projects/linux·blob master · refs

3926 lines · 92.1 KiB

1// SPDX-License-Identifier: GPL-2.0-or-later
2/*
3 * TUN - Universal TUN/TAP device driver.
4 * Copyright (C) 1999-2002 Maxim Krasnyansky <[email protected]>
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 <[email protected]> 2005/08/14
13 * Add TUNSETLINK ioctl to set the link encapsulation
14 *
15 * Mark Smith <[email protected]>
16 * Use eth_random_addr() for tap MAC address.
17 *
18 * Harald Roelle <[email protected]> 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 <[email protected]>
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 <[email protected]>"
34
35#include <linux/module.h>
36#include <linux/errno.h>
37#include <linux/kernel.h>
38#include <linux/sched/signal.h>
39#include <linux/major.h>
40#include <linux/slab.h>
41#include <linux/poll.h>
42#include <linux/fcntl.h>
43#include <linux/init.h>
44#include <linux/skbuff.h>
45#include <linux/netdevice.h>
46#include <linux/etherdevice.h>
47#include <linux/miscdevice.h>
48#include <linux/ethtool.h>
49#include <linux/rtnetlink.h>
50#include <linux/compat.h>
51#include <linux/if.h>
52#include <linux/if_arp.h>
53#include <linux/if_ether.h>
54#include <linux/if_tun.h>
55#include <linux/if_vlan.h>
56#include <linux/crc32.h>
57#include <linux/math.h>
58#include <linux/nsproxy.h>
59#include <linux/virtio_net.h>
60#include <linux/rcupdate.h>
61#include <net/net_namespace.h>
62#include <net/netns/generic.h>
63#include <net/rtnetlink.h>
64#include <net/sock.h>
65#include <net/xdp.h>
66#include <net/ip_tunnels.h>
67#include <linux/seq_file.h>
68#include <linux/uio.h>
69#include <linux/skb_array.h>
70#include <linux/bpf.h>
71#include <linux/bpf_trace.h>
72#include <linux/mutex.h>
73#include <linux/ieee802154.h>
74#include <uapi/linux/if_ltalk.h>
75#include <uapi/linux/if_fddi.h>
76#include <uapi/linux/if_hippi.h>
77#include <uapi/linux/if_fc.h>
78#include <net/ax25.h>
79#include <net/rose.h>
80#include <net/6lowpan.h>
81#include <net/rps.h>
82
83#include <linux/uaccess.h>
84#include <linux/proc_fs.h>
85
86#include "tun_vnet.h"
87
88static void tun_default_link_ksettings(struct net_device *dev,
89 struct ethtool_link_ksettings *cmd);
90
91#define TUN_RX_PAD (NET_IP_ALIGN + NET_SKB_PAD)
92
93/* TUN device flags */
94
95/* IFF_ATTACH_QUEUE is never stored in device flags,
96 * overload it to mean fasync when stored there.
97 */
98#define TUN_FASYNC IFF_ATTACH_QUEUE
99
100#define TUN_FEATURES (IFF_NO_PI | IFF_ONE_QUEUE | IFF_VNET_HDR | \
101 IFF_MULTI_QUEUE | IFF_NAPI | IFF_NAPI_FRAGS | \
102 IFF_BACKPRESSURE)
103
104#define GOODCOPY_LEN 128
105
106#define FLT_EXACT_COUNT 8
107struct tap_filter {
108 unsigned int count; /* Number of addrs. Zero means disabled */
109 u32 mask[2]; /* Mask of the hashed addrs */
110 unsigned char addr[FLT_EXACT_COUNT][ETH_ALEN];
111};
112
113/* MAX_TAP_QUEUES 256 is chosen to allow rx/tx queues to be equal
114 * to max number of VCPUs in guest. */
115#define MAX_TAP_QUEUES 256
116#define MAX_TAP_FLOWS 4096
117
118#define TUN_FLOW_EXPIRE (3 * HZ)
119
120/* A tun_file connects an open character device to a tuntap netdevice. It
121 * also contains all socket related structures (except sock_fprog and tap_filter)
122 * to serve as one transmit queue for tuntap device. The sock_fprog and
123 * tap_filter were kept in tun_struct since they were used for filtering for the
124 * netdevice not for a specific queue (at least I didn't see the requirement for
125 * this).
126 *
127 * RCU usage:
128 * The tun_file and tun_struct are loosely coupled, the pointer from one to the
129 * other can only be read while rcu_read_lock or rtnl_lock is held.
130 */
131struct tun_file {
132 struct sock sk;
133 struct socket socket;
134 struct tun_struct __rcu *tun;
135 struct fasync_struct *fasync;
136 /* only used for fasnyc */
137 unsigned int flags;
138 union {
139 u16 queue_index;
140 unsigned int ifindex;
141 };
142 struct napi_struct napi;
143 bool napi_enabled;
144 bool napi_frags_enabled;
145 struct mutex napi_mutex; /* Protects access to the above napi */
146 struct list_head next;
147 struct tun_struct *detached;
148 struct ptr_ring tx_ring;
149 /* Protected by tx_ring.consumer_lock */
150 int cons_cnt;
151 struct xdp_rxq_info xdp_rxq;
152};
153
154struct tun_page {
155 struct page *page;
156 int count;
157};
158
159struct tun_flow_entry {
160 struct hlist_node hash_link;
161 struct rcu_head rcu;
162 struct tun_struct *tun;
163
164 u32 rxhash;
165 u32 rps_rxhash;
166 int queue_index;
167 unsigned long updated ____cacheline_aligned_in_smp;
168};
169
170#define TUN_NUM_FLOW_ENTRIES 1024
171#define TUN_MASK_FLOW_ENTRIES (TUN_NUM_FLOW_ENTRIES - 1)
172
173struct tun_prog {
174 struct rcu_head rcu;
175 struct bpf_prog *prog;
176};
177
178/* Since the socket were moved to tun_file, to preserve the behavior of persist
179 * device, socket filter, sndbuf and vnet header size were restore when the
180 * file were attached to a persist device.
181 */
182struct tun_struct {
183 struct tun_file __rcu *tfiles[MAX_TAP_QUEUES];
184 unsigned int numqueues;
185 unsigned int flags;
186 kuid_t owner;
187 kgid_t group;
188
189 struct net_device *dev;
190 netdev_features_t set_features;
191#define TUN_USER_FEATURES (NETIF_F_HW_CSUM|NETIF_F_TSO_ECN|NETIF_F_TSO| \
192 NETIF_F_TSO6 | NETIF_F_GSO_UDP_L4 | \
193 NETIF_F_GSO_UDP_TUNNEL | NETIF_F_GSO_UDP_TUNNEL_CSUM)
194
195 int align;
196 int vnet_hdr_sz;
197 int sndbuf;
198 struct tap_filter txflt;
199 struct sock_fprog fprog;
200 /* protected by rtnl lock */
201 bool filter_attached;
202 u32 msg_enable;
203 spinlock_t lock;
204 struct hlist_head flows[TUN_NUM_FLOW_ENTRIES];
205 struct timer_list flow_gc_timer;
206 unsigned long ageing_time;
207 unsigned int numdisabled;
208 struct list_head disabled;
209 void *security;
210 u32 flow_count;
211 u32 rx_batched;
212 atomic_long_t rx_frame_errors;
213 struct bpf_prog __rcu *xdp_prog;
214 struct tun_prog __rcu *steering_prog;
215 struct tun_prog __rcu *filter_prog;
216 struct ethtool_link_ksettings link_ksettings;
217 /* init args */
218 struct file *file;
219 struct ifreq *ifr;
220};
221
222struct veth {
223 __be16 h_vlan_proto;
224 __be16 h_vlan_TCI;
225};
226
227static void tun_flow_init(struct tun_struct *tun);
228static void tun_flow_uninit(struct tun_struct *tun);
229
230static int tun_napi_receive(struct napi_struct *napi, int budget)
231{
232 struct tun_file *tfile = container_of(napi, struct tun_file, napi);
233 struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
234 struct sk_buff_head process_queue;
235 struct sk_buff *skb;
236 int received = 0;
237
238 __skb_queue_head_init(&process_queue);
239
240 spin_lock(&queue->lock);
241 skb_queue_splice_tail_init(queue, &process_queue);
242 spin_unlock(&queue->lock);
243
244 while (received < budget && (skb = __skb_dequeue(&process_queue))) {
245 napi_gro_receive(napi, skb);
246 ++received;
247 }
248
249 if (!skb_queue_empty(&process_queue)) {
250 spin_lock(&queue->lock);
251 skb_queue_splice(&process_queue, queue);
252 spin_unlock(&queue->lock);
253 }
254
255 return received;
256}
257
258static int tun_napi_poll(struct napi_struct *napi, int budget)
259{
260 unsigned int received;
261
262 received = tun_napi_receive(napi, budget);
263
264 if (received < budget)
265 napi_complete_done(napi, received);
266
267 return received;
268}
269
270static void tun_napi_init(struct tun_struct *tun, struct tun_file *tfile,
271 bool napi_en, bool napi_frags)
272{
273 tfile->napi_enabled = napi_en;
274 tfile->napi_frags_enabled = napi_en && napi_frags;
275 if (napi_en) {
276 netif_napi_add_tx(tun->dev, &tfile->napi, tun_napi_poll);
277 napi_enable(&tfile->napi);
278 }
279}
280
281static void tun_napi_enable(struct tun_file *tfile)
282{
283 if (tfile->napi_enabled)
284 napi_enable(&tfile->napi);
285}
286
287static void tun_napi_disable(struct tun_file *tfile)
288{
289 if (tfile->napi_enabled)
290 napi_disable(&tfile->napi);
291}
292
293static void tun_napi_del(struct tun_file *tfile)
294{
295 if (tfile->napi_enabled)
296 netif_napi_del(&tfile->napi);
297}
298
299static bool tun_napi_frags_enabled(const struct tun_file *tfile)
300{
301 return tfile->napi_frags_enabled;
302}
303
304static inline u32 tun_hashfn(u32 rxhash)
305{
306 return rxhash & TUN_MASK_FLOW_ENTRIES;
307}
308
309static struct tun_flow_entry *tun_flow_find(struct hlist_head *head, u32 rxhash)
310{
311 struct tun_flow_entry *e;
312
313 hlist_for_each_entry_rcu(e, head, hash_link) {
314 if (e->rxhash == rxhash)
315 return e;
316 }
317 return NULL;
318}
319
320static struct tun_flow_entry *tun_flow_create(struct tun_struct *tun,
321 struct hlist_head *head,
322 u32 rxhash, u16 queue_index)
323{
324 struct tun_flow_entry *e = kmalloc_obj(*e, GFP_ATOMIC);
325
326 if (e) {
327 netif_info(tun, tx_queued, tun->dev,
328 "create flow: hash %u index %u\n",
329 rxhash, queue_index);
330 e->updated = jiffies;
331 e->rxhash = rxhash;
332 e->rps_rxhash = 0;
333 e->queue_index = queue_index;
334 e->tun = tun;
335 hlist_add_head_rcu(&e->hash_link, head);
336 ++tun->flow_count;
337 }
338 return e;
339}
340
341static void tun_flow_delete(struct tun_struct *tun, struct tun_flow_entry *e)
342{
343 netif_info(tun, tx_queued, tun->dev, "delete flow: hash %u index %u\n",
344 e->rxhash, e->queue_index);
345 hlist_del_rcu(&e->hash_link);
346 kfree_rcu(e, rcu);
347 --tun->flow_count;
348}
349
350static void tun_flow_flush(struct tun_struct *tun)
351{
352 int i;
353
354 spin_lock_bh(&tun->lock);
355 for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
356 struct tun_flow_entry *e;
357 struct hlist_node *n;
358
359 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link)
360 tun_flow_delete(tun, e);
361 }
362 spin_unlock_bh(&tun->lock);
363}
364
365static void tun_flow_delete_by_queue(struct tun_struct *tun, u16 queue_index)
366{
367 int i;
368
369 spin_lock_bh(&tun->lock);
370 for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
371 struct tun_flow_entry *e;
372 struct hlist_node *n;
373
374 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
375 if (e->queue_index == queue_index)
376 tun_flow_delete(tun, e);
377 }
378 }
379 spin_unlock_bh(&tun->lock);
380}
381
382static void tun_flow_cleanup(struct timer_list *t)
383{
384 struct tun_struct *tun = timer_container_of(tun, t, flow_gc_timer);
385 unsigned long delay = tun->ageing_time;
386 unsigned long next_timer = jiffies + delay;
387 unsigned long count = 0;
388 int i;
389
390 spin_lock(&tun->lock);
391 for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
392 struct tun_flow_entry *e;
393 struct hlist_node *n;
394
395 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
396 unsigned long this_timer;
397
398 this_timer = e->updated + delay;
399 if (time_before_eq(this_timer, jiffies)) {
400 tun_flow_delete(tun, e);
401 continue;
402 }
403 count++;
404 if (time_before(this_timer, next_timer))
405 next_timer = this_timer;
406 }
407 }
408
409 if (count)
410 mod_timer(&tun->flow_gc_timer, round_jiffies_up(next_timer));
411 spin_unlock(&tun->lock);
412}
413
414static void tun_flow_update(struct tun_struct *tun, u32 rxhash,
415 struct tun_file *tfile)
416{
417 struct hlist_head *head;
418 struct tun_flow_entry *e;
419 unsigned long delay = tun->ageing_time;
420 u16 queue_index = tfile->queue_index;
421
422 head = &tun->flows[tun_hashfn(rxhash)];
423
424 rcu_read_lock();
425
426 e = tun_flow_find(head, rxhash);
427 if (likely(e)) {
428 /* TODO: keep queueing to old queue until it's empty? */
429 if (READ_ONCE(e->queue_index) != queue_index)
430 WRITE_ONCE(e->queue_index, queue_index);
431 if (e->updated != jiffies)
432 e->updated = jiffies;
433 sock_rps_record_flow_hash(e->rps_rxhash);
434 } else {
435 spin_lock_bh(&tun->lock);
436 if (!tun_flow_find(head, rxhash) &&
437 tun->flow_count < MAX_TAP_FLOWS)
438 tun_flow_create(tun, head, rxhash, queue_index);
439
440 if (!timer_pending(&tun->flow_gc_timer))
441 mod_timer(&tun->flow_gc_timer,
442 round_jiffies_up(jiffies + delay));
443 spin_unlock_bh(&tun->lock);
444 }
445
446 rcu_read_unlock();
447}
448
449/* Save the hash received in the stack receive path and update the
450 * flow_hash table accordingly.
451 */
452static inline void tun_flow_save_rps_rxhash(struct tun_flow_entry *e, u32 hash)
453{
454 if (unlikely(e->rps_rxhash != hash))
455 e->rps_rxhash = hash;
456}
457
458/* We try to identify a flow through its rxhash. The reason that
459 * we do not check rxq no. is because some cards(e.g 82599), chooses
460 * the rxq based on the txq where the last packet of the flow comes. As
461 * the userspace application move between processors, we may get a
462 * different rxq no. here.
463 */
464static u16 tun_automq_select_queue(struct tun_struct *tun, struct sk_buff *skb)
465{
466 struct tun_flow_entry *e;
467 u32 txq, numqueues;
468
469 numqueues = READ_ONCE(tun->numqueues);
470
471 txq = __skb_get_hash_symmetric(skb);
472 e = tun_flow_find(&tun->flows[tun_hashfn(txq)], txq);
473 if (e) {
474 tun_flow_save_rps_rxhash(e, txq);
475 txq = e->queue_index;
476 } else {
477 txq = reciprocal_scale(txq, numqueues);
478 }
479
480 return txq;
481}
482
483static u16 tun_ebpf_select_queue(struct tun_struct *tun, struct sk_buff *skb)
484{
485 struct tun_prog *prog;
486 u32 numqueues;
487 u16 ret = 0;
488
489 numqueues = READ_ONCE(tun->numqueues);
490 if (!numqueues)
491 return 0;
492
493 prog = rcu_dereference(tun->steering_prog);
494 if (prog)
495 ret = bpf_prog_run_clear_cb(prog->prog, skb);
496
497 return ret % numqueues;
498}
499
500static u16 tun_select_queue(struct net_device *dev, struct sk_buff *skb,
501 struct net_device *sb_dev)
502{
503 struct tun_struct *tun = netdev_priv(dev);
504 u16 ret;
505
506 rcu_read_lock();
507 if (rcu_dereference(tun->steering_prog))
508 ret = tun_ebpf_select_queue(tun, skb);
509 else
510 ret = tun_automq_select_queue(tun, skb);
511 rcu_read_unlock();
512
513 return ret;
514}
515
516static inline bool tun_not_capable(struct tun_struct *tun)
517{
518 const struct cred *cred = current_cred();
519 struct net *net = dev_net(tun->dev);
520
521 return ((uid_valid(tun->owner) && !uid_eq(cred->euid, tun->owner)) ||
522 (gid_valid(tun->group) && !in_egroup_p(tun->group))) &&
523 !ns_capable(net->user_ns, CAP_NET_ADMIN);
524}
525
526static void tun_set_real_num_queues(struct tun_struct *tun)
527{
528 netif_set_real_num_tx_queues(tun->dev, tun->numqueues);
529 netif_set_real_num_rx_queues(tun->dev, tun->numqueues);
530}
531
532static void tun_disable_queue(struct tun_struct *tun, struct tun_file *tfile)
533{
534 tfile->detached = tun;
535 list_add_tail(&tfile->next, &tun->disabled);
536 WRITE_ONCE(tun->numdisabled, tun->numdisabled + 1);
537}
538
539static struct tun_struct *tun_enable_queue(struct tun_file *tfile)
540{
541 struct tun_struct *tun = tfile->detached;
542
543 tfile->detached = NULL;
544 list_del_init(&tfile->next);
545 WRITE_ONCE(tun->numdisabled, tun->numdisabled - 1);
546 return tun;
547}
548
549void tun_ptr_free(void *ptr)
550{
551 if (!ptr)
552 return;
553 if (tun_is_xdp_frame(ptr)) {
554 struct xdp_frame *xdpf = tun_ptr_to_xdp(ptr);
555
556 xdp_return_frame(xdpf);
557 } else {
558 __skb_array_destroy_skb(ptr);
559 }
560}
561EXPORT_SYMBOL_GPL(tun_ptr_free);
562
563static void tun_queue_purge(struct tun_file *tfile)
564{
565 void *ptr;
566
567 while ((ptr = ptr_ring_consume(&tfile->tx_ring)) != NULL)
568 tun_ptr_free(ptr);
569
570 skb_queue_purge(&tfile->sk.sk_write_queue);
571 skb_queue_purge(&tfile->sk.sk_error_queue);
572}
573
574static void __tun_detach(struct tun_file *tfile, bool clean)
575{
576 struct tun_file *ntfile;
577 struct tun_struct *tun;
578
579 tun = rtnl_dereference(tfile->tun);
580
581 if (tun && clean) {
582 if (!tfile->detached)
583 tun_napi_disable(tfile);
584 tun_napi_del(tfile);
585 }
586
587 if (tun && !tfile->detached) {
588 u16 index = tfile->queue_index;
589 BUG_ON(index >= tun->numqueues);
590
591 spin_lock(&tfile->tx_ring.consumer_lock);
592 rcu_assign_pointer(tun->tfiles[index],
593 tun->tfiles[tun->numqueues - 1]);
594 spin_unlock(&tfile->tx_ring.consumer_lock);
595 ntfile = rtnl_dereference(tun->tfiles[index]);
596 spin_lock(&ntfile->tx_ring.consumer_lock);
597 ntfile->queue_index = index;
598 ntfile->xdp_rxq.queue_index = index;
599 ntfile->cons_cnt = 0;
600 spin_unlock(&ntfile->tx_ring.consumer_lock);
601 rcu_assign_pointer(tun->tfiles[tun->numqueues - 1],
602 NULL);
603
604 WRITE_ONCE(tun->numqueues, tun->numqueues - 1);
605 if (clean) {
606 RCU_INIT_POINTER(tfile->tun, NULL);
607 sock_put(&tfile->sk);
608 } else {
609 tun_disable_queue(tun, tfile);
610 tun_napi_disable(tfile);
611 }
612
613 synchronize_net();
614 tun_flow_delete_by_queue(tun, tun->numqueues + 1);
615 /* Drop read queue */
616 tun_queue_purge(tfile);
617 spin_lock_bh(&ntfile->tx_ring.consumer_lock);
618 spin_lock(&ntfile->tx_ring.producer_lock);
619 ntfile->cons_cnt = 0;
620 if (netif_running(tun->dev) &&
621 __ptr_ring_empty(&ntfile->tx_ring))
622 netif_wake_subqueue(tun->dev, index);
623 spin_unlock(&ntfile->tx_ring.producer_lock);
624 spin_unlock_bh(&ntfile->tx_ring.consumer_lock);
625 tun_set_real_num_queues(tun);
626 } else if (tfile->detached && clean) {
627 tun = tun_enable_queue(tfile);
628 sock_put(&tfile->sk);
629 }
630
631 if (clean) {
632 if (tun && tun->numqueues == 0 && tun->numdisabled == 0) {
633 netif_carrier_off(tun->dev);
634
635 if (!(tun->flags & IFF_PERSIST) &&
636 tun->dev->reg_state == NETREG_REGISTERED)
637 unregister_netdevice(tun->dev);
638 }
639 if (tun)
640 xdp_rxq_info_unreg(&tfile->xdp_rxq);
641 ptr_ring_cleanup(&tfile->tx_ring, tun_ptr_free);
642 }
643}
644
645static void tun_detach(struct tun_file *tfile, bool clean)
646{
647 struct tun_struct *tun;
648 struct net_device *dev;
649
650 rtnl_lock();
651 tun = rtnl_dereference(tfile->tun);
652 dev = tun ? tun->dev : NULL;
653 __tun_detach(tfile, clean);
654 if (dev)
655 netdev_state_change(dev);
656 rtnl_unlock();
657
658 if (clean)
659 sock_put(&tfile->sk);
660}
661
662static void tun_detach_all(struct net_device *dev)
663{
664 struct tun_struct *tun = netdev_priv(dev);
665 struct tun_file *tfile, *tmp;
666 int i, n = tun->numqueues;
667
668 for (i = 0; i < n; i++) {
669 tfile = rtnl_dereference(tun->tfiles[i]);
670 BUG_ON(!tfile);
671 tun_napi_disable(tfile);
672 tfile->socket.sk->sk_shutdown = RCV_SHUTDOWN;
673 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
674 RCU_INIT_POINTER(tfile->tun, NULL);
675 WRITE_ONCE(tun->numqueues, tun->numqueues - 1);
676 }
677 list_for_each_entry(tfile, &tun->disabled, next) {
678 tfile->socket.sk->sk_shutdown = RCV_SHUTDOWN;
679 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
680 RCU_INIT_POINTER(tfile->tun, NULL);
681 }
682 BUG_ON(tun->numqueues != 0);
683
684 synchronize_net();
685 for (i = 0; i < n; i++) {
686 tfile = rtnl_dereference(tun->tfiles[i]);
687 tun_napi_del(tfile);
688 /* Drop read queue */
689 tun_queue_purge(tfile);
690 xdp_rxq_info_unreg(&tfile->xdp_rxq);
691 sock_put(&tfile->sk);
692 }
693 list_for_each_entry_safe(tfile, tmp, &tun->disabled, next) {
694 tun_napi_del(tfile);
695 tun_enable_queue(tfile);
696 tun_queue_purge(tfile);
697 xdp_rxq_info_unreg(&tfile->xdp_rxq);
698 sock_put(&tfile->sk);
699 }
700 BUG_ON(tun->numdisabled != 0);
701
702 if (tun->flags & IFF_PERSIST)
703 module_put(THIS_MODULE);
704}
705
706static void tun_force_wake_queue(struct tun_struct *tun,
707 struct tun_file *tfile)
708{
709 /* Ensure that the producer can not stop the
710 * queue concurrently by taking locks.
711 */
712 spin_lock_bh(&tfile->tx_ring.consumer_lock);
713 spin_lock(&tfile->tx_ring.producer_lock);
714 tfile->cons_cnt = 0;
715 /* Tested under the locks that tun_net_close() takes, so this can not
716 * undo its stop. tun_net_open() wakes the queues of a device that
717 * comes back up.
718 */
719 if (netif_running(tun->dev))
720 netif_wake_subqueue(tun->dev, tfile->queue_index);
721 spin_unlock(&tfile->tx_ring.producer_lock);
722 spin_unlock_bh(&tfile->tx_ring.consumer_lock);
723}
724
725static int tun_attach(struct tun_struct *tun, struct file *file,
726 bool skip_filter, bool napi, bool napi_frags,
727 bool publish_tun)
728{
729 struct tun_file *tfile = file->private_data;
730 struct net_device *dev = tun->dev;
731 int err;
732
733 err = security_tun_dev_attach(tfile->socket.sk, tun->security);
734 if (err < 0)
735 goto out;
736
737 err = -EINVAL;
738 if (rtnl_dereference(tfile->tun) && !tfile->detached)
739 goto out;
740
741 err = -EBUSY;
742 if (!(tun->flags & IFF_MULTI_QUEUE) && tun->numqueues == 1)
743 goto out;
744
745 err = -E2BIG;
746 if (!tfile->detached &&
747 tun->numqueues + tun->numdisabled == MAX_TAP_QUEUES)
748 goto out;
749
750 err = 0;
751
752 /* Re-attach the filter to persist device */
753 if (!skip_filter && (tun->filter_attached == true)) {
754 lock_sock(tfile->socket.sk);
755 err = sk_attach_filter(&tun->fprog, tfile->socket.sk);
756 release_sock(tfile->socket.sk);
757 if (!err)
758 goto out;
759 }
760
761 if (!tfile->detached &&
762 ptr_ring_resize(&tfile->tx_ring, dev->tx_queue_len,
763 GFP_KERNEL, tun_ptr_free)) {
764 err = -ENOMEM;
765 goto out;
766 }
767
768 spin_lock(&tfile->tx_ring.consumer_lock);
769 tfile->queue_index = tun->numqueues;
770 spin_unlock(&tfile->tx_ring.consumer_lock);
771 tfile->socket.sk->sk_shutdown &= ~RCV_SHUTDOWN;
772 tun_force_wake_queue(tun, tfile);
773
774 if (tfile->detached) {
775 /* Re-attach detached tfile, updating XDP queue_index */
776 WARN_ON(!xdp_rxq_info_is_reg(&tfile->xdp_rxq));
777
778 if (tfile->xdp_rxq.queue_index != tfile->queue_index)
779 tfile->xdp_rxq.queue_index = tfile->queue_index;
780 } else {
781 /* Setup XDP RX-queue info, for new tfile getting attached */
782 err = xdp_rxq_info_reg(&tfile->xdp_rxq,
783 tun->dev, tfile->queue_index, 0);
784 if (err < 0)
785 goto out;
786 err = xdp_rxq_info_reg_mem_model(&tfile->xdp_rxq,
787 MEM_TYPE_PAGE_SHARED, NULL);
788 if (err < 0) {
789 xdp_rxq_info_unreg(&tfile->xdp_rxq);
790 goto out;
791 }
792 err = 0;
793 }
794
795 if (tfile->detached) {
796 tun_enable_queue(tfile);
797 tun_napi_enable(tfile);
798 } else {
799 sock_hold(&tfile->sk);
800 tun_napi_init(tun, tfile, napi, napi_frags);
801 }
802
803 if (rtnl_dereference(tun->xdp_prog))
804 sock_set_flag(&tfile->sk, SOCK_XDP);
805
806 /* device is allowed to go away first, so no need to hold extra
807 * refcnt.
808 */
809
810 /* Publish tfile->tun and tun->tfiles only after we've fully
811 * initialized tfile; otherwise we risk using half-initialized
812 * object.
813 */
814 if (publish_tun)
815 rcu_assign_pointer(tfile->tun, tun);
816 rcu_assign_pointer(tun->tfiles[tun->numqueues], tfile);
817 WRITE_ONCE(tun->numqueues, tun->numqueues + 1);
818 tun_set_real_num_queues(tun);
819out:
820 return err;
821}
822
823static struct tun_struct *tun_get(struct tun_file *tfile)
824{
825 struct tun_struct *tun;
826
827 rcu_read_lock();
828 tun = rcu_dereference(tfile->tun);
829 if (tun)
830 dev_hold(tun->dev);
831 rcu_read_unlock();
832
833 return tun;
834}
835
836static void tun_put(struct tun_struct *tun)
837{
838 dev_put(tun->dev);
839}
840
841/* TAP filtering */
842static void addr_hash_set(u32 *mask, const u8 *addr)
843{
844 int n = ether_crc(ETH_ALEN, addr) >> 26;
845 mask[n >> 5] |= (1 << (n & 31));
846}
847
848static unsigned int addr_hash_test(const u32 *mask, const u8 *addr)
849{
850 int n = ether_crc(ETH_ALEN, addr) >> 26;
851 return mask[n >> 5] & (1 << (n & 31));
852}
853
854static int update_filter(struct tap_filter *filter, void __user *arg)
855{
856 struct { u8 u[ETH_ALEN]; } *addr;
857 struct tun_filter uf;
858 int err, alen, n, nexact;
859
860 if (copy_from_user(&uf, arg, sizeof(uf)))
861 return -EFAULT;
862
863 if (!uf.count) {
864 /* Disabled */
865 filter->count = 0;
866 return 0;
867 }
868
869 alen = ETH_ALEN * uf.count;
870 addr = memdup_user(arg + sizeof(uf), alen);
871 if (IS_ERR(addr))
872 return PTR_ERR(addr);
873
874 /* The filter is updated without holding any locks. Which is
875 * perfectly safe. We disable it first and in the worst
876 * case we'll accept a few undesired packets. */
877 filter->count = 0;
878 wmb();
879
880 /* Use first set of addresses as an exact filter */
881 for (n = 0; n < uf.count && n < FLT_EXACT_COUNT; n++)
882 memcpy(filter->addr[n], addr[n].u, ETH_ALEN);
883
884 nexact = n;
885
886 /* Remaining multicast addresses are hashed,
887 * unicast will leave the filter disabled. */
888 memset(filter->mask, 0, sizeof(filter->mask));
889 for (; n < uf.count; n++) {
890 if (!is_multicast_ether_addr(addr[n].u)) {
891 err = 0; /* no filter */
892 goto free_addr;
893 }
894 addr_hash_set(filter->mask, addr[n].u);
895 }
896
897 /* For ALLMULTI just set the mask to all ones.
898 * This overrides the mask populated above. */
899 if ((uf.flags & TUN_FLT_ALLMULTI))
900 memset(filter->mask, ~0, sizeof(filter->mask));
901
902 /* Now enable the filter */
903 wmb();
904 filter->count = nexact;
905
906 /* Return the number of exact filters */
907 err = nexact;
908free_addr:
909 kfree(addr);
910 return err;
911}
912
913/* Returns: 0 - drop, !=0 - accept */
914static int run_filter(struct tap_filter *filter, const struct sk_buff *skb)
915{
916 /* Cannot use eth_hdr(skb) here because skb_mac_hdr() is incorrect
917 * at this point. */
918 struct ethhdr *eh = (struct ethhdr *) skb->data;
919 int i;
920
921 /* Exact match */
922 for (i = 0; i < filter->count; i++)
923 if (ether_addr_equal(eh->h_dest, filter->addr[i]))
924 return 1;
925
926 /* Inexact match (multicast only) */
927 if (is_multicast_ether_addr(eh->h_dest))
928 return addr_hash_test(filter->mask, eh->h_dest);
929
930 return 0;
931}
932
933/*
934 * Checks whether the packet is accepted or not.
935 * Returns: 0 - drop, !=0 - accept
936 */
937static int check_filter(struct tap_filter *filter, const struct sk_buff *skb)
938{
939 if (!filter->count)
940 return 1;
941
942 return run_filter(filter, skb);
943}
944
945/* Network device part of the driver */
946
947static const struct ethtool_ops tun_ethtool_ops;
948
949static int tun_net_init(struct net_device *dev)
950{
951 struct tun_struct *tun = netdev_priv(dev);
952 struct ifreq *ifr = tun->ifr;
953 int err;
954
955 spin_lock_init(&tun->lock);
956
957 err = security_tun_dev_alloc_security(&tun->security);
958 if (err < 0)
959 return err;
960
961 tun_flow_init(tun);
962
963 dev->pcpu_stat_type = NETDEV_PCPU_STAT_TSTATS;
964 dev->hw_features = NETIF_F_SG | NETIF_F_FRAGLIST |
965 TUN_USER_FEATURES | NETIF_F_HW_VLAN_CTAG_TX |
966 NETIF_F_HW_VLAN_STAG_TX;
967 dev->hw_enc_features = dev->hw_features;
968 dev->features = dev->hw_features;
969 dev->vlan_features = dev->features &
970 ~(NETIF_F_HW_VLAN_CTAG_TX |
971 NETIF_F_HW_VLAN_STAG_TX);
972 dev->lltx = true;
973
974 tun->flags = (tun->flags & ~TUN_FEATURES) |
975 (ifr->ifr_flags & TUN_FEATURES);
976
977 INIT_LIST_HEAD(&tun->disabled);
978 err = tun_attach(tun, tun->file, false, ifr->ifr_flags & IFF_NAPI,
979 ifr->ifr_flags & IFF_NAPI_FRAGS, false);
980 if (err < 0) {
981 tun_flow_uninit(tun);
982 security_tun_dev_free_security(tun->security);
983 return err;
984 }
985 return 0;
986}
987
988/* Net device detach from fd. */
989static void tun_net_uninit(struct net_device *dev)
990{
991 tun_detach_all(dev);
992}
993
994/* Net device open. */
995static int tun_net_open(struct net_device *dev)
996{
997 netif_tx_start_all_queues(dev);
998
999 return 0;
1000}
1001
1002/* Net device close. */
1003static int tun_net_close(struct net_device *dev)
1004{
1005 struct tun_struct *tun = netdev_priv(dev);
1006 struct tun_file *tfile;
1007 int i;
1008
1009 /* netif_running() is already false: take both ring locks to keep the
1010 * wake sites out, so the stop below is the last write to
1011 * __QUEUE_STATE_DRV_XOFF.
1012 */
1013 for (i = 0; i < tun->numqueues; i++) {
1014 tfile = rtnl_dereference(tun->tfiles[i]);
1015
1016 spin_lock_bh(&tfile->tx_ring.consumer_lock);
1017 spin_lock(&tfile->tx_ring.producer_lock);
1018 tfile->cons_cnt = 0;
1019 spin_unlock(&tfile->tx_ring.producer_lock);
1020 spin_unlock_bh(&tfile->tx_ring.consumer_lock);
1021 }
1022
1023 netif_tx_stop_all_queues(dev);
1024 return 0;
1025}
1026
1027/* Net device start xmit */
1028static void tun_automq_xmit(struct tun_struct *tun, struct sk_buff *skb)
1029{
1030#ifdef CONFIG_RPS
1031 if (tun->numqueues == 1 && static_branch_unlikely(&rps_needed)) {
1032 /* Select queue was not called for the skbuff, so we extract the
1033 * RPS hash and save it into the flow_table here.
1034 */
1035 struct tun_flow_entry *e;
1036 __u32 rxhash;
1037
1038 rxhash = __skb_get_hash_symmetric(skb);
1039 e = tun_flow_find(&tun->flows[tun_hashfn(rxhash)], rxhash);
1040 if (e)
1041 tun_flow_save_rps_rxhash(e, rxhash);
1042 }
1043#endif
1044}
1045
1046static unsigned int run_ebpf_filter(struct tun_struct *tun,
1047 struct sk_buff *skb,
1048 int len)
1049{
1050 struct tun_prog *prog = rcu_dereference(tun->filter_prog);
1051
1052 if (prog)
1053 len = bpf_prog_run_clear_cb(prog->prog, skb);
1054
1055 return len;
1056}
1057
1058/* Net device start xmit */
1059static netdev_tx_t tun_net_xmit(struct sk_buff *skb, struct net_device *dev)
1060{
1061 enum skb_drop_reason drop_reason = SKB_DROP_REASON_NOT_SPECIFIED;
1062 struct tun_struct *tun = netdev_priv(dev);
1063 int txq = skb->queue_mapping;
1064 struct netdev_queue *queue;
1065 struct tun_file *tfile;
1066 int len = skb->len;
1067 int ret;
1068
1069 rcu_read_lock();
1070 tfile = rcu_dereference(tun->tfiles[txq]);
1071
1072 /* Drop packet if interface is not attached */
1073 if (!tfile) {
1074 drop_reason = SKB_DROP_REASON_DEV_READY;
1075 goto drop;
1076 }
1077
1078 if (!rcu_dereference(tun->steering_prog))
1079 tun_automq_xmit(tun, skb);
1080
1081 netif_info(tun, tx_queued, tun->dev, "%s %d\n", __func__, skb->len);
1082
1083 /* Drop if the filter does not like it.
1084 * This is a noop if the filter is disabled.
1085 * Filter can be enabled only for the TAP devices. */
1086 if (!check_filter(&tun->txflt, skb)) {
1087 drop_reason = SKB_DROP_REASON_TAP_TXFILTER;
1088 goto drop;
1089 }
1090
1091 if (tfile->socket.sk->sk_filter) {
1092 drop_reason = sk_filter_reason(tfile->socket.sk, skb);
1093 if (drop_reason)
1094 goto drop;
1095 }
1096
1097 len = run_ebpf_filter(tun, skb, len);
1098 if (len == 0) {
1099 drop_reason = SKB_DROP_REASON_TAP_FILTER;
1100 goto drop;
1101 }
1102
1103 if (pskb_trim(skb, len)) {
1104 drop_reason = SKB_DROP_REASON_NOMEM;
1105 goto drop;
1106 }
1107
1108 if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC))) {
1109 drop_reason = SKB_DROP_REASON_SKB_UCOPY_FAULT;
1110 goto drop;
1111 }
1112
1113 skb_tx_timestamp(skb);
1114
1115 /* Orphan the skb - required as we might hang on to it
1116 * for indefinite time.
1117 */
1118 skb_orphan(skb);
1119
1120 nf_reset_ct(skb);
1121
1122 queue = netdev_get_tx_queue(dev, txq);
1123
1124 spin_lock(&tfile->tx_ring.producer_lock);
1125 ret = __ptr_ring_produce(&tfile->tx_ring, skb);
1126 /* Do not touch the queue state of a device that is going down. */
1127 if ((tun->flags & IFF_BACKPRESSURE) && netif_running(dev) &&
1128 !qdisc_txq_has_no_queue(queue) &&
1129 __ptr_ring_check_produce(&tfile->tx_ring) == -ENOSPC) {
1130 netif_tx_stop_queue(queue);
1131 /* Paired with smp_mb() in __tun_wake_queue() */
1132 smp_mb__after_atomic();
1133 if (!__ptr_ring_check_produce(&tfile->tx_ring))
1134 netif_tx_wake_queue(queue);
1135 }
1136 spin_unlock(&tfile->tx_ring.producer_lock);
1137
1138 if (ret) {
1139 /* This should be a rare case if IFF_BACKPRESSURE is enabled and
1140 * a qdisc is present, but can happen due to lltx.
1141 * Since skb_tx_timestamp(), skb_orphan(),
1142 * run_ebpf_filter() and pskb_trim() could have tinkered
1143 * with the SKB, returning NETDEV_TX_BUSY is unsafe and
1144 * we must drop instead.
1145 */
1146 drop_reason = SKB_DROP_REASON_FULL_RING;
1147 goto drop;
1148 }
1149
1150 /* dev->lltx requires to do our own update of trans_start */
1151 txq_trans_cond_update(queue);
1152
1153 /* Notify and wake up reader process */
1154 if (tfile->flags & TUN_FASYNC)
1155 kill_fasync(&tfile->fasync, SIGIO, POLL_IN);
1156 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
1157
1158 rcu_read_unlock();
1159 return NETDEV_TX_OK;
1160
1161drop:
1162 dev_core_stats_tx_dropped_inc(dev);
1163 skb_tx_error(skb);
1164 kfree_skb_reason(skb, drop_reason);
1165 rcu_read_unlock();
1166 return NET_XMIT_DROP;
1167}
1168
1169static void tun_net_mclist(struct net_device *dev)
1170{
1171 /*
1172 * This callback is supposed to deal with mc filter in
1173 * _rx_ path and has nothing to do with the _tx_ path.
1174 * In rx path we always accept everything userspace gives us.
1175 */
1176}
1177
1178static netdev_features_t tun_net_fix_features(struct net_device *dev,
1179 netdev_features_t features)
1180{
1181 struct tun_struct *tun = netdev_priv(dev);
1182
1183 return (features & tun->set_features) | (features & ~TUN_USER_FEATURES);
1184}
1185
1186static void tun_set_headroom(struct net_device *dev, int new_hr)
1187{
1188 struct tun_struct *tun = netdev_priv(dev);
1189 size_t max_headroom;
1190
1191 max_headroom = min_t(size_t, SKB_MAX_HEAD(0), U16_MAX - 1);
1192
1193 if ((tun->flags & TUN_TYPE_MASK) == IFF_TAP)
1194 max_headroom -= ETH_HLEN + NET_IP_ALIGN;
1195 else
1196 max_headroom -= 1;
1197
1198 tun->align = clamp_t(int, new_hr, NET_SKB_PAD, max_headroom);
1199}
1200
1201static void
1202tun_net_get_stats64(struct net_device *dev, struct rtnl_link_stats64 *stats)
1203{
1204 struct tun_struct *tun = netdev_priv(dev);
1205
1206 dev_get_tstats64(dev, stats);
1207
1208 stats->rx_frame_errors +=
1209 (unsigned long)atomic_long_read(&tun->rx_frame_errors);
1210}
1211
1212static int tun_xdp_set(struct net_device *dev, struct bpf_prog *prog,
1213 struct netlink_ext_ack *extack)
1214{
1215 struct tun_struct *tun = netdev_priv(dev);
1216 struct tun_file *tfile;
1217 struct bpf_prog *old_prog;
1218 int i;
1219
1220 old_prog = rtnl_dereference(tun->xdp_prog);
1221 rcu_assign_pointer(tun->xdp_prog, prog);
1222 if (old_prog)
1223 bpf_prog_put(old_prog);
1224
1225 for (i = 0; i < tun->numqueues; i++) {
1226 tfile = rtnl_dereference(tun->tfiles[i]);
1227 if (prog)
1228 sock_set_flag(&tfile->sk, SOCK_XDP);
1229 else
1230 sock_reset_flag(&tfile->sk, SOCK_XDP);
1231 }
1232 list_for_each_entry(tfile, &tun->disabled, next) {
1233 if (prog)
1234 sock_set_flag(&tfile->sk, SOCK_XDP);
1235 else
1236 sock_reset_flag(&tfile->sk, SOCK_XDP);
1237 }
1238
1239 return 0;
1240}
1241
1242static int tun_xdp(struct net_device *dev, struct netdev_bpf *xdp)
1243{
1244 switch (xdp->command) {
1245 case XDP_SETUP_PROG:
1246 return tun_xdp_set(dev, xdp->prog, xdp->extack);
1247 default:
1248 return -EINVAL;
1249 }
1250}
1251
1252static int tun_net_change_carrier(struct net_device *dev, bool new_carrier)
1253{
1254 if (new_carrier) {
1255 struct tun_struct *tun = netdev_priv(dev);
1256
1257 if (!tun->numqueues)
1258 return -EPERM;
1259
1260 netif_carrier_on(dev);
1261 } else {
1262 netif_carrier_off(dev);
1263 }
1264 return 0;
1265}
1266
1267static const struct net_device_ops tun_netdev_ops = {
1268 .ndo_init = tun_net_init,
1269 .ndo_uninit = tun_net_uninit,
1270 .ndo_open = tun_net_open,
1271 .ndo_stop = tun_net_close,
1272 .ndo_start_xmit = tun_net_xmit,
1273 .ndo_fix_features = tun_net_fix_features,
1274 .ndo_select_queue = tun_select_queue,
1275 .ndo_set_rx_headroom = tun_set_headroom,
1276 .ndo_get_stats64 = tun_net_get_stats64,
1277 .ndo_change_carrier = tun_net_change_carrier,
1278};
1279
1280static void __tun_xdp_flush_tfile(struct tun_file *tfile)
1281{
1282 /* Notify and wake up reader process */
1283 if (tfile->flags & TUN_FASYNC)
1284 kill_fasync(&tfile->fasync, SIGIO, POLL_IN);
1285 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
1286}
1287
1288static int tun_xdp_xmit(struct net_device *dev, int n,
1289 struct xdp_frame **frames, u32 flags)
1290{
1291 struct tun_struct *tun = netdev_priv(dev);
1292 struct tun_file *tfile;
1293 u32 numqueues;
1294 int nxmit = 0;
1295 int i;
1296
1297 if (unlikely(flags & ~XDP_XMIT_FLAGS_MASK))
1298 return -EINVAL;
1299
1300 rcu_read_lock();
1301
1302resample:
1303 numqueues = READ_ONCE(tun->numqueues);
1304 if (!numqueues) {
1305 rcu_read_unlock();
1306 return -ENXIO; /* Caller will free/return all frames */
1307 }
1308
1309 tfile = rcu_dereference(tun->tfiles[smp_processor_id() %
1310 numqueues]);
1311 if (unlikely(!tfile))
1312 goto resample;
1313
1314 spin_lock(&tfile->tx_ring.producer_lock);
1315 for (i = 0; i < n; i++) {
1316 struct xdp_frame *xdp = frames[i];
1317 /* Encode the XDP flag into lowest bit for consumer to differ
1318 * XDP buffer from sk_buff.
1319 */
1320 void *frame = tun_xdp_to_ptr(xdp);
1321
1322 if (__ptr_ring_produce(&tfile->tx_ring, frame)) {
1323 dev_core_stats_tx_dropped_inc(dev);
1324 break;
1325 }
1326 nxmit++;
1327 }
1328 spin_unlock(&tfile->tx_ring.producer_lock);
1329
1330 if (flags & XDP_XMIT_FLUSH)
1331 __tun_xdp_flush_tfile(tfile);
1332
1333 rcu_read_unlock();
1334 return nxmit;
1335}
1336
1337static int tun_xdp_tx(struct net_device *dev, struct xdp_buff *xdp)
1338{
1339 struct xdp_frame *frame = xdp_convert_buff_to_frame(xdp);
1340 int nxmit;
1341
1342 if (unlikely(!frame))
1343 return -EOVERFLOW;
1344
1345 nxmit = tun_xdp_xmit(dev, 1, &frame, XDP_XMIT_FLUSH);
1346 if (!nxmit)
1347 xdp_return_frame_rx_napi(frame);
1348 return nxmit;
1349}
1350
1351static const struct net_device_ops tap_netdev_ops = {
1352 .ndo_init = tun_net_init,
1353 .ndo_uninit = tun_net_uninit,
1354 .ndo_open = tun_net_open,
1355 .ndo_stop = tun_net_close,
1356 .ndo_start_xmit = tun_net_xmit,
1357 .ndo_fix_features = tun_net_fix_features,
1358 .ndo_set_rx_mode = tun_net_mclist,
1359 .ndo_set_mac_address = eth_mac_addr,
1360 .ndo_validate_addr = eth_validate_addr,
1361 .ndo_select_queue = tun_select_queue,
1362 .ndo_features_check = passthru_features_check,
1363 .ndo_set_rx_headroom = tun_set_headroom,
1364 .ndo_bpf = tun_xdp,
1365 .ndo_xdp_xmit = tun_xdp_xmit,
1366 .ndo_change_carrier = tun_net_change_carrier,
1367};
1368
1369static void tun_flow_init(struct tun_struct *tun)
1370{
1371 int i;
1372
1373 for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++)
1374 INIT_HLIST_HEAD(&tun->flows[i]);
1375
1376 tun->ageing_time = TUN_FLOW_EXPIRE;
1377 timer_setup(&tun->flow_gc_timer, tun_flow_cleanup, 0);
1378 mod_timer(&tun->flow_gc_timer,
1379 round_jiffies_up(jiffies + tun->ageing_time));
1380}
1381
1382static void tun_flow_uninit(struct tun_struct *tun)
1383{
1384 timer_delete_sync(&tun->flow_gc_timer);
1385 tun_flow_flush(tun);
1386}
1387
1388#define MIN_MTU 68
1389#define MAX_MTU 65535
1390
1391/* Initialize net device. */
1392static void tun_net_initialize(struct net_device *dev)
1393{
1394 struct tun_struct *tun = netdev_priv(dev);
1395
1396 switch (tun->flags & TUN_TYPE_MASK) {
1397 case IFF_TUN:
1398 dev->netdev_ops = &tun_netdev_ops;
1399 dev->header_ops = &ip_tunnel_header_ops;
1400
1401 /* Point-to-Point TUN Device */
1402 dev->hard_header_len = 0;
1403 dev->addr_len = 0;
1404 dev->mtu = 1500;
1405
1406 /* Zero header length */
1407 dev->type = ARPHRD_NONE;
1408 dev->flags = IFF_POINTOPOINT | IFF_NOARP | IFF_MULTICAST;
1409 break;
1410
1411 case IFF_TAP:
1412 dev->netdev_ops = &tap_netdev_ops;
1413 /* Ethernet TAP Device */
1414 ether_setup(dev);
1415 dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1416 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
1417
1418 eth_hw_addr_random(dev);
1419
1420 /* Currently tun does not support XDP, only tap does. */
1421 dev->xdp_features = NETDEV_XDP_ACT_BASIC |
1422 NETDEV_XDP_ACT_REDIRECT |
1423 NETDEV_XDP_ACT_NDO_XMIT;
1424
1425 break;
1426 }
1427
1428 dev->min_mtu = MIN_MTU;
1429 dev->max_mtu = MAX_MTU - dev->hard_header_len;
1430}
1431
1432static bool tun_sock_writeable(struct tun_struct *tun, struct tun_file *tfile)
1433{
1434 struct sock *sk = tfile->socket.sk;
1435
1436 return (tun->dev->flags & IFF_UP) && sock_writeable(sk);
1437}
1438
1439/* Character device part */
1440
1441/* Poll */
1442static __poll_t tun_chr_poll(struct file *file, poll_table *wait)
1443{
1444 struct tun_file *tfile = file->private_data;
1445 struct tun_struct *tun = tun_get(tfile);
1446 struct sock *sk;
1447 __poll_t mask = 0;
1448
1449 if (!tun)
1450 return EPOLLERR;
1451
1452 sk = tfile->socket.sk;
1453
1454 poll_wait(file, sk_sleep(sk), wait);
1455
1456 if (!ptr_ring_empty(&tfile->tx_ring))
1457 mask |= EPOLLIN | EPOLLRDNORM;
1458
1459 /* Make sure SOCKWQ_ASYNC_NOSPACE is set if not writable to
1460 * guarantee EPOLLOUT to be raised by either here or
1461 * tun_sock_write_space(). Then process could get notification
1462 * after it writes to a down device and meets -EIO.
1463 */
1464 if (tun_sock_writeable(tun, tfile) ||
1465 (!test_and_set_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags) &&
1466 tun_sock_writeable(tun, tfile)))
1467 mask |= EPOLLOUT | EPOLLWRNORM;
1468
1469 if (tun->dev->reg_state != NETREG_REGISTERED)
1470 mask = EPOLLERR;
1471
1472 tun_put(tun);
1473 return mask;
1474}
1475
1476static struct sk_buff *tun_napi_alloc_frags(struct tun_file *tfile,
1477 size_t len,
1478 const struct iov_iter *it)
1479{
1480 struct sk_buff *skb;
1481 size_t linear;
1482 int err;
1483 int i;
1484
1485 if (it->nr_segs > MAX_SKB_FRAGS + 1 ||
1486 len > (ETH_MAX_MTU - NET_SKB_PAD - NET_IP_ALIGN))
1487 return ERR_PTR(-EMSGSIZE);
1488
1489 local_bh_disable();
1490 skb = napi_get_frags(&tfile->napi);
1491 local_bh_enable();
1492 if (!skb)
1493 return ERR_PTR(-ENOMEM);
1494
1495 linear = iov_iter_single_seg_count(it);
1496 err = __skb_grow(skb, linear);
1497 if (err)
1498 goto free;
1499
1500 skb->len = len;
1501 skb->data_len = len - linear;
1502 skb->truesize += skb->data_len;
1503
1504 for (i = 1; i < it->nr_segs; i++) {
1505 const struct iovec *iov = iter_iov(it) + i;
1506 size_t fragsz = iov->iov_len;
1507 struct page *page;
1508 void *frag;
1509
1510 if (fragsz == 0 || fragsz > PAGE_SIZE) {
1511 err = -EINVAL;
1512 goto free;
1513 }
1514 frag = netdev_alloc_frag(fragsz);
1515 if (!frag) {
1516 err = -ENOMEM;
1517 goto free;
1518 }
1519 page = virt_to_head_page(frag);
1520 skb_fill_page_desc(skb, i - 1, page,
1521 frag - page_address(page), fragsz);
1522 }
1523
1524 return skb;
1525free:
1526 /* frees skb and all frags allocated with napi_alloc_frag() */
1527 napi_free_frags(&tfile->napi);
1528 return ERR_PTR(err);
1529}
1530
1531/* prepad is the amount to reserve at front. len is length after that.
1532 * linear is a hint as to how much to copy (usually headers). */
1533static struct sk_buff *tun_alloc_skb(struct tun_file *tfile,
1534 size_t prepad, size_t len,
1535 size_t linear, int noblock)
1536{
1537 struct sock *sk = tfile->socket.sk;
1538 struct sk_buff *skb;
1539 int err;
1540
1541 /* Under a page? Don't bother with paged skb. */
1542 if (prepad + len < PAGE_SIZE)
1543 linear = len;
1544
1545 if (len - linear > MAX_SKB_FRAGS * (PAGE_SIZE << PAGE_ALLOC_COSTLY_ORDER))
1546 linear = len - MAX_SKB_FRAGS * (PAGE_SIZE << PAGE_ALLOC_COSTLY_ORDER);
1547 skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
1548 &err, PAGE_ALLOC_COSTLY_ORDER);
1549 if (!skb)
1550 return ERR_PTR(err);
1551
1552 skb_reserve(skb, prepad);
1553 skb_put(skb, linear);
1554 skb->data_len = len - linear;
1555 skb->len += len - linear;
1556
1557 return skb;
1558}
1559
1560static void tun_rx_batched(struct tun_struct *tun, struct tun_file *tfile,
1561 struct sk_buff *skb, int more)
1562{
1563 struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
1564 struct sk_buff_head process_queue;
1565 u32 rx_batched = tun->rx_batched;
1566 bool rcv = false;
1567
1568 if (!rx_batched || (!more && skb_queue_empty(queue))) {
1569 local_bh_disable();
1570 skb_record_rx_queue(skb, tfile->queue_index);
1571 netif_receive_skb(skb);
1572 local_bh_enable();
1573 return;
1574 }
1575
1576 spin_lock(&queue->lock);
1577 if (!more || skb_queue_len(queue) == rx_batched) {
1578 __skb_queue_head_init(&process_queue);
1579 skb_queue_splice_tail_init(queue, &process_queue);
1580 rcv = true;
1581 } else {
1582 __skb_queue_tail(queue, skb);
1583 }
1584 spin_unlock(&queue->lock);
1585
1586 if (rcv) {
1587 struct sk_buff *nskb;
1588
1589 local_bh_disable();
1590 while ((nskb = __skb_dequeue(&process_queue))) {
1591 skb_record_rx_queue(nskb, tfile->queue_index);
1592 netif_receive_skb(nskb);
1593 }
1594 skb_record_rx_queue(skb, tfile->queue_index);
1595 netif_receive_skb(skb);
1596 local_bh_enable();
1597 }
1598}
1599
1600static bool tun_can_build_skb(struct tun_struct *tun, struct tun_file *tfile,
1601 int len, int noblock, bool zerocopy)
1602{
1603 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
1604 return false;
1605
1606 if (tfile->socket.sk->sk_sndbuf != INT_MAX)
1607 return false;
1608
1609 if (!noblock)
1610 return false;
1611
1612 if (zerocopy)
1613 return false;
1614
1615 if (SKB_DATA_ALIGN(len + TUN_RX_PAD + XDP_PACKET_HEADROOM) +
1616 SKB_DATA_ALIGN(sizeof(struct skb_shared_info)) > PAGE_SIZE)
1617 return false;
1618
1619 return true;
1620}
1621
1622static struct sk_buff *__tun_build_skb(struct tun_file *tfile,
1623 struct page_frag *alloc_frag, char *buf,
1624 int buflen, int len, int pad,
1625 int metasize)
1626{
1627 struct sk_buff *skb = build_skb(buf, buflen);
1628
1629 if (!skb)
1630 return ERR_PTR(-ENOMEM);
1631
1632 skb_reserve(skb, pad);
1633 skb_put(skb, len);
1634 if (metasize)
1635 skb_metadata_set(skb, metasize);
1636 skb_set_owner_w(skb, tfile->socket.sk);
1637
1638 get_page(alloc_frag->page);
1639 alloc_frag->offset += buflen;
1640
1641 return skb;
1642}
1643
1644static int tun_xdp_act(struct tun_struct *tun, struct bpf_prog *xdp_prog,
1645 struct xdp_buff *xdp, u32 act)
1646{
1647 int err;
1648
1649 switch (act) {
1650 case XDP_REDIRECT:
1651 err = xdp_do_redirect(tun->dev, xdp, xdp_prog);
1652 if (err) {
1653 dev_core_stats_rx_dropped_inc(tun->dev);
1654 return err;
1655 }
1656 dev_sw_netstats_rx_add(tun->dev, xdp->data_end - xdp->data);
1657 break;
1658 case XDP_TX:
1659 err = tun_xdp_tx(tun->dev, xdp);
1660 if (err < 0) {
1661 dev_core_stats_rx_dropped_inc(tun->dev);
1662 return err;
1663 }
1664 dev_sw_netstats_rx_add(tun->dev, xdp->data_end - xdp->data);
1665 break;
1666 case XDP_PASS:
1667 break;
1668 default:
1669 bpf_warn_invalid_xdp_action(tun->dev, xdp_prog, act);
1670 fallthrough;
1671 case XDP_ABORTED:
1672 trace_xdp_exception(tun->dev, xdp_prog, act);
1673 fallthrough;
1674 case XDP_DROP:
1675 dev_core_stats_rx_dropped_inc(tun->dev);
1676 break;
1677 }
1678
1679 return act;
1680}
1681
1682static struct sk_buff *tun_build_skb(struct tun_struct *tun,
1683 struct tun_file *tfile,
1684 struct iov_iter *from,
1685 struct virtio_net_hdr *hdr,
1686 int len, int *skb_xdp)
1687{
1688 struct page_frag *alloc_frag = &current->task_frag;
1689 struct bpf_net_context __bpf_net_ctx, *bpf_net_ctx;
1690 struct bpf_prog *xdp_prog;
1691 int buflen = SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
1692 char *buf;
1693 size_t copied;
1694 int pad = TUN_RX_PAD;
1695 int metasize = 0;
1696 int err = 0;
1697
1698 rcu_read_lock();
1699 xdp_prog = rcu_dereference(tun->xdp_prog);
1700 if (xdp_prog)
1701 pad += XDP_PACKET_HEADROOM;
1702 buflen += SKB_DATA_ALIGN(len + pad);
1703 rcu_read_unlock();
1704
1705 alloc_frag->offset = ALIGN((u64)alloc_frag->offset, SMP_CACHE_BYTES);
1706 if (unlikely(!skb_page_frag_refill(buflen, alloc_frag, GFP_KERNEL)))
1707 return ERR_PTR(-ENOMEM);
1708
1709 buf = (char *)page_address(alloc_frag->page) + alloc_frag->offset;
1710 copied = copy_page_from_iter(alloc_frag->page,
1711 alloc_frag->offset + pad,
1712 len, from);
1713 if (copied != len)
1714 return ERR_PTR(-EFAULT);
1715
1716 /* There's a small window that XDP may be set after the check
1717 * of xdp_prog above, this should be rare and for simplicity
1718 * we do XDP on skb in case the headroom is not enough.
1719 */
1720 if (hdr->gso_type || !xdp_prog) {
1721 *skb_xdp = 1;
1722 return __tun_build_skb(tfile, alloc_frag, buf, buflen, len,
1723 pad, metasize);
1724 }
1725
1726 *skb_xdp = 0;
1727
1728 local_bh_disable();
1729 rcu_read_lock();
1730 bpf_net_ctx = bpf_net_ctx_set(&__bpf_net_ctx);
1731 xdp_prog = rcu_dereference(tun->xdp_prog);
1732 if (xdp_prog) {
1733 struct xdp_buff xdp;
1734 u32 act;
1735
1736 xdp_init_buff(&xdp, buflen, &tfile->xdp_rxq);
1737 xdp_prepare_buff(&xdp, buf, pad, len, true);
1738
1739 act = bpf_prog_run_xdp(xdp_prog, &xdp);
1740 if (act == XDP_REDIRECT || act == XDP_TX) {
1741 get_page(alloc_frag->page);
1742 alloc_frag->offset += buflen;
1743 }
1744 err = tun_xdp_act(tun, xdp_prog, &xdp, act);
1745 if (err < 0) {
1746 if (act == XDP_REDIRECT || act == XDP_TX)
1747 put_page(alloc_frag->page);
1748 goto out;
1749 }
1750
1751 if (err == XDP_REDIRECT)
1752 xdp_do_flush();
1753 if (err != XDP_PASS)
1754 goto out;
1755
1756 pad = xdp.data - xdp.data_hard_start;
1757 len = xdp.data_end - xdp.data;
1758
1759 /* It is known that the xdp_buff was prepared with metadata
1760 * support, so the metasize will never be negative.
1761 */
1762 metasize = xdp.data - xdp.data_meta;
1763 }
1764 bpf_net_ctx_clear(bpf_net_ctx);
1765 rcu_read_unlock();
1766 local_bh_enable();
1767
1768 return __tun_build_skb(tfile, alloc_frag, buf, buflen, len, pad,
1769 metasize);
1770
1771out:
1772 bpf_net_ctx_clear(bpf_net_ctx);
1773 rcu_read_unlock();
1774 local_bh_enable();
1775 return NULL;
1776}
1777
1778/* Get packet from user space buffer */
1779static ssize_t tun_get_user(struct tun_struct *tun, struct tun_file *tfile,
1780 void *msg_control, struct iov_iter *from,
1781 int noblock, bool more)
1782{
1783 struct tun_pi pi = { 0, cpu_to_be16(ETH_P_IP) };
1784 struct sk_buff *skb;
1785 size_t total_len = iov_iter_count(from);
1786 size_t len = total_len, align = tun->align, linear;
1787 struct virtio_net_hdr_v1_hash_tunnel hdr;
1788 struct virtio_net_hdr *gso;
1789 int good_linear;
1790 int copylen;
1791 int hdr_len = 0;
1792 bool zerocopy = false;
1793 int err;
1794 u32 rxhash = 0;
1795 int skb_xdp = 1;
1796 bool frags = tun_napi_frags_enabled(tfile);
1797 enum skb_drop_reason drop_reason = SKB_DROP_REASON_NOT_SPECIFIED;
1798 netdev_features_t features = 0;
1799
1800 /*
1801 * Keep it easy and always zero the whole buffer, even if the
1802 * tunnel-related field will be touched only when the feature
1803 * is enabled and the hdr size id compatible.
1804 */
1805 memset(&hdr, 0, sizeof(hdr));
1806 gso = (struct virtio_net_hdr *)&hdr;
1807
1808 if (!(tun->flags & IFF_NO_PI)) {
1809 if (len < sizeof(pi))
1810 return -EINVAL;
1811 len -= sizeof(pi);
1812
1813 if (!copy_from_iter_full(&pi, sizeof(pi), from))
1814 return -EFAULT;
1815 }
1816
1817 if (tun->flags & IFF_VNET_HDR) {
1818 int vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
1819
1820 features = tun_vnet_hdr_guest_features(vnet_hdr_sz);
1821 hdr_len = __tun_vnet_hdr_get(vnet_hdr_sz, tun->flags,
1822 features, from, gso);
1823 if (hdr_len < 0)
1824 return hdr_len;
1825
1826 len -= vnet_hdr_sz;
1827 }
1828
1829 if ((tun->flags & TUN_TYPE_MASK) == IFF_TAP) {
1830 align += NET_IP_ALIGN;
1831 if (unlikely(len < ETH_HLEN || (hdr_len && hdr_len < ETH_HLEN)))
1832 return -EINVAL;
1833 }
1834
1835 good_linear = SKB_MAX_HEAD(align);
1836
1837 if (msg_control) {
1838 struct iov_iter i = *from;
1839
1840 /* There are 256 bytes to be copied in skb, so there is
1841 * enough room for skb expand head in case it is used.
1842 * The rest of the buffer is mapped from userspace.
1843 */
1844 copylen = min(hdr_len ? hdr_len : GOODCOPY_LEN, good_linear);
1845 linear = copylen;
1846 iov_iter_advance(&i, copylen);
1847 if (iov_iter_npages(&i, INT_MAX) <= MAX_SKB_FRAGS)
1848 zerocopy = true;
1849 }
1850
1851 if (!frags && tun_can_build_skb(tun, tfile, len, noblock, zerocopy)) {
1852 /* For the packet that is not easy to be processed
1853 * (e.g gso or jumbo packet), we will do it at after
1854 * skb was created with generic XDP routine.
1855 */
1856 skb = tun_build_skb(tun, tfile, from, gso, len, &skb_xdp);
1857 err = PTR_ERR_OR_ZERO(skb);
1858 if (err)
1859 goto drop;
1860 if (!skb)
1861 return total_len;
1862 } else {
1863 if (!zerocopy) {
1864 copylen = len;
1865 linear = min(hdr_len, good_linear);
1866 }
1867
1868 if (frags) {
1869 mutex_lock(&tfile->napi_mutex);
1870 skb = tun_napi_alloc_frags(tfile, copylen, from);
1871 /* tun_napi_alloc_frags() enforces a layout for the skb.
1872 * If zerocopy is enabled, then this layout will be
1873 * overwritten by zerocopy_sg_from_iter().
1874 */
1875 zerocopy = false;
1876 } else {
1877 if (!linear)
1878 linear = min_t(size_t, good_linear, copylen);
1879
1880 skb = tun_alloc_skb(tfile, align, copylen, linear,
1881 noblock);
1882 }
1883
1884 err = PTR_ERR_OR_ZERO(skb);
1885 if (err)
1886 goto drop;
1887
1888 if (zerocopy)
1889 err = zerocopy_sg_from_iter(skb, from);
1890 else
1891 err = skb_copy_datagram_from_iter(skb, 0, from, len);
1892
1893 if (err) {
1894 err = -EFAULT;
1895 drop_reason = SKB_DROP_REASON_SKB_UCOPY_FAULT;
1896 goto drop;
1897 }
1898 }
1899
1900 skb->dev = tun->dev;
1901 switch (tun->flags & TUN_TYPE_MASK) {
1902 case IFF_TUN:
1903 if (tun->flags & IFF_NO_PI) {
1904 u8 ip_version;
1905
1906 if (!pskb_may_pull(skb, 1)) {
1907 err = -EINVAL;
1908 goto drop;
1909 }
1910 ip_version = skb->data[0] >> 4;
1911
1912 switch (ip_version) {
1913 case 4:
1914 pi.proto = htons(ETH_P_IP);
1915 break;
1916 case 6:
1917 pi.proto = htons(ETH_P_IPV6);
1918 break;
1919 default:
1920 err = -EINVAL;
1921 goto drop;
1922 }
1923 }
1924
1925 skb_reset_network_header(skb);
1926 skb->protocol = pi.proto;
1927 break;
1928 case IFF_TAP:
1929 if (!pskb_may_pull(skb, ETH_HLEN)) {
1930 err = -ENOMEM;
1931 drop_reason = SKB_DROP_REASON_HDR_TRUNC;
1932 goto drop;
1933 }
1934 skb_set_network_header(skb, ETH_HLEN);
1935 break;
1936 }
1937
1938 if (tun_vnet_hdr_tnl_to_skb(tun->flags, features, skb, &hdr)) {
1939 atomic_long_inc(&tun->rx_frame_errors);
1940 err = -EINVAL;
1941 goto free_skb;
1942 }
1943
1944 if ((tun->flags & TUN_TYPE_MASK) == IFF_TAP)
1945 skb->protocol = eth_type_trans(skb, tun->dev);
1946
1947 /* copy skb_ubuf_info for callback when skb has no error */
1948 if (zerocopy) {
1949 skb_zcopy_init(skb, msg_control);
1950 } else if (msg_control) {
1951 struct ubuf_info *uarg = msg_control;
1952 uarg->ops->complete(NULL, uarg, false);
1953 }
1954
1955 skb_reset_network_header(skb);
1956 skb_probe_transport_header(skb);
1957 skb_record_rx_queue(skb, tfile->queue_index);
1958
1959 if (skb_xdp) {
1960 struct bpf_prog *xdp_prog;
1961 int ret;
1962
1963 local_bh_disable();
1964 rcu_read_lock();
1965 xdp_prog = rcu_dereference(tun->xdp_prog);
1966 if (xdp_prog) {
1967 ret = do_xdp_generic(xdp_prog, &skb);
1968 if (ret != XDP_PASS) {
1969 rcu_read_unlock();
1970 local_bh_enable();
1971 goto unlock_frags;
1972 }
1973
1974 if (frags && skb != tfile->napi.skb)
1975 tfile->napi.skb = skb;
1976 }
1977 rcu_read_unlock();
1978 local_bh_enable();
1979 }
1980
1981 /* Compute the costly rx hash only if needed for flow updates.
1982 * We may get a very small possibility of OOO during switching, not
1983 * worth to optimize.
1984 */
1985 if (!rcu_access_pointer(tun->steering_prog) && tun->numqueues > 1 &&
1986 !tfile->detached)
1987 rxhash = __skb_get_hash_symmetric(skb);
1988
1989 rcu_read_lock();
1990 if (unlikely(!(tun->dev->flags & IFF_UP))) {
1991 err = -EIO;
1992 rcu_read_unlock();
1993 drop_reason = SKB_DROP_REASON_DEV_READY;
1994 goto drop;
1995 }
1996
1997 if (frags) {
1998 u32 headlen;
1999
2000 /* Exercise flow dissector code path. */
2001 skb_push(skb, ETH_HLEN);
2002 headlen = eth_get_headlen(tun->dev, skb->data,
2003 skb_headlen(skb));
2004
2005 if (unlikely(headlen > skb_headlen(skb))) {
2006 WARN_ON_ONCE(1);
2007 err = -ENOMEM;
2008 dev_core_stats_rx_dropped_inc(tun->dev);
2009napi_busy:
2010 napi_free_frags(&tfile->napi);
2011 rcu_read_unlock();
2012 mutex_unlock(&tfile->napi_mutex);
2013 return err;
2014 }
2015
2016 if (likely(napi_schedule_prep(&tfile->napi))) {
2017 local_bh_disable();
2018 napi_gro_frags(&tfile->napi);
2019 napi_complete(&tfile->napi);
2020 local_bh_enable();
2021 } else {
2022 err = -EBUSY;
2023 goto napi_busy;
2024 }
2025 mutex_unlock(&tfile->napi_mutex);
2026 } else if (tfile->napi_enabled) {
2027 struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
2028 int queue_len;
2029
2030 spin_lock_bh(&queue->lock);
2031
2032 if (unlikely(tfile->detached)) {
2033 spin_unlock_bh(&queue->lock);
2034 rcu_read_unlock();
2035 err = -EBUSY;
2036 goto free_skb;
2037 }
2038
2039 __skb_queue_tail(queue, skb);
2040 queue_len = skb_queue_len(queue);
2041 spin_unlock(&queue->lock);
2042
2043 if (!more || queue_len > NAPI_POLL_WEIGHT)
2044 napi_schedule(&tfile->napi);
2045
2046 local_bh_enable();
2047 } else if (!IS_ENABLED(CONFIG_4KSTACKS)) {
2048 tun_rx_batched(tun, tfile, skb, more);
2049 } else {
2050 netif_rx(skb);
2051 }
2052 rcu_read_unlock();
2053
2054 preempt_disable();
2055 dev_sw_netstats_rx_add(tun->dev, len);
2056 preempt_enable();
2057
2058 if (rxhash)
2059 tun_flow_update(tun, rxhash, tfile);
2060
2061 return total_len;
2062
2063drop:
2064 if (err != -EAGAIN)
2065 dev_core_stats_rx_dropped_inc(tun->dev);
2066
2067free_skb:
2068 if (!IS_ERR_OR_NULL(skb))
2069 kfree_skb_reason(skb, drop_reason);
2070
2071unlock_frags:
2072 if (frags) {
2073 tfile->napi.skb = NULL;
2074 mutex_unlock(&tfile->napi_mutex);
2075 }
2076
2077 return err ?: total_len;
2078}
2079
2080static ssize_t tun_chr_write_iter(struct kiocb *iocb, struct iov_iter *from)
2081{
2082 struct file *file = iocb->ki_filp;
2083 struct tun_file *tfile = file->private_data;
2084 struct tun_struct *tun = tun_get(tfile);
2085 ssize_t result;
2086 int noblock = 0;
2087
2088 if (!tun)
2089 return -EBADFD;
2090
2091 if ((file->f_flags & O_NONBLOCK) || (iocb->ki_flags & IOCB_NOWAIT))
2092 noblock = 1;
2093
2094 result = tun_get_user(tun, tfile, NULL, from, noblock, false);
2095
2096 tun_put(tun);
2097 return result;
2098}
2099
2100static ssize_t tun_put_user_xdp(struct tun_struct *tun,
2101 struct tun_file *tfile,
2102 struct xdp_frame *xdp_frame,
2103 struct iov_iter *iter)
2104{
2105 int vnet_hdr_sz = 0;
2106 size_t size = xdp_frame->len;
2107 ssize_t ret;
2108
2109 if (tun->flags & IFF_VNET_HDR) {
2110 struct virtio_net_hdr gso = { 0 };
2111
2112 vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
2113 ret = tun_vnet_hdr_put(vnet_hdr_sz, iter, &gso);
2114 if (ret)
2115 return ret;
2116 }
2117
2118 ret = copy_to_iter(xdp_frame->data, size, iter) + vnet_hdr_sz;
2119
2120 preempt_disable();
2121 dev_sw_netstats_tx_add(tun->dev, 1, ret);
2122 preempt_enable();
2123
2124 return ret;
2125}
2126
2127/* Put packet to the user space buffer */
2128static ssize_t tun_put_user(struct tun_struct *tun,
2129 struct tun_file *tfile,
2130 struct sk_buff *skb,
2131 struct iov_iter *iter)
2132{
2133 struct tun_pi pi = { 0, skb->protocol };
2134 ssize_t total;
2135 int vlan_offset = 0;
2136 int vlan_hlen = 0;
2137 int vnet_hdr_sz = 0;
2138 int ret;
2139
2140 if (skb_vlan_tag_present(skb))
2141 vlan_hlen = VLAN_HLEN;
2142
2143 if (tun->flags & IFF_VNET_HDR)
2144 vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
2145
2146 total = skb->len + vlan_hlen + vnet_hdr_sz;
2147
2148 if (!(tun->flags & IFF_NO_PI)) {
2149 if (iov_iter_count(iter) < sizeof(pi))
2150 return -EINVAL;
2151
2152 total += sizeof(pi);
2153 if (iov_iter_count(iter) < total) {
2154 /* Packet will be striped */
2155 pi.flags |= TUN_PKT_STRIP;
2156 }
2157
2158 if (copy_to_iter(&pi, sizeof(pi), iter) != sizeof(pi))
2159 return -EFAULT;
2160 }
2161
2162 if (vnet_hdr_sz) {
2163 struct virtio_net_hdr_v1_hash_tunnel hdr;
2164 struct virtio_net_hdr *gso;
2165
2166 memset(&hdr, 0, sizeof(hdr));
2167 ret = tun_vnet_hdr_tnl_from_skb(tun->flags, tun->dev, skb,
2168 &hdr);
2169 if (ret)
2170 return ret;
2171
2172 /*
2173 * Drop the packet if the configured header size is too small
2174 * WRT the enabled offloads.
2175 */
2176 gso = (struct virtio_net_hdr *)&hdr;
2177 ret = __tun_vnet_hdr_put(vnet_hdr_sz, tun->dev->features,
2178 iter, gso);
2179 if (ret)
2180 return ret;
2181 }
2182
2183 if (vlan_hlen) {
2184 int ret;
2185 struct veth veth;
2186
2187 veth.h_vlan_proto = skb->vlan_proto;
2188 veth.h_vlan_TCI = htons(skb_vlan_tag_get(skb));
2189
2190 vlan_offset = offsetof(struct vlan_ethhdr, h_vlan_proto);
2191
2192 ret = skb_copy_datagram_iter(skb, 0, iter, vlan_offset);
2193 if (ret || !iov_iter_count(iter))
2194 goto done;
2195
2196 ret = copy_to_iter(&veth, sizeof(veth), iter);
2197 if (ret != sizeof(veth) || !iov_iter_count(iter))
2198 goto done;
2199 }
2200
2201 skb_copy_datagram_iter(skb, vlan_offset, iter, skb->len - vlan_offset);
2202
2203done:
2204 /* caller is in process context, */
2205 preempt_disable();
2206 dev_sw_netstats_tx_add(tun->dev, 1, skb->len + vlan_hlen);
2207 preempt_enable();
2208
2209 return total;
2210}
2211
2212/* Callers must hold ring.consumer_lock */
2213static void __tun_wake_queue(struct tun_struct *tun,
2214 struct tun_file *tfile, int consumed)
2215{
2216 u16 queue_index = tfile->queue_index;
2217 struct netdev_queue *txq;
2218
2219 lockdep_assert_held(&tfile->tx_ring.consumer_lock);
2220
2221 if (!(tun->flags & IFF_BACKPRESSURE))
2222 return;
2223
2224 /* A stop from tun_net_close() is not backpressure, leave it alone. */
2225 if (unlikely(!netif_running(tun->dev)))
2226 return;
2227
2228 /* Only the current owner of the slot may wake its subqueue. */
2229 if (unlikely(rcu_access_pointer(tun->tfiles[queue_index]) != tfile))
2230 return;
2231
2232 txq = netdev_get_tx_queue(tun->dev, queue_index);
2233
2234 /* Paired with smp_mb__after_atomic() in tun_net_xmit() */
2235 smp_mb();
2236 if (netif_tx_queue_stopped(txq)) {
2237 tfile->cons_cnt += consumed;
2238 if (tfile->cons_cnt >= tfile->tx_ring.size / 2 ||
2239 __ptr_ring_empty(&tfile->tx_ring)) {
2240 netif_tx_wake_queue(txq);
2241 tfile->cons_cnt = 0;
2242 }
2243 }
2244}
2245
2246static void *tun_ring_consume(struct tun_struct *tun, struct tun_file *tfile)
2247{
2248 void *ptr;
2249
2250 spin_lock(&tfile->tx_ring.consumer_lock);
2251 ptr = __ptr_ring_consume(&tfile->tx_ring);
2252 if (ptr)
2253 __tun_wake_queue(tun, tfile, 1);
2254
2255 spin_unlock(&tfile->tx_ring.consumer_lock);
2256 return ptr;
2257}
2258
2259static void *tun_ring_recv(struct tun_struct *tun, struct tun_file *tfile,
2260 int noblock, int *err)
2261{
2262 DECLARE_WAITQUEUE(wait, current);
2263 void *ptr = NULL;
2264 int error = 0;
2265
2266 ptr = tun_ring_consume(tun, tfile);
2267 if (ptr)
2268 goto out;
2269 if (noblock) {
2270 error = -EAGAIN;
2271 goto out;
2272 }
2273
2274 add_wait_queue(&tfile->socket.wq.wait, &wait);
2275
2276 while (1) {
2277 set_current_state(TASK_INTERRUPTIBLE);
2278 ptr = tun_ring_consume(tun, tfile);
2279 if (ptr)
2280 break;
2281 if (signal_pending(current)) {
2282 error = -ERESTARTSYS;
2283 break;
2284 }
2285 if (tfile->socket.sk->sk_shutdown & RCV_SHUTDOWN) {
2286 error = -EFAULT;
2287 break;
2288 }
2289
2290 schedule();
2291 }
2292
2293 __set_current_state(TASK_RUNNING);
2294 remove_wait_queue(&tfile->socket.wq.wait, &wait);
2295
2296out:
2297 *err = error;
2298 return ptr;
2299}
2300
2301static ssize_t tun_do_read(struct tun_struct *tun, struct tun_file *tfile,
2302 struct iov_iter *to,
2303 int noblock, void *ptr)
2304{
2305 ssize_t ret;
2306 int err;
2307
2308 if (!iov_iter_count(to)) {
2309 tun_ptr_free(ptr);
2310 return 0;
2311 }
2312
2313 if (!ptr) {
2314 /* Read frames from ring */
2315 ptr = tun_ring_recv(tun, tfile, noblock, &err);
2316 if (!ptr)
2317 return err;
2318 }
2319
2320 if (tun_is_xdp_frame(ptr)) {
2321 struct xdp_frame *xdpf = tun_ptr_to_xdp(ptr);
2322
2323 ret = tun_put_user_xdp(tun, tfile, xdpf, to);
2324 xdp_return_frame(xdpf);
2325 } else {
2326 struct sk_buff *skb = ptr;
2327
2328 ret = tun_put_user(tun, tfile, skb, to);
2329 if (unlikely(ret < 0))
2330 kfree_skb(skb);
2331 else
2332 consume_skb(skb);
2333 }
2334
2335 return ret;
2336}
2337
2338static ssize_t tun_chr_read_iter(struct kiocb *iocb, struct iov_iter *to)
2339{
2340 struct file *file = iocb->ki_filp;
2341 struct tun_file *tfile = file->private_data;
2342 struct tun_struct *tun = tun_get(tfile);
2343 ssize_t len = iov_iter_count(to), ret;
2344 int noblock = 0;
2345
2346 if (!tun)
2347 return -EBADFD;
2348
2349 if ((file->f_flags & O_NONBLOCK) || (iocb->ki_flags & IOCB_NOWAIT))
2350 noblock = 1;
2351
2352 ret = tun_do_read(tun, tfile, to, noblock, NULL);
2353 ret = min_t(ssize_t, ret, len);
2354 if (ret > 0)
2355 iocb->ki_pos = ret;
2356 tun_put(tun);
2357 return ret;
2358}
2359
2360static void tun_prog_free(struct rcu_head *rcu)
2361{
2362 struct tun_prog *prog = container_of(rcu, struct tun_prog, rcu);
2363
2364 bpf_prog_destroy(prog->prog);
2365 kfree(prog);
2366}
2367
2368static int __tun_set_ebpf(struct tun_struct *tun,
2369 struct tun_prog __rcu **prog_p,
2370 struct bpf_prog *prog)
2371{
2372 struct tun_prog *old, *new = NULL;
2373
2374 if (prog) {
2375 new = kmalloc_obj(*new);
2376 if (!new)
2377 return -ENOMEM;
2378 new->prog = prog;
2379 }
2380
2381 spin_lock_bh(&tun->lock);
2382 old = rcu_dereference_protected(*prog_p,
2383 lockdep_is_held(&tun->lock));
2384 rcu_assign_pointer(*prog_p, new);
2385 spin_unlock_bh(&tun->lock);
2386
2387 if (old)
2388 call_rcu(&old->rcu, tun_prog_free);
2389
2390 return 0;
2391}
2392
2393static void tun_free_netdev(struct net_device *dev)
2394{
2395 struct tun_struct *tun = netdev_priv(dev);
2396
2397 BUG_ON(!(list_empty(&tun->disabled)));
2398
2399 tun_flow_uninit(tun);
2400 security_tun_dev_free_security(tun->security);
2401 __tun_set_ebpf(tun, &tun->steering_prog, NULL);
2402 __tun_set_ebpf(tun, &tun->filter_prog, NULL);
2403}
2404
2405static void tun_setup(struct net_device *dev)
2406{
2407 struct tun_struct *tun = netdev_priv(dev);
2408
2409 tun->owner = INVALID_UID;
2410 tun->group = INVALID_GID;
2411 tun_default_link_ksettings(dev, &tun->link_ksettings);
2412
2413 dev->ethtool_ops = &tun_ethtool_ops;
2414 dev->needs_free_netdev = true;
2415 dev->priv_destructor = tun_free_netdev;
2416 /* We prefer our own queue length */
2417 dev->tx_queue_len = TUN_READQ_SIZE;
2418}
2419
2420/* Trivial set of netlink ops to allow deleting tun or tap
2421 * device with netlink.
2422 */
2423static int tun_validate(struct nlattr *tb[], struct nlattr *data[],
2424 struct netlink_ext_ack *extack)
2425{
2426 NL_SET_ERR_MSG(extack,
2427 "tun/tap creation via rtnetlink is not supported.");
2428 return -EOPNOTSUPP;
2429}
2430
2431static size_t tun_get_size(const struct net_device *dev)
2432{
2433 BUILD_BUG_ON(sizeof(u32) != sizeof(uid_t));
2434 BUILD_BUG_ON(sizeof(u32) != sizeof(gid_t));
2435
2436 return nla_total_size(sizeof(uid_t)) + /* OWNER */
2437 nla_total_size(sizeof(gid_t)) + /* GROUP */
2438 nla_total_size(sizeof(u8)) + /* TYPE */
2439 nla_total_size(sizeof(u8)) + /* PI */
2440 nla_total_size(sizeof(u8)) + /* VNET_HDR */
2441 nla_total_size(sizeof(u8)) + /* PERSIST */
2442 nla_total_size(sizeof(u8)) + /* MULTI_QUEUE */
2443 nla_total_size(sizeof(u32)) + /* NUM_QUEUES */
2444 nla_total_size(sizeof(u32)) + /* NUM_DISABLED_QUEUES */
2445 0;
2446}
2447
2448static int tun_fill_info(struct sk_buff *skb, const struct net_device *dev)
2449{
2450 const struct tun_struct *tun = netdev_priv(dev);
2451 unsigned int flags = READ_ONCE(tun->flags);
2452 kuid_t owner = READ_ONCE(tun->owner);
2453 kgid_t group = READ_ONCE(tun->group);
2454
2455 if (nla_put_u8(skb, IFLA_TUN_TYPE, flags & TUN_TYPE_MASK))
2456 goto nla_put_failure;
2457 if (uid_valid(owner) &&
2458 nla_put_u32(skb, IFLA_TUN_OWNER,
2459 from_kuid_munged(current_user_ns(), owner)))
2460 goto nla_put_failure;
2461 if (gid_valid(group) &&
2462 nla_put_u32(skb, IFLA_TUN_GROUP,
2463 from_kgid_munged(current_user_ns(), group)))
2464 goto nla_put_failure;
2465 if (nla_put_u8(skb, IFLA_TUN_PI, !(flags & IFF_NO_PI)))
2466 goto nla_put_failure;
2467 if (nla_put_u8(skb, IFLA_TUN_VNET_HDR, !!(flags & IFF_VNET_HDR)))
2468 goto nla_put_failure;
2469 if (nla_put_u8(skb, IFLA_TUN_PERSIST, !!(flags & IFF_PERSIST)))
2470 goto nla_put_failure;
2471 if (nla_put_u8(skb, IFLA_TUN_MULTI_QUEUE,
2472 !!(flags & IFF_MULTI_QUEUE)))
2473 goto nla_put_failure;
2474 if (flags & IFF_MULTI_QUEUE) {
2475 if (nla_put_u32(skb, IFLA_TUN_NUM_QUEUES,
2476 READ_ONCE(tun->numqueues)))
2477 goto nla_put_failure;
2478 if (nla_put_u32(skb, IFLA_TUN_NUM_DISABLED_QUEUES,
2479 READ_ONCE(tun->numdisabled)))
2480 goto nla_put_failure;
2481 }
2482
2483 return 0;
2484
2485nla_put_failure:
2486 return -EMSGSIZE;
2487}
2488
2489static struct rtnl_link_ops tun_link_ops __read_mostly = {
2490 .kind = DRV_NAME,
2491 .priv_size = sizeof(struct tun_struct),
2492 .setup = tun_setup,
2493 .validate = tun_validate,
2494 .get_size = tun_get_size,
2495 .fill_info = tun_fill_info,
2496};
2497
2498static void tun_sock_write_space(struct sock *sk)
2499{
2500 struct tun_file *tfile;
2501 wait_queue_head_t *wqueue;
2502
2503 if (!sock_writeable(sk))
2504 return;
2505
2506 if (!test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags))
2507 return;
2508
2509 wqueue = sk_sleep(sk);
2510 if (wqueue && waitqueue_active(wqueue))
2511 wake_up_interruptible_sync_poll(wqueue, EPOLLOUT |
2512 EPOLLWRNORM | EPOLLWRBAND);
2513
2514 tfile = container_of(sk, struct tun_file, sk);
2515 kill_fasync(&tfile->fasync, SIGIO, POLL_OUT);
2516}
2517
2518static void tun_put_page(struct tun_page *tpage)
2519{
2520 if (tpage->page)
2521 __page_frag_cache_drain(tpage->page, tpage->count);
2522}
2523
2524static int tun_xdp_one(struct tun_struct *tun,
2525 struct tun_file *tfile,
2526 struct xdp_buff *xdp, int *flush,
2527 struct tun_page *tpage)
2528{
2529 unsigned int datasize = xdp->data_end - xdp->data;
2530 struct virtio_net_hdr *gso = xdp->data_hard_start;
2531 struct virtio_net_hdr_v1_hash_tunnel *tnl_hdr;
2532 struct bpf_prog *xdp_prog;
2533 struct sk_buff *skb = NULL;
2534 struct sk_buff_head *queue;
2535 netdev_features_t features;
2536 u32 rxhash = 0, act;
2537 int buflen = xdp->frame_sz;
2538 int metasize = 0;
2539 int ret = 0;
2540 bool skb_xdp = false;
2541 struct page *page;
2542
2543 if (unlikely(datasize < ETH_HLEN)) {
2544 put_page(virt_to_head_page(xdp->data));
2545 return -EINVAL;
2546 }
2547
2548 xdp_prog = rcu_dereference(tun->xdp_prog);
2549 if (xdp_prog) {
2550 if (gso->gso_type) {
2551 skb_xdp = true;
2552 goto build;
2553 }
2554
2555 xdp_init_buff(xdp, buflen, &tfile->xdp_rxq);
2556
2557 act = bpf_prog_run_xdp(xdp_prog, xdp);
2558 ret = tun_xdp_act(tun, xdp_prog, xdp, act);
2559 if (ret < 0) {
2560 put_page(virt_to_head_page(xdp->data));
2561 return ret;
2562 }
2563
2564 switch (ret) {
2565 case XDP_REDIRECT:
2566 *flush = true;
2567 fallthrough;
2568 case XDP_TX:
2569 return 0;
2570 case XDP_PASS:
2571 break;
2572 default:
2573 page = virt_to_head_page(xdp->data);
2574 if (tpage->page == page) {
2575 ++tpage->count;
2576 } else {
2577 tun_put_page(tpage);
2578 tpage->page = page;
2579 tpage->count = 1;
2580 }
2581 return 0;
2582 }
2583 }
2584
2585build:
2586 skb = build_skb(xdp->data_hard_start, buflen);
2587 if (!skb) {
2588 put_page(virt_to_head_page(xdp->data));
2589 ret = -ENOMEM;
2590 goto out;
2591 }
2592
2593 skb_reserve(skb, xdp->data - xdp->data_hard_start);
2594 skb_put(skb, xdp->data_end - xdp->data);
2595
2596 /* The externally provided xdp_buff may have no metadata support, which
2597 * is marked by xdp->data_meta being xdp->data + 1. This will lead to a
2598 * metasize of -1 and is the reason why the condition checks for > 0.
2599 */
2600 metasize = xdp->data - xdp->data_meta;
2601 if (metasize > 0)
2602 skb_metadata_set(skb, metasize);
2603
2604 features = tun_vnet_hdr_guest_features(READ_ONCE(tun->vnet_hdr_sz));
2605 tnl_hdr = (struct virtio_net_hdr_v1_hash_tunnel *)gso;
2606 skb->dev = tun->dev;
2607 skb_set_network_header(skb, ETH_HLEN);
2608 if (tun_vnet_hdr_tnl_to_skb(tun->flags, features, skb, tnl_hdr)) {
2609 atomic_long_inc(&tun->rx_frame_errors);
2610 kfree_skb(skb);
2611 ret = -EINVAL;
2612 goto out;
2613 }
2614
2615 skb->protocol = eth_type_trans(skb, tun->dev);
2616 skb_reset_network_header(skb);
2617 skb_probe_transport_header(skb);
2618 skb_record_rx_queue(skb, tfile->queue_index);
2619
2620 if (skb_xdp) {
2621 ret = do_xdp_generic(xdp_prog, &skb);
2622 if (ret != XDP_PASS) {
2623 ret = 0;
2624 goto out;
2625 }
2626 }
2627
2628 if (!rcu_dereference(tun->steering_prog) && tun->numqueues > 1 &&
2629 !tfile->detached)
2630 rxhash = __skb_get_hash_symmetric(skb);
2631
2632 if (tfile->napi_enabled) {
2633 queue = &tfile->sk.sk_write_queue;
2634 spin_lock(&queue->lock);
2635
2636 if (unlikely(tfile->detached)) {
2637 spin_unlock(&queue->lock);
2638 kfree_skb(skb);
2639 return -EBUSY;
2640 }
2641
2642 __skb_queue_tail(queue, skb);
2643 spin_unlock(&queue->lock);
2644 ret = 1;
2645 } else {
2646 netif_receive_skb(skb);
2647 ret = 0;
2648 }
2649
2650 /* No need to disable preemption here since this function is
2651 * always called with bh disabled
2652 */
2653 dev_sw_netstats_rx_add(tun->dev, datasize);
2654
2655 if (rxhash)
2656 tun_flow_update(tun, rxhash, tfile);
2657
2658out:
2659 return ret;
2660}
2661
2662static int tun_sendmsg(struct socket *sock, struct msghdr *m, size_t total_len)
2663{
2664 int ret, i;
2665 struct tun_file *tfile = container_of(sock, struct tun_file, socket);
2666 struct tun_struct *tun = tun_get(tfile);
2667 struct tun_msg_ctl *ctl = m->msg_control;
2668 struct xdp_buff *xdp;
2669
2670 if (!tun)
2671 return -EBADFD;
2672
2673 if (m->msg_controllen == sizeof(struct tun_msg_ctl) &&
2674 ctl && ctl->type == TUN_MSG_PTR) {
2675 struct bpf_net_context __bpf_net_ctx, *bpf_net_ctx;
2676 struct tun_page tpage;
2677 int n = ctl->num;
2678 int flush = 0, queued = 0;
2679
2680 memset(&tpage, 0, sizeof(tpage));
2681
2682 local_bh_disable();
2683 rcu_read_lock();
2684 bpf_net_ctx = bpf_net_ctx_set(&__bpf_net_ctx);
2685
2686 for (i = 0; i < n; i++) {
2687 xdp = &((struct xdp_buff *)ctl->ptr)[i];
2688 ret = tun_xdp_one(tun, tfile, xdp, &flush, &tpage);
2689 if (ret > 0)
2690 queued += ret;
2691 }
2692
2693 if (flush)
2694 xdp_do_flush();
2695
2696 if (tfile->napi_enabled && queued > 0)
2697 napi_schedule(&tfile->napi);
2698
2699 bpf_net_ctx_clear(bpf_net_ctx);
2700 rcu_read_unlock();
2701 local_bh_enable();
2702
2703 tun_put_page(&tpage);
2704
2705 ret = total_len;
2706 goto out;
2707 }
2708
2709 ret = tun_get_user(tun, tfile, ctl ? ctl->ptr : NULL, &m->msg_iter,
2710 m->msg_flags & MSG_DONTWAIT,
2711 m->msg_flags & MSG_MORE);
2712out:
2713 tun_put(tun);
2714 return ret;
2715}
2716
2717static int tun_recvmsg(struct socket *sock, struct msghdr *m, size_t total_len,
2718 int flags)
2719{
2720 struct tun_file *tfile = container_of(sock, struct tun_file, socket);
2721 struct tun_struct *tun = tun_get(tfile);
2722 void *ptr = m->msg_control;
2723 int ret;
2724
2725 if (!tun) {
2726 ret = -EBADFD;
2727 goto out_free;
2728 }
2729
2730 if (flags & ~(MSG_DONTWAIT|MSG_TRUNC|MSG_ERRQUEUE)) {
2731 ret = -EINVAL;
2732 goto out_put_tun;
2733 }
2734 if (flags & MSG_ERRQUEUE) {
2735 ret = sock_recv_errqueue(sock->sk, m, total_len,
2736 SOL_PACKET, TUN_TX_TIMESTAMP);
2737 goto out;
2738 }
2739 ret = tun_do_read(tun, tfile, &m->msg_iter, flags & MSG_DONTWAIT, ptr);
2740 if (ret > (ssize_t)total_len) {
2741 m->msg_flags |= MSG_TRUNC;
2742 ret = flags & MSG_TRUNC ? ret : total_len;
2743 }
2744out:
2745 tun_put(tun);
2746 return ret;
2747
2748out_put_tun:
2749 tun_put(tun);
2750out_free:
2751 tun_ptr_free(ptr);
2752 return ret;
2753}
2754
2755static int tun_ptr_peek_len(void *ptr)
2756{
2757 if (likely(ptr)) {
2758 if (tun_is_xdp_frame(ptr)) {
2759 struct xdp_frame *xdpf = tun_ptr_to_xdp(ptr);
2760
2761 return xdpf->len;
2762 }
2763 return __skb_array_len_with_tag(ptr);
2764 } else {
2765 return 0;
2766 }
2767}
2768
2769static int tun_peek_len(struct socket *sock)
2770{
2771 struct tun_file *tfile = container_of(sock, struct tun_file, socket);
2772 struct tun_struct *tun;
2773 int ret = 0;
2774
2775 tun = tun_get(tfile);
2776 if (!tun)
2777 return 0;
2778
2779 ret = PTR_RING_PEEK_CALL(&tfile->tx_ring, tun_ptr_peek_len);
2780 tun_put(tun);
2781
2782 return ret;
2783}
2784
2785/* Ops structure to mimic raw sockets with tun */
2786static const struct proto_ops tun_socket_ops = {
2787 .peek_len = tun_peek_len,
2788 .sendmsg = tun_sendmsg,
2789 .recvmsg = tun_recvmsg,
2790};
2791
2792static struct proto tun_proto = {
2793 .name = "tun",
2794 .owner = THIS_MODULE,
2795 .obj_size = sizeof(struct tun_file),
2796};
2797
2798static int tun_flags(struct tun_struct *tun)
2799{
2800 return tun->flags & (TUN_FEATURES | IFF_PERSIST | IFF_TUN | IFF_TAP);
2801}
2802
2803static ssize_t tun_flags_show(struct device *dev, struct device_attribute *attr,
2804 char *buf)
2805{
2806 struct tun_struct *tun = netdev_priv(to_net_dev(dev));
2807 return sysfs_emit(buf, "0x%x\n", tun_flags(tun));
2808}
2809
2810static ssize_t owner_show(struct device *dev, struct device_attribute *attr,
2811 char *buf)
2812{
2813 struct tun_struct *tun = netdev_priv(to_net_dev(dev));
2814 return uid_valid(tun->owner)?
2815 sysfs_emit(buf, "%u\n",
2816 from_kuid_munged(current_user_ns(), tun->owner)) :
2817 sysfs_emit(buf, "-1\n");
2818}
2819
2820static ssize_t group_show(struct device *dev, struct device_attribute *attr,
2821 char *buf)
2822{
2823 struct tun_struct *tun = netdev_priv(to_net_dev(dev));
2824 return gid_valid(tun->group) ?
2825 sysfs_emit(buf, "%u\n",
2826 from_kgid_munged(current_user_ns(), tun->group)) :
2827 sysfs_emit(buf, "-1\n");
2828}
2829
2830static DEVICE_ATTR_RO(tun_flags);
2831static DEVICE_ATTR_RO(owner);
2832static DEVICE_ATTR_RO(group);
2833
2834static struct attribute *tun_dev_attrs[] = {
2835 &dev_attr_tun_flags.attr,
2836 &dev_attr_owner.attr,
2837 &dev_attr_group.attr,
2838 NULL
2839};
2840
2841static const struct attribute_group tun_attr_group = {
2842 .attrs = tun_dev_attrs
2843};
2844
2845static int tun_set_iff(struct net *net, struct file *file, struct ifreq *ifr)
2846{
2847 struct tun_struct *tun;
2848 struct tun_file *tfile = file->private_data;
2849 struct tun_file *ntfile;
2850 struct net_device *dev;
2851 int err, i;
2852
2853 if (tfile->detached)
2854 return -EINVAL;
2855
2856 if ((ifr->ifr_flags & IFF_NAPI_FRAGS)) {
2857 if (!capable(CAP_NET_ADMIN))
2858 return -EPERM;
2859
2860 if (!(ifr->ifr_flags & IFF_NAPI) ||
2861 (ifr->ifr_flags & TUN_TYPE_MASK) != IFF_TAP)
2862 return -EINVAL;
2863 }
2864
2865 dev = __dev_get_by_name(net, ifr->ifr_name);
2866 if (dev) {
2867 if (ifr->ifr_flags & IFF_TUN_EXCL)
2868 return -EBUSY;
2869 if ((ifr->ifr_flags & IFF_TUN) && dev->netdev_ops == &tun_netdev_ops)
2870 tun = netdev_priv(dev);
2871 else if ((ifr->ifr_flags & IFF_TAP) && dev->netdev_ops == &tap_netdev_ops)
2872 tun = netdev_priv(dev);
2873 else
2874 return -EINVAL;
2875
2876 if (!!(ifr->ifr_flags & IFF_MULTI_QUEUE) !=
2877 !!(tun->flags & IFF_MULTI_QUEUE))
2878 return -EINVAL;
2879
2880 if (tun_not_capable(tun))
2881 return -EPERM;
2882 err = security_tun_dev_open(tun->security);
2883 if (err < 0)
2884 return err;
2885
2886 err = tun_attach(tun, file, ifr->ifr_flags & IFF_NOFILTER,
2887 ifr->ifr_flags & IFF_NAPI,
2888 ifr->ifr_flags & IFF_NAPI_FRAGS, true);
2889 if (err < 0)
2890 return err;
2891
2892 if (tun->flags & IFF_MULTI_QUEUE &&
2893 (tun->numqueues + tun->numdisabled > 1)) {
2894 /* One or more queue has already been attached, no need
2895 * to initialize the device again.
2896 */
2897 netdev_state_change(dev);
2898 return 0;
2899 }
2900
2901 WRITE_ONCE(tun->flags, (tun->flags & ~TUN_FEATURES) |
2902 (ifr->ifr_flags & TUN_FEATURES));
2903
2904 netdev_state_change(dev);
2905 } else {
2906 char *name;
2907 unsigned long flags = 0;
2908 int queues = ifr->ifr_flags & IFF_MULTI_QUEUE ?
2909 MAX_TAP_QUEUES : 1;
2910
2911 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
2912 return -EPERM;
2913 err = security_tun_dev_create();
2914 if (err < 0)
2915 return err;
2916
2917 /* Set dev type */
2918 if (ifr->ifr_flags & IFF_TUN) {
2919 /* TUN device */
2920 flags |= IFF_TUN;
2921 name = "tun%d";
2922 } else if (ifr->ifr_flags & IFF_TAP) {
2923 /* TAP device */
2924 flags |= IFF_TAP;
2925 name = "tap%d";
2926 } else
2927 return -EINVAL;
2928
2929 if (*ifr->ifr_name)
2930 name = ifr->ifr_name;
2931
2932 dev = alloc_netdev_mqs(sizeof(struct tun_struct), name,
2933 NET_NAME_UNKNOWN, tun_setup, queues,
2934 queues);
2935
2936 if (!dev)
2937 return -ENOMEM;
2938
2939 dev_net_set(dev, net);
2940 dev->rtnl_link_ops = &tun_link_ops;
2941 dev->ifindex = tfile->ifindex;
2942 dev->sysfs_groups[0] = &tun_attr_group;
2943
2944 tun = netdev_priv(dev);
2945 tun->dev = dev;
2946 tun->flags = flags;
2947 tun->txflt.count = 0;
2948 tun->vnet_hdr_sz = sizeof(struct virtio_net_hdr);
2949
2950 tun->align = NET_SKB_PAD;
2951 tun->filter_attached = false;
2952 tun->sndbuf = tfile->socket.sk->sk_sndbuf;
2953 tun->rx_batched = 0;
2954 RCU_INIT_POINTER(tun->steering_prog, NULL);
2955
2956 tun->ifr = ifr;
2957 tun->file = file;
2958
2959 tun_net_initialize(dev);
2960
2961 err = register_netdevice(tun->dev);
2962 if (err < 0) {
2963 free_netdev(dev);
2964 return err;
2965 }
2966 /* free_netdev() won't check refcnt, to avoid race
2967 * with dev_put() we need publish tun after registration.
2968 */
2969 rcu_assign_pointer(tfile->tun, tun);
2970 }
2971
2972 if (ifr->ifr_flags & IFF_NO_CARRIER)
2973 netif_carrier_off(tun->dev);
2974 else
2975 netif_carrier_on(tun->dev);
2976
2977 /* Make sure persistent devices do not get stuck in
2978 * xoff state.
2979 */
2980 for (i = 0; i < tun->numqueues; i++) {
2981 ntfile = rtnl_dereference(tun->tfiles[i]);
2982 tun_force_wake_queue(tun, ntfile);
2983 }
2984
2985 strscpy(ifr->ifr_name, tun->dev->name);
2986 return 0;
2987}
2988
2989static void tun_get_iff(struct tun_struct *tun, struct ifreq *ifr)
2990{
2991 strscpy(ifr->ifr_name, tun->dev->name);
2992
2993 ifr->ifr_flags = tun_flags(tun);
2994
2995}
2996
2997#define PLAIN_GSO (NETIF_F_GSO_UDP_L4 | NETIF_F_TSO | NETIF_F_TSO6)
2998
2999/* This is like a cut-down ethtool ops, except done via tun fd so no
3000 * privs required. */
3001static int set_offload(struct tun_struct *tun, unsigned long arg)
3002{
3003 netdev_features_t features = 0;
3004
3005 if (arg & TUN_F_CSUM) {
3006 features |= NETIF_F_HW_CSUM;
3007 arg &= ~TUN_F_CSUM;
3008
3009 if (arg & (TUN_F_TSO4|TUN_F_TSO6)) {
3010 if (arg & TUN_F_TSO_ECN) {
3011 features |= NETIF_F_TSO_ECN;
3012 arg &= ~TUN_F_TSO_ECN;
3013 }
3014 if (arg & TUN_F_TSO4)
3015 features |= NETIF_F_TSO;
3016 if (arg & TUN_F_TSO6)
3017 features |= NETIF_F_TSO6;
3018 arg &= ~(TUN_F_TSO4|TUN_F_TSO6);
3019 }
3020
3021 arg &= ~TUN_F_UFO;
3022
3023 /* TODO: for now USO4 and USO6 should work simultaneously */
3024 if (arg & TUN_F_USO4 && arg & TUN_F_USO6) {
3025 features |= NETIF_F_GSO_UDP_L4;
3026 arg &= ~(TUN_F_USO4 | TUN_F_USO6);
3027 }
3028
3029 /*
3030 * Tunnel offload is allowed only if some plain offload is
3031 * available, too.
3032 */
3033 if (features & PLAIN_GSO && arg & TUN_F_UDP_TUNNEL_GSO) {
3034 features |= NETIF_F_GSO_UDP_TUNNEL;
3035 if (arg & TUN_F_UDP_TUNNEL_GSO_CSUM)
3036 features |= NETIF_F_GSO_UDP_TUNNEL_CSUM;
3037 arg &= ~(TUN_F_UDP_TUNNEL_GSO |
3038 TUN_F_UDP_TUNNEL_GSO_CSUM);
3039 }
3040 }
3041
3042 /* This gives the user a way to test for new features in future by
3043 * trying to set them. */
3044 if (arg)
3045 return -EINVAL;
3046
3047 tun->set_features = features;
3048 tun->dev->wanted_features &= ~TUN_USER_FEATURES;
3049 tun->dev->wanted_features |= features;
3050 netdev_update_features(tun->dev);
3051
3052 return 0;
3053}
3054
3055static void tun_detach_filter(struct tun_struct *tun, int n)
3056{
3057 int i;
3058 struct tun_file *tfile;
3059
3060 for (i = 0; i < n; i++) {
3061 tfile = rtnl_dereference(tun->tfiles[i]);
3062 lock_sock(tfile->socket.sk);
3063 sk_detach_filter(tfile->socket.sk);
3064 release_sock(tfile->socket.sk);
3065 }
3066
3067 tun->filter_attached = false;
3068}
3069
3070static int tun_attach_filter(struct tun_struct *tun)
3071{
3072 int i, ret = 0;
3073 struct tun_file *tfile;
3074
3075 for (i = 0; i < tun->numqueues; i++) {
3076 tfile = rtnl_dereference(tun->tfiles[i]);
3077 lock_sock(tfile->socket.sk);
3078 ret = sk_attach_filter(&tun->fprog, tfile->socket.sk);
3079 release_sock(tfile->socket.sk);
3080 if (ret) {
3081 tun_detach_filter(tun, i);
3082 return ret;
3083 }
3084 }
3085
3086 tun->filter_attached = true;
3087 return ret;
3088}
3089
3090static void tun_set_sndbuf(struct tun_struct *tun)
3091{
3092 struct tun_file *tfile;
3093 int i;
3094
3095 for (i = 0; i < tun->numqueues; i++) {
3096 tfile = rtnl_dereference(tun->tfiles[i]);
3097 tfile->socket.sk->sk_sndbuf = tun->sndbuf;
3098 }
3099}
3100
3101static int tun_set_queue(struct file *file, struct ifreq *ifr)
3102{
3103 struct tun_file *tfile = file->private_data;
3104 struct tun_struct *tun;
3105 int ret = 0;
3106
3107 rtnl_lock();
3108
3109 if (ifr->ifr_flags & IFF_ATTACH_QUEUE) {
3110 tun = tfile->detached;
3111 if (!tun) {
3112 ret = -EINVAL;
3113 goto unlock;
3114 }
3115 ret = security_tun_dev_attach_queue(tun->security);
3116 if (ret < 0)
3117 goto unlock;
3118 ret = tun_attach(tun, file, false, tun->flags & IFF_NAPI,
3119 tun->flags & IFF_NAPI_FRAGS, true);
3120 } else if (ifr->ifr_flags & IFF_DETACH_QUEUE) {
3121 tun = rtnl_dereference(tfile->tun);
3122 if (!tun || !(tun->flags & IFF_MULTI_QUEUE) || tfile->detached)
3123 ret = -EINVAL;
3124 else
3125 __tun_detach(tfile, false);
3126 } else
3127 ret = -EINVAL;
3128
3129 if (ret >= 0)
3130 netdev_state_change(tun->dev);
3131
3132unlock:
3133 rtnl_unlock();
3134 return ret;
3135}
3136
3137static int tun_set_ebpf(struct tun_struct *tun, struct tun_prog __rcu **prog_p,
3138 void __user *data)
3139{
3140 struct bpf_prog *prog;
3141 int fd;
3142
3143 if (copy_from_user(&fd, data, sizeof(fd)))
3144 return -EFAULT;
3145
3146 if (fd == -1) {
3147 prog = NULL;
3148 } else {
3149 prog = bpf_prog_get_type(fd, BPF_PROG_TYPE_SOCKET_FILTER);
3150 if (IS_ERR(prog))
3151 return PTR_ERR(prog);
3152 }
3153
3154 return __tun_set_ebpf(tun, prog_p, prog);
3155}
3156
3157/* Return correct value for tun->dev->addr_len based on tun->dev->type. */
3158static unsigned char tun_get_addr_len(unsigned short type)
3159{
3160 switch (type) {
3161 case ARPHRD_IP6GRE:
3162 case ARPHRD_TUNNEL6:
3163 return sizeof(struct in6_addr);
3164 case ARPHRD_IPGRE:
3165 case ARPHRD_TUNNEL:
3166 case ARPHRD_SIT:
3167 return 4;
3168 case ARPHRD_ETHER:
3169 return ETH_ALEN;
3170 case ARPHRD_IEEE802154:
3171 case ARPHRD_IEEE802154_MONITOR:
3172 return IEEE802154_EXTENDED_ADDR_LEN;
3173 case ARPHRD_PHONET_PIPE:
3174 case ARPHRD_PPP:
3175 case ARPHRD_NONE:
3176 return 0;
3177 case ARPHRD_6LOWPAN:
3178 return EUI64_ADDR_LEN;
3179 case ARPHRD_FDDI:
3180 return FDDI_K_ALEN;
3181 case ARPHRD_HIPPI:
3182 return HIPPI_ALEN;
3183 case ARPHRD_IEEE802:
3184 return FC_ALEN;
3185 case ARPHRD_ROSE:
3186 return ROSE_ADDR_LEN;
3187 case ARPHRD_NETROM:
3188 return AX25_ADDR_LEN;
3189 case ARPHRD_LOCALTLK:
3190 return LTALK_ALEN;
3191 default:
3192 return 0;
3193 }
3194}
3195
3196static long __tun_chr_ioctl(struct file *file, unsigned int cmd,
3197 unsigned long arg, int ifreq_len)
3198{
3199 struct tun_file *tfile = file->private_data;
3200 struct net *net = sock_net(&tfile->sk);
3201 struct tun_struct *tun;
3202 void __user* argp = (void __user*)arg;
3203 unsigned int carrier;
3204 struct ifreq ifr;
3205 kuid_t owner;
3206 kgid_t group;
3207 int ifindex;
3208 int sndbuf;
3209 int ret;
3210 bool do_notify = false;
3211
3212 if (cmd == TUNSETIFF || cmd == TUNSETQUEUE ||
3213 (_IOC_TYPE(cmd) == SOCK_IOC_TYPE && cmd != SIOCGSKNS)) {
3214 if (copy_from_user(&ifr, argp, ifreq_len))
3215 return -EFAULT;
3216 } else {
3217 memset(&ifr, 0, sizeof(ifr));
3218 }
3219 if (cmd == TUNGETFEATURES) {
3220 /* Currently this just means: "what IFF flags are valid?".
3221 * This is needed because we never checked for invalid flags on
3222 * TUNSETIFF.
3223 */
3224 return put_user(IFF_TUN | IFF_TAP | IFF_NO_CARRIER |
3225 TUN_FEATURES, (unsigned int __user*)argp);
3226 } else if (cmd == TUNSETQUEUE) {
3227 return tun_set_queue(file, &ifr);
3228 } else if (cmd == SIOCGSKNS) {
3229 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
3230 return -EPERM;
3231 return open_related_ns(&net->ns, get_net_ns);
3232 }
3233
3234 rtnl_lock();
3235
3236 tun = tun_get(tfile);
3237 if (cmd == TUNSETIFF) {
3238 ret = -EEXIST;
3239 if (tun)
3240 goto unlock;
3241
3242 ifr.ifr_name[IFNAMSIZ-1] = '\0';
3243
3244 ret = tun_set_iff(net, file, &ifr);
3245
3246 if (ret)
3247 goto unlock;
3248
3249 if (copy_to_user(argp, &ifr, ifreq_len))
3250 ret = -EFAULT;
3251 goto unlock;
3252 }
3253 if (cmd == TUNSETIFINDEX) {
3254 ret = -EPERM;
3255 if (tun)
3256 goto unlock;
3257
3258 ret = -EFAULT;
3259 if (copy_from_user(&ifindex, argp, sizeof(ifindex)))
3260 goto unlock;
3261 ret = -EINVAL;
3262 if (ifindex < 0)
3263 goto unlock;
3264 ret = 0;
3265 tfile->ifindex = ifindex;
3266 goto unlock;
3267 }
3268
3269 ret = -EBADFD;
3270 if (!tun)
3271 goto unlock;
3272
3273 netif_info(tun, drv, tun->dev, "tun_chr_ioctl cmd %u\n", cmd);
3274
3275 net = dev_net(tun->dev);
3276 ret = 0;
3277 switch (cmd) {
3278 case TUNGETIFF:
3279 tun_get_iff(tun, &ifr);
3280
3281 if (tfile->detached)
3282 ifr.ifr_flags |= IFF_DETACH_QUEUE;
3283 if (!tfile->socket.sk->sk_filter)
3284 ifr.ifr_flags |= IFF_NOFILTER;
3285
3286 if (copy_to_user(argp, &ifr, ifreq_len))
3287 ret = -EFAULT;
3288 break;
3289
3290 case TUNSETNOCSUM:
3291 /* Disable/Enable checksum */
3292
3293 /* [unimplemented] */
3294 netif_info(tun, drv, tun->dev, "ignored: set checksum %s\n",
3295 arg ? "disabled" : "enabled");
3296 break;
3297
3298 case TUNSETPERSIST:
3299 /* Disable/Enable persist mode. Keep an extra reference to the
3300 * module to prevent the module being unprobed.
3301 */
3302 if (arg && !(READ_ONCE(tun->flags) & IFF_PERSIST)) {
3303 WRITE_ONCE(tun->flags, READ_ONCE(tun->flags) | IFF_PERSIST);
3304 __module_get(THIS_MODULE);
3305 do_notify = true;
3306 }
3307 if (!arg && (READ_ONCE(tun->flags) & IFF_PERSIST)) {
3308 WRITE_ONCE(tun->flags, READ_ONCE(tun->flags) & ~IFF_PERSIST);
3309 module_put(THIS_MODULE);
3310 do_notify = true;
3311 }
3312
3313 netif_info(tun, drv, tun->dev, "persist %s\n",
3314 arg ? "enabled" : "disabled");
3315 break;
3316
3317 case TUNSETOWNER:
3318 /* Set owner of the device */
3319 owner = make_kuid(current_user_ns(), arg);
3320 if (!uid_valid(owner)) {
3321 ret = -EINVAL;
3322 break;
3323 }
3324 WRITE_ONCE(tun->owner, owner);
3325 do_notify = true;
3326 netif_info(tun, drv, tun->dev, "owner set to %u\n",
3327 from_kuid(&init_user_ns, owner));
3328 break;
3329
3330 case TUNSETGROUP:
3331 /* Set group of the device */
3332 group = make_kgid(current_user_ns(), arg);
3333 if (!gid_valid(group)) {
3334 ret = -EINVAL;
3335 break;
3336 }
3337 WRITE_ONCE(tun->group, group);
3338 do_notify = true;
3339 netif_info(tun, drv, tun->dev, "group set to %u\n",
3340 from_kgid(&init_user_ns, group));
3341 break;
3342
3343 case TUNSETLINK:
3344 /* Only allow setting the type when the interface is down */
3345 if (tun->dev->flags & IFF_UP) {
3346 netif_info(tun, drv, tun->dev,
3347 "Linktype set failed because interface is up\n");
3348 ret = -EBUSY;
3349 } else {
3350 ret = call_netdevice_notifiers(NETDEV_PRE_TYPE_CHANGE,
3351 tun->dev);
3352 ret = notifier_to_errno(ret);
3353 if (ret) {
3354 netif_info(tun, drv, tun->dev,
3355 "Refused to change device type\n");
3356 break;
3357 }
3358 tun->dev->type = (int) arg;
3359 tun->dev->addr_len = tun_get_addr_len(tun->dev->type);
3360 netif_info(tun, drv, tun->dev, "linktype set to %d\n",
3361 tun->dev->type);
3362 call_netdevice_notifiers(NETDEV_POST_TYPE_CHANGE,
3363 tun->dev);
3364 }
3365 break;
3366
3367 case TUNSETDEBUG:
3368 tun->msg_enable = (u32)arg;
3369 break;
3370
3371 case TUNSETOFFLOAD:
3372 ret = set_offload(tun, arg);
3373 break;
3374
3375 case TUNSETTXFILTER:
3376 /* Can be set only for TAPs */
3377 ret = -EINVAL;
3378 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3379 break;
3380 ret = update_filter(&tun->txflt, (void __user *)arg);
3381 break;
3382
3383 case SIOCGIFHWADDR:
3384 /* Get hw address */
3385 netif_get_mac_address(&ifr.ifr_hwaddr, net, tun->dev->name);
3386 if (copy_to_user(argp, &ifr, ifreq_len))
3387 ret = -EFAULT;
3388 break;
3389
3390 case SIOCSIFHWADDR:
3391 /* Set hw address */
3392 if (tun->dev->addr_len > sizeof(ifr.ifr_hwaddr)) {
3393 ret = -EINVAL;
3394 break;
3395 }
3396 ret = dev_set_mac_address_user(tun->dev,
3397 (struct sockaddr_storage *)&ifr.ifr_hwaddr,
3398 NULL);
3399 break;
3400
3401 case TUNGETSNDBUF:
3402 sndbuf = tfile->socket.sk->sk_sndbuf;
3403 if (copy_to_user(argp, &sndbuf, sizeof(sndbuf)))
3404 ret = -EFAULT;
3405 break;
3406
3407 case TUNSETSNDBUF:
3408 if (copy_from_user(&sndbuf, argp, sizeof(sndbuf))) {
3409 ret = -EFAULT;
3410 break;
3411 }
3412 if (sndbuf <= 0) {
3413 ret = -EINVAL;
3414 break;
3415 }
3416
3417 tun->sndbuf = sndbuf;
3418 tun_set_sndbuf(tun);
3419 break;
3420
3421 case TUNATTACHFILTER:
3422 /* Can be set only for TAPs */
3423 ret = -EINVAL;
3424 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3425 break;
3426 ret = -EFAULT;
3427 if (copy_from_user(&tun->fprog, argp, sizeof(tun->fprog)))
3428 break;
3429
3430 ret = tun_attach_filter(tun);
3431 break;
3432
3433 case TUNDETACHFILTER:
3434 /* Can be set only for TAPs */
3435 ret = -EINVAL;
3436 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3437 break;
3438 ret = 0;
3439 tun_detach_filter(tun, tun->numqueues);
3440 break;
3441
3442 case TUNGETFILTER:
3443 ret = -EINVAL;
3444 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3445 break;
3446 ret = -EFAULT;
3447 if (copy_to_user(argp, &tun->fprog, sizeof(tun->fprog)))
3448 break;
3449 ret = 0;
3450 break;
3451
3452 case TUNSETSTEERINGEBPF:
3453 ret = tun_set_ebpf(tun, &tun->steering_prog, argp);
3454 break;
3455
3456 case TUNSETFILTEREBPF:
3457 ret = tun_set_ebpf(tun, &tun->filter_prog, argp);
3458 break;
3459
3460 case TUNSETCARRIER:
3461 ret = -EFAULT;
3462 if (copy_from_user(&carrier, argp, sizeof(carrier)))
3463 goto unlock;
3464
3465 ret = tun_net_change_carrier(tun->dev, (bool)carrier);
3466 break;
3467
3468 case TUNGETDEVNETNS:
3469 ret = -EPERM;
3470 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
3471 goto unlock;
3472 ret = open_related_ns(&net->ns, get_net_ns);
3473 break;
3474
3475 default:
3476 ret = tun_vnet_ioctl(&tun->vnet_hdr_sz, &tun->flags, cmd, argp);
3477 break;
3478 }
3479
3480 if (do_notify)
3481 netdev_state_change(tun->dev);
3482
3483unlock:
3484 rtnl_unlock();
3485 if (tun)
3486 tun_put(tun);
3487 return ret;
3488}
3489
3490static long tun_chr_ioctl(struct file *file,
3491 unsigned int cmd, unsigned long arg)
3492{
3493 return __tun_chr_ioctl(file, cmd, arg, sizeof (struct ifreq));
3494}
3495
3496#ifdef CONFIG_COMPAT
3497static long tun_chr_compat_ioctl(struct file *file,
3498 unsigned int cmd, unsigned long arg)
3499{
3500 switch (cmd) {
3501 case TUNSETIFF:
3502 case TUNGETIFF:
3503 case TUNSETTXFILTER:
3504 case TUNGETSNDBUF:
3505 case TUNSETSNDBUF:
3506 case SIOCGIFHWADDR:
3507 case SIOCSIFHWADDR:
3508 arg = (unsigned long)compat_ptr(arg);
3509 break;
3510 default:
3511 arg = (compat_ulong_t)arg;
3512 break;
3513 }
3514
3515 /*
3516 * compat_ifreq is shorter than ifreq, so we must not access beyond
3517 * the end of that structure. All fields that are used in this
3518 * driver are compatible though, we don't need to convert the
3519 * contents.
3520 */
3521 return __tun_chr_ioctl(file, cmd, arg, sizeof(struct compat_ifreq));
3522}
3523#endif /* CONFIG_COMPAT */
3524
3525static int tun_chr_fasync(int fd, struct file *file, int on)
3526{
3527 struct tun_file *tfile = file->private_data;
3528 int ret;
3529
3530 if (on) {
3531 ret = file_f_owner_allocate(file);
3532 if (ret)
3533 goto out;
3534 }
3535
3536 if ((ret = fasync_helper(fd, file, on, &tfile->fasync)) < 0)
3537 goto out;
3538
3539 if (on) {
3540 __f_setown(file, task_pid(current), PIDTYPE_TGID, 0);
3541 tfile->flags |= TUN_FASYNC;
3542 } else
3543 tfile->flags &= ~TUN_FASYNC;
3544 ret = 0;
3545out:
3546 return ret;
3547}
3548
3549static int tun_chr_open(struct inode *inode, struct file * file)
3550{
3551 struct net *net = current->nsproxy->net_ns;
3552 struct tun_file *tfile;
3553
3554 tfile = (struct tun_file *)sk_alloc(net, AF_UNSPEC, GFP_KERNEL,
3555 &tun_proto, 0);
3556 if (!tfile)
3557 return -ENOMEM;
3558 if (ptr_ring_init(&tfile->tx_ring, 0, GFP_KERNEL)) {
3559 sk_free(&tfile->sk);
3560 return -ENOMEM;
3561 }
3562
3563 mutex_init(&tfile->napi_mutex);
3564 RCU_INIT_POINTER(tfile->tun, NULL);
3565 tfile->flags = 0;
3566 tfile->ifindex = 0;
3567
3568 init_waitqueue_head(&tfile->socket.wq.wait);
3569
3570 tfile->socket.file = file;
3571 tfile->socket.ops = &tun_socket_ops;
3572
3573 sock_init_data_uid(&tfile->socket, &tfile->sk, current_fsuid());
3574
3575 tfile->sk.sk_write_space = tun_sock_write_space;
3576 tfile->sk.sk_sndbuf = INT_MAX;
3577
3578 file->private_data = tfile;
3579 INIT_LIST_HEAD(&tfile->next);
3580
3581 sock_set_flag(&tfile->sk, SOCK_ZEROCOPY);
3582
3583 /* tun groks IOCB_NOWAIT just fine, mark it as such */
3584 file->f_mode |= FMODE_NOWAIT;
3585 return 0;
3586}
3587
3588static int tun_chr_close(struct inode *inode, struct file *file)
3589{
3590 struct tun_file *tfile = file->private_data;
3591
3592 tun_detach(tfile, true);
3593
3594 return 0;
3595}
3596
3597#ifdef CONFIG_PROC_FS
3598static void tun_chr_show_fdinfo(struct seq_file *m, struct file *file)
3599{
3600 struct tun_file *tfile = file->private_data;
3601 struct tun_struct *tun;
3602 struct ifreq ifr;
3603
3604 memset(&ifr, 0, sizeof(ifr));
3605
3606 rtnl_lock();
3607 tun = tun_get(tfile);
3608 if (tun)
3609 tun_get_iff(tun, &ifr);
3610 rtnl_unlock();
3611
3612 if (tun)
3613 tun_put(tun);
3614
3615 seq_printf(m, "iff:\t%s\n", ifr.ifr_name);
3616}
3617#endif
3618
3619static const struct file_operations tun_fops = {
3620 .owner = THIS_MODULE,
3621 .read_iter = tun_chr_read_iter,
3622 .write_iter = tun_chr_write_iter,
3623 .poll = tun_chr_poll,
3624 .unlocked_ioctl = tun_chr_ioctl,
3625#ifdef CONFIG_COMPAT
3626 .compat_ioctl = tun_chr_compat_ioctl,
3627#endif
3628 .open = tun_chr_open,
3629 .release = tun_chr_close,
3630 .fasync = tun_chr_fasync,
3631#ifdef CONFIG_PROC_FS
3632 .show_fdinfo = tun_chr_show_fdinfo,
3633#endif
3634};
3635
3636static struct miscdevice tun_miscdev = {
3637 .minor = TUN_MINOR,
3638 .name = "tun",
3639 .nodename = "net/tun",
3640 .fops = &tun_fops,
3641};
3642
3643/* ethtool interface */
3644
3645static void tun_default_link_ksettings(struct net_device *dev,
3646 struct ethtool_link_ksettings *cmd)
3647{
3648 ethtool_link_ksettings_zero_link_mode(cmd, supported);
3649 ethtool_link_ksettings_zero_link_mode(cmd, advertising);
3650 cmd->base.speed = SPEED_10000;
3651 cmd->base.duplex = DUPLEX_FULL;
3652 cmd->base.port = PORT_TP;
3653 cmd->base.phy_address = 0;
3654 cmd->base.autoneg = AUTONEG_DISABLE;
3655}
3656
3657static int tun_get_link_ksettings(struct net_device *dev,
3658 struct ethtool_link_ksettings *cmd)
3659{
3660 struct tun_struct *tun = netdev_priv(dev);
3661
3662 memcpy(cmd, &tun->link_ksettings, sizeof(*cmd));
3663 return 0;
3664}
3665
3666static int tun_set_link_ksettings(struct net_device *dev,
3667 const struct ethtool_link_ksettings *cmd)
3668{
3669 struct tun_struct *tun = netdev_priv(dev);
3670
3671 memcpy(&tun->link_ksettings, cmd, sizeof(*cmd));
3672 return 0;
3673}
3674
3675static void tun_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
3676{
3677 struct tun_struct *tun = netdev_priv(dev);
3678
3679 strscpy(info->driver, DRV_NAME, sizeof(info->driver));
3680 strscpy(info->version, DRV_VERSION, sizeof(info->version));
3681
3682 switch (tun->flags & TUN_TYPE_MASK) {
3683 case IFF_TUN:
3684 strscpy(info->bus_info, "tun", sizeof(info->bus_info));
3685 break;
3686 case IFF_TAP:
3687 strscpy(info->bus_info, "tap", sizeof(info->bus_info));
3688 break;
3689 }
3690}
3691
3692static u32 tun_get_msglevel(struct net_device *dev)
3693{
3694 struct tun_struct *tun = netdev_priv(dev);
3695
3696 return tun->msg_enable;
3697}
3698
3699static void tun_set_msglevel(struct net_device *dev, u32 value)
3700{
3701 struct tun_struct *tun = netdev_priv(dev);
3702
3703 tun->msg_enable = value;
3704}
3705
3706static int tun_get_coalesce(struct net_device *dev,
3707 struct ethtool_coalesce *ec,
3708 struct kernel_ethtool_coalesce *kernel_coal,
3709 struct netlink_ext_ack *extack)
3710{
3711 struct tun_struct *tun = netdev_priv(dev);
3712
3713 ec->rx_max_coalesced_frames = tun->rx_batched;
3714
3715 return 0;
3716}
3717
3718static int tun_set_coalesce(struct net_device *dev,
3719 struct ethtool_coalesce *ec,
3720 struct kernel_ethtool_coalesce *kernel_coal,
3721 struct netlink_ext_ack *extack)
3722{
3723 struct tun_struct *tun = netdev_priv(dev);
3724
3725 if (ec->rx_max_coalesced_frames > NAPI_POLL_WEIGHT)
3726 tun->rx_batched = NAPI_POLL_WEIGHT;
3727 else
3728 tun->rx_batched = ec->rx_max_coalesced_frames;
3729
3730 return 0;
3731}
3732
3733static void tun_get_channels(struct net_device *dev,
3734 struct ethtool_channels *channels)
3735{
3736 struct tun_struct *tun = netdev_priv(dev);
3737
3738 channels->combined_count = tun->numqueues;
3739 channels->max_combined = tun->flags & IFF_MULTI_QUEUE ? MAX_TAP_QUEUES : 1;
3740}
3741
3742static const struct ethtool_ops tun_ethtool_ops = {
3743 .supported_coalesce_params = ETHTOOL_COALESCE_RX_MAX_FRAMES,
3744 .get_drvinfo = tun_get_drvinfo,
3745 .get_msglevel = tun_get_msglevel,
3746 .set_msglevel = tun_set_msglevel,
3747 .get_link = ethtool_op_get_link,
3748 .get_channels = tun_get_channels,
3749 .get_ts_info = ethtool_op_get_ts_info,
3750 .get_coalesce = tun_get_coalesce,
3751 .set_coalesce = tun_set_coalesce,
3752 .get_link_ksettings = tun_get_link_ksettings,
3753 .set_link_ksettings = tun_set_link_ksettings,
3754};
3755
3756static int tun_queue_resize(struct tun_struct *tun)
3757{
3758 struct net_device *dev = tun->dev;
3759 struct tun_file *tfile;
3760 struct ptr_ring **rings;
3761 int n = tun->numqueues + tun->numdisabled;
3762 int ret, i;
3763
3764 rings = kmalloc_objs(*rings, n);
3765 if (!rings)
3766 return -ENOMEM;
3767
3768 for (i = 0; i < tun->numqueues; i++) {
3769 tfile = rtnl_dereference(tun->tfiles[i]);
3770 rings[i] = &tfile->tx_ring;
3771 }
3772 list_for_each_entry(tfile, &tun->disabled, next)
3773 rings[i++] = &tfile->tx_ring;
3774
3775 ret = ptr_ring_resize_multiple_bh(rings, n,
3776 dev->tx_queue_len, GFP_KERNEL,
3777 tun_ptr_free);
3778
3779 if (!ret) {
3780 for (i = 0; i < tun->numqueues; i++) {
3781 tfile = rtnl_dereference(tun->tfiles[i]);
3782 tun_force_wake_queue(tun, tfile);
3783 }
3784 }
3785
3786 kfree(rings);
3787 return ret;
3788}
3789
3790static int tun_device_event(struct notifier_block *unused,
3791 unsigned long event, void *ptr)
3792{
3793 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
3794 struct tun_struct *tun = netdev_priv(dev);
3795 int i;
3796
3797 if (dev->rtnl_link_ops != &tun_link_ops)
3798 return NOTIFY_DONE;
3799
3800 switch (event) {
3801 case NETDEV_CHANGE_TX_QUEUE_LEN:
3802 if (tun_queue_resize(tun))
3803 return NOTIFY_BAD;
3804 break;
3805 case NETDEV_UP:
3806 for (i = 0; i < tun->numqueues; i++) {
3807 struct tun_file *tfile;
3808
3809 tfile = rtnl_dereference(tun->tfiles[i]);
3810 tfile->socket.sk->sk_write_space(tfile->socket.sk);
3811 }
3812 break;
3813 default:
3814 break;
3815 }
3816
3817 return NOTIFY_DONE;
3818}
3819
3820static struct notifier_block tun_notifier_block __read_mostly = {
3821 .notifier_call = tun_device_event,
3822};
3823
3824static int __init tun_init(void)
3825{
3826 int ret = 0;
3827
3828 pr_info("%s, %s\n", DRV_DESCRIPTION, DRV_VERSION);
3829
3830 ret = rtnl_link_register(&tun_link_ops);
3831 if (ret) {
3832 pr_err("Can't register link_ops\n");
3833 goto err_linkops;
3834 }
3835
3836 ret = misc_register(&tun_miscdev);
3837 if (ret) {
3838 pr_err("Can't register misc device %d\n", TUN_MINOR);
3839 goto err_misc;
3840 }
3841
3842 ret = register_netdevice_notifier(&tun_notifier_block);
3843 if (ret) {
3844 pr_err("Can't register netdevice notifier\n");
3845 goto err_notifier;
3846 }
3847
3848 return 0;
3849
3850err_notifier:
3851 misc_deregister(&tun_miscdev);
3852err_misc:
3853 rtnl_link_unregister(&tun_link_ops);
3854err_linkops:
3855 return ret;
3856}
3857
3858static void __exit tun_cleanup(void)
3859{
3860 misc_deregister(&tun_miscdev);
3861 rtnl_link_unregister(&tun_link_ops);
3862 unregister_netdevice_notifier(&tun_notifier_block);
3863}
3864
3865/* Get an underlying socket object from tun file. Returns error unless file is
3866 * attached to a device. The returned object works like a packet socket, it
3867 * can be used for sock_sendmsg/sock_recvmsg. The caller is responsible for
3868 * holding a reference to the file for as long as the socket is in use. */
3869struct socket *tun_get_socket(struct file *file)
3870{
3871 struct tun_file *tfile;
3872 if (file->f_op != &tun_fops)
3873 return ERR_PTR(-EINVAL);
3874 tfile = file->private_data;
3875 if (!tfile)
3876 return ERR_PTR(-EBADFD);
3877 return &tfile->socket;
3878}
3879EXPORT_SYMBOL_GPL(tun_get_socket);
3880
3881struct ptr_ring *tun_get_tx_ring(struct file *file)
3882{
3883 struct tun_file *tfile;
3884
3885 if (file->f_op != &tun_fops)
3886 return ERR_PTR(-EINVAL);
3887 tfile = file->private_data;
3888 if (!tfile)
3889 return ERR_PTR(-EBADFD);
3890 return &tfile->tx_ring;
3891}
3892EXPORT_SYMBOL_GPL(tun_get_tx_ring);
3893
3894/* Callers must hold ring.consumer_lock */
3895void tun_wake_queue(struct file *file, int consumed)
3896{
3897 struct tun_file *tfile;
3898 struct tun_struct *tun;
3899
3900 if (file->f_op != &tun_fops)
3901 return;
3902
3903 tfile = file->private_data;
3904 if (!tfile)
3905 return;
3906
3907 lockdep_assert_held(&tfile->tx_ring.consumer_lock);
3908
3909 rcu_read_lock();
3910
3911 tun = rcu_dereference(tfile->tun);
3912 if (tun)
3913 __tun_wake_queue(tun, tfile, consumed);
3914
3915 rcu_read_unlock();
3916}
3917EXPORT_SYMBOL_GPL(tun_wake_queue);
3918
3919module_init(tun_init);
3920module_exit(tun_cleanup);
3921MODULE_DESCRIPTION(DRV_DESCRIPTION);
3922MODULE_AUTHOR(DRV_COPYRIGHT);
3923MODULE_LICENSE("GPL");
3924MODULE_ALIAS_MISCDEV(TUN_MINOR);
3925MODULE_ALIAS("devname:net/tun");
3926MODULE_IMPORT_NS("NETDEV_INTERNAL");