eth_hw_addr_random(dev);
1317
1318 break;
1319 }
1320
1321 dev->min_mtu = MIN_MTU;
1322 dev->max_mtu = MAX_MTU - dev->hard_header_len;
1323}
1324
1325static bool tun_sock_writeable(struct tun_struct *tun, struct tun_file *tfile)
1326{
1327 struct sock *sk = tfile->socket.sk;
1328
1329 return (tun->dev->flags & IFF_UP) && sock_writeable(sk);
1330}
1331
1332/* Character device part */
1333
1334/* Poll */
1335static __poll_t tun_chr_poll(struct file *file, poll_table *wait)
1336{
1337 struct tun_file *tfile = file->private_data;
1338 struct tun_struct *tun = tun_get(tfile);
1339 struct sock *sk;
1340 __poll_t mask = 0;
1341
1342 if (!tun)
1343 return EPOLLERR;
1344
1345 sk = tfile->socket.sk;
1346
1347 poll_wait(file, sk_sleep(sk), wait);
1348
1349 if (!ptr_ring_empty(&tfile->tx_ring))
1350 mask |= EPOLLIN | EPOLLRDNORM;
1351
1352 /* Make sure SOCKWQ_ASYNC_NOSPACE is set if not writable to
1353 * guarantee EPOLLOUT to be raised by either here or
1354 * tun_sock_write_space(). Then process could get notification
1355 * after it writes to a down device and meets -EIO.
1356 */
1357 if (tun_sock_writeable(tun, tfile) ||
1358 (!test_and_set_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags) &&
1359 tun_sock_writeable(tun, tfile)))
1360 mask |= EPOLLOUT | EPOLLWRNORM;
1361
1362 if (tun->dev->reg_state != NETREG_REGISTERED)
1363 mask = EPOLLERR;
1364
1365 tun_put(tun);
1366 return mask;
1367}
1368
1369static struct sk_buff *tun_napi_alloc_frags(struct tun_file *tfile,
1370 size_t len,
1371 const struct iov_iter *it)
1372{
1373 struct sk_buff *skb;
1374 size_t linear;
1375 int err;
1376 int i;
1377
1378 if (it->nr_segs > MAX_SKB_FRAGS + 1)
1379 return ERR_PTR(-EMSGSIZE);
1380
1381 local_bh_disable();
1382 skb = napi_get_frags(&tfile->napi);
1383 local_bh_enable();
1384 if (!skb)
1385 return ERR_PTR(-ENOMEM);
1386
1387 linear = iov_iter_single_seg_count(it);
1388 err = __skb_grow(skb, linear);
1389 if (err)
1390 goto free;
1391
1392 skb->len = len;
1393 skb->data_len = len - linear;
1394 skb->truesize += skb->data_len;
1395
1396 for (i = 1; i < it->nr_segs; i++) {
1397 size_t fragsz = it->iov[i].iov_len;
1398 struct page *page;
1399 void *frag;
1400
1401 if (fragsz == 0 || fragsz > PAGE_SIZE) {
1402 err = -EINVAL;
1403 goto free;
1404 }
1405 frag = netdev_alloc_frag(fragsz);
1406 if (!frag) {
1407 err = -ENOMEM;
1408 goto free;
1409 }
1410 page = virt_to_head_page(frag);
1411 skb_fill_page_desc(skb, i - 1, page,
1412 frag - page_address(page), fragsz);
1413 }
1414
1415 return skb;
1416free:
1417 /* frees skb and all frags allocated with napi_alloc_frag() */
1418 napi_free_frags(&tfile->napi);
1419 return ERR_PTR(err);
1420}
1421
1422/* prepad is the amount to reserve at front. len is length after that.
1423 * linear is a hint as to how much to copy (usually headers). */
1424static struct sk_buff *tun_alloc_skb(struct tun_file *tfile,
1425 size_t prepad, size_t len,
1426 size_t linear, int noblock)
1427{
1428 struct sock *sk = tfile->socket.sk;
1429 struct sk_buff *skb;
1430 int err;
1431
1432 /* Under a page? Don't bother with paged skb. */
1433 if (prepad + len < PAGE_SIZE || !linear)
1434 linear = len;
1435
1436 skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
1437 &err, 0);
1438 if (!skb)
1439 return ERR_PTR(err);
1440
1441 skb_reserve(skb, prepad);
1442 skb_put(skb, linear);
1443 skb->data_len = len - linear;
1444 skb->len += len - linear;
1445
1446 return skb;
1447}
1448
1449static void tun_rx_batched(struct tun_struct *tun, struct tun_file *tfile,
1450 struct sk_buff *skb, int more)
1451{
1452 struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
1453 struct sk_buff_head process_queue;
1454 u32 rx_batched = tun->rx_batched;
1455 bool rcv = false;
1456
1457 if (!rx_batched || (!more && skb_queue_empty(queue))) {
1458 local_bh_disable();
1459 skb_record_rx_queue(skb, tfile->queue_index);
1460 netif_receive_skb(skb);
1461 local_bh_enable();
1462 return;
1463 }
1464
1465 spin_lock(&queue->lock);
1466 if (!more || skb_queue_len(queue) == rx_batched) {
1467 __skb_queue_head_init(&process_queue);
1468 skb_queue_splice_tail_init(queue, &process_queue);
1469 rcv = true;
1470 } else {
1471 __skb_queue_tail(queue, skb);
1472 }
1473 spin_unlock(&queue->lock);
1474
1475 if (rcv) {
1476 struct sk_buff *nskb;
1477
1478 local_bh_disable();
1479 while ((nskb = __skb_dequeue(&process_queue))) {
1480 skb_record_rx_queue(nskb, tfile->queue_index);
1481 netif_receive_skb(nskb);
1482 }
1483 skb_record_rx_queue(skb, tfile->queue_index);
1484 netif_receive_skb(skb);
1485 local_bh_enable();
1486 }
1487}
1488
1489static bool tun_can_build_skb(struct tun_struct *tun, struct tun_file *tfile,
1490 int len, int noblock, bool zerocopy)
1491{
1492 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
1493 return false;
1494
1495 if (tfile->socket.sk->sk_sndbuf != INT_MAX)
1496 return false;
1497
1498 if (!noblock)
1499 return false;
1500
1501 if (zerocopy)
1502 return false;
1503
1504 if (SKB_DATA_ALIGN(len + TUN_RX_PAD) +
1505 SKB_DATA_ALIGN(sizeof(struct skb_shared_info)) > PAGE_SIZE)
1506 return false;
1507
1508 return true;
1509}
1510
1511static struct sk_buff *__tun_build_skb(struct tun_file *tfile,
1512 struct page_frag *alloc_frag, char *buf,
1513 int buflen, int len, int pad)
1514{
1515 struct sk_buff *skb = build_skb(buf, buflen);
1516
1517 if (!skb)
1518 return ERR_PTR(-ENOMEM);
1519
1520 skb_reserve(skb, pad);
1521 skb_put(skb, len);
1522 skb_set_owner_w(skb, tfile->socket.sk);
1523
1524 get_page(alloc_frag->page);
1525 alloc_frag->offset += buflen;
1526
1527 return skb;
1528}
1529
1530static int tun_xdp_act(struct tun_struct *tun, struct bpf_prog *xdp_prog,
1531 struct xdp_buff *xdp, u32 act)
1532{
1533 int err;
1534
1535 switch (act) {
1536 case XDP_REDIRECT:
1537 err = xdp_do_redirect(tun->dev, xdp, xdp_prog);
1538 if (err)
1539 return err;
1540 break;
1541 case XDP_TX:
1542 err = tun_xdp_tx(tun->dev, xdp);
1543 if (err < 0)
1544 return err;
1545 break;
1546 case XDP_PASS:
1547 break;
1548 default:
1549 bpf_warn_invalid_xdp_action(act);
1550 fallthrough;
1551 case XDP_ABORTED:
1552 trace_xdp_exception(tun->dev, xdp_prog, act);
1553 fallthrough;
1554 case XDP_DROP:
1555 atomic_long_inc(&tun->dev->rx_dropped);
1556 break;
1557 }
1558
1559 return act;
1560}
1561
1562static struct sk_buff *tun_build_skb(struct tun_struct *tun,
1563 struct tun_file *tfile,
1564 struct iov_iter *from,
1565 struct virtio_net_hdr *hdr,
1566 int len, int *skb_xdp)
1567{
1568 struct page_frag *alloc_frag = ¤t->task_frag;
1569 struct bpf_prog *xdp_prog;
1570 int buflen = SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
1571 char *buf;
1572 size_t copied;
1573 int pad = TUN_RX_PAD;
1574 int err = 0;
1575
1576 rcu_read_lock();
1577 xdp_prog = rcu_dereference(tun->xdp_prog);
1578 if (xdp_prog)
1579 pad += XDP_PACKET_HEADROOM;
1580 buflen += SKB_DATA_ALIGN(len + pad);
1581 rcu_read_unlock();
1582
1583 alloc_frag->offset = ALIGN((u64)alloc_frag->offset, SMP_CACHE_BYTES);
1584 if (unlikely(!skb_page_frag_refill(buflen, alloc_frag, GFP_KERNEL)))
1585 return ERR_PTR(-ENOMEM);
1586
1587 buf = (char *)page_address(alloc_frag->page) + alloc_frag->offset;
1588 copied = copy_page_from_iter(alloc_frag->page,
1589 alloc_frag->offset + pad,
1590 len, from);
1591 if (copied != len)
1592 return ERR_PTR(-EFAULT);
1593
1594 /* There's a small window that XDP may be set after the check
1595 * of xdp_prog above, this should be rare and for simplicity
1596 * we do XDP on skb in case the headroom is not enough.
1597 */
1598 if (hdr->gso_type || !xdp_prog) {
1599 *skb_xdp = 1;
1600 return __tun_build_skb(tfile, alloc_frag, buf, buflen, len,
1601 pad);
1602 }
1603
1604 *skb_xdp = 0;
1605
1606 local_bh_disable();
1607 rcu_read_lock();
1608 xdp_prog = rcu_dereference(tun->xdp_prog);
1609 if (xdp_prog) {
1610 struct xdp_buff xdp;
1611 u32 act;
1612
1613 xdp_init_buff(&xdp, buflen, &tfile->xdp_rxq);
1614 xdp_prepare_buff(&xdp, buf, pad, len, false);
1615
1616 act = bpf_prog_run_xdp(xdp_prog, &xdp);
1617 if (act == XDP_REDIRECT || act == XDP_TX) {
1618 get_page(alloc_frag->page);
1619 alloc_frag->offset += buflen;
1620 }
1621 err = tun_xdp_act(tun, xdp_prog, &xdp, act);
1622 if (err < 0) {
1623 if (act == XDP_REDIRECT || act == XDP_TX)
1624 put_page(alloc_frag->page);
1625 goto out;
1626 }
1627
1628 if (err == XDP_REDIRECT)
1629 xdp_do_flush();
1630 if (err != XDP_PASS)
1631 goto out;
1632
1633 pad = xdp.data - xdp.data_hard_start;
1634 len = xdp.data_end - xdp.data;
1635 }
1636 rcu_read_unlock();
1637 local_bh_enable();
1638
1639 return __tun_build_skb(tfile, alloc_frag, buf, buflen, len, pad);
1640
1641out:
1642 rcu_read_unlock();
1643 local_bh_enable();
1644 return NULL;
1645}
1646
1647/* Get packet from user space buffer */
1648static ssize_t tun_get_user(struct tun_struct *tun, struct tun_file *tfile,
1649 void *msg_control, struct iov_iter *from,
1650 int noblock, bool more)
1651{
1652 struct tun_pi pi = { 0, cpu_to_be16(ETH_P_IP) };
1653 struct sk_buff *skb;
1654 size_t total_len = iov_iter_count(from);
1655 size_t len = total_len, align = tun->align, linear;
1656 struct virtio_net_hdr gso = { 0 };
1657 int good_linear;
1658 int copylen;
1659 bool zerocopy = false;
1660 int err;
1661 u32 rxhash = 0;
1662 int skb_xdp = 1;
1663 bool frags = tun_napi_frags_enabled(tfile);
1664
1665 if (!(tun->flags & IFF_NO_PI)) {
1666 if (len < sizeof(pi))
1667 return -EINVAL;
1668 len -= sizeof(pi);
1669
1670 if (!copy_from_iter_full(&pi, sizeof(pi), from))
1671 return -EFAULT;
1672 }
1673
1674 if (tun->flags & IFF_VNET_HDR) {
1675 int vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
1676
1677 if (len < vnet_hdr_sz)
1678 return -EINVAL;
1679 len -= vnet_hdr_sz;
1680
1681 if (!copy_from_iter_full(&gso, sizeof(gso), from))
1682 return -EFAULT;
1683
1684 if ((gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
1685 tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2 > tun16_to_cpu(tun, gso.hdr_len))
1686 gso.hdr_len = cpu_to_tun16(tun, tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2);
1687
1688 if (tun16_to_cpu(tun, gso.hdr_len) > len)
1689 return -EINVAL;
1690 iov_iter_advance(from, vnet_hdr_sz - sizeof(gso));
1691 }
1692
1693 if ((tun->flags & TUN_TYPE_MASK) == IFF_TAP) {
1694 align += NET_IP_ALIGN;
1695 if (unlikely(len < ETH_HLEN ||
1696 (gso.hdr_len && tun16_to_cpu(tun, gso.hdr_len) < ETH_HLEN)))
1697 return -EINVAL;
1698 }
1699
1700 good_linear = SKB_MAX_HEAD(align);
1701
1702 if (msg_control) {
1703 struct iov_iter i = *from;
1704
1705 /* There are 256 bytes to be copied in skb, so there is
1706 * enough room for skb expand head in case it is used.
1707 * The rest of the buffer is mapped from userspace.
1708 */
1709 copylen = gso.hdr_len ? tun16_to_cpu(tun, gso.hdr_len) : GOODCOPY_LEN;
1710 if (copylen > good_linear)
1711 copylen = good_linear;
1712 linear = copylen;
1713 iov_iter_advance(&i, copylen);
1714 if (iov_iter_npages(&i, INT_MAX) <= MAX_SKB_FRAGS)
1715 zerocopy = true;
1716 }
1717
1718 if (!frags && tun_can_build_skb(tun, tfile, len, noblock, zerocopy)) {
1719 /* For the packet that is not easy to be processed
1720 * (e.g gso or jumbo packet), we will do it at after
1721 * skb was created with generic XDP routine.
1722 */
1723 skb = tun_build_skb(tun, tfile, from, &gso, len, &skb_xdp);
1724 if (IS_ERR(skb)) {
1725 atomic_long_inc(&tun->dev->rx_dropped);
1726 return PTR_ERR(skb);
1727 }
1728 if (!skb)
1729 return total_len;
1730 } else {
1731 if (!zerocopy) {
1732 copylen = len;
1733 if (tun16_to_cpu(tun, gso.hdr_len) > good_linear)
1734 linear = good_linear;
1735 else
1736 linear = tun16_to_cpu(tun, gso.hdr_len);
1737 }
1738
1739 if (frags) {
1740 mutex_lock(&tfile->napi_mutex);
1741 skb = tun_napi_alloc_frags(tfile, copylen, from);
1742 /* tun_napi_alloc_frags() enforces a layout for the skb.
1743 * If zerocopy is enabled, then this layout will be
1744 * overwritten by zerocopy_sg_from_iter().
1745 */
1746 zerocopy = false;
1747 } else {
1748 skb = tun_alloc_skb(tfile, align, copylen, linear,
1749 noblock);
1750 }
1751
1752 if (IS_ERR(skb)) {
1753 if (PTR_ERR(skb) != -EAGAIN)
1754 atomic_long_inc(&tun->dev->rx_dropped);
1755 if (frags)
1756 mutex_unlock(&tfile->napi_mutex);
1757 return PTR_ERR(skb);
1758 }
1759
1760 if (zerocopy)
1761 err = zerocopy_sg_from_iter(skb, from);
1762 else
1763 err = skb_copy_datagram_from_iter(skb, 0, from, len);
1764
1765 if (err) {
1766 err = -EFAULT;
1767drop:
1768 atomic_long_inc(&tun->dev->rx_dropped);
1769 kfree_skb(skb);
1770 if (frags) {
1771 tfile->napi.skb = NULL;
1772 mutex_unlock(&tfile->napi_mutex);
1773 }
1774
1775 return err;
1776 }
1777 }
1778
1779 if (virtio_net_hdr_to_skb(skb, &gso, tun_is_little_endian(tun))) {
1780 atomic_long_inc(&tun->rx_frame_errors);
1781 kfree_skb(skb);
1782 if (frags) {
1783 tfile->napi.skb = NULL;
1784 mutex_unlock(&tfile->napi_mutex);
1785 }
1786
1787 return -EINVAL;
1788 }
1789
1790 switch (tun->flags & TUN_TYPE_MASK) {
1791 case IFF_TUN:
1792 if (tun->flags & IFF_NO_PI) {
1793 u8 ip_version = skb->len ? (skb->data[0] >> 4) : 0;
1794
1795 switch (ip_version) {
1796 case 4:
1797 pi.proto = htons(ETH_P_IP);
1798 break;
1799 case 6:
1800 pi.proto = htons(ETH_P_IPV6);
1801 break;
1802 default:
1803 atomic_long_inc(&tun->dev->rx_dropped);
1804 kfree_skb(skb);
1805 return -EINVAL;
1806 }
1807 }
1808
1809 skb_reset_mac_header(skb);
1810 skb->protocol = pi.proto;
1811 skb->dev = tun->dev;
1812 break;
1813 case IFF_TAP:
1814 if (frags && !pskb_may_pull(skb, ETH_HLEN)) {
1815 err = -ENOMEM;
1816 goto drop;
1817 }
1818 skb->protocol = eth_type_trans(skb, tun->dev);
1819 break;
1820 }
1821
1822 /* copy skb_ubuf_info for callback when skb has no error */
1823 if (zerocopy) {
1824 skb_zcopy_init(skb, msg_control);
1825 } else if (msg_control) {
1826 struct ubuf_info *uarg = msg_control;
1827 uarg->callback(NULL, uarg, false);
1828 }
1829
1830 skb_reset_network_header(skb);
1831 skb_probe_transport_header(skb);
1832 skb_record_rx_queue(skb, tfile->queue_index);
1833
1834 if (skb_xdp) {
1835 struct bpf_prog *xdp_prog;
1836 int ret;
1837
1838 local_bh_disable();
1839 rcu_read_lock();
1840 xdp_prog = rcu_dereference(tun->xdp_prog);
1841 if (xdp_prog) {
1842 ret = do_xdp_generic(xdp_prog, skb);
1843 if (ret != XDP_PASS) {
1844 rcu_read_unlock();
1845 local_bh_enable();
1846 if (frags) {
1847 tfile->napi.skb = NULL;
1848 mutex_unlock(&tfile->napi_mutex);
1849 }
1850 return total_len;
1851 }
1852 }
1853 rcu_read_unlock();
1854 local_bh_enable();
1855 }
1856
1857 /* Compute the costly rx hash only if needed for flow updates.
1858 * We may get a very small possibility of OOO during switching, not
1859 * worth to optimize.
1860 */
1861 if (!rcu_access_pointer(tun->steering_prog) && tun->numqueues > 1 &&
1862 !tfile->detached)
1863 rxhash = __skb_get_hash_symmetric(skb);
1864
1865 rcu_read_lock();
1866 if (unlikely(!(tun->dev->flags & IFF_UP))) {
1867 err = -EIO;
1868 rcu_read_unlock();
1869 goto drop;
1870 }
1871
1872 if (frags) {
1873 u32 headlen;
1874
1875 /* Exercise flow dissector code path. */
1876 skb_push(skb, ETH_HLEN);
1877 headlen = eth_get_headlen(tun->dev, skb->data,
1878 skb_headlen(skb));
1879
1880 if (unlikely(headlen > skb_headlen(skb))) {
1881 atomic_long_inc(&tun->dev->rx_dropped);
1882 napi_free_frags(&tfile->napi);
1883 rcu_read_unlock();
1884 mutex_unlock(&tfile->napi_mutex);
1885 WARN_ON(1);
1886 return -ENOMEM;
1887 }
1888
1889 local_bh_disable();
1890 napi_gro_frags(&tfile->napi);
1891 local_bh_enable();
1892 mutex_unlock(&tfile->napi_mutex);
1893 } else if (tfile->napi_enabled) {
1894 struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
1895 int queue_len;
1896
1897 spin_lock_bh(&queue->lock);
1898 __skb_queue_tail(queue, skb);
1899 queue_len = skb_queue_len(queue);
1900 spin_unlock(&queue->lock);
1901
1902 if (!more || queue_len > NAPI_POLL_WEIGHT)
1903 napi_schedule(&tfile->napi);
1904
1905 local_bh_enable();
1906 } else if (!IS_ENABLED(CONFIG_4KSTACKS)) {
1907 tun_rx_batched(tun, tfile, skb, more);
1908 } else {
1909 netif_rx_ni(skb);
1910 }
1911 rcu_read_unlock();
1912
1913 preempt_disable();
1914 dev_sw_netstats_rx_add(tun->dev, len);
1915 preempt_enable();
1916
1917 if (rxhash)
1918 tun_flow_update(tun, rxhash, tfile);
1919
1920 return total_len;
1921}
1922
1923static ssize_t tun_chr_write_iter(struct kiocb *iocb, struct iov_iter *from)
1924{
1925 struct file *file = iocb->ki_filp;
1926 struct tun_file *tfile = file->private_data;
1927 struct tun_struct *tun = tun_get(tfile);
1928 ssize_t result;
1929 int noblock = 0;
1930
1931 if (!tun)
1932 return -EBADFD;
1933
1934 if ((file->f_flags & O_NONBLOCK) || (iocb->ki_flags & IOCB_NOWAIT))
1935 noblock = 1;
1936
1937 result = tun_get_user(tun, tfile, NULL, from, noblock, false);
1938
1939 tun_put(tun);
1940 return result;
1941}
1942
1943static ssize_t tun_put_user_xdp(struct tun_struct *tun,
1944 struct tun_file *tfile,
1945 struct xdp_frame *xdp_frame,
1946 struct iov_iter *iter)
1947{
1948 int vnet_hdr_sz = 0;
1949 size_t size = xdp_frame->len;
1950 size_t ret;
1951
1952 if (tun->flags & IFF_VNET_HDR) {
1953 struct virtio_net_hdr gso = { 0 };
1954
1955 vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
1956 if (unlikely(iov_iter_count(iter) < vnet_hdr_sz))
1957 return -EINVAL;
1958 if (unlikely(copy_to_iter(&gso, sizeof(gso), iter) !=
1959 sizeof(gso)))
1960 return -EFAULT;
1961 iov_iter_advance(iter, vnet_hdr_sz - sizeof(gso));
1962 }
1963
1964 ret = copy_to_iter(xdp_frame->data, size, iter) + vnet_hdr_sz;
1965
1966 preempt_disable();
1967 dev_sw_netstats_tx_add(tun->dev, 1, ret);
1968 preempt_enable();
1969
1970 return ret;
1971}
1972
1973/* Put packet to the user space buffer */
1974static ssize_t tun_put_user(struct tun_struct *tun,
1975 struct tun_file *tfile,
1976 struct sk_buff *skb,
1977 struct iov_iter *iter)
1978{
1979 struct tun_pi pi = { 0, skb->protocol };
1980 ssize_t total;
1981 int vlan_offset = 0;
1982 int vlan_hlen = 0;
1983 int vnet_hdr_sz = 0;
1984
1985 if (skb_vlan_tag_present(skb))
1986 vlan_hlen = VLAN_HLEN;
1987
1988 if (tun->flags & IFF_VNET_HDR)
1989 vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
1990
1991 total = skb->len + vlan_hlen + vnet_hdr_sz;
1992
1993 if (!(tun->flags & IFF_NO_PI)) {
1994 if (iov_iter_count(iter) < sizeof(pi))
1995 return -EINVAL;
1996
1997 total += sizeof(pi);
1998 if (iov_iter_count(iter) < total) {
1999 /* Packet will be striped */
2000 pi.flags |= TUN_PKT_STRIP;
2001 }
2002
2003 if (copy_to_iter(&pi, sizeof(pi), iter) != sizeof(pi))
2004 return -EFAULT;
2005 }
2006
2007 if (vnet_hdr_sz) {
2008 struct virtio_net_hdr gso;
2009
2010 if (iov_iter_count(iter) < vnet_hdr_sz)
2011 return -EINVAL;
2012
2013 if (virtio_net_hdr_from_skb(skb, &gso,
2014 tun_is_little_endian(tun), true,
2015 vlan_hlen)) {
2016 struct skb_shared_info *sinfo = skb_shinfo(skb);
2017 pr_err("unexpected GSO type: "
2018 "0x%x, gso_size %d, hdr_len %d\n",
2019 sinfo->gso_type, tun16_to_cpu(tun, gso.gso_size),
2020 tun16_to_cpu(tun, gso.hdr_len));
2021 print_hex_dump(KERN_ERR, "tun: ",
2022 DUMP_PREFIX_NONE,
2023 16, 1, skb->head,
2024 min((int)tun16_to_cpu(tun, gso.hdr_len), 64), true);
2025 WARN_ON_ONCE(1);
2026 return -EINVAL;
2027 }
2028
2029 if (copy_to_iter(&gso, sizeof(gso), iter) != sizeof(gso))
2030 return -EFAULT;
2031
2032 iov_iter_advance(iter, vnet_hdr_sz - sizeof(gso));
2033 }
2034
2035 if (vlan_hlen) {
2036 int ret;
2037 struct veth veth;
2038
2039 veth.h_vlan_proto = skb->vlan_proto;
2040 veth.h_vlan_TCI = htons(skb_vlan_tag_get(skb));
2041
2042 vlan_offset = offsetof(struct vlan_ethhdr, h_vlan_proto);
2043
2044 ret = skb_copy_datagram_iter(skb, 0, iter, vlan_offset);
2045 if (ret || !iov_iter_count(iter))
2046 goto done;
2047
2048 ret = copy_to_iter(&veth, sizeof(veth), iter);
2049 if (ret != sizeof(veth) || !iov_iter_count(iter))
2050 goto done;
2051 }
2052
2053 skb_copy_datagram_iter(skb, vlan_offset, iter, skb->len - vlan_offset);
2054
2055done:
2056 /* caller is in process context, */
2057 preempt_disable();
2058 dev_sw_netstats_tx_add(tun->dev, 1, skb->len + vlan_hlen);
2059 preempt_enable();
2060
2061 return total;
2062}
2063
2064static void *tun_ring_recv(struct tun_file *tfile, int noblock, int *err)
2065{
2066 DECLARE_WAITQUEUE(wait, current);
2067 void *ptr = NULL;
2068 int error = 0;
2069
2070 ptr = ptr_ring_consume(&tfile->tx_ring);
2071 if (ptr)
2072 goto out;
2073 if (noblock) {
2074 error = -EAGAIN;
2075 goto out;
2076 }
2077
2078 add_wait_queue(&tfile->socket.wq.wait, &wait);
2079
2080 while (1) {
2081 set_current_state(TASK_INTERRUPTIBLE);
2082 ptr = ptr_ring_consume(&tfile->tx_ring);
2083 if (ptr)
2084 break;
2085 if (signal_pending(current)) {
2086 error = -ERESTARTSYS;
2087 break;
2088 }
2089 if (tfile->socket.sk->sk_shutdown & RCV_SHUTDOWN) {
2090 error = -EFAULT;
2091 break;
2092 }
2093
2094 schedule();
2095 }
2096
2097 __set_current_state(TASK_RUNNING);
2098 remove_wait_queue(&tfile->socket.wq.wait, &wait);
2099
2100out:
2101 *err = error;
2102 return ptr;
2103}
2104
2105static ssize_t tun_do_read(struct tun_struct *tun, struct tun_file *tfile,
2106 struct iov_iter *to,
2107 int noblock, void *ptr)
2108{
2109 ssize_t ret;
2110 int err;
2111
2112 if (!iov_iter_count(to)) {
2113 tun_ptr_free(ptr);
2114 return 0;
2115 }
2116
2117 if (!ptr) {
2118 /* Read frames from ring */
2119 ptr = tun_ring_recv(tfile, noblock, &err);
2120 if (!ptr)
2121 return err;
2122 }
2123
2124 if (tun_is_xdp_frame(ptr)) {
2125 struct xdp_frame *xdpf = tun_ptr_to_xdp(ptr);
2126
2127 ret = tun_put_user_xdp(tun, tfile, xdpf, to);
2128 xdp_return_frame(xdpf);
2129 } else {
2130 struct sk_buff *skb = ptr;
2131
2132 ret = tun_put_user(tun, tfile, skb, to);
2133 if (unlikely(ret < 0))
2134 kfree_skb(skb);
2135 else
2136 consume_skb(skb);
2137 }
2138
2139 return ret;
2140}
2141
2142static ssize_t tun_chr_read_iter(struct kiocb *iocb, struct iov_iter *to)
2143{
2144 struct file *file = iocb->ki_filp;
2145 struct tun_file *tfile = file->private_data;
2146 struct tun_struct *tun = tun_get(tfile);
2147 ssize_t len = iov_iter_count(to), ret;
2148 int noblock = 0;
2149
2150 if (!tun)
2151 return -EBADFD;
2152
2153 if ((file->f_flags & O_NONBLOCK) || (iocb->ki_flags & IOCB_NOWAIT))
2154 noblock = 1;
2155
2156 ret = tun_do_read(tun, tfile, to, noblock, NULL);
2157 ret = min_t(ssize_t, ret, len);
2158 if (ret > 0)
2159 iocb->ki_pos = ret;
2160 tun_put(tun);
2161 return ret;
2162}
2163
2164static void tun_prog_free(struct rcu_head *rcu)
2165{
2166 struct tun_prog *prog = container_of(rcu, struct tun_prog, rcu);
2167
2168 bpf_prog_destroy(prog->prog);
2169 kfree(prog);
2170}
2171
2172static int __tun_set_ebpf(struct tun_struct *tun,
2173 struct tun_prog __rcu **prog_p,
2174 struct bpf_prog *prog)
2175{
2176 struct tun_prog *old, *new = NULL;
2177
2178 if (prog) {
2179 new = kmalloc(sizeof(*new), GFP_KERNEL);
2180 if (!new)
2181 return -ENOMEM;
2182 new->prog = prog;
2183 }
2184
2185 spin_lock_bh(&tun->lock);
2186 old = rcu_dereference_protected(*prog_p,
2187 lockdep_is_held(&tun->lock));
2188 rcu_assign_pointer(*prog_p, new);
2189 spin_unlock_bh(&tun->lock);
2190
2191 if (old)
2192 call_rcu(&old->rcu, tun_prog_free);
2193
2194 return 0;
2195}
2196
2197static void tun_free_netdev(struct net_device *dev)
2198{
2199 struct tun_struct *tun = netdev_priv(dev);
2200
2201 BUG_ON(!(list_empty(&tun->disabled)));
2202
2203 free_percpu(dev->tstats);
2204 /* We clear tstats so that tun_set_iff() can tell if
2205 * tun_free_netdev() has been called from register_netdevice().
2206 */
2207 dev->tstats = NULL;
2208
2209 tun_flow_uninit(tun);
2210 security_tun_dev_free_security(tun->security);
2211 __tun_set_ebpf(tun, &tun->steering_prog, NULL);
2212 __tun_set_ebpf(tun, &tun->filter_prog, NULL);
2213}
2214
2215static void tun_setup(struct net_device *dev)
2216{
2217 struct tun_struct *tun = netdev_priv(dev);
2218
2219 tun->owner = INVALID_UID;
2220 tun->group = INVALID_GID;
2221 tun_default_link_ksettings(dev, &tun->link_ksettings);
2222
2223 dev->ethtool_ops = &tun_ethtool_ops;
2224 dev->needs_free_netdev = true;
2225 dev->priv_destructor = tun_free_netdev;
2226 /* We prefer our own queue length */
2227 dev->tx_queue_len = TUN_READQ_SIZE;
2228}
2229
2230/* Trivial set of netlink ops to allow deleting tun or tap
2231 * device with netlink.
2232 */
2233static int tun_validate(struct nlattr *tb[], struct nlattr *data[],
2234 struct netlink_ext_ack *extack)
2235{
2236 NL_SET_ERR_MSG(extack,
2237 "tun/tap creation via rtnetlink is not supported.");
2238 return -EOPNOTSUPP;
2239}
2240
2241static size_t tun_get_size(const struct net_device *dev)
2242{
2243 BUILD_BUG_ON(sizeof(u32) != sizeof(uid_t));
2244 BUILD_BUG_ON(sizeof(u32) != sizeof(gid_t));
2245
2246 return nla_total_size(sizeof(uid_t)) + /* OWNER */
2247 nla_total_size(sizeof(gid_t)) + /* GROUP */
2248 nla_total_size(sizeof(u8)) + /* TYPE */
2249 nla_total_size(sizeof(u8)) + /* PI */
2250 nla_total_size(sizeof(u8)) + /* VNET_HDR */
2251 nla_total_size(sizeof(u8)) + /* PERSIST */
2252 nla_total_size(sizeof(u8)) + /* MULTI_QUEUE */
2253 nla_total_size(sizeof(u32)) + /* NUM_QUEUES */
2254 nla_total_size(sizeof(u32)) + /* NUM_DISABLED_QUEUES */
2255 0;
2256}
2257
2258static int tun_fill_info(struct sk_buff *skb, const struct net_device *dev)
2259{
2260 struct tun_struct *tun = netdev_priv(dev);
2261
2262 if (nla_put_u8(skb, IFLA_TUN_TYPE, tun->flags & TUN_TYPE_MASK))
2263 goto nla_put_failure;
2264 if (uid_valid(tun->owner) &&
2265 nla_put_u32(skb, IFLA_TUN_OWNER,
2266 from_kuid_munged(current_user_ns(), tun->owner)))
2267 goto nla_put_failure;
2268 if (gid_valid(tun->group) &&
2269 nla_put_u32(skb, IFLA_TUN_GROUP,
2270 from_kgid_munged(current_user_ns(), tun->group)))
2271 goto nla_put_failure;
2272 if (nla_put_u8(skb, IFLA_TUN_PI, !(tun->flags & IFF_NO_PI)))
2273 goto nla_put_failure;
2274 if (nla_put_u8(skb, IFLA_TUN_VNET_HDR, !!(tun->flags & IFF_VNET_HDR)))
2275 goto nla_put_failure;
2276 if (nla_put_u8(skb, IFLA_TUN_PERSIST, !!(tun->flags & IFF_PERSIST)))
2277 goto nla_put_failure;
2278 if (nla_put_u8(skb, IFLA_TUN_MULTI_QUEUE,
2279 !!(tun->flags & IFF_MULTI_QUEUE)))
2280 goto nla_put_failure;
2281 if (tun->flags & IFF_MULTI_QUEUE) {
2282 if (nla_put_u32(skb, IFLA_TUN_NUM_QUEUES, tun->numqueues))
2283 goto nla_put_failure;
2284 if (nla_put_u32(skb, IFLA_TUN_NUM_DISABLED_QUEUES,
2285 tun->numdisabled))
2286 goto nla_put_failure;
2287 }
2288
2289 return 0;
2290
2291nla_put_failure:
2292 return -EMSGSIZE;
2293}
2294
2295static struct rtnl_link_ops tun_link_ops __read_mostly = {
2296 .kind = DRV_NAME,
2297 .priv_size = sizeof(struct tun_struct),
2298 .setup = tun_setup,
2299 .validate = tun_validate,
2300 .get_size = tun_get_size,
2301 .fill_info = tun_fill_info,
2302};
2303
2304static void tun_sock_write_space(struct sock *sk)
2305{
2306 struct tun_file *tfile;
2307 wait_queue_head_t *wqueue;
2308
2309 if (!sock_writeable(sk))
2310 return;
2311
2312 if (!test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags))
2313 return;
2314
2315 wqueue = sk_sleep(sk);
2316 if (wqueue && waitqueue_active(wqueue))
2317 wake_up_interruptible_sync_poll(wqueue, EPOLLOUT |
2318 EPOLLWRNORM | EPOLLWRBAND);
2319
2320 tfile = container_of(sk, struct tun_file, sk);
2321 kill_fasync(&tfile->fasync, SIGIO, POLL_OUT);
2322}
2323
2324static void tun_put_page(struct tun_page *tpage)
2325{
2326 if (tpage->page)
2327 __page_frag_cache_drain(tpage->page, tpage->count);
2328}
2329
2330static int tun_xdp_one(struct tun_struct *tun,
2331 struct tun_file *tfile,
2332 struct xdp_buff *xdp, int *flush,
2333 struct tun_page *tpage)
2334{
2335 unsigned int datasize = xdp->data_end - xdp->data;
2336 struct tun_xdp_hdr *hdr = xdp->data_hard_start;
2337 struct virtio_net_hdr *gso = &hdr->gso;
2338 struct bpf_prog *xdp_prog;
2339 struct sk_buff *skb = NULL;
2340 u32 rxhash = 0, act;
2341 int buflen = hdr->buflen;
2342 int err = 0;
2343 bool skb_xdp = false;
2344 struct page *page;
2345
2346 xdp_prog = rcu_dereference(tun->xdp_prog);
2347 if (xdp_prog) {
2348 if (gso->gso_type) {
2349 skb_xdp = true;
2350 goto build;
2351 }
2352
2353 xdp_init_buff(xdp, buflen, &tfile->xdp_rxq);
2354 xdp_set_data_meta_invalid(xdp);
2355
2356 act = bpf_prog_run_xdp(xdp_prog, xdp);
2357 err = tun_xdp_act(tun, xdp_prog, xdp, act);
2358 if (err < 0) {
2359 put_page(virt_to_head_page(xdp->data));
2360 return err;
2361 }
2362
2363 switch (err) {
2364 case XDP_REDIRECT:
2365 *flush = true;
2366 fallthrough;
2367 case XDP_TX:
2368 return 0;
2369 case XDP_PASS:
2370 break;
2371 default:
2372 page = virt_to_head_page(xdp->data);
2373 if (tpage->page == page) {
2374 ++tpage->count;
2375 } else {
2376 tun_put_page(tpage);
2377 tpage->page = page;
2378 tpage->count = 1;
2379 }
2380 return 0;
2381 }
2382 }
2383
2384build:
2385 skb = build_skb(xdp->data_hard_start, buflen);
2386 if (!skb) {
2387 err = -ENOMEM;
2388 goto out;
2389 }
2390
2391 skb_reserve(skb, xdp->data - xdp->data_hard_start);
2392 skb_put(skb, xdp->data_end - xdp->data);
2393
2394 if (virtio_net_hdr_to_skb(skb, gso, tun_is_little_endian(tun))) {
2395 atomic_long_inc(&tun->rx_frame_errors);
2396 kfree_skb(skb);
2397 err = -EINVAL;
2398 goto out;
2399 }
2400
2401 skb->protocol = eth_type_trans(skb, tun->dev);
2402 skb_reset_network_header(skb);
2403 skb_probe_transport_header(skb);
2404 skb_record_rx_queue(skb, tfile->queue_index);
2405
2406 if (skb_xdp) {
2407 err = do_xdp_generic(xdp_prog, skb);
2408 if (err != XDP_PASS)
2409 goto out;
2410 }
2411
2412 if (!rcu_dereference(tun->steering_prog) && tun->numqueues > 1 &&
2413 !tfile->detached)
2414 rxhash = __skb_get_hash_symmetric(skb);
2415
2416 netif_receive_skb(skb);
2417
2418 /* No need to disable preemption here since this function is
2419 * always called with bh disabled
2420 */
2421 dev_sw_netstats_rx_add(tun->dev, datasize);
2422
2423 if (rxhash)
2424 tun_flow_update(tun, rxhash, tfile);
2425
2426out:
2427 return err;
2428}
2429
2430static int tun_sendmsg(struct socket *sock, struct msghdr *m, size_t total_len)
2431{
2432 int ret, i;
2433 struct tun_file *tfile = container_of(sock, struct tun_file, socket);
2434 struct tun_struct *tun = tun_get(tfile);
2435 struct tun_msg_ctl *ctl = m->msg_control;
2436 struct xdp_buff *xdp;
2437
2438 if (!tun)
2439 return -EBADFD;
2440
2441 if (ctl && (ctl->type == TUN_MSG_PTR)) {
2442 struct tun_page tpage;
2443 int n = ctl->num;
2444 int flush = 0;
2445
2446 memset(&tpage, 0, sizeof(tpage));
2447
2448 local_bh_disable();
2449 rcu_read_lock();
2450
2451 for (i = 0; i < n; i++) {
2452 xdp = &((struct xdp_buff *)ctl->ptr)[i];
2453 tun_xdp_one(tun, tfile, xdp, &flush, &tpage);
2454 }
2455
2456 if (flush)
2457 xdp_do_flush();
2458
2459 rcu_read_unlock();
2460 local_bh_enable();
2461
2462 tun_put_page(&tpage);
2463
2464 ret = total_len;
2465 goto out;
2466 }
2467
2468 ret = tun_get_user(tun, tfile, ctl ? ctl->ptr : NULL, &m->msg_iter,
2469 m->msg_flags & MSG_DONTWAIT,
2470 m->msg_flags & MSG_MORE);
2471out:
2472 tun_put(tun);
2473 return ret;
2474}
2475
2476static int tun_recvmsg(struct socket *sock, struct msghdr *m, size_t total_len,
2477 int flags)
2478{
2479 struct tun_file *tfile = container_of(sock, struct tun_file, socket);
2480 struct tun_struct *tun = tun_get(tfile);
2481 void *ptr = m->msg_control;
2482 int ret;
2483
2484 if (!tun) {
2485 ret = -EBADFD;
2486 goto out_free;
2487 }
2488
2489 if (flags & ~(MSG_DONTWAIT|MSG_TRUNC|MSG_ERRQUEUE)) {
2490 ret = -EINVAL;
2491 goto out_put_tun;
2492 }
2493 if (flags & MSG_ERRQUEUE) {
2494 ret = sock_recv_errqueue(sock->sk, m, total_len,
2495 SOL_PACKET, TUN_TX_TIMESTAMP);
2496 goto out;
2497 }
2498 ret = tun_do_read(tun, tfile, &m->msg_iter, flags & MSG_DONTWAIT, ptr);
2499 if (ret > (ssize_t)total_len) {
2500 m->msg_flags |= MSG_TRUNC;
2501 ret = flags & MSG_TRUNC ? ret : total_len;
2502 }
2503out:
2504 tun_put(tun);
2505 return ret;
2506
2507out_put_tun:
2508 tun_put(tun);
2509out_free:
2510 tun_ptr_free(ptr);
2511 return ret;
2512}
2513
2514static int tun_ptr_peek_len(void *ptr)
2515{
2516 if (likely(ptr)) {
2517 if (tun_is_xdp_frame(ptr)) {
2518 struct xdp_frame *xdpf = tun_ptr_to_xdp(ptr);
2519
2520 return xdpf->len;
2521 }
2522 return __skb_array_len_with_tag(ptr);
2523 } else {
2524 return 0;
2525 }
2526}
2527
2528static int tun_peek_len(struct socket *sock)
2529{
2530 struct tun_file *tfile = container_of(sock, struct tun_file, socket);
2531 struct tun_struct *tun;
2532 int ret = 0;
2533
2534 tun = tun_get(tfile);
2535 if (!tun)
2536 return 0;
2537
2538 ret = PTR_RING_PEEK_CALL(&tfile->tx_ring, tun_ptr_peek_len);
2539 tun_put(tun);
2540
2541 return ret;
2542}
2543
2544/* Ops structure to mimic raw sockets with tun */
2545static const struct proto_ops tun_socket_ops = {
2546 .peek_len = tun_peek_len,
2547 .sendmsg = tun_sendmsg,
2548 .recvmsg = tun_recvmsg,
2549};
2550
2551static struct proto tun_proto = {
2552 .name = "tun",
2553 .owner = THIS_MODULE,
2554 .obj_size = sizeof(struct tun_file),
2555};
2556
2557static int tun_flags(struct tun_struct *tun)
2558{
2559 return tun->flags & (TUN_FEATURES | IFF_PERSIST | IFF_TUN | IFF_TAP);
2560}
2561
2562static ssize_t tun_flags_show(struct device *dev, struct device_attribute *attr,
2563 char *buf)
2564{
2565 struct tun_struct *tun = netdev_priv(to_net_dev(dev));
2566 return sprintf(buf, "0x%x\n", tun_flags(tun));
2567}
2568
2569static ssize_t owner_show(struct device *dev, struct device_attribute *attr,
2570 char *buf)
2571{
2572 struct tun_struct *tun = netdev_priv(to_net_dev(dev));
2573 return uid_valid(tun->owner)?
2574 sprintf(buf, "%u\n",
2575 from_kuid_munged(current_user_ns(), tun->owner)):
2576 sprintf(buf, "-1\n");
2577}
2578
2579static ssize_t group_show(struct device *dev, struct device_attribute *attr,
2580 char *buf)
2581{
2582 struct tun_struct *tun = netdev_priv(to_net_dev(dev));
2583 return gid_valid(tun->group) ?
2584 sprintf(buf, "%u\n",
2585 from_kgid_munged(current_user_ns(), tun->group)):
2586 sprintf(buf, "-1\n");
2587}
2588
2589static DEVICE_ATTR_RO(tun_flags);
2590static DEVICE_ATTR_RO(owner);
2591static DEVICE_ATTR_RO(group);
2592
2593static struct attribute *tun_dev_attrs[] = {
2594 &dev_attr_tun_flags.attr,
2595 &dev_attr_owner.attr,
2596 &dev_attr_group.attr,
2597 NULL
2598};
2599
2600static const struct attribute_group tun_attr_group = {
2601 .attrs = tun_dev_attrs
2602};
2603
2604static int tun_set_iff(struct net *net, struct file *file, struct ifreq *ifr)
2605{
2606 struct tun_struct *tun;
2607 struct tun_file *tfile = file->private_data;
2608 struct net_device *dev;
2609 int err;
2610
2611 if (tfile->detached)
2612 return -EINVAL;
2613
2614 if ((ifr->ifr_flags & IFF_NAPI_FRAGS)) {
2615 if (!capable(CAP_NET_ADMIN))
2616 return -EPERM;
2617
2618 if (!(ifr->ifr_flags & IFF_NAPI) ||
2619 (ifr->ifr_flags & TUN_TYPE_MASK) != IFF_TAP)
2620 return -EINVAL;
2621 }
2622
2623 dev = __dev_get_by_name(net, ifr->ifr_name);
2624 if (dev) {
2625 if (ifr->ifr_flags & IFF_TUN_EXCL)
2626 return -EBUSY;
2627 if ((ifr->ifr_flags & IFF_TUN) && dev->netdev_ops == &tun_netdev_ops)
2628 tun = netdev_priv(dev);
2629 else if ((ifr->ifr_flags & IFF_TAP) && dev->netdev_ops == &tap_netdev_ops)
2630 tun = netdev_priv(dev);
2631 else
2632 return -EINVAL;
2633
2634 if (!!(ifr->ifr_flags & IFF_MULTI_QUEUE) !=
2635 !!(tun->flags & IFF_MULTI_QUEUE))
2636 return -EINVAL;
2637
2638 if (tun_not_capable(tun))
2639 return -EPERM;
2640 err = security_tun_dev_open(tun->security);
2641 if (err < 0)
2642 return err;
2643
2644 err = tun_attach(tun, file, ifr->ifr_flags & IFF_NOFILTER,
2645 ifr->ifr_flags & IFF_NAPI,
2646 ifr->ifr_flags & IFF_NAPI_FRAGS, true);
2647 if (err < 0)
2648 return err;
2649
2650 if (tun->flags & IFF_MULTI_QUEUE &&
2651 (tun->numqueues + tun->numdisabled > 1)) {
2652 /* One or more queue has already been attached, no need
2653 * to initialize the device again.
2654 */
2655 netdev_state_change(dev);
2656 return 0;
2657 }
2658
2659 tun->flags = (tun->flags & ~TUN_FEATURES) |
2660 (ifr->ifr_flags & TUN_FEATURES);
2661
2662 netdev_state_change(dev);
2663 } else {
2664 char *name;
2665 unsigned long flags = 0;
2666 int queues = ifr->ifr_flags & IFF_MULTI_QUEUE ?
2667 MAX_TAP_QUEUES : 1;
2668
2669 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
2670 return -EPERM;
2671 err = security_tun_dev_create();
2672 if (err < 0)
2673 return err;
2674
2675 /* Set dev type */
2676 if (ifr->ifr_flags & IFF_TUN) {
2677 /* TUN device */
2678 flags |= IFF_TUN;
2679 name = "tun%d";
2680 } else if (ifr->ifr_flags & IFF_TAP) {
2681 /* TAP device */
2682 flags |= IFF_TAP;
2683 name = "tap%d";
2684 } else
2685 return -EINVAL;
2686
2687 if (*ifr->ifr_name)
2688 name = ifr->ifr_name;
2689
2690 dev = alloc_netdev_mqs(sizeof(struct tun_struct), name,
2691 NET_NAME_UNKNOWN, tun_setup, queues,
2692 queues);
2693
2694 if (!dev)
2695 return -ENOMEM;
2696
2697 dev_net_set(dev, net);
2698 dev->rtnl_link_ops = &tun_link_ops;
2699 dev->ifindex = tfile->ifindex;
2700 dev->sysfs_groups[0] = &tun_attr_group;
2701
2702 tun = netdev_priv(dev);
2703 tun->dev = dev;
2704 tun->flags = flags;
2705 tun->txflt.count = 0;
2706 tun->vnet_hdr_sz = sizeof(struct virtio_net_hdr);
2707
2708 tun->align = NET_SKB_PAD;
2709 tun->filter_attached = false;
2710 tun->sndbuf = tfile->socket.sk->sk_sndbuf;
2711 tun->rx_batched = 0;
2712 RCU_INIT_POINTER(tun->steering_prog, NULL);
2713
2714 dev->tstats = netdev_alloc_pcpu_stats(struct pcpu_sw_netstats);
2715 if (!dev->tstats) {
2716 err = -ENOMEM;
2717 goto err_free_dev;
2718 }
2719
2720 spin_lock_init(&tun->lock);
2721
2722 err = security_tun_dev_alloc_security(&tun->security);
2723 if (err < 0)
2724 goto err_free_stat;
2725
2726 tun_net_init(dev);
2727 tun_flow_init(tun);
2728
2729 dev->hw_features = NETIF_F_SG | NETIF_F_FRAGLIST |
2730 TUN_USER_FEATURES | NETIF_F_HW_VLAN_CTAG_TX |
2731 NETIF_F_HW_VLAN_STAG_TX;
2732 dev->features = dev->hw_features | NETIF_F_LLTX;
2733 dev->vlan_features = dev->features &
2734 ~(NETIF_F_HW_VLAN_CTAG_TX |
2735 NETIF_F_HW_VLAN_STAG_TX);
2736
2737 tun->flags = (tun->flags & ~TUN_FEATURES) |
2738 (ifr->ifr_flags & TUN_FEATURES);
2739
2740 INIT_LIST_HEAD(&tun->disabled);
2741 err = tun_attach(tun, file, false, ifr->ifr_flags & IFF_NAPI,
2742 ifr->ifr_flags & IFF_NAPI_FRAGS, false);
2743 if (err < 0)
2744 goto err_free_flow;
2745
2746 err = register_netdevice(tun->dev);
2747 if (err < 0)
2748 goto err_detach;
2749 /* free_netdev() won't check refcnt, to avoid race
2750 * with dev_put() we need publish tun after registration.
2751 */
2752 rcu_assign_pointer(tfile->tun, tun);
2753 }
2754
2755 netif_carrier_on(tun->dev);
2756
2757 /* Make sure persistent devices do not get stuck in
2758 * xoff state.
2759 */
2760 if (netif_running(tun->dev))
2761 netif_tx_wake_all_queues(tun->dev);
2762
2763 strcpy(ifr->ifr_name, tun->dev->name);
2764 return 0;
2765
2766err_detach:
2767 tun_detach_all(dev);
2768 /* We are here because register_netdevice() has failed.
2769 * If register_netdevice() already called tun_free_netdev()
2770 * while dealing with the error, dev->stats has been cleared.
2771 */
2772 if (!dev->tstats)
2773 goto err_free_dev;
2774
2775err_free_flow:
2776 tun_flow_uninit(tun);
2777 security_tun_dev_free_security(tun->security);
2778err_free_stat:
2779 free_percpu(dev->tstats);
2780err_free_dev:
2781 free_netdev(dev);
2782 return err;
2783}
2784
2785static void tun_get_iff(struct tun_struct *tun, struct ifreq *ifr)
2786{
2787 strcpy(ifr->ifr_name, tun->dev->name);
2788
2789 ifr->ifr_flags = tun_flags(tun);
2790
2791}
2792
2793/* This is like a cut-down ethtool ops, except done via tun fd so no
2794 * privs required. */
2795static int set_offload(struct tun_struct *tun, unsigned long arg)
2796{
2797 netdev_features_t features = 0;
2798
2799 if (arg & TUN_F_CSUM) {
2800 features |= NETIF_F_HW_CSUM;
2801 arg &= ~TUN_F_CSUM;
2802
2803 if (arg & (TUN_F_TSO4|TUN_F_TSO6)) {
2804 if (arg & TUN_F_TSO_ECN) {
2805 features |= NETIF_F_TSO_ECN;
2806 arg &= ~TUN_F_TSO_ECN;
2807 }
2808 if (arg & TUN_F_TSO4)
2809 features |= NETIF_F_TSO;
2810 if (arg & TUN_F_TSO6)
2811 features |= NETIF_F_TSO6;
2812 arg &= ~(TUN_F_TSO4|TUN_F_TSO6);
2813 }
2814
2815 arg &= ~TUN_F_UFO;
2816 }
2817
2818 /* This gives the user a way to test for new features in future by
2819 * trying to set them. */
2820 if (arg)
2821 return -EINVAL;
2822
2823 tun->set_features = features;
2824 tun->dev->wanted_features &= ~TUN_USER_FEATURES;
2825 tun->dev->wanted_features |= features;
2826 netdev_update_features(tun->dev);
2827
2828 return 0;
2829}
2830
2831static void tun_detach_filter(struct tun_struct *tun, int n)
2832{
2833 int i;
2834 struct tun_file *tfile;
2835
2836 for (i = 0; i < n; i++) {
2837 tfile = rtnl_dereference(tun->tfiles[i]);
2838 lock_sock(tfile->socket.sk);
2839 sk_detach_filter(tfile->socket.sk);
2840 release_sock(tfile->socket.sk);
2841 }
2842
2843 tun->filter_attached = false;
2844}
2845
2846static int tun_attach_filter(struct tun_struct *tun)
2847{
2848 int i, ret = 0;
2849 struct tun_file *tfile;
2850
2851 for (i = 0; i < tun->numqueues; i++) {
2852 tfile = rtnl_dereference(tun->tfiles[i]);
2853 lock_sock(tfile->socket.sk);
2854 ret = sk_attach_filter(&tun->fprog, tfile->socket.sk);
2855 release_sock(tfile->socket.sk);
2856 if (ret) {
2857 tun_detach_filter(tun, i);
2858 return ret;
2859 }
2860 }
2861
2862 tun->filter_attached = true;
2863 return ret;
2864}
2865
2866static void tun_set_sndbuf(struct tun_struct *tun)
2867{
2868 struct tun_file *tfile;
2869 int i;
2870
2871 for (i = 0; i < tun->numqueues; i++) {
2872 tfile = rtnl_dereference(tun->tfiles[i]);
2873 tfile->socket.sk->sk_sndbuf = tun->sndbuf;
2874 }
2875}
2876
2877static int tun_set_queue(struct file *file, struct ifreq *ifr)
2878{
2879 struct tun_file *tfile = file->private_data;
2880 struct tun_struct *tun;
2881 int ret = 0;
2882
2883 rtnl_lock();
2884
2885 if (ifr->ifr_flags & IFF_ATTACH_QUEUE) {
2886 tun = tfile->detached;
2887 if (!tun) {
2888 ret = -EINVAL;
2889 goto unlock;
2890 }
2891 ret = security_tun_dev_attach_queue(tun->security);
2892 if (ret < 0)
2893 goto unlock;
2894 ret = tun_attach(tun, file, false, tun->flags & IFF_NAPI,
2895 tun->flags & IFF_NAPI_FRAGS, true);
2896 } else if (ifr->ifr_flags & IFF_DETACH_QUEUE) {
2897 tun = rtnl_dereference(tfile->tun);
2898 if (!tun || !(tun->flags & IFF_MULTI_QUEUE) || tfile->detached)
2899 ret = -EINVAL;
2900 else
2901 __tun_detach(tfile, false);
2902 } else
2903 ret = -EINVAL;
2904
2905 if (ret >= 0)
2906 netdev_state_change(tun->dev);
2907
2908unlock:
2909 rtnl_unlock();
2910 return ret;
2911}
2912
2913static int tun_set_ebpf(struct tun_struct *tun, struct tun_prog __rcu **prog_p,
2914 void __user *data)
2915{
2916 struct bpf_prog *prog;
2917 int fd;
2918
2919 if (copy_from_user(&fd, data, sizeof(fd)))
2920 return -EFAULT;
2921
2922 if (fd == -1) {
2923 prog = NULL;
2924 } else {
2925 prog = bpf_prog_get_type(fd, BPF_PROG_TYPE_SOCKET_FILTER);
2926 if (IS_ERR(prog))
2927 return PTR_ERR(prog);
2928 }
2929
2930 return __tun_set_ebpf(tun, prog_p, prog);
2931}
2932
2933/* Return correct value for tun->dev->addr_len based on tun->dev->type. */
2934static unsigned char tun_get_addr_len(unsigned short type)
2935{
2936 switch (type) {
2937 case ARPHRD_IP6GRE:
2938 case ARPHRD_TUNNEL6:
2939 return sizeof(struct in6_addr);
2940 case ARPHRD_IPGRE:
2941 case ARPHRD_TUNNEL:
2942 case ARPHRD_SIT:
2943 return 4;
2944 case ARPHRD_ETHER:
2945 return ETH_ALEN;
2946 case ARPHRD_IEEE802154:
2947 case ARPHRD_IEEE802154_MONITOR:
2948 return IEEE802154_EXTENDED_ADDR_LEN;
2949 case ARPHRD_PHONET_PIPE:
2950 case ARPHRD_PPP:
2951 case ARPHRD_NONE:
2952 return 0;
2953 case ARPHRD_6LOWPAN:
2954 return EUI64_ADDR_LEN;
2955 case ARPHRD_FDDI:
2956 return FDDI_K_ALEN;
2957 case ARPHRD_HIPPI:
2958 return HIPPI_ALEN;
2959 case ARPHRD_IEEE802:
2960 return FC_ALEN;
2961 case ARPHRD_ROSE:
2962 return ROSE_ADDR_LEN;
2963 case ARPHRD_NETROM:
2964 return AX25_ADDR_LEN;
2965 case ARPHRD_LOCALTLK:
2966 return LTALK_ALEN;
2967 default:
2968 return 0;
2969 }
2970}
2971
2972static long __tun_chr_ioctl(struct file *file, unsigned int cmd,
2973 unsigned long arg, int ifreq_len)
2974{
2975 struct tun_file *tfile = file->private_data;
2976 struct net *net = sock_net(&tfile->sk);
2977 struct tun_struct *tun;
2978 void __user* argp = (void __user*)arg;
2979 unsigned int ifindex, carrier;
2980 struct ifreq ifr;
2981 kuid_t owner;
2982 kgid_t group;
2983 int sndbuf;
2984 int vnet_hdr_sz;
2985 int le;
2986 int ret;
2987 bool do_notify = false;
2988
2989 if (cmd == TUNSETIFF || cmd == TUNSETQUEUE ||
2990 (_IOC_TYPE(cmd) == SOCK_IOC_TYPE && cmd != SIOCGSKNS)) {
2991 if (copy_from_user(&ifr, argp, ifreq_len))
2992 return -EFAULT;
2993 } else {
2994 memset(&ifr, 0, sizeof(ifr));
2995 }
2996 if (cmd == TUNGETFEATURES) {
2997 /* Currently this just means: "what IFF flags are valid?".
2998 * This is needed because we never checked for invalid flags on
2999 * TUNSETIFF.
3000 */
3001 return put_user(IFF_TUN | IFF_TAP | TUN_FEATURES,
3002 (unsigned int __user*)argp);
3003 } else if (cmd == TUNSETQUEUE) {
3004 return tun_set_queue(file, &ifr);
3005 } else if (cmd == SIOCGSKNS) {
3006 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
3007 return -EPERM;
3008 return open_related_ns(&net->ns, get_net_ns);
3009 }
3010
3011 rtnl_lock();
3012
3013 tun = tun_get(tfile);
3014 if (cmd == TUNSETIFF) {
3015 ret = -EEXIST;
3016 if (tun)
3017 goto unlock;
3018
3019 ifr.ifr_name[IFNAMSIZ-1] = '\0';
3020
3021 ret = tun_set_iff(net, file, &ifr);
3022
3023 if (ret)
3024 goto unlock;
3025
3026 if (copy_to_user(argp, &ifr, ifreq_len))
3027 ret = -EFAULT;
3028 goto unlock;
3029 }
3030 if (cmd == TUNSETIFINDEX) {
3031 ret = -EPERM;
3032 if (tun)
3033 goto unlock;
3034
3035 ret = -EFAULT;
3036 if (copy_from_user(&ifindex, argp, sizeof(ifindex)))
3037 goto unlock;
3038
3039 ret = 0;
3040 tfile->ifindex = ifindex;
3041 goto unlock;
3042 }
3043
3044 ret = -EBADFD;
3045 if (!tun)
3046 goto unlock;
3047
3048 netif_info(tun, drv, tun->dev, "tun_chr_ioctl cmd %u\n", cmd);
3049
3050 net = dev_net(tun->dev);
3051 ret = 0;
3052 switch (cmd) {
3053 case TUNGETIFF:
3054 tun_get_iff(tun, &ifr);
3055
3056 if (tfile->detached)
3057 ifr.ifr_flags |= IFF_DETACH_QUEUE;
3058 if (!tfile->socket.sk->sk_filter)
3059 ifr.ifr_flags |= IFF_NOFILTER;
3060
3061 if (copy_to_user(argp, &ifr, ifreq_len))
3062 ret = -EFAULT;
3063 break;
3064
3065 case TUNSETNOCSUM:
3066 /* Disable/Enable checksum */
3067
3068 /* [unimplemented] */
3069 netif_info(tun, drv, tun->dev, "ignored: set checksum %s\n",
3070 arg ? "disabled" : "enabled");
3071 break;
3072
3073 case TUNSETPERSIST:
3074 /* Disable/Enable persist mode. Keep an extra reference to the
3075 * module to prevent the module being unprobed.
3076 */
3077 if (arg && !(tun->flags & IFF_PERSIST)) {
3078 tun->flags |= IFF_PERSIST;
3079 __module_get(THIS_MODULE);
3080 do_notify = true;
3081 }
3082 if (!arg && (tun->flags & IFF_PERSIST)) {
3083 tun->flags &= ~IFF_PERSIST;
3084 module_put(THIS_MODULE);
3085 do_notify = true;
3086 }
3087
3088 netif_info(tun, drv, tun->dev, "persist %s\n",
3089 arg ? "enabled" : "disabled");
3090 break;
3091
3092 case TUNSETOWNER:
3093 /* Set owner of the device */
3094 owner = make_kuid(current_user_ns(), arg);
3095 if (!uid_valid(owner)) {
3096 ret = -EINVAL;
3097 break;
3098 }
3099 tun->owner = owner;
3100 do_notify = true;
3101 netif_info(tun, drv, tun->dev, "owner set to %u\n",
3102 from_kuid(&init_user_ns, tun->owner));
3103 break;
3104
3105 case TUNSETGROUP:
3106 /* Set group of the device */
3107 group = make_kgid(current_user_ns(), arg);
3108 if (!gid_valid(group)) {
3109 ret = -EINVAL;
3110 break;
3111 }
3112 tun->group = group;
3113 do_notify = true;
3114 netif_info(tun, drv, tun->dev, "group set to %u\n",
3115 from_kgid(&init_user_ns, tun->group));
3116 break;
3117
3118 case TUNSETLINK:
3119 /* Only allow setting the type when the interface is down */
3120 if (tun->dev->flags & IFF_UP) {
3121 netif_info(tun, drv, tun->dev,
3122 "Linktype set failed because interface is up\n");
3123 ret = -EBUSY;
3124 } else {
3125 ret = call_netdevice_notifiers(NETDEV_PRE_TYPE_CHANGE,
3126 tun->dev);
3127 ret = notifier_to_errno(ret);
3128 if (ret) {
3129 netif_info(tun, drv, tun->dev,
3130 "Refused to change device type\n");
3131 break;
3132 }
3133 tun->dev->type = (int) arg;
3134 tun->dev->addr_len = tun_get_addr_len(tun->dev->type);
3135 netif_info(tun, drv, tun->dev, "linktype set to %d\n",
3136 tun->dev->type);
3137 call_netdevice_notifiers(NETDEV_POST_TYPE_CHANGE,
3138 tun->dev);
3139 }
3140 break;
3141
3142 case TUNSETDEBUG:
3143 tun->msg_enable = (u32)arg;
3144 break;
3145
3146 case TUNSETOFFLOAD:
3147 ret = set_offload(tun, arg);
3148 break;
3149
3150 case TUNSETTXFILTER:
3151 /* Can be set only for TAPs */
3152 ret = -EINVAL;
3153 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3154 break;
3155 ret = update_filter(&tun->txflt, (void __user *)arg);
3156 break;
3157
3158 case SIOCGIFHWADDR:
3159 /* Get hw address */
3160 dev_get_mac_address(&ifr.ifr_hwaddr, net, tun->dev->name);
3161 if (copy_to_user(argp, &ifr, ifreq_len))
3162 ret = -EFAULT;
3163 break;
3164
3165 case SIOCSIFHWADDR:
3166 /* Set hw address */
3167 ret = dev_set_mac_address_user(tun->dev, &ifr.ifr_hwaddr, NULL);
3168 break;
3169
3170 case TUNGETSNDBUF:
3171 sndbuf = tfile->socket.sk->sk_sndbuf;
3172 if (copy_to_user(argp, &sndbuf, sizeof(sndbuf)))
3173 ret = -EFAULT;
3174 break;
3175
3176 case TUNSETSNDBUF:
3177 if (copy_from_user(&sndbuf, argp, sizeof(sndbuf))) {
3178 ret = -EFAULT;
3179 break;
3180 }
3181 if (sndbuf <= 0) {
3182 ret = -EINVAL;
3183 break;
3184 }
3185
3186 tun->sndbuf = sndbuf;
3187 tun_set_sndbuf(tun);
3188 break;
3189
3190 case TUNGETVNETHDRSZ:
3191 vnet_hdr_sz = tun->vnet_hdr_sz;
3192 if (copy_to_user(argp, &vnet_hdr_sz, sizeof(vnet_hdr_sz)))
3193 ret = -EFAULT;
3194 break;
3195
3196 case TUNSETVNETHDRSZ:
3197 if (copy_from_user(&vnet_hdr_sz, argp, sizeof(vnet_hdr_sz))) {
3198 ret = -EFAULT;
3199 break;
3200 }
3201 if (vnet_hdr_sz < (int)sizeof(struct virtio_net_hdr)) {
3202 ret = -EINVAL;
3203 break;
3204 }
3205
3206 tun->vnet_hdr_sz = vnet_hdr_sz;
3207 break;
3208
3209 case TUNGETVNETLE:
3210 le = !!(tun->flags & TUN_VNET_LE);
3211 if (put_user(le, (int __user *)argp))
3212 ret = -EFAULT;
3213 break;
3214
3215 case TUNSETVNETLE:
3216 if (get_user(le, (int __user *)argp)) {
3217 ret = -EFAULT;
3218 break;
3219 }
3220 if (le)
3221 tun->flags |= TUN_VNET_LE;
3222 else
3223 tun->flags &= ~TUN_VNET_LE;
3224 break;
3225
3226 case TUNGETVNETBE:
3227 ret = tun_get_vnet_be(tun, argp);
3228 break;
3229
3230 case TUNSETVNETBE:
3231 ret = tun_set_vnet_be(tun, argp);
3232 break;
3233
3234 case TUNATTACHFILTER:
3235 /* Can be set only for TAPs */
3236 ret = -EINVAL;
3237 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3238 break;
3239 ret = -EFAULT;
3240 if (copy_from_user(&tun->fprog, argp, sizeof(tun->fprog)))
3241 break;
3242
3243 ret = tun_attach_filter(tun);
3244 break;
3245
3246 case TUNDETACHFILTER:
3247 /* Can be set only for TAPs */
3248 ret = -EINVAL;
3249 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3250 break;
3251 ret = 0;
3252 tun_detach_filter(tun, tun->numqueues);
3253 break;
3254
3255 case TUNGETFILTER:
3256 ret = -EINVAL;
3257 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3258 break;
3259 ret = -EFAULT;
3260 if (copy_to_user(argp, &tun->fprog, sizeof(tun->fprog)))
3261 break;
3262 ret = 0;
3263 break;
3264
3265 case TUNSETSTEERINGEBPF:
3266 ret = tun_set_ebpf(tun, &tun->steering_prog, argp);
3267 break;
3268
3269 case TUNSETFILTEREBPF:
3270 ret = tun_set_ebpf(tun, &tun->filter_prog, argp);
3271 break;
3272
3273 case TUNSETCARRIER:
3274 ret = -EFAULT;
3275 if (copy_from_user(&carrier, argp, sizeof(carrier)))
3276 goto unlock;
3277
3278 ret = tun_net_change_carrier(tun->dev, (bool)carrier);
3279 break;
3280
3281 case TUNGETDEVNETNS:
3282 ret = -EPERM;
3283 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
3284 goto unlock;
3285 ret = open_related_ns(&net->ns, get_net_ns);
3286 break;
3287
3288 default:
3289 ret = -EINVAL;
3290 break;
3291 }
3292
3293 if (do_notify)
3294 netdev_state_change(tun->dev);
3295
3296unlock:
3297 rtnl_unlock();
3298 if (tun)
3299 tun_put(tun);
3300 return ret;
3301}
3302
3303static long tun_chr_ioctl(struct file *file,
3304 unsigned int cmd, unsigned long arg)
3305{
3306 return __tun_chr_ioctl(file, cmd, arg, sizeof (struct ifreq));
3307}
3308
3309#ifdef CONFIG_COMPAT
3310static long tun_chr_compat_ioctl(struct file *file,
3311 unsigned int cmd, unsigned long arg)
3312{
3313 switch (cmd) {
3314 case TUNSETIFF:
3315 case TUNGETIFF:
3316 case TUNSETTXFILTER:
3317 case TUNGETSNDBUF:
3318 case TUNSETSNDBUF:
3319 case SIOCGIFHWADDR:
3320 case SIOCSIFHWADDR:
3321 arg = (unsigned long)compat_ptr(arg);
3322 break;
3323 default:
3324 arg = (compat_ulong_t)arg;
3325 break;
3326 }
3327
3328 /*
3329 * compat_ifreq is shorter than ifreq, so we must not access beyond
3330 * the end of that structure. All fields that are used in this
3331 * driver are compatible though, we don't need to convert the
3332 * contents.
3333 */
3334 return __tun_chr_ioctl(file, cmd, arg, sizeof(struct compat_ifreq));
3335}
3336#endif /* CONFIG_COMPAT */
3337
3338static int tun_chr_fasync(int fd, struct file *file, int on)
3339{
3340 struct tun_file *tfile = file->private_data;
3341 int ret;
3342
3343 if ((ret = fasync_helper(fd, file, on, &tfile->fasync)) < 0)
3344 goto out;
3345
3346 if (on) {
3347 __f_setown(file, task_pid(current), PIDTYPE_TGID, 0);
3348 tfile->flags |= TUN_FASYNC;
3349 } else
3350 tfile->flags &= ~TUN_FASYNC;
3351 ret = 0;
3352out:
3353 return ret;
3354}
3355
3356static int tun_chr_open(struct inode *inode, struct file * file)
3357{
3358 struct net *net = current->nsproxy->net_ns;
3359 struct tun_file *tfile;
3360
3361 tfile = (struct tun_file *)sk_alloc(net, AF_UNSPEC, GFP_KERNEL,
3362 &tun_proto, 0);
3363 if (!tfile)
3364 return -ENOMEM;
3365 if (ptr_ring_init(&tfile->tx_ring, 0, GFP_KERNEL)) {
3366 sk_free(&tfile->sk);
3367 return -ENOMEM;
3368 }
3369
3370 mutex_init(&tfile->napi_mutex);
3371 RCU_INIT_POINTER(tfile->tun, NULL);
3372 tfile->flags = 0;
3373 tfile->ifindex = 0;
3374
3375 init_waitqueue_head(&tfile->socket.wq.wait);
3376
3377 tfile->socket.file = file;
3378 tfile->socket.ops = &tun_socket_ops;
3379
3380 sock_init_data(&tfile->socket, &tfile->sk);
3381
3382 tfile->sk.sk_write_space = tun_sock_write_space;
3383 tfile->sk.sk_sndbuf = INT_MAX;
3384
3385 file->private_data = tfile;
3386 INIT_LIST_HEAD(&tfile->next);
3387
3388 sock_set_flag(&tfile->sk, SOCK_ZEROCOPY);
3389
3390 return 0;
3391}
3392
3393static int tun_chr_close(struct inode *inode, struct file *file)
3394{
3395 struct tun_file *tfile = file->private_data;
3396
3397 tun_detach(tfile, true);
3398
3399 return 0;
3400}
3401
3402#ifdef CONFIG_PROC_FS
3403static void tun_chr_show_fdinfo(struct seq_file *m, struct file *file)
3404{
3405 struct tun_file *tfile = file->private_data;
3406 struct tun_struct *tun;
3407 struct ifreq ifr;
3408
3409 memset(&ifr, 0, sizeof(ifr));
3410
3411 rtnl_lock();
3412 tun = tun_get(tfile);
3413 if (tun)
3414 tun_get_iff(tun, &ifr);
3415 rtnl_unlock();
3416
3417 if (tun)
3418 tun_put(tun);
3419
3420 seq_printf(m, "iff:\t%s\n", ifr.ifr_name);
3421}
3422#endif
3423
3424static const struct file_operations tun_fops = {
3425 .owner = THIS_MODULE,
3426 .llseek = no_llseek,
3427 .read_iter = tun_chr_read_iter,
3428 .write_iter = tun_chr_write_iter,
3429 .poll = tun_chr_poll,
3430 .unlocked_ioctl = tun_chr_ioctl,
3431#ifdef CONFIG_COMPAT
3432 .compat_ioctl = tun_chr_compat_ioctl,
3433#endif
3434 .open = tun_chr_open,
3435 .release = tun_chr_close,
3436 .fasync = tun_chr_fasync,
3437#ifdef CONFIG_PROC_FS
3438 .show_fdinfo = tun_chr_show_fdinfo,
3439#endif
3440};
3441
3442static struct miscdevice tun_miscdev = {
3443 .minor = TUN_MINOR,
3444 .name = "tun",
3445 .nodename = "net/tun",
3446 .fops = &tun_fops,
3447};
3448
3449/* ethtool interface */
3450
3451static void tun_default_link_ksettings(struct net_device *dev,
3452 struct ethtool_link_ksettings *cmd)
3453{
3454 ethtool_link_ksettings_zero_link_mode(cmd, supported);
3455 ethtool_link_ksettings_zero_link_mode(cmd, advertising);
3456 cmd->base.speed = SPEED_10;
3457 cmd->base.duplex = DUPLEX_FULL;
3458 cmd->base.port = PORT_TP;
3459 cmd->base.phy_address = 0;
3460 cmd->base.autoneg = AUTONEG_DISABLE;
3461}
3462
3463static int tun_get_link_ksettings(struct net_device *dev,
3464 struct ethtool_link_ksettings *cmd)
3465{
3466 struct tun_struct *tun = netdev_priv(dev);
3467
3468 memcpy(cmd, &tun->link_ksettings, sizeof(*cmd));
3469 return 0;
3470}
3471
3472static int tun_set_link_ksettings(struct net_device *dev,
3473 const struct ethtool_link_ksettings *cmd)
3474{
3475 struct tun_struct *tun = netdev_priv(dev);
3476
3477 memcpy(&tun->link_ksettings, cmd, sizeof(*cmd));
3478 return 0;
3479}
3480
3481static void tun_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
3482{
3483 struct tun_struct *tun = netdev_priv(dev);
3484
3485 strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
3486 strlcpy(info->version, DRV_VERSION, sizeof(info->version));
3487
3488 switch (tun->flags & TUN_TYPE_MASK) {
3489 case IFF_TUN:
3490 strlcpy(info->bus_info, "tun", sizeof(info->bus_info));
3491 break;
3492 case IFF_TAP:
3493 strlcpy(info->bus_info, "tap", sizeof(info->bus_info));
3494 break;
3495 }
3496}
3497
3498static u32 tun_get_msglevel(struct net_device *dev)
3499{
3500 struct tun_struct *tun = netdev_priv(dev);
3501
3502 return tun->msg_enable;
3503}
3504
3505static void tun_set_msglevel(struct net_device *dev, u32 value)
3506{
3507 struct tun_struct *tun = netdev_priv(dev);
3508
3509 tun->msg_enable = value;
3510}
3511
3512static int tun_get_coalesce(struct net_device *dev,
3513 struct ethtool_coalesce *ec)
3514{
3515 struct tun_struct *tun = netdev_priv(dev);
3516
3517 ec->rx_max_coalesced_frames = tun->rx_batched;
3518
3519 return 0;
3520}
3521
3522static int tun_set_coalesce(struct net_device *dev,
3523 struct ethtool_coalesce *ec)
3524{
3525 struct tun_struct *tun = netdev_priv(dev);
3526
3527 if (ec->rx_max_coalesced_frames > NAPI_POLL_WEIGHT)
3528 tun->rx_batched = NAPI_POLL_WEIGHT;
3529 else
3530 tun->rx_batched = ec->rx_max_coalesced_frames;
3531
3532 return 0;
3533}
3534
3535static const struct ethtool_ops tun_ethtool_ops = {
3536 .supported_coalesce_params = ETHTOOL_COALESCE_RX_MAX_FRAMES,
3537 .get_drvinfo = tun_get_drvinfo,
3538 .get_msglevel = tun_get_msglevel,
3539 .set_msglevel = tun_set_msglevel,
3540 .get_link = ethtool_op_get_link,
3541 .get_ts_info = ethtool_op_get_ts_info,
3542 .get_coalesce = tun_get_coalesce,
3543 .set_coalesce = tun_set_coalesce,
3544 .get_link_ksettings = tun_get_link_ksettings,
3545 .set_link_ksettings = tun_set_link_ksettings,
3546};
3547
3548static int tun_queue_resize(struct tun_struct *tun)
3549{
3550 struct net_device *dev = tun->dev;
3551 struct tun_file *tfile;
3552 struct ptr_ring **rings;
3553 int n = tun->numqueues + tun->numdisabled;
3554 int ret, i;
3555
3556 rings = kmalloc_array(n, sizeof(*rings), GFP_KERNEL);
3557 if (!rings)
3558 return -ENOMEM;
3559
3560 for (i = 0; i < tun->numqueues; i++) {
3561 tfile = rtnl_dereference(tun->tfiles[i]);
3562 rings[i] = &tfile->tx_ring;
3563 }
3564 list_for_each_entry(tfile, &tun->disabled, next)
3565 rings[i++] = &tfile->tx_ring;
3566
3567 ret = ptr_ring_resize_multiple(rings, n,
3568 dev->tx_queue_len, GFP_KERNEL,
3569 tun_ptr_free);
3570
3571 kfree(rings);
3572 return ret;
3573}
3574
3575static int tun_device_event(struct notifier_block *unused,
3576 unsigned long event, void *ptr)
3577{
3578 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
3579 struct tun_struct *tun = netdev_priv(dev);
3580 int i;
3581
3582 if (dev->rtnl_link_ops != &tun_link_ops)
3583 return NOTIFY_DONE;
3584
3585 switch (event) {
3586 case NETDEV_CHANGE_TX_QUEUE_LEN:
3587 if (tun_queue_resize(tun))
3588 return NOTIFY_BAD;
3589 break;
3590 case NETDEV_UP:
3591 for (i = 0; i < tun->numqueues; i++) {
3592 struct tun_file *tfile;
3593
3594 tfile = rtnl_dereference(tun->tfiles[i]);
3595 tfile->socket.sk->sk_write_space(tfile->socket.sk);
3596 }
3597 break;
3598 default:
3599 break;
3600 }
3601
3602 return NOTIFY_DONE;
3603}
3604
3605static struct notifier_block tun_notifier_block __read_mostly = {
3606 .notifier_call = tun_device_event,
3607};
3608
3609static int __init tun_init(void)
3610{
3611 int ret = 0;
3612
3613 pr_info("%s, %s\n", DRV_DESCRIPTION, DRV_VERSION);
3614
3615 ret = rtnl_link_register(&tun_link_ops);
3616 if (ret) {
3617 pr_err("Can't register link_ops\n");
3618 goto err_linkops;
3619 }
3620
3621 ret = misc_register(&tun_miscdev);
3622 if (ret) {
3623 pr_err("Can't register misc device %d\n", TUN_MINOR);
3624 goto err_misc;
3625 }
3626
3627 ret = register_netdevice_notifier(&tun_notifier_block);
3628 if (ret) {
3629 pr_err("Can't register netdevice notifier\n");
3630 goto err_notifier;
3631 }
3632
3633 return 0;
3634
3635err_notifier:
3636 misc_deregister(&tun_miscdev);
3637err_misc:
3638 rtnl_link_unregister(&tun_link_ops);
3639err_linkops:
3640 return ret;
3641}
3642
3643static void tun_cleanup(void)
3644{
3645 misc_deregister(&tun_miscdev);
3646 rtnl_link_unregister(&tun_link_ops);
3647 unregister_netdevice_notifier(&tun_notifier_block);
3648}
3649
3650/* Get an underlying socket object from tun file. Returns error unless file is
3651 * attached to a device. The returned object works like a packet socket, it
3652 * can be used for sock_sendmsg/sock_recvmsg. The caller is responsible for
3653 * holding a reference to the file for as long as the socket is in use. */
3654struct socket *tun_get_socket(struct file *file)
3655{
3656 struct tun_file *tfile;
3657 if (file->f_op != &tun_fops)
3658 return ERR_PTR(-EINVAL);
3659 tfile = file->private_data;
3660 if (!tfile)
3661 return ERR_PTR(-EBADFD);
3662 return &tfile->socket;
3663}
3664EXPORT_SYMBOL_GPL(tun_get_socket);
3665
3666struct ptr_ring *tun_get_tx_ring(struct file *file)
3667{
3668 struct tun_file *tfile;
3669
3670 if (file->f_op != &tun_fops)
3671 return ERR_PTR(-EINVAL);
3672 tfile = file->private_data;
3673 if (!tfile)
3674 return ERR_PTR(-EBADFD);
3675 return &tfile->tx_ring;
3676}
3677EXPORT_SYMBOL_GPL(tun_get_tx_ring);
3678
3679module_init(tun_init);
3680module_exit(tun_cleanup);
3681MODULE_DESCRIPTION(DRV_DESCRIPTION);
3682MODULE_AUTHOR(DRV_COPYRIGHT);
3683MODULE_LICENSE("GPL");
3684MODULE_ALIAS_MISCDEV(TUN_MINOR);
3685MODULE_ALIAS("devname:net/tun");