Loading...
1/*
2 * TUN - Universal TUN/TAP device driver.
3 * Copyright (C) 1999-2002 Maxim Krasnyansky <maxk@qualcomm.com>
4 *
5 * This program is free software; you can redistribute it and/or modify
6 * it under the terms of the GNU General Public License as published by
7 * the Free Software Foundation; either version 2 of the License, or
8 * (at your option) any later version.
9 *
10 * This program is distributed in the hope that it will be useful,
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
13 * GNU General Public License for more details.
14 *
15 * $Id: tun.c,v 1.15 2002/03/01 02:44:24 maxk Exp $
16 */
17
18/*
19 * Changes:
20 *
21 * Mike Kershaw <dragorn@kismetwireless.net> 2005/08/14
22 * Add TUNSETLINK ioctl to set the link encapsulation
23 *
24 * Mark Smith <markzzzsmith@yahoo.com.au>
25 * Use eth_random_addr() for tap MAC address.
26 *
27 * Harald Roelle <harald.roelle@ifi.lmu.de> 2004/04/20
28 * Fixes in packet dropping, queue length setting and queue wakeup.
29 * Increased default tx queue length.
30 * Added ethtool API.
31 * Minor cleanups
32 *
33 * Daniel Podlejski <underley@underley.eu.org>
34 * Modifications for 2.3.99-pre5 kernel.
35 */
36
37#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
38
39#define DRV_NAME "tun"
40#define DRV_VERSION "1.6"
41#define DRV_DESCRIPTION "Universal TUN/TAP device driver"
42#define DRV_COPYRIGHT "(C) 1999-2004 Max Krasnyansky <maxk@qualcomm.com>"
43
44#include <linux/module.h>
45#include <linux/errno.h>
46#include <linux/kernel.h>
47#include <linux/sched/signal.h>
48#include <linux/major.h>
49#include <linux/slab.h>
50#include <linux/poll.h>
51#include <linux/fcntl.h>
52#include <linux/init.h>
53#include <linux/skbuff.h>
54#include <linux/netdevice.h>
55#include <linux/etherdevice.h>
56#include <linux/miscdevice.h>
57#include <linux/ethtool.h>
58#include <linux/rtnetlink.h>
59#include <linux/compat.h>
60#include <linux/if.h>
61#include <linux/if_arp.h>
62#include <linux/if_ether.h>
63#include <linux/if_tun.h>
64#include <linux/if_vlan.h>
65#include <linux/crc32.h>
66#include <linux/nsproxy.h>
67#include <linux/virtio_net.h>
68#include <linux/rcupdate.h>
69#include <net/net_namespace.h>
70#include <net/netns/generic.h>
71#include <net/rtnetlink.h>
72#include <net/sock.h>
73#include <linux/seq_file.h>
74#include <linux/uio.h>
75#include <linux/skb_array.h>
76#include <linux/bpf.h>
77#include <linux/bpf_trace.h>
78#include <linux/mutex.h>
79
80#include <linux/uaccess.h>
81#include <linux/proc_fs.h>
82
83/* Uncomment to enable debugging */
84/* #define TUN_DEBUG 1 */
85
86#ifdef TUN_DEBUG
87static int debug;
88
89#define tun_debug(level, tun, fmt, args...) \
90do { \
91 if (tun->debug) \
92 netdev_printk(level, tun->dev, fmt, ##args); \
93} while (0)
94#define DBG1(level, fmt, args...) \
95do { \
96 if (debug == 2) \
97 printk(level fmt, ##args); \
98} while (0)
99#else
100#define tun_debug(level, tun, fmt, args...) \
101do { \
102 if (0) \
103 netdev_printk(level, tun->dev, fmt, ##args); \
104} while (0)
105#define DBG1(level, fmt, args...) \
106do { \
107 if (0) \
108 printk(level fmt, ##args); \
109} while (0)
110#endif
111
112#define TUN_HEADROOM 256
113#define TUN_RX_PAD (NET_IP_ALIGN + NET_SKB_PAD)
114
115/* TUN device flags */
116
117/* IFF_ATTACH_QUEUE is never stored in device flags,
118 * overload it to mean fasync when stored there.
119 */
120#define TUN_FASYNC IFF_ATTACH_QUEUE
121/* High bits in flags field are unused. */
122#define TUN_VNET_LE 0x80000000
123#define TUN_VNET_BE 0x40000000
124
125#define TUN_FEATURES (IFF_NO_PI | IFF_ONE_QUEUE | IFF_VNET_HDR | \
126 IFF_MULTI_QUEUE | IFF_NAPI | IFF_NAPI_FRAGS)
127
128#define GOODCOPY_LEN 128
129
130#define FLT_EXACT_COUNT 8
131struct tap_filter {
132 unsigned int count; /* Number of addrs. Zero means disabled */
133 u32 mask[2]; /* Mask of the hashed addrs */
134 unsigned char addr[FLT_EXACT_COUNT][ETH_ALEN];
135};
136
137/* MAX_TAP_QUEUES 256 is chosen to allow rx/tx queues to be equal
138 * to max number of VCPUs in guest. */
139#define MAX_TAP_QUEUES 256
140#define MAX_TAP_FLOWS 4096
141
142#define TUN_FLOW_EXPIRE (3 * HZ)
143
144struct tun_pcpu_stats {
145 u64 rx_packets;
146 u64 rx_bytes;
147 u64 tx_packets;
148 u64 tx_bytes;
149 struct u64_stats_sync syncp;
150 u32 rx_dropped;
151 u32 tx_dropped;
152 u32 rx_frame_errors;
153};
154
155/* A tun_file connects an open character device to a tuntap netdevice. It
156 * also contains all socket related structures (except sock_fprog and tap_filter)
157 * to serve as one transmit queue for tuntap device. The sock_fprog and
158 * tap_filter were kept in tun_struct since they were used for filtering for the
159 * netdevice not for a specific queue (at least I didn't see the requirement for
160 * this).
161 *
162 * RCU usage:
163 * The tun_file and tun_struct are loosely coupled, the pointer from one to the
164 * other can only be read while rcu_read_lock or rtnl_lock is held.
165 */
166struct tun_file {
167 struct sock sk;
168 struct socket socket;
169 struct socket_wq wq;
170 struct tun_struct __rcu *tun;
171 struct fasync_struct *fasync;
172 /* only used for fasnyc */
173 unsigned int flags;
174 union {
175 u16 queue_index;
176 unsigned int ifindex;
177 };
178 struct napi_struct napi;
179 bool napi_enabled;
180 struct mutex napi_mutex; /* Protects access to the above napi */
181 struct list_head next;
182 struct tun_struct *detached;
183 struct ptr_ring tx_ring;
184 struct xdp_rxq_info xdp_rxq;
185};
186
187struct tun_flow_entry {
188 struct hlist_node hash_link;
189 struct rcu_head rcu;
190 struct tun_struct *tun;
191
192 u32 rxhash;
193 u32 rps_rxhash;
194 int queue_index;
195 unsigned long updated;
196};
197
198#define TUN_NUM_FLOW_ENTRIES 1024
199
200struct tun_prog {
201 struct rcu_head rcu;
202 struct bpf_prog *prog;
203};
204
205/* Since the socket were moved to tun_file, to preserve the behavior of persist
206 * device, socket filter, sndbuf and vnet header size were restore when the
207 * file were attached to a persist device.
208 */
209struct tun_struct {
210 struct tun_file __rcu *tfiles[MAX_TAP_QUEUES];
211 unsigned int numqueues;
212 unsigned int flags;
213 kuid_t owner;
214 kgid_t group;
215
216 struct net_device *dev;
217 netdev_features_t set_features;
218#define TUN_USER_FEATURES (NETIF_F_HW_CSUM|NETIF_F_TSO_ECN|NETIF_F_TSO| \
219 NETIF_F_TSO6)
220
221 int align;
222 int vnet_hdr_sz;
223 int sndbuf;
224 struct tap_filter txflt;
225 struct sock_fprog fprog;
226 /* protected by rtnl lock */
227 bool filter_attached;
228#ifdef TUN_DEBUG
229 int debug;
230#endif
231 spinlock_t lock;
232 struct hlist_head flows[TUN_NUM_FLOW_ENTRIES];
233 struct timer_list flow_gc_timer;
234 unsigned long ageing_time;
235 unsigned int numdisabled;
236 struct list_head disabled;
237 void *security;
238 u32 flow_count;
239 u32 rx_batched;
240 struct tun_pcpu_stats __percpu *pcpu_stats;
241 struct bpf_prog __rcu *xdp_prog;
242 struct tun_prog __rcu *steering_prog;
243 struct tun_prog __rcu *filter_prog;
244};
245
246struct veth {
247 __be16 h_vlan_proto;
248 __be16 h_vlan_TCI;
249};
250
251bool tun_is_xdp_buff(void *ptr)
252{
253 return (unsigned long)ptr & TUN_XDP_FLAG;
254}
255EXPORT_SYMBOL(tun_is_xdp_buff);
256
257void *tun_xdp_to_ptr(void *ptr)
258{
259 return (void *)((unsigned long)ptr | TUN_XDP_FLAG);
260}
261EXPORT_SYMBOL(tun_xdp_to_ptr);
262
263void *tun_ptr_to_xdp(void *ptr)
264{
265 return (void *)((unsigned long)ptr & ~TUN_XDP_FLAG);
266}
267EXPORT_SYMBOL(tun_ptr_to_xdp);
268
269static int tun_napi_receive(struct napi_struct *napi, int budget)
270{
271 struct tun_file *tfile = container_of(napi, struct tun_file, napi);
272 struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
273 struct sk_buff_head process_queue;
274 struct sk_buff *skb;
275 int received = 0;
276
277 __skb_queue_head_init(&process_queue);
278
279 spin_lock(&queue->lock);
280 skb_queue_splice_tail_init(queue, &process_queue);
281 spin_unlock(&queue->lock);
282
283 while (received < budget && (skb = __skb_dequeue(&process_queue))) {
284 napi_gro_receive(napi, skb);
285 ++received;
286 }
287
288 if (!skb_queue_empty(&process_queue)) {
289 spin_lock(&queue->lock);
290 skb_queue_splice(&process_queue, queue);
291 spin_unlock(&queue->lock);
292 }
293
294 return received;
295}
296
297static int tun_napi_poll(struct napi_struct *napi, int budget)
298{
299 unsigned int received;
300
301 received = tun_napi_receive(napi, budget);
302
303 if (received < budget)
304 napi_complete_done(napi, received);
305
306 return received;
307}
308
309static void tun_napi_init(struct tun_struct *tun, struct tun_file *tfile,
310 bool napi_en)
311{
312 tfile->napi_enabled = napi_en;
313 if (napi_en) {
314 netif_napi_add(tun->dev, &tfile->napi, tun_napi_poll,
315 NAPI_POLL_WEIGHT);
316 napi_enable(&tfile->napi);
317 mutex_init(&tfile->napi_mutex);
318 }
319}
320
321static void tun_napi_disable(struct tun_struct *tun, struct tun_file *tfile)
322{
323 if (tfile->napi_enabled)
324 napi_disable(&tfile->napi);
325}
326
327static void tun_napi_del(struct tun_struct *tun, struct tun_file *tfile)
328{
329 if (tfile->napi_enabled)
330 netif_napi_del(&tfile->napi);
331}
332
333static bool tun_napi_frags_enabled(const struct tun_struct *tun)
334{
335 return READ_ONCE(tun->flags) & IFF_NAPI_FRAGS;
336}
337
338#ifdef CONFIG_TUN_VNET_CROSS_LE
339static inline bool tun_legacy_is_little_endian(struct tun_struct *tun)
340{
341 return tun->flags & TUN_VNET_BE ? false :
342 virtio_legacy_is_little_endian();
343}
344
345static long tun_get_vnet_be(struct tun_struct *tun, int __user *argp)
346{
347 int be = !!(tun->flags & TUN_VNET_BE);
348
349 if (put_user(be, argp))
350 return -EFAULT;
351
352 return 0;
353}
354
355static long tun_set_vnet_be(struct tun_struct *tun, int __user *argp)
356{
357 int be;
358
359 if (get_user(be, argp))
360 return -EFAULT;
361
362 if (be)
363 tun->flags |= TUN_VNET_BE;
364 else
365 tun->flags &= ~TUN_VNET_BE;
366
367 return 0;
368}
369#else
370static inline bool tun_legacy_is_little_endian(struct tun_struct *tun)
371{
372 return virtio_legacy_is_little_endian();
373}
374
375static long tun_get_vnet_be(struct tun_struct *tun, int __user *argp)
376{
377 return -EINVAL;
378}
379
380static long tun_set_vnet_be(struct tun_struct *tun, int __user *argp)
381{
382 return -EINVAL;
383}
384#endif /* CONFIG_TUN_VNET_CROSS_LE */
385
386static inline bool tun_is_little_endian(struct tun_struct *tun)
387{
388 return tun->flags & TUN_VNET_LE ||
389 tun_legacy_is_little_endian(tun);
390}
391
392static inline u16 tun16_to_cpu(struct tun_struct *tun, __virtio16 val)
393{
394 return __virtio16_to_cpu(tun_is_little_endian(tun), val);
395}
396
397static inline __virtio16 cpu_to_tun16(struct tun_struct *tun, u16 val)
398{
399 return __cpu_to_virtio16(tun_is_little_endian(tun), val);
400}
401
402static inline u32 tun_hashfn(u32 rxhash)
403{
404 return rxhash & 0x3ff;
405}
406
407static struct tun_flow_entry *tun_flow_find(struct hlist_head *head, u32 rxhash)
408{
409 struct tun_flow_entry *e;
410
411 hlist_for_each_entry_rcu(e, head, hash_link) {
412 if (e->rxhash == rxhash)
413 return e;
414 }
415 return NULL;
416}
417
418static struct tun_flow_entry *tun_flow_create(struct tun_struct *tun,
419 struct hlist_head *head,
420 u32 rxhash, u16 queue_index)
421{
422 struct tun_flow_entry *e = kmalloc(sizeof(*e), GFP_ATOMIC);
423
424 if (e) {
425 tun_debug(KERN_INFO, tun, "create flow: hash %u index %u\n",
426 rxhash, queue_index);
427 e->updated = jiffies;
428 e->rxhash = rxhash;
429 e->rps_rxhash = 0;
430 e->queue_index = queue_index;
431 e->tun = tun;
432 hlist_add_head_rcu(&e->hash_link, head);
433 ++tun->flow_count;
434 }
435 return e;
436}
437
438static void tun_flow_delete(struct tun_struct *tun, struct tun_flow_entry *e)
439{
440 tun_debug(KERN_INFO, tun, "delete flow: hash %u index %u\n",
441 e->rxhash, e->queue_index);
442 hlist_del_rcu(&e->hash_link);
443 kfree_rcu(e, rcu);
444 --tun->flow_count;
445}
446
447static void tun_flow_flush(struct tun_struct *tun)
448{
449 int i;
450
451 spin_lock_bh(&tun->lock);
452 for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
453 struct tun_flow_entry *e;
454 struct hlist_node *n;
455
456 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link)
457 tun_flow_delete(tun, e);
458 }
459 spin_unlock_bh(&tun->lock);
460}
461
462static void tun_flow_delete_by_queue(struct tun_struct *tun, u16 queue_index)
463{
464 int i;
465
466 spin_lock_bh(&tun->lock);
467 for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
468 struct tun_flow_entry *e;
469 struct hlist_node *n;
470
471 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
472 if (e->queue_index == queue_index)
473 tun_flow_delete(tun, e);
474 }
475 }
476 spin_unlock_bh(&tun->lock);
477}
478
479static void tun_flow_cleanup(struct timer_list *t)
480{
481 struct tun_struct *tun = from_timer(tun, t, flow_gc_timer);
482 unsigned long delay = tun->ageing_time;
483 unsigned long next_timer = jiffies + delay;
484 unsigned long count = 0;
485 int i;
486
487 tun_debug(KERN_INFO, tun, "tun_flow_cleanup\n");
488
489 spin_lock(&tun->lock);
490 for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
491 struct tun_flow_entry *e;
492 struct hlist_node *n;
493
494 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
495 unsigned long this_timer;
496
497 this_timer = e->updated + delay;
498 if (time_before_eq(this_timer, jiffies)) {
499 tun_flow_delete(tun, e);
500 continue;
501 }
502 count++;
503 if (time_before(this_timer, next_timer))
504 next_timer = this_timer;
505 }
506 }
507
508 if (count)
509 mod_timer(&tun->flow_gc_timer, round_jiffies_up(next_timer));
510 spin_unlock(&tun->lock);
511}
512
513static void tun_flow_update(struct tun_struct *tun, u32 rxhash,
514 struct tun_file *tfile)
515{
516 struct hlist_head *head;
517 struct tun_flow_entry *e;
518 unsigned long delay = tun->ageing_time;
519 u16 queue_index = tfile->queue_index;
520
521 if (!rxhash)
522 return;
523 else
524 head = &tun->flows[tun_hashfn(rxhash)];
525
526 rcu_read_lock();
527
528 /* We may get a very small possibility of OOO during switching, not
529 * worth to optimize.*/
530 if (tun->numqueues == 1 || tfile->detached)
531 goto unlock;
532
533 e = tun_flow_find(head, rxhash);
534 if (likely(e)) {
535 /* TODO: keep queueing to old queue until it's empty? */
536 e->queue_index = queue_index;
537 e->updated = jiffies;
538 sock_rps_record_flow_hash(e->rps_rxhash);
539 } else {
540 spin_lock_bh(&tun->lock);
541 if (!tun_flow_find(head, rxhash) &&
542 tun->flow_count < MAX_TAP_FLOWS)
543 tun_flow_create(tun, head, rxhash, queue_index);
544
545 if (!timer_pending(&tun->flow_gc_timer))
546 mod_timer(&tun->flow_gc_timer,
547 round_jiffies_up(jiffies + delay));
548 spin_unlock_bh(&tun->lock);
549 }
550
551unlock:
552 rcu_read_unlock();
553}
554
555/**
556 * Save the hash received in the stack receive path and update the
557 * flow_hash table accordingly.
558 */
559static inline void tun_flow_save_rps_rxhash(struct tun_flow_entry *e, u32 hash)
560{
561 if (unlikely(e->rps_rxhash != hash))
562 e->rps_rxhash = hash;
563}
564
565/* We try to identify a flow through its rxhash first. The reason that
566 * we do not check rxq no. is because some cards(e.g 82599), chooses
567 * the rxq based on the txq where the last packet of the flow comes. As
568 * the userspace application move between processors, we may get a
569 * different rxq no. here. If we could not get rxhash, then we would
570 * hope the rxq no. may help here.
571 */
572static u16 tun_automq_select_queue(struct tun_struct *tun, struct sk_buff *skb)
573{
574 struct tun_flow_entry *e;
575 u32 txq = 0;
576 u32 numqueues = 0;
577
578 numqueues = READ_ONCE(tun->numqueues);
579
580 txq = __skb_get_hash_symmetric(skb);
581 if (txq) {
582 e = tun_flow_find(&tun->flows[tun_hashfn(txq)], txq);
583 if (e) {
584 tun_flow_save_rps_rxhash(e, txq);
585 txq = e->queue_index;
586 } else
587 /* use multiply and shift instead of expensive divide */
588 txq = ((u64)txq * numqueues) >> 32;
589 } else if (likely(skb_rx_queue_recorded(skb))) {
590 txq = skb_get_rx_queue(skb);
591 while (unlikely(txq >= numqueues))
592 txq -= numqueues;
593 }
594
595 return txq;
596}
597
598static u16 tun_ebpf_select_queue(struct tun_struct *tun, struct sk_buff *skb)
599{
600 struct tun_prog *prog;
601 u16 ret = 0;
602
603 prog = rcu_dereference(tun->steering_prog);
604 if (prog)
605 ret = bpf_prog_run_clear_cb(prog->prog, skb);
606
607 return ret % tun->numqueues;
608}
609
610static u16 tun_select_queue(struct net_device *dev, struct sk_buff *skb,
611 void *accel_priv, select_queue_fallback_t fallback)
612{
613 struct tun_struct *tun = netdev_priv(dev);
614 u16 ret;
615
616 rcu_read_lock();
617 if (rcu_dereference(tun->steering_prog))
618 ret = tun_ebpf_select_queue(tun, skb);
619 else
620 ret = tun_automq_select_queue(tun, skb);
621 rcu_read_unlock();
622
623 return ret;
624}
625
626static inline bool tun_not_capable(struct tun_struct *tun)
627{
628 const struct cred *cred = current_cred();
629 struct net *net = dev_net(tun->dev);
630
631 return ((uid_valid(tun->owner) && !uid_eq(cred->euid, tun->owner)) ||
632 (gid_valid(tun->group) && !in_egroup_p(tun->group))) &&
633 !ns_capable(net->user_ns, CAP_NET_ADMIN);
634}
635
636static void tun_set_real_num_queues(struct tun_struct *tun)
637{
638 netif_set_real_num_tx_queues(tun->dev, tun->numqueues);
639 netif_set_real_num_rx_queues(tun->dev, tun->numqueues);
640}
641
642static void tun_disable_queue(struct tun_struct *tun, struct tun_file *tfile)
643{
644 tfile->detached = tun;
645 list_add_tail(&tfile->next, &tun->disabled);
646 ++tun->numdisabled;
647}
648
649static struct tun_struct *tun_enable_queue(struct tun_file *tfile)
650{
651 struct tun_struct *tun = tfile->detached;
652
653 tfile->detached = NULL;
654 list_del_init(&tfile->next);
655 --tun->numdisabled;
656 return tun;
657}
658
659void tun_ptr_free(void *ptr)
660{
661 if (!ptr)
662 return;
663 if (tun_is_xdp_buff(ptr)) {
664 struct xdp_buff *xdp = tun_ptr_to_xdp(ptr);
665
666 put_page(virt_to_head_page(xdp->data));
667 } else {
668 __skb_array_destroy_skb(ptr);
669 }
670}
671EXPORT_SYMBOL_GPL(tun_ptr_free);
672
673static void tun_queue_purge(struct tun_file *tfile)
674{
675 void *ptr;
676
677 while ((ptr = ptr_ring_consume(&tfile->tx_ring)) != NULL)
678 tun_ptr_free(ptr);
679
680 skb_queue_purge(&tfile->sk.sk_write_queue);
681 skb_queue_purge(&tfile->sk.sk_error_queue);
682}
683
684static void __tun_detach(struct tun_file *tfile, bool clean)
685{
686 struct tun_file *ntfile;
687 struct tun_struct *tun;
688
689 tun = rtnl_dereference(tfile->tun);
690
691 if (tun && clean) {
692 tun_napi_disable(tun, tfile);
693 tun_napi_del(tun, tfile);
694 }
695
696 if (tun && !tfile->detached) {
697 u16 index = tfile->queue_index;
698 BUG_ON(index >= tun->numqueues);
699
700 rcu_assign_pointer(tun->tfiles[index],
701 tun->tfiles[tun->numqueues - 1]);
702 ntfile = rtnl_dereference(tun->tfiles[index]);
703 ntfile->queue_index = index;
704
705 --tun->numqueues;
706 if (clean) {
707 RCU_INIT_POINTER(tfile->tun, NULL);
708 sock_put(&tfile->sk);
709 } else
710 tun_disable_queue(tun, tfile);
711
712 synchronize_net();
713 tun_flow_delete_by_queue(tun, tun->numqueues + 1);
714 /* Drop read queue */
715 tun_queue_purge(tfile);
716 tun_set_real_num_queues(tun);
717 } else if (tfile->detached && clean) {
718 tun = tun_enable_queue(tfile);
719 sock_put(&tfile->sk);
720 }
721
722 if (clean) {
723 if (tun && tun->numqueues == 0 && tun->numdisabled == 0) {
724 netif_carrier_off(tun->dev);
725
726 if (!(tun->flags & IFF_PERSIST) &&
727 tun->dev->reg_state == NETREG_REGISTERED)
728 unregister_netdevice(tun->dev);
729 }
730 if (tun)
731 xdp_rxq_info_unreg(&tfile->xdp_rxq);
732 ptr_ring_cleanup(&tfile->tx_ring, tun_ptr_free);
733 sock_put(&tfile->sk);
734 }
735}
736
737static void tun_detach(struct tun_file *tfile, bool clean)
738{
739 struct tun_struct *tun;
740 struct net_device *dev;
741
742 rtnl_lock();
743 tun = rtnl_dereference(tfile->tun);
744 dev = tun ? tun->dev : NULL;
745 __tun_detach(tfile, clean);
746 if (dev)
747 netdev_state_change(dev);
748 rtnl_unlock();
749}
750
751static void tun_detach_all(struct net_device *dev)
752{
753 struct tun_struct *tun = netdev_priv(dev);
754 struct tun_file *tfile, *tmp;
755 int i, n = tun->numqueues;
756
757 for (i = 0; i < n; i++) {
758 tfile = rtnl_dereference(tun->tfiles[i]);
759 BUG_ON(!tfile);
760 tun_napi_disable(tun, tfile);
761 tfile->socket.sk->sk_shutdown = RCV_SHUTDOWN;
762 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
763 RCU_INIT_POINTER(tfile->tun, NULL);
764 --tun->numqueues;
765 }
766 list_for_each_entry(tfile, &tun->disabled, next) {
767 tfile->socket.sk->sk_shutdown = RCV_SHUTDOWN;
768 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
769 RCU_INIT_POINTER(tfile->tun, NULL);
770 }
771 BUG_ON(tun->numqueues != 0);
772
773 synchronize_net();
774 for (i = 0; i < n; i++) {
775 tfile = rtnl_dereference(tun->tfiles[i]);
776 tun_napi_del(tun, tfile);
777 /* Drop read queue */
778 tun_queue_purge(tfile);
779 xdp_rxq_info_unreg(&tfile->xdp_rxq);
780 sock_put(&tfile->sk);
781 }
782 list_for_each_entry_safe(tfile, tmp, &tun->disabled, next) {
783 tun_enable_queue(tfile);
784 tun_queue_purge(tfile);
785 xdp_rxq_info_unreg(&tfile->xdp_rxq);
786 sock_put(&tfile->sk);
787 }
788 BUG_ON(tun->numdisabled != 0);
789
790 if (tun->flags & IFF_PERSIST)
791 module_put(THIS_MODULE);
792}
793
794static int tun_attach(struct tun_struct *tun, struct file *file,
795 bool skip_filter, bool napi)
796{
797 struct tun_file *tfile = file->private_data;
798 struct net_device *dev = tun->dev;
799 int err;
800
801 err = security_tun_dev_attach(tfile->socket.sk, tun->security);
802 if (err < 0)
803 goto out;
804
805 err = -EINVAL;
806 if (rtnl_dereference(tfile->tun) && !tfile->detached)
807 goto out;
808
809 err = -EBUSY;
810 if (!(tun->flags & IFF_MULTI_QUEUE) && tun->numqueues == 1)
811 goto out;
812
813 err = -E2BIG;
814 if (!tfile->detached &&
815 tun->numqueues + tun->numdisabled == MAX_TAP_QUEUES)
816 goto out;
817
818 err = 0;
819
820 /* Re-attach the filter to persist device */
821 if (!skip_filter && (tun->filter_attached == true)) {
822 lock_sock(tfile->socket.sk);
823 err = sk_attach_filter(&tun->fprog, tfile->socket.sk);
824 release_sock(tfile->socket.sk);
825 if (!err)
826 goto out;
827 }
828
829 if (!tfile->detached &&
830 ptr_ring_resize(&tfile->tx_ring, dev->tx_queue_len,
831 GFP_KERNEL, tun_ptr_free)) {
832 err = -ENOMEM;
833 goto out;
834 }
835
836 tfile->queue_index = tun->numqueues;
837 tfile->socket.sk->sk_shutdown &= ~RCV_SHUTDOWN;
838
839 if (tfile->detached) {
840 /* Re-attach detached tfile, updating XDP queue_index */
841 WARN_ON(!xdp_rxq_info_is_reg(&tfile->xdp_rxq));
842
843 if (tfile->xdp_rxq.queue_index != tfile->queue_index)
844 tfile->xdp_rxq.queue_index = tfile->queue_index;
845 } else {
846 /* Setup XDP RX-queue info, for new tfile getting attached */
847 err = xdp_rxq_info_reg(&tfile->xdp_rxq,
848 tun->dev, tfile->queue_index);
849 if (err < 0)
850 goto out;
851 err = 0;
852 }
853
854 rcu_assign_pointer(tfile->tun, tun);
855 rcu_assign_pointer(tun->tfiles[tun->numqueues], tfile);
856 tun->numqueues++;
857
858 if (tfile->detached) {
859 tun_enable_queue(tfile);
860 } else {
861 sock_hold(&tfile->sk);
862 tun_napi_init(tun, tfile, napi);
863 }
864
865 tun_set_real_num_queues(tun);
866
867 /* device is allowed to go away first, so no need to hold extra
868 * refcnt.
869 */
870
871out:
872 return err;
873}
874
875static struct tun_struct *tun_get(struct tun_file *tfile)
876{
877 struct tun_struct *tun;
878
879 rcu_read_lock();
880 tun = rcu_dereference(tfile->tun);
881 if (tun)
882 dev_hold(tun->dev);
883 rcu_read_unlock();
884
885 return tun;
886}
887
888static void tun_put(struct tun_struct *tun)
889{
890 dev_put(tun->dev);
891}
892
893/* TAP filtering */
894static void addr_hash_set(u32 *mask, const u8 *addr)
895{
896 int n = ether_crc(ETH_ALEN, addr) >> 26;
897 mask[n >> 5] |= (1 << (n & 31));
898}
899
900static unsigned int addr_hash_test(const u32 *mask, const u8 *addr)
901{
902 int n = ether_crc(ETH_ALEN, addr) >> 26;
903 return mask[n >> 5] & (1 << (n & 31));
904}
905
906static int update_filter(struct tap_filter *filter, void __user *arg)
907{
908 struct { u8 u[ETH_ALEN]; } *addr;
909 struct tun_filter uf;
910 int err, alen, n, nexact;
911
912 if (copy_from_user(&uf, arg, sizeof(uf)))
913 return -EFAULT;
914
915 if (!uf.count) {
916 /* Disabled */
917 filter->count = 0;
918 return 0;
919 }
920
921 alen = ETH_ALEN * uf.count;
922 addr = memdup_user(arg + sizeof(uf), alen);
923 if (IS_ERR(addr))
924 return PTR_ERR(addr);
925
926 /* The filter is updated without holding any locks. Which is
927 * perfectly safe. We disable it first and in the worst
928 * case we'll accept a few undesired packets. */
929 filter->count = 0;
930 wmb();
931
932 /* Use first set of addresses as an exact filter */
933 for (n = 0; n < uf.count && n < FLT_EXACT_COUNT; n++)
934 memcpy(filter->addr[n], addr[n].u, ETH_ALEN);
935
936 nexact = n;
937
938 /* Remaining multicast addresses are hashed,
939 * unicast will leave the filter disabled. */
940 memset(filter->mask, 0, sizeof(filter->mask));
941 for (; n < uf.count; n++) {
942 if (!is_multicast_ether_addr(addr[n].u)) {
943 err = 0; /* no filter */
944 goto free_addr;
945 }
946 addr_hash_set(filter->mask, addr[n].u);
947 }
948
949 /* For ALLMULTI just set the mask to all ones.
950 * This overrides the mask populated above. */
951 if ((uf.flags & TUN_FLT_ALLMULTI))
952 memset(filter->mask, ~0, sizeof(filter->mask));
953
954 /* Now enable the filter */
955 wmb();
956 filter->count = nexact;
957
958 /* Return the number of exact filters */
959 err = nexact;
960free_addr:
961 kfree(addr);
962 return err;
963}
964
965/* Returns: 0 - drop, !=0 - accept */
966static int run_filter(struct tap_filter *filter, const struct sk_buff *skb)
967{
968 /* Cannot use eth_hdr(skb) here because skb_mac_hdr() is incorrect
969 * at this point. */
970 struct ethhdr *eh = (struct ethhdr *) skb->data;
971 int i;
972
973 /* Exact match */
974 for (i = 0; i < filter->count; i++)
975 if (ether_addr_equal(eh->h_dest, filter->addr[i]))
976 return 1;
977
978 /* Inexact match (multicast only) */
979 if (is_multicast_ether_addr(eh->h_dest))
980 return addr_hash_test(filter->mask, eh->h_dest);
981
982 return 0;
983}
984
985/*
986 * Checks whether the packet is accepted or not.
987 * Returns: 0 - drop, !=0 - accept
988 */
989static int check_filter(struct tap_filter *filter, const struct sk_buff *skb)
990{
991 if (!filter->count)
992 return 1;
993
994 return run_filter(filter, skb);
995}
996
997/* Network device part of the driver */
998
999static const struct ethtool_ops tun_ethtool_ops;
1000
1001/* Net device detach from fd. */
1002static void tun_net_uninit(struct net_device *dev)
1003{
1004 tun_detach_all(dev);
1005}
1006
1007/* Net device open. */
1008static int tun_net_open(struct net_device *dev)
1009{
1010 struct tun_struct *tun = netdev_priv(dev);
1011 int i;
1012
1013 netif_tx_start_all_queues(dev);
1014
1015 for (i = 0; i < tun->numqueues; i++) {
1016 struct tun_file *tfile;
1017
1018 tfile = rtnl_dereference(tun->tfiles[i]);
1019 tfile->socket.sk->sk_write_space(tfile->socket.sk);
1020 }
1021
1022 return 0;
1023}
1024
1025/* Net device close. */
1026static int tun_net_close(struct net_device *dev)
1027{
1028 netif_tx_stop_all_queues(dev);
1029 return 0;
1030}
1031
1032/* Net device start xmit */
1033static void tun_automq_xmit(struct tun_struct *tun, struct sk_buff *skb)
1034{
1035#ifdef CONFIG_RPS
1036 if (tun->numqueues == 1 && static_key_false(&rps_needed)) {
1037 /* Select queue was not called for the skbuff, so we extract the
1038 * RPS hash and save it into the flow_table here.
1039 */
1040 __u32 rxhash;
1041
1042 rxhash = __skb_get_hash_symmetric(skb);
1043 if (rxhash) {
1044 struct tun_flow_entry *e;
1045 e = tun_flow_find(&tun->flows[tun_hashfn(rxhash)],
1046 rxhash);
1047 if (e)
1048 tun_flow_save_rps_rxhash(e, rxhash);
1049 }
1050 }
1051#endif
1052}
1053
1054static unsigned int run_ebpf_filter(struct tun_struct *tun,
1055 struct sk_buff *skb,
1056 int len)
1057{
1058 struct tun_prog *prog = rcu_dereference(tun->filter_prog);
1059
1060 if (prog)
1061 len = bpf_prog_run_clear_cb(prog->prog, skb);
1062
1063 return len;
1064}
1065
1066/* Net device start xmit */
1067static netdev_tx_t tun_net_xmit(struct sk_buff *skb, struct net_device *dev)
1068{
1069 struct tun_struct *tun = netdev_priv(dev);
1070 int txq = skb->queue_mapping;
1071 struct tun_file *tfile;
1072 int len = skb->len;
1073
1074 rcu_read_lock();
1075 tfile = rcu_dereference(tun->tfiles[txq]);
1076
1077 /* Drop packet if interface is not attached */
1078 if (txq >= tun->numqueues)
1079 goto drop;
1080
1081 if (!rcu_dereference(tun->steering_prog))
1082 tun_automq_xmit(tun, skb);
1083
1084 tun_debug(KERN_INFO, tun, "tun_net_xmit %d\n", skb->len);
1085
1086 BUG_ON(!tfile);
1087
1088 /* Drop if the filter does not like it.
1089 * This is a noop if the filter is disabled.
1090 * Filter can be enabled only for the TAP devices. */
1091 if (!check_filter(&tun->txflt, skb))
1092 goto drop;
1093
1094 if (tfile->socket.sk->sk_filter &&
1095 sk_filter(tfile->socket.sk, skb))
1096 goto drop;
1097
1098 len = run_ebpf_filter(tun, skb, len);
1099 if (len == 0 || pskb_trim(skb, len))
1100 goto drop;
1101
1102 if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC)))
1103 goto drop;
1104
1105 skb_tx_timestamp(skb);
1106
1107 /* Orphan the skb - required as we might hang on to it
1108 * for indefinite time.
1109 */
1110 skb_orphan(skb);
1111
1112 nf_reset(skb);
1113
1114 if (ptr_ring_produce(&tfile->tx_ring, skb))
1115 goto drop;
1116
1117 /* Notify and wake up reader process */
1118 if (tfile->flags & TUN_FASYNC)
1119 kill_fasync(&tfile->fasync, SIGIO, POLL_IN);
1120 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
1121
1122 rcu_read_unlock();
1123 return NETDEV_TX_OK;
1124
1125drop:
1126 this_cpu_inc(tun->pcpu_stats->tx_dropped);
1127 skb_tx_error(skb);
1128 kfree_skb(skb);
1129 rcu_read_unlock();
1130 return NET_XMIT_DROP;
1131}
1132
1133static void tun_net_mclist(struct net_device *dev)
1134{
1135 /*
1136 * This callback is supposed to deal with mc filter in
1137 * _rx_ path and has nothing to do with the _tx_ path.
1138 * In rx path we always accept everything userspace gives us.
1139 */
1140}
1141
1142static netdev_features_t tun_net_fix_features(struct net_device *dev,
1143 netdev_features_t features)
1144{
1145 struct tun_struct *tun = netdev_priv(dev);
1146
1147 return (features & tun->set_features) | (features & ~TUN_USER_FEATURES);
1148}
1149#ifdef CONFIG_NET_POLL_CONTROLLER
1150static void tun_poll_controller(struct net_device *dev)
1151{
1152 /*
1153 * Tun only receives frames when:
1154 * 1) the char device endpoint gets data from user space
1155 * 2) the tun socket gets a sendmsg call from user space
1156 * If NAPI is not enabled, since both of those are synchronous
1157 * operations, we are guaranteed never to have pending data when we poll
1158 * for it so there is nothing to do here but return.
1159 * We need this though so netpoll recognizes us as an interface that
1160 * supports polling, which enables bridge devices in virt setups to
1161 * still use netconsole
1162 * If NAPI is enabled, however, we need to schedule polling for all
1163 * queues unless we are using napi_gro_frags(), which we call in
1164 * process context and not in NAPI context.
1165 */
1166 struct tun_struct *tun = netdev_priv(dev);
1167
1168 if (tun->flags & IFF_NAPI) {
1169 struct tun_file *tfile;
1170 int i;
1171
1172 if (tun_napi_frags_enabled(tun))
1173 return;
1174
1175 rcu_read_lock();
1176 for (i = 0; i < tun->numqueues; i++) {
1177 tfile = rcu_dereference(tun->tfiles[i]);
1178 if (tfile->napi_enabled)
1179 napi_schedule(&tfile->napi);
1180 }
1181 rcu_read_unlock();
1182 }
1183 return;
1184}
1185#endif
1186
1187static void tun_set_headroom(struct net_device *dev, int new_hr)
1188{
1189 struct tun_struct *tun = netdev_priv(dev);
1190
1191 if (new_hr < NET_SKB_PAD)
1192 new_hr = NET_SKB_PAD;
1193
1194 tun->align = new_hr;
1195}
1196
1197static void
1198tun_net_get_stats64(struct net_device *dev, struct rtnl_link_stats64 *stats)
1199{
1200 u32 rx_dropped = 0, tx_dropped = 0, rx_frame_errors = 0;
1201 struct tun_struct *tun = netdev_priv(dev);
1202 struct tun_pcpu_stats *p;
1203 int i;
1204
1205 for_each_possible_cpu(i) {
1206 u64 rxpackets, rxbytes, txpackets, txbytes;
1207 unsigned int start;
1208
1209 p = per_cpu_ptr(tun->pcpu_stats, i);
1210 do {
1211 start = u64_stats_fetch_begin(&p->syncp);
1212 rxpackets = p->rx_packets;
1213 rxbytes = p->rx_bytes;
1214 txpackets = p->tx_packets;
1215 txbytes = p->tx_bytes;
1216 } while (u64_stats_fetch_retry(&p->syncp, start));
1217
1218 stats->rx_packets += rxpackets;
1219 stats->rx_bytes += rxbytes;
1220 stats->tx_packets += txpackets;
1221 stats->tx_bytes += txbytes;
1222
1223 /* u32 counters */
1224 rx_dropped += p->rx_dropped;
1225 rx_frame_errors += p->rx_frame_errors;
1226 tx_dropped += p->tx_dropped;
1227 }
1228 stats->rx_dropped = rx_dropped;
1229 stats->rx_frame_errors = rx_frame_errors;
1230 stats->tx_dropped = tx_dropped;
1231}
1232
1233static int tun_xdp_set(struct net_device *dev, struct bpf_prog *prog,
1234 struct netlink_ext_ack *extack)
1235{
1236 struct tun_struct *tun = netdev_priv(dev);
1237 struct bpf_prog *old_prog;
1238
1239 old_prog = rtnl_dereference(tun->xdp_prog);
1240 rcu_assign_pointer(tun->xdp_prog, prog);
1241 if (old_prog)
1242 bpf_prog_put(old_prog);
1243
1244 return 0;
1245}
1246
1247static u32 tun_xdp_query(struct net_device *dev)
1248{
1249 struct tun_struct *tun = netdev_priv(dev);
1250 const struct bpf_prog *xdp_prog;
1251
1252 xdp_prog = rtnl_dereference(tun->xdp_prog);
1253 if (xdp_prog)
1254 return xdp_prog->aux->id;
1255
1256 return 0;
1257}
1258
1259static int tun_xdp(struct net_device *dev, struct netdev_bpf *xdp)
1260{
1261 switch (xdp->command) {
1262 case XDP_SETUP_PROG:
1263 return tun_xdp_set(dev, xdp->prog, xdp->extack);
1264 case XDP_QUERY_PROG:
1265 xdp->prog_id = tun_xdp_query(dev);
1266 xdp->prog_attached = !!xdp->prog_id;
1267 return 0;
1268 default:
1269 return -EINVAL;
1270 }
1271}
1272
1273static const struct net_device_ops tun_netdev_ops = {
1274 .ndo_uninit = tun_net_uninit,
1275 .ndo_open = tun_net_open,
1276 .ndo_stop = tun_net_close,
1277 .ndo_start_xmit = tun_net_xmit,
1278 .ndo_fix_features = tun_net_fix_features,
1279 .ndo_select_queue = tun_select_queue,
1280#ifdef CONFIG_NET_POLL_CONTROLLER
1281 .ndo_poll_controller = tun_poll_controller,
1282#endif
1283 .ndo_set_rx_headroom = tun_set_headroom,
1284 .ndo_get_stats64 = tun_net_get_stats64,
1285};
1286
1287static int tun_xdp_xmit(struct net_device *dev, struct xdp_buff *xdp)
1288{
1289 struct tun_struct *tun = netdev_priv(dev);
1290 struct xdp_buff *buff = xdp->data_hard_start;
1291 int headroom = xdp->data - xdp->data_hard_start;
1292 struct tun_file *tfile;
1293 u32 numqueues;
1294 int ret = 0;
1295
1296 /* Assure headroom is available and buff is properly aligned */
1297 if (unlikely(headroom < sizeof(*xdp) || tun_is_xdp_buff(xdp)))
1298 return -ENOSPC;
1299
1300 *buff = *xdp;
1301
1302 rcu_read_lock();
1303
1304 numqueues = READ_ONCE(tun->numqueues);
1305 if (!numqueues) {
1306 ret = -ENOSPC;
1307 goto out;
1308 }
1309
1310 tfile = rcu_dereference(tun->tfiles[smp_processor_id() %
1311 numqueues]);
1312 /* Encode the XDP flag into lowest bit for consumer to differ
1313 * XDP buffer from sk_buff.
1314 */
1315 if (ptr_ring_produce(&tfile->tx_ring, tun_xdp_to_ptr(buff))) {
1316 this_cpu_inc(tun->pcpu_stats->tx_dropped);
1317 ret = -ENOSPC;
1318 }
1319
1320out:
1321 rcu_read_unlock();
1322 return ret;
1323}
1324
1325static void tun_xdp_flush(struct net_device *dev)
1326{
1327 struct tun_struct *tun = netdev_priv(dev);
1328 struct tun_file *tfile;
1329 u32 numqueues;
1330
1331 rcu_read_lock();
1332
1333 numqueues = READ_ONCE(tun->numqueues);
1334 if (!numqueues)
1335 goto out;
1336
1337 tfile = rcu_dereference(tun->tfiles[smp_processor_id() %
1338 numqueues]);
1339 /* Notify and wake up reader process */
1340 if (tfile->flags & TUN_FASYNC)
1341 kill_fasync(&tfile->fasync, SIGIO, POLL_IN);
1342 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
1343
1344out:
1345 rcu_read_unlock();
1346}
1347
1348static const struct net_device_ops tap_netdev_ops = {
1349 .ndo_uninit = tun_net_uninit,
1350 .ndo_open = tun_net_open,
1351 .ndo_stop = tun_net_close,
1352 .ndo_start_xmit = tun_net_xmit,
1353 .ndo_fix_features = tun_net_fix_features,
1354 .ndo_set_rx_mode = tun_net_mclist,
1355 .ndo_set_mac_address = eth_mac_addr,
1356 .ndo_validate_addr = eth_validate_addr,
1357 .ndo_select_queue = tun_select_queue,
1358#ifdef CONFIG_NET_POLL_CONTROLLER
1359 .ndo_poll_controller = tun_poll_controller,
1360#endif
1361 .ndo_features_check = passthru_features_check,
1362 .ndo_set_rx_headroom = tun_set_headroom,
1363 .ndo_get_stats64 = tun_net_get_stats64,
1364 .ndo_bpf = tun_xdp,
1365 .ndo_xdp_xmit = tun_xdp_xmit,
1366 .ndo_xdp_flush = tun_xdp_flush,
1367};
1368
1369static void tun_flow_init(struct tun_struct *tun)
1370{
1371 int i;
1372
1373 for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++)
1374 INIT_HLIST_HEAD(&tun->flows[i]);
1375
1376 tun->ageing_time = TUN_FLOW_EXPIRE;
1377 timer_setup(&tun->flow_gc_timer, tun_flow_cleanup, 0);
1378 mod_timer(&tun->flow_gc_timer,
1379 round_jiffies_up(jiffies + tun->ageing_time));
1380}
1381
1382static void tun_flow_uninit(struct tun_struct *tun)
1383{
1384 del_timer_sync(&tun->flow_gc_timer);
1385 tun_flow_flush(tun);
1386}
1387
1388#define MIN_MTU 68
1389#define MAX_MTU 65535
1390
1391/* Initialize net device. */
1392static void tun_net_init(struct net_device *dev)
1393{
1394 struct tun_struct *tun = netdev_priv(dev);
1395
1396 switch (tun->flags & TUN_TYPE_MASK) {
1397 case IFF_TUN:
1398 dev->netdev_ops = &tun_netdev_ops;
1399
1400 /* Point-to-Point TUN Device */
1401 dev->hard_header_len = 0;
1402 dev->addr_len = 0;
1403 dev->mtu = 1500;
1404
1405 /* Zero header length */
1406 dev->type = ARPHRD_NONE;
1407 dev->flags = IFF_POINTOPOINT | IFF_NOARP | IFF_MULTICAST;
1408 break;
1409
1410 case IFF_TAP:
1411 dev->netdev_ops = &tap_netdev_ops;
1412 /* Ethernet TAP Device */
1413 ether_setup(dev);
1414 dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1415 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
1416
1417 eth_hw_addr_random(dev);
1418
1419 break;
1420 }
1421
1422 dev->min_mtu = MIN_MTU;
1423 dev->max_mtu = MAX_MTU - dev->hard_header_len;
1424}
1425
1426static bool tun_sock_writeable(struct tun_struct *tun, struct tun_file *tfile)
1427{
1428 struct sock *sk = tfile->socket.sk;
1429
1430 return (tun->dev->flags & IFF_UP) && sock_writeable(sk);
1431}
1432
1433/* Character device part */
1434
1435/* Poll */
1436static __poll_t tun_chr_poll(struct file *file, poll_table *wait)
1437{
1438 struct tun_file *tfile = file->private_data;
1439 struct tun_struct *tun = tun_get(tfile);
1440 struct sock *sk;
1441 __poll_t mask = 0;
1442
1443 if (!tun)
1444 return EPOLLERR;
1445
1446 sk = tfile->socket.sk;
1447
1448 tun_debug(KERN_INFO, tun, "tun_chr_poll\n");
1449
1450 poll_wait(file, sk_sleep(sk), wait);
1451
1452 if (!ptr_ring_empty(&tfile->tx_ring))
1453 mask |= EPOLLIN | EPOLLRDNORM;
1454
1455 /* Make sure SOCKWQ_ASYNC_NOSPACE is set if not writable to
1456 * guarantee EPOLLOUT to be raised by either here or
1457 * tun_sock_write_space(). Then process could get notification
1458 * after it writes to a down device and meets -EIO.
1459 */
1460 if (tun_sock_writeable(tun, tfile) ||
1461 (!test_and_set_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags) &&
1462 tun_sock_writeable(tun, tfile)))
1463 mask |= EPOLLOUT | EPOLLWRNORM;
1464
1465 if (tun->dev->reg_state != NETREG_REGISTERED)
1466 mask = EPOLLERR;
1467
1468 tun_put(tun);
1469 return mask;
1470}
1471
1472static struct sk_buff *tun_napi_alloc_frags(struct tun_file *tfile,
1473 size_t len,
1474 const struct iov_iter *it)
1475{
1476 struct sk_buff *skb;
1477 size_t linear;
1478 int err;
1479 int i;
1480
1481 if (it->nr_segs > MAX_SKB_FRAGS + 1)
1482 return ERR_PTR(-ENOMEM);
1483
1484 local_bh_disable();
1485 skb = napi_get_frags(&tfile->napi);
1486 local_bh_enable();
1487 if (!skb)
1488 return ERR_PTR(-ENOMEM);
1489
1490 linear = iov_iter_single_seg_count(it);
1491 err = __skb_grow(skb, linear);
1492 if (err)
1493 goto free;
1494
1495 skb->len = len;
1496 skb->data_len = len - linear;
1497 skb->truesize += skb->data_len;
1498
1499 for (i = 1; i < it->nr_segs; i++) {
1500 struct page_frag *pfrag = ¤t->task_frag;
1501 size_t fragsz = it->iov[i].iov_len;
1502
1503 if (fragsz == 0 || fragsz > PAGE_SIZE) {
1504 err = -EINVAL;
1505 goto free;
1506 }
1507
1508 if (!skb_page_frag_refill(fragsz, pfrag, GFP_KERNEL)) {
1509 err = -ENOMEM;
1510 goto free;
1511 }
1512
1513 skb_fill_page_desc(skb, i - 1, pfrag->page,
1514 pfrag->offset, fragsz);
1515 page_ref_inc(pfrag->page);
1516 pfrag->offset += fragsz;
1517 }
1518
1519 return skb;
1520free:
1521 /* frees skb and all frags allocated with napi_alloc_frag() */
1522 napi_free_frags(&tfile->napi);
1523 return ERR_PTR(err);
1524}
1525
1526/* prepad is the amount to reserve at front. len is length after that.
1527 * linear is a hint as to how much to copy (usually headers). */
1528static struct sk_buff *tun_alloc_skb(struct tun_file *tfile,
1529 size_t prepad, size_t len,
1530 size_t linear, int noblock)
1531{
1532 struct sock *sk = tfile->socket.sk;
1533 struct sk_buff *skb;
1534 int err;
1535
1536 /* Under a page? Don't bother with paged skb. */
1537 if (prepad + len < PAGE_SIZE || !linear)
1538 linear = len;
1539
1540 skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
1541 &err, 0);
1542 if (!skb)
1543 return ERR_PTR(err);
1544
1545 skb_reserve(skb, prepad);
1546 skb_put(skb, linear);
1547 skb->data_len = len - linear;
1548 skb->len += len - linear;
1549
1550 return skb;
1551}
1552
1553static void tun_rx_batched(struct tun_struct *tun, struct tun_file *tfile,
1554 struct sk_buff *skb, int more)
1555{
1556 struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
1557 struct sk_buff_head process_queue;
1558 u32 rx_batched = tun->rx_batched;
1559 bool rcv = false;
1560
1561 if (!rx_batched || (!more && skb_queue_empty(queue))) {
1562 local_bh_disable();
1563 netif_receive_skb(skb);
1564 local_bh_enable();
1565 return;
1566 }
1567
1568 spin_lock(&queue->lock);
1569 if (!more || skb_queue_len(queue) == rx_batched) {
1570 __skb_queue_head_init(&process_queue);
1571 skb_queue_splice_tail_init(queue, &process_queue);
1572 rcv = true;
1573 } else {
1574 __skb_queue_tail(queue, skb);
1575 }
1576 spin_unlock(&queue->lock);
1577
1578 if (rcv) {
1579 struct sk_buff *nskb;
1580
1581 local_bh_disable();
1582 while ((nskb = __skb_dequeue(&process_queue)))
1583 netif_receive_skb(nskb);
1584 netif_receive_skb(skb);
1585 local_bh_enable();
1586 }
1587}
1588
1589static bool tun_can_build_skb(struct tun_struct *tun, struct tun_file *tfile,
1590 int len, int noblock, bool zerocopy)
1591{
1592 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
1593 return false;
1594
1595 if (tfile->socket.sk->sk_sndbuf != INT_MAX)
1596 return false;
1597
1598 if (!noblock)
1599 return false;
1600
1601 if (zerocopy)
1602 return false;
1603
1604 if (SKB_DATA_ALIGN(len + TUN_RX_PAD) +
1605 SKB_DATA_ALIGN(sizeof(struct skb_shared_info)) > PAGE_SIZE)
1606 return false;
1607
1608 return true;
1609}
1610
1611static struct sk_buff *tun_build_skb(struct tun_struct *tun,
1612 struct tun_file *tfile,
1613 struct iov_iter *from,
1614 struct virtio_net_hdr *hdr,
1615 int len, int *skb_xdp)
1616{
1617 struct page_frag *alloc_frag = ¤t->task_frag;
1618 struct sk_buff *skb;
1619 struct bpf_prog *xdp_prog;
1620 int buflen = SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
1621 unsigned int delta = 0;
1622 char *buf;
1623 size_t copied;
1624 int err, pad = TUN_RX_PAD;
1625
1626 rcu_read_lock();
1627 xdp_prog = rcu_dereference(tun->xdp_prog);
1628 if (xdp_prog)
1629 pad += TUN_HEADROOM;
1630 buflen += SKB_DATA_ALIGN(len + pad);
1631 rcu_read_unlock();
1632
1633 alloc_frag->offset = ALIGN((u64)alloc_frag->offset, SMP_CACHE_BYTES);
1634 if (unlikely(!skb_page_frag_refill(buflen, alloc_frag, GFP_KERNEL)))
1635 return ERR_PTR(-ENOMEM);
1636
1637 buf = (char *)page_address(alloc_frag->page) + alloc_frag->offset;
1638 copied = copy_page_from_iter(alloc_frag->page,
1639 alloc_frag->offset + pad,
1640 len, from);
1641 if (copied != len)
1642 return ERR_PTR(-EFAULT);
1643
1644 /* There's a small window that XDP may be set after the check
1645 * of xdp_prog above, this should be rare and for simplicity
1646 * we do XDP on skb in case the headroom is not enough.
1647 */
1648 if (hdr->gso_type || !xdp_prog)
1649 *skb_xdp = 1;
1650 else
1651 *skb_xdp = 0;
1652
1653 local_bh_disable();
1654 rcu_read_lock();
1655 xdp_prog = rcu_dereference(tun->xdp_prog);
1656 if (xdp_prog && !*skb_xdp) {
1657 struct xdp_buff xdp;
1658 void *orig_data;
1659 u32 act;
1660
1661 xdp.data_hard_start = buf;
1662 xdp.data = buf + pad;
1663 xdp_set_data_meta_invalid(&xdp);
1664 xdp.data_end = xdp.data + len;
1665 xdp.rxq = &tfile->xdp_rxq;
1666 orig_data = xdp.data;
1667 act = bpf_prog_run_xdp(xdp_prog, &xdp);
1668
1669 switch (act) {
1670 case XDP_REDIRECT:
1671 get_page(alloc_frag->page);
1672 alloc_frag->offset += buflen;
1673 err = xdp_do_redirect(tun->dev, &xdp, xdp_prog);
1674 xdp_do_flush_map();
1675 if (err)
1676 goto err_redirect;
1677 rcu_read_unlock();
1678 local_bh_enable();
1679 return NULL;
1680 case XDP_TX:
1681 get_page(alloc_frag->page);
1682 alloc_frag->offset += buflen;
1683 if (tun_xdp_xmit(tun->dev, &xdp))
1684 goto err_redirect;
1685 tun_xdp_flush(tun->dev);
1686 rcu_read_unlock();
1687 local_bh_enable();
1688 return NULL;
1689 case XDP_PASS:
1690 delta = orig_data - xdp.data;
1691 break;
1692 default:
1693 bpf_warn_invalid_xdp_action(act);
1694 /* fall through */
1695 case XDP_ABORTED:
1696 trace_xdp_exception(tun->dev, xdp_prog, act);
1697 /* fall through */
1698 case XDP_DROP:
1699 goto err_xdp;
1700 }
1701 }
1702
1703 skb = build_skb(buf, buflen);
1704 if (!skb) {
1705 rcu_read_unlock();
1706 local_bh_enable();
1707 return ERR_PTR(-ENOMEM);
1708 }
1709
1710 skb_reserve(skb, pad - delta);
1711 skb_put(skb, len + delta);
1712 get_page(alloc_frag->page);
1713 alloc_frag->offset += buflen;
1714
1715 rcu_read_unlock();
1716 local_bh_enable();
1717
1718 return skb;
1719
1720err_redirect:
1721 put_page(alloc_frag->page);
1722err_xdp:
1723 rcu_read_unlock();
1724 local_bh_enable();
1725 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1726 return NULL;
1727}
1728
1729/* Get packet from user space buffer */
1730static ssize_t tun_get_user(struct tun_struct *tun, struct tun_file *tfile,
1731 void *msg_control, struct iov_iter *from,
1732 int noblock, bool more)
1733{
1734 struct tun_pi pi = { 0, cpu_to_be16(ETH_P_IP) };
1735 struct sk_buff *skb;
1736 size_t total_len = iov_iter_count(from);
1737 size_t len = total_len, align = tun->align, linear;
1738 struct virtio_net_hdr gso = { 0 };
1739 struct tun_pcpu_stats *stats;
1740 int good_linear;
1741 int copylen;
1742 bool zerocopy = false;
1743 int err;
1744 u32 rxhash = 0;
1745 int skb_xdp = 1;
1746 bool frags = tun_napi_frags_enabled(tun);
1747
1748 if (!(tun->dev->flags & IFF_UP))
1749 return -EIO;
1750
1751 if (!(tun->flags & IFF_NO_PI)) {
1752 if (len < sizeof(pi))
1753 return -EINVAL;
1754 len -= sizeof(pi);
1755
1756 if (!copy_from_iter_full(&pi, sizeof(pi), from))
1757 return -EFAULT;
1758 }
1759
1760 if (tun->flags & IFF_VNET_HDR) {
1761 int vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
1762
1763 if (len < vnet_hdr_sz)
1764 return -EINVAL;
1765 len -= vnet_hdr_sz;
1766
1767 if (!copy_from_iter_full(&gso, sizeof(gso), from))
1768 return -EFAULT;
1769
1770 if ((gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
1771 tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2 > tun16_to_cpu(tun, gso.hdr_len))
1772 gso.hdr_len = cpu_to_tun16(tun, tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2);
1773
1774 if (tun16_to_cpu(tun, gso.hdr_len) > len)
1775 return -EINVAL;
1776 iov_iter_advance(from, vnet_hdr_sz - sizeof(gso));
1777 }
1778
1779 if ((tun->flags & TUN_TYPE_MASK) == IFF_TAP) {
1780 align += NET_IP_ALIGN;
1781 if (unlikely(len < ETH_HLEN ||
1782 (gso.hdr_len && tun16_to_cpu(tun, gso.hdr_len) < ETH_HLEN)))
1783 return -EINVAL;
1784 }
1785
1786 good_linear = SKB_MAX_HEAD(align);
1787
1788 if (msg_control) {
1789 struct iov_iter i = *from;
1790
1791 /* There are 256 bytes to be copied in skb, so there is
1792 * enough room for skb expand head in case it is used.
1793 * The rest of the buffer is mapped from userspace.
1794 */
1795 copylen = gso.hdr_len ? tun16_to_cpu(tun, gso.hdr_len) : GOODCOPY_LEN;
1796 if (copylen > good_linear)
1797 copylen = good_linear;
1798 linear = copylen;
1799 iov_iter_advance(&i, copylen);
1800 if (iov_iter_npages(&i, INT_MAX) <= MAX_SKB_FRAGS)
1801 zerocopy = true;
1802 }
1803
1804 if (!frags && tun_can_build_skb(tun, tfile, len, noblock, zerocopy)) {
1805 /* For the packet that is not easy to be processed
1806 * (e.g gso or jumbo packet), we will do it at after
1807 * skb was created with generic XDP routine.
1808 */
1809 skb = tun_build_skb(tun, tfile, from, &gso, len, &skb_xdp);
1810 if (IS_ERR(skb)) {
1811 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1812 return PTR_ERR(skb);
1813 }
1814 if (!skb)
1815 return total_len;
1816 } else {
1817 if (!zerocopy) {
1818 copylen = len;
1819 if (tun16_to_cpu(tun, gso.hdr_len) > good_linear)
1820 linear = good_linear;
1821 else
1822 linear = tun16_to_cpu(tun, gso.hdr_len);
1823 }
1824
1825 if (frags) {
1826 mutex_lock(&tfile->napi_mutex);
1827 skb = tun_napi_alloc_frags(tfile, copylen, from);
1828 /* tun_napi_alloc_frags() enforces a layout for the skb.
1829 * If zerocopy is enabled, then this layout will be
1830 * overwritten by zerocopy_sg_from_iter().
1831 */
1832 zerocopy = false;
1833 } else {
1834 skb = tun_alloc_skb(tfile, align, copylen, linear,
1835 noblock);
1836 }
1837
1838 if (IS_ERR(skb)) {
1839 if (PTR_ERR(skb) != -EAGAIN)
1840 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1841 if (frags)
1842 mutex_unlock(&tfile->napi_mutex);
1843 return PTR_ERR(skb);
1844 }
1845
1846 if (zerocopy)
1847 err = zerocopy_sg_from_iter(skb, from);
1848 else
1849 err = skb_copy_datagram_from_iter(skb, 0, from, len);
1850
1851 if (err) {
1852 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1853 kfree_skb(skb);
1854 if (frags) {
1855 tfile->napi.skb = NULL;
1856 mutex_unlock(&tfile->napi_mutex);
1857 }
1858
1859 return -EFAULT;
1860 }
1861 }
1862
1863 if (virtio_net_hdr_to_skb(skb, &gso, tun_is_little_endian(tun))) {
1864 this_cpu_inc(tun->pcpu_stats->rx_frame_errors);
1865 kfree_skb(skb);
1866 if (frags) {
1867 tfile->napi.skb = NULL;
1868 mutex_unlock(&tfile->napi_mutex);
1869 }
1870
1871 return -EINVAL;
1872 }
1873
1874 switch (tun->flags & TUN_TYPE_MASK) {
1875 case IFF_TUN:
1876 if (tun->flags & IFF_NO_PI) {
1877 u8 ip_version = skb->len ? (skb->data[0] >> 4) : 0;
1878
1879 switch (ip_version) {
1880 case 4:
1881 pi.proto = htons(ETH_P_IP);
1882 break;
1883 case 6:
1884 pi.proto = htons(ETH_P_IPV6);
1885 break;
1886 default:
1887 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1888 kfree_skb(skb);
1889 return -EINVAL;
1890 }
1891 }
1892
1893 skb_reset_mac_header(skb);
1894 skb->protocol = pi.proto;
1895 skb->dev = tun->dev;
1896 break;
1897 case IFF_TAP:
1898 if (!frags)
1899 skb->protocol = eth_type_trans(skb, tun->dev);
1900 break;
1901 }
1902
1903 /* copy skb_ubuf_info for callback when skb has no error */
1904 if (zerocopy) {
1905 skb_shinfo(skb)->destructor_arg = msg_control;
1906 skb_shinfo(skb)->tx_flags |= SKBTX_DEV_ZEROCOPY;
1907 skb_shinfo(skb)->tx_flags |= SKBTX_SHARED_FRAG;
1908 } else if (msg_control) {
1909 struct ubuf_info *uarg = msg_control;
1910 uarg->callback(uarg, false);
1911 }
1912
1913 skb_reset_network_header(skb);
1914 skb_probe_transport_header(skb, 0);
1915
1916 if (skb_xdp) {
1917 struct bpf_prog *xdp_prog;
1918 int ret;
1919
1920 local_bh_disable();
1921 rcu_read_lock();
1922 xdp_prog = rcu_dereference(tun->xdp_prog);
1923 if (xdp_prog) {
1924 ret = do_xdp_generic(xdp_prog, skb);
1925 if (ret != XDP_PASS) {
1926 rcu_read_unlock();
1927 local_bh_enable();
1928 return total_len;
1929 }
1930 }
1931 rcu_read_unlock();
1932 local_bh_enable();
1933 }
1934
1935 rcu_read_lock();
1936 if (!rcu_dereference(tun->steering_prog))
1937 rxhash = __skb_get_hash_symmetric(skb);
1938 rcu_read_unlock();
1939
1940 if (frags) {
1941 /* Exercise flow dissector code path. */
1942 u32 headlen = eth_get_headlen(skb->data, skb_headlen(skb));
1943
1944 if (unlikely(headlen > skb_headlen(skb))) {
1945 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1946 napi_free_frags(&tfile->napi);
1947 mutex_unlock(&tfile->napi_mutex);
1948 WARN_ON(1);
1949 return -ENOMEM;
1950 }
1951
1952 local_bh_disable();
1953 napi_gro_frags(&tfile->napi);
1954 local_bh_enable();
1955 mutex_unlock(&tfile->napi_mutex);
1956 } else if (tfile->napi_enabled) {
1957 struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
1958 int queue_len;
1959
1960 spin_lock_bh(&queue->lock);
1961 __skb_queue_tail(queue, skb);
1962 queue_len = skb_queue_len(queue);
1963 spin_unlock(&queue->lock);
1964
1965 if (!more || queue_len > NAPI_POLL_WEIGHT)
1966 napi_schedule(&tfile->napi);
1967
1968 local_bh_enable();
1969 } else if (!IS_ENABLED(CONFIG_4KSTACKS)) {
1970 tun_rx_batched(tun, tfile, skb, more);
1971 } else {
1972 netif_rx_ni(skb);
1973 }
1974
1975 stats = get_cpu_ptr(tun->pcpu_stats);
1976 u64_stats_update_begin(&stats->syncp);
1977 stats->rx_packets++;
1978 stats->rx_bytes += len;
1979 u64_stats_update_end(&stats->syncp);
1980 put_cpu_ptr(stats);
1981
1982 if (rxhash)
1983 tun_flow_update(tun, rxhash, tfile);
1984
1985 return total_len;
1986}
1987
1988static ssize_t tun_chr_write_iter(struct kiocb *iocb, struct iov_iter *from)
1989{
1990 struct file *file = iocb->ki_filp;
1991 struct tun_file *tfile = file->private_data;
1992 struct tun_struct *tun = tun_get(tfile);
1993 ssize_t result;
1994
1995 if (!tun)
1996 return -EBADFD;
1997
1998 result = tun_get_user(tun, tfile, NULL, from,
1999 file->f_flags & O_NONBLOCK, false);
2000
2001 tun_put(tun);
2002 return result;
2003}
2004
2005static ssize_t tun_put_user_xdp(struct tun_struct *tun,
2006 struct tun_file *tfile,
2007 struct xdp_buff *xdp,
2008 struct iov_iter *iter)
2009{
2010 int vnet_hdr_sz = 0;
2011 size_t size = xdp->data_end - xdp->data;
2012 struct tun_pcpu_stats *stats;
2013 size_t ret;
2014
2015 if (tun->flags & IFF_VNET_HDR) {
2016 struct virtio_net_hdr gso = { 0 };
2017
2018 vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
2019 if (unlikely(iov_iter_count(iter) < vnet_hdr_sz))
2020 return -EINVAL;
2021 if (unlikely(copy_to_iter(&gso, sizeof(gso), iter) !=
2022 sizeof(gso)))
2023 return -EFAULT;
2024 iov_iter_advance(iter, vnet_hdr_sz - sizeof(gso));
2025 }
2026
2027 ret = copy_to_iter(xdp->data, size, iter) + vnet_hdr_sz;
2028
2029 stats = get_cpu_ptr(tun->pcpu_stats);
2030 u64_stats_update_begin(&stats->syncp);
2031 stats->tx_packets++;
2032 stats->tx_bytes += ret;
2033 u64_stats_update_end(&stats->syncp);
2034 put_cpu_ptr(tun->pcpu_stats);
2035
2036 return ret;
2037}
2038
2039/* Put packet to the user space buffer */
2040static ssize_t tun_put_user(struct tun_struct *tun,
2041 struct tun_file *tfile,
2042 struct sk_buff *skb,
2043 struct iov_iter *iter)
2044{
2045 struct tun_pi pi = { 0, skb->protocol };
2046 struct tun_pcpu_stats *stats;
2047 ssize_t total;
2048 int vlan_offset = 0;
2049 int vlan_hlen = 0;
2050 int vnet_hdr_sz = 0;
2051
2052 if (skb_vlan_tag_present(skb))
2053 vlan_hlen = VLAN_HLEN;
2054
2055 if (tun->flags & IFF_VNET_HDR)
2056 vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
2057
2058 total = skb->len + vlan_hlen + vnet_hdr_sz;
2059
2060 if (!(tun->flags & IFF_NO_PI)) {
2061 if (iov_iter_count(iter) < sizeof(pi))
2062 return -EINVAL;
2063
2064 total += sizeof(pi);
2065 if (iov_iter_count(iter) < total) {
2066 /* Packet will be striped */
2067 pi.flags |= TUN_PKT_STRIP;
2068 }
2069
2070 if (copy_to_iter(&pi, sizeof(pi), iter) != sizeof(pi))
2071 return -EFAULT;
2072 }
2073
2074 if (vnet_hdr_sz) {
2075 struct virtio_net_hdr gso;
2076
2077 if (iov_iter_count(iter) < vnet_hdr_sz)
2078 return -EINVAL;
2079
2080 if (virtio_net_hdr_from_skb(skb, &gso,
2081 tun_is_little_endian(tun), true)) {
2082 struct skb_shared_info *sinfo = skb_shinfo(skb);
2083 pr_err("unexpected GSO type: "
2084 "0x%x, gso_size %d, hdr_len %d\n",
2085 sinfo->gso_type, tun16_to_cpu(tun, gso.gso_size),
2086 tun16_to_cpu(tun, gso.hdr_len));
2087 print_hex_dump(KERN_ERR, "tun: ",
2088 DUMP_PREFIX_NONE,
2089 16, 1, skb->head,
2090 min((int)tun16_to_cpu(tun, gso.hdr_len), 64), true);
2091 WARN_ON_ONCE(1);
2092 return -EINVAL;
2093 }
2094
2095 if (copy_to_iter(&gso, sizeof(gso), iter) != sizeof(gso))
2096 return -EFAULT;
2097
2098 iov_iter_advance(iter, vnet_hdr_sz - sizeof(gso));
2099 }
2100
2101 if (vlan_hlen) {
2102 int ret;
2103 struct veth veth;
2104
2105 veth.h_vlan_proto = skb->vlan_proto;
2106 veth.h_vlan_TCI = htons(skb_vlan_tag_get(skb));
2107
2108 vlan_offset = offsetof(struct vlan_ethhdr, h_vlan_proto);
2109
2110 ret = skb_copy_datagram_iter(skb, 0, iter, vlan_offset);
2111 if (ret || !iov_iter_count(iter))
2112 goto done;
2113
2114 ret = copy_to_iter(&veth, sizeof(veth), iter);
2115 if (ret != sizeof(veth) || !iov_iter_count(iter))
2116 goto done;
2117 }
2118
2119 skb_copy_datagram_iter(skb, vlan_offset, iter, skb->len - vlan_offset);
2120
2121done:
2122 /* caller is in process context, */
2123 stats = get_cpu_ptr(tun->pcpu_stats);
2124 u64_stats_update_begin(&stats->syncp);
2125 stats->tx_packets++;
2126 stats->tx_bytes += skb->len + vlan_hlen;
2127 u64_stats_update_end(&stats->syncp);
2128 put_cpu_ptr(tun->pcpu_stats);
2129
2130 return total;
2131}
2132
2133static void *tun_ring_recv(struct tun_file *tfile, int noblock, int *err)
2134{
2135 DECLARE_WAITQUEUE(wait, current);
2136 void *ptr = NULL;
2137 int error = 0;
2138
2139 ptr = ptr_ring_consume(&tfile->tx_ring);
2140 if (ptr)
2141 goto out;
2142 if (noblock) {
2143 error = -EAGAIN;
2144 goto out;
2145 }
2146
2147 add_wait_queue(&tfile->wq.wait, &wait);
2148 current->state = TASK_INTERRUPTIBLE;
2149
2150 while (1) {
2151 ptr = ptr_ring_consume(&tfile->tx_ring);
2152 if (ptr)
2153 break;
2154 if (signal_pending(current)) {
2155 error = -ERESTARTSYS;
2156 break;
2157 }
2158 if (tfile->socket.sk->sk_shutdown & RCV_SHUTDOWN) {
2159 error = -EFAULT;
2160 break;
2161 }
2162
2163 schedule();
2164 }
2165
2166 current->state = TASK_RUNNING;
2167 remove_wait_queue(&tfile->wq.wait, &wait);
2168
2169out:
2170 *err = error;
2171 return ptr;
2172}
2173
2174static ssize_t tun_do_read(struct tun_struct *tun, struct tun_file *tfile,
2175 struct iov_iter *to,
2176 int noblock, void *ptr)
2177{
2178 ssize_t ret;
2179 int err;
2180
2181 tun_debug(KERN_INFO, tun, "tun_do_read\n");
2182
2183 if (!iov_iter_count(to)) {
2184 tun_ptr_free(ptr);
2185 return 0;
2186 }
2187
2188 if (!ptr) {
2189 /* Read frames from ring */
2190 ptr = tun_ring_recv(tfile, noblock, &err);
2191 if (!ptr)
2192 return err;
2193 }
2194
2195 if (tun_is_xdp_buff(ptr)) {
2196 struct xdp_buff *xdp = tun_ptr_to_xdp(ptr);
2197
2198 ret = tun_put_user_xdp(tun, tfile, xdp, to);
2199 put_page(virt_to_head_page(xdp->data));
2200 } else {
2201 struct sk_buff *skb = ptr;
2202
2203 ret = tun_put_user(tun, tfile, skb, to);
2204 if (unlikely(ret < 0))
2205 kfree_skb(skb);
2206 else
2207 consume_skb(skb);
2208 }
2209
2210 return ret;
2211}
2212
2213static ssize_t tun_chr_read_iter(struct kiocb *iocb, struct iov_iter *to)
2214{
2215 struct file *file = iocb->ki_filp;
2216 struct tun_file *tfile = file->private_data;
2217 struct tun_struct *tun = tun_get(tfile);
2218 ssize_t len = iov_iter_count(to), ret;
2219
2220 if (!tun)
2221 return -EBADFD;
2222 ret = tun_do_read(tun, tfile, to, file->f_flags & O_NONBLOCK, NULL);
2223 ret = min_t(ssize_t, ret, len);
2224 if (ret > 0)
2225 iocb->ki_pos = ret;
2226 tun_put(tun);
2227 return ret;
2228}
2229
2230static void tun_prog_free(struct rcu_head *rcu)
2231{
2232 struct tun_prog *prog = container_of(rcu, struct tun_prog, rcu);
2233
2234 bpf_prog_destroy(prog->prog);
2235 kfree(prog);
2236}
2237
2238static int __tun_set_ebpf(struct tun_struct *tun,
2239 struct tun_prog __rcu **prog_p,
2240 struct bpf_prog *prog)
2241{
2242 struct tun_prog *old, *new = NULL;
2243
2244 if (prog) {
2245 new = kmalloc(sizeof(*new), GFP_KERNEL);
2246 if (!new)
2247 return -ENOMEM;
2248 new->prog = prog;
2249 }
2250
2251 spin_lock_bh(&tun->lock);
2252 old = rcu_dereference_protected(*prog_p,
2253 lockdep_is_held(&tun->lock));
2254 rcu_assign_pointer(*prog_p, new);
2255 spin_unlock_bh(&tun->lock);
2256
2257 if (old)
2258 call_rcu(&old->rcu, tun_prog_free);
2259
2260 return 0;
2261}
2262
2263static void tun_free_netdev(struct net_device *dev)
2264{
2265 struct tun_struct *tun = netdev_priv(dev);
2266
2267 BUG_ON(!(list_empty(&tun->disabled)));
2268 free_percpu(tun->pcpu_stats);
2269 tun_flow_uninit(tun);
2270 security_tun_dev_free_security(tun->security);
2271 __tun_set_ebpf(tun, &tun->steering_prog, NULL);
2272 __tun_set_ebpf(tun, &tun->filter_prog, NULL);
2273}
2274
2275static void tun_setup(struct net_device *dev)
2276{
2277 struct tun_struct *tun = netdev_priv(dev);
2278
2279 tun->owner = INVALID_UID;
2280 tun->group = INVALID_GID;
2281
2282 dev->ethtool_ops = &tun_ethtool_ops;
2283 dev->needs_free_netdev = true;
2284 dev->priv_destructor = tun_free_netdev;
2285 /* We prefer our own queue length */
2286 dev->tx_queue_len = TUN_READQ_SIZE;
2287}
2288
2289/* Trivial set of netlink ops to allow deleting tun or tap
2290 * device with netlink.
2291 */
2292static int tun_validate(struct nlattr *tb[], struct nlattr *data[],
2293 struct netlink_ext_ack *extack)
2294{
2295 return -EINVAL;
2296}
2297
2298static size_t tun_get_size(const struct net_device *dev)
2299{
2300 BUILD_BUG_ON(sizeof(u32) != sizeof(uid_t));
2301 BUILD_BUG_ON(sizeof(u32) != sizeof(gid_t));
2302
2303 return nla_total_size(sizeof(uid_t)) + /* OWNER */
2304 nla_total_size(sizeof(gid_t)) + /* GROUP */
2305 nla_total_size(sizeof(u8)) + /* TYPE */
2306 nla_total_size(sizeof(u8)) + /* PI */
2307 nla_total_size(sizeof(u8)) + /* VNET_HDR */
2308 nla_total_size(sizeof(u8)) + /* PERSIST */
2309 nla_total_size(sizeof(u8)) + /* MULTI_QUEUE */
2310 nla_total_size(sizeof(u32)) + /* NUM_QUEUES */
2311 nla_total_size(sizeof(u32)) + /* NUM_DISABLED_QUEUES */
2312 0;
2313}
2314
2315static int tun_fill_info(struct sk_buff *skb, const struct net_device *dev)
2316{
2317 struct tun_struct *tun = netdev_priv(dev);
2318
2319 if (nla_put_u8(skb, IFLA_TUN_TYPE, tun->flags & TUN_TYPE_MASK))
2320 goto nla_put_failure;
2321 if (uid_valid(tun->owner) &&
2322 nla_put_u32(skb, IFLA_TUN_OWNER,
2323 from_kuid_munged(current_user_ns(), tun->owner)))
2324 goto nla_put_failure;
2325 if (gid_valid(tun->group) &&
2326 nla_put_u32(skb, IFLA_TUN_GROUP,
2327 from_kgid_munged(current_user_ns(), tun->group)))
2328 goto nla_put_failure;
2329 if (nla_put_u8(skb, IFLA_TUN_PI, !(tun->flags & IFF_NO_PI)))
2330 goto nla_put_failure;
2331 if (nla_put_u8(skb, IFLA_TUN_VNET_HDR, !!(tun->flags & IFF_VNET_HDR)))
2332 goto nla_put_failure;
2333 if (nla_put_u8(skb, IFLA_TUN_PERSIST, !!(tun->flags & IFF_PERSIST)))
2334 goto nla_put_failure;
2335 if (nla_put_u8(skb, IFLA_TUN_MULTI_QUEUE,
2336 !!(tun->flags & IFF_MULTI_QUEUE)))
2337 goto nla_put_failure;
2338 if (tun->flags & IFF_MULTI_QUEUE) {
2339 if (nla_put_u32(skb, IFLA_TUN_NUM_QUEUES, tun->numqueues))
2340 goto nla_put_failure;
2341 if (nla_put_u32(skb, IFLA_TUN_NUM_DISABLED_QUEUES,
2342 tun->numdisabled))
2343 goto nla_put_failure;
2344 }
2345
2346 return 0;
2347
2348nla_put_failure:
2349 return -EMSGSIZE;
2350}
2351
2352static struct rtnl_link_ops tun_link_ops __read_mostly = {
2353 .kind = DRV_NAME,
2354 .priv_size = sizeof(struct tun_struct),
2355 .setup = tun_setup,
2356 .validate = tun_validate,
2357 .get_size = tun_get_size,
2358 .fill_info = tun_fill_info,
2359};
2360
2361static void tun_sock_write_space(struct sock *sk)
2362{
2363 struct tun_file *tfile;
2364 wait_queue_head_t *wqueue;
2365
2366 if (!sock_writeable(sk))
2367 return;
2368
2369 if (!test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags))
2370 return;
2371
2372 wqueue = sk_sleep(sk);
2373 if (wqueue && waitqueue_active(wqueue))
2374 wake_up_interruptible_sync_poll(wqueue, EPOLLOUT |
2375 EPOLLWRNORM | EPOLLWRBAND);
2376
2377 tfile = container_of(sk, struct tun_file, sk);
2378 kill_fasync(&tfile->fasync, SIGIO, POLL_OUT);
2379}
2380
2381static int tun_sendmsg(struct socket *sock, struct msghdr *m, size_t total_len)
2382{
2383 int ret;
2384 struct tun_file *tfile = container_of(sock, struct tun_file, socket);
2385 struct tun_struct *tun = tun_get(tfile);
2386
2387 if (!tun)
2388 return -EBADFD;
2389
2390 ret = tun_get_user(tun, tfile, m->msg_control, &m->msg_iter,
2391 m->msg_flags & MSG_DONTWAIT,
2392 m->msg_flags & MSG_MORE);
2393 tun_put(tun);
2394 return ret;
2395}
2396
2397static int tun_recvmsg(struct socket *sock, struct msghdr *m, size_t total_len,
2398 int flags)
2399{
2400 struct tun_file *tfile = container_of(sock, struct tun_file, socket);
2401 struct tun_struct *tun = tun_get(tfile);
2402 void *ptr = m->msg_control;
2403 int ret;
2404
2405 if (!tun) {
2406 ret = -EBADFD;
2407 goto out_free;
2408 }
2409
2410 if (flags & ~(MSG_DONTWAIT|MSG_TRUNC|MSG_ERRQUEUE)) {
2411 ret = -EINVAL;
2412 goto out_put_tun;
2413 }
2414 if (flags & MSG_ERRQUEUE) {
2415 ret = sock_recv_errqueue(sock->sk, m, total_len,
2416 SOL_PACKET, TUN_TX_TIMESTAMP);
2417 goto out;
2418 }
2419 ret = tun_do_read(tun, tfile, &m->msg_iter, flags & MSG_DONTWAIT, ptr);
2420 if (ret > (ssize_t)total_len) {
2421 m->msg_flags |= MSG_TRUNC;
2422 ret = flags & MSG_TRUNC ? ret : total_len;
2423 }
2424out:
2425 tun_put(tun);
2426 return ret;
2427
2428out_put_tun:
2429 tun_put(tun);
2430out_free:
2431 tun_ptr_free(ptr);
2432 return ret;
2433}
2434
2435static int tun_ptr_peek_len(void *ptr)
2436{
2437 if (likely(ptr)) {
2438 if (tun_is_xdp_buff(ptr)) {
2439 struct xdp_buff *xdp = tun_ptr_to_xdp(ptr);
2440
2441 return xdp->data_end - xdp->data;
2442 }
2443 return __skb_array_len_with_tag(ptr);
2444 } else {
2445 return 0;
2446 }
2447}
2448
2449static int tun_peek_len(struct socket *sock)
2450{
2451 struct tun_file *tfile = container_of(sock, struct tun_file, socket);
2452 struct tun_struct *tun;
2453 int ret = 0;
2454
2455 tun = tun_get(tfile);
2456 if (!tun)
2457 return 0;
2458
2459 ret = PTR_RING_PEEK_CALL(&tfile->tx_ring, tun_ptr_peek_len);
2460 tun_put(tun);
2461
2462 return ret;
2463}
2464
2465/* Ops structure to mimic raw sockets with tun */
2466static const struct proto_ops tun_socket_ops = {
2467 .peek_len = tun_peek_len,
2468 .sendmsg = tun_sendmsg,
2469 .recvmsg = tun_recvmsg,
2470};
2471
2472static struct proto tun_proto = {
2473 .name = "tun",
2474 .owner = THIS_MODULE,
2475 .obj_size = sizeof(struct tun_file),
2476};
2477
2478static int tun_flags(struct tun_struct *tun)
2479{
2480 return tun->flags & (TUN_FEATURES | IFF_PERSIST | IFF_TUN | IFF_TAP);
2481}
2482
2483static ssize_t tun_show_flags(struct device *dev, struct device_attribute *attr,
2484 char *buf)
2485{
2486 struct tun_struct *tun = netdev_priv(to_net_dev(dev));
2487 return sprintf(buf, "0x%x\n", tun_flags(tun));
2488}
2489
2490static ssize_t tun_show_owner(struct device *dev, struct device_attribute *attr,
2491 char *buf)
2492{
2493 struct tun_struct *tun = netdev_priv(to_net_dev(dev));
2494 return uid_valid(tun->owner)?
2495 sprintf(buf, "%u\n",
2496 from_kuid_munged(current_user_ns(), tun->owner)):
2497 sprintf(buf, "-1\n");
2498}
2499
2500static ssize_t tun_show_group(struct device *dev, struct device_attribute *attr,
2501 char *buf)
2502{
2503 struct tun_struct *tun = netdev_priv(to_net_dev(dev));
2504 return gid_valid(tun->group) ?
2505 sprintf(buf, "%u\n",
2506 from_kgid_munged(current_user_ns(), tun->group)):
2507 sprintf(buf, "-1\n");
2508}
2509
2510static DEVICE_ATTR(tun_flags, 0444, tun_show_flags, NULL);
2511static DEVICE_ATTR(owner, 0444, tun_show_owner, NULL);
2512static DEVICE_ATTR(group, 0444, tun_show_group, NULL);
2513
2514static struct attribute *tun_dev_attrs[] = {
2515 &dev_attr_tun_flags.attr,
2516 &dev_attr_owner.attr,
2517 &dev_attr_group.attr,
2518 NULL
2519};
2520
2521static const struct attribute_group tun_attr_group = {
2522 .attrs = tun_dev_attrs
2523};
2524
2525static int tun_set_iff(struct net *net, struct file *file, struct ifreq *ifr)
2526{
2527 struct tun_struct *tun;
2528 struct tun_file *tfile = file->private_data;
2529 struct net_device *dev;
2530 int err;
2531
2532 if (tfile->detached)
2533 return -EINVAL;
2534
2535 if ((ifr->ifr_flags & IFF_NAPI_FRAGS)) {
2536 if (!capable(CAP_NET_ADMIN))
2537 return -EPERM;
2538
2539 if (!(ifr->ifr_flags & IFF_NAPI) ||
2540 (ifr->ifr_flags & TUN_TYPE_MASK) != IFF_TAP)
2541 return -EINVAL;
2542 }
2543
2544 dev = __dev_get_by_name(net, ifr->ifr_name);
2545 if (dev) {
2546 if (ifr->ifr_flags & IFF_TUN_EXCL)
2547 return -EBUSY;
2548 if ((ifr->ifr_flags & IFF_TUN) && dev->netdev_ops == &tun_netdev_ops)
2549 tun = netdev_priv(dev);
2550 else if ((ifr->ifr_flags & IFF_TAP) && dev->netdev_ops == &tap_netdev_ops)
2551 tun = netdev_priv(dev);
2552 else
2553 return -EINVAL;
2554
2555 if (!!(ifr->ifr_flags & IFF_MULTI_QUEUE) !=
2556 !!(tun->flags & IFF_MULTI_QUEUE))
2557 return -EINVAL;
2558
2559 if (tun_not_capable(tun))
2560 return -EPERM;
2561 err = security_tun_dev_open(tun->security);
2562 if (err < 0)
2563 return err;
2564
2565 err = tun_attach(tun, file, ifr->ifr_flags & IFF_NOFILTER,
2566 ifr->ifr_flags & IFF_NAPI);
2567 if (err < 0)
2568 return err;
2569
2570 if (tun->flags & IFF_MULTI_QUEUE &&
2571 (tun->numqueues + tun->numdisabled > 1)) {
2572 /* One or more queue has already been attached, no need
2573 * to initialize the device again.
2574 */
2575 netdev_state_change(dev);
2576 return 0;
2577 }
2578
2579 tun->flags = (tun->flags & ~TUN_FEATURES) |
2580 (ifr->ifr_flags & TUN_FEATURES);
2581
2582 netdev_state_change(dev);
2583 } else {
2584 char *name;
2585 unsigned long flags = 0;
2586 int queues = ifr->ifr_flags & IFF_MULTI_QUEUE ?
2587 MAX_TAP_QUEUES : 1;
2588
2589 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
2590 return -EPERM;
2591 err = security_tun_dev_create();
2592 if (err < 0)
2593 return err;
2594
2595 /* Set dev type */
2596 if (ifr->ifr_flags & IFF_TUN) {
2597 /* TUN device */
2598 flags |= IFF_TUN;
2599 name = "tun%d";
2600 } else if (ifr->ifr_flags & IFF_TAP) {
2601 /* TAP device */
2602 flags |= IFF_TAP;
2603 name = "tap%d";
2604 } else
2605 return -EINVAL;
2606
2607 if (*ifr->ifr_name)
2608 name = ifr->ifr_name;
2609
2610 dev = alloc_netdev_mqs(sizeof(struct tun_struct), name,
2611 NET_NAME_UNKNOWN, tun_setup, queues,
2612 queues);
2613
2614 if (!dev)
2615 return -ENOMEM;
2616 err = dev_get_valid_name(net, dev, name);
2617 if (err < 0)
2618 goto err_free_dev;
2619
2620 dev_net_set(dev, net);
2621 dev->rtnl_link_ops = &tun_link_ops;
2622 dev->ifindex = tfile->ifindex;
2623 dev->sysfs_groups[0] = &tun_attr_group;
2624
2625 tun = netdev_priv(dev);
2626 tun->dev = dev;
2627 tun->flags = flags;
2628 tun->txflt.count = 0;
2629 tun->vnet_hdr_sz = sizeof(struct virtio_net_hdr);
2630
2631 tun->align = NET_SKB_PAD;
2632 tun->filter_attached = false;
2633 tun->sndbuf = tfile->socket.sk->sk_sndbuf;
2634 tun->rx_batched = 0;
2635 RCU_INIT_POINTER(tun->steering_prog, NULL);
2636
2637 tun->pcpu_stats = netdev_alloc_pcpu_stats(struct tun_pcpu_stats);
2638 if (!tun->pcpu_stats) {
2639 err = -ENOMEM;
2640 goto err_free_dev;
2641 }
2642
2643 spin_lock_init(&tun->lock);
2644
2645 err = security_tun_dev_alloc_security(&tun->security);
2646 if (err < 0)
2647 goto err_free_stat;
2648
2649 tun_net_init(dev);
2650 tun_flow_init(tun);
2651
2652 dev->hw_features = NETIF_F_SG | NETIF_F_FRAGLIST |
2653 TUN_USER_FEATURES | NETIF_F_HW_VLAN_CTAG_TX |
2654 NETIF_F_HW_VLAN_STAG_TX;
2655 dev->features = dev->hw_features | NETIF_F_LLTX;
2656 dev->vlan_features = dev->features &
2657 ~(NETIF_F_HW_VLAN_CTAG_TX |
2658 NETIF_F_HW_VLAN_STAG_TX);
2659
2660 tun->flags = (tun->flags & ~TUN_FEATURES) |
2661 (ifr->ifr_flags & TUN_FEATURES);
2662
2663 INIT_LIST_HEAD(&tun->disabled);
2664 err = tun_attach(tun, file, false, ifr->ifr_flags & IFF_NAPI);
2665 if (err < 0)
2666 goto err_free_flow;
2667
2668 err = register_netdevice(tun->dev);
2669 if (err < 0)
2670 goto err_detach;
2671 }
2672
2673 netif_carrier_on(tun->dev);
2674
2675 tun_debug(KERN_INFO, tun, "tun_set_iff\n");
2676
2677 /* Make sure persistent devices do not get stuck in
2678 * xoff state.
2679 */
2680 if (netif_running(tun->dev))
2681 netif_tx_wake_all_queues(tun->dev);
2682
2683 strcpy(ifr->ifr_name, tun->dev->name);
2684 return 0;
2685
2686err_detach:
2687 tun_detach_all(dev);
2688 /* register_netdevice() already called tun_free_netdev() */
2689 goto err_free_dev;
2690
2691err_free_flow:
2692 tun_flow_uninit(tun);
2693 security_tun_dev_free_security(tun->security);
2694err_free_stat:
2695 free_percpu(tun->pcpu_stats);
2696err_free_dev:
2697 free_netdev(dev);
2698 return err;
2699}
2700
2701static void tun_get_iff(struct net *net, struct tun_struct *tun,
2702 struct ifreq *ifr)
2703{
2704 tun_debug(KERN_INFO, tun, "tun_get_iff\n");
2705
2706 strcpy(ifr->ifr_name, tun->dev->name);
2707
2708 ifr->ifr_flags = tun_flags(tun);
2709
2710}
2711
2712/* This is like a cut-down ethtool ops, except done via tun fd so no
2713 * privs required. */
2714static int set_offload(struct tun_struct *tun, unsigned long arg)
2715{
2716 netdev_features_t features = 0;
2717
2718 if (arg & TUN_F_CSUM) {
2719 features |= NETIF_F_HW_CSUM;
2720 arg &= ~TUN_F_CSUM;
2721
2722 if (arg & (TUN_F_TSO4|TUN_F_TSO6)) {
2723 if (arg & TUN_F_TSO_ECN) {
2724 features |= NETIF_F_TSO_ECN;
2725 arg &= ~TUN_F_TSO_ECN;
2726 }
2727 if (arg & TUN_F_TSO4)
2728 features |= NETIF_F_TSO;
2729 if (arg & TUN_F_TSO6)
2730 features |= NETIF_F_TSO6;
2731 arg &= ~(TUN_F_TSO4|TUN_F_TSO6);
2732 }
2733
2734 arg &= ~TUN_F_UFO;
2735 }
2736
2737 /* This gives the user a way to test for new features in future by
2738 * trying to set them. */
2739 if (arg)
2740 return -EINVAL;
2741
2742 tun->set_features = features;
2743 tun->dev->wanted_features &= ~TUN_USER_FEATURES;
2744 tun->dev->wanted_features |= features;
2745 netdev_update_features(tun->dev);
2746
2747 return 0;
2748}
2749
2750static void tun_detach_filter(struct tun_struct *tun, int n)
2751{
2752 int i;
2753 struct tun_file *tfile;
2754
2755 for (i = 0; i < n; i++) {
2756 tfile = rtnl_dereference(tun->tfiles[i]);
2757 lock_sock(tfile->socket.sk);
2758 sk_detach_filter(tfile->socket.sk);
2759 release_sock(tfile->socket.sk);
2760 }
2761
2762 tun->filter_attached = false;
2763}
2764
2765static int tun_attach_filter(struct tun_struct *tun)
2766{
2767 int i, ret = 0;
2768 struct tun_file *tfile;
2769
2770 for (i = 0; i < tun->numqueues; i++) {
2771 tfile = rtnl_dereference(tun->tfiles[i]);
2772 lock_sock(tfile->socket.sk);
2773 ret = sk_attach_filter(&tun->fprog, tfile->socket.sk);
2774 release_sock(tfile->socket.sk);
2775 if (ret) {
2776 tun_detach_filter(tun, i);
2777 return ret;
2778 }
2779 }
2780
2781 tun->filter_attached = true;
2782 return ret;
2783}
2784
2785static void tun_set_sndbuf(struct tun_struct *tun)
2786{
2787 struct tun_file *tfile;
2788 int i;
2789
2790 for (i = 0; i < tun->numqueues; i++) {
2791 tfile = rtnl_dereference(tun->tfiles[i]);
2792 tfile->socket.sk->sk_sndbuf = tun->sndbuf;
2793 }
2794}
2795
2796static int tun_set_queue(struct file *file, struct ifreq *ifr)
2797{
2798 struct tun_file *tfile = file->private_data;
2799 struct tun_struct *tun;
2800 int ret = 0;
2801
2802 rtnl_lock();
2803
2804 if (ifr->ifr_flags & IFF_ATTACH_QUEUE) {
2805 tun = tfile->detached;
2806 if (!tun) {
2807 ret = -EINVAL;
2808 goto unlock;
2809 }
2810 ret = security_tun_dev_attach_queue(tun->security);
2811 if (ret < 0)
2812 goto unlock;
2813 ret = tun_attach(tun, file, false, tun->flags & IFF_NAPI);
2814 } else if (ifr->ifr_flags & IFF_DETACH_QUEUE) {
2815 tun = rtnl_dereference(tfile->tun);
2816 if (!tun || !(tun->flags & IFF_MULTI_QUEUE) || tfile->detached)
2817 ret = -EINVAL;
2818 else
2819 __tun_detach(tfile, false);
2820 } else
2821 ret = -EINVAL;
2822
2823 if (ret >= 0)
2824 netdev_state_change(tun->dev);
2825
2826unlock:
2827 rtnl_unlock();
2828 return ret;
2829}
2830
2831static int tun_set_ebpf(struct tun_struct *tun, struct tun_prog **prog_p,
2832 void __user *data)
2833{
2834 struct bpf_prog *prog;
2835 int fd;
2836
2837 if (copy_from_user(&fd, data, sizeof(fd)))
2838 return -EFAULT;
2839
2840 if (fd == -1) {
2841 prog = NULL;
2842 } else {
2843 prog = bpf_prog_get_type(fd, BPF_PROG_TYPE_SOCKET_FILTER);
2844 if (IS_ERR(prog))
2845 return PTR_ERR(prog);
2846 }
2847
2848 return __tun_set_ebpf(tun, prog_p, prog);
2849}
2850
2851static long __tun_chr_ioctl(struct file *file, unsigned int cmd,
2852 unsigned long arg, int ifreq_len)
2853{
2854 struct tun_file *tfile = file->private_data;
2855 struct tun_struct *tun;
2856 void __user* argp = (void __user*)arg;
2857 struct ifreq ifr;
2858 struct net *net;
2859 kuid_t owner;
2860 kgid_t group;
2861 int sndbuf;
2862 int vnet_hdr_sz;
2863 unsigned int ifindex;
2864 int le;
2865 int ret;
2866 bool do_notify = false;
2867
2868 if (cmd == TUNSETIFF || cmd == TUNSETQUEUE ||
2869 (_IOC_TYPE(cmd) == SOCK_IOC_TYPE && cmd != SIOCGSKNS)) {
2870 if (copy_from_user(&ifr, argp, ifreq_len))
2871 return -EFAULT;
2872 } else {
2873 memset(&ifr, 0, sizeof(ifr));
2874 }
2875 if (cmd == TUNGETFEATURES) {
2876 /* Currently this just means: "what IFF flags are valid?".
2877 * This is needed because we never checked for invalid flags on
2878 * TUNSETIFF.
2879 */
2880 return put_user(IFF_TUN | IFF_TAP | TUN_FEATURES,
2881 (unsigned int __user*)argp);
2882 } else if (cmd == TUNSETQUEUE)
2883 return tun_set_queue(file, &ifr);
2884
2885 ret = 0;
2886 rtnl_lock();
2887
2888 tun = tun_get(tfile);
2889 net = sock_net(&tfile->sk);
2890 if (cmd == TUNSETIFF) {
2891 ret = -EEXIST;
2892 if (tun)
2893 goto unlock;
2894
2895 ifr.ifr_name[IFNAMSIZ-1] = '\0';
2896
2897 ret = tun_set_iff(net, file, &ifr);
2898
2899 if (ret)
2900 goto unlock;
2901
2902 if (copy_to_user(argp, &ifr, ifreq_len))
2903 ret = -EFAULT;
2904 goto unlock;
2905 }
2906 if (cmd == TUNSETIFINDEX) {
2907 ret = -EPERM;
2908 if (tun)
2909 goto unlock;
2910
2911 ret = -EFAULT;
2912 if (copy_from_user(&ifindex, argp, sizeof(ifindex)))
2913 goto unlock;
2914
2915 ret = 0;
2916 tfile->ifindex = ifindex;
2917 goto unlock;
2918 }
2919 if (cmd == SIOCGSKNS) {
2920 ret = -EPERM;
2921 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
2922 goto unlock;
2923
2924 ret = open_related_ns(&net->ns, get_net_ns);
2925 goto unlock;
2926 }
2927
2928 ret = -EBADFD;
2929 if (!tun)
2930 goto unlock;
2931
2932 tun_debug(KERN_INFO, tun, "tun_chr_ioctl cmd %u\n", cmd);
2933
2934 ret = 0;
2935 switch (cmd) {
2936 case TUNGETIFF:
2937 tun_get_iff(current->nsproxy->net_ns, tun, &ifr);
2938
2939 if (tfile->detached)
2940 ifr.ifr_flags |= IFF_DETACH_QUEUE;
2941 if (!tfile->socket.sk->sk_filter)
2942 ifr.ifr_flags |= IFF_NOFILTER;
2943
2944 if (copy_to_user(argp, &ifr, ifreq_len))
2945 ret = -EFAULT;
2946 break;
2947
2948 case TUNSETNOCSUM:
2949 /* Disable/Enable checksum */
2950
2951 /* [unimplemented] */
2952 tun_debug(KERN_INFO, tun, "ignored: set checksum %s\n",
2953 arg ? "disabled" : "enabled");
2954 break;
2955
2956 case TUNSETPERSIST:
2957 /* Disable/Enable persist mode. Keep an extra reference to the
2958 * module to prevent the module being unprobed.
2959 */
2960 if (arg && !(tun->flags & IFF_PERSIST)) {
2961 tun->flags |= IFF_PERSIST;
2962 __module_get(THIS_MODULE);
2963 do_notify = true;
2964 }
2965 if (!arg && (tun->flags & IFF_PERSIST)) {
2966 tun->flags &= ~IFF_PERSIST;
2967 module_put(THIS_MODULE);
2968 do_notify = true;
2969 }
2970
2971 tun_debug(KERN_INFO, tun, "persist %s\n",
2972 arg ? "enabled" : "disabled");
2973 break;
2974
2975 case TUNSETOWNER:
2976 /* Set owner of the device */
2977 owner = make_kuid(current_user_ns(), arg);
2978 if (!uid_valid(owner)) {
2979 ret = -EINVAL;
2980 break;
2981 }
2982 tun->owner = owner;
2983 do_notify = true;
2984 tun_debug(KERN_INFO, tun, "owner set to %u\n",
2985 from_kuid(&init_user_ns, tun->owner));
2986 break;
2987
2988 case TUNSETGROUP:
2989 /* Set group of the device */
2990 group = make_kgid(current_user_ns(), arg);
2991 if (!gid_valid(group)) {
2992 ret = -EINVAL;
2993 break;
2994 }
2995 tun->group = group;
2996 do_notify = true;
2997 tun_debug(KERN_INFO, tun, "group set to %u\n",
2998 from_kgid(&init_user_ns, tun->group));
2999 break;
3000
3001 case TUNSETLINK:
3002 /* Only allow setting the type when the interface is down */
3003 if (tun->dev->flags & IFF_UP) {
3004 tun_debug(KERN_INFO, tun,
3005 "Linktype set failed because interface is up\n");
3006 ret = -EBUSY;
3007 } else {
3008 tun->dev->type = (int) arg;
3009 tun_debug(KERN_INFO, tun, "linktype set to %d\n",
3010 tun->dev->type);
3011 ret = 0;
3012 }
3013 break;
3014
3015#ifdef TUN_DEBUG
3016 case TUNSETDEBUG:
3017 tun->debug = arg;
3018 break;
3019#endif
3020 case TUNSETOFFLOAD:
3021 ret = set_offload(tun, arg);
3022 break;
3023
3024 case TUNSETTXFILTER:
3025 /* Can be set only for TAPs */
3026 ret = -EINVAL;
3027 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3028 break;
3029 ret = update_filter(&tun->txflt, (void __user *)arg);
3030 break;
3031
3032 case SIOCGIFHWADDR:
3033 /* Get hw address */
3034 memcpy(ifr.ifr_hwaddr.sa_data, tun->dev->dev_addr, ETH_ALEN);
3035 ifr.ifr_hwaddr.sa_family = tun->dev->type;
3036 if (copy_to_user(argp, &ifr, ifreq_len))
3037 ret = -EFAULT;
3038 break;
3039
3040 case SIOCSIFHWADDR:
3041 /* Set hw address */
3042 tun_debug(KERN_DEBUG, tun, "set hw address: %pM\n",
3043 ifr.ifr_hwaddr.sa_data);
3044
3045 ret = dev_set_mac_address(tun->dev, &ifr.ifr_hwaddr);
3046 break;
3047
3048 case TUNGETSNDBUF:
3049 sndbuf = tfile->socket.sk->sk_sndbuf;
3050 if (copy_to_user(argp, &sndbuf, sizeof(sndbuf)))
3051 ret = -EFAULT;
3052 break;
3053
3054 case TUNSETSNDBUF:
3055 if (copy_from_user(&sndbuf, argp, sizeof(sndbuf))) {
3056 ret = -EFAULT;
3057 break;
3058 }
3059 if (sndbuf <= 0) {
3060 ret = -EINVAL;
3061 break;
3062 }
3063
3064 tun->sndbuf = sndbuf;
3065 tun_set_sndbuf(tun);
3066 break;
3067
3068 case TUNGETVNETHDRSZ:
3069 vnet_hdr_sz = tun->vnet_hdr_sz;
3070 if (copy_to_user(argp, &vnet_hdr_sz, sizeof(vnet_hdr_sz)))
3071 ret = -EFAULT;
3072 break;
3073
3074 case TUNSETVNETHDRSZ:
3075 if (copy_from_user(&vnet_hdr_sz, argp, sizeof(vnet_hdr_sz))) {
3076 ret = -EFAULT;
3077 break;
3078 }
3079 if (vnet_hdr_sz < (int)sizeof(struct virtio_net_hdr)) {
3080 ret = -EINVAL;
3081 break;
3082 }
3083
3084 tun->vnet_hdr_sz = vnet_hdr_sz;
3085 break;
3086
3087 case TUNGETVNETLE:
3088 le = !!(tun->flags & TUN_VNET_LE);
3089 if (put_user(le, (int __user *)argp))
3090 ret = -EFAULT;
3091 break;
3092
3093 case TUNSETVNETLE:
3094 if (get_user(le, (int __user *)argp)) {
3095 ret = -EFAULT;
3096 break;
3097 }
3098 if (le)
3099 tun->flags |= TUN_VNET_LE;
3100 else
3101 tun->flags &= ~TUN_VNET_LE;
3102 break;
3103
3104 case TUNGETVNETBE:
3105 ret = tun_get_vnet_be(tun, argp);
3106 break;
3107
3108 case TUNSETVNETBE:
3109 ret = tun_set_vnet_be(tun, argp);
3110 break;
3111
3112 case TUNATTACHFILTER:
3113 /* Can be set only for TAPs */
3114 ret = -EINVAL;
3115 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3116 break;
3117 ret = -EFAULT;
3118 if (copy_from_user(&tun->fprog, argp, sizeof(tun->fprog)))
3119 break;
3120
3121 ret = tun_attach_filter(tun);
3122 break;
3123
3124 case TUNDETACHFILTER:
3125 /* Can be set only for TAPs */
3126 ret = -EINVAL;
3127 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3128 break;
3129 ret = 0;
3130 tun_detach_filter(tun, tun->numqueues);
3131 break;
3132
3133 case TUNGETFILTER:
3134 ret = -EINVAL;
3135 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3136 break;
3137 ret = -EFAULT;
3138 if (copy_to_user(argp, &tun->fprog, sizeof(tun->fprog)))
3139 break;
3140 ret = 0;
3141 break;
3142
3143 case TUNSETSTEERINGEBPF:
3144 ret = tun_set_ebpf(tun, &tun->steering_prog, argp);
3145 break;
3146
3147 case TUNSETFILTEREBPF:
3148 ret = tun_set_ebpf(tun, &tun->filter_prog, argp);
3149 break;
3150
3151 default:
3152 ret = -EINVAL;
3153 break;
3154 }
3155
3156 if (do_notify)
3157 netdev_state_change(tun->dev);
3158
3159unlock:
3160 rtnl_unlock();
3161 if (tun)
3162 tun_put(tun);
3163 return ret;
3164}
3165
3166static long tun_chr_ioctl(struct file *file,
3167 unsigned int cmd, unsigned long arg)
3168{
3169 return __tun_chr_ioctl(file, cmd, arg, sizeof (struct ifreq));
3170}
3171
3172#ifdef CONFIG_COMPAT
3173static long tun_chr_compat_ioctl(struct file *file,
3174 unsigned int cmd, unsigned long arg)
3175{
3176 switch (cmd) {
3177 case TUNSETIFF:
3178 case TUNGETIFF:
3179 case TUNSETTXFILTER:
3180 case TUNGETSNDBUF:
3181 case TUNSETSNDBUF:
3182 case SIOCGIFHWADDR:
3183 case SIOCSIFHWADDR:
3184 arg = (unsigned long)compat_ptr(arg);
3185 break;
3186 default:
3187 arg = (compat_ulong_t)arg;
3188 break;
3189 }
3190
3191 /*
3192 * compat_ifreq is shorter than ifreq, so we must not access beyond
3193 * the end of that structure. All fields that are used in this
3194 * driver are compatible though, we don't need to convert the
3195 * contents.
3196 */
3197 return __tun_chr_ioctl(file, cmd, arg, sizeof(struct compat_ifreq));
3198}
3199#endif /* CONFIG_COMPAT */
3200
3201static int tun_chr_fasync(int fd, struct file *file, int on)
3202{
3203 struct tun_file *tfile = file->private_data;
3204 int ret;
3205
3206 if ((ret = fasync_helper(fd, file, on, &tfile->fasync)) < 0)
3207 goto out;
3208
3209 if (on) {
3210 __f_setown(file, task_pid(current), PIDTYPE_PID, 0);
3211 tfile->flags |= TUN_FASYNC;
3212 } else
3213 tfile->flags &= ~TUN_FASYNC;
3214 ret = 0;
3215out:
3216 return ret;
3217}
3218
3219static int tun_chr_open(struct inode *inode, struct file * file)
3220{
3221 struct net *net = current->nsproxy->net_ns;
3222 struct tun_file *tfile;
3223
3224 DBG1(KERN_INFO, "tunX: tun_chr_open\n");
3225
3226 tfile = (struct tun_file *)sk_alloc(net, AF_UNSPEC, GFP_KERNEL,
3227 &tun_proto, 0);
3228 if (!tfile)
3229 return -ENOMEM;
3230 if (ptr_ring_init(&tfile->tx_ring, 0, GFP_KERNEL)) {
3231 sk_free(&tfile->sk);
3232 return -ENOMEM;
3233 }
3234
3235 RCU_INIT_POINTER(tfile->tun, NULL);
3236 tfile->flags = 0;
3237 tfile->ifindex = 0;
3238
3239 init_waitqueue_head(&tfile->wq.wait);
3240 RCU_INIT_POINTER(tfile->socket.wq, &tfile->wq);
3241
3242 tfile->socket.file = file;
3243 tfile->socket.ops = &tun_socket_ops;
3244
3245 sock_init_data(&tfile->socket, &tfile->sk);
3246
3247 tfile->sk.sk_write_space = tun_sock_write_space;
3248 tfile->sk.sk_sndbuf = INT_MAX;
3249
3250 file->private_data = tfile;
3251 INIT_LIST_HEAD(&tfile->next);
3252
3253 sock_set_flag(&tfile->sk, SOCK_ZEROCOPY);
3254
3255 return 0;
3256}
3257
3258static int tun_chr_close(struct inode *inode, struct file *file)
3259{
3260 struct tun_file *tfile = file->private_data;
3261
3262 tun_detach(tfile, true);
3263
3264 return 0;
3265}
3266
3267#ifdef CONFIG_PROC_FS
3268static void tun_chr_show_fdinfo(struct seq_file *m, struct file *file)
3269{
3270 struct tun_file *tfile = file->private_data;
3271 struct tun_struct *tun;
3272 struct ifreq ifr;
3273
3274 memset(&ifr, 0, sizeof(ifr));
3275
3276 rtnl_lock();
3277 tun = tun_get(tfile);
3278 if (tun)
3279 tun_get_iff(current->nsproxy->net_ns, tun, &ifr);
3280 rtnl_unlock();
3281
3282 if (tun)
3283 tun_put(tun);
3284
3285 seq_printf(m, "iff:\t%s\n", ifr.ifr_name);
3286}
3287#endif
3288
3289static const struct file_operations tun_fops = {
3290 .owner = THIS_MODULE,
3291 .llseek = no_llseek,
3292 .read_iter = tun_chr_read_iter,
3293 .write_iter = tun_chr_write_iter,
3294 .poll = tun_chr_poll,
3295 .unlocked_ioctl = tun_chr_ioctl,
3296#ifdef CONFIG_COMPAT
3297 .compat_ioctl = tun_chr_compat_ioctl,
3298#endif
3299 .open = tun_chr_open,
3300 .release = tun_chr_close,
3301 .fasync = tun_chr_fasync,
3302#ifdef CONFIG_PROC_FS
3303 .show_fdinfo = tun_chr_show_fdinfo,
3304#endif
3305};
3306
3307static struct miscdevice tun_miscdev = {
3308 .minor = TUN_MINOR,
3309 .name = "tun",
3310 .nodename = "net/tun",
3311 .fops = &tun_fops,
3312};
3313
3314/* ethtool interface */
3315
3316static int tun_get_link_ksettings(struct net_device *dev,
3317 struct ethtool_link_ksettings *cmd)
3318{
3319 ethtool_link_ksettings_zero_link_mode(cmd, supported);
3320 ethtool_link_ksettings_zero_link_mode(cmd, advertising);
3321 cmd->base.speed = SPEED_10;
3322 cmd->base.duplex = DUPLEX_FULL;
3323 cmd->base.port = PORT_TP;
3324 cmd->base.phy_address = 0;
3325 cmd->base.autoneg = AUTONEG_DISABLE;
3326 return 0;
3327}
3328
3329static void tun_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
3330{
3331 struct tun_struct *tun = netdev_priv(dev);
3332
3333 strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
3334 strlcpy(info->version, DRV_VERSION, sizeof(info->version));
3335
3336 switch (tun->flags & TUN_TYPE_MASK) {
3337 case IFF_TUN:
3338 strlcpy(info->bus_info, "tun", sizeof(info->bus_info));
3339 break;
3340 case IFF_TAP:
3341 strlcpy(info->bus_info, "tap", sizeof(info->bus_info));
3342 break;
3343 }
3344}
3345
3346static u32 tun_get_msglevel(struct net_device *dev)
3347{
3348#ifdef TUN_DEBUG
3349 struct tun_struct *tun = netdev_priv(dev);
3350 return tun->debug;
3351#else
3352 return -EOPNOTSUPP;
3353#endif
3354}
3355
3356static void tun_set_msglevel(struct net_device *dev, u32 value)
3357{
3358#ifdef TUN_DEBUG
3359 struct tun_struct *tun = netdev_priv(dev);
3360 tun->debug = value;
3361#endif
3362}
3363
3364static int tun_get_coalesce(struct net_device *dev,
3365 struct ethtool_coalesce *ec)
3366{
3367 struct tun_struct *tun = netdev_priv(dev);
3368
3369 ec->rx_max_coalesced_frames = tun->rx_batched;
3370
3371 return 0;
3372}
3373
3374static int tun_set_coalesce(struct net_device *dev,
3375 struct ethtool_coalesce *ec)
3376{
3377 struct tun_struct *tun = netdev_priv(dev);
3378
3379 if (ec->rx_max_coalesced_frames > NAPI_POLL_WEIGHT)
3380 tun->rx_batched = NAPI_POLL_WEIGHT;
3381 else
3382 tun->rx_batched = ec->rx_max_coalesced_frames;
3383
3384 return 0;
3385}
3386
3387static const struct ethtool_ops tun_ethtool_ops = {
3388 .get_drvinfo = tun_get_drvinfo,
3389 .get_msglevel = tun_get_msglevel,
3390 .set_msglevel = tun_set_msglevel,
3391 .get_link = ethtool_op_get_link,
3392 .get_ts_info = ethtool_op_get_ts_info,
3393 .get_coalesce = tun_get_coalesce,
3394 .set_coalesce = tun_set_coalesce,
3395 .get_link_ksettings = tun_get_link_ksettings,
3396};
3397
3398static int tun_queue_resize(struct tun_struct *tun)
3399{
3400 struct net_device *dev = tun->dev;
3401 struct tun_file *tfile;
3402 struct ptr_ring **rings;
3403 int n = tun->numqueues + tun->numdisabled;
3404 int ret, i;
3405
3406 rings = kmalloc_array(n, sizeof(*rings), GFP_KERNEL);
3407 if (!rings)
3408 return -ENOMEM;
3409
3410 for (i = 0; i < tun->numqueues; i++) {
3411 tfile = rtnl_dereference(tun->tfiles[i]);
3412 rings[i] = &tfile->tx_ring;
3413 }
3414 list_for_each_entry(tfile, &tun->disabled, next)
3415 rings[i++] = &tfile->tx_ring;
3416
3417 ret = ptr_ring_resize_multiple(rings, n,
3418 dev->tx_queue_len, GFP_KERNEL,
3419 tun_ptr_free);
3420
3421 kfree(rings);
3422 return ret;
3423}
3424
3425static int tun_device_event(struct notifier_block *unused,
3426 unsigned long event, void *ptr)
3427{
3428 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
3429 struct tun_struct *tun = netdev_priv(dev);
3430
3431 if (dev->rtnl_link_ops != &tun_link_ops)
3432 return NOTIFY_DONE;
3433
3434 switch (event) {
3435 case NETDEV_CHANGE_TX_QUEUE_LEN:
3436 if (tun_queue_resize(tun))
3437 return NOTIFY_BAD;
3438 break;
3439 default:
3440 break;
3441 }
3442
3443 return NOTIFY_DONE;
3444}
3445
3446static struct notifier_block tun_notifier_block __read_mostly = {
3447 .notifier_call = tun_device_event,
3448};
3449
3450static int __init tun_init(void)
3451{
3452 int ret = 0;
3453
3454 pr_info("%s, %s\n", DRV_DESCRIPTION, DRV_VERSION);
3455
3456 ret = rtnl_link_register(&tun_link_ops);
3457 if (ret) {
3458 pr_err("Can't register link_ops\n");
3459 goto err_linkops;
3460 }
3461
3462 ret = misc_register(&tun_miscdev);
3463 if (ret) {
3464 pr_err("Can't register misc device %d\n", TUN_MINOR);
3465 goto err_misc;
3466 }
3467
3468 ret = register_netdevice_notifier(&tun_notifier_block);
3469 if (ret) {
3470 pr_err("Can't register netdevice notifier\n");
3471 goto err_notifier;
3472 }
3473
3474 return 0;
3475
3476err_notifier:
3477 misc_deregister(&tun_miscdev);
3478err_misc:
3479 rtnl_link_unregister(&tun_link_ops);
3480err_linkops:
3481 return ret;
3482}
3483
3484static void tun_cleanup(void)
3485{
3486 misc_deregister(&tun_miscdev);
3487 rtnl_link_unregister(&tun_link_ops);
3488 unregister_netdevice_notifier(&tun_notifier_block);
3489}
3490
3491/* Get an underlying socket object from tun file. Returns error unless file is
3492 * attached to a device. The returned object works like a packet socket, it
3493 * can be used for sock_sendmsg/sock_recvmsg. The caller is responsible for
3494 * holding a reference to the file for as long as the socket is in use. */
3495struct socket *tun_get_socket(struct file *file)
3496{
3497 struct tun_file *tfile;
3498 if (file->f_op != &tun_fops)
3499 return ERR_PTR(-EINVAL);
3500 tfile = file->private_data;
3501 if (!tfile)
3502 return ERR_PTR(-EBADFD);
3503 return &tfile->socket;
3504}
3505EXPORT_SYMBOL_GPL(tun_get_socket);
3506
3507struct ptr_ring *tun_get_tx_ring(struct file *file)
3508{
3509 struct tun_file *tfile;
3510
3511 if (file->f_op != &tun_fops)
3512 return ERR_PTR(-EINVAL);
3513 tfile = file->private_data;
3514 if (!tfile)
3515 return ERR_PTR(-EBADFD);
3516 return &tfile->tx_ring;
3517}
3518EXPORT_SYMBOL_GPL(tun_get_tx_ring);
3519
3520module_init(tun_init);
3521module_exit(tun_cleanup);
3522MODULE_DESCRIPTION(DRV_DESCRIPTION);
3523MODULE_AUTHOR(DRV_COPYRIGHT);
3524MODULE_LICENSE("GPL");
3525MODULE_ALIAS_MISCDEV(TUN_MINOR);
3526MODULE_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/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");