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/nsproxy.h>
58#include <linux/virtio_net.h>
59#include <linux/rcupdate.h>
60#include <net/net_namespace.h>
61#include <net/netns/generic.h>
62#include <net/rtnetlink.h>
63#include <net/sock.h>
64#include <net/xdp.h>
65#include <net/ip_tunnels.h>
66#include <linux/seq_file.h>
67#include <linux/uio.h>
68#include <linux/skb_array.h>
69#include <linux/bpf.h>
70#include <linux/bpf_trace.h>
71#include <linux/mutex.h>
72
73#include <linux/uaccess.h>
74#include <linux/proc_fs.h>
75
76static void tun_default_link_ksettings(struct net_device *dev,
77 struct ethtool_link_ksettings *cmd);
78
79#define TUN_RX_PAD (NET_IP_ALIGN + NET_SKB_PAD)
80
81/* TUN device flags */
82
83/* IFF_ATTACH_QUEUE is never stored in device flags,
84 * overload it to mean fasync when stored there.
85 */
86#define TUN_FASYNC IFF_ATTACH_QUEUE
87/* High bits in flags field are unused. */
88#define TUN_VNET_LE 0x80000000
89#define TUN_VNET_BE 0x40000000
90
91#define TUN_FEATURES (IFF_NO_PI | IFF_ONE_QUEUE | IFF_VNET_HDR | \
92 IFF_MULTI_QUEUE | IFF_NAPI | IFF_NAPI_FRAGS)
93
94#define GOODCOPY_LEN 128
95
96#define FLT_EXACT_COUNT 8
97struct tap_filter {
98 unsigned int count; /* Number of addrs. Zero means disabled */
99 u32 mask[2]; /* Mask of the hashed addrs */
100 unsigned char addr[FLT_EXACT_COUNT][ETH_ALEN];
101};
102
103/* MAX_TAP_QUEUES 256 is chosen to allow rx/tx queues to be equal
104 * to max number of VCPUs in guest. */
105#define MAX_TAP_QUEUES 256
106#define MAX_TAP_FLOWS 4096
107
108#define TUN_FLOW_EXPIRE (3 * HZ)
109
110struct tun_pcpu_stats {
111 u64_stats_t rx_packets;
112 u64_stats_t rx_bytes;
113 u64_stats_t tx_packets;
114 u64_stats_t tx_bytes;
115 struct u64_stats_sync syncp;
116 u32 rx_dropped;
117 u32 tx_dropped;
118 u32 rx_frame_errors;
119};
120
121/* A tun_file connects an open character device to a tuntap netdevice. It
122 * also contains all socket related structures (except sock_fprog and tap_filter)
123 * to serve as one transmit queue for tuntap device. The sock_fprog and
124 * tap_filter were kept in tun_struct since they were used for filtering for the
125 * netdevice not for a specific queue (at least I didn't see the requirement for
126 * this).
127 *
128 * RCU usage:
129 * The tun_file and tun_struct are loosely coupled, the pointer from one to the
130 * other can only be read while rcu_read_lock or rtnl_lock is held.
131 */
132struct tun_file {
133 struct sock sk;
134 struct socket socket;
135 struct tun_struct __rcu *tun;
136 struct fasync_struct *fasync;
137 /* only used for fasnyc */
138 unsigned int flags;
139 union {
140 u16 queue_index;
141 unsigned int ifindex;
142 };
143 struct napi_struct napi;
144 bool napi_enabled;
145 bool napi_frags_enabled;
146 struct mutex napi_mutex; /* Protects access to the above napi */
147 struct list_head next;
148 struct tun_struct *detached;
149 struct ptr_ring tx_ring;
150 struct xdp_rxq_info xdp_rxq;
151};
152
153struct tun_page {
154 struct page *page;
155 int count;
156};
157
158struct tun_flow_entry {
159 struct hlist_node hash_link;
160 struct rcu_head rcu;
161 struct tun_struct *tun;
162
163 u32 rxhash;
164 u32 rps_rxhash;
165 int queue_index;
166 unsigned long updated ____cacheline_aligned_in_smp;
167};
168
169#define TUN_NUM_FLOW_ENTRIES 1024
170#define TUN_MASK_FLOW_ENTRIES (TUN_NUM_FLOW_ENTRIES - 1)
171
172struct tun_prog {
173 struct rcu_head rcu;
174 struct bpf_prog *prog;
175};
176
177/* Since the socket were moved to tun_file, to preserve the behavior of persist
178 * device, socket filter, sndbuf and vnet header size were restore when the
179 * file were attached to a persist device.
180 */
181struct tun_struct {
182 struct tun_file __rcu *tfiles[MAX_TAP_QUEUES];
183 unsigned int numqueues;
184 unsigned int flags;
185 kuid_t owner;
186 kgid_t group;
187
188 struct net_device *dev;
189 netdev_features_t set_features;
190#define TUN_USER_FEATURES (NETIF_F_HW_CSUM|NETIF_F_TSO_ECN|NETIF_F_TSO| \
191 NETIF_F_TSO6)
192
193 int align;
194 int vnet_hdr_sz;
195 int sndbuf;
196 struct tap_filter txflt;
197 struct sock_fprog fprog;
198 /* protected by rtnl lock */
199 bool filter_attached;
200 u32 msg_enable;
201 spinlock_t lock;
202 struct hlist_head flows[TUN_NUM_FLOW_ENTRIES];
203 struct timer_list flow_gc_timer;
204 unsigned long ageing_time;
205 unsigned int numdisabled;
206 struct list_head disabled;
207 void *security;
208 u32 flow_count;
209 u32 rx_batched;
210 struct tun_pcpu_stats __percpu *pcpu_stats;
211 struct bpf_prog __rcu *xdp_prog;
212 struct tun_prog __rcu *steering_prog;
213 struct tun_prog __rcu *filter_prog;
214 struct ethtool_link_ksettings link_ksettings;
215};
216
217struct veth {
218 __be16 h_vlan_proto;
219 __be16 h_vlan_TCI;
220};
221
222bool tun_is_xdp_frame(void *ptr)
223{
224 return (unsigned long)ptr & TUN_XDP_FLAG;
225}
226EXPORT_SYMBOL(tun_is_xdp_frame);
227
228void *tun_xdp_to_ptr(void *ptr)
229{
230 return (void *)((unsigned long)ptr | TUN_XDP_FLAG);
231}
232EXPORT_SYMBOL(tun_xdp_to_ptr);
233
234void *tun_ptr_to_xdp(void *ptr)
235{
236 return (void *)((unsigned long)ptr & ~TUN_XDP_FLAG);
237}
238EXPORT_SYMBOL(tun_ptr_to_xdp);
239
240static int tun_napi_receive(struct napi_struct *napi, int budget)
241{
242 struct tun_file *tfile = container_of(napi, struct tun_file, napi);
243 struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
244 struct sk_buff_head process_queue;
245 struct sk_buff *skb;
246 int received = 0;
247
248 __skb_queue_head_init(&process_queue);
249
250 spin_lock(&queue->lock);
251 skb_queue_splice_tail_init(queue, &process_queue);
252 spin_unlock(&queue->lock);
253
254 while (received < budget && (skb = __skb_dequeue(&process_queue))) {
255 napi_gro_receive(napi, skb);
256 ++received;
257 }
258
259 if (!skb_queue_empty(&process_queue)) {
260 spin_lock(&queue->lock);
261 skb_queue_splice(&process_queue, queue);
262 spin_unlock(&queue->lock);
263 }
264
265 return received;
266}
267
268static int tun_napi_poll(struct napi_struct *napi, int budget)
269{
270 unsigned int received;
271
272 received = tun_napi_receive(napi, budget);
273
274 if (received < budget)
275 napi_complete_done(napi, received);
276
277 return received;
278}
279
280static void tun_napi_init(struct tun_struct *tun, struct tun_file *tfile,
281 bool napi_en, bool napi_frags)
282{
283 tfile->napi_enabled = napi_en;
284 tfile->napi_frags_enabled = napi_en && napi_frags;
285 if (napi_en) {
286 netif_tx_napi_add(tun->dev, &tfile->napi, tun_napi_poll,
287 NAPI_POLL_WEIGHT);
288 napi_enable(&tfile->napi);
289 }
290}
291
292static void tun_napi_disable(struct tun_file *tfile)
293{
294 if (tfile->napi_enabled)
295 napi_disable(&tfile->napi);
296}
297
298static void tun_napi_del(struct tun_file *tfile)
299{
300 if (tfile->napi_enabled)
301 netif_napi_del(&tfile->napi);
302}
303
304static bool tun_napi_frags_enabled(const struct tun_file *tfile)
305{
306 return tfile->napi_frags_enabled;
307}
308
309#ifdef CONFIG_TUN_VNET_CROSS_LE
310static inline bool tun_legacy_is_little_endian(struct tun_struct *tun)
311{
312 return tun->flags & TUN_VNET_BE ? false :
313 virtio_legacy_is_little_endian();
314}
315
316static long tun_get_vnet_be(struct tun_struct *tun, int __user *argp)
317{
318 int be = !!(tun->flags & TUN_VNET_BE);
319
320 if (put_user(be, argp))
321 return -EFAULT;
322
323 return 0;
324}
325
326static long tun_set_vnet_be(struct tun_struct *tun, int __user *argp)
327{
328 int be;
329
330 if (get_user(be, argp))
331 return -EFAULT;
332
333 if (be)
334 tun->flags |= TUN_VNET_BE;
335 else
336 tun->flags &= ~TUN_VNET_BE;
337
338 return 0;
339}
340#else
341static inline bool tun_legacy_is_little_endian(struct tun_struct *tun)
342{
343 return virtio_legacy_is_little_endian();
344}
345
346static long tun_get_vnet_be(struct tun_struct *tun, int __user *argp)
347{
348 return -EINVAL;
349}
350
351static long tun_set_vnet_be(struct tun_struct *tun, int __user *argp)
352{
353 return -EINVAL;
354}
355#endif /* CONFIG_TUN_VNET_CROSS_LE */
356
357static inline bool tun_is_little_endian(struct tun_struct *tun)
358{
359 return tun->flags & TUN_VNET_LE ||
360 tun_legacy_is_little_endian(tun);
361}
362
363static inline u16 tun16_to_cpu(struct tun_struct *tun, __virtio16 val)
364{
365 return __virtio16_to_cpu(tun_is_little_endian(tun), val);
366}
367
368static inline __virtio16 cpu_to_tun16(struct tun_struct *tun, u16 val)
369{
370 return __cpu_to_virtio16(tun_is_little_endian(tun), val);
371}
372
373static inline u32 tun_hashfn(u32 rxhash)
374{
375 return rxhash & TUN_MASK_FLOW_ENTRIES;
376}
377
378static struct tun_flow_entry *tun_flow_find(struct hlist_head *head, u32 rxhash)
379{
380 struct tun_flow_entry *e;
381
382 hlist_for_each_entry_rcu(e, head, hash_link) {
383 if (e->rxhash == rxhash)
384 return e;
385 }
386 return NULL;
387}
388
389static struct tun_flow_entry *tun_flow_create(struct tun_struct *tun,
390 struct hlist_head *head,
391 u32 rxhash, u16 queue_index)
392{
393 struct tun_flow_entry *e = kmalloc(sizeof(*e), GFP_ATOMIC);
394
395 if (e) {
396 netif_info(tun, tx_queued, tun->dev,
397 "create flow: hash %u index %u\n",
398 rxhash, queue_index);
399 e->updated = jiffies;
400 e->rxhash = rxhash;
401 e->rps_rxhash = 0;
402 e->queue_index = queue_index;
403 e->tun = tun;
404 hlist_add_head_rcu(&e->hash_link, head);
405 ++tun->flow_count;
406 }
407 return e;
408}
409
410static void tun_flow_delete(struct tun_struct *tun, struct tun_flow_entry *e)
411{
412 netif_info(tun, tx_queued, tun->dev, "delete flow: hash %u index %u\n",
413 e->rxhash, e->queue_index);
414 hlist_del_rcu(&e->hash_link);
415 kfree_rcu(e, rcu);
416 --tun->flow_count;
417}
418
419static void tun_flow_flush(struct tun_struct *tun)
420{
421 int i;
422
423 spin_lock_bh(&tun->lock);
424 for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
425 struct tun_flow_entry *e;
426 struct hlist_node *n;
427
428 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link)
429 tun_flow_delete(tun, e);
430 }
431 spin_unlock_bh(&tun->lock);
432}
433
434static void tun_flow_delete_by_queue(struct tun_struct *tun, u16 queue_index)
435{
436 int i;
437
438 spin_lock_bh(&tun->lock);
439 for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
440 struct tun_flow_entry *e;
441 struct hlist_node *n;
442
443 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
444 if (e->queue_index == queue_index)
445 tun_flow_delete(tun, e);
446 }
447 }
448 spin_unlock_bh(&tun->lock);
449}
450
451static void tun_flow_cleanup(struct timer_list *t)
452{
453 struct tun_struct *tun = from_timer(tun, t, flow_gc_timer);
454 unsigned long delay = tun->ageing_time;
455 unsigned long next_timer = jiffies + delay;
456 unsigned long count = 0;
457 int i;
458
459 spin_lock(&tun->lock);
460 for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
461 struct tun_flow_entry *e;
462 struct hlist_node *n;
463
464 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
465 unsigned long this_timer;
466
467 this_timer = e->updated + delay;
468 if (time_before_eq(this_timer, jiffies)) {
469 tun_flow_delete(tun, e);
470 continue;
471 }
472 count++;
473 if (time_before(this_timer, next_timer))
474 next_timer = this_timer;
475 }
476 }
477
478 if (count)
479 mod_timer(&tun->flow_gc_timer, round_jiffies_up(next_timer));
480 spin_unlock(&tun->lock);
481}
482
483static void tun_flow_update(struct tun_struct *tun, u32 rxhash,
484 struct tun_file *tfile)
485{
486 struct hlist_head *head;
487 struct tun_flow_entry *e;
488 unsigned long delay = tun->ageing_time;
489 u16 queue_index = tfile->queue_index;
490
491 head = &tun->flows[tun_hashfn(rxhash)];
492
493 rcu_read_lock();
494
495 e = tun_flow_find(head, rxhash);
496 if (likely(e)) {
497 /* TODO: keep queueing to old queue until it's empty? */
498 if (READ_ONCE(e->queue_index) != queue_index)
499 WRITE_ONCE(e->queue_index, queue_index);
500 if (e->updated != jiffies)
501 e->updated = jiffies;
502 sock_rps_record_flow_hash(e->rps_rxhash);
503 } else {
504 spin_lock_bh(&tun->lock);
505 if (!tun_flow_find(head, rxhash) &&
506 tun->flow_count < MAX_TAP_FLOWS)
507 tun_flow_create(tun, head, rxhash, queue_index);
508
509 if (!timer_pending(&tun->flow_gc_timer))
510 mod_timer(&tun->flow_gc_timer,
511 round_jiffies_up(jiffies + delay));
512 spin_unlock_bh(&tun->lock);
513 }
514
515 rcu_read_unlock();
516}
517
518/* Save the hash received in the stack receive path and update the
519 * flow_hash table accordingly.
520 */
521static inline void tun_flow_save_rps_rxhash(struct tun_flow_entry *e, u32 hash)
522{
523 if (unlikely(e->rps_rxhash != hash))
524 e->rps_rxhash = hash;
525}
526
527/* We try to identify a flow through its rxhash. The reason that
528 * we do not check rxq no. is because some cards(e.g 82599), chooses
529 * the rxq based on the txq where the last packet of the flow comes. As
530 * the userspace application move between processors, we may get a
531 * different rxq no. here.
532 */
533static u16 tun_automq_select_queue(struct tun_struct *tun, struct sk_buff *skb)
534{
535 struct tun_flow_entry *e;
536 u32 txq = 0;
537 u32 numqueues = 0;
538
539 numqueues = READ_ONCE(tun->numqueues);
540
541 txq = __skb_get_hash_symmetric(skb);
542 e = tun_flow_find(&tun->flows[tun_hashfn(txq)], txq);
543 if (e) {
544 tun_flow_save_rps_rxhash(e, txq);
545 txq = e->queue_index;
546 } else {
547 /* use multiply and shift instead of expensive divide */
548 txq = ((u64)txq * numqueues) >> 32;
549 }
550
551 return txq;
552}
553
554static u16 tun_ebpf_select_queue(struct tun_struct *tun, struct sk_buff *skb)
555{
556 struct tun_prog *prog;
557 u32 numqueues;
558 u16 ret = 0;
559
560 numqueues = READ_ONCE(tun->numqueues);
561 if (!numqueues)
562 return 0;
563
564 prog = rcu_dereference(tun->steering_prog);
565 if (prog)
566 ret = bpf_prog_run_clear_cb(prog->prog, skb);
567
568 return ret % numqueues;
569}
570
571static u16 tun_select_queue(struct net_device *dev, struct sk_buff *skb,
572 struct net_device *sb_dev)
573{
574 struct tun_struct *tun = netdev_priv(dev);
575 u16 ret;
576
577 rcu_read_lock();
578 if (rcu_dereference(tun->steering_prog))
579 ret = tun_ebpf_select_queue(tun, skb);
580 else
581 ret = tun_automq_select_queue(tun, skb);
582 rcu_read_unlock();
583
584 return ret;
585}
586
587static inline bool tun_not_capable(struct tun_struct *tun)
588{
589 const struct cred *cred = current_cred();
590 struct net *net = dev_net(tun->dev);
591
592 return ((uid_valid(tun->owner) && !uid_eq(cred->euid, tun->owner)) ||
593 (gid_valid(tun->group) && !in_egroup_p(tun->group))) &&
594 !ns_capable(net->user_ns, CAP_NET_ADMIN);
595}
596
597static void tun_set_real_num_queues(struct tun_struct *tun)
598{
599 netif_set_real_num_tx_queues(tun->dev, tun->numqueues);
600 netif_set_real_num_rx_queues(tun->dev, tun->numqueues);
601}
602
603static void tun_disable_queue(struct tun_struct *tun, struct tun_file *tfile)
604{
605 tfile->detached = tun;
606 list_add_tail(&tfile->next, &tun->disabled);
607 ++tun->numdisabled;
608}
609
610static struct tun_struct *tun_enable_queue(struct tun_file *tfile)
611{
612 struct tun_struct *tun = tfile->detached;
613
614 tfile->detached = NULL;
615 list_del_init(&tfile->next);
616 --tun->numdisabled;
617 return tun;
618}
619
620void tun_ptr_free(void *ptr)
621{
622 if (!ptr)
623 return;
624 if (tun_is_xdp_frame(ptr)) {
625 struct xdp_frame *xdpf = tun_ptr_to_xdp(ptr);
626
627 xdp_return_frame(xdpf);
628 } else {
629 __skb_array_destroy_skb(ptr);
630 }
631}
632EXPORT_SYMBOL_GPL(tun_ptr_free);
633
634static void tun_queue_purge(struct tun_file *tfile)
635{
636 void *ptr;
637
638 while ((ptr = ptr_ring_consume(&tfile->tx_ring)) != NULL)
639 tun_ptr_free(ptr);
640
641 skb_queue_purge(&tfile->sk.sk_write_queue);
642 skb_queue_purge(&tfile->sk.sk_error_queue);
643}
644
645static void __tun_detach(struct tun_file *tfile, bool clean)
646{
647 struct tun_file *ntfile;
648 struct tun_struct *tun;
649
650 tun = rtnl_dereference(tfile->tun);
651
652 if (tun && clean) {
653 tun_napi_disable(tfile);
654 tun_napi_del(tfile);
655 }
656
657 if (tun && !tfile->detached) {
658 u16 index = tfile->queue_index;
659 BUG_ON(index >= tun->numqueues);
660
661 rcu_assign_pointer(tun->tfiles[index],
662 tun->tfiles[tun->numqueues - 1]);
663 ntfile = rtnl_dereference(tun->tfiles[index]);
664 ntfile->queue_index = index;
665 rcu_assign_pointer(tun->tfiles[tun->numqueues - 1],
666 NULL);
667
668 --tun->numqueues;
669 if (clean) {
670 RCU_INIT_POINTER(tfile->tun, NULL);
671 sock_put(&tfile->sk);
672 } else
673 tun_disable_queue(tun, tfile);
674
675 synchronize_net();
676 tun_flow_delete_by_queue(tun, tun->numqueues + 1);
677 /* Drop read queue */
678 tun_queue_purge(tfile);
679 tun_set_real_num_queues(tun);
680 } else if (tfile->detached && clean) {
681 tun = tun_enable_queue(tfile);
682 sock_put(&tfile->sk);
683 }
684
685 if (clean) {
686 if (tun && tun->numqueues == 0 && tun->numdisabled == 0) {
687 netif_carrier_off(tun->dev);
688
689 if (!(tun->flags & IFF_PERSIST) &&
690 tun->dev->reg_state == NETREG_REGISTERED)
691 unregister_netdevice(tun->dev);
692 }
693 if (tun)
694 xdp_rxq_info_unreg(&tfile->xdp_rxq);
695 ptr_ring_cleanup(&tfile->tx_ring, tun_ptr_free);
696 sock_put(&tfile->sk);
697 }
698}
699
700static void tun_detach(struct tun_file *tfile, bool clean)
701{
702 struct tun_struct *tun;
703 struct net_device *dev;
704
705 rtnl_lock();
706 tun = rtnl_dereference(tfile->tun);
707 dev = tun ? tun->dev : NULL;
708 __tun_detach(tfile, clean);
709 if (dev)
710 netdev_state_change(dev);
711 rtnl_unlock();
712}
713
714static void tun_detach_all(struct net_device *dev)
715{
716 struct tun_struct *tun = netdev_priv(dev);
717 struct tun_file *tfile, *tmp;
718 int i, n = tun->numqueues;
719
720 for (i = 0; i < n; i++) {
721 tfile = rtnl_dereference(tun->tfiles[i]);
722 BUG_ON(!tfile);
723 tun_napi_disable(tfile);
724 tfile->socket.sk->sk_shutdown = RCV_SHUTDOWN;
725 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
726 RCU_INIT_POINTER(tfile->tun, NULL);
727 --tun->numqueues;
728 }
729 list_for_each_entry(tfile, &tun->disabled, next) {
730 tfile->socket.sk->sk_shutdown = RCV_SHUTDOWN;
731 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
732 RCU_INIT_POINTER(tfile->tun, NULL);
733 }
734 BUG_ON(tun->numqueues != 0);
735
736 synchronize_net();
737 for (i = 0; i < n; i++) {
738 tfile = rtnl_dereference(tun->tfiles[i]);
739 tun_napi_del(tfile);
740 /* Drop read queue */
741 tun_queue_purge(tfile);
742 xdp_rxq_info_unreg(&tfile->xdp_rxq);
743 sock_put(&tfile->sk);
744 }
745 list_for_each_entry_safe(tfile, tmp, &tun->disabled, next) {
746 tun_enable_queue(tfile);
747 tun_queue_purge(tfile);
748 xdp_rxq_info_unreg(&tfile->xdp_rxq);
749 sock_put(&tfile->sk);
750 }
751 BUG_ON(tun->numdisabled != 0);
752
753 if (tun->flags & IFF_PERSIST)
754 module_put(THIS_MODULE);
755}
756
757static int tun_attach(struct tun_struct *tun, struct file *file,
758 bool skip_filter, bool napi, bool napi_frags,
759 bool publish_tun)
760{
761 struct tun_file *tfile = file->private_data;
762 struct net_device *dev = tun->dev;
763 int err;
764
765 err = security_tun_dev_attach(tfile->socket.sk, tun->security);
766 if (err < 0)
767 goto out;
768
769 err = -EINVAL;
770 if (rtnl_dereference(tfile->tun) && !tfile->detached)
771 goto out;
772
773 err = -EBUSY;
774 if (!(tun->flags & IFF_MULTI_QUEUE) && tun->numqueues == 1)
775 goto out;
776
777 err = -E2BIG;
778 if (!tfile->detached &&
779 tun->numqueues + tun->numdisabled == MAX_TAP_QUEUES)
780 goto out;
781
782 err = 0;
783
784 /* Re-attach the filter to persist device */
785 if (!skip_filter && (tun->filter_attached == true)) {
786 lock_sock(tfile->socket.sk);
787 err = sk_attach_filter(&tun->fprog, tfile->socket.sk);
788 release_sock(tfile->socket.sk);
789 if (!err)
790 goto out;
791 }
792
793 if (!tfile->detached &&
794 ptr_ring_resize(&tfile->tx_ring, dev->tx_queue_len,
795 GFP_KERNEL, tun_ptr_free)) {
796 err = -ENOMEM;
797 goto out;
798 }
799
800 tfile->queue_index = tun->numqueues;
801 tfile->socket.sk->sk_shutdown &= ~RCV_SHUTDOWN;
802
803 if (tfile->detached) {
804 /* Re-attach detached tfile, updating XDP queue_index */
805 WARN_ON(!xdp_rxq_info_is_reg(&tfile->xdp_rxq));
806
807 if (tfile->xdp_rxq.queue_index != tfile->queue_index)
808 tfile->xdp_rxq.queue_index = tfile->queue_index;
809 } else {
810 /* Setup XDP RX-queue info, for new tfile getting attached */
811 err = xdp_rxq_info_reg(&tfile->xdp_rxq,
812 tun->dev, tfile->queue_index);
813 if (err < 0)
814 goto out;
815 err = xdp_rxq_info_reg_mem_model(&tfile->xdp_rxq,
816 MEM_TYPE_PAGE_SHARED, NULL);
817 if (err < 0) {
818 xdp_rxq_info_unreg(&tfile->xdp_rxq);
819 goto out;
820 }
821 err = 0;
822 }
823
824 if (tfile->detached) {
825 tun_enable_queue(tfile);
826 } else {
827 sock_hold(&tfile->sk);
828 tun_napi_init(tun, tfile, napi, napi_frags);
829 }
830
831 if (rtnl_dereference(tun->xdp_prog))
832 sock_set_flag(&tfile->sk, SOCK_XDP);
833
834 /* device is allowed to go away first, so no need to hold extra
835 * refcnt.
836 */
837
838 /* Publish tfile->tun and tun->tfiles only after we've fully
839 * initialized tfile; otherwise we risk using half-initialized
840 * object.
841 */
842 if (publish_tun)
843 rcu_assign_pointer(tfile->tun, tun);
844 rcu_assign_pointer(tun->tfiles[tun->numqueues], tfile);
845 tun->numqueues++;
846 tun_set_real_num_queues(tun);
847out:
848 return err;
849}
850
851static struct tun_struct *tun_get(struct tun_file *tfile)
852{
853 struct tun_struct *tun;
854
855 rcu_read_lock();
856 tun = rcu_dereference(tfile->tun);
857 if (tun)
858 dev_hold(tun->dev);
859 rcu_read_unlock();
860
861 return tun;
862}
863
864static void tun_put(struct tun_struct *tun)
865{
866 dev_put(tun->dev);
867}
868
869/* TAP filtering */
870static void addr_hash_set(u32 *mask, const u8 *addr)
871{
872 int n = ether_crc(ETH_ALEN, addr) >> 26;
873 mask[n >> 5] |= (1 << (n & 31));
874}
875
876static unsigned int addr_hash_test(const u32 *mask, const u8 *addr)
877{
878 int n = ether_crc(ETH_ALEN, addr) >> 26;
879 return mask[n >> 5] & (1 << (n & 31));
880}
881
882static int update_filter(struct tap_filter *filter, void __user *arg)
883{
884 struct { u8 u[ETH_ALEN]; } *addr;
885 struct tun_filter uf;
886 int err, alen, n, nexact;
887
888 if (copy_from_user(&uf, arg, sizeof(uf)))
889 return -EFAULT;
890
891 if (!uf.count) {
892 /* Disabled */
893 filter->count = 0;
894 return 0;
895 }
896
897 alen = ETH_ALEN * uf.count;
898 addr = memdup_user(arg + sizeof(uf), alen);
899 if (IS_ERR(addr))
900 return PTR_ERR(addr);
901
902 /* The filter is updated without holding any locks. Which is
903 * perfectly safe. We disable it first and in the worst
904 * case we'll accept a few undesired packets. */
905 filter->count = 0;
906 wmb();
907
908 /* Use first set of addresses as an exact filter */
909 for (n = 0; n < uf.count && n < FLT_EXACT_COUNT; n++)
910 memcpy(filter->addr[n], addr[n].u, ETH_ALEN);
911
912 nexact = n;
913
914 /* Remaining multicast addresses are hashed,
915 * unicast will leave the filter disabled. */
916 memset(filter->mask, 0, sizeof(filter->mask));
917 for (; n < uf.count; n++) {
918 if (!is_multicast_ether_addr(addr[n].u)) {
919 err = 0; /* no filter */
920 goto free_addr;
921 }
922 addr_hash_set(filter->mask, addr[n].u);
923 }
924
925 /* For ALLMULTI just set the mask to all ones.
926 * This overrides the mask populated above. */
927 if ((uf.flags & TUN_FLT_ALLMULTI))
928 memset(filter->mask, ~0, sizeof(filter->mask));
929
930 /* Now enable the filter */
931 wmb();
932 filter->count = nexact;
933
934 /* Return the number of exact filters */
935 err = nexact;
936free_addr:
937 kfree(addr);
938 return err;
939}
940
941/* Returns: 0 - drop, !=0 - accept */
942static int run_filter(struct tap_filter *filter, const struct sk_buff *skb)
943{
944 /* Cannot use eth_hdr(skb) here because skb_mac_hdr() is incorrect
945 * at this point. */
946 struct ethhdr *eh = (struct ethhdr *) skb->data;
947 int i;
948
949 /* Exact match */
950 for (i = 0; i < filter->count; i++)
951 if (ether_addr_equal(eh->h_dest, filter->addr[i]))
952 return 1;
953
954 /* Inexact match (multicast only) */
955 if (is_multicast_ether_addr(eh->h_dest))
956 return addr_hash_test(filter->mask, eh->h_dest);
957
958 return 0;
959}
960
961/*
962 * Checks whether the packet is accepted or not.
963 * Returns: 0 - drop, !=0 - accept
964 */
965static int check_filter(struct tap_filter *filter, const struct sk_buff *skb)
966{
967 if (!filter->count)
968 return 1;
969
970 return run_filter(filter, skb);
971}
972
973/* Network device part of the driver */
974
975static const struct ethtool_ops tun_ethtool_ops;
976
977/* Net device detach from fd. */
978static void tun_net_uninit(struct net_device *dev)
979{
980 tun_detach_all(dev);
981}
982
983/* Net device open. */
984static int tun_net_open(struct net_device *dev)
985{
986 netif_tx_start_all_queues(dev);
987
988 return 0;
989}
990
991/* Net device close. */
992static int tun_net_close(struct net_device *dev)
993{
994 netif_tx_stop_all_queues(dev);
995 return 0;
996}
997
998/* Net device start xmit */
999static void tun_automq_xmit(struct tun_struct *tun, struct sk_buff *skb)
1000{
1001#ifdef CONFIG_RPS
1002 if (tun->numqueues == 1 && static_branch_unlikely(&rps_needed)) {
1003 /* Select queue was not called for the skbuff, so we extract the
1004 * RPS hash and save it into the flow_table here.
1005 */
1006 struct tun_flow_entry *e;
1007 __u32 rxhash;
1008
1009 rxhash = __skb_get_hash_symmetric(skb);
1010 e = tun_flow_find(&tun->flows[tun_hashfn(rxhash)], rxhash);
1011 if (e)
1012 tun_flow_save_rps_rxhash(e, rxhash);
1013 }
1014#endif
1015}
1016
1017static unsigned int run_ebpf_filter(struct tun_struct *tun,
1018 struct sk_buff *skb,
1019 int len)
1020{
1021 struct tun_prog *prog = rcu_dereference(tun->filter_prog);
1022
1023 if (prog)
1024 len = bpf_prog_run_clear_cb(prog->prog, skb);
1025
1026 return len;
1027}
1028
1029/* Net device start xmit */
1030static netdev_tx_t tun_net_xmit(struct sk_buff *skb, struct net_device *dev)
1031{
1032 struct tun_struct *tun = netdev_priv(dev);
1033 int txq = skb->queue_mapping;
1034 struct tun_file *tfile;
1035 int len = skb->len;
1036
1037 rcu_read_lock();
1038 tfile = rcu_dereference(tun->tfiles[txq]);
1039
1040 /* Drop packet if interface is not attached */
1041 if (!tfile)
1042 goto drop;
1043
1044 if (!rcu_dereference(tun->steering_prog))
1045 tun_automq_xmit(tun, skb);
1046
1047 netif_info(tun, tx_queued, tun->dev, "%s %d\n", __func__, skb->len);
1048
1049 /* Drop if the filter does not like it.
1050 * This is a noop if the filter is disabled.
1051 * Filter can be enabled only for the TAP devices. */
1052 if (!check_filter(&tun->txflt, skb))
1053 goto drop;
1054
1055 if (tfile->socket.sk->sk_filter &&
1056 sk_filter(tfile->socket.sk, skb))
1057 goto drop;
1058
1059 len = run_ebpf_filter(tun, skb, len);
1060 if (len == 0 || pskb_trim(skb, len))
1061 goto drop;
1062
1063 if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC)))
1064 goto drop;
1065
1066 skb_tx_timestamp(skb);
1067
1068 /* Orphan the skb - required as we might hang on to it
1069 * for indefinite time.
1070 */
1071 skb_orphan(skb);
1072
1073 nf_reset_ct(skb);
1074
1075 if (ptr_ring_produce(&tfile->tx_ring, skb))
1076 goto drop;
1077
1078 /* Notify and wake up reader process */
1079 if (tfile->flags & TUN_FASYNC)
1080 kill_fasync(&tfile->fasync, SIGIO, POLL_IN);
1081 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
1082
1083 rcu_read_unlock();
1084 return NETDEV_TX_OK;
1085
1086drop:
1087 this_cpu_inc(tun->pcpu_stats->tx_dropped);
1088 skb_tx_error(skb);
1089 kfree_skb(skb);
1090 rcu_read_unlock();
1091 return NET_XMIT_DROP;
1092}
1093
1094static void tun_net_mclist(struct net_device *dev)
1095{
1096 /*
1097 * This callback is supposed to deal with mc filter in
1098 * _rx_ path and has nothing to do with the _tx_ path.
1099 * In rx path we always accept everything userspace gives us.
1100 */
1101}
1102
1103static netdev_features_t tun_net_fix_features(struct net_device *dev,
1104 netdev_features_t features)
1105{
1106 struct tun_struct *tun = netdev_priv(dev);
1107
1108 return (features & tun->set_features) | (features & ~TUN_USER_FEATURES);
1109}
1110
1111static void tun_set_headroom(struct net_device *dev, int new_hr)
1112{
1113 struct tun_struct *tun = netdev_priv(dev);
1114
1115 if (new_hr < NET_SKB_PAD)
1116 new_hr = NET_SKB_PAD;
1117
1118 tun->align = new_hr;
1119}
1120
1121static void
1122tun_net_get_stats64(struct net_device *dev, struct rtnl_link_stats64 *stats)
1123{
1124 u32 rx_dropped = 0, tx_dropped = 0, rx_frame_errors = 0;
1125 struct tun_struct *tun = netdev_priv(dev);
1126 struct tun_pcpu_stats *p;
1127 int i;
1128
1129 for_each_possible_cpu(i) {
1130 u64 rxpackets, rxbytes, txpackets, txbytes;
1131 unsigned int start;
1132
1133 p = per_cpu_ptr(tun->pcpu_stats, i);
1134 do {
1135 start = u64_stats_fetch_begin(&p->syncp);
1136 rxpackets = u64_stats_read(&p->rx_packets);
1137 rxbytes = u64_stats_read(&p->rx_bytes);
1138 txpackets = u64_stats_read(&p->tx_packets);
1139 txbytes = u64_stats_read(&p->tx_bytes);
1140 } while (u64_stats_fetch_retry(&p->syncp, start));
1141
1142 stats->rx_packets += rxpackets;
1143 stats->rx_bytes += rxbytes;
1144 stats->tx_packets += txpackets;
1145 stats->tx_bytes += txbytes;
1146
1147 /* u32 counters */
1148 rx_dropped += p->rx_dropped;
1149 rx_frame_errors += p->rx_frame_errors;
1150 tx_dropped += p->tx_dropped;
1151 }
1152 stats->rx_dropped = rx_dropped;
1153 stats->rx_frame_errors = rx_frame_errors;
1154 stats->tx_dropped = tx_dropped;
1155}
1156
1157static int tun_xdp_set(struct net_device *dev, struct bpf_prog *prog,
1158 struct netlink_ext_ack *extack)
1159{
1160 struct tun_struct *tun = netdev_priv(dev);
1161 struct tun_file *tfile;
1162 struct bpf_prog *old_prog;
1163 int i;
1164
1165 old_prog = rtnl_dereference(tun->xdp_prog);
1166 rcu_assign_pointer(tun->xdp_prog, prog);
1167 if (old_prog)
1168 bpf_prog_put(old_prog);
1169
1170 for (i = 0; i < tun->numqueues; i++) {
1171 tfile = rtnl_dereference(tun->tfiles[i]);
1172 if (prog)
1173 sock_set_flag(&tfile->sk, SOCK_XDP);
1174 else
1175 sock_reset_flag(&tfile->sk, SOCK_XDP);
1176 }
1177 list_for_each_entry(tfile, &tun->disabled, next) {
1178 if (prog)
1179 sock_set_flag(&tfile->sk, SOCK_XDP);
1180 else
1181 sock_reset_flag(&tfile->sk, SOCK_XDP);
1182 }
1183
1184 return 0;
1185}
1186
1187static int tun_xdp(struct net_device *dev, struct netdev_bpf *xdp)
1188{
1189 switch (xdp->command) {
1190 case XDP_SETUP_PROG:
1191 return tun_xdp_set(dev, xdp->prog, xdp->extack);
1192 default:
1193 return -EINVAL;
1194 }
1195}
1196
1197static int tun_net_change_carrier(struct net_device *dev, bool new_carrier)
1198{
1199 if (new_carrier) {
1200 struct tun_struct *tun = netdev_priv(dev);
1201
1202 if (!tun->numqueues)
1203 return -EPERM;
1204
1205 netif_carrier_on(dev);
1206 } else {
1207 netif_carrier_off(dev);
1208 }
1209 return 0;
1210}
1211
1212static const struct net_device_ops tun_netdev_ops = {
1213 .ndo_uninit = tun_net_uninit,
1214 .ndo_open = tun_net_open,
1215 .ndo_stop = tun_net_close,
1216 .ndo_start_xmit = tun_net_xmit,
1217 .ndo_fix_features = tun_net_fix_features,
1218 .ndo_select_queue = tun_select_queue,
1219 .ndo_set_rx_headroom = tun_set_headroom,
1220 .ndo_get_stats64 = tun_net_get_stats64,
1221 .ndo_change_carrier = tun_net_change_carrier,
1222};
1223
1224static void __tun_xdp_flush_tfile(struct tun_file *tfile)
1225{
1226 /* Notify and wake up reader process */
1227 if (tfile->flags & TUN_FASYNC)
1228 kill_fasync(&tfile->fasync, SIGIO, POLL_IN);
1229 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
1230}
1231
1232static int tun_xdp_xmit(struct net_device *dev, int n,
1233 struct xdp_frame **frames, u32 flags)
1234{
1235 struct tun_struct *tun = netdev_priv(dev);
1236 struct tun_file *tfile;
1237 u32 numqueues;
1238 int drops = 0;
1239 int cnt = n;
1240 int i;
1241
1242 if (unlikely(flags & ~XDP_XMIT_FLAGS_MASK))
1243 return -EINVAL;
1244
1245 rcu_read_lock();
1246
1247resample:
1248 numqueues = READ_ONCE(tun->numqueues);
1249 if (!numqueues) {
1250 rcu_read_unlock();
1251 return -ENXIO; /* Caller will free/return all frames */
1252 }
1253
1254 tfile = rcu_dereference(tun->tfiles[smp_processor_id() %
1255 numqueues]);
1256 if (unlikely(!tfile))
1257 goto resample;
1258
1259 spin_lock(&tfile->tx_ring.producer_lock);
1260 for (i = 0; i < n; i++) {
1261 struct xdp_frame *xdp = frames[i];
1262 /* Encode the XDP flag into lowest bit for consumer to differ
1263 * XDP buffer from sk_buff.
1264 */
1265 void *frame = tun_xdp_to_ptr(xdp);
1266
1267 if (__ptr_ring_produce(&tfile->tx_ring, frame)) {
1268 this_cpu_inc(tun->pcpu_stats->tx_dropped);
1269 xdp_return_frame_rx_napi(xdp);
1270 drops++;
1271 }
1272 }
1273 spin_unlock(&tfile->tx_ring.producer_lock);
1274
1275 if (flags & XDP_XMIT_FLUSH)
1276 __tun_xdp_flush_tfile(tfile);
1277
1278 rcu_read_unlock();
1279 return cnt - drops;
1280}
1281
1282static int tun_xdp_tx(struct net_device *dev, struct xdp_buff *xdp)
1283{
1284 struct xdp_frame *frame = xdp_convert_buff_to_frame(xdp);
1285
1286 if (unlikely(!frame))
1287 return -EOVERFLOW;
1288
1289 return tun_xdp_xmit(dev, 1, &frame, XDP_XMIT_FLUSH);
1290}
1291
1292static const struct net_device_ops tap_netdev_ops = {
1293 .ndo_uninit = tun_net_uninit,
1294 .ndo_open = tun_net_open,
1295 .ndo_stop = tun_net_close,
1296 .ndo_start_xmit = tun_net_xmit,
1297 .ndo_fix_features = tun_net_fix_features,
1298 .ndo_set_rx_mode = tun_net_mclist,
1299 .ndo_set_mac_address = eth_mac_addr,
1300 .ndo_validate_addr = eth_validate_addr,
1301 .ndo_select_queue = tun_select_queue,
1302 .ndo_features_check = passthru_features_check,
1303 .ndo_set_rx_headroom = tun_set_headroom,
1304 .ndo_get_stats64 = tun_net_get_stats64,
1305 .ndo_bpf = tun_xdp,
1306 .ndo_xdp_xmit = tun_xdp_xmit,
1307 .ndo_change_carrier = tun_net_change_carrier,
1308};
1309
1310static void tun_flow_init(struct tun_struct *tun)
1311{
1312 int i;
1313
1314 for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++)
1315 INIT_HLIST_HEAD(&tun->flows[i]);
1316
1317 tun->ageing_time = TUN_FLOW_EXPIRE;
1318 timer_setup(&tun->flow_gc_timer, tun_flow_cleanup, 0);
1319 mod_timer(&tun->flow_gc_timer,
1320 round_jiffies_up(jiffies + tun->ageing_time));
1321}
1322
1323static void tun_flow_uninit(struct tun_struct *tun)
1324{
1325 del_timer_sync(&tun->flow_gc_timer);
1326 tun_flow_flush(tun);
1327}
1328
1329#define MIN_MTU 68
1330#define MAX_MTU 65535
1331
1332/* Initialize net device. */
1333static void tun_net_init(struct net_device *dev)
1334{
1335 struct tun_struct *tun = netdev_priv(dev);
1336
1337 switch (tun->flags & TUN_TYPE_MASK) {
1338 case IFF_TUN:
1339 dev->netdev_ops = &tun_netdev_ops;
1340 dev->header_ops = &ip_tunnel_header_ops;
1341
1342 /* Point-to-Point TUN Device */
1343 dev->hard_header_len = 0;
1344 dev->addr_len = 0;
1345 dev->mtu = 1500;
1346
1347 /* Zero header length */
1348 dev->type = ARPHRD_NONE;
1349 dev->flags = IFF_POINTOPOINT | IFF_NOARP | IFF_MULTICAST;
1350 break;
1351
1352 case IFF_TAP:
1353 dev->netdev_ops = &tap_netdev_ops;
1354 /* Ethernet TAP Device */
1355 ether_setup(dev);
1356 dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1357 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
1358
1359 eth_hw_addr_random(dev);
1360
1361 break;
1362 }
1363
1364 dev->min_mtu = MIN_MTU;
1365 dev->max_mtu = MAX_MTU - dev->hard_header_len;
1366}
1367
1368static bool tun_sock_writeable(struct tun_struct *tun, struct tun_file *tfile)
1369{
1370 struct sock *sk = tfile->socket.sk;
1371
1372 return (tun->dev->flags & IFF_UP) && sock_writeable(sk);
1373}
1374
1375/* Character device part */
1376
1377/* Poll */
1378static __poll_t tun_chr_poll(struct file *file, poll_table *wait)
1379{
1380 struct tun_file *tfile = file->private_data;
1381 struct tun_struct *tun = tun_get(tfile);
1382 struct sock *sk;
1383 __poll_t mask = 0;
1384
1385 if (!tun)
1386 return EPOLLERR;
1387
1388 sk = tfile->socket.sk;
1389
1390 poll_wait(file, sk_sleep(sk), wait);
1391
1392 if (!ptr_ring_empty(&tfile->tx_ring))
1393 mask |= EPOLLIN | EPOLLRDNORM;
1394
1395 /* Make sure SOCKWQ_ASYNC_NOSPACE is set if not writable to
1396 * guarantee EPOLLOUT to be raised by either here or
1397 * tun_sock_write_space(). Then process could get notification
1398 * after it writes to a down device and meets -EIO.
1399 */
1400 if (tun_sock_writeable(tun, tfile) ||
1401 (!test_and_set_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags) &&
1402 tun_sock_writeable(tun, tfile)))
1403 mask |= EPOLLOUT | EPOLLWRNORM;
1404
1405 if (tun->dev->reg_state != NETREG_REGISTERED)
1406 mask = EPOLLERR;
1407
1408 tun_put(tun);
1409 return mask;
1410}
1411
1412static struct sk_buff *tun_napi_alloc_frags(struct tun_file *tfile,
1413 size_t len,
1414 const struct iov_iter *it)
1415{
1416 struct sk_buff *skb;
1417 size_t linear;
1418 int err;
1419 int i;
1420
1421 if (it->nr_segs > MAX_SKB_FRAGS + 1)
1422 return ERR_PTR(-ENOMEM);
1423
1424 local_bh_disable();
1425 skb = napi_get_frags(&tfile->napi);
1426 local_bh_enable();
1427 if (!skb)
1428 return ERR_PTR(-ENOMEM);
1429
1430 linear = iov_iter_single_seg_count(it);
1431 err = __skb_grow(skb, linear);
1432 if (err)
1433 goto free;
1434
1435 skb->len = len;
1436 skb->data_len = len - linear;
1437 skb->truesize += skb->data_len;
1438
1439 for (i = 1; i < it->nr_segs; i++) {
1440 size_t fragsz = it->iov[i].iov_len;
1441 struct page *page;
1442 void *frag;
1443
1444 if (fragsz == 0 || fragsz > PAGE_SIZE) {
1445 err = -EINVAL;
1446 goto free;
1447 }
1448 frag = netdev_alloc_frag(fragsz);
1449 if (!frag) {
1450 err = -ENOMEM;
1451 goto free;
1452 }
1453 page = virt_to_head_page(frag);
1454 skb_fill_page_desc(skb, i - 1, page,
1455 frag - page_address(page), fragsz);
1456 }
1457
1458 return skb;
1459free:
1460 /* frees skb and all frags allocated with napi_alloc_frag() */
1461 napi_free_frags(&tfile->napi);
1462 return ERR_PTR(err);
1463}
1464
1465/* prepad is the amount to reserve at front. len is length after that.
1466 * linear is a hint as to how much to copy (usually headers). */
1467static struct sk_buff *tun_alloc_skb(struct tun_file *tfile,
1468 size_t prepad, size_t len,
1469 size_t linear, int noblock)
1470{
1471 struct sock *sk = tfile->socket.sk;
1472 struct sk_buff *skb;
1473 int err;
1474
1475 /* Under a page? Don't bother with paged skb. */
1476 if (prepad + len < PAGE_SIZE || !linear)
1477 linear = len;
1478
1479 skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
1480 &err, 0);
1481 if (!skb)
1482 return ERR_PTR(err);
1483
1484 skb_reserve(skb, prepad);
1485 skb_put(skb, linear);
1486 skb->data_len = len - linear;
1487 skb->len += len - linear;
1488
1489 return skb;
1490}
1491
1492static void tun_rx_batched(struct tun_struct *tun, struct tun_file *tfile,
1493 struct sk_buff *skb, int more)
1494{
1495 struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
1496 struct sk_buff_head process_queue;
1497 u32 rx_batched = tun->rx_batched;
1498 bool rcv = false;
1499
1500 if (!rx_batched || (!more && skb_queue_empty(queue))) {
1501 local_bh_disable();
1502 skb_record_rx_queue(skb, tfile->queue_index);
1503 netif_receive_skb(skb);
1504 local_bh_enable();
1505 return;
1506 }
1507
1508 spin_lock(&queue->lock);
1509 if (!more || skb_queue_len(queue) == rx_batched) {
1510 __skb_queue_head_init(&process_queue);
1511 skb_queue_splice_tail_init(queue, &process_queue);
1512 rcv = true;
1513 } else {
1514 __skb_queue_tail(queue, skb);
1515 }
1516 spin_unlock(&queue->lock);
1517
1518 if (rcv) {
1519 struct sk_buff *nskb;
1520
1521 local_bh_disable();
1522 while ((nskb = __skb_dequeue(&process_queue))) {
1523 skb_record_rx_queue(nskb, tfile->queue_index);
1524 netif_receive_skb(nskb);
1525 }
1526 skb_record_rx_queue(skb, tfile->queue_index);
1527 netif_receive_skb(skb);
1528 local_bh_enable();
1529 }
1530}
1531
1532static bool tun_can_build_skb(struct tun_struct *tun, struct tun_file *tfile,
1533 int len, int noblock, bool zerocopy)
1534{
1535 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
1536 return false;
1537
1538 if (tfile->socket.sk->sk_sndbuf != INT_MAX)
1539 return false;
1540
1541 if (!noblock)
1542 return false;
1543
1544 if (zerocopy)
1545 return false;
1546
1547 if (SKB_DATA_ALIGN(len + TUN_RX_PAD) +
1548 SKB_DATA_ALIGN(sizeof(struct skb_shared_info)) > PAGE_SIZE)
1549 return false;
1550
1551 return true;
1552}
1553
1554static struct sk_buff *__tun_build_skb(struct tun_file *tfile,
1555 struct page_frag *alloc_frag, char *buf,
1556 int buflen, int len, int pad)
1557{
1558 struct sk_buff *skb = build_skb(buf, buflen);
1559
1560 if (!skb)
1561 return ERR_PTR(-ENOMEM);
1562
1563 skb_reserve(skb, pad);
1564 skb_put(skb, len);
1565 skb_set_owner_w(skb, tfile->socket.sk);
1566
1567 get_page(alloc_frag->page);
1568 alloc_frag->offset += buflen;
1569
1570 return skb;
1571}
1572
1573static int tun_xdp_act(struct tun_struct *tun, struct bpf_prog *xdp_prog,
1574 struct xdp_buff *xdp, u32 act)
1575{
1576 int err;
1577
1578 switch (act) {
1579 case XDP_REDIRECT:
1580 err = xdp_do_redirect(tun->dev, xdp, xdp_prog);
1581 if (err)
1582 return err;
1583 break;
1584 case XDP_TX:
1585 err = tun_xdp_tx(tun->dev, xdp);
1586 if (err < 0)
1587 return err;
1588 break;
1589 case XDP_PASS:
1590 break;
1591 default:
1592 bpf_warn_invalid_xdp_action(act);
1593 fallthrough;
1594 case XDP_ABORTED:
1595 trace_xdp_exception(tun->dev, xdp_prog, act);
1596 fallthrough;
1597 case XDP_DROP:
1598 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1599 break;
1600 }
1601
1602 return act;
1603}
1604
1605static struct sk_buff *tun_build_skb(struct tun_struct *tun,
1606 struct tun_file *tfile,
1607 struct iov_iter *from,
1608 struct virtio_net_hdr *hdr,
1609 int len, int *skb_xdp)
1610{
1611 struct page_frag *alloc_frag = ¤t->task_frag;
1612 struct bpf_prog *xdp_prog;
1613 int buflen = SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
1614 char *buf;
1615 size_t copied;
1616 int pad = TUN_RX_PAD;
1617 int err = 0;
1618
1619 rcu_read_lock();
1620 xdp_prog = rcu_dereference(tun->xdp_prog);
1621 if (xdp_prog)
1622 pad += XDP_PACKET_HEADROOM;
1623 buflen += SKB_DATA_ALIGN(len + pad);
1624 rcu_read_unlock();
1625
1626 alloc_frag->offset = ALIGN((u64)alloc_frag->offset, SMP_CACHE_BYTES);
1627 if (unlikely(!skb_page_frag_refill(buflen, alloc_frag, GFP_KERNEL)))
1628 return ERR_PTR(-ENOMEM);
1629
1630 buf = (char *)page_address(alloc_frag->page) + alloc_frag->offset;
1631 copied = copy_page_from_iter(alloc_frag->page,
1632 alloc_frag->offset + pad,
1633 len, from);
1634 if (copied != len)
1635 return ERR_PTR(-EFAULT);
1636
1637 /* There's a small window that XDP may be set after the check
1638 * of xdp_prog above, this should be rare and for simplicity
1639 * we do XDP on skb in case the headroom is not enough.
1640 */
1641 if (hdr->gso_type || !xdp_prog) {
1642 *skb_xdp = 1;
1643 return __tun_build_skb(tfile, alloc_frag, buf, buflen, len,
1644 pad);
1645 }
1646
1647 *skb_xdp = 0;
1648
1649 local_bh_disable();
1650 rcu_read_lock();
1651 xdp_prog = rcu_dereference(tun->xdp_prog);
1652 if (xdp_prog) {
1653 struct xdp_buff xdp;
1654 u32 act;
1655
1656 xdp.data_hard_start = buf;
1657 xdp.data = buf + pad;
1658 xdp_set_data_meta_invalid(&xdp);
1659 xdp.data_end = xdp.data + len;
1660 xdp.rxq = &tfile->xdp_rxq;
1661 xdp.frame_sz = buflen;
1662
1663 act = bpf_prog_run_xdp(xdp_prog, &xdp);
1664 if (act == XDP_REDIRECT || act == XDP_TX) {
1665 get_page(alloc_frag->page);
1666 alloc_frag->offset += buflen;
1667 }
1668 err = tun_xdp_act(tun, xdp_prog, &xdp, act);
1669 if (err < 0) {
1670 if (act == XDP_REDIRECT || act == XDP_TX)
1671 put_page(alloc_frag->page);
1672 goto out;
1673 }
1674
1675 if (err == XDP_REDIRECT)
1676 xdp_do_flush();
1677 if (err != XDP_PASS)
1678 goto out;
1679
1680 pad = xdp.data - xdp.data_hard_start;
1681 len = xdp.data_end - xdp.data;
1682 }
1683 rcu_read_unlock();
1684 local_bh_enable();
1685
1686 return __tun_build_skb(tfile, alloc_frag, buf, buflen, len, pad);
1687
1688out:
1689 rcu_read_unlock();
1690 local_bh_enable();
1691 return NULL;
1692}
1693
1694/* Get packet from user space buffer */
1695static ssize_t tun_get_user(struct tun_struct *tun, struct tun_file *tfile,
1696 void *msg_control, struct iov_iter *from,
1697 int noblock, bool more)
1698{
1699 struct tun_pi pi = { 0, cpu_to_be16(ETH_P_IP) };
1700 struct sk_buff *skb;
1701 size_t total_len = iov_iter_count(from);
1702 size_t len = total_len, align = tun->align, linear;
1703 struct virtio_net_hdr gso = { 0 };
1704 struct tun_pcpu_stats *stats;
1705 int good_linear;
1706 int copylen;
1707 bool zerocopy = false;
1708 int err;
1709 u32 rxhash = 0;
1710 int skb_xdp = 1;
1711 bool frags = tun_napi_frags_enabled(tfile);
1712
1713 if (!(tun->flags & IFF_NO_PI)) {
1714 if (len < sizeof(pi))
1715 return -EINVAL;
1716 len -= sizeof(pi);
1717
1718 if (!copy_from_iter_full(&pi, sizeof(pi), from))
1719 return -EFAULT;
1720 }
1721
1722 if (tun->flags & IFF_VNET_HDR) {
1723 int vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
1724
1725 if (len < vnet_hdr_sz)
1726 return -EINVAL;
1727 len -= vnet_hdr_sz;
1728
1729 if (!copy_from_iter_full(&gso, sizeof(gso), from))
1730 return -EFAULT;
1731
1732 if ((gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
1733 tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2 > tun16_to_cpu(tun, gso.hdr_len))
1734 gso.hdr_len = cpu_to_tun16(tun, tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2);
1735
1736 if (tun16_to_cpu(tun, gso.hdr_len) > len)
1737 return -EINVAL;
1738 iov_iter_advance(from, vnet_hdr_sz - sizeof(gso));
1739 }
1740
1741 if ((tun->flags & TUN_TYPE_MASK) == IFF_TAP) {
1742 align += NET_IP_ALIGN;
1743 if (unlikely(len < ETH_HLEN ||
1744 (gso.hdr_len && tun16_to_cpu(tun, gso.hdr_len) < ETH_HLEN)))
1745 return -EINVAL;
1746 }
1747
1748 good_linear = SKB_MAX_HEAD(align);
1749
1750 if (msg_control) {
1751 struct iov_iter i = *from;
1752
1753 /* There are 256 bytes to be copied in skb, so there is
1754 * enough room for skb expand head in case it is used.
1755 * The rest of the buffer is mapped from userspace.
1756 */
1757 copylen = gso.hdr_len ? tun16_to_cpu(tun, gso.hdr_len) : GOODCOPY_LEN;
1758 if (copylen > good_linear)
1759 copylen = good_linear;
1760 linear = copylen;
1761 iov_iter_advance(&i, copylen);
1762 if (iov_iter_npages(&i, INT_MAX) <= MAX_SKB_FRAGS)
1763 zerocopy = true;
1764 }
1765
1766 if (!frags && tun_can_build_skb(tun, tfile, len, noblock, zerocopy)) {
1767 /* For the packet that is not easy to be processed
1768 * (e.g gso or jumbo packet), we will do it at after
1769 * skb was created with generic XDP routine.
1770 */
1771 skb = tun_build_skb(tun, tfile, from, &gso, len, &skb_xdp);
1772 if (IS_ERR(skb)) {
1773 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1774 return PTR_ERR(skb);
1775 }
1776 if (!skb)
1777 return total_len;
1778 } else {
1779 if (!zerocopy) {
1780 copylen = len;
1781 if (tun16_to_cpu(tun, gso.hdr_len) > good_linear)
1782 linear = good_linear;
1783 else
1784 linear = tun16_to_cpu(tun, gso.hdr_len);
1785 }
1786
1787 if (frags) {
1788 mutex_lock(&tfile->napi_mutex);
1789 skb = tun_napi_alloc_frags(tfile, copylen, from);
1790 /* tun_napi_alloc_frags() enforces a layout for the skb.
1791 * If zerocopy is enabled, then this layout will be
1792 * overwritten by zerocopy_sg_from_iter().
1793 */
1794 zerocopy = false;
1795 } else {
1796 skb = tun_alloc_skb(tfile, align, copylen, linear,
1797 noblock);
1798 }
1799
1800 if (IS_ERR(skb)) {
1801 if (PTR_ERR(skb) != -EAGAIN)
1802 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1803 if (frags)
1804 mutex_unlock(&tfile->napi_mutex);
1805 return PTR_ERR(skb);
1806 }
1807
1808 if (zerocopy)
1809 err = zerocopy_sg_from_iter(skb, from);
1810 else
1811 err = skb_copy_datagram_from_iter(skb, 0, from, len);
1812
1813 if (err) {
1814 err = -EFAULT;
1815drop:
1816 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1817 kfree_skb(skb);
1818 if (frags) {
1819 tfile->napi.skb = NULL;
1820 mutex_unlock(&tfile->napi_mutex);
1821 }
1822
1823 return err;
1824 }
1825 }
1826
1827 if (virtio_net_hdr_to_skb(skb, &gso, tun_is_little_endian(tun))) {
1828 this_cpu_inc(tun->pcpu_stats->rx_frame_errors);
1829 kfree_skb(skb);
1830 if (frags) {
1831 tfile->napi.skb = NULL;
1832 mutex_unlock(&tfile->napi_mutex);
1833 }
1834
1835 return -EINVAL;
1836 }
1837
1838 switch (tun->flags & TUN_TYPE_MASK) {
1839 case IFF_TUN:
1840 if (tun->flags & IFF_NO_PI) {
1841 u8 ip_version = skb->len ? (skb->data[0] >> 4) : 0;
1842
1843 switch (ip_version) {
1844 case 4:
1845 pi.proto = htons(ETH_P_IP);
1846 break;
1847 case 6:
1848 pi.proto = htons(ETH_P_IPV6);
1849 break;
1850 default:
1851 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1852 kfree_skb(skb);
1853 return -EINVAL;
1854 }
1855 }
1856
1857 skb_reset_mac_header(skb);
1858 skb->protocol = pi.proto;
1859 skb->dev = tun->dev;
1860 break;
1861 case IFF_TAP:
1862 if (frags && !pskb_may_pull(skb, ETH_HLEN)) {
1863 err = -ENOMEM;
1864 goto drop;
1865 }
1866 skb->protocol = eth_type_trans(skb, tun->dev);
1867 break;
1868 }
1869
1870 /* copy skb_ubuf_info for callback when skb has no error */
1871 if (zerocopy) {
1872 skb_shinfo(skb)->destructor_arg = msg_control;
1873 skb_shinfo(skb)->tx_flags |= SKBTX_DEV_ZEROCOPY;
1874 skb_shinfo(skb)->tx_flags |= SKBTX_SHARED_FRAG;
1875 } else if (msg_control) {
1876 struct ubuf_info *uarg = msg_control;
1877 uarg->callback(uarg, false);
1878 }
1879
1880 skb_reset_network_header(skb);
1881 skb_probe_transport_header(skb);
1882 skb_record_rx_queue(skb, tfile->queue_index);
1883
1884 if (skb_xdp) {
1885 struct bpf_prog *xdp_prog;
1886 int ret;
1887
1888 local_bh_disable();
1889 rcu_read_lock();
1890 xdp_prog = rcu_dereference(tun->xdp_prog);
1891 if (xdp_prog) {
1892 ret = do_xdp_generic(xdp_prog, skb);
1893 if (ret != XDP_PASS) {
1894 rcu_read_unlock();
1895 local_bh_enable();
1896 if (frags) {
1897 tfile->napi.skb = NULL;
1898 mutex_unlock(&tfile->napi_mutex);
1899 }
1900 return total_len;
1901 }
1902 }
1903 rcu_read_unlock();
1904 local_bh_enable();
1905 }
1906
1907 /* Compute the costly rx hash only if needed for flow updates.
1908 * We may get a very small possibility of OOO during switching, not
1909 * worth to optimize.
1910 */
1911 if (!rcu_access_pointer(tun->steering_prog) && tun->numqueues > 1 &&
1912 !tfile->detached)
1913 rxhash = __skb_get_hash_symmetric(skb);
1914
1915 rcu_read_lock();
1916 if (unlikely(!(tun->dev->flags & IFF_UP))) {
1917 err = -EIO;
1918 rcu_read_unlock();
1919 goto drop;
1920 }
1921
1922 if (frags) {
1923 u32 headlen;
1924
1925 /* Exercise flow dissector code path. */
1926 skb_push(skb, ETH_HLEN);
1927 headlen = eth_get_headlen(tun->dev, skb->data,
1928 skb_headlen(skb));
1929
1930 if (unlikely(headlen > skb_headlen(skb))) {
1931 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1932 napi_free_frags(&tfile->napi);
1933 rcu_read_unlock();
1934 mutex_unlock(&tfile->napi_mutex);
1935 WARN_ON(1);
1936 return -ENOMEM;
1937 }
1938
1939 local_bh_disable();
1940 napi_gro_frags(&tfile->napi);
1941 local_bh_enable();
1942 mutex_unlock(&tfile->napi_mutex);
1943 } else if (tfile->napi_enabled) {
1944 struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
1945 int queue_len;
1946
1947 spin_lock_bh(&queue->lock);
1948 __skb_queue_tail(queue, skb);
1949 queue_len = skb_queue_len(queue);
1950 spin_unlock(&queue->lock);
1951
1952 if (!more || queue_len > NAPI_POLL_WEIGHT)
1953 napi_schedule(&tfile->napi);
1954
1955 local_bh_enable();
1956 } else if (!IS_ENABLED(CONFIG_4KSTACKS)) {
1957 tun_rx_batched(tun, tfile, skb, more);
1958 } else {
1959 netif_rx_ni(skb);
1960 }
1961 rcu_read_unlock();
1962
1963 stats = get_cpu_ptr(tun->pcpu_stats);
1964 u64_stats_update_begin(&stats->syncp);
1965 u64_stats_inc(&stats->rx_packets);
1966 u64_stats_add(&stats->rx_bytes, len);
1967 u64_stats_update_end(&stats->syncp);
1968 put_cpu_ptr(stats);
1969
1970 if (rxhash)
1971 tun_flow_update(tun, rxhash, tfile);
1972
1973 return total_len;
1974}
1975
1976static ssize_t tun_chr_write_iter(struct kiocb *iocb, struct iov_iter *from)
1977{
1978 struct file *file = iocb->ki_filp;
1979 struct tun_file *tfile = file->private_data;
1980 struct tun_struct *tun = tun_get(tfile);
1981 ssize_t result;
1982
1983 if (!tun)
1984 return -EBADFD;
1985
1986 result = tun_get_user(tun, tfile, NULL, from,
1987 file->f_flags & O_NONBLOCK, false);
1988
1989 tun_put(tun);
1990 return result;
1991}
1992
1993static ssize_t tun_put_user_xdp(struct tun_struct *tun,
1994 struct tun_file *tfile,
1995 struct xdp_frame *xdp_frame,
1996 struct iov_iter *iter)
1997{
1998 int vnet_hdr_sz = 0;
1999 size_t size = xdp_frame->len;
2000 struct tun_pcpu_stats *stats;
2001 size_t ret;
2002
2003 if (tun->flags & IFF_VNET_HDR) {
2004 struct virtio_net_hdr gso = { 0 };
2005
2006 vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
2007 if (unlikely(iov_iter_count(iter) < vnet_hdr_sz))
2008 return -EINVAL;
2009 if (unlikely(copy_to_iter(&gso, sizeof(gso), iter) !=
2010 sizeof(gso)))
2011 return -EFAULT;
2012 iov_iter_advance(iter, vnet_hdr_sz - sizeof(gso));
2013 }
2014
2015 ret = copy_to_iter(xdp_frame->data, size, iter) + vnet_hdr_sz;
2016
2017 stats = get_cpu_ptr(tun->pcpu_stats);
2018 u64_stats_update_begin(&stats->syncp);
2019 u64_stats_inc(&stats->tx_packets);
2020 u64_stats_add(&stats->tx_bytes, ret);
2021 u64_stats_update_end(&stats->syncp);
2022 put_cpu_ptr(tun->pcpu_stats);
2023
2024 return ret;
2025}
2026
2027/* Put packet to the user space buffer */
2028static ssize_t tun_put_user(struct tun_struct *tun,
2029 struct tun_file *tfile,
2030 struct sk_buff *skb,
2031 struct iov_iter *iter)
2032{
2033 struct tun_pi pi = { 0, skb->protocol };
2034 struct tun_pcpu_stats *stats;
2035 ssize_t total;
2036 int vlan_offset = 0;
2037 int vlan_hlen = 0;
2038 int vnet_hdr_sz = 0;
2039
2040 if (skb_vlan_tag_present(skb))
2041 vlan_hlen = VLAN_HLEN;
2042
2043 if (tun->flags & IFF_VNET_HDR)
2044 vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
2045
2046 total = skb->len + vlan_hlen + vnet_hdr_sz;
2047
2048 if (!(tun->flags & IFF_NO_PI)) {
2049 if (iov_iter_count(iter) < sizeof(pi))
2050 return -EINVAL;
2051
2052 total += sizeof(pi);
2053 if (iov_iter_count(iter) < total) {
2054 /* Packet will be striped */
2055 pi.flags |= TUN_PKT_STRIP;
2056 }
2057
2058 if (copy_to_iter(&pi, sizeof(pi), iter) != sizeof(pi))
2059 return -EFAULT;
2060 }
2061
2062 if (vnet_hdr_sz) {
2063 struct virtio_net_hdr gso;
2064
2065 if (iov_iter_count(iter) < vnet_hdr_sz)
2066 return -EINVAL;
2067
2068 if (virtio_net_hdr_from_skb(skb, &gso,
2069 tun_is_little_endian(tun), true,
2070 vlan_hlen)) {
2071 struct skb_shared_info *sinfo = skb_shinfo(skb);
2072 pr_err("unexpected GSO type: "
2073 "0x%x, gso_size %d, hdr_len %d\n",
2074 sinfo->gso_type, tun16_to_cpu(tun, gso.gso_size),
2075 tun16_to_cpu(tun, gso.hdr_len));
2076 print_hex_dump(KERN_ERR, "tun: ",
2077 DUMP_PREFIX_NONE,
2078 16, 1, skb->head,
2079 min((int)tun16_to_cpu(tun, gso.hdr_len), 64), true);
2080 WARN_ON_ONCE(1);
2081 return -EINVAL;
2082 }
2083
2084 if (copy_to_iter(&gso, sizeof(gso), iter) != sizeof(gso))
2085 return -EFAULT;
2086
2087 iov_iter_advance(iter, vnet_hdr_sz - sizeof(gso));
2088 }
2089
2090 if (vlan_hlen) {
2091 int ret;
2092 struct veth veth;
2093
2094 veth.h_vlan_proto = skb->vlan_proto;
2095 veth.h_vlan_TCI = htons(skb_vlan_tag_get(skb));
2096
2097 vlan_offset = offsetof(struct vlan_ethhdr, h_vlan_proto);
2098
2099 ret = skb_copy_datagram_iter(skb, 0, iter, vlan_offset);
2100 if (ret || !iov_iter_count(iter))
2101 goto done;
2102
2103 ret = copy_to_iter(&veth, sizeof(veth), iter);
2104 if (ret != sizeof(veth) || !iov_iter_count(iter))
2105 goto done;
2106 }
2107
2108 skb_copy_datagram_iter(skb, vlan_offset, iter, skb->len - vlan_offset);
2109
2110done:
2111 /* caller is in process context, */
2112 stats = get_cpu_ptr(tun->pcpu_stats);
2113 u64_stats_update_begin(&stats->syncp);
2114 u64_stats_inc(&stats->tx_packets);
2115 u64_stats_add(&stats->tx_bytes, skb->len + vlan_hlen);
2116 u64_stats_update_end(&stats->syncp);
2117 put_cpu_ptr(tun->pcpu_stats);
2118
2119 return total;
2120}
2121
2122static void *tun_ring_recv(struct tun_file *tfile, int noblock, int *err)
2123{
2124 DECLARE_WAITQUEUE(wait, current);
2125 void *ptr = NULL;
2126 int error = 0;
2127
2128 ptr = ptr_ring_consume(&tfile->tx_ring);
2129 if (ptr)
2130 goto out;
2131 if (noblock) {
2132 error = -EAGAIN;
2133 goto out;
2134 }
2135
2136 add_wait_queue(&tfile->socket.wq.wait, &wait);
2137
2138 while (1) {
2139 set_current_state(TASK_INTERRUPTIBLE);
2140 ptr = ptr_ring_consume(&tfile->tx_ring);
2141 if (ptr)
2142 break;
2143 if (signal_pending(current)) {
2144 error = -ERESTARTSYS;
2145 break;
2146 }
2147 if (tfile->socket.sk->sk_shutdown & RCV_SHUTDOWN) {
2148 error = -EFAULT;
2149 break;
2150 }
2151
2152 schedule();
2153 }
2154
2155 __set_current_state(TASK_RUNNING);
2156 remove_wait_queue(&tfile->socket.wq.wait, &wait);
2157
2158out:
2159 *err = error;
2160 return ptr;
2161}
2162
2163static ssize_t tun_do_read(struct tun_struct *tun, struct tun_file *tfile,
2164 struct iov_iter *to,
2165 int noblock, void *ptr)
2166{
2167 ssize_t ret;
2168 int err;
2169
2170 if (!iov_iter_count(to)) {
2171 tun_ptr_free(ptr);
2172 return 0;
2173 }
2174
2175 if (!ptr) {
2176 /* Read frames from ring */
2177 ptr = tun_ring_recv(tfile, noblock, &err);
2178 if (!ptr)
2179 return err;
2180 }
2181
2182 if (tun_is_xdp_frame(ptr)) {
2183 struct xdp_frame *xdpf = tun_ptr_to_xdp(ptr);
2184
2185 ret = tun_put_user_xdp(tun, tfile, xdpf, to);
2186 xdp_return_frame(xdpf);
2187 } else {
2188 struct sk_buff *skb = ptr;
2189
2190 ret = tun_put_user(tun, tfile, skb, to);
2191 if (unlikely(ret < 0))
2192 kfree_skb(skb);
2193 else
2194 consume_skb(skb);
2195 }
2196
2197 return ret;
2198}
2199
2200static ssize_t tun_chr_read_iter(struct kiocb *iocb, struct iov_iter *to)
2201{
2202 struct file *file = iocb->ki_filp;
2203 struct tun_file *tfile = file->private_data;
2204 struct tun_struct *tun = tun_get(tfile);
2205 ssize_t len = iov_iter_count(to), ret;
2206
2207 if (!tun)
2208 return -EBADFD;
2209 ret = tun_do_read(tun, tfile, to, file->f_flags & O_NONBLOCK, NULL);
2210 ret = min_t(ssize_t, ret, len);
2211 if (ret > 0)
2212 iocb->ki_pos = ret;
2213 tun_put(tun);
2214 return ret;
2215}
2216
2217static void tun_prog_free(struct rcu_head *rcu)
2218{
2219 struct tun_prog *prog = container_of(rcu, struct tun_prog, rcu);
2220
2221 bpf_prog_destroy(prog->prog);
2222 kfree(prog);
2223}
2224
2225static int __tun_set_ebpf(struct tun_struct *tun,
2226 struct tun_prog __rcu **prog_p,
2227 struct bpf_prog *prog)
2228{
2229 struct tun_prog *old, *new = NULL;
2230
2231 if (prog) {
2232 new = kmalloc(sizeof(*new), GFP_KERNEL);
2233 if (!new)
2234 return -ENOMEM;
2235 new->prog = prog;
2236 }
2237
2238 spin_lock_bh(&tun->lock);
2239 old = rcu_dereference_protected(*prog_p,
2240 lockdep_is_held(&tun->lock));
2241 rcu_assign_pointer(*prog_p, new);
2242 spin_unlock_bh(&tun->lock);
2243
2244 if (old)
2245 call_rcu(&old->rcu, tun_prog_free);
2246
2247 return 0;
2248}
2249
2250static void tun_free_netdev(struct net_device *dev)
2251{
2252 struct tun_struct *tun = netdev_priv(dev);
2253
2254 BUG_ON(!(list_empty(&tun->disabled)));
2255
2256 free_percpu(tun->pcpu_stats);
2257 /* We clear pcpu_stats so that tun_set_iff() can tell if
2258 * tun_free_netdev() has been called from register_netdevice().
2259 */
2260 tun->pcpu_stats = NULL;
2261
2262 tun_flow_uninit(tun);
2263 security_tun_dev_free_security(tun->security);
2264 __tun_set_ebpf(tun, &tun->steering_prog, NULL);
2265 __tun_set_ebpf(tun, &tun->filter_prog, NULL);
2266}
2267
2268static void tun_setup(struct net_device *dev)
2269{
2270 struct tun_struct *tun = netdev_priv(dev);
2271
2272 tun->owner = INVALID_UID;
2273 tun->group = INVALID_GID;
2274 tun_default_link_ksettings(dev, &tun->link_ksettings);
2275
2276 dev->ethtool_ops = &tun_ethtool_ops;
2277 dev->needs_free_netdev = true;
2278 dev->priv_destructor = tun_free_netdev;
2279 /* We prefer our own queue length */
2280 dev->tx_queue_len = TUN_READQ_SIZE;
2281}
2282
2283/* Trivial set of netlink ops to allow deleting tun or tap
2284 * device with netlink.
2285 */
2286static int tun_validate(struct nlattr *tb[], struct nlattr *data[],
2287 struct netlink_ext_ack *extack)
2288{
2289 NL_SET_ERR_MSG(extack,
2290 "tun/tap creation via rtnetlink is not supported.");
2291 return -EOPNOTSUPP;
2292}
2293
2294static size_t tun_get_size(const struct net_device *dev)
2295{
2296 BUILD_BUG_ON(sizeof(u32) != sizeof(uid_t));
2297 BUILD_BUG_ON(sizeof(u32) != sizeof(gid_t));
2298
2299 return nla_total_size(sizeof(uid_t)) + /* OWNER */
2300 nla_total_size(sizeof(gid_t)) + /* GROUP */
2301 nla_total_size(sizeof(u8)) + /* TYPE */
2302 nla_total_size(sizeof(u8)) + /* PI */
2303 nla_total_size(sizeof(u8)) + /* VNET_HDR */
2304 nla_total_size(sizeof(u8)) + /* PERSIST */
2305 nla_total_size(sizeof(u8)) + /* MULTI_QUEUE */
2306 nla_total_size(sizeof(u32)) + /* NUM_QUEUES */
2307 nla_total_size(sizeof(u32)) + /* NUM_DISABLED_QUEUES */
2308 0;
2309}
2310
2311static int tun_fill_info(struct sk_buff *skb, const struct net_device *dev)
2312{
2313 struct tun_struct *tun = netdev_priv(dev);
2314
2315 if (nla_put_u8(skb, IFLA_TUN_TYPE, tun->flags & TUN_TYPE_MASK))
2316 goto nla_put_failure;
2317 if (uid_valid(tun->owner) &&
2318 nla_put_u32(skb, IFLA_TUN_OWNER,
2319 from_kuid_munged(current_user_ns(), tun->owner)))
2320 goto nla_put_failure;
2321 if (gid_valid(tun->group) &&
2322 nla_put_u32(skb, IFLA_TUN_GROUP,
2323 from_kgid_munged(current_user_ns(), tun->group)))
2324 goto nla_put_failure;
2325 if (nla_put_u8(skb, IFLA_TUN_PI, !(tun->flags & IFF_NO_PI)))
2326 goto nla_put_failure;
2327 if (nla_put_u8(skb, IFLA_TUN_VNET_HDR, !!(tun->flags & IFF_VNET_HDR)))
2328 goto nla_put_failure;
2329 if (nla_put_u8(skb, IFLA_TUN_PERSIST, !!(tun->flags & IFF_PERSIST)))
2330 goto nla_put_failure;
2331 if (nla_put_u8(skb, IFLA_TUN_MULTI_QUEUE,
2332 !!(tun->flags & IFF_MULTI_QUEUE)))
2333 goto nla_put_failure;
2334 if (tun->flags & IFF_MULTI_QUEUE) {
2335 if (nla_put_u32(skb, IFLA_TUN_NUM_QUEUES, tun->numqueues))
2336 goto nla_put_failure;
2337 if (nla_put_u32(skb, IFLA_TUN_NUM_DISABLED_QUEUES,
2338 tun->numdisabled))
2339 goto nla_put_failure;
2340 }
2341
2342 return 0;
2343
2344nla_put_failure:
2345 return -EMSGSIZE;
2346}
2347
2348static struct rtnl_link_ops tun_link_ops __read_mostly = {
2349 .kind = DRV_NAME,
2350 .priv_size = sizeof(struct tun_struct),
2351 .setup = tun_setup,
2352 .validate = tun_validate,
2353 .get_size = tun_get_size,
2354 .fill_info = tun_fill_info,
2355};
2356
2357static void tun_sock_write_space(struct sock *sk)
2358{
2359 struct tun_file *tfile;
2360 wait_queue_head_t *wqueue;
2361
2362 if (!sock_writeable(sk))
2363 return;
2364
2365 if (!test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags))
2366 return;
2367
2368 wqueue = sk_sleep(sk);
2369 if (wqueue && waitqueue_active(wqueue))
2370 wake_up_interruptible_sync_poll(wqueue, EPOLLOUT |
2371 EPOLLWRNORM | EPOLLWRBAND);
2372
2373 tfile = container_of(sk, struct tun_file, sk);
2374 kill_fasync(&tfile->fasync, SIGIO, POLL_OUT);
2375}
2376
2377static void tun_put_page(struct tun_page *tpage)
2378{
2379 if (tpage->page)
2380 __page_frag_cache_drain(tpage->page, tpage->count);
2381}
2382
2383static int tun_xdp_one(struct tun_struct *tun,
2384 struct tun_file *tfile,
2385 struct xdp_buff *xdp, int *flush,
2386 struct tun_page *tpage)
2387{
2388 unsigned int datasize = xdp->data_end - xdp->data;
2389 struct tun_xdp_hdr *hdr = xdp->data_hard_start;
2390 struct virtio_net_hdr *gso = &hdr->gso;
2391 struct tun_pcpu_stats *stats;
2392 struct bpf_prog *xdp_prog;
2393 struct sk_buff *skb = NULL;
2394 u32 rxhash = 0, act;
2395 int buflen = hdr->buflen;
2396 int err = 0;
2397 bool skb_xdp = false;
2398 struct page *page;
2399
2400 xdp_prog = rcu_dereference(tun->xdp_prog);
2401 if (xdp_prog) {
2402 if (gso->gso_type) {
2403 skb_xdp = true;
2404 goto build;
2405 }
2406 xdp_set_data_meta_invalid(xdp);
2407 xdp->rxq = &tfile->xdp_rxq;
2408 xdp->frame_sz = buflen;
2409
2410 act = bpf_prog_run_xdp(xdp_prog, xdp);
2411 err = tun_xdp_act(tun, xdp_prog, xdp, act);
2412 if (err < 0) {
2413 put_page(virt_to_head_page(xdp->data));
2414 return err;
2415 }
2416
2417 switch (err) {
2418 case XDP_REDIRECT:
2419 *flush = true;
2420 fallthrough;
2421 case XDP_TX:
2422 return 0;
2423 case XDP_PASS:
2424 break;
2425 default:
2426 page = virt_to_head_page(xdp->data);
2427 if (tpage->page == page) {
2428 ++tpage->count;
2429 } else {
2430 tun_put_page(tpage);
2431 tpage->page = page;
2432 tpage->count = 1;
2433 }
2434 return 0;
2435 }
2436 }
2437
2438build:
2439 skb = build_skb(xdp->data_hard_start, buflen);
2440 if (!skb) {
2441 err = -ENOMEM;
2442 goto out;
2443 }
2444
2445 skb_reserve(skb, xdp->data - xdp->data_hard_start);
2446 skb_put(skb, xdp->data_end - xdp->data);
2447
2448 if (virtio_net_hdr_to_skb(skb, gso, tun_is_little_endian(tun))) {
2449 this_cpu_inc(tun->pcpu_stats->rx_frame_errors);
2450 kfree_skb(skb);
2451 err = -EINVAL;
2452 goto out;
2453 }
2454
2455 skb->protocol = eth_type_trans(skb, tun->dev);
2456 skb_reset_network_header(skb);
2457 skb_probe_transport_header(skb);
2458 skb_record_rx_queue(skb, tfile->queue_index);
2459
2460 if (skb_xdp) {
2461 err = do_xdp_generic(xdp_prog, skb);
2462 if (err != XDP_PASS)
2463 goto out;
2464 }
2465
2466 if (!rcu_dereference(tun->steering_prog) && tun->numqueues > 1 &&
2467 !tfile->detached)
2468 rxhash = __skb_get_hash_symmetric(skb);
2469
2470 netif_receive_skb(skb);
2471
2472 /* No need for get_cpu_ptr() here since this function is
2473 * always called with bh disabled
2474 */
2475 stats = this_cpu_ptr(tun->pcpu_stats);
2476 u64_stats_update_begin(&stats->syncp);
2477 u64_stats_inc(&stats->rx_packets);
2478 u64_stats_add(&stats->rx_bytes, datasize);
2479 u64_stats_update_end(&stats->syncp);
2480
2481 if (rxhash)
2482 tun_flow_update(tun, rxhash, tfile);
2483
2484out:
2485 return err;
2486}
2487
2488static int tun_sendmsg(struct socket *sock, struct msghdr *m, size_t total_len)
2489{
2490 int ret, i;
2491 struct tun_file *tfile = container_of(sock, struct tun_file, socket);
2492 struct tun_struct *tun = tun_get(tfile);
2493 struct tun_msg_ctl *ctl = m->msg_control;
2494 struct xdp_buff *xdp;
2495
2496 if (!tun)
2497 return -EBADFD;
2498
2499 if (ctl && (ctl->type == TUN_MSG_PTR)) {
2500 struct tun_page tpage;
2501 int n = ctl->num;
2502 int flush = 0;
2503
2504 memset(&tpage, 0, sizeof(tpage));
2505
2506 local_bh_disable();
2507 rcu_read_lock();
2508
2509 for (i = 0; i < n; i++) {
2510 xdp = &((struct xdp_buff *)ctl->ptr)[i];
2511 tun_xdp_one(tun, tfile, xdp, &flush, &tpage);
2512 }
2513
2514 if (flush)
2515 xdp_do_flush();
2516
2517 rcu_read_unlock();
2518 local_bh_enable();
2519
2520 tun_put_page(&tpage);
2521
2522 ret = total_len;
2523 goto out;
2524 }
2525
2526 ret = tun_get_user(tun, tfile, ctl ? ctl->ptr : NULL, &m->msg_iter,
2527 m->msg_flags & MSG_DONTWAIT,
2528 m->msg_flags & MSG_MORE);
2529out:
2530 tun_put(tun);
2531 return ret;
2532}
2533
2534static int tun_recvmsg(struct socket *sock, struct msghdr *m, size_t total_len,
2535 int flags)
2536{
2537 struct tun_file *tfile = container_of(sock, struct tun_file, socket);
2538 struct tun_struct *tun = tun_get(tfile);
2539 void *ptr = m->msg_control;
2540 int ret;
2541
2542 if (!tun) {
2543 ret = -EBADFD;
2544 goto out_free;
2545 }
2546
2547 if (flags & ~(MSG_DONTWAIT|MSG_TRUNC|MSG_ERRQUEUE)) {
2548 ret = -EINVAL;
2549 goto out_put_tun;
2550 }
2551 if (flags & MSG_ERRQUEUE) {
2552 ret = sock_recv_errqueue(sock->sk, m, total_len,
2553 SOL_PACKET, TUN_TX_TIMESTAMP);
2554 goto out;
2555 }
2556 ret = tun_do_read(tun, tfile, &m->msg_iter, flags & MSG_DONTWAIT, ptr);
2557 if (ret > (ssize_t)total_len) {
2558 m->msg_flags |= MSG_TRUNC;
2559 ret = flags & MSG_TRUNC ? ret : total_len;
2560 }
2561out:
2562 tun_put(tun);
2563 return ret;
2564
2565out_put_tun:
2566 tun_put(tun);
2567out_free:
2568 tun_ptr_free(ptr);
2569 return ret;
2570}
2571
2572static int tun_ptr_peek_len(void *ptr)
2573{
2574 if (likely(ptr)) {
2575 if (tun_is_xdp_frame(ptr)) {
2576 struct xdp_frame *xdpf = tun_ptr_to_xdp(ptr);
2577
2578 return xdpf->len;
2579 }
2580 return __skb_array_len_with_tag(ptr);
2581 } else {
2582 return 0;
2583 }
2584}
2585
2586static int tun_peek_len(struct socket *sock)
2587{
2588 struct tun_file *tfile = container_of(sock, struct tun_file, socket);
2589 struct tun_struct *tun;
2590 int ret = 0;
2591
2592 tun = tun_get(tfile);
2593 if (!tun)
2594 return 0;
2595
2596 ret = PTR_RING_PEEK_CALL(&tfile->tx_ring, tun_ptr_peek_len);
2597 tun_put(tun);
2598
2599 return ret;
2600}
2601
2602/* Ops structure to mimic raw sockets with tun */
2603static const struct proto_ops tun_socket_ops = {
2604 .peek_len = tun_peek_len,
2605 .sendmsg = tun_sendmsg,
2606 .recvmsg = tun_recvmsg,
2607};
2608
2609static struct proto tun_proto = {
2610 .name = "tun",
2611 .owner = THIS_MODULE,
2612 .obj_size = sizeof(struct tun_file),
2613};
2614
2615static int tun_flags(struct tun_struct *tun)
2616{
2617 return tun->flags & (TUN_FEATURES | IFF_PERSIST | IFF_TUN | IFF_TAP);
2618}
2619
2620static ssize_t tun_show_flags(struct device *dev, struct device_attribute *attr,
2621 char *buf)
2622{
2623 struct tun_struct *tun = netdev_priv(to_net_dev(dev));
2624 return sprintf(buf, "0x%x\n", tun_flags(tun));
2625}
2626
2627static ssize_t tun_show_owner(struct device *dev, struct device_attribute *attr,
2628 char *buf)
2629{
2630 struct tun_struct *tun = netdev_priv(to_net_dev(dev));
2631 return uid_valid(tun->owner)?
2632 sprintf(buf, "%u\n",
2633 from_kuid_munged(current_user_ns(), tun->owner)):
2634 sprintf(buf, "-1\n");
2635}
2636
2637static ssize_t tun_show_group(struct device *dev, struct device_attribute *attr,
2638 char *buf)
2639{
2640 struct tun_struct *tun = netdev_priv(to_net_dev(dev));
2641 return gid_valid(tun->group) ?
2642 sprintf(buf, "%u\n",
2643 from_kgid_munged(current_user_ns(), tun->group)):
2644 sprintf(buf, "-1\n");
2645}
2646
2647static DEVICE_ATTR(tun_flags, 0444, tun_show_flags, NULL);
2648static DEVICE_ATTR(owner, 0444, tun_show_owner, NULL);
2649static DEVICE_ATTR(group, 0444, tun_show_group, NULL);
2650
2651static struct attribute *tun_dev_attrs[] = {
2652 &dev_attr_tun_flags.attr,
2653 &dev_attr_owner.attr,
2654 &dev_attr_group.attr,
2655 NULL
2656};
2657
2658static const struct attribute_group tun_attr_group = {
2659 .attrs = tun_dev_attrs
2660};
2661
2662static int tun_set_iff(struct net *net, struct file *file, struct ifreq *ifr)
2663{
2664 struct tun_struct *tun;
2665 struct tun_file *tfile = file->private_data;
2666 struct net_device *dev;
2667 int err;
2668
2669 if (tfile->detached)
2670 return -EINVAL;
2671
2672 if ((ifr->ifr_flags & IFF_NAPI_FRAGS)) {
2673 if (!capable(CAP_NET_ADMIN))
2674 return -EPERM;
2675
2676 if (!(ifr->ifr_flags & IFF_NAPI) ||
2677 (ifr->ifr_flags & TUN_TYPE_MASK) != IFF_TAP)
2678 return -EINVAL;
2679 }
2680
2681 dev = __dev_get_by_name(net, ifr->ifr_name);
2682 if (dev) {
2683 if (ifr->ifr_flags & IFF_TUN_EXCL)
2684 return -EBUSY;
2685 if ((ifr->ifr_flags & IFF_TUN) && dev->netdev_ops == &tun_netdev_ops)
2686 tun = netdev_priv(dev);
2687 else if ((ifr->ifr_flags & IFF_TAP) && dev->netdev_ops == &tap_netdev_ops)
2688 tun = netdev_priv(dev);
2689 else
2690 return -EINVAL;
2691
2692 if (!!(ifr->ifr_flags & IFF_MULTI_QUEUE) !=
2693 !!(tun->flags & IFF_MULTI_QUEUE))
2694 return -EINVAL;
2695
2696 if (tun_not_capable(tun))
2697 return -EPERM;
2698 err = security_tun_dev_open(tun->security);
2699 if (err < 0)
2700 return err;
2701
2702 err = tun_attach(tun, file, ifr->ifr_flags & IFF_NOFILTER,
2703 ifr->ifr_flags & IFF_NAPI,
2704 ifr->ifr_flags & IFF_NAPI_FRAGS, true);
2705 if (err < 0)
2706 return err;
2707
2708 if (tun->flags & IFF_MULTI_QUEUE &&
2709 (tun->numqueues + tun->numdisabled > 1)) {
2710 /* One or more queue has already been attached, no need
2711 * to initialize the device again.
2712 */
2713 netdev_state_change(dev);
2714 return 0;
2715 }
2716
2717 tun->flags = (tun->flags & ~TUN_FEATURES) |
2718 (ifr->ifr_flags & TUN_FEATURES);
2719
2720 netdev_state_change(dev);
2721 } else {
2722 char *name;
2723 unsigned long flags = 0;
2724 int queues = ifr->ifr_flags & IFF_MULTI_QUEUE ?
2725 MAX_TAP_QUEUES : 1;
2726
2727 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
2728 return -EPERM;
2729 err = security_tun_dev_create();
2730 if (err < 0)
2731 return err;
2732
2733 /* Set dev type */
2734 if (ifr->ifr_flags & IFF_TUN) {
2735 /* TUN device */
2736 flags |= IFF_TUN;
2737 name = "tun%d";
2738 } else if (ifr->ifr_flags & IFF_TAP) {
2739 /* TAP device */
2740 flags |= IFF_TAP;
2741 name = "tap%d";
2742 } else
2743 return -EINVAL;
2744
2745 if (*ifr->ifr_name)
2746 name = ifr->ifr_name;
2747
2748 dev = alloc_netdev_mqs(sizeof(struct tun_struct), name,
2749 NET_NAME_UNKNOWN, tun_setup, queues,
2750 queues);
2751
2752 if (!dev)
2753 return -ENOMEM;
2754
2755 dev_net_set(dev, net);
2756 dev->rtnl_link_ops = &tun_link_ops;
2757 dev->ifindex = tfile->ifindex;
2758 dev->sysfs_groups[0] = &tun_attr_group;
2759
2760 tun = netdev_priv(dev);
2761 tun->dev = dev;
2762 tun->flags = flags;
2763 tun->txflt.count = 0;
2764 tun->vnet_hdr_sz = sizeof(struct virtio_net_hdr);
2765
2766 tun->align = NET_SKB_PAD;
2767 tun->filter_attached = false;
2768 tun->sndbuf = tfile->socket.sk->sk_sndbuf;
2769 tun->rx_batched = 0;
2770 RCU_INIT_POINTER(tun->steering_prog, NULL);
2771
2772 tun->pcpu_stats = netdev_alloc_pcpu_stats(struct tun_pcpu_stats);
2773 if (!tun->pcpu_stats) {
2774 err = -ENOMEM;
2775 goto err_free_dev;
2776 }
2777
2778 spin_lock_init(&tun->lock);
2779
2780 err = security_tun_dev_alloc_security(&tun->security);
2781 if (err < 0)
2782 goto err_free_stat;
2783
2784 tun_net_init(dev);
2785 tun_flow_init(tun);
2786
2787 dev->hw_features = NETIF_F_SG | NETIF_F_FRAGLIST |
2788 TUN_USER_FEATURES | NETIF_F_HW_VLAN_CTAG_TX |
2789 NETIF_F_HW_VLAN_STAG_TX;
2790 dev->features = dev->hw_features | NETIF_F_LLTX;
2791 dev->vlan_features = dev->features &
2792 ~(NETIF_F_HW_VLAN_CTAG_TX |
2793 NETIF_F_HW_VLAN_STAG_TX);
2794
2795 tun->flags = (tun->flags & ~TUN_FEATURES) |
2796 (ifr->ifr_flags & TUN_FEATURES);
2797
2798 INIT_LIST_HEAD(&tun->disabled);
2799 err = tun_attach(tun, file, false, ifr->ifr_flags & IFF_NAPI,
2800 ifr->ifr_flags & IFF_NAPI_FRAGS, false);
2801 if (err < 0)
2802 goto err_free_flow;
2803
2804 err = register_netdevice(tun->dev);
2805 if (err < 0)
2806 goto err_detach;
2807 /* free_netdev() won't check refcnt, to aovid race
2808 * with dev_put() we need publish tun after registration.
2809 */
2810 rcu_assign_pointer(tfile->tun, tun);
2811 }
2812
2813 netif_carrier_on(tun->dev);
2814
2815 /* Make sure persistent devices do not get stuck in
2816 * xoff state.
2817 */
2818 if (netif_running(tun->dev))
2819 netif_tx_wake_all_queues(tun->dev);
2820
2821 strcpy(ifr->ifr_name, tun->dev->name);
2822 return 0;
2823
2824err_detach:
2825 tun_detach_all(dev);
2826 /* We are here because register_netdevice() has failed.
2827 * If register_netdevice() already called tun_free_netdev()
2828 * while dealing with the error, tun->pcpu_stats has been cleared.
2829 */
2830 if (!tun->pcpu_stats)
2831 goto err_free_dev;
2832
2833err_free_flow:
2834 tun_flow_uninit(tun);
2835 security_tun_dev_free_security(tun->security);
2836err_free_stat:
2837 free_percpu(tun->pcpu_stats);
2838err_free_dev:
2839 free_netdev(dev);
2840 return err;
2841}
2842
2843static void tun_get_iff(struct tun_struct *tun, struct ifreq *ifr)
2844{
2845 strcpy(ifr->ifr_name, tun->dev->name);
2846
2847 ifr->ifr_flags = tun_flags(tun);
2848
2849}
2850
2851/* This is like a cut-down ethtool ops, except done via tun fd so no
2852 * privs required. */
2853static int set_offload(struct tun_struct *tun, unsigned long arg)
2854{
2855 netdev_features_t features = 0;
2856
2857 if (arg & TUN_F_CSUM) {
2858 features |= NETIF_F_HW_CSUM;
2859 arg &= ~TUN_F_CSUM;
2860
2861 if (arg & (TUN_F_TSO4|TUN_F_TSO6)) {
2862 if (arg & TUN_F_TSO_ECN) {
2863 features |= NETIF_F_TSO_ECN;
2864 arg &= ~TUN_F_TSO_ECN;
2865 }
2866 if (arg & TUN_F_TSO4)
2867 features |= NETIF_F_TSO;
2868 if (arg & TUN_F_TSO6)
2869 features |= NETIF_F_TSO6;
2870 arg &= ~(TUN_F_TSO4|TUN_F_TSO6);
2871 }
2872
2873 arg &= ~TUN_F_UFO;
2874 }
2875
2876 /* This gives the user a way to test for new features in future by
2877 * trying to set them. */
2878 if (arg)
2879 return -EINVAL;
2880
2881 tun->set_features = features;
2882 tun->dev->wanted_features &= ~TUN_USER_FEATURES;
2883 tun->dev->wanted_features |= features;
2884 netdev_update_features(tun->dev);
2885
2886 return 0;
2887}
2888
2889static void tun_detach_filter(struct tun_struct *tun, int n)
2890{
2891 int i;
2892 struct tun_file *tfile;
2893
2894 for (i = 0; i < n; i++) {
2895 tfile = rtnl_dereference(tun->tfiles[i]);
2896 lock_sock(tfile->socket.sk);
2897 sk_detach_filter(tfile->socket.sk);
2898 release_sock(tfile->socket.sk);
2899 }
2900
2901 tun->filter_attached = false;
2902}
2903
2904static int tun_attach_filter(struct tun_struct *tun)
2905{
2906 int i, ret = 0;
2907 struct tun_file *tfile;
2908
2909 for (i = 0; i < tun->numqueues; i++) {
2910 tfile = rtnl_dereference(tun->tfiles[i]);
2911 lock_sock(tfile->socket.sk);
2912 ret = sk_attach_filter(&tun->fprog, tfile->socket.sk);
2913 release_sock(tfile->socket.sk);
2914 if (ret) {
2915 tun_detach_filter(tun, i);
2916 return ret;
2917 }
2918 }
2919
2920 tun->filter_attached = true;
2921 return ret;
2922}
2923
2924static void tun_set_sndbuf(struct tun_struct *tun)
2925{
2926 struct tun_file *tfile;
2927 int i;
2928
2929 for (i = 0; i < tun->numqueues; i++) {
2930 tfile = rtnl_dereference(tun->tfiles[i]);
2931 tfile->socket.sk->sk_sndbuf = tun->sndbuf;
2932 }
2933}
2934
2935static int tun_set_queue(struct file *file, struct ifreq *ifr)
2936{
2937 struct tun_file *tfile = file->private_data;
2938 struct tun_struct *tun;
2939 int ret = 0;
2940
2941 rtnl_lock();
2942
2943 if (ifr->ifr_flags & IFF_ATTACH_QUEUE) {
2944 tun = tfile->detached;
2945 if (!tun) {
2946 ret = -EINVAL;
2947 goto unlock;
2948 }
2949 ret = security_tun_dev_attach_queue(tun->security);
2950 if (ret < 0)
2951 goto unlock;
2952 ret = tun_attach(tun, file, false, tun->flags & IFF_NAPI,
2953 tun->flags & IFF_NAPI_FRAGS, true);
2954 } else if (ifr->ifr_flags & IFF_DETACH_QUEUE) {
2955 tun = rtnl_dereference(tfile->tun);
2956 if (!tun || !(tun->flags & IFF_MULTI_QUEUE) || tfile->detached)
2957 ret = -EINVAL;
2958 else
2959 __tun_detach(tfile, false);
2960 } else
2961 ret = -EINVAL;
2962
2963 if (ret >= 0)
2964 netdev_state_change(tun->dev);
2965
2966unlock:
2967 rtnl_unlock();
2968 return ret;
2969}
2970
2971static int tun_set_ebpf(struct tun_struct *tun, struct tun_prog __rcu **prog_p,
2972 void __user *data)
2973{
2974 struct bpf_prog *prog;
2975 int fd;
2976
2977 if (copy_from_user(&fd, data, sizeof(fd)))
2978 return -EFAULT;
2979
2980 if (fd == -1) {
2981 prog = NULL;
2982 } else {
2983 prog = bpf_prog_get_type(fd, BPF_PROG_TYPE_SOCKET_FILTER);
2984 if (IS_ERR(prog))
2985 return PTR_ERR(prog);
2986 }
2987
2988 return __tun_set_ebpf(tun, prog_p, prog);
2989}
2990
2991static long __tun_chr_ioctl(struct file *file, unsigned int cmd,
2992 unsigned long arg, int ifreq_len)
2993{
2994 struct tun_file *tfile = file->private_data;
2995 struct net *net = sock_net(&tfile->sk);
2996 struct tun_struct *tun;
2997 void __user* argp = (void __user*)arg;
2998 unsigned int ifindex, carrier;
2999 struct ifreq ifr;
3000 kuid_t owner;
3001 kgid_t group;
3002 int sndbuf;
3003 int vnet_hdr_sz;
3004 int le;
3005 int ret;
3006 bool do_notify = false;
3007
3008 if (cmd == TUNSETIFF || cmd == TUNSETQUEUE ||
3009 (_IOC_TYPE(cmd) == SOCK_IOC_TYPE && cmd != SIOCGSKNS)) {
3010 if (copy_from_user(&ifr, argp, ifreq_len))
3011 return -EFAULT;
3012 } else {
3013 memset(&ifr, 0, sizeof(ifr));
3014 }
3015 if (cmd == TUNGETFEATURES) {
3016 /* Currently this just means: "what IFF flags are valid?".
3017 * This is needed because we never checked for invalid flags on
3018 * TUNSETIFF.
3019 */
3020 return put_user(IFF_TUN | IFF_TAP | TUN_FEATURES,
3021 (unsigned int __user*)argp);
3022 } else if (cmd == TUNSETQUEUE) {
3023 return tun_set_queue(file, &ifr);
3024 } else if (cmd == SIOCGSKNS) {
3025 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
3026 return -EPERM;
3027 return open_related_ns(&net->ns, get_net_ns);
3028 }
3029
3030 ret = 0;
3031 rtnl_lock();
3032
3033 tun = tun_get(tfile);
3034 if (cmd == TUNSETIFF) {
3035 ret = -EEXIST;
3036 if (tun)
3037 goto unlock;
3038
3039 ifr.ifr_name[IFNAMSIZ-1] = '\0';
3040
3041 ret = tun_set_iff(net, file, &ifr);
3042
3043 if (ret)
3044 goto unlock;
3045
3046 if (copy_to_user(argp, &ifr, ifreq_len))
3047 ret = -EFAULT;
3048 goto unlock;
3049 }
3050 if (cmd == TUNSETIFINDEX) {
3051 ret = -EPERM;
3052 if (tun)
3053 goto unlock;
3054
3055 ret = -EFAULT;
3056 if (copy_from_user(&ifindex, argp, sizeof(ifindex)))
3057 goto unlock;
3058
3059 ret = 0;
3060 tfile->ifindex = ifindex;
3061 goto unlock;
3062 }
3063
3064 ret = -EBADFD;
3065 if (!tun)
3066 goto unlock;
3067
3068 netif_info(tun, drv, tun->dev, "tun_chr_ioctl cmd %u\n", cmd);
3069
3070 net = dev_net(tun->dev);
3071 ret = 0;
3072 switch (cmd) {
3073 case TUNGETIFF:
3074 tun_get_iff(tun, &ifr);
3075
3076 if (tfile->detached)
3077 ifr.ifr_flags |= IFF_DETACH_QUEUE;
3078 if (!tfile->socket.sk->sk_filter)
3079 ifr.ifr_flags |= IFF_NOFILTER;
3080
3081 if (copy_to_user(argp, &ifr, ifreq_len))
3082 ret = -EFAULT;
3083 break;
3084
3085 case TUNSETNOCSUM:
3086 /* Disable/Enable checksum */
3087
3088 /* [unimplemented] */
3089 netif_info(tun, drv, tun->dev, "ignored: set checksum %s\n",
3090 arg ? "disabled" : "enabled");
3091 break;
3092
3093 case TUNSETPERSIST:
3094 /* Disable/Enable persist mode. Keep an extra reference to the
3095 * module to prevent the module being unprobed.
3096 */
3097 if (arg && !(tun->flags & IFF_PERSIST)) {
3098 tun->flags |= IFF_PERSIST;
3099 __module_get(THIS_MODULE);
3100 do_notify = true;
3101 }
3102 if (!arg && (tun->flags & IFF_PERSIST)) {
3103 tun->flags &= ~IFF_PERSIST;
3104 module_put(THIS_MODULE);
3105 do_notify = true;
3106 }
3107
3108 netif_info(tun, drv, tun->dev, "persist %s\n",
3109 arg ? "enabled" : "disabled");
3110 break;
3111
3112 case TUNSETOWNER:
3113 /* Set owner of the device */
3114 owner = make_kuid(current_user_ns(), arg);
3115 if (!uid_valid(owner)) {
3116 ret = -EINVAL;
3117 break;
3118 }
3119 tun->owner = owner;
3120 do_notify = true;
3121 netif_info(tun, drv, tun->dev, "owner set to %u\n",
3122 from_kuid(&init_user_ns, tun->owner));
3123 break;
3124
3125 case TUNSETGROUP:
3126 /* Set group of the device */
3127 group = make_kgid(current_user_ns(), arg);
3128 if (!gid_valid(group)) {
3129 ret = -EINVAL;
3130 break;
3131 }
3132 tun->group = group;
3133 do_notify = true;
3134 netif_info(tun, drv, tun->dev, "group set to %u\n",
3135 from_kgid(&init_user_ns, tun->group));
3136 break;
3137
3138 case TUNSETLINK:
3139 /* Only allow setting the type when the interface is down */
3140 if (tun->dev->flags & IFF_UP) {
3141 netif_info(tun, drv, tun->dev,
3142 "Linktype set failed because interface is up\n");
3143 ret = -EBUSY;
3144 } else {
3145 tun->dev->type = (int) arg;
3146 netif_info(tun, drv, tun->dev, "linktype set to %d\n",
3147 tun->dev->type);
3148 ret = 0;
3149 }
3150 break;
3151
3152 case TUNSETDEBUG:
3153 tun->msg_enable = (u32)arg;
3154 break;
3155
3156 case TUNSETOFFLOAD:
3157 ret = set_offload(tun, arg);
3158 break;
3159
3160 case TUNSETTXFILTER:
3161 /* Can be set only for TAPs */
3162 ret = -EINVAL;
3163 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3164 break;
3165 ret = update_filter(&tun->txflt, (void __user *)arg);
3166 break;
3167
3168 case SIOCGIFHWADDR:
3169 /* Get hw address */
3170 memcpy(ifr.ifr_hwaddr.sa_data, tun->dev->dev_addr, ETH_ALEN);
3171 ifr.ifr_hwaddr.sa_family = tun->dev->type;
3172 if (copy_to_user(argp, &ifr, ifreq_len))
3173 ret = -EFAULT;
3174 break;
3175
3176 case SIOCSIFHWADDR:
3177 /* Set hw address */
3178 ret = dev_set_mac_address(tun->dev, &ifr.ifr_hwaddr, NULL);
3179 break;
3180
3181 case TUNGETSNDBUF:
3182 sndbuf = tfile->socket.sk->sk_sndbuf;
3183 if (copy_to_user(argp, &sndbuf, sizeof(sndbuf)))
3184 ret = -EFAULT;
3185 break;
3186
3187 case TUNSETSNDBUF:
3188 if (copy_from_user(&sndbuf, argp, sizeof(sndbuf))) {
3189 ret = -EFAULT;
3190 break;
3191 }
3192 if (sndbuf <= 0) {
3193 ret = -EINVAL;
3194 break;
3195 }
3196
3197 tun->sndbuf = sndbuf;
3198 tun_set_sndbuf(tun);
3199 break;
3200
3201 case TUNGETVNETHDRSZ:
3202 vnet_hdr_sz = tun->vnet_hdr_sz;
3203 if (copy_to_user(argp, &vnet_hdr_sz, sizeof(vnet_hdr_sz)))
3204 ret = -EFAULT;
3205 break;
3206
3207 case TUNSETVNETHDRSZ:
3208 if (copy_from_user(&vnet_hdr_sz, argp, sizeof(vnet_hdr_sz))) {
3209 ret = -EFAULT;
3210 break;
3211 }
3212 if (vnet_hdr_sz < (int)sizeof(struct virtio_net_hdr)) {
3213 ret = -EINVAL;
3214 break;
3215 }
3216
3217 tun->vnet_hdr_sz = vnet_hdr_sz;
3218 break;
3219
3220 case TUNGETVNETLE:
3221 le = !!(tun->flags & TUN_VNET_LE);
3222 if (put_user(le, (int __user *)argp))
3223 ret = -EFAULT;
3224 break;
3225
3226 case TUNSETVNETLE:
3227 if (get_user(le, (int __user *)argp)) {
3228 ret = -EFAULT;
3229 break;
3230 }
3231 if (le)
3232 tun->flags |= TUN_VNET_LE;
3233 else
3234 tun->flags &= ~TUN_VNET_LE;
3235 break;
3236
3237 case TUNGETVNETBE:
3238 ret = tun_get_vnet_be(tun, argp);
3239 break;
3240
3241 case TUNSETVNETBE:
3242 ret = tun_set_vnet_be(tun, argp);
3243 break;
3244
3245 case TUNATTACHFILTER:
3246 /* Can be set only for TAPs */
3247 ret = -EINVAL;
3248 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3249 break;
3250 ret = -EFAULT;
3251 if (copy_from_user(&tun->fprog, argp, sizeof(tun->fprog)))
3252 break;
3253
3254 ret = tun_attach_filter(tun);
3255 break;
3256
3257 case TUNDETACHFILTER:
3258 /* Can be set only for TAPs */
3259 ret = -EINVAL;
3260 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3261 break;
3262 ret = 0;
3263 tun_detach_filter(tun, tun->numqueues);
3264 break;
3265
3266 case TUNGETFILTER:
3267 ret = -EINVAL;
3268 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3269 break;
3270 ret = -EFAULT;
3271 if (copy_to_user(argp, &tun->fprog, sizeof(tun->fprog)))
3272 break;
3273 ret = 0;
3274 break;
3275
3276 case TUNSETSTEERINGEBPF:
3277 ret = tun_set_ebpf(tun, &tun->steering_prog, argp);
3278 break;
3279
3280 case TUNSETFILTEREBPF:
3281 ret = tun_set_ebpf(tun, &tun->filter_prog, argp);
3282 break;
3283
3284 case TUNSETCARRIER:
3285 ret = -EFAULT;
3286 if (copy_from_user(&carrier, argp, sizeof(carrier)))
3287 goto unlock;
3288
3289 ret = tun_net_change_carrier(tun->dev, (bool)carrier);
3290 break;
3291
3292 case TUNGETDEVNETNS:
3293 ret = -EPERM;
3294 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
3295 goto unlock;
3296 ret = open_related_ns(&net->ns, get_net_ns);
3297 break;
3298
3299 default:
3300 ret = -EINVAL;
3301 break;
3302 }
3303
3304 if (do_notify)
3305 netdev_state_change(tun->dev);
3306
3307unlock:
3308 rtnl_unlock();
3309 if (tun)
3310 tun_put(tun);
3311 return ret;
3312}
3313
3314static long tun_chr_ioctl(struct file *file,
3315 unsigned int cmd, unsigned long arg)
3316{
3317 return __tun_chr_ioctl(file, cmd, arg, sizeof (struct ifreq));
3318}
3319
3320#ifdef CONFIG_COMPAT
3321static long tun_chr_compat_ioctl(struct file *file,
3322 unsigned int cmd, unsigned long arg)
3323{
3324 switch (cmd) {
3325 case TUNSETIFF:
3326 case TUNGETIFF:
3327 case TUNSETTXFILTER:
3328 case TUNGETSNDBUF:
3329 case TUNSETSNDBUF:
3330 case SIOCGIFHWADDR:
3331 case SIOCSIFHWADDR:
3332 arg = (unsigned long)compat_ptr(arg);
3333 break;
3334 default:
3335 arg = (compat_ulong_t)arg;
3336 break;
3337 }
3338
3339 /*
3340 * compat_ifreq is shorter than ifreq, so we must not access beyond
3341 * the end of that structure. All fields that are used in this
3342 * driver are compatible though, we don't need to convert the
3343 * contents.
3344 */
3345 return __tun_chr_ioctl(file, cmd, arg, sizeof(struct compat_ifreq));
3346}
3347#endif /* CONFIG_COMPAT */
3348
3349static int tun_chr_fasync(int fd, struct file *file, int on)
3350{
3351 struct tun_file *tfile = file->private_data;
3352 int ret;
3353
3354 if ((ret = fasync_helper(fd, file, on, &tfile->fasync)) < 0)
3355 goto out;
3356
3357 if (on) {
3358 __f_setown(file, task_pid(current), PIDTYPE_TGID, 0);
3359 tfile->flags |= TUN_FASYNC;
3360 } else
3361 tfile->flags &= ~TUN_FASYNC;
3362 ret = 0;
3363out:
3364 return ret;
3365}
3366
3367static int tun_chr_open(struct inode *inode, struct file * file)
3368{
3369 struct net *net = current->nsproxy->net_ns;
3370 struct tun_file *tfile;
3371
3372 tfile = (struct tun_file *)sk_alloc(net, AF_UNSPEC, GFP_KERNEL,
3373 &tun_proto, 0);
3374 if (!tfile)
3375 return -ENOMEM;
3376 if (ptr_ring_init(&tfile->tx_ring, 0, GFP_KERNEL)) {
3377 sk_free(&tfile->sk);
3378 return -ENOMEM;
3379 }
3380
3381 mutex_init(&tfile->napi_mutex);
3382 RCU_INIT_POINTER(tfile->tun, NULL);
3383 tfile->flags = 0;
3384 tfile->ifindex = 0;
3385
3386 init_waitqueue_head(&tfile->socket.wq.wait);
3387
3388 tfile->socket.file = file;
3389 tfile->socket.ops = &tun_socket_ops;
3390
3391 sock_init_data(&tfile->socket, &tfile->sk);
3392
3393 tfile->sk.sk_write_space = tun_sock_write_space;
3394 tfile->sk.sk_sndbuf = INT_MAX;
3395
3396 file->private_data = tfile;
3397 INIT_LIST_HEAD(&tfile->next);
3398
3399 sock_set_flag(&tfile->sk, SOCK_ZEROCOPY);
3400
3401 return 0;
3402}
3403
3404static int tun_chr_close(struct inode *inode, struct file *file)
3405{
3406 struct tun_file *tfile = file->private_data;
3407
3408 tun_detach(tfile, true);
3409
3410 return 0;
3411}
3412
3413#ifdef CONFIG_PROC_FS
3414static void tun_chr_show_fdinfo(struct seq_file *m, struct file *file)
3415{
3416 struct tun_file *tfile = file->private_data;
3417 struct tun_struct *tun;
3418 struct ifreq ifr;
3419
3420 memset(&ifr, 0, sizeof(ifr));
3421
3422 rtnl_lock();
3423 tun = tun_get(tfile);
3424 if (tun)
3425 tun_get_iff(tun, &ifr);
3426 rtnl_unlock();
3427
3428 if (tun)
3429 tun_put(tun);
3430
3431 seq_printf(m, "iff:\t%s\n", ifr.ifr_name);
3432}
3433#endif
3434
3435static const struct file_operations tun_fops = {
3436 .owner = THIS_MODULE,
3437 .llseek = no_llseek,
3438 .read_iter = tun_chr_read_iter,
3439 .write_iter = tun_chr_write_iter,
3440 .poll = tun_chr_poll,
3441 .unlocked_ioctl = tun_chr_ioctl,
3442#ifdef CONFIG_COMPAT
3443 .compat_ioctl = tun_chr_compat_ioctl,
3444#endif
3445 .open = tun_chr_open,
3446 .release = tun_chr_close,
3447 .fasync = tun_chr_fasync,
3448#ifdef CONFIG_PROC_FS
3449 .show_fdinfo = tun_chr_show_fdinfo,
3450#endif
3451};
3452
3453static struct miscdevice tun_miscdev = {
3454 .minor = TUN_MINOR,
3455 .name = "tun",
3456 .nodename = "net/tun",
3457 .fops = &tun_fops,
3458};
3459
3460/* ethtool interface */
3461
3462static void tun_default_link_ksettings(struct net_device *dev,
3463 struct ethtool_link_ksettings *cmd)
3464{
3465 ethtool_link_ksettings_zero_link_mode(cmd, supported);
3466 ethtool_link_ksettings_zero_link_mode(cmd, advertising);
3467 cmd->base.speed = SPEED_10;
3468 cmd->base.duplex = DUPLEX_FULL;
3469 cmd->base.port = PORT_TP;
3470 cmd->base.phy_address = 0;
3471 cmd->base.autoneg = AUTONEG_DISABLE;
3472}
3473
3474static int tun_get_link_ksettings(struct net_device *dev,
3475 struct ethtool_link_ksettings *cmd)
3476{
3477 struct tun_struct *tun = netdev_priv(dev);
3478
3479 memcpy(cmd, &tun->link_ksettings, sizeof(*cmd));
3480 return 0;
3481}
3482
3483static int tun_set_link_ksettings(struct net_device *dev,
3484 const struct ethtool_link_ksettings *cmd)
3485{
3486 struct tun_struct *tun = netdev_priv(dev);
3487
3488 memcpy(&tun->link_ksettings, cmd, sizeof(*cmd));
3489 return 0;
3490}
3491
3492static void tun_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
3493{
3494 struct tun_struct *tun = netdev_priv(dev);
3495
3496 strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
3497 strlcpy(info->version, DRV_VERSION, sizeof(info->version));
3498
3499 switch (tun->flags & TUN_TYPE_MASK) {
3500 case IFF_TUN:
3501 strlcpy(info->bus_info, "tun", sizeof(info->bus_info));
3502 break;
3503 case IFF_TAP:
3504 strlcpy(info->bus_info, "tap", sizeof(info->bus_info));
3505 break;
3506 }
3507}
3508
3509static u32 tun_get_msglevel(struct net_device *dev)
3510{
3511 struct tun_struct *tun = netdev_priv(dev);
3512
3513 return tun->msg_enable;
3514}
3515
3516static void tun_set_msglevel(struct net_device *dev, u32 value)
3517{
3518 struct tun_struct *tun = netdev_priv(dev);
3519
3520 tun->msg_enable = value;
3521}
3522
3523static int tun_get_coalesce(struct net_device *dev,
3524 struct ethtool_coalesce *ec)
3525{
3526 struct tun_struct *tun = netdev_priv(dev);
3527
3528 ec->rx_max_coalesced_frames = tun->rx_batched;
3529
3530 return 0;
3531}
3532
3533static int tun_set_coalesce(struct net_device *dev,
3534 struct ethtool_coalesce *ec)
3535{
3536 struct tun_struct *tun = netdev_priv(dev);
3537
3538 if (ec->rx_max_coalesced_frames > NAPI_POLL_WEIGHT)
3539 tun->rx_batched = NAPI_POLL_WEIGHT;
3540 else
3541 tun->rx_batched = ec->rx_max_coalesced_frames;
3542
3543 return 0;
3544}
3545
3546static const struct ethtool_ops tun_ethtool_ops = {
3547 .supported_coalesce_params = ETHTOOL_COALESCE_RX_MAX_FRAMES,
3548 .get_drvinfo = tun_get_drvinfo,
3549 .get_msglevel = tun_get_msglevel,
3550 .set_msglevel = tun_set_msglevel,
3551 .get_link = ethtool_op_get_link,
3552 .get_ts_info = ethtool_op_get_ts_info,
3553 .get_coalesce = tun_get_coalesce,
3554 .set_coalesce = tun_set_coalesce,
3555 .get_link_ksettings = tun_get_link_ksettings,
3556 .set_link_ksettings = tun_set_link_ksettings,
3557};
3558
3559static int tun_queue_resize(struct tun_struct *tun)
3560{
3561 struct net_device *dev = tun->dev;
3562 struct tun_file *tfile;
3563 struct ptr_ring **rings;
3564 int n = tun->numqueues + tun->numdisabled;
3565 int ret, i;
3566
3567 rings = kmalloc_array(n, sizeof(*rings), GFP_KERNEL);
3568 if (!rings)
3569 return -ENOMEM;
3570
3571 for (i = 0; i < tun->numqueues; i++) {
3572 tfile = rtnl_dereference(tun->tfiles[i]);
3573 rings[i] = &tfile->tx_ring;
3574 }
3575 list_for_each_entry(tfile, &tun->disabled, next)
3576 rings[i++] = &tfile->tx_ring;
3577
3578 ret = ptr_ring_resize_multiple(rings, n,
3579 dev->tx_queue_len, GFP_KERNEL,
3580 tun_ptr_free);
3581
3582 kfree(rings);
3583 return ret;
3584}
3585
3586static int tun_device_event(struct notifier_block *unused,
3587 unsigned long event, void *ptr)
3588{
3589 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
3590 struct tun_struct *tun = netdev_priv(dev);
3591 int i;
3592
3593 if (dev->rtnl_link_ops != &tun_link_ops)
3594 return NOTIFY_DONE;
3595
3596 switch (event) {
3597 case NETDEV_CHANGE_TX_QUEUE_LEN:
3598 if (tun_queue_resize(tun))
3599 return NOTIFY_BAD;
3600 break;
3601 case NETDEV_UP:
3602 for (i = 0; i < tun->numqueues; i++) {
3603 struct tun_file *tfile;
3604
3605 tfile = rtnl_dereference(tun->tfiles[i]);
3606 tfile->socket.sk->sk_write_space(tfile->socket.sk);
3607 }
3608 break;
3609 default:
3610 break;
3611 }
3612
3613 return NOTIFY_DONE;
3614}
3615
3616static struct notifier_block tun_notifier_block __read_mostly = {
3617 .notifier_call = tun_device_event,
3618};
3619
3620static int __init tun_init(void)
3621{
3622 int ret = 0;
3623
3624 pr_info("%s, %s\n", DRV_DESCRIPTION, DRV_VERSION);
3625
3626 ret = rtnl_link_register(&tun_link_ops);
3627 if (ret) {
3628 pr_err("Can't register link_ops\n");
3629 goto err_linkops;
3630 }
3631
3632 ret = misc_register(&tun_miscdev);
3633 if (ret) {
3634 pr_err("Can't register misc device %d\n", TUN_MINOR);
3635 goto err_misc;
3636 }
3637
3638 ret = register_netdevice_notifier(&tun_notifier_block);
3639 if (ret) {
3640 pr_err("Can't register netdevice notifier\n");
3641 goto err_notifier;
3642 }
3643
3644 return 0;
3645
3646err_notifier:
3647 misc_deregister(&tun_miscdev);
3648err_misc:
3649 rtnl_link_unregister(&tun_link_ops);
3650err_linkops:
3651 return ret;
3652}
3653
3654static void tun_cleanup(void)
3655{
3656 misc_deregister(&tun_miscdev);
3657 rtnl_link_unregister(&tun_link_ops);
3658 unregister_netdevice_notifier(&tun_notifier_block);
3659}
3660
3661/* Get an underlying socket object from tun file. Returns error unless file is
3662 * attached to a device. The returned object works like a packet socket, it
3663 * can be used for sock_sendmsg/sock_recvmsg. The caller is responsible for
3664 * holding a reference to the file for as long as the socket is in use. */
3665struct socket *tun_get_socket(struct file *file)
3666{
3667 struct tun_file *tfile;
3668 if (file->f_op != &tun_fops)
3669 return ERR_PTR(-EINVAL);
3670 tfile = file->private_data;
3671 if (!tfile)
3672 return ERR_PTR(-EBADFD);
3673 return &tfile->socket;
3674}
3675EXPORT_SYMBOL_GPL(tun_get_socket);
3676
3677struct ptr_ring *tun_get_tx_ring(struct file *file)
3678{
3679 struct tun_file *tfile;
3680
3681 if (file->f_op != &tun_fops)
3682 return ERR_PTR(-EINVAL);
3683 tfile = file->private_data;
3684 if (!tfile)
3685 return ERR_PTR(-EBADFD);
3686 return &tfile->tx_ring;
3687}
3688EXPORT_SYMBOL_GPL(tun_get_tx_ring);
3689
3690module_init(tun_init);
3691module_exit(tun_cleanup);
3692MODULE_DESCRIPTION(DRV_DESCRIPTION);
3693MODULE_AUTHOR(DRV_COPYRIGHT);
3694MODULE_LICENSE("GPL");
3695MODULE_ALIAS_MISCDEV(TUN_MINOR);
3696MODULE_ALIAS("devname:net/tun");