1/*
2 * TUN - Universal TUN/TAP device driver.
3 * Copyright (C) 1999-2002 Maxim Krasnyansky <maxk@qualcomm.com>
4 *
5 * This program is free software; you can redistribute it and/or modify
6 * it under the terms of the GNU General Public License as published by
7 * the Free Software Foundation; either version 2 of the License, or
8 * (at your option) any later version.
9 *
10 * This program is distributed in the hope that it will be useful,
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
13 * GNU General Public License for more details.
14 *
15 * $Id: tun.c,v 1.15 2002/03/01 02:44:24 maxk Exp $
16 */
17
18/*
19 * Changes:
20 *
21 * Mike Kershaw <dragorn@kismetwireless.net> 2005/08/14
22 * Add TUNSETLINK ioctl to set the link encapsulation
23 *
24 * Mark Smith <markzzzsmith@yahoo.com.au>
25 * Use eth_random_addr() for tap MAC address.
26 *
27 * Harald Roelle <harald.roelle@ifi.lmu.de> 2004/04/20
28 * Fixes in packet dropping, queue length setting and queue wakeup.
29 * Increased default tx queue length.
30 * Added ethtool API.
31 * Minor cleanups
32 *
33 * Daniel Podlejski <underley@underley.eu.org>
34 * Modifications for 2.3.99-pre5 kernel.
35 */
36
37#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
38
39#define DRV_NAME "tun"
40#define DRV_VERSION "1.6"
41#define DRV_DESCRIPTION "Universal TUN/TAP device driver"
42#define DRV_COPYRIGHT "(C) 1999-2004 Max Krasnyansky <maxk@qualcomm.com>"
43
44#include <linux/module.h>
45#include <linux/errno.h>
46#include <linux/kernel.h>
47#include <linux/sched/signal.h>
48#include <linux/major.h>
49#include <linux/slab.h>
50#include <linux/poll.h>
51#include <linux/fcntl.h>
52#include <linux/init.h>
53#include <linux/skbuff.h>
54#include <linux/netdevice.h>
55#include <linux/etherdevice.h>
56#include <linux/miscdevice.h>
57#include <linux/ethtool.h>
58#include <linux/rtnetlink.h>
59#include <linux/compat.h>
60#include <linux/if.h>
61#include <linux/if_arp.h>
62#include <linux/if_ether.h>
63#include <linux/if_tun.h>
64#include <linux/if_vlan.h>
65#include <linux/crc32.h>
66#include <linux/nsproxy.h>
67#include <linux/virtio_net.h>
68#include <linux/rcupdate.h>
69#include <net/net_namespace.h>
70#include <net/netns/generic.h>
71#include <net/rtnetlink.h>
72#include <net/sock.h>
73#include <net/xdp.h>
74#include <linux/seq_file.h>
75#include <linux/uio.h>
76#include <linux/skb_array.h>
77#include <linux/bpf.h>
78#include <linux/bpf_trace.h>
79#include <linux/mutex.h>
80
81#include <linux/uaccess.h>
82#include <linux/proc_fs.h>
83
84static void tun_default_link_ksettings(struct net_device *dev,
85 struct ethtool_link_ksettings *cmd);
86
87/* Uncomment to enable debugging */
88/* #define TUN_DEBUG 1 */
89
90#ifdef TUN_DEBUG
91static int debug;
92
93#define tun_debug(level, tun, fmt, args...) \
94do { \
95 if (tun->debug) \
96 netdev_printk(level, tun->dev, fmt, ##args); \
97} while (0)
98#define DBG1(level, fmt, args...) \
99do { \
100 if (debug == 2) \
101 printk(level fmt, ##args); \
102} while (0)
103#else
104#define tun_debug(level, tun, fmt, args...) \
105do { \
106 if (0) \
107 netdev_printk(level, tun->dev, fmt, ##args); \
108} while (0)
109#define DBG1(level, fmt, args...) \
110do { \
111 if (0) \
112 printk(level fmt, ##args); \
113} while (0)
114#endif
115
116#define TUN_RX_PAD (NET_IP_ALIGN + NET_SKB_PAD)
117
118/* TUN device flags */
119
120/* IFF_ATTACH_QUEUE is never stored in device flags,
121 * overload it to mean fasync when stored there.
122 */
123#define TUN_FASYNC IFF_ATTACH_QUEUE
124/* High bits in flags field are unused. */
125#define TUN_VNET_LE 0x80000000
126#define TUN_VNET_BE 0x40000000
127
128#define TUN_FEATURES (IFF_NO_PI | IFF_ONE_QUEUE | IFF_VNET_HDR | \
129 IFF_MULTI_QUEUE | IFF_NAPI | IFF_NAPI_FRAGS)
130
131#define GOODCOPY_LEN 128
132
133#define FLT_EXACT_COUNT 8
134struct tap_filter {
135 unsigned int count; /* Number of addrs. Zero means disabled */
136 u32 mask[2]; /* Mask of the hashed addrs */
137 unsigned char addr[FLT_EXACT_COUNT][ETH_ALEN];
138};
139
140/* MAX_TAP_QUEUES 256 is chosen to allow rx/tx queues to be equal
141 * to max number of VCPUs in guest. */
142#define MAX_TAP_QUEUES 256
143#define MAX_TAP_FLOWS 4096
144
145#define TUN_FLOW_EXPIRE (3 * HZ)
146
147struct tun_pcpu_stats {
148 u64 rx_packets;
149 u64 rx_bytes;
150 u64 tx_packets;
151 u64 tx_bytes;
152 struct u64_stats_sync syncp;
153 u32 rx_dropped;
154 u32 tx_dropped;
155 u32 rx_frame_errors;
156};
157
158/* A tun_file connects an open character device to a tuntap netdevice. It
159 * also contains all socket related structures (except sock_fprog and tap_filter)
160 * to serve as one transmit queue for tuntap device. The sock_fprog and
161 * tap_filter were kept in tun_struct since they were used for filtering for the
162 * netdevice not for a specific queue (at least I didn't see the requirement for
163 * this).
164 *
165 * RCU usage:
166 * The tun_file and tun_struct are loosely coupled, the pointer from one to the
167 * other can only be read while rcu_read_lock or rtnl_lock is held.
168 */
169struct tun_file {
170 struct sock sk;
171 struct socket socket;
172 struct socket_wq wq;
173 struct tun_struct __rcu *tun;
174 struct fasync_struct *fasync;
175 /* only used for fasnyc */
176 unsigned int flags;
177 union {
178 u16 queue_index;
179 unsigned int ifindex;
180 };
181 struct napi_struct napi;
182 bool napi_enabled;
183 bool napi_frags_enabled;
184 struct mutex napi_mutex; /* Protects access to the above napi */
185 struct list_head next;
186 struct tun_struct *detached;
187 struct ptr_ring tx_ring;
188 struct xdp_rxq_info xdp_rxq;
189};
190
191struct tun_page {
192 struct page *page;
193 int count;
194};
195
196struct tun_flow_entry {
197 struct hlist_node hash_link;
198 struct rcu_head rcu;
199 struct tun_struct *tun;
200
201 u32 rxhash;
202 u32 rps_rxhash;
203 int queue_index;
204 unsigned long updated ____cacheline_aligned_in_smp;
205};
206
207#define TUN_NUM_FLOW_ENTRIES 1024
208#define TUN_MASK_FLOW_ENTRIES (TUN_NUM_FLOW_ENTRIES - 1)
209
210struct tun_prog {
211 struct rcu_head rcu;
212 struct bpf_prog *prog;
213};
214
215/* Since the socket were moved to tun_file, to preserve the behavior of persist
216 * device, socket filter, sndbuf and vnet header size were restore when the
217 * file were attached to a persist device.
218 */
219struct tun_struct {
220 struct tun_file __rcu *tfiles[MAX_TAP_QUEUES];
221 unsigned int numqueues;
222 unsigned int flags;
223 kuid_t owner;
224 kgid_t group;
225
226 struct net_device *dev;
227 netdev_features_t set_features;
228#define TUN_USER_FEATURES (NETIF_F_HW_CSUM|NETIF_F_TSO_ECN|NETIF_F_TSO| \
229 NETIF_F_TSO6)
230
231 int align;
232 int vnet_hdr_sz;
233 int sndbuf;
234 struct tap_filter txflt;
235 struct sock_fprog fprog;
236 /* protected by rtnl lock */
237 bool filter_attached;
238#ifdef TUN_DEBUG
239 int debug;
240#endif
241 spinlock_t lock;
242 struct hlist_head flows[TUN_NUM_FLOW_ENTRIES];
243 struct timer_list flow_gc_timer;
244 unsigned long ageing_time;
245 unsigned int numdisabled;
246 struct list_head disabled;
247 void *security;
248 u32 flow_count;
249 u32 rx_batched;
250 struct tun_pcpu_stats __percpu *pcpu_stats;
251 struct bpf_prog __rcu *xdp_prog;
252 struct tun_prog __rcu *steering_prog;
253 struct tun_prog __rcu *filter_prog;
254 struct ethtool_link_ksettings link_ksettings;
255};
256
257struct veth {
258 __be16 h_vlan_proto;
259 __be16 h_vlan_TCI;
260};
261
262bool tun_is_xdp_frame(void *ptr)
263{
264 return (unsigned long)ptr & TUN_XDP_FLAG;
265}
266EXPORT_SYMBOL(tun_is_xdp_frame);
267
268void *tun_xdp_to_ptr(void *ptr)
269{
270 return (void *)((unsigned long)ptr | TUN_XDP_FLAG);
271}
272EXPORT_SYMBOL(tun_xdp_to_ptr);
273
274void *tun_ptr_to_xdp(void *ptr)
275{
276 return (void *)((unsigned long)ptr & ~TUN_XDP_FLAG);
277}
278EXPORT_SYMBOL(tun_ptr_to_xdp);
279
280static int tun_napi_receive(struct napi_struct *napi, int budget)
281{
282 struct tun_file *tfile = container_of(napi, struct tun_file, napi);
283 struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
284 struct sk_buff_head process_queue;
285 struct sk_buff *skb;
286 int received = 0;
287
288 __skb_queue_head_init(&process_queue);
289
290 spin_lock(&queue->lock);
291 skb_queue_splice_tail_init(queue, &process_queue);
292 spin_unlock(&queue->lock);
293
294 while (received < budget && (skb = __skb_dequeue(&process_queue))) {
295 napi_gro_receive(napi, skb);
296 ++received;
297 }
298
299 if (!skb_queue_empty(&process_queue)) {
300 spin_lock(&queue->lock);
301 skb_queue_splice(&process_queue, queue);
302 spin_unlock(&queue->lock);
303 }
304
305 return received;
306}
307
308static int tun_napi_poll(struct napi_struct *napi, int budget)
309{
310 unsigned int received;
311
312 received = tun_napi_receive(napi, budget);
313
314 if (received < budget)
315 napi_complete_done(napi, received);
316
317 return received;
318}
319
320static void tun_napi_init(struct tun_struct *tun, struct tun_file *tfile,
321 bool napi_en, bool napi_frags)
322{
323 tfile->napi_enabled = napi_en;
324 tfile->napi_frags_enabled = napi_en && napi_frags;
325 if (napi_en) {
326 netif_napi_add(tun->dev, &tfile->napi, tun_napi_poll,
327 NAPI_POLL_WEIGHT);
328 napi_enable(&tfile->napi);
329 }
330}
331
332static void tun_napi_disable(struct tun_file *tfile)
333{
334 if (tfile->napi_enabled)
335 napi_disable(&tfile->napi);
336}
337
338static void tun_napi_del(struct tun_file *tfile)
339{
340 if (tfile->napi_enabled)
341 netif_napi_del(&tfile->napi);
342}
343
344static bool tun_napi_frags_enabled(const struct tun_file *tfile)
345{
346 return tfile->napi_frags_enabled;
347}
348
349#ifdef CONFIG_TUN_VNET_CROSS_LE
350static inline bool tun_legacy_is_little_endian(struct tun_struct *tun)
351{
352 return tun->flags & TUN_VNET_BE ? false :
353 virtio_legacy_is_little_endian();
354}
355
356static long tun_get_vnet_be(struct tun_struct *tun, int __user *argp)
357{
358 int be = !!(tun->flags & TUN_VNET_BE);
359
360 if (put_user(be, argp))
361 return -EFAULT;
362
363 return 0;
364}
365
366static long tun_set_vnet_be(struct tun_struct *tun, int __user *argp)
367{
368 int be;
369
370 if (get_user(be, argp))
371 return -EFAULT;
372
373 if (be)
374 tun->flags |= TUN_VNET_BE;
375 else
376 tun->flags &= ~TUN_VNET_BE;
377
378 return 0;
379}
380#else
381static inline bool tun_legacy_is_little_endian(struct tun_struct *tun)
382{
383 return virtio_legacy_is_little_endian();
384}
385
386static long tun_get_vnet_be(struct tun_struct *tun, int __user *argp)
387{
388 return -EINVAL;
389}
390
391static long tun_set_vnet_be(struct tun_struct *tun, int __user *argp)
392{
393 return -EINVAL;
394}
395#endif /* CONFIG_TUN_VNET_CROSS_LE */
396
397static inline bool tun_is_little_endian(struct tun_struct *tun)
398{
399 return tun->flags & TUN_VNET_LE ||
400 tun_legacy_is_little_endian(tun);
401}
402
403static inline u16 tun16_to_cpu(struct tun_struct *tun, __virtio16 val)
404{
405 return __virtio16_to_cpu(tun_is_little_endian(tun), val);
406}
407
408static inline __virtio16 cpu_to_tun16(struct tun_struct *tun, u16 val)
409{
410 return __cpu_to_virtio16(tun_is_little_endian(tun), val);
411}
412
413static inline u32 tun_hashfn(u32 rxhash)
414{
415 return rxhash & TUN_MASK_FLOW_ENTRIES;
416}
417
418static struct tun_flow_entry *tun_flow_find(struct hlist_head *head, u32 rxhash)
419{
420 struct tun_flow_entry *e;
421
422 hlist_for_each_entry_rcu(e, head, hash_link) {
423 if (e->rxhash == rxhash)
424 return e;
425 }
426 return NULL;
427}
428
429static struct tun_flow_entry *tun_flow_create(struct tun_struct *tun,
430 struct hlist_head *head,
431 u32 rxhash, u16 queue_index)
432{
433 struct tun_flow_entry *e = kmalloc(sizeof(*e), GFP_ATOMIC);
434
435 if (e) {
436 tun_debug(KERN_INFO, tun, "create flow: hash %u index %u\n",
437 rxhash, queue_index);
438 e->updated = jiffies;
439 e->rxhash = rxhash;
440 e->rps_rxhash = 0;
441 e->queue_index = queue_index;
442 e->tun = tun;
443 hlist_add_head_rcu(&e->hash_link, head);
444 ++tun->flow_count;
445 }
446 return e;
447}
448
449static void tun_flow_delete(struct tun_struct *tun, struct tun_flow_entry *e)
450{
451 tun_debug(KERN_INFO, tun, "delete flow: hash %u index %u\n",
452 e->rxhash, e->queue_index);
453 hlist_del_rcu(&e->hash_link);
454 kfree_rcu(e, rcu);
455 --tun->flow_count;
456}
457
458static void tun_flow_flush(struct tun_struct *tun)
459{
460 int i;
461
462 spin_lock_bh(&tun->lock);
463 for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
464 struct tun_flow_entry *e;
465 struct hlist_node *n;
466
467 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link)
468 tun_flow_delete(tun, e);
469 }
470 spin_unlock_bh(&tun->lock);
471}
472
473static void tun_flow_delete_by_queue(struct tun_struct *tun, u16 queue_index)
474{
475 int i;
476
477 spin_lock_bh(&tun->lock);
478 for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
479 struct tun_flow_entry *e;
480 struct hlist_node *n;
481
482 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
483 if (e->queue_index == queue_index)
484 tun_flow_delete(tun, e);
485 }
486 }
487 spin_unlock_bh(&tun->lock);
488}
489
490static void tun_flow_cleanup(struct timer_list *t)
491{
492 struct tun_struct *tun = from_timer(tun, t, flow_gc_timer);
493 unsigned long delay = tun->ageing_time;
494 unsigned long next_timer = jiffies + delay;
495 unsigned long count = 0;
496 int i;
497
498 tun_debug(KERN_INFO, tun, "tun_flow_cleanup\n");
499
500 spin_lock(&tun->lock);
501 for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
502 struct tun_flow_entry *e;
503 struct hlist_node *n;
504
505 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
506 unsigned long this_timer;
507
508 this_timer = e->updated + delay;
509 if (time_before_eq(this_timer, jiffies)) {
510 tun_flow_delete(tun, e);
511 continue;
512 }
513 count++;
514 if (time_before(this_timer, next_timer))
515 next_timer = this_timer;
516 }
517 }
518
519 if (count)
520 mod_timer(&tun->flow_gc_timer, round_jiffies_up(next_timer));
521 spin_unlock(&tun->lock);
522}
523
524static void tun_flow_update(struct tun_struct *tun, u32 rxhash,
525 struct tun_file *tfile)
526{
527 struct hlist_head *head;
528 struct tun_flow_entry *e;
529 unsigned long delay = tun->ageing_time;
530 u16 queue_index = tfile->queue_index;
531
532 head = &tun->flows[tun_hashfn(rxhash)];
533
534 rcu_read_lock();
535
536 e = tun_flow_find(head, rxhash);
537 if (likely(e)) {
538 /* TODO: keep queueing to old queue until it's empty? */
539 if (e->queue_index != queue_index)
540 e->queue_index = queue_index;
541 if (e->updated != jiffies)
542 e->updated = jiffies;
543 sock_rps_record_flow_hash(e->rps_rxhash);
544 } else {
545 spin_lock_bh(&tun->lock);
546 if (!tun_flow_find(head, rxhash) &&
547 tun->flow_count < MAX_TAP_FLOWS)
548 tun_flow_create(tun, head, rxhash, queue_index);
549
550 if (!timer_pending(&tun->flow_gc_timer))
551 mod_timer(&tun->flow_gc_timer,
552 round_jiffies_up(jiffies + delay));
553 spin_unlock_bh(&tun->lock);
554 }
555
556 rcu_read_unlock();
557}
558
559/**
560 * Save the hash received in the stack receive path and update the
561 * flow_hash table accordingly.
562 */
563static inline void tun_flow_save_rps_rxhash(struct tun_flow_entry *e, u32 hash)
564{
565 if (unlikely(e->rps_rxhash != hash))
566 e->rps_rxhash = hash;
567}
568
569/* We try to identify a flow through its rxhash. The reason that
570 * we do not check rxq no. is because some cards(e.g 82599), chooses
571 * the rxq based on the txq where the last packet of the flow comes. As
572 * the userspace application move between processors, we may get a
573 * different rxq no. here.
574 */
575static u16 tun_automq_select_queue(struct tun_struct *tun, struct sk_buff *skb)
576{
577 struct tun_flow_entry *e;
578 u32 txq = 0;
579 u32 numqueues = 0;
580
581 numqueues = READ_ONCE(tun->numqueues);
582
583 txq = __skb_get_hash_symmetric(skb);
584 e = tun_flow_find(&tun->flows[tun_hashfn(txq)], txq);
585 if (e) {
586 tun_flow_save_rps_rxhash(e, txq);
587 txq = e->queue_index;
588 } else {
589 /* use multiply and shift instead of expensive divide */
590 txq = ((u64)txq * numqueues) >> 32;
591 }
592
593 return txq;
594}
595
596static u16 tun_ebpf_select_queue(struct tun_struct *tun, struct sk_buff *skb)
597{
598 struct tun_prog *prog;
599 u16 ret = 0;
600
601 prog = rcu_dereference(tun->steering_prog);
602 if (prog)
603 ret = bpf_prog_run_clear_cb(prog->prog, skb);
604
605 return ret % tun->numqueues;
606}
607
608static u16 tun_select_queue(struct net_device *dev, struct sk_buff *skb,
609 struct net_device *sb_dev,
610 select_queue_fallback_t fallback)
611{
612 struct tun_struct *tun = netdev_priv(dev);
613 u16 ret;
614
615 rcu_read_lock();
616 if (rcu_dereference(tun->steering_prog))
617 ret = tun_ebpf_select_queue(tun, skb);
618 else
619 ret = tun_automq_select_queue(tun, skb);
620 rcu_read_unlock();
621
622 return ret;
623}
624
625static inline bool tun_not_capable(struct tun_struct *tun)
626{
627 const struct cred *cred = current_cred();
628 struct net *net = dev_net(tun->dev);
629
630 return ((uid_valid(tun->owner) && !uid_eq(cred->euid, tun->owner)) ||
631 (gid_valid(tun->group) && !in_egroup_p(tun->group))) &&
632 !ns_capable(net->user_ns, CAP_NET_ADMIN);
633}
634
635static void tun_set_real_num_queues(struct tun_struct *tun)
636{
637 netif_set_real_num_tx_queues(tun->dev, tun->numqueues);
638 netif_set_real_num_rx_queues(tun->dev, tun->numqueues);
639}
640
641static void tun_disable_queue(struct tun_struct *tun, struct tun_file *tfile)
642{
643 tfile->detached = tun;
644 list_add_tail(&tfile->next, &tun->disabled);
645 ++tun->numdisabled;
646}
647
648static struct tun_struct *tun_enable_queue(struct tun_file *tfile)
649{
650 struct tun_struct *tun = tfile->detached;
651
652 tfile->detached = NULL;
653 list_del_init(&tfile->next);
654 --tun->numdisabled;
655 return tun;
656}
657
658void tun_ptr_free(void *ptr)
659{
660 if (!ptr)
661 return;
662 if (tun_is_xdp_frame(ptr)) {
663 struct xdp_frame *xdpf = tun_ptr_to_xdp(ptr);
664
665 xdp_return_frame(xdpf);
666 } else {
667 __skb_array_destroy_skb(ptr);
668 }
669}
670EXPORT_SYMBOL_GPL(tun_ptr_free);
671
672static void tun_queue_purge(struct tun_file *tfile)
673{
674 void *ptr;
675
676 while ((ptr = ptr_ring_consume(&tfile->tx_ring)) != NULL)
677 tun_ptr_free(ptr);
678
679 skb_queue_purge(&tfile->sk.sk_write_queue);
680 skb_queue_purge(&tfile->sk.sk_error_queue);
681}
682
683static void __tun_detach(struct tun_file *tfile, bool clean)
684{
685 struct tun_file *ntfile;
686 struct tun_struct *tun;
687
688 tun = rtnl_dereference(tfile->tun);
689
690 if (tun && clean) {
691 tun_napi_disable(tfile);
692 tun_napi_del(tfile);
693 }
694
695 if (tun && !tfile->detached) {
696 u16 index = tfile->queue_index;
697 BUG_ON(index >= tun->numqueues);
698
699 rcu_assign_pointer(tun->tfiles[index],
700 tun->tfiles[tun->numqueues - 1]);
701 ntfile = rtnl_dereference(tun->tfiles[index]);
702 ntfile->queue_index = index;
703
704 --tun->numqueues;
705 if (clean) {
706 RCU_INIT_POINTER(tfile->tun, NULL);
707 sock_put(&tfile->sk);
708 } else
709 tun_disable_queue(tun, tfile);
710
711 synchronize_net();
712 tun_flow_delete_by_queue(tun, tun->numqueues + 1);
713 /* Drop read queue */
714 tun_queue_purge(tfile);
715 tun_set_real_num_queues(tun);
716 } else if (tfile->detached && clean) {
717 tun = tun_enable_queue(tfile);
718 sock_put(&tfile->sk);
719 }
720
721 if (clean) {
722 if (tun && tun->numqueues == 0 && tun->numdisabled == 0) {
723 netif_carrier_off(tun->dev);
724
725 if (!(tun->flags & IFF_PERSIST) &&
726 tun->dev->reg_state == NETREG_REGISTERED)
727 unregister_netdevice(tun->dev);
728 }
729 if (tun)
730 xdp_rxq_info_unreg(&tfile->xdp_rxq);
731 ptr_ring_cleanup(&tfile->tx_ring, tun_ptr_free);
732 sock_put(&tfile->sk);
733 }
734}
735
736static void tun_detach(struct tun_file *tfile, bool clean)
737{
738 struct tun_struct *tun;
739 struct net_device *dev;
740
741 rtnl_lock();
742 tun = rtnl_dereference(tfile->tun);
743 dev = tun ? tun->dev : NULL;
744 __tun_detach(tfile, clean);
745 if (dev)
746 netdev_state_change(dev);
747 rtnl_unlock();
748}
749
750static void tun_detach_all(struct net_device *dev)
751{
752 struct tun_struct *tun = netdev_priv(dev);
753 struct tun_file *tfile, *tmp;
754 int i, n = tun->numqueues;
755
756 for (i = 0; i < n; i++) {
757 tfile = rtnl_dereference(tun->tfiles[i]);
758 BUG_ON(!tfile);
759 tun_napi_disable(tfile);
760 tfile->socket.sk->sk_shutdown = RCV_SHUTDOWN;
761 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
762 RCU_INIT_POINTER(tfile->tun, NULL);
763 --tun->numqueues;
764 }
765 list_for_each_entry(tfile, &tun->disabled, next) {
766 tfile->socket.sk->sk_shutdown = RCV_SHUTDOWN;
767 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
768 RCU_INIT_POINTER(tfile->tun, NULL);
769 }
770 BUG_ON(tun->numqueues != 0);
771
772 synchronize_net();
773 for (i = 0; i < n; i++) {
774 tfile = rtnl_dereference(tun->tfiles[i]);
775 tun_napi_del(tfile);
776 /* Drop read queue */
777 tun_queue_purge(tfile);
778 xdp_rxq_info_unreg(&tfile->xdp_rxq);
779 sock_put(&tfile->sk);
780 }
781 list_for_each_entry_safe(tfile, tmp, &tun->disabled, next) {
782 tun_enable_queue(tfile);
783 tun_queue_purge(tfile);
784 xdp_rxq_info_unreg(&tfile->xdp_rxq);
785 sock_put(&tfile->sk);
786 }
787 BUG_ON(tun->numdisabled != 0);
788
789 if (tun->flags & IFF_PERSIST)
790 module_put(THIS_MODULE);
791}
792
793static int tun_attach(struct tun_struct *tun, struct file *file,
794 bool skip_filter, bool napi, bool napi_frags)
795{
796 struct tun_file *tfile = file->private_data;
797 struct net_device *dev = tun->dev;
798 int err;
799
800 err = security_tun_dev_attach(tfile->socket.sk, tun->security);
801 if (err < 0)
802 goto out;
803
804 err = -EINVAL;
805 if (rtnl_dereference(tfile->tun) && !tfile->detached)
806 goto out;
807
808 err = -EBUSY;
809 if (!(tun->flags & IFF_MULTI_QUEUE) && tun->numqueues == 1)
810 goto out;
811
812 err = -E2BIG;
813 if (!tfile->detached &&
814 tun->numqueues + tun->numdisabled == MAX_TAP_QUEUES)
815 goto out;
816
817 err = 0;
818
819 /* Re-attach the filter to persist device */
820 if (!skip_filter && (tun->filter_attached == true)) {
821 lock_sock(tfile->socket.sk);
822 err = sk_attach_filter(&tun->fprog, tfile->socket.sk);
823 release_sock(tfile->socket.sk);
824 if (!err)
825 goto out;
826 }
827
828 if (!tfile->detached &&
829 ptr_ring_resize(&tfile->tx_ring, dev->tx_queue_len,
830 GFP_KERNEL, tun_ptr_free)) {
831 err = -ENOMEM;
832 goto out;
833 }
834
835 tfile->queue_index = tun->numqueues;
836 tfile->socket.sk->sk_shutdown &= ~RCV_SHUTDOWN;
837
838 if (tfile->detached) {
839 /* Re-attach detached tfile, updating XDP queue_index */
840 WARN_ON(!xdp_rxq_info_is_reg(&tfile->xdp_rxq));
841
842 if (tfile->xdp_rxq.queue_index != tfile->queue_index)
843 tfile->xdp_rxq.queue_index = tfile->queue_index;
844 } else {
845 /* Setup XDP RX-queue info, for new tfile getting attached */
846 err = xdp_rxq_info_reg(&tfile->xdp_rxq,
847 tun->dev, tfile->queue_index);
848 if (err < 0)
849 goto out;
850 err = xdp_rxq_info_reg_mem_model(&tfile->xdp_rxq,
851 MEM_TYPE_PAGE_SHARED, NULL);
852 if (err < 0) {
853 xdp_rxq_info_unreg(&tfile->xdp_rxq);
854 goto out;
855 }
856 err = 0;
857 }
858
859 if (tfile->detached) {
860 tun_enable_queue(tfile);
861 } else {
862 sock_hold(&tfile->sk);
863 tun_napi_init(tun, tfile, napi, napi_frags);
864 }
865
866 if (rtnl_dereference(tun->xdp_prog))
867 sock_set_flag(&tfile->sk, SOCK_XDP);
868
869 /* device is allowed to go away first, so no need to hold extra
870 * refcnt.
871 */
872
873 /* Publish tfile->tun and tun->tfiles only after we've fully
874 * initialized tfile; otherwise we risk using half-initialized
875 * object.
876 */
877 rcu_assign_pointer(tfile->tun, tun);
878 rcu_assign_pointer(tun->tfiles[tun->numqueues], tfile);
879 tun->numqueues++;
880 tun_set_real_num_queues(tun);
881out:
882 return err;
883}
884
885static struct tun_struct *tun_get(struct tun_file *tfile)
886{
887 struct tun_struct *tun;
888
889 rcu_read_lock();
890 tun = rcu_dereference(tfile->tun);
891 if (tun)
892 dev_hold(tun->dev);
893 rcu_read_unlock();
894
895 return tun;
896}
897
898static void tun_put(struct tun_struct *tun)
899{
900 dev_put(tun->dev);
901}
902
903/* TAP filtering */
904static void addr_hash_set(u32 *mask, const u8 *addr)
905{
906 int n = ether_crc(ETH_ALEN, addr) >> 26;
907 mask[n >> 5] |= (1 << (n & 31));
908}
909
910static unsigned int addr_hash_test(const u32 *mask, const u8 *addr)
911{
912 int n = ether_crc(ETH_ALEN, addr) >> 26;
913 return mask[n >> 5] & (1 << (n & 31));
914}
915
916static int update_filter(struct tap_filter *filter, void __user *arg)
917{
918 struct { u8 u[ETH_ALEN]; } *addr;
919 struct tun_filter uf;
920 int err, alen, n, nexact;
921
922 if (copy_from_user(&uf, arg, sizeof(uf)))
923 return -EFAULT;
924
925 if (!uf.count) {
926 /* Disabled */
927 filter->count = 0;
928 return 0;
929 }
930
931 alen = ETH_ALEN * uf.count;
932 addr = memdup_user(arg + sizeof(uf), alen);
933 if (IS_ERR(addr))
934 return PTR_ERR(addr);
935
936 /* The filter is updated without holding any locks. Which is
937 * perfectly safe. We disable it first and in the worst
938 * case we'll accept a few undesired packets. */
939 filter->count = 0;
940 wmb();
941
942 /* Use first set of addresses as an exact filter */
943 for (n = 0; n < uf.count && n < FLT_EXACT_COUNT; n++)
944 memcpy(filter->addr[n], addr[n].u, ETH_ALEN);
945
946 nexact = n;
947
948 /* Remaining multicast addresses are hashed,
949 * unicast will leave the filter disabled. */
950 memset(filter->mask, 0, sizeof(filter->mask));
951 for (; n < uf.count; n++) {
952 if (!is_multicast_ether_addr(addr[n].u)) {
953 err = 0; /* no filter */
954 goto free_addr;
955 }
956 addr_hash_set(filter->mask, addr[n].u);
957 }
958
959 /* For ALLMULTI just set the mask to all ones.
960 * This overrides the mask populated above. */
961 if ((uf.flags & TUN_FLT_ALLMULTI))
962 memset(filter->mask, ~0, sizeof(filter->mask));
963
964 /* Now enable the filter */
965 wmb();
966 filter->count = nexact;
967
968 /* Return the number of exact filters */
969 err = nexact;
970free_addr:
971 kfree(addr);
972 return err;
973}
974
975/* Returns: 0 - drop, !=0 - accept */
976static int run_filter(struct tap_filter *filter, const struct sk_buff *skb)
977{
978 /* Cannot use eth_hdr(skb) here because skb_mac_hdr() is incorrect
979 * at this point. */
980 struct ethhdr *eh = (struct ethhdr *) skb->data;
981 int i;
982
983 /* Exact match */
984 for (i = 0; i < filter->count; i++)
985 if (ether_addr_equal(eh->h_dest, filter->addr[i]))
986 return 1;
987
988 /* Inexact match (multicast only) */
989 if (is_multicast_ether_addr(eh->h_dest))
990 return addr_hash_test(filter->mask, eh->h_dest);
991
992 return 0;
993}
994
995/*
996 * Checks whether the packet is accepted or not.
997 * Returns: 0 - drop, !=0 - accept
998 */
999static int check_filter(struct tap_filter *filter, const struct sk_buff *skb)
1000{
1001 if (!filter->count)
1002 return 1;
1003
1004 return run_filter(filter, skb);
1005}
1006
1007/* Network device part of the driver */
1008
1009static const struct ethtool_ops tun_ethtool_ops;
1010
1011/* Net device detach from fd. */
1012static void tun_net_uninit(struct net_device *dev)
1013{
1014 tun_detach_all(dev);
1015}
1016
1017/* Net device open. */
1018static int tun_net_open(struct net_device *dev)
1019{
1020 struct tun_struct *tun = netdev_priv(dev);
1021 int i;
1022
1023 netif_tx_start_all_queues(dev);
1024
1025 for (i = 0; i < tun->numqueues; i++) {
1026 struct tun_file *tfile;
1027
1028 tfile = rtnl_dereference(tun->tfiles[i]);
1029 tfile->socket.sk->sk_write_space(tfile->socket.sk);
1030 }
1031
1032 return 0;
1033}
1034
1035/* Net device close. */
1036static int tun_net_close(struct net_device *dev)
1037{
1038 netif_tx_stop_all_queues(dev);
1039 return 0;
1040}
1041
1042/* Net device start xmit */
1043static void tun_automq_xmit(struct tun_struct *tun, struct sk_buff *skb)
1044{
1045#ifdef CONFIG_RPS
1046 if (tun->numqueues == 1 && static_key_false(&rps_needed)) {
1047 /* Select queue was not called for the skbuff, so we extract the
1048 * RPS hash and save it into the flow_table here.
1049 */
1050 struct tun_flow_entry *e;
1051 __u32 rxhash;
1052
1053 rxhash = __skb_get_hash_symmetric(skb);
1054 e = tun_flow_find(&tun->flows[tun_hashfn(rxhash)], rxhash);
1055 if (e)
1056 tun_flow_save_rps_rxhash(e, rxhash);
1057 }
1058#endif
1059}
1060
1061static unsigned int run_ebpf_filter(struct tun_struct *tun,
1062 struct sk_buff *skb,
1063 int len)
1064{
1065 struct tun_prog *prog = rcu_dereference(tun->filter_prog);
1066
1067 if (prog)
1068 len = bpf_prog_run_clear_cb(prog->prog, skb);
1069
1070 return len;
1071}
1072
1073/* Net device start xmit */
1074static netdev_tx_t tun_net_xmit(struct sk_buff *skb, struct net_device *dev)
1075{
1076 struct tun_struct *tun = netdev_priv(dev);
1077 int txq = skb->queue_mapping;
1078 struct tun_file *tfile;
1079 int len = skb->len;
1080
1081 rcu_read_lock();
1082 tfile = rcu_dereference(tun->tfiles[txq]);
1083
1084 /* Drop packet if interface is not attached */
1085 if (txq >= tun->numqueues)
1086 goto drop;
1087
1088 if (!rcu_dereference(tun->steering_prog))
1089 tun_automq_xmit(tun, skb);
1090
1091 tun_debug(KERN_INFO, tun, "tun_net_xmit %d\n", skb->len);
1092
1093 BUG_ON(!tfile);
1094
1095 /* Drop if the filter does not like it.
1096 * This is a noop if the filter is disabled.
1097 * Filter can be enabled only for the TAP devices. */
1098 if (!check_filter(&tun->txflt, skb))
1099 goto drop;
1100
1101 if (tfile->socket.sk->sk_filter &&
1102 sk_filter(tfile->socket.sk, skb))
1103 goto drop;
1104
1105 len = run_ebpf_filter(tun, skb, len);
1106 if (len == 0 || pskb_trim(skb, len))
1107 goto drop;
1108
1109 if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC)))
1110 goto drop;
1111
1112 skb_tx_timestamp(skb);
1113
1114 /* Orphan the skb - required as we might hang on to it
1115 * for indefinite time.
1116 */
1117 skb_orphan(skb);
1118
1119 nf_reset(skb);
1120
1121 if (ptr_ring_produce(&tfile->tx_ring, skb))
1122 goto drop;
1123
1124 /* Notify and wake up reader process */
1125 if (tfile->flags & TUN_FASYNC)
1126 kill_fasync(&tfile->fasync, SIGIO, POLL_IN);
1127 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
1128
1129 rcu_read_unlock();
1130 return NETDEV_TX_OK;
1131
1132drop:
1133 this_cpu_inc(tun->pcpu_stats->tx_dropped);
1134 skb_tx_error(skb);
1135 kfree_skb(skb);
1136 rcu_read_unlock();
1137 return NET_XMIT_DROP;
1138}
1139
1140static void tun_net_mclist(struct net_device *dev)
1141{
1142 /*
1143 * This callback is supposed to deal with mc filter in
1144 * _rx_ path and has nothing to do with the _tx_ path.
1145 * In rx path we always accept everything userspace gives us.
1146 */
1147}
1148
1149static netdev_features_t tun_net_fix_features(struct net_device *dev,
1150 netdev_features_t features)
1151{
1152 struct tun_struct *tun = netdev_priv(dev);
1153
1154 return (features & tun->set_features) | (features & ~TUN_USER_FEATURES);
1155}
1156
1157static void tun_set_headroom(struct net_device *dev, int new_hr)
1158{
1159 struct tun_struct *tun = netdev_priv(dev);
1160
1161 if (new_hr < NET_SKB_PAD)
1162 new_hr = NET_SKB_PAD;
1163
1164 tun->align = new_hr;
1165}
1166
1167static void
1168tun_net_get_stats64(struct net_device *dev, struct rtnl_link_stats64 *stats)
1169{
1170 u32 rx_dropped = 0, tx_dropped = 0, rx_frame_errors = 0;
1171 struct tun_struct *tun = netdev_priv(dev);
1172 struct tun_pcpu_stats *p;
1173 int i;
1174
1175 for_each_possible_cpu(i) {
1176 u64 rxpackets, rxbytes, txpackets, txbytes;
1177 unsigned int start;
1178
1179 p = per_cpu_ptr(tun->pcpu_stats, i);
1180 do {
1181 start = u64_stats_fetch_begin(&p->syncp);
1182 rxpackets = p->rx_packets;
1183 rxbytes = p->rx_bytes;
1184 txpackets = p->tx_packets;
1185 txbytes = p->tx_bytes;
1186 } while (u64_stats_fetch_retry(&p->syncp, start));
1187
1188 stats->rx_packets += rxpackets;
1189 stats->rx_bytes += rxbytes;
1190 stats->tx_packets += txpackets;
1191 stats->tx_bytes += txbytes;
1192
1193 /* u32 counters */
1194 rx_dropped += p->rx_dropped;
1195 rx_frame_errors += p->rx_frame_errors;
1196 tx_dropped += p->tx_dropped;
1197 }
1198 stats->rx_dropped = rx_dropped;
1199 stats->rx_frame_errors = rx_frame_errors;
1200 stats->tx_dropped = tx_dropped;
1201}
1202
1203static int tun_xdp_set(struct net_device *dev, struct bpf_prog *prog,
1204 struct netlink_ext_ack *extack)
1205{
1206 struct tun_struct *tun = netdev_priv(dev);
1207 struct tun_file *tfile;
1208 struct bpf_prog *old_prog;
1209 int i;
1210
1211 old_prog = rtnl_dereference(tun->xdp_prog);
1212 rcu_assign_pointer(tun->xdp_prog, prog);
1213 if (old_prog)
1214 bpf_prog_put(old_prog);
1215
1216 for (i = 0; i < tun->numqueues; i++) {
1217 tfile = rtnl_dereference(tun->tfiles[i]);
1218 if (prog)
1219 sock_set_flag(&tfile->sk, SOCK_XDP);
1220 else
1221 sock_reset_flag(&tfile->sk, SOCK_XDP);
1222 }
1223 list_for_each_entry(tfile, &tun->disabled, next) {
1224 if (prog)
1225 sock_set_flag(&tfile->sk, SOCK_XDP);
1226 else
1227 sock_reset_flag(&tfile->sk, SOCK_XDP);
1228 }
1229
1230 return 0;
1231}
1232
1233static u32 tun_xdp_query(struct net_device *dev)
1234{
1235 struct tun_struct *tun = netdev_priv(dev);
1236 const struct bpf_prog *xdp_prog;
1237
1238 xdp_prog = rtnl_dereference(tun->xdp_prog);
1239 if (xdp_prog)
1240 return xdp_prog->aux->id;
1241
1242 return 0;
1243}
1244
1245static int tun_xdp(struct net_device *dev, struct netdev_bpf *xdp)
1246{
1247 switch (xdp->command) {
1248 case XDP_SETUP_PROG:
1249 return tun_xdp_set(dev, xdp->prog, xdp->extack);
1250 case XDP_QUERY_PROG:
1251 xdp->prog_id = tun_xdp_query(dev);
1252 return 0;
1253 default:
1254 return -EINVAL;
1255 }
1256}
1257
1258static int tun_net_change_carrier(struct net_device *dev, bool new_carrier)
1259{
1260 if (new_carrier) {
1261 struct tun_struct *tun = netdev_priv(dev);
1262
1263 if (!tun->numqueues)
1264 return -EPERM;
1265
1266 netif_carrier_on(dev);
1267 } else {
1268 netif_carrier_off(dev);
1269 }
1270 return 0;
1271}
1272
1273static const struct net_device_ops tun_netdev_ops = {
1274 .ndo_uninit = tun_net_uninit,
1275 .ndo_open = tun_net_open,
1276 .ndo_stop = tun_net_close,
1277 .ndo_start_xmit = tun_net_xmit,
1278 .ndo_fix_features = tun_net_fix_features,
1279 .ndo_select_queue = tun_select_queue,
1280 .ndo_set_rx_headroom = tun_set_headroom,
1281 .ndo_get_stats64 = tun_net_get_stats64,
1282 .ndo_change_carrier = tun_net_change_carrier,
1283};
1284
1285static void __tun_xdp_flush_tfile(struct tun_file *tfile)
1286{
1287 /* Notify and wake up reader process */
1288 if (tfile->flags & TUN_FASYNC)
1289 kill_fasync(&tfile->fasync, SIGIO, POLL_IN);
1290 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
1291}
1292
1293static int tun_xdp_xmit(struct net_device *dev, int n,
1294 struct xdp_frame **frames, u32 flags)
1295{
1296 struct tun_struct *tun = netdev_priv(dev);
1297 struct tun_file *tfile;
1298 u32 numqueues;
1299 int drops = 0;
1300 int cnt = n;
1301 int i;
1302
1303 if (unlikely(flags & ~XDP_XMIT_FLAGS_MASK))
1304 return -EINVAL;
1305
1306 rcu_read_lock();
1307
1308 numqueues = READ_ONCE(tun->numqueues);
1309 if (!numqueues) {
1310 rcu_read_unlock();
1311 return -ENXIO; /* Caller will free/return all frames */
1312 }
1313
1314 tfile = rcu_dereference(tun->tfiles[smp_processor_id() %
1315 numqueues]);
1316
1317 spin_lock(&tfile->tx_ring.producer_lock);
1318 for (i = 0; i < n; i++) {
1319 struct xdp_frame *xdp = frames[i];
1320 /* Encode the XDP flag into lowest bit for consumer to differ
1321 * XDP buffer from sk_buff.
1322 */
1323 void *frame = tun_xdp_to_ptr(xdp);
1324
1325 if (__ptr_ring_produce(&tfile->tx_ring, frame)) {
1326 this_cpu_inc(tun->pcpu_stats->tx_dropped);
1327 xdp_return_frame_rx_napi(xdp);
1328 drops++;
1329 }
1330 }
1331 spin_unlock(&tfile->tx_ring.producer_lock);
1332
1333 if (flags & XDP_XMIT_FLUSH)
1334 __tun_xdp_flush_tfile(tfile);
1335
1336 rcu_read_unlock();
1337 return cnt - drops;
1338}
1339
1340static int tun_xdp_tx(struct net_device *dev, struct xdp_buff *xdp)
1341{
1342 struct xdp_frame *frame = convert_to_xdp_frame(xdp);
1343
1344 if (unlikely(!frame))
1345 return -EOVERFLOW;
1346
1347 return tun_xdp_xmit(dev, 1, &frame, XDP_XMIT_FLUSH);
1348}
1349
1350static const struct net_device_ops tap_netdev_ops = {
1351 .ndo_uninit = tun_net_uninit,
1352 .ndo_open = tun_net_open,
1353 .ndo_stop = tun_net_close,
1354 .ndo_start_xmit = tun_net_xmit,
1355 .ndo_fix_features = tun_net_fix_features,
1356 .ndo_set_rx_mode = tun_net_mclist,
1357 .ndo_set_mac_address = eth_mac_addr,
1358 .ndo_validate_addr = eth_validate_addr,
1359 .ndo_select_queue = tun_select_queue,
1360 .ndo_features_check = passthru_features_check,
1361 .ndo_set_rx_headroom = tun_set_headroom,
1362 .ndo_get_stats64 = tun_net_get_stats64,
1363 .ndo_bpf = tun_xdp,
1364 .ndo_xdp_xmit = tun_xdp_xmit,
1365 .ndo_change_carrier = tun_net_change_carrier,
1366};
1367
1368static void tun_flow_init(struct tun_struct *tun)
1369{
1370 int i;
1371
1372 for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++)
1373 INIT_HLIST_HEAD(&tun->flows[i]);
1374
1375 tun->ageing_time = TUN_FLOW_EXPIRE;
1376 timer_setup(&tun->flow_gc_timer, tun_flow_cleanup, 0);
1377 mod_timer(&tun->flow_gc_timer,
1378 round_jiffies_up(jiffies + tun->ageing_time));
1379}
1380
1381static void tun_flow_uninit(struct tun_struct *tun)
1382{
1383 del_timer_sync(&tun->flow_gc_timer);
1384 tun_flow_flush(tun);
1385}
1386
1387#define MIN_MTU 68
1388#define MAX_MTU 65535
1389
1390/* Initialize net device. */
1391static void tun_net_init(struct net_device *dev)
1392{
1393 struct tun_struct *tun = netdev_priv(dev);
1394
1395 switch (tun->flags & TUN_TYPE_MASK) {
1396 case IFF_TUN:
1397 dev->netdev_ops = &tun_netdev_ops;
1398
1399 /* Point-to-Point TUN Device */
1400 dev->hard_header_len = 0;
1401 dev->addr_len = 0;
1402 dev->mtu = 1500;
1403
1404 /* Zero header length */
1405 dev->type = ARPHRD_NONE;
1406 dev->flags = IFF_POINTOPOINT | IFF_NOARP | IFF_MULTICAST;
1407 break;
1408
1409 case IFF_TAP:
1410 dev->netdev_ops = &tap_netdev_ops;
1411 /* Ethernet TAP Device */
1412 ether_setup(dev);
1413 dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1414 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
1415
1416 eth_hw_addr_random(dev);
1417
1418 break;
1419 }
1420
1421 dev->min_mtu = MIN_MTU;
1422 dev->max_mtu = MAX_MTU - dev->hard_header_len;
1423}
1424
1425static bool tun_sock_writeable(struct tun_struct *tun, struct tun_file *tfile)
1426{
1427 struct sock *sk = tfile->socket.sk;
1428
1429 return (tun->dev->flags & IFF_UP) && sock_writeable(sk);
1430}
1431
1432/* Character device part */
1433
1434/* Poll */
1435static __poll_t tun_chr_poll(struct file *file, poll_table *wait)
1436{
1437 struct tun_file *tfile = file->private_data;
1438 struct tun_struct *tun = tun_get(tfile);
1439 struct sock *sk;
1440 __poll_t mask = 0;
1441
1442 if (!tun)
1443 return EPOLLERR;
1444
1445 sk = tfile->socket.sk;
1446
1447 tun_debug(KERN_INFO, tun, "tun_chr_poll\n");
1448
1449 poll_wait(file, sk_sleep(sk), wait);
1450
1451 if (!ptr_ring_empty(&tfile->tx_ring))
1452 mask |= EPOLLIN | EPOLLRDNORM;
1453
1454 /* Make sure SOCKWQ_ASYNC_NOSPACE is set if not writable to
1455 * guarantee EPOLLOUT to be raised by either here or
1456 * tun_sock_write_space(). Then process could get notification
1457 * after it writes to a down device and meets -EIO.
1458 */
1459 if (tun_sock_writeable(tun, tfile) ||
1460 (!test_and_set_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags) &&
1461 tun_sock_writeable(tun, tfile)))
1462 mask |= EPOLLOUT | EPOLLWRNORM;
1463
1464 if (tun->dev->reg_state != NETREG_REGISTERED)
1465 mask = EPOLLERR;
1466
1467 tun_put(tun);
1468 return mask;
1469}
1470
1471static struct sk_buff *tun_napi_alloc_frags(struct tun_file *tfile,
1472 size_t len,
1473 const struct iov_iter *it)
1474{
1475 struct sk_buff *skb;
1476 size_t linear;
1477 int err;
1478 int i;
1479
1480 if (it->nr_segs > MAX_SKB_FRAGS + 1)
1481 return ERR_PTR(-ENOMEM);
1482
1483 local_bh_disable();
1484 skb = napi_get_frags(&tfile->napi);
1485 local_bh_enable();
1486 if (!skb)
1487 return ERR_PTR(-ENOMEM);
1488
1489 linear = iov_iter_single_seg_count(it);
1490 err = __skb_grow(skb, linear);
1491 if (err)
1492 goto free;
1493
1494 skb->len = len;
1495 skb->data_len = len - linear;
1496 skb->truesize += skb->data_len;
1497
1498 for (i = 1; i < it->nr_segs; i++) {
1499 size_t fragsz = it->iov[i].iov_len;
1500 struct page *page;
1501 void *frag;
1502
1503 if (fragsz == 0 || fragsz > PAGE_SIZE) {
1504 err = -EINVAL;
1505 goto free;
1506 }
1507 frag = netdev_alloc_frag(fragsz);
1508 if (!frag) {
1509 err = -ENOMEM;
1510 goto free;
1511 }
1512 page = virt_to_head_page(frag);
1513 skb_fill_page_desc(skb, i - 1, page,
1514 frag - page_address(page), fragsz);
1515 }
1516
1517 return skb;
1518free:
1519 /* frees skb and all frags allocated with napi_alloc_frag() */
1520 napi_free_frags(&tfile->napi);
1521 return ERR_PTR(err);
1522}
1523
1524/* prepad is the amount to reserve at front. len is length after that.
1525 * linear is a hint as to how much to copy (usually headers). */
1526static struct sk_buff *tun_alloc_skb(struct tun_file *tfile,
1527 size_t prepad, size_t len,
1528 size_t linear, int noblock)
1529{
1530 struct sock *sk = tfile->socket.sk;
1531 struct sk_buff *skb;
1532 int err;
1533
1534 /* Under a page? Don't bother with paged skb. */
1535 if (prepad + len < PAGE_SIZE || !linear)
1536 linear = len;
1537
1538 skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
1539 &err, 0);
1540 if (!skb)
1541 return ERR_PTR(err);
1542
1543 skb_reserve(skb, prepad);
1544 skb_put(skb, linear);
1545 skb->data_len = len - linear;
1546 skb->len += len - linear;
1547
1548 return skb;
1549}
1550
1551static void tun_rx_batched(struct tun_struct *tun, struct tun_file *tfile,
1552 struct sk_buff *skb, int more)
1553{
1554 struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
1555 struct sk_buff_head process_queue;
1556 u32 rx_batched = tun->rx_batched;
1557 bool rcv = false;
1558
1559 if (!rx_batched || (!more && skb_queue_empty(queue))) {
1560 local_bh_disable();
1561 skb_record_rx_queue(skb, tfile->queue_index);
1562 netif_receive_skb(skb);
1563 local_bh_enable();
1564 return;
1565 }
1566
1567 spin_lock(&queue->lock);
1568 if (!more || skb_queue_len(queue) == rx_batched) {
1569 __skb_queue_head_init(&process_queue);
1570 skb_queue_splice_tail_init(queue, &process_queue);
1571 rcv = true;
1572 } else {
1573 __skb_queue_tail(queue, skb);
1574 }
1575 spin_unlock(&queue->lock);
1576
1577 if (rcv) {
1578 struct sk_buff *nskb;
1579
1580 local_bh_disable();
1581 while ((nskb = __skb_dequeue(&process_queue))) {
1582 skb_record_rx_queue(nskb, tfile->queue_index);
1583 netif_receive_skb(nskb);
1584 }
1585 skb_record_rx_queue(skb, tfile->queue_index);
1586 netif_receive_skb(skb);
1587 local_bh_enable();
1588 }
1589}
1590
1591static bool tun_can_build_skb(struct tun_struct *tun, struct tun_file *tfile,
1592 int len, int noblock, bool zerocopy)
1593{
1594 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
1595 return false;
1596
1597 if (tfile->socket.sk->sk_sndbuf != INT_MAX)
1598 return false;
1599
1600 if (!noblock)
1601 return false;
1602
1603 if (zerocopy)
1604 return false;
1605
1606 if (SKB_DATA_ALIGN(len + TUN_RX_PAD) +
1607 SKB_DATA_ALIGN(sizeof(struct skb_shared_info)) > PAGE_SIZE)
1608 return false;
1609
1610 return true;
1611}
1612
1613static struct sk_buff *__tun_build_skb(struct page_frag *alloc_frag, char *buf,
1614 int buflen, int len, int pad)
1615{
1616 struct sk_buff *skb = build_skb(buf, buflen);
1617
1618 if (!skb)
1619 return ERR_PTR(-ENOMEM);
1620
1621 skb_reserve(skb, pad);
1622 skb_put(skb, len);
1623
1624 get_page(alloc_frag->page);
1625 alloc_frag->offset += buflen;
1626
1627 return skb;
1628}
1629
1630static int tun_xdp_act(struct tun_struct *tun, struct bpf_prog *xdp_prog,
1631 struct xdp_buff *xdp, u32 act)
1632{
1633 int err;
1634
1635 switch (act) {
1636 case XDP_REDIRECT:
1637 err = xdp_do_redirect(tun->dev, xdp, xdp_prog);
1638 if (err)
1639 return err;
1640 break;
1641 case XDP_TX:
1642 err = tun_xdp_tx(tun->dev, xdp);
1643 if (err < 0)
1644 return err;
1645 break;
1646 case XDP_PASS:
1647 break;
1648 default:
1649 bpf_warn_invalid_xdp_action(act);
1650 /* fall through */
1651 case XDP_ABORTED:
1652 trace_xdp_exception(tun->dev, xdp_prog, act);
1653 /* fall through */
1654 case XDP_DROP:
1655 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1656 break;
1657 }
1658
1659 return act;
1660}
1661
1662static struct sk_buff *tun_build_skb(struct tun_struct *tun,
1663 struct tun_file *tfile,
1664 struct iov_iter *from,
1665 struct virtio_net_hdr *hdr,
1666 int len, int *skb_xdp)
1667{
1668 struct page_frag *alloc_frag = &current->task_frag;
1669 struct bpf_prog *xdp_prog;
1670 int buflen = SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
1671 char *buf;
1672 size_t copied;
1673 int pad = TUN_RX_PAD;
1674 int err = 0;
1675
1676 rcu_read_lock();
1677 xdp_prog = rcu_dereference(tun->xdp_prog);
1678 if (xdp_prog)
1679 pad += XDP_PACKET_HEADROOM;
1680 buflen += SKB_DATA_ALIGN(len + pad);
1681 rcu_read_unlock();
1682
1683 alloc_frag->offset = ALIGN((u64)alloc_frag->offset, SMP_CACHE_BYTES);
1684 if (unlikely(!skb_page_frag_refill(buflen, alloc_frag, GFP_KERNEL)))
1685 return ERR_PTR(-ENOMEM);
1686
1687 buf = (char *)page_address(alloc_frag->page) + alloc_frag->offset;
1688 copied = copy_page_from_iter(alloc_frag->page,
1689 alloc_frag->offset + pad,
1690 len, from);
1691 if (copied != len)
1692 return ERR_PTR(-EFAULT);
1693
1694 /* There's a small window that XDP may be set after the check
1695 * of xdp_prog above, this should be rare and for simplicity
1696 * we do XDP on skb in case the headroom is not enough.
1697 */
1698 if (hdr->gso_type || !xdp_prog) {
1699 *skb_xdp = 1;
1700 return __tun_build_skb(alloc_frag, buf, buflen, len, pad);
1701 }
1702
1703 *skb_xdp = 0;
1704
1705 local_bh_disable();
1706 rcu_read_lock();
1707 xdp_prog = rcu_dereference(tun->xdp_prog);
1708 if (xdp_prog) {
1709 struct xdp_buff xdp;
1710 u32 act;
1711
1712 xdp.data_hard_start = buf;
1713 xdp.data = buf + pad;
1714 xdp_set_data_meta_invalid(&xdp);
1715 xdp.data_end = xdp.data + len;
1716 xdp.rxq = &tfile->xdp_rxq;
1717
1718 act = bpf_prog_run_xdp(xdp_prog, &xdp);
1719 if (act == XDP_REDIRECT || act == XDP_TX) {
1720 get_page(alloc_frag->page);
1721 alloc_frag->offset += buflen;
1722 }
1723 err = tun_xdp_act(tun, xdp_prog, &xdp, act);
1724 if (err < 0)
1725 goto err_xdp;
1726 if (err == XDP_REDIRECT)
1727 xdp_do_flush_map();
1728 if (err != XDP_PASS)
1729 goto out;
1730
1731 pad = xdp.data - xdp.data_hard_start;
1732 len = xdp.data_end - xdp.data;
1733 }
1734 rcu_read_unlock();
1735 local_bh_enable();
1736
1737 return __tun_build_skb(alloc_frag, buf, buflen, len, pad);
1738
1739err_xdp:
1740 put_page(alloc_frag->page);
1741out:
1742 rcu_read_unlock();
1743 local_bh_enable();
1744 return NULL;
1745}
1746
1747/* Get packet from user space buffer */
1748static ssize_t tun_get_user(struct tun_struct *tun, struct tun_file *tfile,
1749 void *msg_control, struct iov_iter *from,
1750 int noblock, bool more)
1751{
1752 struct tun_pi pi = { 0, cpu_to_be16(ETH_P_IP) };
1753 struct sk_buff *skb;
1754 size_t total_len = iov_iter_count(from);
1755 size_t len = total_len, align = tun->align, linear;
1756 struct virtio_net_hdr gso = { 0 };
1757 struct tun_pcpu_stats *stats;
1758 int good_linear;
1759 int copylen;
1760 bool zerocopy = false;
1761 int err;
1762 u32 rxhash = 0;
1763 int skb_xdp = 1;
1764 bool frags = tun_napi_frags_enabled(tfile);
1765
1766 if (!(tun->dev->flags & IFF_UP))
1767 return -EIO;
1768
1769 if (!(tun->flags & IFF_NO_PI)) {
1770 if (len < sizeof(pi))
1771 return -EINVAL;
1772 len -= sizeof(pi);
1773
1774 if (!copy_from_iter_full(&pi, sizeof(pi), from))
1775 return -EFAULT;
1776 }
1777
1778 if (tun->flags & IFF_VNET_HDR) {
1779 int vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
1780
1781 if (len < vnet_hdr_sz)
1782 return -EINVAL;
1783 len -= vnet_hdr_sz;
1784
1785 if (!copy_from_iter_full(&gso, sizeof(gso), from))
1786 return -EFAULT;
1787
1788 if ((gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
1789 tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2 > tun16_to_cpu(tun, gso.hdr_len))
1790 gso.hdr_len = cpu_to_tun16(tun, tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2);
1791
1792 if (tun16_to_cpu(tun, gso.hdr_len) > len)
1793 return -EINVAL;
1794 iov_iter_advance(from, vnet_hdr_sz - sizeof(gso));
1795 }
1796
1797 if ((tun->flags & TUN_TYPE_MASK) == IFF_TAP) {
1798 align += NET_IP_ALIGN;
1799 if (unlikely(len < ETH_HLEN ||
1800 (gso.hdr_len && tun16_to_cpu(tun, gso.hdr_len) < ETH_HLEN)))
1801 return -EINVAL;
1802 }
1803
1804 good_linear = SKB_MAX_HEAD(align);
1805
1806 if (msg_control) {
1807 struct iov_iter i = *from;
1808
1809 /* There are 256 bytes to be copied in skb, so there is
1810 * enough room for skb expand head in case it is used.
1811 * The rest of the buffer is mapped from userspace.
1812 */
1813 copylen = gso.hdr_len ? tun16_to_cpu(tun, gso.hdr_len) : GOODCOPY_LEN;
1814 if (copylen > good_linear)
1815 copylen = good_linear;
1816 linear = copylen;
1817 iov_iter_advance(&i, copylen);
1818 if (iov_iter_npages(&i, INT_MAX) <= MAX_SKB_FRAGS)
1819 zerocopy = true;
1820 }
1821
1822 if (!frags && tun_can_build_skb(tun, tfile, len, noblock, zerocopy)) {
1823 /* For the packet that is not easy to be processed
1824 * (e.g gso or jumbo packet), we will do it at after
1825 * skb was created with generic XDP routine.
1826 */
1827 skb = tun_build_skb(tun, tfile, from, &gso, len, &skb_xdp);
1828 if (IS_ERR(skb)) {
1829 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1830 return PTR_ERR(skb);
1831 }
1832 if (!skb)
1833 return total_len;
1834 } else {
1835 if (!zerocopy) {
1836 copylen = len;
1837 if (tun16_to_cpu(tun, gso.hdr_len) > good_linear)
1838 linear = good_linear;
1839 else
1840 linear = tun16_to_cpu(tun, gso.hdr_len);
1841 }
1842
1843 if (frags) {
1844 mutex_lock(&tfile->napi_mutex);
1845 skb = tun_napi_alloc_frags(tfile, copylen, from);
1846 /* tun_napi_alloc_frags() enforces a layout for the skb.
1847 * If zerocopy is enabled, then this layout will be
1848 * overwritten by zerocopy_sg_from_iter().
1849 */
1850 zerocopy = false;
1851 } else {
1852 skb = tun_alloc_skb(tfile, align, copylen, linear,
1853 noblock);
1854 }
1855
1856 if (IS_ERR(skb)) {
1857 if (PTR_ERR(skb) != -EAGAIN)
1858 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1859 if (frags)
1860 mutex_unlock(&tfile->napi_mutex);
1861 return PTR_ERR(skb);
1862 }
1863
1864 if (zerocopy)
1865 err = zerocopy_sg_from_iter(skb, from);
1866 else
1867 err = skb_copy_datagram_from_iter(skb, 0, from, len);
1868
1869 if (err) {
1870 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1871 kfree_skb(skb);
1872 if (frags) {
1873 tfile->napi.skb = NULL;
1874 mutex_unlock(&tfile->napi_mutex);
1875 }
1876
1877 return -EFAULT;
1878 }
1879 }
1880
1881 if (virtio_net_hdr_to_skb(skb, &gso, tun_is_little_endian(tun))) {
1882 this_cpu_inc(tun->pcpu_stats->rx_frame_errors);
1883 kfree_skb(skb);
1884 if (frags) {
1885 tfile->napi.skb = NULL;
1886 mutex_unlock(&tfile->napi_mutex);
1887 }
1888
1889 return -EINVAL;
1890 }
1891
1892 switch (tun->flags & TUN_TYPE_MASK) {
1893 case IFF_TUN:
1894 if (tun->flags & IFF_NO_PI) {
1895 u8 ip_version = skb->len ? (skb->data[0] >> 4) : 0;
1896
1897 switch (ip_version) {
1898 case 4:
1899 pi.proto = htons(ETH_P_IP);
1900 break;
1901 case 6:
1902 pi.proto = htons(ETH_P_IPV6);
1903 break;
1904 default:
1905 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1906 kfree_skb(skb);
1907 return -EINVAL;
1908 }
1909 }
1910
1911 skb_reset_mac_header(skb);
1912 skb->protocol = pi.proto;
1913 skb->dev = tun->dev;
1914 break;
1915 case IFF_TAP:
1916 if (!frags)
1917 skb->protocol = eth_type_trans(skb, tun->dev);
1918 break;
1919 }
1920
1921 /* copy skb_ubuf_info for callback when skb has no error */
1922 if (zerocopy) {
1923 skb_shinfo(skb)->destructor_arg = msg_control;
1924 skb_shinfo(skb)->tx_flags |= SKBTX_DEV_ZEROCOPY;
1925 skb_shinfo(skb)->tx_flags |= SKBTX_SHARED_FRAG;
1926 } else if (msg_control) {
1927 struct ubuf_info *uarg = msg_control;
1928 uarg->callback(uarg, false);
1929 }
1930
1931 skb_reset_network_header(skb);
1932 skb_probe_transport_header(skb);
1933
1934 if (skb_xdp) {
1935 struct bpf_prog *xdp_prog;
1936 int ret;
1937
1938 local_bh_disable();
1939 rcu_read_lock();
1940 xdp_prog = rcu_dereference(tun->xdp_prog);
1941 if (xdp_prog) {
1942 ret = do_xdp_generic(xdp_prog, skb);
1943 if (ret != XDP_PASS) {
1944 rcu_read_unlock();
1945 local_bh_enable();
1946 return total_len;
1947 }
1948 }
1949 rcu_read_unlock();
1950 local_bh_enable();
1951 }
1952
1953 /* Compute the costly rx hash only if needed for flow updates.
1954 * We may get a very small possibility of OOO during switching, not
1955 * worth to optimize.
1956 */
1957 if (!rcu_access_pointer(tun->steering_prog) && tun->numqueues > 1 &&
1958 !tfile->detached)
1959 rxhash = __skb_get_hash_symmetric(skb);
1960
1961 if (frags) {
1962 /* Exercise flow dissector code path. */
1963 u32 headlen = eth_get_headlen(skb->data, skb_headlen(skb));
1964
1965 if (unlikely(headlen > skb_headlen(skb))) {
1966 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1967 napi_free_frags(&tfile->napi);
1968 mutex_unlock(&tfile->napi_mutex);
1969 WARN_ON(1);
1970 return -ENOMEM;
1971 }
1972
1973 local_bh_disable();
1974 napi_gro_frags(&tfile->napi);
1975 local_bh_enable();
1976 mutex_unlock(&tfile->napi_mutex);
1977 } else if (tfile->napi_enabled) {
1978 struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
1979 int queue_len;
1980
1981 spin_lock_bh(&queue->lock);
1982 __skb_queue_tail(queue, skb);
1983 queue_len = skb_queue_len(queue);
1984 spin_unlock(&queue->lock);
1985
1986 if (!more || queue_len > NAPI_POLL_WEIGHT)
1987 napi_schedule(&tfile->napi);
1988
1989 local_bh_enable();
1990 } else if (!IS_ENABLED(CONFIG_4KSTACKS)) {
1991 tun_rx_batched(tun, tfile, skb, more);
1992 } else {
1993 netif_rx_ni(skb);
1994 }
1995
1996 stats = get_cpu_ptr(tun->pcpu_stats);
1997 u64_stats_update_begin(&stats->syncp);
1998 stats->rx_packets++;
1999 stats->rx_bytes += len;
2000 u64_stats_update_end(&stats->syncp);
2001 put_cpu_ptr(stats);
2002
2003 if (rxhash)
2004 tun_flow_update(tun, rxhash, tfile);
2005
2006 return total_len;
2007}
2008
2009static ssize_t tun_chr_write_iter(struct kiocb *iocb, struct iov_iter *from)
2010{
2011 struct file *file = iocb->ki_filp;
2012 struct tun_file *tfile = file->private_data;
2013 struct tun_struct *tun = tun_get(tfile);
2014 ssize_t result;
2015
2016 if (!tun)
2017 return -EBADFD;
2018
2019 result = tun_get_user(tun, tfile, NULL, from,
2020 file->f_flags & O_NONBLOCK, false);
2021
2022 tun_put(tun);
2023 return result;
2024}
2025
2026static ssize_t tun_put_user_xdp(struct tun_struct *tun,
2027 struct tun_file *tfile,
2028 struct xdp_frame *xdp_frame,
2029 struct iov_iter *iter)
2030{
2031 int vnet_hdr_sz = 0;
2032 size_t size = xdp_frame->len;
2033 struct tun_pcpu_stats *stats;
2034 size_t ret;
2035
2036 if (tun->flags & IFF_VNET_HDR) {
2037 struct virtio_net_hdr gso = { 0 };
2038
2039 vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
2040 if (unlikely(iov_iter_count(iter) < vnet_hdr_sz))
2041 return -EINVAL;
2042 if (unlikely(copy_to_iter(&gso, sizeof(gso), iter) !=
2043 sizeof(gso)))
2044 return -EFAULT;
2045 iov_iter_advance(iter, vnet_hdr_sz - sizeof(gso));
2046 }
2047
2048 ret = copy_to_iter(xdp_frame->data, size, iter) + vnet_hdr_sz;
2049
2050 stats = get_cpu_ptr(tun->pcpu_stats);
2051 u64_stats_update_begin(&stats->syncp);
2052 stats->tx_packets++;
2053 stats->tx_bytes += ret;
2054 u64_stats_update_end(&stats->syncp);
2055 put_cpu_ptr(tun->pcpu_stats);
2056
2057 return ret;
2058}
2059
2060/* Put packet to the user space buffer */
2061static ssize_t tun_put_user(struct tun_struct *tun,
2062 struct tun_file *tfile,
2063 struct sk_buff *skb,
2064 struct iov_iter *iter)
2065{
2066 struct tun_pi pi = { 0, skb->protocol };
2067 struct tun_pcpu_stats *stats;
2068 ssize_t total;
2069 int vlan_offset = 0;
2070 int vlan_hlen = 0;
2071 int vnet_hdr_sz = 0;
2072
2073 if (skb_vlan_tag_present(skb))
2074 vlan_hlen = VLAN_HLEN;
2075
2076 if (tun->flags & IFF_VNET_HDR)
2077 vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
2078
2079 total = skb->len + vlan_hlen + vnet_hdr_sz;
2080
2081 if (!(tun->flags & IFF_NO_PI)) {
2082 if (iov_iter_count(iter) < sizeof(pi))
2083 return -EINVAL;
2084
2085 total += sizeof(pi);
2086 if (iov_iter_count(iter) < total) {
2087 /* Packet will be striped */
2088 pi.flags |= TUN_PKT_STRIP;
2089 }
2090
2091 if (copy_to_iter(&pi, sizeof(pi), iter) != sizeof(pi))
2092 return -EFAULT;
2093 }
2094
2095 if (vnet_hdr_sz) {
2096 struct virtio_net_hdr gso;
2097
2098 if (iov_iter_count(iter) < vnet_hdr_sz)
2099 return -EINVAL;
2100
2101 if (virtio_net_hdr_from_skb(skb, &gso,
2102 tun_is_little_endian(tun), true,
2103 vlan_hlen)) {
2104 struct skb_shared_info *sinfo = skb_shinfo(skb);
2105 pr_err("unexpected GSO type: "
2106 "0x%x, gso_size %d, hdr_len %d\n",
2107 sinfo->gso_type, tun16_to_cpu(tun, gso.gso_size),
2108 tun16_to_cpu(tun, gso.hdr_len));
2109 print_hex_dump(KERN_ERR, "tun: ",
2110 DUMP_PREFIX_NONE,
2111 16, 1, skb->head,
2112 min((int)tun16_to_cpu(tun, gso.hdr_len), 64), true);
2113 WARN_ON_ONCE(1);
2114 return -EINVAL;
2115 }
2116
2117 if (copy_to_iter(&gso, sizeof(gso), iter) != sizeof(gso))
2118 return -EFAULT;
2119
2120 iov_iter_advance(iter, vnet_hdr_sz - sizeof(gso));
2121 }
2122
2123 if (vlan_hlen) {
2124 int ret;
2125 struct veth veth;
2126
2127 veth.h_vlan_proto = skb->vlan_proto;
2128 veth.h_vlan_TCI = htons(skb_vlan_tag_get(skb));
2129
2130 vlan_offset = offsetof(struct vlan_ethhdr, h_vlan_proto);
2131
2132 ret = skb_copy_datagram_iter(skb, 0, iter, vlan_offset);
2133 if (ret || !iov_iter_count(iter))
2134 goto done;
2135
2136 ret = copy_to_iter(&veth, sizeof(veth), iter);
2137 if (ret != sizeof(veth) || !iov_iter_count(iter))
2138 goto done;
2139 }
2140
2141 skb_copy_datagram_iter(skb, vlan_offset, iter, skb->len - vlan_offset);
2142
2143done:
2144 /* caller is in process context, */
2145 stats = get_cpu_ptr(tun->pcpu_stats);
2146 u64_stats_update_begin(&stats->syncp);
2147 stats->tx_packets++;
2148 stats->tx_bytes += skb->len + vlan_hlen;
2149 u64_stats_update_end(&stats->syncp);
2150 put_cpu_ptr(tun->pcpu_stats);
2151
2152 return total;
2153}
2154
2155static void *tun_ring_recv(struct tun_file *tfile, int noblock, int *err)
2156{
2157 DECLARE_WAITQUEUE(wait, current);
2158 void *ptr = NULL;
2159 int error = 0;
2160
2161 ptr = ptr_ring_consume(&tfile->tx_ring);
2162 if (ptr)
2163 goto out;
2164 if (noblock) {
2165 error = -EAGAIN;
2166 goto out;
2167 }
2168
2169 add_wait_queue(&tfile->wq.wait, &wait);
2170
2171 while (1) {
2172 set_current_state(TASK_INTERRUPTIBLE);
2173 ptr = ptr_ring_consume(&tfile->tx_ring);
2174 if (ptr)
2175 break;
2176 if (signal_pending(current)) {
2177 error = -ERESTARTSYS;
2178 break;
2179 }
2180 if (tfile->socket.sk->sk_shutdown & RCV_SHUTDOWN) {
2181 error = -EFAULT;
2182 break;
2183 }
2184
2185 schedule();
2186 }
2187
2188 __set_current_state(TASK_RUNNING);
2189 remove_wait_queue(&tfile->wq.wait, &wait);
2190
2191out:
2192 *err = error;
2193 return ptr;
2194}
2195
2196static ssize_t tun_do_read(struct tun_struct *tun, struct tun_file *tfile,
2197 struct iov_iter *to,
2198 int noblock, void *ptr)
2199{
2200 ssize_t ret;
2201 int err;
2202
2203 tun_debug(KERN_INFO, tun, "tun_do_read\n");
2204
2205 if (!iov_iter_count(to)) {
2206 tun_ptr_free(ptr);
2207 return 0;
2208 }
2209
2210 if (!ptr) {
2211 /* Read frames from ring */
2212 ptr = tun_ring_recv(tfile, noblock, &err);
2213 if (!ptr)
2214 return err;
2215 }
2216
2217 if (tun_is_xdp_frame(ptr)) {
2218 struct xdp_frame *xdpf = tun_ptr_to_xdp(ptr);
2219
2220 ret = tun_put_user_xdp(tun, tfile, xdpf, to);
2221 xdp_return_frame(xdpf);
2222 } else {
2223 struct sk_buff *skb = ptr;
2224
2225 ret = tun_put_user(tun, tfile, skb, to);
2226 if (unlikely(ret < 0))
2227 kfree_skb(skb);
2228 else
2229 consume_skb(skb);
2230 }
2231
2232 return ret;
2233}
2234
2235static ssize_t tun_chr_read_iter(struct kiocb *iocb, struct iov_iter *to)
2236{
2237 struct file *file = iocb->ki_filp;
2238 struct tun_file *tfile = file->private_data;
2239 struct tun_struct *tun = tun_get(tfile);
2240 ssize_t len = iov_iter_count(to), ret;
2241
2242 if (!tun)
2243 return -EBADFD;
2244 ret = tun_do_read(tun, tfile, to, file->f_flags & O_NONBLOCK, NULL);
2245 ret = min_t(ssize_t, ret, len);
2246 if (ret > 0)
2247 iocb->ki_pos = ret;
2248 tun_put(tun);
2249 return ret;
2250}
2251
2252static void tun_prog_free(struct rcu_head *rcu)
2253{
2254 struct tun_prog *prog = container_of(rcu, struct tun_prog, rcu);
2255
2256 bpf_prog_destroy(prog->prog);
2257 kfree(prog);
2258}
2259
2260static int __tun_set_ebpf(struct tun_struct *tun,
2261 struct tun_prog __rcu **prog_p,
2262 struct bpf_prog *prog)
2263{
2264 struct tun_prog *old, *new = NULL;
2265
2266 if (prog) {
2267 new = kmalloc(sizeof(*new), GFP_KERNEL);
2268 if (!new)
2269 return -ENOMEM;
2270 new->prog = prog;
2271 }
2272
2273 spin_lock_bh(&tun->lock);
2274 old = rcu_dereference_protected(*prog_p,
2275 lockdep_is_held(&tun->lock));
2276 rcu_assign_pointer(*prog_p, new);
2277 spin_unlock_bh(&tun->lock);
2278
2279 if (old)
2280 call_rcu(&old->rcu, tun_prog_free);
2281
2282 return 0;
2283}
2284
2285static void tun_free_netdev(struct net_device *dev)
2286{
2287 struct tun_struct *tun = netdev_priv(dev);
2288
2289 BUG_ON(!(list_empty(&tun->disabled)));
2290 free_percpu(tun->pcpu_stats);
2291 tun_flow_uninit(tun);
2292 security_tun_dev_free_security(tun->security);
2293 __tun_set_ebpf(tun, &tun->steering_prog, NULL);
2294 __tun_set_ebpf(tun, &tun->filter_prog, NULL);
2295}
2296
2297static void tun_setup(struct net_device *dev)
2298{
2299 struct tun_struct *tun = netdev_priv(dev);
2300
2301 tun->owner = INVALID_UID;
2302 tun->group = INVALID_GID;
2303 tun_default_link_ksettings(dev, &tun->link_ksettings);
2304
2305 dev->ethtool_ops = &tun_ethtool_ops;
2306 dev->needs_free_netdev = true;
2307 dev->priv_destructor = tun_free_netdev;
2308 /* We prefer our own queue length */
2309 dev->tx_queue_len = TUN_READQ_SIZE;
2310}
2311
2312/* Trivial set of netlink ops to allow deleting tun or tap
2313 * device with netlink.
2314 */
2315static int tun_validate(struct nlattr *tb[], struct nlattr *data[],
2316 struct netlink_ext_ack *extack)
2317{
2318 NL_SET_ERR_MSG(extack,
2319 "tun/tap creation via rtnetlink is not supported.");
2320 return -EOPNOTSUPP;
2321}
2322
2323static size_t tun_get_size(const struct net_device *dev)
2324{
2325 BUILD_BUG_ON(sizeof(u32) != sizeof(uid_t));
2326 BUILD_BUG_ON(sizeof(u32) != sizeof(gid_t));
2327
2328 return nla_total_size(sizeof(uid_t)) + /* OWNER */
2329 nla_total_size(sizeof(gid_t)) + /* GROUP */
2330 nla_total_size(sizeof(u8)) + /* TYPE */
2331 nla_total_size(sizeof(u8)) + /* PI */
2332 nla_total_size(sizeof(u8)) + /* VNET_HDR */
2333 nla_total_size(sizeof(u8)) + /* PERSIST */
2334 nla_total_size(sizeof(u8)) + /* MULTI_QUEUE */
2335 nla_total_size(sizeof(u32)) + /* NUM_QUEUES */
2336 nla_total_size(sizeof(u32)) + /* NUM_DISABLED_QUEUES */
2337 0;
2338}
2339
2340static int tun_fill_info(struct sk_buff *skb, const struct net_device *dev)
2341{
2342 struct tun_struct *tun = netdev_priv(dev);
2343
2344 if (nla_put_u8(skb, IFLA_TUN_TYPE, tun->flags & TUN_TYPE_MASK))
2345 goto nla_put_failure;
2346 if (uid_valid(tun->owner) &&
2347 nla_put_u32(skb, IFLA_TUN_OWNER,
2348 from_kuid_munged(current_user_ns(), tun->owner)))
2349 goto nla_put_failure;
2350 if (gid_valid(tun->group) &&
2351 nla_put_u32(skb, IFLA_TUN_GROUP,
2352 from_kgid_munged(current_user_ns(), tun->group)))
2353 goto nla_put_failure;
2354 if (nla_put_u8(skb, IFLA_TUN_PI, !(tun->flags & IFF_NO_PI)))
2355 goto nla_put_failure;
2356 if (nla_put_u8(skb, IFLA_TUN_VNET_HDR, !!(tun->flags & IFF_VNET_HDR)))
2357 goto nla_put_failure;
2358 if (nla_put_u8(skb, IFLA_TUN_PERSIST, !!(tun->flags & IFF_PERSIST)))
2359 goto nla_put_failure;
2360 if (nla_put_u8(skb, IFLA_TUN_MULTI_QUEUE,
2361 !!(tun->flags & IFF_MULTI_QUEUE)))
2362 goto nla_put_failure;
2363 if (tun->flags & IFF_MULTI_QUEUE) {
2364 if (nla_put_u32(skb, IFLA_TUN_NUM_QUEUES, tun->numqueues))
2365 goto nla_put_failure;
2366 if (nla_put_u32(skb, IFLA_TUN_NUM_DISABLED_QUEUES,
2367 tun->numdisabled))
2368 goto nla_put_failure;
2369 }
2370
2371 return 0;
2372
2373nla_put_failure:
2374 return -EMSGSIZE;
2375}
2376
2377static struct rtnl_link_ops tun_link_ops __read_mostly = {
2378 .kind = DRV_NAME,
2379 .priv_size = sizeof(struct tun_struct),
2380 .setup = tun_setup,
2381 .validate = tun_validate,
2382 .get_size = tun_get_size,
2383 .fill_info = tun_fill_info,
2384};
2385
2386static void tun_sock_write_space(struct sock *sk)
2387{
2388 struct tun_file *tfile;
2389 wait_queue_head_t *wqueue;
2390
2391 if (!sock_writeable(sk))
2392 return;
2393
2394 if (!test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags))
2395 return;
2396
2397 wqueue = sk_sleep(sk);
2398 if (wqueue && waitqueue_active(wqueue))
2399 wake_up_interruptible_sync_poll(wqueue, EPOLLOUT |
2400 EPOLLWRNORM | EPOLLWRBAND);
2401
2402 tfile = container_of(sk, struct tun_file, sk);
2403 kill_fasync(&tfile->fasync, SIGIO, POLL_OUT);
2404}
2405
2406static void tun_put_page(struct tun_page *tpage)
2407{
2408 if (tpage->page)
2409 __page_frag_cache_drain(tpage->page, tpage->count);
2410}
2411
2412static int tun_xdp_one(struct tun_struct *tun,
2413 struct tun_file *tfile,
2414 struct xdp_buff *xdp, int *flush,
2415 struct tun_page *tpage)
2416{
2417 unsigned int datasize = xdp->data_end - xdp->data;
2418 struct tun_xdp_hdr *hdr = xdp->data_hard_start;
2419 struct virtio_net_hdr *gso = &hdr->gso;
2420 struct tun_pcpu_stats *stats;
2421 struct bpf_prog *xdp_prog;
2422 struct sk_buff *skb = NULL;
2423 u32 rxhash = 0, act;
2424 int buflen = hdr->buflen;
2425 int err = 0;
2426 bool skb_xdp = false;
2427 struct page *page;
2428
2429 xdp_prog = rcu_dereference(tun->xdp_prog);
2430 if (xdp_prog) {
2431 if (gso->gso_type) {
2432 skb_xdp = true;
2433 goto build;
2434 }
2435 xdp_set_data_meta_invalid(xdp);
2436 xdp->rxq = &tfile->xdp_rxq;
2437
2438 act = bpf_prog_run_xdp(xdp_prog, xdp);
2439 err = tun_xdp_act(tun, xdp_prog, xdp, act);
2440 if (err < 0) {
2441 put_page(virt_to_head_page(xdp->data));
2442 return err;
2443 }
2444
2445 switch (err) {
2446 case XDP_REDIRECT:
2447 *flush = true;
2448 /* fall through */
2449 case XDP_TX:
2450 return 0;
2451 case XDP_PASS:
2452 break;
2453 default:
2454 page = virt_to_head_page(xdp->data);
2455 if (tpage->page == page) {
2456 ++tpage->count;
2457 } else {
2458 tun_put_page(tpage);
2459 tpage->page = page;
2460 tpage->count = 1;
2461 }
2462 return 0;
2463 }
2464 }
2465
2466build:
2467 skb = build_skb(xdp->data_hard_start, buflen);
2468 if (!skb) {
2469 err = -ENOMEM;
2470 goto out;
2471 }
2472
2473 skb_reserve(skb, xdp->data - xdp->data_hard_start);
2474 skb_put(skb, xdp->data_end - xdp->data);
2475
2476 if (virtio_net_hdr_to_skb(skb, gso, tun_is_little_endian(tun))) {
2477