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